ILLUMINATING THE DEVELOPMENTAL ETIOLOGY OF YOUTH RESILIENCE By Alexandra Y. Vazquez A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the degree of Psychology – Doctor of Philosophy 2024 ABSTRACT Decades of studies have demonstrated that residing in impoverished and dangerous neighborhoods places youth at risk for many maladaptive outcomes, including poor academic performance and psychopathology (Campbell et al., 2000; Winslow & Shaw, 2007; Wodtke et al., 2011). Even so, a large proportion of exposed youth (approximately 40-60%; Vanderbilt- Adriance & Shaw, 2008) evidence resilience, or successful adaptation and competence in the face of adversity. The high prevalence of resilience is a cause for optimism, not only because it focuses scientific attention on children’s strengths in the face of chronic stressors, but also because resilient youth provide a model of successful adaptation that could be used to enhance policy and interventions for non-resilient youth facing similar adversity. Despite its promise, however, little is known about the developmental trajectories that characterize resilience or the specific biological mechanisms underlying those trajectories. The studies in this dissertation thus sought to bridge these gaps in the resilience literature. Study 1 employed variable and person-centered approaches to elucidate the development of social and psychological resilience and identify the socioecological factors that promote their development. Generally, youth were characterized by increasing trajectories of social and psychological resilience; when examined using growth mixture models, however, we found evidence of three psychological resilience trajectories and two social resilience trajectories. Parental nurturance, neighborhood social cohesion, neighborhood informal social control, and sex predicted psychological resilience trajectory class membership, while race/ethnicity and parental nurturance predicted social resilience trajectory class membership. Study 2 focused on evaluating the role of DNA Methylation (DNAm) as a specific biological mechanism undergirding youth resilience. We performed a series of methylome-wide association analyses to evaluate whether DNAm was associated with social and psychological resilience growth factors. We identified only one differentially methylated probe (DMP) for the social resilience intercept and multiple suggestive DMPs for each outcome. We then leveraged a monozygotic twin difference design to circumvent genetic confounds; only five top DMPs for the social resilience intercept were significant, suggesting that these were environmental in origin while remaining DMPs for all outcomes appear to be genetically or developmentally mediated. Finally, pathway analyses revealed multiple enriched pathways implicated in the social resilience intercept and slope. Our findings ultimately suggest that DNAm may play at most a modest role in the development of youth resilience, particularly in the psychological domain. This set of studies ultimately facilitated greater insight into the development of resilience during prominent developmental transitions. While we identified socioecological factors that promote resilience development, we were unable to identify clear biological mechanisms undergirding resilience trajectories over time. Finally, future directions for the resilience literature more broadly are discussed, including work that: 1) further elucidates developmental patterns of resilience, 2) investigates other potential biological mechanisms influencing resilience, and 3) begins to meaningfully incorporate multicultural and social justice considerations. Copyright by ALEXANDRA Y. VAZQUEZ 2024 This dissertation is dedicated to all who fight to have a rich and meaningful life despite the adversity they have endured. “I can be changed by what happens to me. But I refuse to be reduced by it.” ― Maya Angelou v ACKNOWLEDGEMENTS I am immensely grateful to the many people who have supported and guided me in the completion of this dissertation and my doctoral studies more broadly, without whom, this work would not have been possible. First and foremost, thank you to my advisor and doctoral guidance committee chair, Dr. S. Alexandra Burt. I feel so incredibly fortunate to have a mentor who is as brilliant as she is compassionate. I would not be the scholar I am today without her guidance and dedicated mentorship. Thank you also to my doctoral guidance committee members, Dr. Brent Donnellan, Dr. Shaunna Clark, and Dr. Alytia Levendosky for lending their expertise and feedback on this dissertation. My completion of the analyses throughout this dissertation would not have been possible without Dr. Clark’s expert guidance. I would also like to thank my mentor and dear friend, Dr. Elizabeth Shewark, who has been seminal in helping me navigate the finer points of research throughout graduate school, and most critically, has endlessly cheered me on and helped me cultivate my identity as a researcher. In addition, I am deeply appreciative of the time that Dr. Levendosky has dedicated to helping me grow into a well-rounded psychologist. As clinical supervisors, both Dr. Levendosky’s and Dr. Joshua Turchan’s mastery of psychodynamic therapy has been deeply influential in my development as a scientist-practitioner. I am also forever grateful for the support of my MSU peers, including Alexandra, Chris, Aksheya, Dom, Diondra, Evan, Lauren, Sarah, and Megan. These friends brought balance and joy to the at times, painstaking graduate school experience. This dissertation was also completed during my doctoral internship, during which time I am grateful to have had the company of my vi cohort members, Kennedy and Arielle, who provided the sarcasm and levity I needed to push through weeks of intensive clinical work and writing. Most importantly, words cannot express how profoundly grateful I am for the support of my family. All four of my grandparents, my mother, and my father paved the way for me to pursue a graduate education. I would not be where I am or who I am without their sacrifices and love. To my sisters, your unconditional friendship is such a blessing, thank you for always making your presence and support felt from afar. I am likewise grateful to have loving and playful nephews, a beautiful newborn niece, and my cat Joon to pull me away from work for much needed breaks. Finally, thank you to my funding sources, including the Eunice Kennedy Shriver National Institute of Child Health and Human Development, the National Science Foundation, and the Michigan State University Graduate School. vii TABLE OF CONTENTS GENERAL INTRODUCTION........................................................................................................1 STUDY 1: DEVELOPMENTAL TRAJECTORIES OF MULTIDIMENSIONAL YOUTH RESILIENCE ..............................................................................................................................10 STUDY 2: EXPLORING DNA METHYLATION AS A PREDICTOR OF YOUTH RESILIENCE TRAJECTORIES...................................................................................................44 GENERAL DISCUSSION ...........................................................................................................81 REFERENCES…..........................................................................................................................97 APPENDIX…………………………………………………… .................................................108 viii GENERAL INTRODUCTION Two central qualities that characterize the construct of resilience include its dynamism and its multi-dimensionality. The multidimensional taxonomy of individual resilience (MITR; Miller-Graff, 2020) proposes that there are two fairly common and theoretically grounded approaches to conceptualizing resilience (Miller-Graff, 2020). The first emphasizes individual characteristics (e.g., agency) and external resources (e.g., familial support) that are likely to bolster resilience; such approaches typically regard resilience as a process, sometimes referred to as generative resilience (Miller-Graff, 2020). The second approach is quite distinct in that it focuses only on observable and quantifiable outcomes indicative of successful adaptation, or manifested resilience (Miller-Graff, 2020). The latter approach regards internal and external resources as promotive factors that may foster resilient outcomes but are not necessarily indicative of resilience itself. Given the observable and quantifiable nature of manifested resilience, the studies in this dissertation will employ this conceptualization. There is now robust evidence that manifested resilience may be best conceptualized as multifaceted, such that youth can demonstrate adaptiveness in one area but not in others (Luthar, 2006a; Masten & Curtis, 2000). For example, an individual might demonstrate psychological resilience but not social resilience. Consistent with this, prior studies have demonstrated that the prevalence rate of resilience varies across domains. In our previous work, for instance, my coauthors and I made use of factor analyses to provide empirical evidence of the theorized presence of three distinct domains of resilience to neighborhood disadvantage: psychological health, social engagement, and scholastic success (S. A. Burt et al., 2021). We found that the prevalence rates of resilience varied across domains, ranging from 52% for psychological resilience to 87% for scholastic or academic resilience among a predominantly White sample 1 (79%; 55% male, 45% female; Burt et al., 2021). However, only 39% of youth exhibited resilience across domains (S. A. Burt et al., 2021). These findings are consistent with other literature as well. Yoon and colleagues (Yoon, Maguire-Jack, et al., 2021) examined 771 racially and ethnically diverse adolescents (42% White, 22% Black, 25% Hispanic, 11% Other; 57% female, 43% male) in families investigated by child protective services, and found prevalence rates ranging from 78% for ‘externalizing resilience’ to 95% for social resilience. In short, although very few studies examine multiple domains of resilience, those that do typically see variability in their prevalence rates. In addition to being multifaceted, manifest resilience (Miller-Graff, 2020) is also widely regarded as a dynamic construct that can develop and shift over time (Luthar et al., 2015; Masten & Cicchetti, 2016). Despite this, we note that the majority of empirical studies on resilience are cross-sectional and assess youth during middle childhood (Werner, 2012). Multi-wave examinations of resilience development are relatively rare. A handful of studies have investigated two-time points via preliminary analyses. For instance, in the aforementioned study (S. A. Burt et al., 2021), my colleagues and I preliminarily examined stability from middle childhood to early adolescence (i.e., across two-time points), and found that while most domains of resilience were fairly stable, there was considerable variability; academic resilience demonstrated more stability than other domains while social resilience demonstrated less stability (S. A. Burt et al., 2021). Longitudinal examinations across two waves of data collected 18 months apart by Yoon et al (Yoon, Maguire-Jack, et al., 2021) similarly revealed that the cognitive resilience domain was most stable (i.e., approximately 74% meeting criteria at both waves), followed by the social resilience domain (i.e., approximately 73% meeting criteria at both waves), the internalizing resilience domain (i.e., 61% meeting criteria at both waves), and the externalizing resilience 2 domain (i.e., approximately 56% meeting criteria at both waves; Yoon et al., 2021). Finally, a two-wave study of racially diverse young children (55% Black, 26% White, 13% Other, 6% Hispanic; 49% male, 51% female; Dubowitz et al., 2016) experiencing maltreatment also found that while resilience is generally quite stable, the degree of stability varied across domains. They specifically found that developmental resilience (i.e., those meeting psychomotor and cognitive/intelligence developmental milestones) was the most stable (persisting in 66% of the sample), followed by behavioral resilience (persisting in 60% of the sample), and social resilience (persisting in 38% of the sample). Taken together, these results suggest that a thorough examination of resilience trajectories across three or more time points would likely yield divergent results across domains. To the best of our knowledge, only two studies have used advanced longitudinal models (e.g., growth curve models) to investigate resilient outcomes across three or more time points. Yoon and colleagues (Yoon, Sattler, et al., 2021) conducted a study on the development of resilience over 36 months in the aforementioned national longitudinal sample of 771 racially and ethnically diverse families investigated by child protective services for child abuse or neglect. They employed latent growth curve models to investigate changes in resilience across 3 waves from early to late adolescence. Results suggest that adolescents exhibited a significant increase in resilience across the 36-month period. In addition, Yoon and colleagues (Yoon, Sattler, et al., 2021) also demonstrated that caregiver-child relationships predicted higher initial levels of resilience and slower increases over time, whereas deviant peer affiliation and the receipt of behavioral health services predicted lower initial levels of resilience and more rapid increases. Another study by this research group (Sattler et al., 2023) investigated trajectories of resilience in early childhood among 1,699 racially and ethnically diverse families (42% White, 28% Black, 3 30% Hispanic/Other; 50% female, 50% male) investigated by child protective services. While resilience was conceptualized multidimensionally, a composite was used to perform repeated measures latent class analysis. They uncovered three trajectories characterized by 1) stable and low levels of resilience (31.1% of the sample), 2) increasing levels of resilience (17.4% of the sample), and 3) decreasing levels of resilience (51.5% of the sample). Item response probability estimates demonstrated that the proportion of participants in each class varied across domains of resilience. Such findings collectively suggest that, although we may expect resilience to generally increase over time, there is significant variability across persons and potentially developmental windows in that general trend. Although the above studies examining 3+ time points provide important insight into the development of resilience, we note that none have performed separate analyses for multiple resilience domains to see if their trajectory classes vary. What’s more, studies have yet to examine trajectories of resilience during prominent developmental transitions (e.g., middle childhood to adolescence). There thus remains a clear need for a longitudinal examination of resilience and promotive factors that employs advanced statistical models and independently examines multiple domains of resilience across multiple developmental periods. Study 1 of this dissertation sought to do just this, using growth curve and growth mixture models to identify developmental trajectories of domain-specific resilience to neighborhood disadvantage from middle childhood to emerging adulthood, while also investigating the role of parental nurturance and neighborhood social processes in these trajectories. The origins of resilience. Historically, resilience researchers have focused all but exclusively on environmental factors that are thought to buffer against the negative effects of stress. This research has consistently highlighted the role of particular environmental experiences 4 in fostering resilience, namely parental warmth, familial cohesion, and neighborhood cohesion (Masten & Cicchetti, 2016). Although such work has provided critical insights for the field, it is noteworthy that it has by and large failed to incorporate individual biological or genetic influences. This is a rather surprising omission, as there is now overwhelming evidence of pervasive genetic influences on human outcomes, including resilience. For example, our prior work revealed significant genetic influences on child resilience (Vazquez et al., 2021), while also highlighting the clear importance of environmental influences. Consistent with the multi-faceted nature of youth resilience, however, these heritability estimates varied considerably across resilience domains: 22% for social resilience and 40% for psychological resilience (Vazquez et al., 2021). Fortunately, contemporary theories have also begun to incorporate considerations of biological influences. The biopsychosocial framework, for instance (Feder et al., 2019), highlights the interactions between genes and the environment, incorporating epigenetics, neural circuity, individual factors (e.g., self‐efficacy), and interventions (e.g., youth developmental programs). More specifically, the biopsychosocial model posits that genes and environmental influences shape gene expression via epigenetic mechanisms, which in turn impact neural circuitry and the stress response system. Next, the model proposes that neural circuitry and stress response systems have a bidirectional relationship with psychological factors (e.g., self-efficacy, cognitive reappraisal) and collectively predict resilience. Importantly, the biopsychosocial model also includes therapeutic and preventative interventions aimed at promoting resilience and notes that these experiences can then further shape neural circuitry, stress response systems, and psychological factors. Given these theoretical advancements, it would be important for our 5 understanding of the developmental origins of resilience to meaningfully incorporate individual genetic and biological influences into empirical studies of resilience-promotive factors. One proposed mechanism through which environmental factors may enable resilience is by buffering against the biological embedding of stress (i.e., DNA methylation). DNA methylation (DNAm) involves the silencing or activation of genes and is thought to be predominantly (although not exclusively) a product of the environment. Numerous studies have found evidence that DNAm resulting from environmental stressors predicts alterations in the stress-response system (Smith et al., 2017), poor physical health (Notterman & Mitchell, 2015), and depression (Sun et al., 2013a). As such, it is plausible that DNAm serves as a biological mediator of the relationship between environmental protective factors and resilience. Indeed, studies using animal models have provided strong support for this possibility (Szyf et al., 2005; Zhang et al., 2010). For instance, in Weaver and colleagues’ (Weaver et al., 2004) seminal paper, maternal care (i.e., licking and grooming and arched back nursing) was found to directly alter the DNA methylome and stress-response system of rat pups in the first week of life, with behavioral and biological effects persisting into adulthood (although the methylomic alterations were also found to be reversible via a histone deacetylase inhibitor trichostatin A). In addition, DNAm of the Crf promoter (Elliott et al., 2010) and epigenetic alterations to the blood-brain barrier (such that its integrity was maintained in the face of stress; Dudek et al., 2020) have been found to distinguish behaviorally resilient and non-resilient mice following exposure to social stress. While these animal studies inform our understanding of the role of DNAm in resilience, translation of these findings to human resilience is limited largely because human studies of DNAm cannot determine direct causality. Instead, human studies of DNAm must rely on biomarkers of resilience. Until recently, only three published studies had examined methylomic 6 biomarkers of resilience in a human sample (Milaniak et al., 2017; Miller et al., 2020), two of which examined DNAm in only one or two specific gene regions. Milaniak and colleagues (2017) found that DNAm in the oxytocin receptor gene at birth predicted psychological resilience (i.e., a lack of conduct problems) to prenatal environmental stressors. Similarly, Miller and colleagues (2020) found that DNAm of sites located on the NR3C1 and FKBP5 genes predicted psychological resilience. Although these studies begin to provide proof of concept for the idea that DNAm may at least partially underlie resilience to adversity, they were notably limited by their focus on a few specific gene regions despite the availability of methylome-wide arrays that interrogate methylation at roughly 850,000 CpG sites. Lu and colleagues performed the first methylome-wide association study (MWAS) of resilience (Lu et al., 2023; Vazquez, Burt, et al., 2024). Unfortunately, this study had a number of critical limitations, most notably the failure to correct for multiple testing despite performing analyses on approximately 850,000 CpG sites. They also did not include blood cell type proportions as a covariate in their analyses, an approach that leads to inflated test statistics (McGregor et al., 2016). What’s more, none of these studies (Lu et al., 2023; Milaniak et al., 2017; Miller et al., 2020) accounted for possible genetic confounds. Namely, while DNAm is heavily influenced by the environment, it is also known to be influenced by genetics. Thus, what may appear to be environmentally induced DNAm may instead reflect genetic confounds, thereby undercutting conclusions. The gold standard for overcoming this uncertainty in human studies lies in the use of a monozygotic (MZ) twin difference design (S. A. Burt et al., 2006). MZ twins are genetically identical and yet can and do have different methylomes as a function of their unique environmental experiences (Fraga et al., 2005). Unfortunately, most twin studies include relatively few youths exposed to adversity (and even fewer who demonstrate resilience to that 7 adversity). As such, little is known regarding the role DNAm may play in the origins of human resilience. My Master’s thesis (Vazquez, Burt, et al., 2024) began the process of filling this key gap in the literature. While my analyses were notably underpowered, results revealed a few DNA methylome-wide significant differentially methylated probes (DMPs) for social and academic resilience and several suggestive DMPs for all domains as well as general resilience. Pathway analyses suggested that DNAm in pathways related to DNA repair and transcription, and initiation of RNA Polymerase III were implicated in academic resilience, while those related to T-cell receptor signaling were implicated in social resilience. These analyses also highlighted the role of the BRF1 gene and the HLA region in academic and social resilience, respectively. However, only a handful of methylome-wide significant/suggestive DMPs differed significantly across discordant MZ twin pairs, suggesting that only these DMPs were environmental in origin. Study 2 of this dissertation aimed to build upon this work by identifying DMPs that are longitudinally predictive of resilience. Overview and impact of the proposed study. The overarching goal of this dissertation was to elucidate the development and underpinnings of multi-domain resilience to neighborhood disadvantage from middle childhood to late adolescence. Study 1 leveraged three waves of data collected from twin families exposed to neighborhood disadvantage to model developmental trajectories of social and psychological resilience, respectively, from middle childhood to emerging adulthood. We employed variable and person-centered methods to identify general patterns of domain-specific resilience development, as well as trajectory profiles that reflect the diversity of resilience development. We then investigated socioecological predictors of these trajectory profiles. For Study 2, we conducted a series of MWAS using DNAm measured in 8 middle childhood to identify differentially methylated probes (DMPs) associated with social and psychological resilience trajectory growth factors, with the ultimate goal of identifying longitudinally predictive epigenetic mechanisms underlying resilience development. We then evaluated the biological underpinnings of those DMPs that predicted resilience growth factors via pathway analyses. As a final step, we assessed the origins of significant and/or suggestive DMPs by performing twin difference analyses among monozygotic twin pairs. Collectively, the present studies meaningfully extended resilience science by illuminating the developmental trajectories of two resilience domains as well as the longitudinally predictive power of socioecological promotive factors and DNAm. 9 STUDY 1: DEVELOPMENTAL TRAJECTORIES OF MULTIDIMENSIONAL YOUTH RESILIENCE Abstract Youth resilience is widely considered a dynamic construct, meaning that youth may experience fluctuations in their ability to adapt to adversity in response to developmental as well as socioecological influences. Importantly, resilience is also multidimensional, in that the degree to which youth can successfully adapt may vary across domains of resilience. To date, however, we know relatively little about the developmental patterns of different forms of resilience over time, particularly during critical developmental transitions. The present study leveraged longitudinal data from the Michigan State University Twin Registry (N = 798) to investigate trajectories of youth social and psychological resilience to neighborhood disadvantage, respectively, from middle childhood to emerging adulthood. Growth curve models demonstrated that, on average, youth were characterized by moderate to high baseline levels of social and psychological resilience, respectively, with linear increases in each domain over time. However, growth mixture models revealed variability in trajectories of resilience such that social resilience was characterized by two trajectories (moderate-increasing and moderate-stable), while psychological resilience was characterized by three (moderate-increasing, high-increasing, and high-decreasing). Demographic characteristics (i.e., race/ethnicity or sex), parental nurturance, and neighborhood social processes significantly predicted trajectory class membership for social and/or psychological resilience. Our findings highlight the importance of employing both multidimensional and person-centered approaches to examining youth resilience. 10 Neighborhood adversity is a form of chronic adversity that poses a threat to healthy youth development (Alvarado, 2016; Campbell et al., 2000; Duncan et al., 2006; Raposa et al., 2014; Wodtke et al., 2011). Despite this risk, however, many youth (40-60%; Vanderbilt-Adriance & Shaw, 2008) are able to successfully adapt and exhibit positive mental health and competency in the face of adversity, or resilience (Luthar, 2000; Masten, 2001). Although historically regarded as a static trait, resilience is now widely considered a dynamic construct (Miller-Graff, 2022; Vazquez et al., 2021). In other words, resilience can fluctuate in response to developmental and environmental influences over time. Despite the importance of dynamism in modern conceptualizations of resilience, however, we still know little about the patterns of resilience across development, and how these patterns are influenced by promotive factors. Such research would be quite important as it can be leveraged to inform the timing and targets of prevention and intervention efforts that support healthy development among youth in disadvantaged neighborhood contexts. Background Only a handful of studies have examined the development of resilience over time. Among these, most conceptualized resilience as unidimensional, despite robust evidence that resilience is a multidimensional construct (Miller-Graff, 2022; Vazquez et al., 2021). Indeed, the various domains of resilience (e.g., social and psychological) demonstrate distinct prevalence rates (Burt et al., 2021; Dubowitz et al., 2016; Yoon, Maguire-Jack, et al., 2021) and etiologies (Vazquez et al., 2021). Other work has suggested that these differences across dimensions may extend to the stability of resilience over time as well (S. A. Burt et al., 2021; Dubowitz et al., 2016; Masten et al., 1999). Critically, however, conclusions regarding which domain is most stable were inconsistent, likely due to differences in participant ages, assessment windows, and/or resilience 11 domain conceptualizations. Generally, however, studies examining change across two waves of data found that all resilience domains and general resilience were relatively stable during adolescence (S. A. Burt et al., 2021; Dubowitz et al., 2016; Masten et al., 1999; Yoon, Maguire- Jack, et al., 2021). Studies employing >3 waves of data and advanced longitudinal methods can provide a more nuanced picture of the developmental trajectories of resilience over time. Unfortunately, such studies are particularly scarce. When examining general resilience (i.e., as a composite of social, cognitive, emotional, and behavioral domains of functioning) to maltreatment, Yoon et al. found that adolescent resilience exhibited significant linear increases across a 36-month period (Yoon, Sattler, et al., 2021), rather than the stable trajectory suggested by other work. When examining socioecological influences on this trajectory, they found that better caregiver-child relationships predicted higher baseline resilience and slower increases across adolescence. The inverse was seen for the effect of deviant peer affiliation and engagement in behavioral health services such that youth had lower resilience at baseline but evidenced more rapid increases. Yoon et al. relied on variable-centered methods, specifically growth curve modeling. Growth curve models summarize data across all participants, providing a mean initial level or baseline and a rate of change. The variances therefore represent individual differences in the intercept and slope, respectively. Although this provides important and meaningful information, it assumes that everyone in the sample follows the same general trajectory pattern and therefore inherently overlooks potential heterogeneity in developmental trajectories of resilience. Person-centered approaches such as growth mixture modeling, overcome this limitation. In growth mixture models, participants are grouped into unobserved trajectory classes that can differ in their intercept and/or 12 slope. This data-driven approach can illuminate individual differences in patterns of resilience over time and in doing so, improve our understanding of the various ways resilience may develop over time and the promotive factors that uniquely contribute to different resilience trajectory patterns. For instance, while variable-centered approaches may show that youth generally have stable trajectories of resilience over time, a person-centered approach may reveal a subset of youth that exhibit decreases or increases in resilience across development. To date, only one study has leveraged a person-centered approach to investigate individual differences in trajectories of resilience to maltreatment (Sattler et al., 2023). Three trajectories of resilience across early childhood were identified: 1) stable and low levels of resilience (31.1%), 2) increasing levels of resilience (17.4%), and 3) decreasing levels of resilience (51.5%). Notably, and contrary to Yoon and colleagues’ (2021) results for adolescent resilience development, the increasing trajectory was least common in early childhood. Moreover, item response probability estimates revealed significant variability across resilience domains, with varying class proportions for behavioral (i.e., assessed via behaviors such as eating, dressing, toileting, general safety, and household tasks), language (i.e., assessed via auditory comprehension and expressive communication), and cognitive (i.e., assessed via a standardized assessment of cognitive development) resilience. The most prominent trajectory thus appeared to vary across domains. Finally, Sattler et al., (2023) found that youth were more likely to have increasing trajectories if they had greater levels of caregiver cognitive stimulation. While extant work has made strides to elucidate the development of resilience, all prior studies examined resilience within a specific developmental stage. Indeed, we know of no published studies to date that have investigated resilience trajectories across developmental transitions (i.e., from childhood into adolescence or adolescence into early adulthood). This is 13 particularly surprising in light of extensive evidence that these transitions are critical periods for cognitive, social, and emotional (e.g., self-regulation) growth (Buttelmann & Karbach, 2017; Carr, 2011), as well as expanded agency and autonomy (e.g., greater choice in friends and activities). Studies examining resilience across developmental transitions are therefore important to improve our understanding of how youth resilience develops over time. Given the paucity of literature specific to resilience across developmental transitions, we must rely on studies examining social and psychological development more broadly to inform our hypotheses regarding the development of resilience. For instance, extant literature demonstrates generally linear patterns of social skills acquisition across adolescence (Blakemore, 2012; Carr, 2011) with marked individual differences in the timing of social skill acquisition (Anderson-Butcher et al., 2018; Oshri et al., 2017). In addition, the prevalence of psychopathology increases dramatically from childhood into adolescence (Costello et al., 2011). We might therefore expect common resilience trajectories to mirror those of social competence and psychopathology, whereas others may demonstrate trajectories unique to resilient youth. Indeed, given variability in the timing of developmental milestones, the severity of adverse exposures, and the presence of promotive factors, we would expect to see multiple trajectories of resilience for each domain. We can also seek to identify specific promotive factors that predict membership in each trajectory class, such as nurturing parenting or neighborhood social processes, both of which have been demonstrated to promote adaptive self-regulation in the face of adversity (Bernier et al., 2010; Sampson et al., 1997). Current Study The current study leveraged three waves of data in a sample of twin families exposed to moderate to severe neighborhood adversity to identify trajectories of resilience to neighborhood 14 disadvantage. In light of clear evidence of the multidimensionality of resilience, we examined social and psychological resilience separately to assess whether developmental trajectories of resilience and the salience of promotive factors varied across domains. We specifically employed variable and person-centered growth modeling to investigate the trajectories of resilience during critical transitional periods of youth development (i.e., from middle childhood to emerging adulthood). In doing so, we sought to identify the general pattern of resilience development across our sample, while also identifying youth characterized by unique developmental trajectories. We expected that we would observe multiple trajectory classes for both social and psychological resilience. Based on extant literature on social skills acquisition (Blakemore, 2012; Carr, 2011) and resilience development in adolescence (Yoon, Sattler, et al., 2021), we anticipated that the most prevalent social resilience trajectory would generally increase from middle childhood to late adolescence. In addition, psychopathology literature (Costello et al., 2011) suggests that we should expect a psychological resilience trajectory that demonstrates decreases from middle childhood to late adolescence. Meanwhile, literature on resilience trajectories across adolescence also suggests that we might see an increasing psychological resilience trajectory. Beyond this, however, we were unable to make well-grounded hypotheses regarding the number of trajectory classes or the characterization of other potential trajectory classes, although we did expect parental nurturance and neighborhood social processes to emerge as factors that promote increasing as well as high and stable trajectories of youth resilience over time. 15 Participants Methods We examined data from the Twin Study of Behavioral and Emotional Development in Children (TBED-C) and the Michigan Twin Neurogenetics Study (MTwiNS), studies embedded within the broader Michigan State University Twin Registry (MSUTR; Burt & Klump, 2013, 2019). The TBED-C (S. A. Burt et al., 2016, 2019) was recruited between 2008 and 2015 and included two arms: (1) a population-based arm (N=528) and (2) an ‘under-resourced’ arm (N=502). Both arms were identified via birth records and recruited in collaboration with the Michigan Department of Health and Human Services (MDHHS; Burt & Klump, 2013). The under-resourced arm, however, was restricted to families residing in neighborhoods with neighborhood poverty at or above the Census mean of 10.5% (the mean at study onset). Across the two arms, a majority (58.1%) of TBED-C families had household incomes that fell below the living wage in Michigan, 52.8% of the twins attended schools where more than 50% of students received free or reduced-price lunches, and 47.3% of families were exposed to severe or frequent community violence (S. A. Burt & Klump, 2019). Twins were in middle childhood, between ages 6 and 10 when they participated in the TBED-C (Wave 1), although a small handful had turned 11 by the time they participated. TBED- C families who met criteria for the under-resourced arm of the sample are currently being reassessed twice during adolescence as a part of the MTwiNS. The first follow-up assessment (wave 2) took place approximately 4-6 years following initial participation in the TBED-C, during which most twins were in early-to-mid adolescence, though they ranged in age from 7 to 20 (data are currently available for 354 families). The second follow-up assessment (wave 3) took place approximately 18 to 24 months after wave 2 when the majority of twins were in mid- 16 to-late adolescence, though they ranged in age from 10 to 22 (data are currently available for 188 families). Parents provided informed consent and children provided informed assent. Because resilience is conditional on exposure to adversity, the present study included only MTwiNS families (i.e., those from the under-resourced arm of the TBED-C). The final analytic sample therefore included 798 youth within 399 families, all of whom had at least two waves of data. The mean Area Deprivation Index decile (a Census-derived composite that includes 17 employment, housing quality, poverty, and education measures; Kind & Buckingham, 2018) for these participants was 5.17 (SD = 2.53), indicating that on average, participants were residing in neighborhoods with an ADI of 51.7%. Zygosity was established at wave 1 via caregiver report on a physical similarity questionnaire that has demonstrated over 95% accuracy (Peeters et al., 1998). Participants in the current study included 153 monozygotic twin pairs and 246 dizygotic twin pairs; 51% were female and 49% were male. The majority of participants identified as White (77.9%), 12.3% identified as Black, 1.8% as Native American, 1% as Latinx, and 7% as ‘Other’ or a racial group prominent in less than 1% of the sample (i.e., Asian or Pacific Islander). Demographics are displayed in Table 1. Measures Social Resilience. We measured social resilience via the social competence subscale of the Child Behavior Checklist (CBCL; Achenbach & Rescorla, 2001) at all three waves (wave 1 α = 0.569, wave 2 α = 0.629, wave 3 α = 0.596). Mothers reported on their child’s involvement in organizations, number of friends, contact with friends, behavior with others, and behavior alone (e.g., About how many times a week does your child do things with any friends outside of 17 regular school hours?) during the previous 6 months. This subscale included 10 items assessed using various scalings (a 3-point Likert scale, 4-point Likert scale, and dichotomous 2-point scale were used). These were then weighted by the ASEBA scoring program to compute a total sum score ranging from 0 to 14. Higher scores were indicative of better functioning or ‘more’ social resilience. Table 1. Sample Demographic Characteristics – Study 1 Demographic Variable N % Sex Male Female Race/Ethnicity White Hispanic/Latinx Black Native American/Native Alaskan Asian/Pacific Islander Pacific Islander Other Total 407 391 622 8 98 14 10 46 798 51% 49% 77.9% 0.5% 12.3% 1.8% 1.3% 5.8% 100% Note. Abbreviations: N = sample size, SD = standard deviation. The mean age at wave 1 = 7.9 (SD = 1.4), at wave 2 = 14.5 (SD = 2.2), and at wave 3 = 16.1 (SD = 2.3). Psychological Resilience. Psychological resilience was likewise measured using the CBCL (Achenbach & Rescorla, 2001) at all three waves, specifically via the total problems subscale (wave 1 α = 0.932. wave 2 α = 0.933, wave 3 α = 0.933). Mothers rated their children’s behavior on a three-point Likert scale ranging from (0) not true to (2) very true/often true across 103 items assessing emotional (e.g., There is very little he/she enjoys) and behavioral problems (e.g., Destroys things belonging to his/her family or others). The total problems subscale was 18 reverse-scored for analyses so that higher values reflect greater psychological resilience (the highest score across all three waves was used to reverse score items). Neighborhood Social Processes. Mothers also completed the Neighborhood Matters Questionnaire (Henry et al., 2014), which includes a Social Cohesion scale as well as an Informal Social Control scale. The Social Cohesion scale (wave 1 α=0.922, wave 2 α=0.911) includes 13 items on a 5-point Likert scale from ‘strongly agree’ (1) to ‘strongly disagree’ (5) assessing perceptions of support and help among neighbors in their community. The Informal Social Control scale (wave 1 α=0.864, wave 2 α=0.959) includes 13 items on a 4-point Likert scale from ‘do nothing’ (1) to ‘do something directly’ (4) assessing perceptions of whether neighbors will maintain social order. To assess the collective effects of neighborhood social processes during middle childhood and adolescence, we computed mean composites of neighborhood social cohesion and informal social control using subscale scores from waves 1 and 2. Parental Nurturance. We measured parental nurturance via mother-report on the parental involvement subscale of the Parental Environment Questionnaire (PEQ; Elkins et al., 1997). Though this subscale’s name implies a focus on involvement, its items specifically assess parental investments in communication, closeness, and support in the parent-child relationship (e.g. “I praise my child when he/she does something well”; “My child talks about his/her concerns and experiences with me”). The parental involvement subscale (α=0.72) includes 12 items on a 4-point Likert scale from definitely true (1) to definitely false (4). To evaluate the overarching influence of parental nurturance across middle childhood and adolescence, we created a mean composite using scores from waves 1 and 2. 19 Puberty. Puberty was assessed via the Pubertal Development Scale (Carskadon & Acebo, 1993), completed by mothers. This scale includes three items for boys and girls regarding growth spurts, body hair, and skin changes. In addition, two items assess breast development and menstruation for girls, while two items assess voice changes and facial hair for boys. All items are scored from 1 (not yet started) to 4 (complete). The mean of five items for boys (wave 2 α = 0.835, wave 3 α = 0.854) and girls (wave 2 α = 0.835, wave 3 α = 0.713) is computed and examined as a continuous indicator of pubertal development. Analytic Strategy Variable and person-centered longitudinal models were fit using Mplus version 8.6 (Muthén & Muthén, 1998-2019). Full Information Maximum Likelihood (FIML) was used to handle missing data as this approach has been demonstrated to handle larger amounts of missing data adequately and produce estimates with greater efficiency and less bias than other missing data techniques (Lang & Little, 2018). The Mplus ‘cluster’ command was used to account for the nesting of twins within families; specifically, this command created separate sets of regression coefficients for each family within the analysis by incorporating family ID as a categorical predictor within the model. Separate models were fit for psychological and social resilience, respectively, to allow for comparison of the trajectories of distinct domains of resilience. Given the composition of our sample, race/ethnicity was dichotomized into 0 (minoritized) and 1 (White) to ease the interpretation of its confounding effect. Variable-Centered Analyses. Latent Growth Curve Models (LGCM) were used to assess change in psychological and social resilience over time. This approach is grounded in Structural Equation Modeling (SEM; Kline, 2011), estimating the intercept (initial value) and slope (rate of change) of an outcome as latent variables to model their trajectories over time. We used time 20 scores to allow for individually-varying times of observations at each wave due to the wide and overlapping age ranges of participants across waves. We imputed age for a small number of participants missing age data at wave 2 or wave 3 so that our time scores were complete for the entire sample. We did so by adding the mean number of years between waves to the age of participants at wave 1 (available for all participants). Intercepts were centered at age six, the youngest age in our sample. We first fit unconditional growth curve models to assess change in social and psychological resilience across development, respectively, allowing for interindividual variation in the rate of annual change over time. Intercept-only models (including three parameters: intercept mean, intercept variance, and residual variance) and linear growth models (including six parameters: (intercept and slope means, intercept and slope variances and their covariance, and residual variance) were compared for each respective outcome. Better model fit was indicated by lower values on the following indices: Akaike information criterion (AIC; (Akaike, 1987), Bayesian information criterion (BIC; Raftery, 1995), and sample-size adjusted Bayesian information criterion (saBIC; Sclove, 1987). The best-fitting model was then used for our conditional growth curve model analyses in which sex and race/ethnicity were examined as time- invariant predictors and puberty at wave 2 and wave 3 was examined as a time-varying predictor. Puberty was an important covariate to include due to its association with trauma exposure and youth outcomes, particularly mental health. Literature has demonstrated that early life trauma is associated with the early onset puberty, which has in turn been linked to greater risk for the development of psychopathology (Colich et al., 2022; Marshall, 2016). Person-Centered Analyses. We employed Growth Mixture Models (GMM) to group individuals based on their patterns of social and psychological resilience, respectively, from 21 middle childhood to emerging adulthood. GMM is a semi-parametric statistical technique designed for classifying longitudinal data (Andruff et al., 2009; Collins & Lanza, 2009; Duncan et al., 2006) and provides an empirically based, data-driven method for grouping the development of resilience into profiles or trajectory classes. GMMs use a categorical latent variable to capture different classes of trajectories based on growth factors (i.e., intercept and slope). We first fit a one-class model and progressively increased the number of classes until we encountered very small class sizes (<2%) or encountered significant modeling problems (e.g., convergence issues). These models allowed trajectory classes to have differing intercepts and slopes but constrained the intercept and slope variances to be the same across classes. Class enumeration was ultimately determined using statistical and substantive criteria (Nylund et al., 2007). As before, lower values on the AIC, BIC, and saBIC indicated superior model fit (we were unable to obtain and use the BLRT or Lo-Mendal Rubin fit statistics as they are only available with TYPE=MIXTURE in Mplus). Substantive criteria in class enumeration included interpretability and uniqueness of the classes, as well as the size of the trajectory classes, with very small classes (<5%) typically avoided as they can indicate local solutions that are difficult to replicate. Small classes were retained if they had a very strong theoretical meaning. Trajectory class names correspond to a description of the level of the intercept followed by the direction of the slope. Next, we examined predictors of class membership, specifically ethnicity/race, sex, and three promotive factors (all time-invariant). Our promotive factors included neighborhood social cohesion, neighborhood informal social control, and parental nurturance. Of note, we did not include puberty as a covariate in these analyses because its time-varying nature led to difficulties maintaining the class structures and proportions. Although we initially intended to allow 22 predictor effects to vary by class and predict growth factors, we were unable to do so in a way that preserved the original class membership meanings and proportions when using the one-step method. We therefore used the 3-step method to ensure that the resilience trajectory class meanings were not influenced by the inclusion of predictors. Briefly, Step 1 involves identifying the best-fitting unconditional GMM. In Step 2, the estimated probability of being in a given trajectory class is computed for each individual in the best-fitting model. Finally, in Step 3, the inverse logit of the probabilities computed in Step 2 are used as weights in the multinomial logistic regression between predictors and resilience trajectory class membership. We used the Mplus ‘R3STEP’ auxiliary feature to automate this method. As a final step, we performed sensitivity analyses to test whether our results persist when restricting analyses to those with extreme adversity exposures. Participants were considered to be experiencing substantial ‘disadvantage’ if they met criteria for 2 of 3 indicators: family poverty, neighborhood poverty, and exposure to community violence (as used in Burt, Slawinski, & Klump, 2018). Family poverty was measured using maternal reports of total family income; those families with a combined income of $55K or less (below the living wage in the state at that time of assessment) met criteria for this indicator. The mean family income in these data was $40,000-$45,000 (versus $72,027 in the population-based arm of the sample). Neighborhood poverty was assessed using Census data; participants residing in Census tracts where 20% or more of households were below the 2008 poverty line met criteria for this indicator. Mean neighborhood poverty level in these data was 21.7%, compared to 11.4% in the population-based arm of the sample. Finally, exposure to community violence was assessed via maternal reports on the indirect violence scale of the KID-SAVE (Flowers, Hastings, & Kelley, 2000); participants who endorsed 30% or more of the items on this scale met criteria for this indicator (53% in the 23 Table 2. Descriptive Statistics and Correlations 1. Psy Resilience Wave 1 1. - 2. 3. 4. 5. 6. 7. 8. 9. 0.479** 0.37** 0.226** 0.155** 0.116* 0.158** 0.155** 0.108** 2. Psy Resilience Wave 2 0.479** - 0.659** 0.177** 0.273** 0.266** 0.27** 0.187** 0.131** 3. Psy Resilience Wave 3 0.370** 0.659** - 0.093* 0.145* 0.326** 0.245** 0.137** 0.218** 4. Soc Resilience Wave 1 0.226** 0.177** 0.093* - 0.424** 0.257** 0.143** 0.202** 0.106** 5. Soc Resilience Wave 2 0.155** 0.273** 0.145* 0.424** - 0.555** 0.274** 0.190** 0.150** 6. Soc Resilience Wave 3 0.116* 0.266** 0.326** 0.257** 0.555** - 0.186** 0.145** 0.201** 7. Parental Nurturance 0.158** 0.270** 0.245** 0.143** 0.274** 0.186** - 0.179** 0.235** 8. Neigh. Soc. Cohesion 0.155** 0.187** 0.137** 0.202** 0.190** 0.145** 0.179** - 0.459** 9. Neigh. Inf. Soc. Control 0.108** 0.131** 0.218** 0.106** 0.150** 0.201** 0.235** 0.459** - 42.533 49.843 11.358 2.524 32-48 0-48 8.263 15-65 0-65 792 2.040 0.5-13 0-52 768 Mean SD Actual Range Possible Range 108.033 114.303 116.882 17.248 15.052 14.102 7.378 2.394 18-130 0-130 9-130 0-130 53-130 0.5-13.5 0-130 0-14 8.488 2.650 0-14 0-14 8.437 2.706 2-14 0-14 N Note. **p<.001, *p<.05. Abbreviations: Psy = psychological, Soc = social, Neigh = neighborhood, Inf = informal. 797 793 474 789 551 429 798 24 current sample and 7% in the population-based arm of the sample). 394 twins within 197 families met these criteria and were therefore included in the sensitivity analyses. Descriptives Results Descriptive statistics and correlations for social resilience, psychological resilience, and promotive factors are shown in Table 2. Youth had relatively high mean levels of psychological resilience and moderate levels of social resilience (relative to the max score on each scale). Mothers reported high mean levels of parental nurturance and neighborhood social cohesion, and relatively low mean levels of neighborhood informal social control. Finally, all variables were significantly correlated with one another. Variable-Centered Analyses We first fit unconditional linear LGCMs for social and psychological resilience, respectively, to assess change over time. Relative to the intercept-only social and psychological resilience models, the linear growth models fit the data better for both domains of resilience (see Table 3). We thus used the linear models as our final unconditional models. Social Resilience. The mean level of social resilience at age six was 7.131 (SE = 0.127, p<.001). On average, social resilience demonstrated linear increases by a factor of 0.141 (SE = 0.016, p<.001) annually. Youth did not, however, demonstrate between-person variation in either their initial level (b = 1.457, SE = 0.881, p = 0.098) of social resilience or their rate of change (b = 0.018, SE = 0.011, p = 0.109). The estimated intercept-slope covariance was likewise non- significant (b = 0.044, SE = 0.091, 0.633), indicating that there was no relationship between youths’ baseline social resilience and their rate of change. 25 Next, we fit conditional linear growth curve models in which we included three promotive factors (parental nurturance, neighborhood informal social control, and neighborhood social cohesion), two time-invariant covariates (sex and race/ethnicity) and one time-varying covariate (puberty); see Table 3. Unlike the unconditional model, neither the mean intercept nor the mean slope were significantly different from zero, indicating stable trajectories of social resilience across development. The intercept and slope variances as well as the intercept-slope covariance were all once again non-significant. Neighborhood social cohesion and parental nurturance were the only significant predictors of social resilience trajectories. Specifically, youth in more cohesive neighborhoods had higher initial levels of social resilience (b = 0.05, SE = 0.02, p = 0.002), while youth with more nurturing parents had steeper increases in social resilience over time (b = 0.01, SE = 0.01, p = 0.044). Sex, race/ethnicity, and neighborhood informal social control did not significantly predict social resilience trajectories. Finally, puberty was not predictive of concurrent resilience at either wave 1 or 2. Of note, when our covariates were included in the model without the promotive factors, the mean intercept and slope both remained significant (see Table S1). Psychological Resilience. At age six, youth had a mean psychological resilience score of 106.886 (SE = 0.917, p<.001). The mean slope was 0.885 (SE = 0.095, p<.001), indicating significant linear increases across development annually. Youth demonstrated significant between-person variation in their initial level (b = 207.773, SE = 58.576, p<.001) of psychological resilience and in their rate of change (b = 1.474, SE = 0.605, p = 0.015). The estimated intercept-slope covariance was non-significant but trending (b = -10.335, SE = 5.805, 0.075), indicating that youth with higher baseline levels of psychological resilience were 26 Table 3. Growth Curve Models Outcome Parameter Estimate S.E. p-value Estimate S.E. p-value Unconditional Model Conditional Model Social Intercept Resilience Mean Variance Ethnicity Sex Parental Nurturance Neigh. Soc. Cohesion Neigh. Inf. Soc. Control Slope Mean Variance Ethnicity Sex Parental Nurturance Neigh. Soc. Cohesion Neigh. Inf. Soc. Control Time Varying Covariates Puberty2 Puberty3 7.131 1.457 0.127 <.001 0.881 0.098 - - - - - - - - - - - - - - - 0.141 0.018 0.016 <.001 0.011 0.109 - - - - - - - - - - - - - - - - - - - - - 27 0.758 1.125 0.462 0.156 0.093 0.054 -0.065 -0.488 0.018 -0.003 -0.049 0.014 -0.002 0.017 -0.041 -0.160 1.974 0.851 0.310 0.234 0.051 0.017 0.074 0.298 0.011 0.046 0.032 0.007 0.002 0.009 0.691 0.186 0.137 0.506 0.069 0.002 0.383 0.101 0.092 0.942 0.132 0.044 0.325 0.065 0.114 0.128 0.719 0.213 Table 3 (cont’d) Social Resilience Intercept-Slope Covariance 0.044 0.091 0.633 0.035 0.090 0.697 Model fit Statistics AIC BIC saBIC 8139.264 8176.721 8151.316 10338.592 10454.622 10375.235 Psychological Intercept Resilience Mean Variance Ethnicity Sex Parental Nurturance Neigh. Soc. Cohesion Neigh. Inf. Soc. Control Slope Mean Variance Ethnicity Sex Parental Nurturance Neigh. Soc. Cohesion Neigh. Inf. Soc. Control 106.886 0.917 <.001 56.416 16.325 0.001 207.773 58.576 <.001 212.787 57.416 <.001 - - - - - - - - - - - - - - - 0.885 1.474 0.095 <.001 0.605 0.015 - - - - - - - - - - - - - - - 28 2.688 4.930 0.825 0.290 -0.257 -1.607 1.565 -0.567 -0.913 -0.052 -0.004 0.090 2.317 1.612 0.385 0.136 0.514 1.679 0.590 0.223 0.168 0.040 0.013 0.054 0.246 0.002 0.032 0.033 0.617 0.339 0.008 0.011 <.001 0.196 0.753 0.095 Table 3 (cont’d) Psychological Time Varying Covariates Resilience Puberty2 Puberty3 - - - - - - 0.690 1.137 0.468 0.529 0.141 0.032 Intercept-Slope Covariance -10.335 5.805 0.075 -12.189 5.733 0.034 Model fit Statistics AIC BIC saBIC 16761.035 16798.492 16733.088 18521.027 18637.057 18557.671 Note. Results are provided for the unconditional and conditional growth curve models for each resilience domain, respectively. Abbreviations: SE = Standard Error, AIC = Akaike information criterion, BIC = Bayesian information criterion, saBIC = sample-size adjusted Bayesian information criterion. Gender= 1(female), Ethnicity = 1(White). 29 somewhat more likely to have lower slope values (and vice-versa), indicating either decreases over time or slower increases over time. We then fit a conditional linear growth curve model in which our three promotive factors and three covariates were included as described above (see Table 3). When compared to the unconditional model, the mean intercept remained significant (b = 56.42, SE = 16.33, p = 0.001), however the mean slope became non-significant (b = -1.61, SE = 1.68, p = 0.339; the slope did however remain significant when only covariates, excluding promotive factors, were added to the conditional model, see Table S1), indicating stability in psychological resilience over time. Unlike the unconditional model, the intercept and slope variances were significant, suggesting between-person variation in the psychological resilience intercept (b = 212.79, SE = 57.42, p<.001) and slope (b = 1.57, SE = 0.59, p = 0.008). The estimated intercept-slope covariance was also significant (b = -12.19, SE = 5.73, p = 0.03), suggesting that youth with higher baselines levels of psychological resilience were more likely to have smaller slope values and vice-versa. Sex was a significant predictor of the psychological resilience intercept (b = 4.93, SE = 1.61, p = 0.002) and slope (b = -0.91, SE = 0.17, p<.001) such that females evidenced significantly higher initial levels and smaller slope values. Race/ethnicity also significantly predicted the psychological resilience slope (b = -0.57, SE = 0.22, p = 0.011), such that White youth had smaller slope values compared to minoritized youth. In addition, youth with higher levels of parental nurturance (b = 0.83, SE = 0.39, p=0.03) and neighborhood social cohesion (b = 0.29, SE = 0.14, p = 0.033) were found to have higher baseline levels of psychological resilience, respectively. Puberty scores at wave 3 also significantly predicted psychological resilience at wave 3 (b = 1.14, SE = 0.53, p = 0.032; the effect of puberty at wave 2 on concurrent 30 psychological resilience was non-significant). Neighborhood informal social control did not significantly predict psychological resilience trajectories. Sensitivity analyses confirm that the unconditional LGCM results for both social and psychological resilience were consistent among youth exposed to more severe neighborhood disadvantage (See Table S2). There were, however, some differences in the effects of demographics and promotive factors such that neighborhood social cohesion and race/ethnicity no longer predicted psychological resilience trajectories, and all three promotive factors predicted social resilience trajectories (specifically the intercept, while only informal social control predicted the slope). Person-Centered Analyses We fit unconditional GMMs for social and psychological resilience, respectively, to identify unique classes of resilience trajectories from middle childhood to emerging adulthood. We compared the respective fit of models from a one-class through to a five-class solution for each resilience domain. Social Resilience. Based on information criteria, model fit was best for the two-class solution (see Table 4). This conclusion was further supported as the three-class solution had two classes that were not meaningfully distinct (one of which was very small); specifically, this solution was comprised of two stable trajectory classes, one of which comprised 4.1% of the sample and started somewhat higher than the other trajectory which included 39% of the sample. In addition, the four and five-class solutions introduced issues with singularity that we were unable to resolve. The two-class solution was thus selected based on statistical and substantive criteria. As demonstrated in Figure 1, the two-class model included a moderate-increasing (i.e., 31 Table 4. Growth Mixture Model Fit Statistics # of groups # free parameters AIC BIC saBIC Log-likelihood (df) Entropy 1 2 3 4 5 1 2 3 4 5 Social Resilience 8139.264 8176.721 8120.656 8172.159 8125.088 8190.638 - - - - 8151.316 8137.228 8146.180 - - Psychological Resilience 16761.035 16798.492 16773.088 16543.971 16595.474 16560.543 16416.918 16482.467 16438.010 16358.494 16438.090 16384.106 16232.852 16417.494 16353.983 8 11 14 - - 8 11 14 17 20 -4061.632 () -4049.328 () -4048.544 () - - -8372.518 () -8260.985 () -8194.459 () -8162.247 () -8141.926 () - 0.491 0.580 - - - 0.914 0.898 0.870 0.864 Smallest group (%N) 100% 42.4% 4.1% - - 100% 8% 7.1% 2.3% 0.5% Note. The four and five-class solutions for the social resilience model were not identified and thus introduced issues with singularity. Abbreviations: SE = Standard Error, AIC = Akaike information criterion, BIC = Bayesian information criterion, saBIC = sample-size adjusted Bayesian information criterion, df = degrees of freedom, N = sample size. 32 Table 5. Social Resilience Growth Mixture Models Parameter Intercept-Slope Covariance Estimate S.E. p-value -0.075 0.094 0.421 1.303 0.947 0.169 0.009 0.010 0.391 Class 1 of 2: Moderate, Increasing (57.6%) 7.739 0.235 <.001 0.240 0.031 <.001 Class 2 of 2: Moderate, Stable (42.4%) 6.326 0.301 <.001 0.004 0.034 0.903 Increasing vs. Stable (reference) Variance I Variance S Mean I Mean S Mean I Mean S Ethnicity Sex 0.730 0.353 0.038 0.006 0.278 0.984 0.223 0.067 0.001 0.027 0.019 0.157 Parental Nurturance Neighborhood Social Cohesion Neighborhood Informal Social Control 0.012 0.096 0.898 Intercept -11.153 2.732 <.001 Note. I = Intercept, S = Slope. Gender= 1(Female); Ethnicity = 1(White). The results presented here are in logits. Race/ethnicity was a significant predictor such that White youth were 2.1x (derived from odds ratios) more likely to be in the increasing group than the stable group compared to minoritized youth. In addition, youth receiving more nurturing parenting were 1.3x more likely to be in the increasing group than the stable group. Sex, neighborhood social cohesion, and neighborhood informal social control were not significant predictors of social resilience class membership. 33 e c n e i l i s e R l a i c o S 12 10 8 6 4 2 0 Social Resilience Class 1: moderate, increasing (57.6%) Class 2: moderate, stable (42.4%) 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 Age Figure 1. Social Resilience 2-Class Growth Mixture Model Profile Plot intercept-slope) trajectory representing most (57.6%) of the sample, as well as a moderate-stable trajectory (42.4%). Next, we examined predictors of class membership (see Table 5). Sensitivity analyses confirm that the social resilience trajectory profiles and approximate proportions persist among participants exposed to more extreme levels of neighborhood disadvantage (see Tables S3). Moreover, parental nurturance remained a significant predictor of class membership in this population and neighborhood social cohesion also emerged as a significant predictor; youth in more cohesive neighborhoods were more likely to have increasing levels of social resilience over time as opposed to stable levels. Psychological Resilience. Model fit continued to improve as the number of classes increased (see Table 3). However, the four-class and five-class models included one or more classes comprised of a small proportion of the full sample that was not meaningfully distinct from the other trajectories. Specifically, the four-class solution included two increasing classes, one of which was comprised of 2.2% of the sample and had a lower baseline than a class comprising 13% of the sample. Similarly, the five-class solution included an increasing class 34 comprised of <1% of the sample, with a similar slope but lower baseline than a class with 3.1% of the sample. We thus proceeded with the three-class solution. As shown in Figure 2, the three- class solution included a high-increasing trajectory representing most of the sample (87.7%), a high-decreasing trajectory (7%), and a moderate-increasing trajectory (7.1%). We then examined predictors of class membership (see Table 6). Psychological Resilience 140 120 100 80 60 40 20 0 e c n e i l i s e R l a c i g o l o h c y s P Class 1: high, increasing (85.8%) Class 2: high, decreasing (7.1%) Class 3: moderate, increasing (7.2%) 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 Age Figure 2. Psychological Resilience 3-Class Growth Mixture Model Profile Plot Sex predicted membership such that females were 5.9x and 11x (derived from odds ratios) more likely to be in the decreasing class than the high-increasing or moderate-increasing class, respectively, as compared to males. In addition, parental nurturance predicted class membership such that youth with higher levels of parental nurturance were 1.2x and 1.4x more likely to be in the high-increasing group than in either the moderate-increasing or high- decreasing group, respectively, and were 1.2x more likely to be in the moderate-increasing group than the high-decreasing group. Neighborhood social cohesion was also a significant predictor of class membership; youth in neighborhoods with greater social cohesion were 1.1x more likely to be in the high-increasing group than the moderate-increasing group. Finally, Neighborhood 35 informal social control likewise predicted class membership such that youth in neighborhoods with greater informal social control were 1.3x more likely to be in the moderate-increasing class than the high-decreasing class. Race/ethnicity was not predictive of psychological resilience class membership. Table 6. Psychological Resilience Growth Mixture Models Parameter Estimate S.E. p-value Intercept-Slope Covariance -3.703 3.171 92.242 35.120 0.307 0.307 0.243 0.009 0.318 Variance I Variance S Mean I Mean S Mean I Mean S Mean I Mean S Ethnicity Sex Class 1 of 3: High, Increasing (85.8%) 110.579 1.249 0.885 0.104 <.001 <.001 Class 2 of 3: High, Decreasing (7.1%) 108.570 3.708 -2.557 0.378 <.001 <.001 Class 3 of 3: Moderate, Increasing (7.2%) 57.250 8.795 4.598 0.687 <.001 <.001 High Decreasing vs. High Increasing (reference) Parental Nurturance Neighborhood Social Cohesion Neighborhood Informal Social Control Intercept 0.519 0.539 1.781 0.585 0.336 0.002 -0.327 0.087 <.001 -0.015 0.032 -0.135 0.116 0.649 0.246 11.788 3.366 <.001 36 Table 6 (cont’d) High Decreasing vs. Moderate Increasing (reference) Ethnicity Sex Parental Nurturance Neighborhood Social Cohesion Neighborhood Informal Social Control Intercept Ethnicity Sex Parental Nurturance Neighborhood Social Cohesion Neighborhood Informal Social Control Intercept 0.637 0.631 0.313 2.394 0.672 <.001 -0.185 0.101 0.047 0.039 -0.272 0.135 -5.246 3.030 0.118 0.483 0.612 0.417 0.141 0.068 0.062 0.028 -0.137 0.121 -5.246 3.030 0.066 0.229 0.045 0.083 0.807 0.142 0.039 0.027 0.256 0.083 High Increasing vs. Moderate Increasing (reference) Note. I = Intercept, S = Slope. Gender= 1(Female); Ethnicity = 1(White). The results presented here are in logits. Once again, sensitivity analyses demonstrated that these psychological resilience trajectory profiles and approximate proportions were robust, even among participants exposed to the most significant levels of neighborhood disadvantage (see Table S4). The pattern of results for predictors of trajectory classes was also relatively consistent, with parental nurturance, sex, and neighborhood informal social control all emerging as significant predictors of psychological resilience class membership. However, neighborhood social cohesion was no longer a significant predictor (though it was trending; p<.10). 37 Discussion We used variable and person-centered approaches to investigate the development of multidimensional resilience to neighborhood disadvantage from middle childhood to emerging adulthood. Unconditional LGCMs revealed that on average, youth had high initial levels of social and psychological resilience, respectively, and that both domains of resilience were characterized by significant linear growth over time. While this pattern of results persisted when accounting for the confounding effects of sex and ethnicity, neither slope was found to be significant when specific promotive factors were included in the model. This suggests that our promotive factors may account for the significant positive mean slope in youth resilience across development. When we grouped individuals based on their resilience trajectories using GMMs, we found evidence of two classes for social resilience and three classes for psychological resilience. The predominant (57.6%) social resilience class was characterized by high initial levels that increased over time, while the second class (42.4%) was characterized by high initial levels and stability over time. Notably, White youth (as compared to minorized youth) and youth receiving more nurturing parenting were more likely to demonstrate increased resilience over time as opposed to stable levels. For psychological resilience, the majority of the sample (85.8%) was also characterized by high initial levels that increased over time, while the remainder of the sample was characterized by either a moderate baseline with an increasing trajectory (7.2%) or a high baseline with a decreasing trajectory (7.1%). Consistent with our LGCM results, our GMMs demonstrated that females were more likely to have trajectories of psychological resilience with high initial levels that decrease over time. In addition, youth in more cohesive neighborhoods were more likely to have high-increasing than moderate-increasing trajectories and youth in 38 neighborhoods with more informal social control were more likely to have moderate-increasing than high-decreasing trajectories of psychological resilience. Finally, youth with more nurturing parents were more likely to have high initial levels of psychological resilience and demonstrate increases in their psychological resilience over time. Notably, sensitivity analyses confirmed that our pattern of GMM results was consistent among participants exposed to more extreme levels of adversity. In revealing multiple unique trajectory classes for each domain of resilience, these results highlight the importance of investigating individual differences in multiple domains of youth resilience across developmental transitions. Our results were consistent with Yoon and colleagues’ (Yoon, Sattler, et al., 2021) work which demonstrated that on average, resilience increased during the course of adolescence. Our findings were, however, fairly distinct from the class proportions evidenced by Sattler and colleagues (Sattler et al., 2023) in that they found a decreasing trajectory to be the most common trajectory while we found it to be the least common for psychological resilience and did not find evidence of this trajectory for social resilience. That said, our sample ages were quite distinct from those in Sattler et al. and we did nonetheless show consistent evidence of multiple resilience trajectories and variable patterns of resilience across domains. Indeed, our work demonstrating unique trajectory classes and class proportions across resilience domains is consistent with a multidimensional conceptualization of resilience (Miller- Graff, 2022; Vazquez et al., 2021). Moreover, our finding of a sizable social resilience class characterized by stability across development is consistent with literature suggesting that resilience can be stable during adolescence (S. A. Burt et al., 2021; Dubowitz et al., 2016; Masten et al., 1999). Our results for psychological resilience, however, demonstrated that this pattern is specific to specific types of resilience and does not extend to psychological resilience. 39 This inconsistency may also reflect the narrow developmental period (i.e., adolescence) examined in prior studies (S. A. Burt et al., 2021; Dubowitz et al., 2016; Masten et al., 1999) as compared to the longer developmental window (i.e., middle childhood to emerging adulthood) that we investigated. Importantly, we also identified parental nurturance as a strong predictor of more favorable trajectories of social and psychological resilience over time. This finding aligns well with extant evidence that quality caregiver-child relationships and positive caregiver behaviors are associated with youth resilience (Sattler et al., 2023; Vazquez, Shewark, et al., 2024; Yoon, Sattler, et al., 2021). Neighborhood social processes were also identified as predictors of increasing trajectories of youth psychological resilience. This findings is consistent with literature demonstrating that neighborhood collective efficacy promotes positive youth outcomes (Sampson et al., 1997). In light of this, it is surprising to note that this effect did not extend to social resilience for our GMMs. Given that our LGCMs did evidence an association between neighborhood social cohesion and youth social resilience at baseline, our limited findings for neighborhood effects in the GMM may be a result of the more limited heterogeneity for this domain’s trajectories. Indeed, the baseline levels of social resilience were quite similar across trajectory classes, and we did not see any evidence of a decreasing trajectory class. We suspect that this relatively limited heterogeneity may be due to the narrow nature of our measure of social resilience in that the social competence CBCL subscale does not capture the nuance of social functioning in the context of developmental progression (e.g., the closeness or quality of these friendships). Our GMM results also suggest that youth demographics influence resilience trajectories such that White youth had more favorable trajectories of social resilience across development 40 than racially/ethnically minoritized youth. Interestingly, the effects of race/ethnicity on social resilience were masked when we examined the general pattern of social resilience across the sample, highlighting the important nuance provided by GMMs. Given evidence that minoritized youth, particularly those residing in disadvantaged contexts, are more likely to experience discrimination (Assari & Caldwell, 2018; Denise, 2012; Neblett Jr. et al., 2012; Prince et al., 2018), systemic inequities, or toxicant exposure (as a result of historical redlining; Taylor, 2014), it is possible that these results reflect the compounding and intersecting stressors and challenges faced by this population. That being said, this effect did not carry over to the psychological domain of resilience in our GMMs. Finally, we found that females were substantially more likely to evidence unfavorable trajectories (i.e., decreasing) of psychological resilience over time as compared to males. This is consistent with evidence that females exhibit higher levels of internalizing disorders in adolescence than males (Zahn-Waxler et al., 2008). Moreover, this increase may be reflective of the compounding or intersecting experiences females are more likely to endure than males, such as gender-based discrimination or inequities. Notably, however, this effect was not seen for social resilience trajectories. Limitations The present study examined resilience across key developmental transitions and identified multiple social and psychological resilience trajectories as well as promotive factors predictive of these trajectories. Nonetheless, there were clear limitations. Due to the limited racial/ethnic as well as geographic diversity of our sample, the generalizability of our findings are somewhat limited. In addition, our use of secondary data introduced some measurement limitations. For instance, we relied on mothers as our sole informants to indicate resilience and the presence of promotive factors, potentially introducing informant bias. Although we chose 41 mothers as informants to maintain consistency in measurement across waves, future work should endeavor to incorporate more informants as well as more objective measures of resilience (e.g., structured interviews or coded behavioral paradigms) consistently across data collection waves. In addition to concerns regarding informant bias, the use of mother-reports on the CBCL to assess resilience may be limited in that the degree to which mothers accurately report youth social functioning and mental health may diminish across the course of development. As youth gain more independence and agency in late adolescence and early adulthood, it is likely that their perception regarding their interpersonal relationships and mental wellbeing becomes increasingly discordant from that of their mothers. Indeed, both perspectives are likely to provide rich, though somewhat distinct information regarding youth resilience. In addition, our reliance on the CBCL may also have restricted our ability to assess developmentally specific aspects of youth resilience. For instance, what we consider to be positive social behaviors and indicators of positive social relationships vary based on the developmental stage of youth. It is therefore likely that our reliance on the same measure for youth from middle childhood to early adulthood interfered with our ability to robustly assess resilience within developmental context. Next, we were also limited in our use of a deficit measure of psychological resilience as we did not have strength-based measures at wave 1. Given the consensus that resilience should be indicated by both the absence of negative outcomes and the presence of positive outcomes, future work should collect strength-based measures at all waves. Finally, although three waves of data are the minimum required for many advanced longitudinal methods, future studies would benefit from additional data collection waves that would enable exploration of non-linear growth patterns of resilience over time. 42 Implications The present study is the first to identify longitudinal trajectories of multidimensional resilience across salient developmental periods. Growth curve models indicated that youth evidenced high levels of social and psychological resilience in middle childhood that continued to increase across development. Our incorporation of person-centered approaches, however, highlighted the diversity of resilience trajectories. Growth mixture models revealed that the general increasing trend characterized one of two trajectory classes for social resilience and one of three trajectory classes for psychological resilience. These results suggest that there are indeed multiple developmental patterns of resilience across development that vary both within and between domains. Interventions and policies aiming to promote the development of youth resilience should thus target youth in middle childhood and adolescence. We specifically identified parental nurturance as a key intervention target that supports progressively increasing trajectories of psychological and social resilience across youth development, as well as neighborhood social cohesion and informal social control as intervention targets that specifically promote the development of greater psychological resilience over time. Interventions and/or policy advocates aiming to promote resilience should evaluate whether widely accessible psychoeducation, parenting skills, and resources to caregivers that facilitate greater engagement in nurturing parenting also act to increase youth resilience. In addition, it would be worth illuminating the extent to which youth in disadvantaged neighborhoods would benefit from community improvements that foster greater cohesion and social control. 43 STUDY 2: EXPLORING DNA METHYLATION AS A PREDICTOR OF YOUTH RESILIENCE TRAJECTORIES Abstract Twin studies have highlighted the presence of both genetic and environmental contributions to resilience. To date, however, the specific mechanistic pathways underlying resilience remain unknown. One key possibility involves epigenetics, specifically DNA methylation (DNAm). The present study investigated this possibility by exploring DNAm biomarkers of youth resilience trajectories. We leveraged a longitudinal dataset (i.e., three waves: middle childhood, adolescence, and emerging adulthood) from the Michigan State University Twin Registry (MSUTR), restricting the sample to families residing in moderately-to-severely disadvantaged neighborhoods (N = 1722 individuals at wave 1). Psychological and social resilience were assessed using the Child Behavior Checklist. Growth curve models were first conducted to elucidate the longitudinal trajectories of psychological and social resilience, respectively. We then performed a series of methylome-wide association analyses (MWA), examining DNAm as a predictor of social and psychological resilience trajectories in a subset of the sample (N = 244 individuals, i.e., those who had DNAm data available). Results revealed only one significant DMP for the social resilience intercept. We did, however, have several suggestive DMPs for each outcome, ranging from four to twenty-seven. We performed pathway analyses to understand the biological underpinnings of our top differentially methylated probes (DMPs). Finally, to strengthen causal inferences and circumvent genetic confounds, we performed monozygotic (MZ) twin difference analyses. Overall, our results suggest that DNAm may play a negligible role in the development of youth resilience. 44 Neighborhood disadvantage includes characteristics such as high rates of poverty, high toxicant exposure, exposure to community violence, limited access to physical and built resources, and low social cohesion (Jutte et al., 2015; Wodtke et al., 2011). There is robust evidence that chronic exposure to this form of adversity places youth at a higher risk for negative long-term physical outcomes (Cubbin et al., 2006; Jutte et al., 2015) and mental health (Diez Roux & Mair, 2010; Jutte et al., 2015) and predicts lower life expectancy as well as fewer life chances (Evans et al., 2012; Haley et al., 2012; Jutte et al., 2015; Lavizzo-Mourey, 2012). Even so, a remarkably high proportion of youth (40-60%) evidence positive adjustment and competent functioning in the face of chronic exposure to adversity (Vanderbilt-Adriance & Shaw, 2008a), or resilience. Contemporary literature conceptualizes resilience as a dynamic construct that may fluctuate over time in response to socioecological, biological, and developmental influences. The biopsychosocial model of resilience (Feder et al., 2019) specifies that socioecological and biological factors have a bidirectional and transactional relationship such that they impact one another over time and collectively shape youth resilience. In recent decades, an epigenetic process known as DNA methylation (DNAm) has emerged as a promising mechanism that may undergird such transactions. DNAm can result in malleable changes in gene expression (i.e., silencing or activating genes) that are often environmentally induced. Emerging literature suggests that DNAm may be associated with both adversity exposure as well as with resilience to adversity. However, the cross-sectional nature of extant human studies has impeded our ability to determine the direction of causation in this relationship. It also remains unclear whether the relationship between DNAm and youth resilience persists over time. Longitudinal research is therefore needed to assess whether DNAm might predict the development of youth resilience over time. 45 Background Contemporary literature widely recognizes resilience as a multidimensional construct such that youth might demonstrate varying degrees of resilience across areas of functioning (e.g., social versus psychological domains; S. A. Burt et al., 2021; Miller-Graff, 2022). Indeed, the etiology of resilience has been found to vary across domains (Vazquez et al., 2021). Despite this, most extant studies have conceptualized resilience as unidimensional, often using a composite of ‘general resilience’. Similarly, despite widespread agreement that resilience is a dynamic construct, the vast majority of research on resilience is cross-sectional, thereby limiting our understanding of the typical pattern of youth resilience across development. Moreover, the handful of studies that have examined the stability of resilience over time have done so using two waves of data. These studies collectively suggest that resilience is fairly stable over time (S. A. Burt et al., 2021; Dubowitz et al., 2016; Yoon, Maguire-Jack, et al., 2021). Although these studies have been informative, their inclusion of only two waves of data prohibits the use of more advanced longitudinal methods that can further elucidate resilience trajectories and the effect of socioecological and biological influences on resilience development. We know of only two studies that have leveraged three or more waves of data and advanced longitudinal models to investigate resilient outcomes. Yoon and colleagues (Yoon, Sattler, et al., 2021) for instance, used latent growth curve models to investigate resilience to maltreatment using a general resilience composite (i.e., including emotional, social, behavioral, and cognitive areas of functioning) measured in adolescence. They found that over a span of 36 months, adolescents demonstrated significant linear increases in their levels of resilience. They further found that youth with better caregiver-child relationships were more likely to have higher 46 initial levels of resilience and to demonstrate lower increases in resilience over time (perhaps reflecting regression to the mean). By contrast, youth with greater levels of deviant peer affiliation and youth receiving behavioral health services, respectively, were more likely to have lower initial levels of resilience and more rapid increases in resilience over time. As another example, Sattler and colleagues (Sattler et al., 2023) employed repeated measures latent class analysis to identify multiple trajectories of resilience in early childhood. This study incorporated behavioral, language, and cognitive domains of resilience into a composite of resilience. The authors found evidence of three trajectory classes; the first was characterized by low and stable levels of resilience (31%), the second was characterized by increasing levels of resilience (17%), and the third was characterized by decreasing levels of resilience (52%). Item response probability estimates revealed that the proportion of young children in each class varied across resilience domains. Sattler et al. (2023) also investigated the impact of caregiver cognitive stimulation and found that youth with greater levels of stimulation were more likely to have increasing trajectories of resilience as opposed to decreasing or stable trajectories. Notably, this study’s finding that a decreasing trajectory was most common in early childhood stands in contrast to Yoon and colleagues’ (Yoon, Sattler, et al., 2021) finding that adolescents typically increase in their trajectories over time. That being said, this discrepancy could be attributed to their differing conceptualization of resilience as well as the markedly different developmental stage of participants. Because developmental work on resilience is limited, we may also leverage literature on related constructs to inform our expectations of resilience trajectories. For example, literature suggests that youth social skills acquisition typically exhibits increasing trajectories across adolescence (Blakemore, 2012; Carr, 2011). We might therefore expect to see similarly 47 increasing trajectories of social resilience across development for some youth. Longitudinal literature on psychopathology can likewise provide insight regarding the development of psychological resilience. In light of increasing prevalence rates of psychopathology from childhood into adolescence (Costello et al., 2011), we might expect some youth to exhibit the inverse (i.e., decreasing trajectories) for psychological resilience. While extant longitudinal studies begin to provide insight into the patterns of resilience development and its promotive factors, they are limited in three key ways. First, both resilience studies used composites of general resilience despite robust evidence of its multidimensionality. Second, these studies focused on the development of resilience within a particular developmental stage and did not include key developmental transitions (such as the transition from childhood to adolescence). Third, despite theoretical and empirical literature underscoring the role of genetic/biological influences in the etiology of resilience (Feder et al., 2019; Kim-Cohen et al., 2004; Vazquez et al., 2021), we know of no longitudinal study investigating the role of specific biological mechanisms in the development of resilience. One particularly promising mechanism that embodies the synergistic effects of socioecological and biological mechanisms is DNAm. DNAm resulting from exposure to stress has been found to predict negative physical and mental health outcomes (Notterman & Mitchell, 2015; Smith et al., 2017; Sun et al., 2013b). For example, Su and colleagues (Su et al., 2019) investigated a sample of 329 young adults and found evidence that differential DNAm partially mediated the relationship between stress (measured via the Adverse Childhood Experience questionnaire and the Perceived Stress Scale) and depression. Other scholars have theorized that exposure to promotive socioecological factors may likewise result in resilience promotive DNAm alterations. Indeed, a number of studies have 48 investigated components of this theory and yielded promising results. Early work in this area used animal models and found that DNAm in mice exposed to adversity differentiated those who exhibited behavioral (Elliott et al., 2010) and physiological (Szyf et al., 2005; Weaver et al., 2004) signs of resilience from those mice who did not. More recently, human studies have also provided empirical support for this theory. For instance, DNAm in the oxytocin receptor gene measured at birth was found to be associated with subsequent psychological resilience in middle childhood (Milaniak et al., 2017). As another example, psychological resilience in adulthood was linked to DNAm in the NR3C1 and FKBP5 genes. Such findings point to possible links between resilience and DNAm in particular genes. Other work has sought to evaluate this possibility via methylome-wide analyses. Lu and colleagues (Lu et al., 2023) performed a Methylome-Wide Association Study (MWAS) to assess DNAm across the entire methylome in a sample of 78 individuals and identified three differentially methylated probes (DMPs) associated with psychological resilience. Finally, our lab already examined the current sample cross-sectionally, conducting MWAS analyses of multidimensional resilience, evaluating the concurrent association between DNAm and youth resilience in middle childhood (Vazquez, Burt, et al., 2024). Our results revealed evidence of 90 suggestive DMPs associated with psychological resilience as well as six significant and 54 suggestive DMPs associated with social resilience. We then employed MZ twin difference analyses to evaluate the origins of these DMPs. Because MZ co-twins share 100% of their DNA, any divergence in their methylome is assumed to be environmentally induced (Fraga et al., 2005), thereby making the MZ twin difference design the gold standard for overcoming genetic confounds in DNAm analyses in humans. Our findings confirmed that four of the DMPs 49 associated with social resilience were environmentally mediated, and that the remainder appeared to instead be the result of genetic or developmental mediation. In sum, extant work provides promising evidence that DNAm may influence youth resilience. That being said, only four studies have investigated this association in living humans, and all but one were confounded by genetics, thereby limiting etiologic interpretations. What’s more, because the majority of these studies focused exclusively on the psychological domain of resilience, we still know little about the association between DNAm and other domains of resilience. In addition, half of these studies focused on specific gene regions, despite the availability of methylome-wide arrays that provide quantification at over 850,000 CpG sites (i.e., regions of the DNA sequence where cytosine is followed by guanine; methylation is possible in these regions). Finally, all prior work investigating DNAm and resilience in humans has been cross-sectional, thereby limiting our understanding of whether or how DNAm may impact developmental patterns of resilience over time. Current Study The present study sought to investigate patterns of resilience development over time and to identify potential DNAm biomarkers of that development. To do so, we employed a longitudinal and genetically informed sample of twin families exposed to moderate to severe levels of neighborhood disadvantage. This unique sample enabled us to model trajectories of the social and psychological domains of resilience using three waves of data collected from middle childhood to emerging adulthood. This allowed us to build on past work by elucidating patterns of resilience across key developmental periods. Consistent with prior literature on adolescent resilience (Yoon, Sattler, et al., 2021) and social skills acquisition (Blakemore, 2012; Carr, 2011), we hypothesized that youth would, on average, exhibit linear increases in social resilience over 50 time. For psychological resilience, however, our hypotheses are less clear in light of conflicting evidence from work on resilience (Yoon, Sattler, et al., 2021) and psychopathology (Costello et al., 2011). What’s more, while prior work has exclusively examined socioecological predictors of resilience trajectories, we sought to investigate the role of a key biological mechanism. Specifically, we performed a series of MWAS to examine DNAm in middle childhood as a predictor of social and psychological resilience trajectory growth factors, respectively. We expected to identify a number of significant and suggestive DMPs associated with each respective outcome. Next, we used enrichment analyses to inform our understanding of the biological underpinnings of top DMPs. Finally, we leveraged the genetically informed nature of our sample to perform monozygotic (MZ) twin difference analyses and assess whether top DMPs are in fact environmental in origin. Of note, the original plan for this dissertation was to examine the same sample evaluated in Study 1. Unfortunately, there were significant delays in the processing of the DNAm data by the lab (see methods section for details), and thus we elected to re-analyze the smaller sample already evaluated cross-sectionally as part of Vazquez et al. (2024). Methods Participants We examined a longitudinal sample from the Twin Study of Behavioral and Emotional Development in Children (TBED-C) and the Michigan Twin Neurogenetics Study (MTwiNS). These studies are embedded within the Michigan State University Twin Registry (MSUTR; Burt & Klump, 2019), which identified participants using birth records in collaboration with the 51 Michigan Department of Health and Human Services (MDHHS; Burt & Klump, 2013). The TBED-C (wave 1 in the current study) recruited families with twins in middle childhood (i.e., ages 6 to 10, although a small handful had turned 11 by the time they participated) between 2008 and 2015; the sample incorporates two arms: (1) a population-based arm (N=528) and (2) an ‘under-resourced’ (N=502). Recruitment for the under-resourced arm of the sample was restricted to families within neighborhoods experiencing above average levels of neighborhood poverty (i.e., > 10.5%, the census mean at study onset). This approach successfully enriched the TBED-C sample for families experiencing disadvantage; 58% of TBED-C families had household incomes below the Michigan living wage, approximately half of families were exposed to frequent or severe levels of community violence, and 53% of twins attended schools where approximately half of the students were eligible to receive free or reduced-price lunch. MTwiNS is currently reassessing TBED-C families that meet criteria for the under- resourced arm of the sample. MTwiNS includes two additional waves of data collection in which twins are reassessed in early-to-mid adolescence (i.e., wave 2; approximately 4-6 years after their participation in the TBED-C) and again in mid-to-late adolescence (wave 3; approximately 18-24 months after wave 2). For all waves of data collection, parents provided informed consent and youth provided informed assent. Given that the presence of resilience is conditional on exposure to adversity, the present study only included families residing in disadvantaged neighborhood contexts. This was determined using the Area Deprivation Index (ADI), a Census-derived composite that incorporates 17 measures collected between 2008 and 2012 that assess employment, poverty, housing quality, and education (Kind & Buckingham, 2018). The ADI provides a deprivation index score for each block group, which is then used to create percentiles that demonstrate a 52 block group’s level of deprivation compared to all block groups in the state of Michigan. Higher scores on the ADI are thus indicative of greater deprivation. The full analytic sample was restricted to families with an ADI of 20 or greater (wave 1 N=1722, wave 2 N=765, wave 3 N=444), consistent with literature suggesting that families above this threshold experience the negative effects of neighborhood deprivation (Galster, 2018). Participants in our full analytic sample had a mean ADI of 50 at wave 1. Twin zygosity was determined via caregiver report on a physical similarity questionnaire (with 95% accuracy; Peeters et al., 1998) completed at wave 1. Twin participants in the current study included 350 monozygotic (MZ) twin pairs, 521 dizygotic (DZ) twin pairs, and one singleton. The racial breakdown of our sample was 80.9% White, 10.6% Black, and 8.5% ‘Other’ or a racial group prominent in <1% of the sample; 52% of participants identified as female and 48% as male. DNAm data are currently only available for a subset of TBED-C families (these samples were assayed as part of a pilot study for a funding application); thus, the second portion of our analyses were performed with only a subset of the full analytic sample (N = 244 individuals). This subsample included 103 MZ pairs, 18 DZ pairs, and two singletons; 32% female and 68% male. The majority of this subsample identified as White (77.9%), 11.2% identified as Black, 2.4% as Native American, 1.6% as Pacific Islander, and 6.9% as ‘Other’ or a racial group prominent in <1% of the subsample. Demographic characteristics for the full and DNAm samples are displayed in Table 1. Due to the previously mentioned delays in the processing of the DNAm data by the lab, Study 2 utilized different inclusion criteria than Study 1. Specifically, Study 1 was restricted to participants who participated in both the TBED-C and MTwiNS studies (N = 798 individuals) with at least 2 waves of data and thus necessarily resided in an under-resourced neighborhood. 53 However, due to unforeseen DNAm processing delays, Study 2 was restricted to a smaller number of TBED-C participants (N=276 individuals) who were assayed for our pilot DNAm study. As a portion of these participants did not meet inclusion criteria for Study 1, we used different inclusion criteria for Study 2 – specifically, we required only one wave of data and a lower threshold for neighborhood disadvantage, namely, an ADI>20 as described above. Table 1. Sample Demographic Characteristics – Study 2 Demographic Variable Sex Male Female Race/Ethnicity White Hispanic/Latinx Black Native American/Native Alaskan Asian Pacific Islander Other Total Full Analytic Sample MWAS Sample N % N % 904 838 52% 48% 167 77 1404 80.6% 190 14 182 18 10 4 110 1742 0.8% 10.4% 1% 0.6% 0.2% 6.3% 100% 2 27 5 2 4 14 244 68.4% 31.6% 77.9% 0.8% 11.1% 2% 0.8% 1.6% 5.7% 100% Note. Abbreviations: N = sample size, SD = standard deviation. The mean age at wave 1 = 8 (SD = 1.5), at wave 2 = 14.5 (SD = 1.8), and at wave 3 = 16 (SD = 2.1). Measures Social Resilience. Social resilience was measured using the Child Behavior Checklist’s (CBCL; Achenbach & Rescorla, 2001) social competence subscale (wave 1 α = 0.56, wave 2 α = 0.55, wave 3 α = 0.55) which was collected at all three waves. This subscale included 10 items assessed using various scalings (i.e., a 3-point Likert scale, 4-point Likert scale, and dichotomous 54 2-point scale were used) completed by mothers regarding their child’s involvement in organizations, number of friends, contact with friends, behavior with others, and behavior alone (e.g., About how many times a week does your child do things with any friends outside of regular school hours?) during the prior six months. Items were then weighted by the ASEBA scoring program to compute a total sum score ranging from 0 to 14. Youth with higher scores on the social competency subscale were considered to have greater levels of social resilience. Psychological Resilience. Psychological resilience was also assessed via the CBCL (Achenbach & Rescorla, 2001), using the total problems subscale (wave 1 α = 0.94, wave 2 α = 0.94, wave 3 α = 0.92), which was also completed by mothers at all three waves. The subscale included 103 items regarding youths’ emotional (e.g., There is very little he/she enjoys) and behavioral problems (e.g., Destroys things belonging to his/her family or others), assessed on a three-point Likert scale ranging from (0) not true to (2) very true/often true. We reverse-scored the total problems subscale for analyses so that higher scores corresponded to greater levels of psychological resilience. DNA Methylation. Saliva samples were collected from youth during wave one (in middle childhood) using Oragene collection kits (DNA Genotek). DNA was then extracted from these samples via processes consistent with the Oragene Laboratory Protocol Manual Purification of DNA. Sodium bisulfite conversion was then performed on the extracted DNA by the University of Michigan Sequencing Core, after which the Infinium Human Methylation EPIC Bead Chip (Illumina) v1 was used to assay methylation in the converted DNA. We completed thorough quality control procedures and intra-sample normalization via the Chip Analysis Methylation Pipeline for Illumina HumanMethylation450 and EPIC (ChAMP) Bioconductor package (Butcher & Beck, 2015; Morris et al., 2014). We removed samples with 55 (1) a high proportion of failed probes (>10%; n=1), (2) detection p-values above 0.01 (n=86415 probes), (3) bead counts less than 3 in at least 5% of samples (n=3608 probes), (4) probes that aligned to multiple locations (cross-hybridizing probes; Nordlund et al., 2013), (5) probes not located at CpG sites (n=2242), (6) probes that overlapped with single nucleotide polymorphisms (SNPs), or (7) probes located on sex chromosomes (n=12,610). In addition, we employed the ComBat function from the Surrogate Variable Analysis Bioconductor package to correct for batch effects by slide and then array (Leek, 2020). We also estimated cell type proportions for the most common cell types in saliva via the Epigenetic Dissection of Intra-Sample-Heterogeneity (EpiDISH) Bioconductor package (Zheng, Breeze, Beck, & Teschendorff, 2018). These procedures ultimately yielded DNAm values (log2 methylated/unmethylated DNA at a specific probe, i.e., M-values) across 728,396 CpG sites for 276 participants from the TBED-C. Among those with DNAm data, 244 participants met criteria for the present study. Analytic Strategy Growth Curve Models. We fit Latent Growth Curve Models (LGCM) using Mplus version 8.6 (Muthén & Muthén, 1998-2019) to examine change in social and psychological resilience from middle childhood to emerging adulthood. LGCMs are grounded in Structural Equation Modeling (SEM; Kline, 2011) and estimate both the intercept (initial value) and slope (rate of change) of a given outcome as latent variables to model their trajectories over time. Due to the wide and overlapping ages of participants across our three waves of data, we allowed for individually-varying times of observation at all waves using time scores. To ensure that no participants had missing data for our time scores, we imputed the age for a small number of participants missing this data at waves two or three. Specifically, for wave two we added the mean number of years between waves one and two to the age of participants at wave one and for 56 wave three we added the mean number of years between waves one and three to the age of participants at wave one. Intercepts were centered at the youngest age in our sample, age six. We first fit unconditional growth curve models for social and psychological resilience, respectively, using three waves of data. Intercept-only models and linear growth models were compared for each respective outcome. Intercept-only models incorporate three parameters: intercept mean, intercept variance, and residual variance. Linear growth models incorporate six parameters: intercept and slope means, intercept and slope variances and their covariance, and residual variance. Better model fit was indicated by lower values on the Akaike Information Criterion (AIC; Akaike, 1987), Bayesian Information Criterion (BIC; Raftery, 1995), and sample-size adjusted Bayesian Information Criterion (saBIC; Sclove, 1987). We then fit conditional growth curve models by adding our demographic covariates (i.e., sex, race/ethnicity) to our best-fitting unconditional models. In light of the racial composition of our sample, we used a dichotomized indicator for race/ethnicity (0=minoritized; 1=White) to ease the interpretation of its confounding effect. To handle missing data, we employed Full Information Maximum Likelihood (FIML) as this approach is equipped to handle larger amounts of missing data and yield estimates with greater efficiency and less bias than alternative missing data techniques (Lang & Little, 2018). To account for the nesting of twins within families, we used the ‘cluster’ command in Mplus. Trajectories of psychological and social resilience were examined separately to allow for potential differences across these domains. Finally, we extracted the intercept and slope factors from the social and psychological resilience models for use in subsequent analyses. Methylome-Wide Association Studies. We performed a Methylome Wide Association Study in which ordinary least square (OLS) regressions were used to identify DNAm sites 57 associated with youth social and psychological resilience trajectory parameters (i.e., LGCM intercepts and slopes), or Differentially Methylated Probes (DMPs). Separate models were fit for each domain of resilience as well as each LGCM parameter (a total of four models). These models were fit in R, version 4.1. The sandwich package in R (Zeileis, 2006) was used to correct the standard errors within a heteroskedasticity-consistent covariance matrix estimator, which in turn accounted for the nesting of twins within families. We included sex, age, zygosity, race/ethnicity, four estimated cell-type proportions (i.e., those correlated with our outcomes: Epi, B, NK, and Neutro), and two principal components as covariates to account for potential confounding effects. Principal components were calculated for the control probes using the ‘prcomp’ R function. Consistent with past literature, methylome-wide significant DMPs were those with a p-value of P<9x10-8 (Mansell et al., 2019), whereas suggestive DMPs were those with a p-value of P<1x10-5 (Lander & Kruglyak, 1995). Enriched Pathway Analyses. To elucidate the biological pathways that are implicated in social and psychological resilience, we tested for overrepresentation of top MWAS genes (i.e., those that significant and suggestive DMPs were located in) in biological pathways within the KEGG Pathway, GO Biological Processes, Reactome Gene Sets, Canonical Pathways, CORUM, and WikiPathways databases using Metascape (Zhou et al., 2019). Enriched pathways were identified using a cut-point of p<.01 and were only considered if at least three genes from the top MWAS results were present in the pathway and the enrichment factor (i.e., the ratio of the observed counts to the counts expected by chance) was >1.5. Monozygotic Twin Difference Analyses. To eliminate genetic confounds and strengthen causal inferences, we also performed MZ twin difference analyses in R. MZ co-twins are genetically identical, and thus any differences in their epigenome must be environmentally 58 engendered. We therefore computed difference scores of our four outcomes and DNAm for the top DMPs (i.e., those that were significant or suggestive) across all MZ co-twins. We then used linear regression models to test for associations between co-twin difference scores of DNAm and co-twin difference scores for each respective resilience growth factor (i.e., slopes and intercepts for each domain), while controlling for demographic covariates (i.e., sex, age, and ethnicity). Finally, we employed a 95% statistical significance threshold (p < 0.05). As a caveat, this dissertation did originally intend to incorporate Growth Mixture Models (GMMs) into Study 2 and investigate whether DMPs predicted resilience class memberships. However, because GMMs further partition the sample into trajectory classes, we would not have been reasonably powered to perform MWAS given the more limited sample size available. We have thus focused here only on the variable-centered analyses. Results Descriptives Descriptive statistics and correlations for social and psychological resilience at all three waves are provided in Table 2. Youth evidenced high mean levels of psychological and moderate mean levels of social resilience (relative to the max score on each respective subscale), at all waves. Growth Curve Models We first compared the fit of unconditional linear growth and intercept-only models of social and psychological resilience, respectively. Linear growth models evidenced a better fit to the data for both social (AIC = 11657.14, BIC = 11700.63, saBIC = 11675.21) and psychological (AIC = 23896.95, BIC = 23940.52, saBIC = 23915.10) resilience, as compared to the intercept- only social (AIC = 11812.69, BIC = 11839.87, saBIC = 11823.98) and psychological (AIC = 59 24097.19, BIC = 24121.42, saBIC = 24105.53) resilience models. We therefore proceeded with the unconditional linear growth models. Table 2. Descriptive Statistics and Correlations 1. Psy Resilience T1 1. - 2. 3. 4. 5. 6. .480** .359** .228** .128** .109* 2. Psy Resilience T2 .480** - .653** .169** .237** .248** 3. Psy Resilience T3 .359** .653** - 0.088 .137* .316** 4. Soc Resilience T1 .228** .169** 5. Soc Resilience T2 .128** .237** 0.088 .137* - .410** .233** .410** - .549** 6. Soc Resilience T3 .109* .248** .316** .233** .549** Mean SD Actual Range Possible Range N Note. **p<.001, *p<.05 107.37 114.72 116.86 18.18 0-130 0-130 1722 15.30 9-130 0-130 765 7.44 2.42 13.76 55-130 0-13.5 0-130 444 0-14 1707 8.42 2.66 0-14 0-14 519 - 8.37 2.69 2-14 0-14 397 Social Resilience. As shown in Figure 1a, at age six, youth had a mean social resilience score of 7.13 (SE = 0.09, p<.001) and generally evidenced linear increases annually by a factor of 0.15 (SE = 0.02, p<.001). Youth demonstrated trending levels of between-person variation in their initial level (b = 1.55, SE = 0.81, p = 0.054) of social resilience. They did not, however, demonstrate between-person variation in their rate of change (b = 0.016, SE = 0.01, p = 0.195). Similarly, there was no relationship between the mean baseline level of social resilience and the rate of change, as evidenced by a non-significant intercept-slope covariance (b = 0.03, SE = 0.09, p = 0.716). We also fit a conditional linear growth curve model (see Table 3) to account for potential confounding effects of sex and race/ethnicity. As in the unconditional model, the mean social resilience intercept and slope were significant, while the intercept variance was trending 60 and the slope variance was non-significant. Neither sex nor race/ethnicity significantly predicted the social resilience intercept or slope. Figure 1. Unconditional Linear Growth Curve Models Psychological Resilience. As shown in Figure 1b, on average, youth demonstrated baseline psychological resilience scores equal to 105.99 (SE = 0.68, p<.001). Psychological resilience increased linearly by an average factor of 0.95 (SE = 0.09, p<.001) annually. The between-person variation was trending for the baseline level of psychological resilience (b = 61 97.493, SE = 54.14, p = 0.07), while it was non-significant for the rate of change (b = 0.10, SE = 0.60, p = 0.87). Finally, the intercept-slope covariance (b = 2.93, SE = 5.62, p = 0.60) indicated a lack of an association between psychological resilience scores at baseline and the rate of change. Next, as with social resilience, we fit a conditional linear growth curve model for psychological resilience in which two covariates were included: sex and race/ethnicity. Consistent with the unconditional model, the mean psychological resilience intercept and slope were significant. Unlike the unconditional model, however, youth evidenced significant between- person variation in both the intercept and slope (see Table 3). Sex was a significant predictor of the intercept and slope such that, on average, females had higher initial levels of psychological resilience relative to males, and their level of psychological resilience increased at a slower rate. Race/ethnicity did not predict the psychological resilience intercept, but it did predict the slope such that White participants demonstrated slower increases in social resilience relative to racially/ethnically minoritized participants. Finally, unlike the unconditional model, the intercept and slope were significantly correlated suggesting that youth who had higher baseline levels of psychological resilience were more likely to have more rapid increases in psychological resilience over time. Methylome-Wide Association Studies We performed MWA analyses for four outcomes: the social resilience intercept, social resilience slope, psychological resilience intercept, and psychological resilience slope. Quantile- quantile (QQ) plots (Figure 2) depict the number of probes deviating from the line of expected points based on the null hypothesis for each outcome. These QQ plots and lambda values suggest that our results are not inflated for any of our outcomes: psychological resilience intercept λ = 62 Table 3. Growth Curve Model Results Outcome Parameter Estimate S.E. p-value Estimate S.E. p-value Unconditional Model Conditional Model Psychological Intercept Resilience Mean Variance Ethnicity Sex Slope Mean Variance Ethnicity Sex 105.990 0.676 <.001 101.857 1.549 <.001 97.403 54.137 0.072 107.182 13.391 <.001 - - 0.952 0.099 - - - - - - 0.089 0.596 <.001 0.868 - - - - Intercept-Slope Covariance 2.928 5.621 0.603 Model fit Statistics AIC BIC saBIC 23896.952 23940.515 23915.100 1.614 0.822 0.161 <.001 0.170 0.005 0.197 0.044 0.248 <.001 <.001 0.025 <.001 <.001 2.265 4.701 1.727 0.057 -0.440 -0.839 2.469 23861.013 23926.358 23888.235 Social Intercept Resilience Mean Variance Ethnicity Sex 7.125 1.554 0.094 0.805 <.001 0.054 - - - - 63 - - 6.757 1.472 0.321 0.229 0.237 0.804 0.244 0.172 <.001 0.067 0.189 0.183 Table 3 (cont’d) Social Slope Resilience Mean Variance Ethnicity Sex 0.152 0.016 0.016 0.012 <.001 0.195 - - - - - - Intercept-Slope Covariance 0.031 0.085 0.716 Model fit Statistics AIC BIC saBIC 11657.144 11700.628 11675.213 0.147 0.014 0.025 -0.031 0.037 0.047 0.012 0.047 0.031 0.085 0.002 0.243 0.592 0.308 0.660 11652.822 11718.047 11679.925 Note. Results are provided for the unconditional and conditional growth curve models for each resilience domain, respectively. SE = Standard Error. Gender= 1(female), Ethnicity = 1(White). 64 Figure 2. Quantile-Quantile plots 65 0.9380, psychological resilience slope λ = 0.9353, social resilience intercept λ = 1.006, social resilience slope λ = 0.919. The test statistics and location of the top significant and suggestive DMPs are provided in Table 4. Results revealed no methylome-wide significant DMPs for either the psychological resilience intercept or slope, although we did find four DMPs (the same four) above the suggestive threshold for both psychological resilience models. The top suggestive DMP for both psychological resilience models was located in the TROVE2 gene, which encodes a Y RNA binding protein (Ro60; Heck et al., 2017) and is associated with emotional memory capacity, traumatic memory, and the development of PTSD (Heck et al., 2017). In addition, another top suggestive DMP for both models was located in ALCAM, a gene that encodes activated leukocyte cell adhesion molecules. ALCAM plays a key role in regulating and maintaining the stability of the blood-brain barrier (Lécuyer et al., 2017), promoting the growth of midbrain dopamine neurons (which facilitate reinforcement learning; Bye et al., 2019), and supporting the proliferation of T-cells (Zimmerman et al., 2006). What’s more, DMPs located on ALCAM have been found to be associated with narrative exposure therapy treatment (Carleial et al., 2021). For the social resilience intercept model, there was one significant DMP and 27 suggestive DMPs. The significant DMP was located in DPYSL2, a gene that encodes for a protein involved in a range of cellular processes important for brain formation and organization (e.g., cell migration, dendritic spine development, synaptic plasticity; Desprez et al., 2023). The literature suggests that polymorphisms or pathogenic genetic variants of DPYSL2 may contribute to the development of serious mental illness (i.e., schizophrenia; Desprez et al., 2023; Lee et al., 2015; Pham et al., 2016) and neurodevelopmental disorders (e.g., intellectual disability; Desprez et al., 2023). In addition, one of the top DMPs was located in the ACOX2 gene, which encodes 66 Table 4. Top Ten Significant and/or Suggestive Methylation Wide Association Study Differentially Methylated Probes Model Probe Chr Start Beta T-value P-value Gene Genomic Features Social Resilience Intercept cg12835868 3 4841393 -5.7907 -5.0152 1.06E-06 ITPR1 Intron cg05193970 3 58523648 3.9806 5.474 1.15E-07 ACOX2 cg24858031 3 27810553 -4.5504 -4.9895 1.19E-06 cg13663091 22 50584128 5.3789 4.9349 1.54E-06 MOV10L1 cg12218771 1 6696165 13.0458 4.912 1.71E-06 DNAJC11 cg06250135 8 26481821 -13.8191 -5.8285 1.88E-08 DPYSL2 cg07384913 3 63637117 8.5982 4.8792 1.99E-06 SNTN Exon Intron Exon cg13020870 15 37173016 4.7212 4.8751 2.03E-06 LOC145845 Intron; CpG island cg12426196 1 161008825 -17.1219 -4.8128 2.70E-06 TSTD1 CpG island cg15391116 6 133273587 11.603 4.8036 2.81E-06 cg19032306 10 94051795 -18.275 -4.7838 3.07E-06 MARCH5 Intron cg14707269 3 69435338 -37.8041 -4.7335 3.86E-06 FRMD4B Intron; CpG island cg13979085 3 58523569 3.8144 4.7304 3.91E-06 ACOX2 cg10258721 7 94023441 -3.0328 -4.7217 4.07E-06 COL1A2 cg07831612 15 67437032 -5.4246 -4.7127 4.23E-06 SMAD3 Intron cg16626280 5 128430254 -46.4344 -5.2057 4.28E-07 ISOC1 CpG island cg24557248 1 155990768 -28.2467 -4.7082 4.32E-06 SSR2 Exon; CpG island 67 Table 4 (cont’d) Social Resilience Intercept cg19593182 6 30176075 11.0234 4.6932 4.62E-06 TRIM26 Intron cg26330116 15 37172999 4.1833 4.6866 4.76E-06 LOC145845 Intron; CpG island cg10985094 17 3631481 6.927 4.6303 6.11E-06 ITGAE Intron cg17409297 1 31170296 4.7229 4.6256 6.24E-06 cg16400090 16 1235101 21.8311 4.6116 6.63E-06 CACNA1H cg21973667 13 73634423 -66.7777 -5.0968 7.21E-07 KLF5 cg26119884 2 228192346 3.2459 4.5768 7.73E-06 MFF cg17509337 1 41200810 3.8502 4.5627 8.22E-06 NFYC cg19584551 10 24721828 5.962 4.5402 9.06E-06 KIAA1217 cg25100061 20 57735524 -3.8386 -4.5363 9.22E-06 ZNF831 cg21594961 3 3152915 1.7892 4.5193 9.93E-06 IL5RA cg04002187 5 40835753 0.6986 4.9891 1.20E-06 RPL37 cg12426196 1 161008825 -1.3601 -4.9729 1.29E-06 TSTD1 Intron Intron Intron Intron Intron Exon Intron CpG island CpG island Social Resilience Slope cg13663091 22 50584128 0.4271 4.9705 1.30E-06 MOV10L1 Exon cg24858031 3 27810553 -0.3270 -4.9620 1.36E-06 cg16626280 5 128430254 -3.2611 -4.9524 1.42E-06 ISOC1 CpG island cg17509337 1 41200810 0.2826 4.8257 2.54E-06 NFYC Intron 68 Table 4 (cont’d) Social Resilience Slope Psychological Resilience Intercept Psychological Resilience Slope cg10258721 7 94023441 -0.2303 -4.8248 2.55E-06 COL1A2 cg07576664 10 34408653 0.2616 4.7884 3.01E-06 PARD3 Exon cg26262605 1 93427690 -2.3743 -4.7593 3.43E-06 FAM69A Exon; CpG island cg06250135 8 26481821 -0.9025 -5.2414 3.60E-07 DPYSL2 cg07890549 2 225681274 0.4184 4.7446 3.67E-06 DOCK10 Exon Intron cg26593997 16 18801737 -1.5112 -4.7064 4.35E-06 RPS15A Intron; CpG island cg13020870 15 37173016 0.3280 4.6243 6.27E-06 LOC145845 Intron; CpG island cg07211768 12 115102289 0.3452 4.5573 8.41E-06 cg03226204 8 81086267 1.1225 4.5409 9.04E-06 TPD52 cg23971069 1 214158685 0.9609 4.5202 9.89E-06 PROX1-AS1 cg14630503 1 193038286 188.1095 4.9011 1.80E-06 TROVE2 cg07677545 3 105088087 -384.9823 -4.6599 5.36E-06 ALCAM cg09838701 3 121839599 -230.5418 -4.5786 7.67E-06 CD86 cg19639041 4 1569745 -46.1309 -4.5754 7.77E-06 Intron Intron Exon Intron Exon cg14630503 1 193038286 5.7507 4.8682 2.09E-06 TROVE2 Exon cg07677545 3 105088087 -11.7992 -4.6500 5.60E-06 ALCAM Intron, CpG island cg09838701 3 121839599 -7.0567 -4.5706 7.94E-06 CD86 Exon 69 Table 4 (cont’d) Psychological Resilience Slope cg19639041 4 1569745 -1.4102 -4.5561 8.46E-06 Note. ‘Probe’ is the name of the CpG probe in the human reference genome hg19/GRCh37, ‘Chr’ is Chromosome, ‘Start’ is the base pair location of the probe, ‘Gene’ is the gene the probe is located in, and ‘Genomic Feature’ indicates if the probe is located in an intron, exon, or CpG island. Also shown are the signed test statistic values for regression: ‘T-values’, ‘P- values’, and ‘Beta’ or regression coefficient. The methylome-wide significant (P-value < 9 x 10-8) and/or suggestive (P-value < 1 x 10-5) MWAS DMPs are displayed for each outcome. 70 the branched-chain acyl-CoA oxidase enzyme. Notably, a deficiency in this enzyme has been linked to neurological dysfunction such as severe cognitive disability (Vilarinho et al., 2016). Finally, there were zero significant DMPs and 16 suggestive DMPs for the social resilience slope model. Among the top suggestive results was a DMP located in the aforementioned DPYSL2 gene. In addition, another top DMP was located in ISOCI, a protein coding gene which has been implicated in late-onset Alzheimer's disease (Jiang et al., 2016). Enriched Pathways Most of the significant/suggestive DMPs for each of our outcomes were located in unique genes: 22 of 28 for the social resilience intercept, 14 of 16 for the social resilience slope, 3 of 4 for the psychological resilience intercept, 3 of 4 for the psychological resilience slope. All pathways that were found to be significantly enriched are listed in Table 5. Given the limited number of significant or suggestive DMPs for our psychological resilience outcomes (i.e., four for each), it not surprising that we did not find any significantly enriched pathways for either the psychological resilience intercept or the psychological resilience slope. In sharp contrast, the top DMPs for the social resilience intercept were located in genes that are implicated in nine pathways. The SMAD3 gene, which encodes for a protein that is a transcription factor and tumor suppressor involved in regulating gene activity and cell proliferation, was involved in all nine pathways. Among the top pathways were the PID MYC repress pathway and the apoptotic signaling pathway. MYC is a proto-oncogene and transcription factor involved in key biological processes such as cell growth, cell proliferation, and apoptotic cell death. The apoptotic signaling pathway facilitates balance between cell 71 Table 5. Enriched Pathways Model Pathway p-value q-value Gene Overlap PID MYC repress pathway Log10(-4.96) Log10(-0.61) Apoptotic signaling pathway Log10(-2.89) Log10(0.00) Cytokine signaling in immune system Log10(-2.73) Log10(0.00) COL1A2, SMAD3, NFYC, IL5RA, TRIM26, KIAA1217 ITPR1, SMAD3, MFF, MARCHF5 COL1A2, IL5RA, SMAD3, TRIM26 Social Resilience Intercept Signaling by interleukins Log10(-2.39) Log10(0.00) COL1A2, IL5RA, SMAD3 Skeletal system development Log10(-2.32) Log10(0.00) COL1A2, SMAD3, KIAA1217 Positive regulation of establishment of protein localization Log10(-2.84) Log10(0.00) ITPR1, SMAD3, MFF Positive regulation of protein localization Log10(-2.36) Log10(0.00) ITPR1, SMAD3, MFF Positive regulation of organelle organization Log10(-2.30) Log10(0.00) SMAD3, MARCHF5, MFF Regulation of establishment of protein localization Log10(-2.24) Log10(0.00) ITPR1, SMAD3, MFF 72 Table 5 (cont’d) Metabolism of amino acids and derivatives Log10(-3.34) Log10(0.00) RPL37, RPS15A, TSTD1 Social Resilience Slope Axon guidance Log10(-2.83) Log10(0.00) DPYSL2, RPL37, RPS15A Nervous system development Log10(-2.78) Log10(0.00) DPYSL2, RPL37, RPS15A Note. ‘Pathway’ is the name of the significantly enriched pathway from the databases assessed in Metascape, and ‘Gene Overlap’ provides the names of genes from the MWAS that are present in the pathway. Also shown are the signed test statistic values for the pathway analyses in log base 10 format: ‘p-value’ and ‘q-value’. There were no significantly enriched pathways for the psychological resilience intercept or slope. 73 creation and cell death. Notably, both of these pathways are implicated in the development of cancer. The social resilience slope DMPs yielded three significantly enriched pathways; the RPL37 gene—which encodes a ribosomal protein—was present in all three, while the aforementioned DPYSL2 gene was present in two. The top pathway was involved in the metabolism of amino acids and derivatives which ultimately provides amino acids that facilitate protein synthesis. The other two pathways were involved in the nervous system: the axon guidance pathway and the nervous system development pathway. The former specifically helps axons reach their targets while the latter is more broadly involved in a range of nervous system functions (e.g., cell migration, axon and dendrite formation, synapse elaboration). Monozygotic Twin Differences The last step of our analysis was to assess whether the top DMPs (i.e., suggestive and significant DMPs) yielded from the MWAS for each resilience outcome were environmentally engendered, via MZ twin difference analyses. Although they were strongly correlated (r = 0.623, p<.001) in their rank-ordering across the sample, we also found that nearly all MZ pairs (96.6%) had different baseline levels of psychological resilience, with a mean co-twin difference of 59.6% of the typical phenotypic standard deviation across the full sample. Similarly, 96.6% of co-twins also had different, though strongly correlated (r = 0.623, p<.001) psychological resilience slopes, with a mean co-twin difference of 59.4% of the typical phenotypic standard deviation across the sample. This pattern extended to social resilience, to a slightly lesser extent. The majority (83.9%) of twins evidenced different levels of social resilience at baseline, with strong twin intraclass correlations (r = 0.834, p<.001) and mean co-twin differences of 41% of the typical standard deviation of the sample. For the social resilience slope, although 83.1% of 74 twins diverged, twin intraclass correlations were strong (r = 0.851, p<.001) and co-twin differences were only 5.6% of the typical standard deviation across the sample. Finally, there were no co-twins with identical DNAm scores across all 728,396 CpG sites. Table 6 includes the results of our MZ twin difference analyses. Co-twin differences for five DMPs were significantly associated with co-twin differences in their social resilience intercept. Among them were DMPs located on the aforementioned ACOX2 and DPYSL2 genes. In addition, another significant DMP was located on the FRMD4B gene, which encodes a protein that is believed to function as a scaffolding protein such that it is involved in organizing and stabilizing other intracellular proteins. No co-twins had DMP differences that were significantly associated with co-twin differences in any of our other outcomes (i.e., the social resilience slope, psychological resilience intercept, and psychological resilience slope). Discussion The present study sought to elucidate the trajectories of youth social and psychological resilience over time and identify methylomic biomarkers of those trajectories. We used a longitudinal sample from the MSUTR and found that, on average, youth evidenced high baseline levels of psychological resilience and moderate baseline levels of social resilience. Consistent with prior literature on resilience development during adolescence (Yoon, Sattler, et al., 2021), we found that youth generally exhibited linear increases in each domain of resilience from middle childhood to emerging adulthood. While neither sex nor race/ethnicity were significantly associated with social resilience, both were associated with psychological resilience trajectories. This finding coincides with literature demonstrating divergent prevalence rates of psychological resilience based on sex (particularly in adolescence; Zahn-Waxler et al., 2008) and culture (Kirmayer & Ryder, 2016). Our results demonstrating higher initial levels of psychological 75 Table 6. Significant MZ Twin Difference DMPs Model Probe Chr Start Beta T-value P-value Gene Genomic Features Social Resilience Intercept cg15391116 6 133273587 5.5436 2.8643 0.0051 cg19584551 10 24721828 -2.2516 -2.6289 0.0100 KIAA1217 Intron cg05193970 3 58523648 -2.9726 -2.4754 0.0151 ACOX2 cg14707269 3 69435338 14.9494 2.2607 0.0261 FRMD4B Intron, CpG Island cg06250135 8 26481821 5.9667 2.0003 0.0483 DPYSL2 Exon Note. ‘Probe’ is the name of the CpG probe in the human reference genome hg19/GRCh37, ‘Chr’ is Chromosome, ‘Start’ is the base pair location of the probe, ‘Gene’ is the gene the probe is located in, and ‘Genomic Feature’ indicates if the probe is located in an intron, exon, or CpG island. Also shown are the signed test statistic values for regression: ‘T-values’, ‘P- values’, and ‘Beta’ or regression coefficient. 76 resilience that increase at a slower rate in females compared to males coincide with literature suggesting that males exhibit higher rates of ‘early-onset disorders’ (e.g., conduct disorder) whereas females have higher rates of ‘adolescent-onset disorders’ (i.e., internalizing disorders; Zahn-Waxler et al., 2008). What’s more, in light of research suggesting that DNAm may be implicated in resilience cross-sectionally, we performed a series of MWAS analyses to assess whether DNAm in middle childhood may be longitudinally predictive of resilience trajectories. Findings were relatively stronger for social resilience than psychological resilience. The only methylome-wide significant DMP was found for the social resilience intercept, along with multiple suggestive DMPs: twenty-seven for the intercept, and sixteen for the slope. By contrast, we did not find any methylome-wide DMPs for psychological resilience and only four suggestive. Moreover, we found evidence of significantly enriched pathways only for social resilience. Nine significantly enriched pathways were found for the social resilience intercept, many of which were involved in the positive regulation of protein localization, RNA transcription, and regulating cell proliferation and apoptotic death. In addition, we found evidence of three enriched pathways for the social resilience slope, two of which were implicated in the development of the nervous system. These results collectively imply that DNAm in middle childhood may be more likely to play a role in social resilience trajectories than psychological resilience trajectories, although even here, only one DMP was significant at the methylome-wide level. Our mostly null results for psychological resilience could suggest that DNAm may be limited in its ability to predict subsequent resilience in this domain. However, given that DNAm is known to be malleable, particularly in response to pubertal development (Han et al., 2019), it is also possible that DNAm 77 measured at a single time-point during middle childhood does not adequately capture the later DNAm alterations that impact the development of psychological resilience. Finally, our MZ twin difference analyses revealed that co-twin differences in five DMPs were significantly associated with co-twin differences in the social resilience intercept. Because MZ co-twins share 100% of their DNA, these significant results point directly towards environmentally induced DNAm alterations. Likewise, they suggest that associations between social resilience and the remaining top MWAS DMPs may instead reflect genetic or developmental processes affecting DNAm as opposed to causal environmental processes. Alternatively, they may be a function of family-wide influences on DNAm or MZ differences that were too small to capture environmental mediation. The present study's use of a genetically informed longitudinal sample of youth exposed to significant neighborhood adversity uniquely positioned us to investigate resilience across development and its associations with DNAm. Even so, there were important limitations to our study. For instance, due to the racial breakdown of our sample and our focus on recruiting families residing in mid-Michigan, our findings may not generalize to racially diverse communities or to individuals in other geographic regions. In addition, our approach to measuring resilience could be improved in three ways. First, we only employed a deficit-focused measure of psychological resilience to ensure measurement consistency across all the waves. Consistent with the strengths-based emphasis of resilience science, future work should strive to incorporate measures that capture the presence of positive psychological outcomes as opposed to solely measuring the absence of negative outcomes. Second, we relied exclusively on mother informant report to measure resilience due to missing teacher and youth self-report at particular waves (i.e., middle childhood for self-report and emerging adulthood for teacher-report). To 78 minimize potential bias and error, studies should incorporate additional informants as well as more objective measures of resilience (i.e., clinical structured interviews, observed paradigms/tasks). Third, the present study was only able to examine two domains of resilience despite evidence of three or more such domains in the broader literature (Miller-Graff, 2020; Vazquez et al., 2021). Beyond measurement issues, our study also had analytic limitations. As one example, our MWAS sample was small in size and thus we were likely underpowered to detect DMPs with small effect sizes. We therefore relied on extracted estimates of resilience growth factors in our MWAS to preserve power as opposed to estimating them directly in the same model as the MWAS. Moreover, given that DNAm can be tissue-specific, etiologic interpretations regarding our top DMPs from saliva tissue warrant caution as they may not be reflective of processes in the brain per se. Finally, we were unable to include a replication sample in our analyses and thus, our findings are provisional and should be investigated further. Unfortunately, we are not aware of another existing longitudinal twin sample with high rates of adversity and DNAm data with which to perform replication analyses at this time. Indeed, the field would benefit greatly from the recruitment of additional diverse and longitudinal twin samples that include the collection of bio-samples. In addition, as mentioned previously, we were unable to take a person-centered approach to investigating the role of DNAm in resilience trajectories in the present study due to unexpected power constraints. In light of the findings from Study 1 demonstrating heterogeneity in resilience trajectories, it is accordingly possible that DNAm associations were obscured by the variable centered approach we employed. For instance, nearly half of the sample in Study 1 was found to evidence stable trajectories of social resilience across development; perhaps we would 79 have seen a greater number of DMPs that distinguished between increasing and stable trajectories of social resilience with a person-centered approach. Similarly, given that psychological resilience was found to have three trajectory classes in Study 1, perhaps our very limited findings for this outcome relate to its unexplored heterogeneity. Future work is needed to investigate these possibilities using person centered methods. Implications Our study is the first to investigate DNAm as an epigenetic biomarker of developmental trajectories of youth resilience. Our growth curve models demonstrated that on average, youth have moderate-to-high levels of resilience in middle childhood, and generally increase in their levels of social and psychological resilience over time. These findings demonstrate that resilience is indeed malleable across adolescence such that most youth continue to increase in their degree of resiliency across youth development. Moreover, our MWAS results suggest that DNAm may play a relatively larger role in social resilience development than in the development of psychological resilience. Indeed, some of the top DMPs for the social resilience intercept did appear to be environmental in origin and were implicated in a number of enriched pathways. Nonetheless, given that the bulk of our MWAS results were null, we must also acknowledge the possibility that DNAm may play only a minimal role in subsequent youth resilience trajectories, particularly in the psychological domain. 80 GENERAL DISCUSSION The studies in this dissertation sought to expand the knowledge base regarding youth resilience by addressing three understudied areas in the resilience literature. First, consistent with contemporary conceptualizations that recognize the dynamic nature of resilience, we investigated longitudinal trajectories of youth resilience during critical developmental transitions. Second, in light of empirical evidence supporting the multidimensionality of youth resilience, we examined trajectories of two domains of resilience independently. Third, we investigated the influence of socioecological and biological promotive factors on resilience trajectories, including parental nurturance, neighborhood social processes, and DNA Methylation (DNAm). Summary of Results and Implications Study 1 was novel in its examination of youth resilience from middle childhood to emerging adulthood and in its use of both variable and person-centered methodological approaches. Our results demonstrated that on average, youth were characterized by high levels of psychological resilience in middle childhood that continued to increase over the course of development. Similarly, youth generally exhibited moderate levels of social resilience in middle childhood and linear increases thereafter. Notably, however, we found evidence of three profiles of youth psychological resilience development and two profiles of youth social resilience development. Parental nurturance and demographic characteristics (race/ethnicity or sex) predicted social and psychological resilience trajectories, while neighborhood social processes predicted only psychological resilience trajectories. Collectively, our results demonstrate that youth resilience typically increases in prevalence across development and does so in part as a function of parental nurturance, demographic variables, and neighborhood social processes. 81 Our results for social resilience augment past literature by demonstrating that while the majority of youth had increasing trajectories of social resilience, a sizable proportion also remained stable. That being said, we had expected to see more variability in social resilience trajectories, including perhaps a trajectory with lower baseline levels or a negative slope. One potential reason for the unexpected absence of this trajectory could relate to our operationalization of this construct. The CBCL’s social competence subscale assesses the quantity of friendships, frequency of social involvement, and behavior in different contexts. While this scale captures youths’ social skills, it could be too narrow in that it has limited capacity to assess the quality of youths’ social relationships. That is, a child might have sufficient social skills to form friendships, but if they are unable to foster and maintain positive social relationships, perhaps we would not consider this child to have high levels of social resilience. Future work might consider incorporating youths’ ability to specifically maintain positive social relationships into their conceptualization and measurement of social resilience. Even so, consistent with prior literature and our hypotheses, we found that greater parental nurturance influenced social resilience trajectory class membership. Parental nurturance may thus serve as an important intervention target for promoting increases in social resilience across development. By contrast, and contrary to our hypotheses, neighborhood social processes did not influence social resilience trajectory profile membership. It is possible that these influences were too distal to impact youth social development. Notably, a more proximal socioecological factor that has been implicated in resilience that we did not assess is peer support (Stewart & Sun, 2004); future work should consider the role of peer support in fostering the development of youth social resilience. Finally, race/ethnicity was also included as a covariate and we found that White youth were more likely to evidence increasing trajectories of resilience 82 than racially/ethnically minoritized youth. We suspect that these results reflect the intersecting identities and compounding stressors that racially and ethnically minoritized youth in disadvantaged neighborhoods face (e.g., discrimination, systemic inequity, high toxicant exposure). For psychological resilience, youth evidenced three trajectory profiles. The least common trajectory was characterized by high baseline psychological resilience with decreases across development. The second most common was a trajectory with a moderate baseline and increases over time, while the most common trajectory had high baseline levels of psychological resilience and increases over time. Interestingly, our finding that an increasing trajectory of psychological resilience was most common is consistent with prior resilience literature (Yoon, Sattler, et al., 2021), but stands in contrast to literature on psychopathology more broadly. That is, literature has demonstrated that rates of psychopathology typically increase during adolescence (Costello et al., 2011). Although one might reasonably argue that psychological resilience and psychopathology are different constructs and therefore have distinct trajectories, it is also true that in practice, the majority of resilience literature (studies 1 and 2 included) fails to adequately measure resilience in a way that is meaningfully distinct from the absence of psychopathology. Because the majority of resilience research is performed via secondary data analysis, many scholars rely on measuring the absence of psychopathology to the exclusion of measures of positive psychological outcomes (e.g., well-being). With that in mind, it is somewhat surprising that on average (and most often, in the case of our GMMs), youth demonstrated increasing trajectories of psychological resilience (i.e., reflecting lower scores on the total problems scale of the CBCL). 83 One possible explanation for this pattern of results is that the youth in our studies may evidence a distinct pattern from the general population. Given the chronic and pervasive nature of neighborhood disadvantage, it is possible that youth in such contexts are forced to learn to adapt and cope with adversity earlier in life than unexposed youth. If this were true, we would likely still see some variability in the ease with which youth are able to adapt. Perhaps, some youth would quickly adapt, whereas others would demonstrate a delay in their ability to bounce back, and others might be unable to adapt and instead progressively decrease in their level of resilience. Indeed, this pattern is consistent with the three trajectory classes evidenced by our psychological resilience growth mixture models. When investigating socioecological promotive factors of psychological resilience, our results were consistent with prior work as well as our social resilience results in that we found that greater parental nurturance promoted high-increasing trajectories of psychological resilience as opposed to either a high-decreasing or moderate-increasing trajectories. In addition, youth with higher levels of neighborhood informal social control were more likely to have moderate- increasing trajectories of psychological resilience than they were to have high-decreasing trajectories. Interestingly, this finding did not extend to the high-increasing trajectory class, suggesting informal social control may be more promotive for youth who have lower initial levels of psychological resilience compared to those with higher initial levels. In addition, youth with high neighborhood social cohesion were more likely to have high-increasing as opposed to moderate-increasing trajectories of psychological resilience. This pattern of results suggests that for youth who had increasing trajectories of resilience, neighborhood social cohesion effectively promoted persistently high levels of youth resilience. Finally, we examined sex as a covariate and found that females were more likely to be in the high decreasing trajectory class as opposed 84 to either increasing trajectory class. Similar to race/ethnicity, these findings can be interpreted in light of the compounding stressors and barriers that females in disadvantaged neighborhoods face (e.g., discrimination, inequity, and higher rates of harassment and sexual violence as compared to males). Study 2 examined the trajectory of psychological and social domains of resilience via a variable-centered approach and incorporated the examination of DNAm as a predictor of resilience growth factors via a series of methylome-wide association studies (MWAS). Although the sample for this study was somewhat distinct from that of Study 1 (due to different inclusion criteria), the pattern of the growth curve model results was consistent with those of Study 1. Our MWAS results further revealed differences across domains with somewhat clearer evidence of methylomic associations with the development of social resilience than of psychological resilience. We found one significant differentially methylated probe (DMP) for the social resilience intercept and multiple suggestive DMPs for the social resilience intercept (N=27) and slope (N=16). Pathway analyses additionally revealed nine enriched pathways for the social resilience slope and three for the social resilience intercept. Many of the enriched pathways for the social resilience intercept were implicated in RNA transcription as well as regulation of cell proliferation, apoptotic death, and protein localization. For the social resilience slope, a theme emerged of enriched pathways involved in aspects of nervous system development. What’s more, monozygotic (MZ) twin difference analyses suggested that the DNAm of five probes (all for the social resilience intercept) was environmentally induced, while the remainder were the result of genetic, developmental, or family-wide influences. These results notably differ from those of psychological resilience. Indeed, we found no significant DMPs or enriched pathways, and only four suggestive DMPs for the intercept and the slope. 85 That said, study 2 provided less evidence of associations between DNAm in middle childhood and resilience trajectories thereafter than we had expected. Given that prior animal literature and cross-sectional human studies supported our hypotheses, it is possible that our study was merely ill equipped to capture this association. For instance, DNAm is known to be malleable throughout development, particularly during pubertal development when adolescents experience a recalibration of their stress response system and experience a developmentally appropriate period of heightened emotion dysregulation – all of which are highly relevant for youth resilience. It is therefore possible that middle childhood (the time period during which we measured DNAm) was too early to capture salient DNAm alterations for resilience to adversity. Future studies should conduct an examination of the relationship between resilience and DNAm at different points. Presently, our lab is assaying bloodspot and saliva samples for 500 twin families (1000 twins) at birth, middle childhood, early adolescence, and late adolescence/emerging adulthood. Using these data, we hope to eventually perform a cross-lagged panel model in which DNAm and resilience are simultaneously modeled over time. This would illuminate whether DNAm at a particular time is most predictive of concurrent or subsequent resilience. In addition, we would be able to examine DNAm measured at different waves as a time-varying predictor of youth resilience trajectories. Another important limitation of Study 2 that may relate to our results is our use of salivary samples to assess DNAm. Although the use of saliva samples to measure DNAm in living humans is not uncommon due to the relative ease of biosample collection, DNAm is tissue-specific and thus DNAm measured in saliva may not reflect DNAm in other tissues. Studies examining cross-tissue concordance in DNAm have shown that there is correspondence between saliva and brain tissue, but only at a very small number of CpGs. Thus, for the majority 86 of CpGs, DNAm in saliva tissue does not correspond to those in brain tissue. This is an important limitation of study 2 given that our interpretations regarding the biological basis of significant and suggestive DMPs are typically tied to neurobiological processes and how they may relate to youth psychological and behavioral outcomes. Unfortunately, there are few options for overcoming this limitation when studying DNAm in living humans. One option would be to measure DNAm in the brain tissue of deceased humans. However, we know of no brain studies to date that have collected data regarding resilience or risk/promotive factors. In addition, because methylome-wide arrays assess DNAm across over 850,000 CpG sites, they require correction for multiple testing. Given our sample size, we may thus have been underpowered to detect significance for smaller effect sizes. The field would accordingly benefit from studies with a larger sample size to clarify the association between DNAm and youth resilience trajectories. Unfortunately, genetically informed work is hindered by the lack of sufficiently diverse (socioeconomically and racially/ethnically) twin samples. That being said, there are diverse longitudinal samples of singletons with DNAm data that can at minimum perform MWA analyses using similar measures to assess resilience (e.g., The Future of Families and Child Wellbeing Study). All that said, it is also important to acknowledge that our findings may also simply reflect a limited relationship between DNAm and developmental trajectories of resilience. Collectively, the studies in this dissertation elucidated developmental patterns of youth resilience and the potential role of promotive factors. Indeed, while we identified multiple socioecological promotive factors of resilience, we were unable to identify strong biological promotive factors. Additional work is therefore needed to investigate the biological mechanisms underlying resilience development. 87 Developmental Considerations FUTURE DIRECTIONS Developmental literature suggests that there are sensitive developmental periods during which socioecological influences may be most impactful. Despite this, to my knowledge, the developmental timing of adverse exposures and promotive factors—which resilience scholars suspect may offset the risks associated with adversity—has seldom been considered. It is plausible that the timing of promotive factors and shifts in the strength of their importance to youth outcomes also impact resilience trajectories, particularly regarding how their presence coincides with the timing of adverse exposures and developmental milestones. To address the impact of the developmental timing of adverse exposures and promotive factors, future multi- wave studies of resilience should incorporate a thorough evaluation of each at all waves. This data could be examined using a multivariate random-intercept cross-lagged panel analysis in which the impact of adversity and promotive factors on resilience over time is modeled. Ideally, these studies would also fully assess the types of adverse exposures. The definition of ‘adversity’ varies substantially across studies and is often described quite minimally. Often, studies will focus on a single form of adversity such as neighborhood disadvantage (as in Studies 1 and 2) or maltreatment (as in Sattler et al., 2023 and Yoon et al., 2021). Despite the likelihood that participants have been exposed to multiple forms of adversity to varying degrees, this is rarely thoroughly assessed. It is entirely possible that the type of adversity (as well as its severity and frequency, e.g., acute versus chronic) might shape youth resilience development. For instance, youth exposed to acute adversities may experience greater non-linear fluctuations in their resilience trajectories and may be more responsive to certain types of promotive factors (e.g., parental nurturance). Conversely, youth exposed to more chronic forms of adversity may 88 have progressively learned to adapt to their environment and may be more likely to have shared their experiences of adversities with others (e.g., community violence) and therefore benefit more from contextual promotive factors (e.g., neighborhood social processes). One way in which this could be investigated is to thoroughly assess lifetime history of a wide range of potentially traumatic or adverse events, including an assessment of the frequency and severity of each event. To facilitate developmental modeling, researchers should perform multi-wave data collection of resilience indicators and adversity across development. The potential differential impact of acute versus chronic adversity exposure could subsequently be teased apart with these data via a multi- group growth curve model. If the model evidences a better fit when parameters are allowed to vary between groups (i.e., those exposed to acute adversity versus those exposed to chronic adversity) compared to when they are fixed to be zero, this would suggest that the groups do in fact have divergent trajectories of resilience. This model could also be leveraged to examine whether youth exposed to distinct forms of adversity evidence similar or distinct trajectories of resilience. Next, it would be important to study specific developmental transitions. Among the most important contributions of the studies in this dissertation is the examination of resilience during critical developmental transitions. While we were able to examine resilience from middle- childhood to emerging adulthood, this leaves a great deal of adult development unexamined. Indeed, early adulthood is a period of significant agentic and identity development, during which new stressors and responsibilities are encountered. To date, it remains unclear how resilience develops through the rest of emerging adulthood into middle adulthood and older adulthood. Does the presence of promotive factors earlier in development continue to impact trajectories of resilience across adulthood? If so, are those that occur in early childhood, middle childhood, or 89 adolescence most salient? Do these early promotive factors prepare individuals to adapt to adversity more successfully in adulthood? Or are promotive factors that are present concurrently or in recent history most relevant for resilience in adulthood? These are important questions that resilience scholars could begin to answer by collecting data on resilience, adverse exposures, and promotive factors across the lifespan. Indeed, leveraging data across five or more waves would enable the detection of more nuanced and non-linear trajectories of resilience. Biological Mechanism of Resilience Study 2 suggests that DNAm accounts for little to no variance in the development of resilience. The question therefore remains, what other biological processes might be influencing youth resilience? The stress response system is a natural candidate given its prominent role in reactions to adversity. Indeed, the HPA axis has often been discussed in relation to resilience (Feder et al., 2019). While there is robust evidence of a link between HPA axis activity and psychopathology as well as social competence (Raymond et al., 2018), scholars have seldom taken a strengths-based approach to investigate the role of the HPA-axis on youth development. Does the HPA axis impact concurrent as well as longitudinal trajectories of youth resilience? Is this occurring through a GxE process? One way in which these questions could be answered is by collecting longitudinal data on youth resilience from twin families and employing a stress induction paradigm during and after which participant’s stress responses (e.g., cortisol levels, heart rate) are measured. These data could be analyzed in a number of ways. First, phenotypic analyses could be performed to simply assess the relationship between the stress response system and resilience using structural equation modeling. A moderated mediation model could be conducted in which adversity is the independent variable, resilience is the dependent variable, cortisol is the mediator, and promotive factors are modeled as moderators in the relationship 90 between adversity and cortisol levels. This would allow us to test whether promotive factors buffer against the harmful effects of adversity on the stress response system, ultimately enabling resilience in the face of adversity. Alternatively, using a genotype-by-environment interaction model, cortisol levels could be analyzed as a moderator of genetic and environmental influences on youth resilience. This could be modeled cross-sectionally at each wave and/or by using growth factors of resilience growth curve models as outcomes (i.e., the intercept and slope). One final biologic pathway that may undergird resilience development involves neural structure, function, and connectivity (Feder et al., 2019; Bezek et al., 2023). Neuroimaging studies conducted in adults suggest that decreased activation of brain regions implicated in threat appraisal (e.g., amygdala) and increased activation of prefrontal regions related to cognitive control and emotion regulation support greater adaptation to stress and adversity (Feder et al., 2019). Moreover, greater volumes of gray matter in the hippocampus and prefrontal cortices have been linked to youth resilience (K. B. Burt et al., 2016; Morey et al., 2016). Beyond the structure and function of particular brain regions, larger brain networks that support higher level cognitive processes have also been implicated in resilience (Iadipaolo et al., 2018). Specifically, these include the frontoparietal network and the default mode network, which are involved in cognitive control and self-regulation. While extant literature is promising, much of this work has been restricted to one domain of resilience and none have examined neural correlates of resilience development. There is thus a clear need for additional work examining neural correlates of longitudinal resilience trajectories. Cultural Considerations One notable area of the resilience literature that remains relatively under-explored involves cultural considerations, including extant work on the foundational conceptualizations of 91 resilience, socioecological promotive factors, and potential differences in trajectories. Contemporary conceptualizations of resilience and theorized promotive factors are predominantly grounded in Western culture. Given nuanced cultural differences in values and lifestyle, it stands to reason that the very idea of what resilience is, and the factors that give rise to it, might also be culturally bound. In other words, the definition of “successful adaptation” and the resources/factors that promote this resilience seem likely to vary based on the cultural lens through which they are viewed. For instance, familismo or familism (i.e., a value set in which the interests of the family are prioritized) is a salient value in Latinx culture that has been demonstrated to serve as a protective factor for youth exposed to adversity (Piña-Watson et al., 2019). As another example, in some cultures, family goes beyond the nuclear unit and includes extended family. For these youth, family cohesion and support may be equally or more salient promotive factors than parental support and nurturance. There is also evidence that amongst collectivistic cultures, general belief systems (as opposed to personal belief systems) and perspective-taking are influential in fostering psychological resilience, whereas they are less influential in individualistic cultures (Özcan & Bulus, 2022; Wu et al., 2011). Cultural considerations must also be applied to resilience development. In light of varying cultural norms and values regarding transitions to adolescence and early adulthood, trajectories of resilience could potentially look differently across cultural contexts. As one example, increased levels of agency and independence are often valued as youth transition into adolescence and adulthood. However, this might look quite different across cultures. In individualistic cultures, youth may be expected to move out of their parent’s home by their late teens or early twenties and the idea of them living with their parents into their late 20s or 30s might be indicative of a ‘failure to launch’. By contrast, collectivistic cultures might expect 92 young adults to remain in the home, taking on greater familial and financial responsibilities; for these youth, moving out of their parent’s home before they are settled into their career and/or own family may even be frowned upon. Although some work has examined differences in promotive or protective factors between individualistic and collectivistic cultures, I know of no extant work that has investigated cultural differences in resilience conceptualizations or in promotive factors for resilience development. How might this be done? Scholars could seek to identify cross-cultural differences in resilience using qualitative or quantitative methods. Given the novelty of this work, however, it may be best explored using a mixed methods approach. The qualitative portion could involve a series of focus groups or individual interviews in which participants are asked to describe resilience and what it looks like in their culture. Participants could also be asked to describe socioecological influences that they feel support resilience for members of their cultural group. It would also be important to inquire whether and how culturally specific definitions of resilience and relevant socioecological influences might be influenced by developmental stage. Although it would be ideal to ultimately include participants from a number of international cultural communities in and outside of the U.S., one could pilot this study in a diverse U.S. city that has multiple large ethnic enclaves. The city of Chicago for instance, has enclaves of Polish, Mexican, Puerto Rican, and Chinese communities, among others. A thematic analysis could be performed on participant responses to identify recurrent themes within and across cultures. For the quantitative portion of this study, researchers could develop and administer measures that ask participants to score the extent to which they believe different characteristics are reflective of resilience in their culture and the salience of different socioecological factors in promoting resilience in their culture. This measure could be informed by literature on resilience broadly as 93 well as literature that discusses values and strengths of minoritized cultures in the United States. This approach would allow researchers to quantitatively examine the extent to which current resilience conceptualizations and frequently examined promotive factors resonate with individuals across different cultures and identify culturally salient additions that have been historically omitted from resilience science. A mixed methods design would augment this work by allowing researchers to uncover themes that may not be represented in extant published literature. Another area of resilience science that would benefit from thoughtful consideration of culture and identity concerns the types of adversity that are commonly examined in resilience literature. Certain types of adversity, particularly those that are uniquely relevant to marginalized populations have been largely overlooked. While there is emerging literature regarding resilience amongst marginalized populations, it is typically siloed from mainstream resilience literature and omitted from reviews of youth resilience. One prominent example is discrimination, including micro and macro-aggressions as well as historical structural and systemic discrimination based on race, ethnicity, gender, sex, class, and nationality. These forms of adversity are chronic and pervasive throughout the United States and have been shown to take a marked toll on the well- being of minoritized individuals and communities. Unique promotive factors such as parental racial or ethnic socialization (i.e., messages that parents provide their children regarding perceptions of and interactions with own’s own and other racial groups; Brown & Krishnakumar, 2007), have, however, been demonstrated to buffer against some of these harmful effects (Schires et al., 2020; Wang & Huguley, 2012). Another form of adversity that pertains to marginalized communities is acculturation stress. Acculturation stress, commonly experienced by immigrant populations, refers to stress 94 related to the physical and psychological toll of transitioning to a new culture. One specific form of acculturation stress is intergenerational acculturative conflict, which refers to conflict that arises in a family system due to differences in the cultural values held by family members (typically those from different generations). For instance, immigrant parents and/or grandparents may hold strong values that align with the culture in their home country, while youth born or raised in a different country may hold conflicting values. Youth in multicultural families often experience internal conflict regarding their values and sense of belonging; this experience can have profound effects on youth development (Ward & Szabó, 2023). Scholars who conduct resilience research on diverse populations would thus benefit from including measures that assess exposure to a wider array of adversities, with particular attention paid to those that uniquely affect marginalized populations. Indeed, the impact of most, if not all adversities are compounded by marginalized identities due to additional barriers introduced by historically biased systems/processes and implicitly biased socialization. Given this, it is critically important for resilience scholars to also begin incorporating a consideration of intersectionality. One way to begin incorporating intersectionality would be to include an interaction between identity and promotive factors or adverse exposures into models of youth resilience. Relatedly, it would be important for scholars to be thoughtful about their conceptualizations and language surrounding resilience to adversities related to inequities and bias. Literature on decolonizing mental health emphasizes a shift away from advocating for ‘adaptation’ to oppressive systems so as to not legitimize or normalize these experiences and systems (Phillips et al., 2015). Instead, decolonization literature aligns with liberation theory, empowering individuals to reflect on social systems of oppression and practice ‘resistance’, while also developing ways to cope effectively (Johnson & Friedman, 2014). This theory holds 95 space for the duality of personal and social transformation. Recently, resilience literature along these lines has emerged where resilience is defined as “overcoming adversity, whilst also potentially subtly changing, or even dramatically transforming, (aspects of) that adversity” (Hart et al., 2016; Haynes et al., 2024). Additional work that expands on this conceptualization and more generally integrates postcolonial psychological theories with mainstream resilience literature is greatly needed to advance the inclusivity of resilience theory and research for marginalized populations. In conclusion, there are many opportunities to build on and expand extant resilience literature. The literature base on resilience development is still in its early stages and would benefit greatly from additional longitudinal work. Relatedly, research focusing on the timing and type of adversity would greatly improve our understanding of the intricacies of how and when youth respond to different adversities. Moreover, although the field is slowly gaining insight into the biological mechanisms that underlie youth resilience, work in this area that honors the multidimensionality and dynamic nature of resilience remains limited. Finally, among the most under-studied areas of mainstream resilience science are culture and identity. There is a great need for a bridging of literature on resilience among marginalized populations, postcolonial theory, and resilience literature more broadly. Research that meaningfully incorporates culture and a wider range of adversities into foundational conceptualizations of resilience through into the promotive factors and developmental trajectories of resilience is greatly needed. 96 REFERENCES Achenbach, T. M., & Rescorla, L. A. (2001). Manual for the ASEBA school-age forms & profiles: Child behavior checklist for ages 6-18, teacher’s report form, youth self-report: an integrated system of multi-informant assessment. University of Vermont, research center for children youth & families. Akaike, H. (1987). Factor analysis and AIC. Psychometrika, 52(3), 317–332. https://doi.org/10.1007/BF02294359 Alvarado, S. E. (2016). Neighborhood disadvantage and obesity across childhood and adolescence: Evidence from the NLSY children and young adults cohort (1986–2010). Social Science Research, 57, 80–98. https://doi.org/10.1016/j.ssresearch.2016.01.008 Anderson-Butcher, D., Martin, E., Paluta, L., & Gould, D. (2018). Patterns of social skill development over-time among clusters of LiFEsports participants. Children and Youth Services Review, 87, 17–25. https://doi.org/10.1016/j.childyouth.2018.01.044 Andruff, H., Carraro, N., Thompson, A., Gaudreau, P., & Louvet, B. (2009). Latent class growth modeling: A tutorial. Tutorials in Quantitative Methods for Psychology, 5(1), 11–24. https://doi.org/10.20982/tqmp.05.1.p011 Assari, S., & Caldwell, C. H. (2018). Social determinants of perceived discrimination among black youth: Intersection of ethnicity and gender. Children, 5(2), Article 2. https://doi.org/10.3390/children5020024 Bernier, A., Carlson, S. M., & Whipple, N. (2010). From external regulation to self-regulation: Early parenting precursors of young children’s executive functioning. Child Development, 81(1), 326–339. https://doi.org/10.1111/j.1467-8624.2009.01397.x Blakemore, S.-J. (2012). Development of the social brain in adolescence. Journal of the Royal Society of Medicine, 105(3), 111–116. https://doi.org/10.1258/jrsm.2011.110221 Brown, T. L., & Krishnakumar, A. (2007). Development and validation of the adolescent racial and ethnic socialization scale (ARESS) in African American families. Journal of Youth and Adolescence, 36(8), 1072–1085. https://doi.org/10.1007/s10964-007-9197-z Burt, K. B., Whelan, R., Conrod, P. J., Banaschewski, T., Barker, G. J., Bokde, A. L. W., Bromberg, U., Büchel, C., Fauth-Bühler, M., Flor, H., Galinowski, A., Gallinat, J., Gowland, P., Heinz, A., Ittermann, B., Mann, K., Nees, F., Papadopoulos-Orfanos, D., Paus, T., … Consortium, the I. (2016). Structural brain correlates of adolescent resilience. Journal of Child Psychology and Psychiatry, 57(11), 1287–1296. https://doi.org/10.1111/jcpp.12552 Burt, S. A., & Klump, K. L. (2013). The Michigan State University Twin Registry (MSUTR): An update. Twin Research and Human Genetics, 16(1), 344–350. https://doi.org/10.1017/thg.2012.87 97 Burt, S. A., & Klump, K. L. (2019). The Michigan State University Twin Registry (MSUTR): 15 years of twin and family research. Twin Research and Human Genetics, 22(6), 741–745. https://doi.org/10.1017/thg.2019.57 Burt, S. A., Klump, K. L., Gorman-Smith, D., & Neiderhiser, J. M. (2016). Neighborhood disadvantage alters the origins of children’s nonaggressive conduct problems. Clinical Psychological Science, 4(3), 511–526. https://doi.org/10.1177/2167702615618164 Burt, S. A., Klump, K. L., Vazquez, A. Y., Shewark, E. A., & Hyde, L. W. (2021). Identifying patterns of youth resilience to neighborhood disadvantage. Research in Human Development, 18(3), 181–196. https://doi.org/10.1080/15427609.2021.1935607 Burt, S. A., McGue, M., Iacono, W. G., & Krueger, R. F. (2006). Differential parent-child relationships and adolescent externalizing symptoms: Cross-lagged analyses within a monozygotic twin differences design. Developmental Psychology, 42(6), 1289–1298. https://doi.org/10.1037/0012-1649.42.6.1289 Burt, S. A., Pearson, A. L., Rzotkiewicz, A., Klump, K. L., & Neiderhiser, J. M. (2019). It really does take a village: The role of neighbors in the etiology of nonaggressive rule-breaking behavior. Development and Psychopathology, 31(2), 713–725. https://doi.org/10.1017/S0954579418000366 Buttelmann, F., & Karbach, J. (2017). Development and plasticity of cognitive flexibility in early and middle childhood. Frontiers in Psychology, 8. https://doi.org/10.3389/fpsyg.2017.01040 Bye, C. R., Rytova, V., Alsanie, W. F., Parish, C. L., & Thompson, L. H. (2019). Axonal growth of midbrain dopamine neurons is modulated by the cell adhesion molecule ALCAM through trans-heterophilic interactions with L1cam, Chl1, and semaphorins. Journal of Neuroscience, 39(34), 6656–6667. https://doi.org/10.1523/JNEUROSCI.0278-19.2019 Campbell, S. B., Shaw, D. S., & Gilliom, M. (2000). Early externalizing behavior problems: Toddlers and preschoolers at risk for later maladjustment. Development and Psychopathology, 12(3), 467–488. https://doi.org/10.1017/S0954579400003114 Carleial, S., Nätt, D., Unternährer, E., Elbert, T., Robjant, K., Wilker, S., Vukojevic, V., Kolassa, I.-T., Zeller, A. C., & Koebach, A. (2021). DNA methylation changes following narrative exposure therapy in a randomized controlled trial with female former child soldiers. Scientific Reports, 11(1), 18493. https://doi.org/10.1038/s41598-021-98067-9 Carr, A. (2011). Social and emotional development in middle childhood. In Child Psychology and Psychiatry (pp. 56–61). John Wiley & Sons, Ltd. https://doi.org/10.1002/9781119993971.ch10 Carskadon, M. A., & Acebo, C. (1993). A self-administered rating scale for pubertal development. Journal of Adolescent Health, 14(3), 190–195. https://doi.org/10.1016/1054-139X(93)90004-9 98 Colich, N. L., & McLaughlin, K. A. (2022). Accelerated pubertal development as a mechanism linking trauma exposure with depression and anxiety in adolescence. Current Opinion in Psychology, 46, 101338. https://doi.org/10.1016/j.copsyc.2022.101338 Collins, L. M., & Lanza, S. T. (2009). Latent class and latent transition analysis: With applications in the social, behavioral, and health sciences. John Wiley & Sons. Costello, E. J., Copeland, W., & Angold, A. (2011). Trends in psychopathology across the adolescent years: What changes when children become adolescents, and when adolescents become adults? Journal of Child Psychology and Psychiatry, 52(10), 1015– 1025. https://doi.org/10.1111/j.1469-7610.2011.02446.x Cubbin, C., Sundquist, K., Ahlén, H., Johansson, S.-E., Winkleby, M. A., & Sundquist, J. (2006). Neighborhood deprivation and cardiovascular disease risk factors: Protective and harmful effects. Scandinavian Journal of Public Health, 34(3), 228–237. Denise, E. J. (2012). Multiple forms of perceived discrimination and health among adolescents and young adults. Journal of Health and Social Behavior, 53(2), 199–214. https://doi.org/10.1177/0022146512444289 Desprez, F., Ung, D. C., Vourc’h, P., Jeanne, M., & Laumonnier, F. (2023). Contribution of the dihydropyrimidinase-like proteins family in synaptic physiology and in neurodevelopmental disorders. Frontiers in Neuroscience, 17. https://doi.org/10.3389/fnins.2023.1154446 Diez Roux, A. V., & Mair, C. (2010). Neighborhoods and health. Annals of the New York Academy of Sciences, 1186(1), 125–145. https://doi.org/10.1111/j.1749- 6632.2009.05333.x Dubowitz, H., Thompson, R., Proctor, L., Metzger, R., Black, M. M., English, D., Poole, G., & Magder, L. (2016). Adversity, maltreatment, and resilience in young children. Academic Pediatrics, 16(3), 233–239. https://doi.org/10.1016/j.acap.2015.12.005 Dudek, K. A., Dion-Albert, L., Lebel, M., LeClair, K., Labrecque, S., Tuck, E., Ferrer Perez, C., Golden, S. A., Tamminga, C., Turecki, G., Mechawar, N., Russo, S. J., & Menard, C. (2020). Molecular adaptations of the blood–brain barrier promote stress resilience vs. depression. Proceedings of the National Academy of Sciences, 117(6), 3326–3336. https://doi.org/10.1073/pnas.1914655117 Duncan, T. E., Duncan, S. C., & Strycker, L. A. (2006). An introduction to latent variable growth curve modeling: Concepts, issues, and applications (2nd ed.). Routledge. https://doi.org/10.4324/9780203879962 Elkins, I. J., McGue, M., & Iacono, W. G. (1997). Genetic and environmental influences on parent–son relationships: Evidence for increasing genetic influence during adolescence. Developmental Psychology, 33(2), 351–363. https://doi.org/10.1037/0012-1649.33.2.351 99 Elliott, E., Ezra-Nevo, G., Regev, L., Neufeld-Cohen, A., & Chen, A. (2010). Resilience to social stress coincides with functional DNA methylation of the Crf gene in adult mice. Nature Neuroscience, 13(11), Article 11. https://doi.org/10.1038/nn.2642 Evans, B., Zimmerman, E., Woolf, S., & Haley, A. (2012). Social Determinants of Health and Crime in Post-Katrina Orleans Parish. Richmond, VA: Virginia Commonwealth University, Center on Human Needs. Feder, A., Fred-Torres, S., Southwick, S. M., & Charney, D. S. (2019). The biology of human resilience: Opportunities for enhancing resilience across the life span. Biological Psychiatry, 86(6), 443–453. https://doi.org/10.1016/j.biopsych.2019.07.012 Fraga, M. F., Ballestar, E., Paz, M. F., Ropero, S., Setien, F., Ballestar, M. L., Heine-Suñer, D., Cigudosa, J. C., Urioste, M., Benitez, J., Boix-Chornet, M., Sanchez-Aguilera, A., Ling, C., Carlsson, E., Poulsen, P., Vaag, A., Stephan, Z., Spector, T. D., Wu, Y.-Z., … Esteller, M. (2005). Epigenetic differences arise during the lifetime of monozygotic twins. Proceedings of the National Academy of Sciences, 102(30), 10604–10609. https://doi.org/10.1073/pnas.0500398102 Galster, G. C. (2018). Nonlinear and threshold effects related to neighborhood: Implications for planning and policy. Journal of Planning Literature, 33(4), 492–508. https://doi.org/10.1177/0885412218793693 Haley, A., Zimmerman, E., Woolf, S., & Evans, B. (2012). Neighborhood-Level Determinants of Life Expectancy in Oakland, CA. Virginia Commonwealth University, Center on Human Needs. Han, L., Zhang, H., Kaushal, A., Rezwan, F. I., Kadalayil, L., Karmaus, W., Henderson, A. J., Relton, C. L., Ring, S., Arshad, S. H., Ewart, S. L., & Holloway, J. W. (2019). Changes in DNA methylation from pre- to post-adolescence are associated with pubertal exposures. Clinical Epigenetics, 11(1), 176. https://doi.org/10.1186/s13148-019-0780-4 Hart, A., Gagnon, E., Eryigit-Madzwamuse, S., Cameron, J., Aranda, K., Rathbone, A., & Heaver, B. (2016). Uniting resilience research and practice with an inequalities approach. Sage Open, 6(4), 2158244016682477. https://doi.org/10.1177/2158244016682477 Haynes, P., Hart, A., Eryigit-Madzwamuse, S., Wood, M., Maitland, J., & Cameron, J. (2024). The contribution of a complex systems-based approach to progressive social resilience. Health, 28(5), 754–774. https://doi.org/10.1177/13634593231195784 Heck, A., Milnik, A., Vukojevic, V., Petrovska, J., Egli, T., Singer, J., Escobar, P., Sengstag, T., Coynel, D., Freytag, V., Fastenrath, M., Demougin, P., Loos, E., Hartmann, F., Schicktanz, N., Delarue Bizzini, B., Vogler, C., Kolassa, I.-T., Wilker, S., … Papassotiropoulos, A. (2017). Exome sequencing of healthy phenotypic extremes links TROVE2 to emotional memory and PTSD. Nature Human Behaviour, 1(4), 1–10. https://doi.org/10.1038/s41562-017-0081 100 Henry, D., Gorman-Smith, D., Schoeny, M., & Tolan, P. (2014). “Neighborhood Matters”: Assessment of neighborhood social processes. American Journal of Community Psychology, 54(3), 187–204. https://doi.org/10.1007/s10464-014-9681-z Iadipaolo, A. S., Marusak, H. A., Paulisin, S. M., Sala-Hamrick, K., Crespo, L. M., Elrahal, F., Peters, C., Brown, S., & Rabinak, C. A. (2018). Distinct neural correlates of trait resilience within core neurocognitive networks in at-risk children and adolescents. NeuroImage: Clinical, 20, 24–34. https://doi.org/10.1016/j.nicl.2018.06.026 Jiang, S., Tang, L., Zhao, N., Yang, W., Qiu, Y., & Chen, H.-Z. (2016). A Systems View of the Differences between APOE ε4 Carriers and Non-carriers in Alzheimer’s Disease. Frontiers in Aging Neuroscience, 8. https://doi.org/10.3389/fnagi.2016.00171 Johnson, C. V., & Friedman, H. L. (2014). The Praeger handbook of social justice and psychology: [3 volumes]. Bloomsbury Publishing USA. Jutte, D. P., Miller, J. L., & Erickson, D. J. (2015). Neighborhood adversity, child health, and the role for community development. Pediatrics, 135(Supplement_2), S48–S57. https://doi.org/10.1542/peds.2014-3549F Kim-Cohen, J., Moffitt, T. E., Caspi, A., & Taylor, A. (2004). Genetic and environmental processes in young children’s resilience and vulnerability to socioeconomic deprivation. Child Development, 75(3), 651–668. https://doi.org/10.1111/j.1467-8624.2004.00699.x Kind, A. J. H., & Buckingham, W. R. (2018). Making neighborhood-disadvantage metrics accessible—the neighborhood atlas. The New England Journal of Medicine, 378(26), 2456–2458. https://doi.org/10.1056/NEJMp1802313 Kline, R. B. (2023). Principles and practice of structural equation modeling. Guilford Publications. Lander, E., & Kruglyak, L. (1995). Genetic dissection of complex traits: Guidelines for interpreting and reporting linkage results. Nature Genetics, 11(3), 241–247. https://doi.org/10.1038/ng1195-241 Lang, K. M., & Little, T. D. (2018). Principled missing data treatments. Prevention Science, 19(3), 284–294. https://doi.org/10.1007/s11121-016-0644-5 Lavizzo-Mourey, R. (2012). Why health, poverty, and community development are inseparable. Investing in what works for America’s communities: Essays on people, place & purpose, Edited by Nancy O. Andrews and David J. Erickson, 215–225. Lécuyer, M.-A., Saint-Laurent, O., Bourbonnière, L., Larouche, S., Larochelle, C., Michel, L., Charabati, M., Abadier, M., Zandee, S., Haghayegh Jahromi, N., Gowing, E., Pittet, C., Lyck, R., Engelhardt, B., & Prat, A. (2017). Dual role of ALCAM in neuroinflammation and blood–brain barrier homeostasis. Proceedings of the National Academy of Sciences, 114(4), E524–E533. https://doi.org/10.1073/pnas.1614336114 101 Lee, H., Joo, J., Nah, S.-S., Kim, J. W., Kim, H.-K., Kwon, J.-T., Lee, H.-Y., Kim, Y. O., & Kim, H.-J. (2015). Changes in Dpysl2 expression are associated with prenatally stressed rat offspring and susceptibility to schizophrenia in humans. International Journal of Molecular Medicine, 35(6), 1574–1586. https://doi.org/10.3892/ijmm.2015.2161 Lu, A. K.-M., Hsieh, S., Yang, C.-T., Wang, X.-Y., & Lin, S.-H. (2023). DNA methylation signature of psychological resilience in young adults: Constructing a methylation risk score using a machine learning method. Frontiers in Genetics, 13. https://www.frontiersin.org/articles/10.3389/fgene.2022.1046700 Luthar, S. S. (2006a). Resilience in development: A synthesis of research across five decades. In Developmental psychopathology: Risk, disorder, and adaptation, Vol. 3, 2nd ed (pp. 739– 795). John Wiley & Sons, Inc. Luthar, S. S., Grossman, E. J., & Small, P. J. (2015). Resilience and adversity. In Handbook of child psychology and developmental science: Socioemotional processes, Vol. 3, 7th ed (pp. 247–286). John Wiley & Sons, Inc. https://doi.org/10.1002/9781118963418.childpsy307 Mansell, G., Gorrie-Stone, T. J., Bao, Y., Kumari, M., Schalkwyk, L. S., Mill, J., & Hannon, E. (2019). Guidance for DNA methylation studies: Statistical insights from the Illumina EPIC array. BMC Genomics, 20(1), 366. https://doi.org/10.1186/s12864-019-5761-7 Marshall, A. D. (2016). Developmental timing of trauma exposure relative to puberty and the nature of psychopathology among adolescent girls. Journal of the American Academy of Child & Adolescent Psychiatry, 55(1), 25-32. https://doi.org/10.1016/j.jaac.2015.10.004 Masten, A. S. (2001). Ordinary magic. Resilience processes in development. The American Psychologist, 56(3), 227–238. https://doi.org/10.1037//0003-066x.56.3.227 Masten, A. S., & Cicchetti, D. (2016). Resilience in development: Progress and transformation. In Developmental psychopathology: Risk, resilience, and intervention, Vol. 4, 3rd ed (pp. 271–333). John Wiley & Sons, Inc. https://doi.org/10.1002/9781119125556.devpsy406 Masten, A. S., & Curtis, W. J. (2000). Integrating competence and psychopathology: Pathways toward a comprehensive science of adaptation in development. Development and Psychopathology, 12(3), 529–550. https://doi.org/10.1017/S095457940000314X Masten, A. S., Hubbard, J. J., Gest, S. D., Tellegen, A., Garmezy, N., & Ramirez, M. (1999). Competence in the context of adversity: Pathways to resilience and maladaptation from childhood to late adolescence. Development and Psychopathology, 11(1), 143–169. https://doi.org/10.1017/S0954579499001996 McGregor, K., Bernatsky, S., Colmegna, I., Hudson, M., Pastinen, T., Labbe, A., & Greenwood, C. M. (2016). An evaluation of methods correcting for cell-type heterogeneity in DNA methylation studies. Genome Biology, 17, 1-17. https://doi.org/10.1186/s13059-016- 0935-y 102 Milaniak, I., Cecil, C. A. M., Barker, E. D., Relton, C. L., Gaunt, T. R., McArdle, W., & Jaffee, S. R. (2017). Variation in DNA methylation of the oxytocin receptor gene predicts children’s resilience to prenatal stress. Development and Psychopathology, 29(5), 1663– 1674. https://doi.org/10.1017/S0954579417001316 Miller, O., Shakespeare-Finch, J., Bruenig, D., & Mehta, D. (2020). DNA methylation of NR3C1 and FKBP5 is associated with posttraumatic stress disorder, posttraumatic growth, and resilience. Psychological Trauma: Theory, Research, Practice, and Policy, 12(7), 750– 755. http://dx.doi.org.proxy2.cl.msu.edu/10.1037/tra0000574 Miller-Graff, L. E. (2022). The Multidimensional Taxonomy of Individual Resilience. Trauma, Violence, & Abuse, 23(2), 660–675. https://doi.org/10.1177/1524838020967329 Morey, R. A., Haswell, C. C., Hooper, S. R., & De Bellis, M. D. (2016). Amygdala, hippocampus, and ventral medial prefrontal cortex volumes differ in maltreated youth with and without chronic Posttraumatic Stress Disorder. Neuropsychopharmacology, 41(3), Article 3. https://doi.org/10.1038/npp.2015.205 Muthén, L. K., & Muthén, B. O. (1998-2017). Mplus user’s guide. 8th Edition. Los Angeles, CA: Muthén & Muthén. Neblett Jr., E. W., Rivas-Drake, D., & Umaña-Taylor, A. J. (2012). The promise of racial and ethnic protective factors in promoting ethnic minority youth development. Child Development Perspectives, 6(3), 295–303. https://doi.org/10.1111/j.1750- 8606.2012.00239.x Notterman, D. A., & Mitchell, C. (2015). Epigenetics and understanding the impact of social determinants of health. Pediatric Clinics of North America, 62(5), 1227–1240. https://doi.org/10.1016/j.pcl.2015.05.012 Nylund, K. L., Asparouhov, T., & Muthén, B. O. (2007). Deciding on the number of classes in latent class analysis and growth mixture modeling: A Monte Carlo simulation study. Structural Equation Modeling: A Multidisciplinary Journal, 14(4), 535–569. https://doi.org/10.1080/10705510701575396 Oshri, A., Topple, T. A., & Carlson, M. W. (2017). Positive youth development and resilience: growth patterns of social skills among youth investigated for maltreatment. Child Development, 88(4), 1087–1099. https://doi.org/10.1111/cdev.12865 Özcan, B., & Bulus, M. (2022). Protective factors associated with academic resilience of adolescents in individualist and collectivist cultures: Evidence from PISA 2018 large scale assessment. Current Psychology, 41(4), 1740–1756. https://doi.org/10.1007/s12144- 022-02944-z Peeters, H., Van Gestel, S., Vlietinck, R., Derom, C., & Derom, R. (1998). Validation of a telephone zygosity questionnaire in twins of known zygosity. Behavior Genetics, 28(3), 159–163. https://doi.org/10.1023/A:1021416112215 103 Pham, X., Song, G., Lao, S., Goff, L., Zhu, H., Valle, D., & Avramopoulos, D. (2016). The DPYSL2 gene connects mTOR and schizophrenia. Translational Psychiatry, 6(11), e933– e933. https://doi.org/10.1038/tp.2016.204 Phillips, N. L., Adams, G., & Salter, P. S. (2015). Beyond adaptation: Decolonizing approaches to coping with oppression. Journal of Social and Political Psychology, 3(1), Article 1. https://doi.org/10.5964/jspp.v3i1.310 Piña-Watson, B., Gonzalez, I. M., & Manzo, G. (2019). Mexican-descent adolescent resilience through familismo in the context of intergeneration acculturation conflict on depressive symptoms. Translational Issues in Psychological Science, 5(4), 326–334. https://doi.org/10.1037/tps0000210 Prince, D. M., Rocha, A., & Nurius, P. S. (2018). Multiple disadvantage and discrimination: implications for adolescent health and education. Social Work Research, 42(3), 169–179. https://doi.org/10.1093/swr/svy016 Raftery, A. E. (1995). Bayesian model selection in social research. Sociological Methodology, 25, 111–163. https://doi.org/10.2307/271063 Raposa, E. B., Hammen, C. L., Brennan, P. A., O’Callaghan, F., & Najman, J. M. (2014). Early adversity and health outcomes in young adulthood: The role of ongoing stress. Health Psychology, 33(5), 410–418. https://doi.org/10.1037/a0032752 Raymond, C., Marin, M.-F., Majeur, D., & Lupien, S. (2018). Early child adversity and psychopathology in adulthood: HPA axis and cognitive dysregulations as potential mechanisms. Progress in Neuro-Psychopharmacology and Biological Psychiatry, 85, 152–160. https://doi.org/10.1016/j.pnpbp.2017.07.015 Sampson, R. J., Raudenbush, S. W., & Earls, F. (1997). Neighborhoods and violent crime: A multilevel study of collective efficacy. Science, 277(5328), 918–924. https://doi.org/10.1126/science.277.5328.918 Sattler, K., Yoon, S., & Lutolli, A. (2023). Trajectories of resilience among young children involved with child protective services. Development and Psychopathology, 36(1), 467- 477. https://doi.org/10.1017/S095457942200133X Schires, S. M., Buchanan, N. T., Lee, R. M., McGue, M., Iacono, W. G., & Burt, S. A. (2020). Discrimination and ethnic-racial socialization among youth adopted from South Korea into White American families. Child Development, 91(1), e42–e58. https://doi.org/10.1111/cdev.13167 Sclove, S. L. (1987). Application of model-selection criteria to some problems in multivariate analysis. Psychometrika, 52(3), 333–343. https://doi.org/10.1007/BF02294360 Smith, J. A., Zhao, W., Wang, X., Ratliff, S. M., Mukherjee, B., Kardia, S. L. R., Liu, Y., Roux, A. V. D., & Needham, B. L. (2017). Neighborhood characteristics influence DNA methylation of genes involved in stress response and inflammation: The multi-ethnic 104 study of atherosclerosis. Epigenetics, 12(8), 662–673. https://doi.org/10.1080/15592294.2017.1341026 Stewart, D., & Sun, J. (2004). How can we build resilience in primary school aged children? The importance of social support from adults and peers in family, school and community settings. Asia Pacific Journal of Public Health, 16(S1), 37–41. https://doi.org/10.1177/101053950401600S10 Su, S., Hao, G., Youssef, N. A., Stefanek, M., Treiber, F., Wang, X., Pollock, J., & Harshfield, G. (2019). DNA methylation as a mediator of the association between stress and depression: A genome-wide DNA methylation study. The FASEB Journal, 33(S1), 593.14-593.14. https://doi.org/10.1096/fasebj.2019.33.1_supplement.593.14 Sun, H., Kennedy, P. J., & Nestler, E. J. (2013). Epigenetics of the depressed brain: Role of histone acetylation and methylation. Neuropsychopharmacology, 38(1), Article 1. https://doi.org/10.1038/npp.2012.73 Szyf, M., Weaver, I. C. G., Champagne, F. A., Diorio, J., & Meaney, M. J. (2005). Maternal programming of steroid receptor expression and phenotype through DNA methylation in the rat. Frontiers in Neuroendocrinology, 26(3), 139–162. https://doi.org/10.1016/j.yfrne.2005.10.002 Taylor, D. (2014). Toxic communities: Environmental racism, industrial pollution, and residential mobility. In Toxic communities. New York University Press. https://doi.org/10.18574/nyu/9781479805150.001.0001 Vanderbilt-Adriance, E., & Shaw, D. S. (2008a). Conceptualizing and re-evaluating resilience across levels of risk, time, and domains of competence. Clinical Child and Family Psychology Review, 11(1), 30–58. Vanderbilt-Adriance, E., & Shaw, D. S. (2008b). Protective factors and the development of resilience in the context of neighborhood disadvantage. Journal of Abnormal Child Psychology, 36(6), 887–901. https://doi.org/10.1007/s10802-008-9220-1 Vazquez, A. Y., Burt, S. A., Mitchell, C., Klump, K. L., Hyde, L. W., & Clark, S. L. (2024). Methylome-wide association study of multidimensional resilience. Development and Psychopathology, 1–12. https://doi.org/10.1017/S0954579424001330 Vazquez, A. Y., Shewark, E. A., Clark, D. A., Klump, K. L., Hyde, L. W., & Burt, S. A. (2021). The etiology of resilience to disadvantage. JCPP Advances, 1(3), e12033. https://doi.org/10.1002/jcv2.12033 Vazquez, A. Y., Shewark, E. A., Hyde, L. W., Klump, K. L., & Burt, S. A. (2024). Parental nurturance moderates the etiology of youth resilience. Behavior genetics, 54(1), 137-149. https://doi.org/10.1007/s10519-023-10150-1 Vilarinho, S., Sari, S., Mazzacuva, F., Bilgüvar, K., Esendagli-Yilmaz, G., Jain, D., Akyol, G., Dalgiç, B., Günel, M., Clayton, P. T., & Lifton, R. P. (2016). ACOX2 deficiency: A 105 disorder of bile acid synthesis with transaminase elevation, liver fibrosis, ataxia, and cognitive impairment. Proceedings of the National Academy of Sciences, 113(40), 11289– 11293. https://doi.org/10.1073/pnas.1613228113 Wang, M.-T., & Huguley, J. P. (2012). Parental racial socialization as a moderator of the effects of racial discrimination on educational success among African American adolescents. Child Development, 83(5), 1716–1731. https://doi.org/10.1111/j.1467-8624.2012.01808.x Ward, C., & Szabó, Á. (2023). Acculturation, cultural identity and well-being. Nature Reviews Psychology, 2(5), 267–282. https://doi.org/10.1038/s44159-023-00171-2 Weaver, I. C. G., Cervoni, N., Champagne, F. A., D’Alessio, A. C., Sharma, S., Seckl, J. R., Dymov, S., Szyf, M., & Meaney, M. J. (2004). Epigenetic programming by maternal behavior. Nature Neuroscience, 7(8), Article 8. https://doi.org/10.1038/nn1276 Werner, E. E. (2012). Children and war: Risk, resilience, and recovery. Development and Psychopathology, 24(2), 553–558. https://doi.org/10.1017/S0954579412000156 Winslow, E. B., & Shaw, D. S. (2007). Impact of neighborhood disadvantage on overt behavior problems during early childhood. Aggressive Behavior, 33(3), 207–219. https://doi.org/10.1002/ab.20178 Wodtke, G. T., Harding, D. J., & Elwert, F. (2011). Neighborhood effects in temporal perspective: The impact of long-term exposure to concentrated disadvantage on high school graduation. American Sociological Review, 76(5), 713–736. https://doi.org/10.1177/0003122411420816 Wu, M. S., Yan, X., Zhou, C., Chen, Y., Li, J., Zhu, Z., Shen, X., & Han, B. (2011). General belief in a just world and resilience: Evidence from a collectivistic culture. European Journal of Personality, 25(6), 431–442. https://doi.org/10.1002/per.807 Yoon, S., Maguire-Jack, K., Knox, J., & Ploss, A. (2021). Socio-ecological predictors of resilience development over time among youth with a history of maltreatment. Child Maltreatment, 26(2), 162–171. https://doi.org/10.1177/1077559520981151 Yoon, S., Sattler, K., Knox, J., & Xin, Y. (2021). Longitudinal examination of resilience among child welfare-involved adolescents: The roles of caregiver–child relationships and deviant peer affiliation. Development and Psychopathology, 1–10. https://doi.org/10.1017/S0954579421000924 Zeileis, A. (2006). Object-oriented computation of sandwich estimators. Journal of Statistical Software, 16, 1–16. https://doi.org/10.18637/jss.v016.i09 Zhang, T.-Y., Hellstrom, I. C., Bagot, R. C., Wen, X., Diorio, J., & Meaney, M. J. (2010). Maternal care and DNA methylation of a glutamic acid decarboxylase 1 promoter in rat hippocampus. Journal of Neuroscience, 30(39), 13130–13137. https://doi.org/10.1523/JNEUROSCI.1039-10.2010 106 Zhou, Y., Zhou, B., Pache, L., Chang, M., Khodabakhshi, A. H., Tanaseichuk, O., Benner, C., & Chanda, S. K. (2019). Metascape provides a biologist-oriented resource for the analysis of systems-level datasets. Nature Communications, 10(1), 1523. https://doi.org/10.1038/s41467-019-09234-6 Zimmerman, A. W., Joosten, B., Torensma, R., Parnes, J. R., van Leeuwen, F. N., & Figdor, C. G. (2006). Long-term engagement of CD6 and ALCAM is essential for T-cell proliferation induced by dendritic cells. Blood, 107(8), 3212–3220. https://doi.org/10.1182/blood-2005-09-3881 107 APPENDIX Table S1. Conditional Growth Curve Model with only Covariates Outcome Parameter Estimate S.E. p-value Conditional Model Psychological Intercept Resilience Mean Variance Ethnicity Sex Slope Mean Variance Ethnicity Sex Time Invariant Covariates Puberty2 Puberty3 102.874 219.438 3.030 4.922 1.215 1.510 -0.495 -0.842 0.969 1.482 2.135 57.627 2.204 1.608 0.280 0.574 0.215 0.164 0.470 0.530 <.001 <.001 0.169 0.002 <.001 0.009 0.021 <.001 0.039 0.005 Social Resilience Intercept-Slope Covariance -11.330 5.629 0.044 Model fit Statistics AIC BIC saBIC Intercept Mean Variance Ethnicity Sex Slope Mean Variance Ethnicity Sex 19455.062 19544.022 19483.686 6.450 1.197 0.755 0.107 0.190 0.016 -0.015 -0.029 0.293 0.856 0.305 0.232 0.062 0.011 0.045 0.032 <.001 0.162 0.013 0.645 0.002 0.138 0.731 0.362 108 Table S1 (cont’d) Social Resilience Time Invariant Covariates Puberty2 Puberty3 Intercept-Slope Covariance Model fit Statistics 2.735 26.195 0.066 0.765 0.782 0.089 <.001 <.001 0.457 AIC BIC saBIC 10859.064 10948.024 10887.688 Note. Abbreviations: SE = Standard Error, AIC = Akaike information criterion, BIC = Bayesian information criterion, saBIC = sample-size adjusted Bayesian information criterion. Gender= 1(female), Ethnicity = 1(White). 109 Table S2. Growth Curve Model Sensitivity Analysis Outcome Parameter Social Intercept Resilience Mean Variance Ethnicity Sex Parental Nurturance Neigh. Soc. Cohesion Neigh. Inf. Soc. Control Slope Mean Variance Ethnicity Sex Parental Nurturance Neigh. Soc. Cohesion Neigh. Inf. Soc. Control Time Varying Covariates Puberty2 Puberty3 Unconditional Model Conditional Model Estimate S.E. p-value Estimate S.E. p-value 6.729 1.431 0.174 1.098 <.001 0.193 - - - - - - - - - - - - - - - 0.156 0.016 0.023 0.016 <.001 0.320 - - - - - - - - - - - - - - - - - - - - - 110 -0.905 1.031 0.055 0.108 0.154 0.066 -0.202 -0.194 0.021 -0.019 -0.070 0.006 -0.003 0.031 0.014 -0.100 2.361 1.123 0.378 0.305 0.063 0.019 0.081 0.377 0.016 0.054 0.046 0.009 0.003 0.012 0.701 0.359 0.884 0.724 0.014 0.001 0.013 0.607 0.194 0.729 0.128 0.533 0.316 0.011 0.169 0.183 0.932 0.584 Table S2 (cont’d) Social Intercept-Slope Covariance 0.036 0.118 0.763 0.024 0.122 0.844 Resilience Model fit Statistics AIC BIC saBIC Psychological Intercept Resilience Mean Variance Ethnicity Sex Parental Nurturance Neigh. Soc. Cohesion Neigh. Inf. Soc. Control Slope Mean Variance Ethnicity Sex Parental Nurturance Neigh. Soc. Cohesion Neigh. Inf. Soc. Control 4085.118 4116.929 4091.545 5027.373 5125.075 5045.759 104.735 1.578 <.001 27.960 25.552 0.274 368.833 101.117 <.001 339.677 95.427 <.001 - - - - - - - - - - - - - - - 1.102 2.608 0.159 0.946 <.001 0.006 - - - - - - - - - - - - - - - 111 0.817 8.294 1.553 0.331 -0.768 -0.903 2.418 -0.369 -1.197 0.024 -0.009 0.141 2.936 2.629 0.619 0.201 0.771 2.530 0.883 0.287 0.265 0.062 0.018 0.075 0.781 0.002 0.012 0.101 0.319 0.721 0.006 0.200 <.001 0.702 0.612 0.059 Table S2 (cont’d) Psychological Time Varying Covariates Resilience Puberty2 Puberty3 - - - - - - 1.522 2.004 0.711 0.768 0.032 0.009 Intercept-Slope Covariance -24.015 9.475 0.011 -23.235 9.001 0.010 Model fit Statistics AIC BIC 8535.130 8566.941 9114.879 9212.581 8541.557 Note. Abbreviations: SE = Standard Error, AIC = Akaike information criterion, BIC = Bayesian information criterion, saBIC = sample-size adjusted Bayesian information criterion. Gender= 1(female), Ethnicity = 1(White). 9133.266 saBIC 112 Variance I Variance S Mean I Mean S Mean I Mean S Ethnicity Sex Table S3. Social Resilience Growth Mixture Models Sensitivity Analyses Parameter Estimate S.E. p-value Intercept-Slope Covariance -0.031 0.111 0.779 0.479 1.055 0.650 0.009 0.013 0.518 Class 1 of 2: High, Increasing (56%) 7.554 0.298 <.001 0.226 0.050 <.001 Class 2 of 2: High, Stable (44%) Increasing vs. Stable 5.680 0.476 <.001 0.066 0.049 0.176 0.077 0.450 0.864 -0.239 0.401 0.552 0.241 0.078 0.002 Parental Nurturance Neighborhood Social Cohesion 0.054 0.025 0.030 Neighborhood Informal Social Control -0.135 0.104 0.194 Intercept -10.893 3.032 <.001 Note. The results presented here are in logits. I = Intercept, S = Slope. Gender= 1(Female); Ethnicity = 1(White). 113 Table S4. Psychological Resilience Growth Mixture Models Sensitivity Analyses Parameter Estimate S.E. p-value Intercept-Slope Covariance -5.880 3.913 0.133 Variance I Variance S Mean I Mean S Mean I Mean S Mean I Mean S Ethnicity Sex Intercept Ethnicity Sex 120.609 42.746 0.005 0.479 0.412 0.245 Class 1 of 3: High Decreasing (7.7%) 108.463 5.118 <.001 -2.273 0.466 <.001 Class 2 of 3: Moderate Increasing (9.8%) 53.617 8.665 <.001 4.969 0.818 <.001 Class 3 of 3: High Increasing (82.5%) 110.276 1.656 <.001 0.976 0.138 <.001 High Decreasing vs. Moderate Increasing (reference) 0.019 0.797 0.981 2.629 0.807 0.001 -0.110 0.137 0.422 Parental Nurturance Neighborhood Social Cohesion 0.043 0.045 0.338 Neighborhood Informal Social Control -0.330 0.154 0.031 High Decreasing vs. High Increasing (reference) 4.294 5.370 0.424 0.688 0.711 0.333 1.727 0.772 0.250 -0.352 0.140 0.012 Parental Nurturance Neighborhood Social Cohesion -0.017 0.041 0.676 Neighborhood Informal Social Control -0.155 0.163 0.341 Intercept 13.161 6.155 0.032 114 Table S4 (cont’d) Moderate Increasing vs. High Increasing (reference) Ethnicity Sex Parental Nurturance 0.670 0.606 0.269 -0.902 0.502 0.073 -0.242 0.076 0.001 Neighborhood Social Cohesion -0.061 0.031 0.052 Neighborhood Informal Social Control 0.175 0.156 0.261 Intercept 8.868 3.473 0.011 Note. The results presented here are in logits. I = Intercept, S = Slope. Gender= 1(Female); Ethnicity = 1(White). 115 Social Resilience (SA) Class 1: high, increasing (56%) Class 2: high, stable (44%) e c n e i l i s e R l a i c o S 12 10 8 6 4 2 0 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 Age Figure S1. Sensitivity Analyses for Social Resilience Growth Mixture Model Psychological Resilience (SA) e c n e i l i s e R l a c i g o l o h c y s P 140 120 100 80 60 40 20 0 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 Age Class 1: high, decreasing (7.7%) Class 2: moderate, increasing (9.8%) Class 3: high, increasing (82.5%) Figure S2. Sensitivity Analyses for Psychological Resilience Growth Mixture Model 116