7

Brain Health Across the Life Span

This chapter focuses on brain health throughout the life span, with respect to typical brain development as well as the development of psychiatric disorders. The session explored how brain health and resilience change across the life span and how researchers have measured these changes. Presenters and panelists also discussed the signals that changes in vulnerabilities and opportunities can provide about brain health and resilience at various life stages. An overview of early adversity, emotional processing, and the neural bases of psychiatric illness was provided by Deanna Barch, chair and professor of psychological and brain sciences, professor of radiology, and Gregory B. Couch Professor of Psychiatry at Washington University in St. Louis. Nim Tottenham, professor in the Department of Psychology at Columbia University, looked at the effect of early-life stress on neurodevelopment. Ted Satterthwaite, assistant professor in the Department of Psychiatry at the University of Pennsylvania School of Medicine, examined how the integration of complex and personalized data can be used to understand normal and abnormal brain network development. Brain network aging and health across the adult life span was described by Gagan Wig, associate professor of behavioral and brain sciences at the Center for Vital Longevity at the University of Texas at Dallas.

EARLY ADVERSITY, EMOTIONAL PROCESSING, AND THE NEURAL BASES OF PSYCHIATRIC ILLNESS

Barch’s presentation explored sensitive periods in which environmental influences have particularly strong relationships to brain health. Studying how these influences affect mental health later in life can shed light on the temporality of those influences—meaning, how early in the process of brain development they have an effect. Many of the factors that appear to be critical seem to be emerging earlier and earlier, she

noted, which underlines the importance of investigating neonatal maternal health. Environmental factors that are not consistent with the expected input at specific developmental stages—such as lack of parental support early in life—might be most detrimental to brain health (Gabard-Durnam and McLaughlin, 2019). Similarly, the presence of input that should not be happening at a given time, owing to various types of adversity, may also be very damaging at certain periods of life. This likely interacts with what is happening in the brain during these periods. Myelination and experience-dependent processes, such as pruning,¹ can vary by brain region. This suggests that there is developmental specificity to the effects of adversity and nurturance on mental health; these effects vary by region and may be differently susceptible at various points in development. Understanding these differences can guide decisions about when to intervene for optimal success.

Maternal Support and Brain Development

Barch focused on maternal support and brain development, with the caveat that paternal support is also important, but good quality measures of paternal support are very limited. A rich body of literature from rodent and nonhuman primate studies clearly demonstrates that the presence of a nurturing caregiver early in life has a powerful effect on hippocampal development and function. This occurs through epigenetic mechanisms that are modulated by various aspects of early caregiving (Fish et al., 2004; Liu et al., 1997; Meaney, 2001; Szyf et al., 2005). Animal models have been able to elucidate some of the causal effects of maternal support because investigators are able to experimentally manipulate this variable.

Carrying out similar research in humans is challenging, because the factors that drive the presence or absence of maternal support could also be contributing to brain development in children through genetic processes or other factors that are difficult to tease apart. However, available data in humans are consistent with the animal study data in suggesting that early experiences of maternal support—or conversely, of abuse, neglect, or adversity—also affect human hippocampal development (Bremner et al., 1997; Driessen et al., 2000; Stein et al., 1997). The hippocampus is a structure that is dense with glucocorticoid receptors and is important in stress regulation and stress modulation through its integral role in the hypothalamic–pituitary–adrenal axis stress response. It has been suggested that reductions and disruptions to hippocampal volume and function lead to maladaptive stress reactivity later in life, which makes it more difficult for a child to engage in appropriate emotion

___________________

¹ A normal developmental process in which the connections between neurons are reduced.

regulation and coping. Later in life, this can contribute to affective psychopathologies such as depression and anxiety (Luby et al., 2016).

Effect of Preschool Maternal Support on the Trajectory of Hippocampal Volume

In humans, it is not yet clear whether there are “sensitive” periods during which parental nurturance to brain development is either more or less important. To explore this question, Barch described a long-term longitudinal study that recruited a sample of around 300 preschoolers between the ages of 3 and 5 years that began in 2002 and is still going on today (Luby et al., 2016). Each year, the participants receive intensive assessments of psychopathology, home factors, and observational, objectively coded measures of structured maternal support and parent–child interaction. Longitudinal neuroimaging began when the children were 7 or 8 years of age (it is currently in its fifth wave), and investigators are still following the participants using a wide range of behavioral assessments.

Neuroimaging data were used to look at the trajectories of hippocampal development in the participants between the ages of roughly 7 and 16 years. A multilevel linear model allows for looking at the entire trajectory of hippocampal volume development across multiple waves, while controlling for factors such as whole-brain gray matter. The investigators also looked at whether the measures of preschool maternal support and school-age maternal support have main effects (i.e., overall hippocampal volume) or interactions over time. The latter are interactions with changes in hippocampal volume over time as children grow. In the preschool age range, they found an upward slope typical of hippocampal volume development in the growth period. The only effect that holds is that of preschool maternal support on the slope of hippocampal volume, which positively predicts children’s self-report management of sad emotions. The greater the preschool maternal support, the steeper the increase in hippocampal volume development over time (see Figure 7-1). This suggests that preschool is a sensitive period for the influence of maternal support on the trajectory of hippocampal development.

The findings for the longitudinal preschool study are consistent with the animal literature suggesting that early maternal support has a positive effect on hippocampal structure and function.² The importance of larger hippocampal volume to emotional regulation was demonstrated when

___________________

² Barch noted that animal models typically do not measure hippocampal volume; they tend to measure more molecular and cellular processes related to hippocampal development—so the human and animal study results do not mirror each other exactly, but they are consistent.

the study participants were in mid-adolescence. The participants with the steepest upward growth of hippocampal volume reported being the most effective at managing their negative emotions. This relationship was supported by both the self-reported and the parent-reported measures of emotion regulation. Steeper hippocampal volume growth was also associated with better episodic memory function in later adolescence.

Adverse Childhood Experiences and Interactions with Maternal Support

The outcome of the trajectory of hippocampal growth on later behaviors can be tied to early maternal support effects, said Barch. Although maternal support is important, many other adversities can occur during early childhood. In theory, these types of adverse childhood experiences (ACEs) could be connected to maternal support; for example, maternal support could be affected by parental psychopathology, also producing an ACE. However, other ACEs are relatively independent of parental behavior, such as poverty and exposure to trauma not perpetrated by parents. Research on the effect

of ACEs on child development has generated good data that early childhood adversity also relates to the structural and functional development of limbic regions including the hippocampus, the amygdala, the basal ganglia, and cortical regions (Carrión et al., 2010; Edmiston et al., 2011; Hanson et al., 2015; McDermott et al., 2019; Rao et al., 2010).

More information is needed on the potentially interactive effects of early childhood adversity and caregiver support, as well as the developmental timing in which these two things have their strongest relationships to brain outcomes. Some evidence suggests that maternal support may have protective effects for children, in the sense that children may be buffered from some of the effects of nonmaternal-related ACEs by having strong maternal support. This could be attributable to a resilience factor, but it could also be the result of the additive contributions that maternal support provides to a child’s brain health and development.³

In recent work, Barch and colleagues analyzed data from a longitudinal study to look for independent or interactive effects of maternal support and ACEs on brain development in preschool- and school-aged children. In this study, they looked at the hippocampus but also looked more broadly at various subcortical and cortical brain regions, including the hippocampus, the amygdala, the subgenual cingulate cortex, and the caudate. Maternal support was assessed using the measure described in the previous section. ACEs were defined as poverty (defined as an income-to-needs ratio of less than 1), traumatic life events,⁴ and parental psychiatric disorders (e.g., suicidality, parental substance use disorder, or other parental psychiatric disorders).⁵ The neuroimaging data were used to estimate the trajectories of hippocampal, amygdala, and caudate volume by preschool-age ACEs and school-age maternal support. This revealed interesting interactions between maternal support and ACEs, with some differential effects on specific developmental periods.

The estimated trajectories of hippocampal volume by preschool ACEs and school-age maternal support show that those two factors interact. For

___________________

³ Barch, D. Workshop presentation—Early Adversity, Emotional Processing, and the Neural Bases of Psychiatric Illness. Available at http://www.nationalacademies.org/hmd/Activities/Aging/BrainHealthAcrossTheLifeSpanWorkshop/2019-JUN-26.aspx (accessed March 12, 2020).

⁴ Traumatic life events included parent arrest; parent hospitalization; crash with motor vehicle, plane, or boat; accidental burning, poisoning, or drowning; attacked by an animal; death of adult loved one; death of sibling or peer; domestic violence; hospitalized, visited emergency department, or had invasive medical procedure; man-made disaster; natural disaster; physical abuse; sexual abuse, sexual assault, or rape; witnessed someone threatened with harm, seriously injured, or killed; physical violence or event causing death or severe harm; or other traumatic life event.

⁵ Barch noted that each of those ACEs could have its own independent effect, but they were aggregated in this study.

school-age children with low maternal support (i.e., one standardization below the mean), no particularly strong differential effect of high versus low preschool ACEs was observed. However, participants with mean school-age maternal support have some differentiation, with children with lower preschool ACEs having a steeper increase in hippocampal volume than children with higher preschool ACEs. At one standard deviation above the mean—the strongest school-age maternal support—there is greater differentiation among the effects of preschool ACEs. This is not a buffering pattern, Barch said. In a buffering pattern, high maternal support would see little effect of preschool ACEs—the effect would only be strong for children with low maternal support. Instead, these results suggest that optimal brain health requires both factors to be present: low preschool ACEs as well as strong maternal support that continues at least into school age.

Estimated trajectories of amygdala volume by preschool ACEs and school-age maternal support show a pattern that is somewhat similar. In participants with the lowest maternal support, there was some differentiation among low ACEs versus high ACEs that becomes stronger in participants with greater maternal support. The strongest differentiation was seen in participants with high school-age maternal support. The largest amygdala volumes were associated with low preschool ACEs and high school-age maternal support, suggesting that both factors need to be present to optimize brain development.

Not every brain region shows the same effect, however. Estimated trajectories of caudate volume by preschool ACEs and preschool maternal support show a different pattern. Participants with many ACEs start out with a smaller caudate volume, but regardless of the number of ACEs all participants show a downward decline in caudate volume. That is, differences between low and high ACEs are present from very early on but do not change over time. Independent of that effect, there was a main effect of preschool maternal support showing a similar pattern. That is, participants with low preschool maternal support start low, but do not show a difference in decline compared to participants with high preschool maternal support. This phenomenon is different in the hippocampus and the amygdala, regions in which the effects of ACEs and maternal support interact on the trajectories of volume over time.

Timing of Mental Health Challenges

Barch turned to the timing with which children develop mental health problems and the relationship that has to brain development. Mental health issues can arise anywhere across the life span—children as young as 3 years of age can have clinical depression and anxiety. Some evidence

suggests that earlier onset of mental health issues is associated with especially poor outcomes. Two possible explanations are that (1) the issues occur during key developmental periods or (2) the issues disrupt the child’s normative developmental experiences, because early onset of mental health issues is associated with greater chronicity. A child who begins to have mental health problems at a very early age but does not receive treatment has a high likelihood of continuing to have mental health problems. Potentially, this could be a type of experience-dependent learning. Living with depression colors a child’s developmental experiences and may change the types of learning experiences they have (Gabard-Durnam and McLaughlin, 2019). However, the timing of the onset of depression may also be interacting with various phases of brain development. Different brain areas, functions, and processes mature at different time points, which may also interact with when a child is experiencing depression.

Barch and colleagues explored these questions using data from the longitudinal study, which included measures of depression from school age to late adolescence.⁶ Evidence shows that depression is associated with disruptions in reward processing—in reward anticipation and, in some cases, reward receipt. Barch focused on cue-related brain activity that her team observed by having participants complete a reward anticipation task in the neuroimaging sessions. The overall pattern of brain activity across the entire sample was as expected, with activity in both dorsal and ventral striatum as well as ventral medial prefrontal cortex and visual cortex.

Next, the investigators looked across the entire circuit of brain regions thought to be important for reward processing (including the dorsal and the ventral striatum, and the dorsal and rostral anterior cingulate versus specific brain regions) to see if it was related to a child’s current depression versus cumulative level of depression. Then they separated out depression at preschool, school age, and later adolescence to look at differential effects. The only relationship seen with current depression was activity in the nucleus accumbens, where greater depression was associated with reduced activity in that region. Looking at cumulative depression revealed much broader effects—reduction in activity in the circuit as a whole as well as in almost every brain region, with a broader effect related to longer cumulative depression.

Depression during the preschool period shows a broad effect in the caudate, the putamen, and the dorsal and the rostral anterior cingulate, but

___________________

⁶ Barch, D. Workshop presentation—Early Adversity, Emotional Processing, and the Neural Bases of Psychiatric Illness. Available at http://www.nationalacademies.org/hmd/Activities/Aging/BrainHealthAcrossTheLifeSpanWorkshop/2019-JUN-26.aspx (accessed March 12, 2020).

no effect with the nucleus accumbens. Depression in school-age children shows an effect that is less broad, while adolescent depression was associated with the nucleus accumbens. The same pattern of effects is observed when all three age ranges are included, indicating that these are the differential effects of preschool, school age, and adolescent depression. This suggests that earlier-onset depression is associated with broader effects in the cortical limbic circuit, even when controlling for current depression.

Barch shared a hypothesis related to the different effects of cumulative versus current depression. Current depression severity was associated with hyporeactivity of the ventral striatum to anticipation of reward. If such an association between current depressed mood state and ventral striatal hyporeactivity to reward anticipation is present across development, then repeated experience of depression that starts early in childhood could lead to downstream hyporeactivity of a broader cortico-striatal circuit. An early onset of depression may disrupt this network as the child is developing, with a cascading and broad effect.

Brain Health and Resilience

Early environmental and emotional experiences relate to brain development in ways that are consistent with both experience-expectant and experience-dependent processes. In addition to considering how to measure brain health or resilience, researchers should consider how to measure the factors that promote brain health and resilience. Figure 7-2 is a simplistic model of how violations of experience-expectant input at specific developmental stages, in conjunction with early-occurring mental health challenges (which are not unrelated), can contribute to disrupted development of limbic and cortical regions. These regions are associated with subsequent poor emotion regulation and stress responsivity, which may contribute to mental health and physical health challenges in adolescence and adulthood.

Discussion

A participant noted that there appeared to be a negative slope among children with high ACEs as well as high maternal support. Barch agreed that it looks like a negative slope, but it is not a particularly strong effect. The stronger effect is driven by the positive slope of having low ACEs and high maternal support. This was not exactly the pattern Barch’s group predicted. They expected to see more of a buffering effect (i.e., very little effect of ACEs with high maternal support); instead, there seems to be more of an additive effect. Another participant asked about potential mechanisms underlying resilience in children with high ACEs who do not receive maternal support, such as self-reliance. Barch replied that she has not looked into this possibility, but she speculated that in children who are removed from a home with a poor maternal relationship, for example, other nurturing caregivers or family members may have an effect on “promoting” resilience.

THE IMPACT OF EARLY-LIFE STRESS ON NEURODEVELOPMENT

Nim Tottenham, professor in the Department of Psychology at Columbia University, presented on the impact of early-life stress on neurodevelopment. Brain health is age dependent, context dependent, and age appropriate with respect to plasticity and to the tendency or the ability to coordinate with parental cues at age-appropriate times. Humans have a long developmental period for brain development, so it is necessary to pay attention to and support families to improve the brain health of children. Childhood adversity is a leading cause of adult mental health problems, contributing to about one-third of mental illness according to conservative estimates (Kessler et al., 2010). Childhood adversity is one of the major stressors that a developing brain can experience. The large corpus of basic neuroscientific evidence from animal models shows that brain regions such as the amygdala, hippocampus, and prefrontal cortex—which make up the fundamental circuits underlying emotion regulation processes—are highly susceptible to the effects of stress in adulthood and even more so in early life. This is due in part to the rapid growth and potential sensitive periods during this phase of life; it is also due in part to the large extent to which development depends on input from a highly variable outside world (Liston et al., 2006; Magarinos and McEwen, 1995; Mitra et al., 2005; Vyas et al., 2002).

Influence of Caregivers on Neurodevelopment

Tottenham’s presentation explored how input from caregivers in this process affects neurodevelopment. This neurobiology does not develop in isolation; it develops in special species-expected context. Humans spend more time with their parents than other species. In addition to fulfilling basic needs, parents also influence the way a child’s brain learns as well as the way the brain constructs itself over time (Tottenham, 2012). Typical children and adolescents have a very robust amygdala response to emotional stimuli early in life. This tends to happen in the absence of the more mature connections among amygdala, prefrontal cortex, and hippocampus that are seen later in adolescence and adulthood. This points to an early period when the nature of the neurobiology may be such that it is highly amenable to influences like individual variations in caregiving (Gabard-Durnam et al., 2014, 2016; Gee et al., 2013b; Silvers et al., 2016; Tottenham and Galván, 2016).

Parental Buffering of Aversive Learning in Rats and Humans

Data from rodent pups show that a parent’s regulated presence buffers amygdala activity and aversive learning. During a certain period in postnatal life, the functioning of the amygdala is dependent on the presence or absence of a regulated parent. For example, when the mother rat is in the nest, her presence leads to a neural hormonal cascade that essentially quiets the activity of the pup’s amygdala during fear learning. Those processes reverse when the mother is outside of the nest: the same-aged pup’s amygdala will now be engaged during fear learning (Moriceau and Sullivan, 2006). That plays out behaviorally in an important way, as demonstrated by placing a peppermint odor that has been paired with a foot shock into one of the arms of a Y maze. If the rat pup acquires that pairing in the absence of its mother, when the amygdala is free to mediate learning, then it avoids the arm that has the odor. However, if the rat pup learns the pairing in the presence of its mother, the mother’s presence will block amygdala engagement. This allows competitive learning systems to mediate learning such that fear learning is blocked. These animals tend to show a relative preference for the peppermint odor associated with the mother’s presence. This form of learning happens in part because it plays an important role in attachment learning, which must occur during this specific period of life. This suggests that the developmental state of the young altricial brain is designed to coordinate with parental input at certain moments in development.

To investigate whether the type of effect on learning observed in the rats is similarly present humans, a study was conducted with preschool-aged children (Tottenham et al., 2019). The children were presented with

a blue square that co-terminated with a terrible noise as well as a triangle that was not paired with any noise. Children learned these pairings either alone or in the physical presence of a parent. Researchers did not observe an effect of the parent’s presence during the acquisition phase. After acquisition, the children were placed without their parents into a human Y maze with the triangle on one door and the square on the other in order to look for a behavioral tendency to approach one door over the other.

Children who had been conditioned alone without a parent were more likely than not to avoid the square door, indicating avoidance learning. However, children who had been conditioned in the presence of a parent were more likely than not to show a preference for the square door, which was similar to the behavior seen in the rat pups. This was a within-subjects design, so the same children’s learning seemed to be affected by the parent. No effect was seen during acquisition, suggesting that the parent is not simply calming the child, but changing the nature of the learning that occurs. Variability across children is partially explained by cortisol levels, Tottenham added. Children with higher levels of cortisol production were less likely to show this effect of the parent, suggesting some sort of biological constraints on this type of learning.

In a separate study, children were scanned while they looked at pictures of their parents relative to other people’s parents. Investigators found that pictures of parents during childhood were effective in dampening the activity of the amygdala (Gee et al., 2014). Although it is not clear whether these are exactly the same processes as shown in rodents, there are some compelling parallels. These types of data suggest that during sensitive periods of childhood, a parent can potentially have large ramifications on the nature of emotion regulation neurobiology that is observed later on in adulthood, reflecting a scaffolding effect. This research provides a foundation for asking questions about what happens to the model when there are severe aberrations in the early caregiving environment.

Effect of Emotional Neglect or Rejection by Caregiver

Children can experience many types of maltreatment, but Tottenham focused on emotional neglect and/or rejection by the caregiver. Unlike the more obvious signs present in physical neglect or abuse, emotional neglect and rejection effects can be less obvious and often overlooked. This is a pernicious form of maltreatment that is highly comorbid with other forms of maltreatment that children experience. The absence of parental input early in life is not the same as the absence of a threat to the infant. It is a failure to receive needed parent–child intimacy, support, and serve–return dynamic, as well as being a significant stressor during brain development. When considering the role of stress during development, it

is important to consider the “stress chronotype” that a child has or experiences—for example, whether the stress experience was limited to the infant period, childhood, adolescence, or whether the stress was chronic (Tottenham and Galván, 2016). Studying different stress chronotypes is valuable because they characterize the experience of many children who are subject to severe adversity. It is important to bear in mind that in many cases, children are still living in those environments at the time of assessment.

Early Parental Deprivation and Amygdala Responsivity

Tottenham focused on children who experienced a major early-life stress that was terminated, then followed by a relative absence of stress. In this case, they studied children who experienced early institutional care, which is an extreme form of caregiving neglect or deprivation, and were subsequently adopted into families that provided a very enriched caregiving environment. These children had a significant initial developmental risk followed by a significant rebound in a number of domains after adoption. However, there was significant heterogeneity in their outcomes. At the group level early-life stress is a tremendous risk factor, but there are many individual differences. When children struggle, they are most likely to struggle in the domain broadly defined as emotion regulation, much of which occurs prior to the formation of explicit memory.

Tottenham presented data from children who were placed in institutional care at or near birth and then adopted by their second birthday (Tottenham et al., 2011). They found that previous institutional care is associated with elevated symptoms in the domains of internalizing or externalizing problems in a sample of 373 adolescents, but the data are heterogeneous. To explore what discriminates children at the top of the range from the bottom, they assessed the participants’ amygdala responsivity. Previous studies had found evidence of amygdala hyperresponsivity to emotional cues, such as fear faces following early institutional care. These responsivity patterns have been associated with many of the internalizing problems, as well as with behaviors that can be measured in the laboratory. For instance, children with stronger amygdala signals to fear faces were also less likely to make eye contact as measured by eye-tracking measures and as measured by live dyadic interactions (Tottenham et al., 2011).

Amygdala Hyperactivity Reduces Developmental Plasticity

Functional connectivity between the amygdala and medial prefrontal cortex reveals age-by-caregiving group interactions. Children with

low-risk, no adversity backgrounds were likely to show a more childlike pattern of connectivity between amygdala and prefrontal cortex. Later, in adolescence, a more inhibitory pattern or anticorrelated relationship between the amygdala and prefrontal cortex is typically observed (Gee et al., 2013a). In children with a history of previous institutional care, the connectivity pattern does not look like typically raised children. Instead, it more closely resembles the patterns seen in adolescents and in adults. Tottenham posited that strong amygdala reactivity early in life as a result of stress may actually instantiate earlier formation of these connections with the prefrontal cortex through an activity-based process, leading to a reduction in developmental plasticity (see Figure 7-3). This may represent one means by which early-life stress is reducing neuroplasticity during childhood and adolescence.

It has been posited that these windows of plasticity—or sensitive periods—can be moved by different life experiences (Werker and Hensch, 2015). Caregiving adversity may shift some of these moments of plasticity or truncate them at earlier points within the circuits that have been most affected by early-life adversity (Callaghan and Tottenham, 2016). Tottenham suggested that this may be happening through activity-based processes that have certain immediate benefits to the individual. Evidence suggests that overall, previous institutional care is associated with higher anxiety. However, children in the previous institutional care group show a more adult-like pattern of amygdala-prefrontal connectivity, which

indicates lower separation anxiety relative to their peers who did not show that change.

These data suggest that there might be some advantages to emotion regulation processes following early-life stress with this adaptation (Gee et al., 2013a). When the children were followed over 5 years, the children who showed the more adult-like pattern of connectivity were those who were likely to retain the higher anxiety phenotype over time. One hypothesis is that children who showed the more childlike phenotypes through some transactional processes with parents may actually be invoking different caregiving behaviors that could have some benefits on these phenotypes over the long term.

Truncated plasticity also places limits on the developing brain’s capacity to respond to parental cues. Unlike the typically raised rat pups that showed the bizarre preference learning in the presence of the parent, animals that had experienced early maltreatment did not show this buffering effect by the parent. Instead, they avoided the negative stimulus. This effect was mediated by the amygdala, such that the presence of the parent was less effective in buffering the amygdala during the fear learning (Moriceau et al., 2009). Similarly, typically raised children showed a decrease in amygdala reactivity to their parents versus strangers that was not seen in children following institutional care. At the group level, this suggests that the parent was less able to modulate the amygdala, but the data show large individual differences; some children might have been showing this amygdala dampening to their parental cues even following very early institutional care.

These individual differences were interrogated longitudinally by splitting the postinstitutional care group into two subgroups: (1) those who showed lowering of the amygdala at time 1, and (2) those who did not, despite having comparable levels of anxiety at initial assessment. Over the 2-year period, those who showed the dampening of the amygdala at time 1 showed the decreases in anxiety over that 2-year period.⁷ Those who showed the buffering effects were children who reported higher attachment security. This suggests that even despite the significant adversity, a family can have some powerful effects on shifting this neurobiology.

Addressing Heterogeneity Among People Exposed to Adversity

Tottenham concluded by noting that early adversity significantly increases the risk for poor mental health, but there is tremendous

___________________

⁷ Available at http://www.nationalacademies.org/hmd/Activities/Aging/BrainHealthAcrossTheLifeSpanWorkshop/2019-JUN-26.aspx (accessed March 12, 2020).

heterogeneity within groups of people who are exposed to adversity that warrants investigation. Some of these differences might be viewed as potential developmental adaptations, some of which may work for the individuals and others against. The immediate goal is to better understand independent variables in studies of early-life stress, while also considering how environmental needs develop and change with age. This heterogeneity is present within subgroupings of people who experienced early adversity, such as people who experienced domestic foster care and people who experienced international adoption with institutional care. This indicates that there may be specific experiences that transcend these traditional recruitment boundaries.

Her group’s current approach is to invite children who have experienced various types of caregiving experiences—both positive experiences and adversities—and, through data-driven processes, to cluster those children either on their brain behavior phenotypes or on their caregiving experiences. These clusters can also include adversities that are not related to caregiving per se. Preliminary data are beginning to reveal some evidence of meaningful clusters, although there is still heterogeneity within the clusters and there are some factors that transcend these different boundaries.

INTEGRATING COMPLEX AND PERSONALIZED DATA TO UNDERSTAND NORMAL AND ABNORMAL BRAIN NETWORK DEVELOPMENT

Satterthwaite described two studies that integrate complex and personalized measures of network development throughout youth and adolescence with a focus on brain health. He explained that the rationale for studying brain development is increasingly clear. Convergent lines of evidence from animal models, human epidemiological studies, and translational studies suggest that most major neuropsychiatric conditions can be conceptualized as disorders of development. This domain of research seeks to understand how the brain develops normally and then to understand how abnormal patterns of brain development are associated with different forms of psychopathology.

The ultimate goal of describing major mental illness in terms of abnormal trajectories of brain development is to allow for earlier diagnosis and intervention with more effective treatments in order to achieve improved functional outcomes for people living with these conditions. This work requires large-scale data that allow for sampling across multiple age ranges in both healthy and affected children. In his presentation, Satterthwaite presented neuroimaging data from a complex large-scale initiative called the Philadelphia Neurodevelopmental Cohort (PNC), which

focuses on characterizing brain and behavior interaction with genetics. The cohort included 1,600 children aged 8–22 years, with a balanced mix of males and females and of Caucasians and African Americans that reflects the local Philadelphia population (Satterthwaite et al., 2014).

Network Modularity as a Key Measure of Brain Health

Satterthwaite used selection neuroimaging data from the PNC to focus on structural and functional brain networks. Unlike neurological conditions in which there is a clear lesion and focality, psychiatric conditions are increasingly conceptualized as connectopathies—that is, disorders of how the brain communicates. Brain networks can be measured both structurally and functionally. Structural brain networks can be reconstructed with diffusion imaging tractography techniques, while functional brain networks can be estimated with functional magnetic resonance imaging (fMRI).

Network modularity is a key feature of brain development as well as aging. Although measures of brain health are not yet sufficiently refined to be clinically actionable, available data suggest that network modularity is a key measure (Satterthwaite et al., 2014). A network module is a collection of brain regions that are tightly connected to each other and weakly connected to other parts of the brain. These correspond to functional subsystems of the brain, as defined by conversion evidence from task fMRI, lesion studies, and animal studies. This modularity can be visualized using spring-embedded rendering to depict how the tightly connected brain regions are brought together and the weakly connected brain regions are pushed apart, highlighting the network modules.

One of the most widely replicated findings in developmental cognitive neuroscience is that this modularity evolves dramatically throughout childhood and adolescence (Fair et al., 2007). Intramodular connections are much more likely to strengthen than weaken with age, which causes modules to become more defined during adolescence (Satterthwaite et al., 2014). Satterthwaite’s group recently showed that structural brain networks undergo a similar process of modular segregation—modules become refined and prominent, with more connectivity within a module and less connectivity between modules. These modules are present but are relatively indistinct in younger children, but they become much more defined as the children grow up. This modular segregation has functional consequences. The development of executive function during this period is mediated by the degree to which this network topology develops in the brain, with modules shifting toward more specialized functions (Baum et al., 2017). However, this type of analysis only looks at simple, low-dimensional summary measures of network structure and

only at executive function, which is just a single domain of cognition. A more clinically relevant approach would be to look across all clinical domains in psychiatry and simultaneously map these to abnormalities in the high-dimensional topology of the connectome.

Data Integration to Understand Abnormal Network Development

Satterthwaite described a recent study that used machine-learning techniques to integrate complex clinical data with high-dimensional imaging data. First, he described how the approach used in this project is a departure from the case-control design of clinical studies (e.g., comparing a person with a diagnosis of depression to a healthy person), which does not address two central challenges in integrative psychiatric neuroscience. The conceptual challenge is that clinical diagnostic categories as codified in clinical practice are not clean biological classes, owing to vast heterogeneity and the frequency of comorbidities. In other words, they do not carve nature at the joints in a clear way. For example, depression is a large category that is unlikely to represent a single biological phenomenon, given the amount of heterogeneity that has been demonstrated and the number of comorbidities that occur with the condition, such as anxiety.

Dimensionality

The methodological challenge is that the data are highly dimensional. Typical case-control designs that ignore heterogeneity and comorbidity are often very hypothesis driven, thus they miss the opportunity to collect other rich data. Instead, this project takes a discovery science approach using machine learning to define data-driven links between functional brain networks and psychiatric symptoms. He said that in essence, this “lets the brain teach us what the dimensions should be.” He explained that typical functional networks use atlases with hundreds of nodes that cover the entire brain; when this is taken to a connectivity matrix, connections between each of these nodes can create a common network with tens of thousands of edges (Xia et al., 2018). This is relatively high-dimensional data.

A problem that is just as substantial, but even more commonly ignored, is that clinical data also have reasonable dimensionality. Standardized psychiatric interviews include hundreds of questions about symptoms, but most psychiatric imaging studies take a very reduced look at these data—for example, by looking only at amygdala connectivity and the rest of the brain, instead of looking at the entire connectivity matrix. Similarly, a clinical case-control study of depression would typically ignore

other symptoms and collapse the depression items into a single categorical diagnosis.

A common alternative is to try to integrate the data through data reduction, for instance, by looking parsimoniously at a small number of brain networks instead of looking at all 35,000 connections or by summarizing itemwise clinical data into a four- or five-factor model. Although this approach is reasonable, it is based on the assumption—which is not well supported by evidence—that the biological scale of the abnormalities matches with the scale of data reduction. Another alternative is to try to directly integrate high-dimensional brain data with available granular clinical data using sparse canonical correlation analysis (sCCA).⁸

Identifying Linked Dimensions of Psychopathology and Functional Connectivity

The approach taken in the study Satterthwaite presented uses sCCA to build a linear combination of brain features that predict a linear combination of clinical features in a data-driven manner (Xia et al., 2018). Through a process of permutation testing and correction for multiple comparisons, sCCA can be used to identify linked dimensions of psychopathology and functional connectivity. In this case, the study identified the dimensions of mood, psychosis, fear, and externalizing behavior. A first step in looking at the data was to plot the brain connectivity score versus the clinical dimension score, which revealed tight relationships between the brain and the clinical dimension and allowed for identifying the most highly rated clinical item in each of the dimensions. For example, the most highly rated item in the mood dimension is “feeling sad” (Xia et al., 2018). Boot-strap resampling anyalysis was used to understand which clinical items significantly contribute to each dimension. Figure 7-4 is a ring plot that is laid out with classic discrete clinical diagnoses in the outer ring. In the inner rings are the loadings of the clinical items in each dimension.

Investigators found that the clinical item loadings accord with clinical experience, but they also cross diagnostic boundaries (Xia et al., 2018). These data-driven dimensions of psychopathology generally cohere with the clinical diagnostic categories to a great extent, but they also bleed across them in a graded way. The psychosis dimension has significant loadings in the mania items, for instance, which makes sense given the

___________________

⁸ sCCA is an updated version of an older statistical technique called canonical correlation analysis, which is limited by the requirement to have more samples than features in the model. sCCA imposes sparsity constraints that allow the model to have more features than samples.

genetic overlap between bipolar disorder and psychotic disorders. Furthermore, Satterthwaite and colleagues found that specific differences of functional connectivity define each dimension, but there are key features that are present across each dimension, such as a loss of modular segregation.

Studies of typical brain development show that modular segregation evolves throughout childhood and adolescence and supports executive function. In these data-driven dimensions of psychopathology, each of these dimensions is associated with a loss of this normative modular segregation (Xia et al., 2018). A replication sample generated largely convergent results with the initial discovery sample, although the psychosis dimension was not replicated. Together, this indicates that data-driven

dimensions of psychopathology can link abnormalities and functional connectivity. Modular segregation is a common feature across the dimensions, suggesting that it is an important aspect of brain health as the brain develops (Xia et al., 2018).

Personalized Development of Network Topography

The second study presented by Satterthwaite was designed to work toward personalizing measures of functional connectivity. Typically, all the brain images from participants in a study are registered to a standard group atlas, with an implicit assumption that networks are in the same anatomical location for each person. Satterthwaite’s study addresses a major limitation of the first study he presented as well as most studies of functional brain networks conducted until recently: the idea that the functional networks in the brain are laid out similarly across individuals.

Functional Topography Varies Across Individuals

However, recent evidence from multiple groups has shown that this assumption is demonstrably false. For example, work using the Midnight Scan Club⁹ dataset shows remarkable heterogeneity in the spatial layout of these functional networks on the anatomic cortex, which has been replicated in multiple datasets using independent methods (Bijsterbosch et al., 2018; Gordon et al., 2017; Kong et al., 2019; Li et al., 2017). Furthermore, these studies reveal the heterogeneity in functional topography, or the spatial layout of these functional networks, varies by location in the brain. Functional topography varies most across individuals in the higher-order association cortex, which are the regions of the brain most relevant in psychiatric illness. Across different sets, the most variability is present in the frontoparietal control network and in the ventral attention network. However, it is not yet understood how this individualized functional topography evolves in development or how it associates with important domains of healthy behavior, such as executive function.

Nonnegative matrix factorization is a machine-learning technique that can be used for identifying brain networks in individuals (Li et al., 2018), and it allows for defining 17 networks per person on a subject-specific basis (Cui et al., 2020). Single-subject data show that the spatial layout of these networks varies between individuals. For example, looking into frontoparietal networks and the ventral attention networks either on a continuously loaded basis or on a binary basis shows that they vary

___________________

⁹ See the Midnight Scan Club dataset at https://www.openfmri.org/dataset/ds000224 (accessed November 13, 2019).

in terms of their placement on the anatomic cortex (Cui et al., 2020). The networks with the highest variability are in the frontoparietal control network and the ventral attention system, which are key networks for executive function (Cui et al., 2020).

To explore how these highly variable networks evolve throughout the adolescent period, which is a critical period for the development of brain health and the period in which many neuropsychiatric symptoms begin to emerge, Cui and colleagues created a total network representation. This is a simple measure of each of the individualized networks that summarizes how much space on an individual cortex each functional network occupies. They found that total network representation did not associate with age (which was surprising), but it was strongly associated with executive function (Cui et al., 2020). Children and adolescents who perform better in executive tasks have more total network representation allotted to systems that are important for executive function, like the frontoparietal control network and the ventral attention network.

To move beyond the simple summary measure to the overall multivariate pattern of how the functional topography is laid out in the brain, researchers used split-half data and machine-learning techniques to predict executive function in completely unseen data with a relatively high degree of accuracy (Cui et al., 2020). The most important weights were seen in the ventral attention system and in the frontoparietal control network, suggesting that the individualized layout of functional network topography could be a very important measure of brain health in adolescents. Finally, they looked at whether the overall complex pattern of functional topography was associated with age. Even though the summary measure of the total network representation did not seem to change with age, it was possible to predict age from the functional network topography even more accurately than executive function could be predicted. In fact, it was refinement within association networks that predicted age. The total size of these networks did not change over the age span, but the borders of the network were sharpened. This suggests that these networks are differentiating throughout the developmental period, in a process reminiscent of the network-wide segregation process. Satterthwaite added that these highly variable brain networks, with functional topography that varies across individuals, have certain fundamental properties of cortical organization.

Variability in the frontoparietal control network and the ventral attention system align with a high degree of evolutionary expansion, low cortical myelin content, and high cerebral blood flow (Cui et al., 2020). This is consistent with an account whereby association networks evolve and become untethered from rigid developmental programs. This process is probably beneficial in many ways, because it allows for interindividual

variability and adaptability as children are developing. However, it also comes at a metabolic cost and could potentially lead to higher vulnerability to neuropsychiatric syndromes during the critical period of adolescence.

BRAIN NETWORK AGING AND HEALTH ACROSS THE ADULT LIFE SPAN

Gagan Wig, associate professor of behavioral and brain sciences at the Center for Vital Longevity at the University of Texas at Dallas, presented on brain network aging and health across the adult life span, with a focus on novel measures of brain health. Even in the absence of disease, aging is associated with progressive changes in cognition (Park and Reuter-Lorenz, 2009). However, simply comparing endpoints between younger and older adults is inadequate to understanding cognitive health decline. Aging also varies widely across individuals, with longitudinal data demonstrating how perceptual speed declines within an individual over age at very different rates (Wilson et al., 2002). Although variance can be helpful in understanding how steeper declines might be related to degenerative processes versus typical aging, understanding the parameter space of healthy aging requires inquiry into the brain associates that accompany those behavioral observations.

Multiple imaging-based measures of the brain can be used to characterize aging, including functional activation (Cabeza et al., 1997), structure of gray matter (Raz et al., 1997) and white matter (O’Sullivan et al., 2001), metabolism (Oh et al., 2016), and dopamine binding. However, none of these measures is considered a standard measure of brain health, because it is not yet well understood how the information that is embedded in the signals provided by these tests correspond to individual variation in brain health and potential health outcomes.

Wig suggested that understanding brain health and risk of decline will require examining multiple brain measures together as well as identifying novel measures. His group conducts work based on the hypothesis that resting-state brain network organization is an important biomarker (for lack of a better term) of age-related cognitive decline. The network approach may be the most appropriate framework that is currently available for understanding cognition. Analyzing resting-state networks involves four main steps: identifying parcellated nodes of the brain, extracting the time course for every node and computing pairwise correlations, building node-to-node correlation matrices for each subject, and then graphing a theoretic analysis of network structure (Bullmore and Sporns, 2009; Wig et al., 2011). The work he described focuses on the graph structure in terms of aging.

Desirable Features of Measures of Health in Aging Brains

To provide context for thinking about measures of brain health, Wig outlined a set of desirable features for measures of brain health as people age: ease of collection, reliability, validity, and changeability. Validity of brain health measures can be characterized in different ways, such as having continuous variation across the adult life span and not just at the endpoints or being related to cognition, even in “typical” ranges. It could also be described as being moderated by measures related to general health, environment, and lifestyle or by its predictive value in warning of impending dysfunction or adverse event. Changeability is desirable because it can be modified in an ideal situation.

Differences in Brain Network Organization Across the Adult Life Span

Evidence is emerging that resting-state brain networks in young adults are organized into communities—described by Satterthwaite in terms of modularity—that correspond to functionally distinct brain systems (Power et al., 2011). Wig’s group is applying that observation in the context of healthy aging and how that organization differs across the adult life span. They are using the Dallas Lifespan Brain Study dataset, which includes data from more than 300 subjects sampled across a broad segment of the adult life span collected via T1, DTI, and BOLD functional scans (multiple tasks and rest). Starting with the rest data, Wig’s group has worked to minimize known sources of variance by minimizing the influence of movement and ensuring quality control of the T1 and BOLD data, which are both susceptible to movement-related artifacts (Power et al., 2014; Savalia et al., 2017).

The topography of large-scale functional brain systems is largely consistent across the healthy adult life span, indicating that healthy aging is not accompanied by a massive reconfiguration of basic modular structure (Han et al., 2018).¹⁰ Inherent in this description of functional systems is the idea that modularity requires dense within-system connections and sparser between-system connections (Power et al., 2011). Although this concept may underpin functional specialization of a network, the segregation of network communities can change over time.

___________________

¹⁰ To avoid the limitation of fixed node atlases in network descriptions (described by Satterthwaite) they create parcellations at every age cohort level.

Effect of Aging on Network Segregation

Having network communities that are segregated, but still able to communicate with one another, requires a fine balance of connections both within and between the communities. This confers functional specialization as well as some interaction between them (Wig, 2017). As functionally specialized communities in the network become more segregated, they become increasingly disconnected; if they become less segregated, they run the risk of becoming undifferentiated.

Wig’s group hypothesized that functionally specialized networks in healthy young adults would be situated roughly in the middle of this spectrum. To explore whether aging has an effect on this basic property, they took the nodes as a function of the system and classified them according to whether the connections were within or between systems and used the weighted average of the various connection types as the measure of segregation. They found that older age was associated with decreased segregation of large-scale brain systems. With increasing age, the connectivity within the functional systems decreases but the connectivity between systems increases (Chan et al., 2014).

Figure 7-5 is a spring-bedding diagram showing that increasingly sparser connectivity is associated with increasing age fanning outward (i.e., in the blue system, which corresponds to the visual system). The interactions between the association systems (i.e., within the circle in purple, yellow, and green, corresponding to the cingulo-opercular control system, frontal-parietal control system, and dorsal attention system, respectively) tend to increase, leading to the observation of decreasing segregation.

Reliability of the Relationship Between Brain System Segregation and Adult Age

With respect to reliability of the relationship between system segregation and adult age, the age-versus-system segregation relationship has now been observed across multiple studies using different methods (Betzel et al., 2014; Cassady et al., 2019; Chong et al., 2019; Geerligs et al., 2014; Grady et al., 2016; Han et al., 2018; King et al., 2017; Shaw et al., 2015; Song et al., 2014; Spreng et al., 2016; Zonneveld et al., 2019). Wig’s laboratory has also carried out independent replications based on other data sets that support the relationship between the system segregation measure and age. Additionally, his group has looked at whether system segregation is a reliable measure of an individual’s brain network organization by using resting-state data from young adults collected on two different days, finding the measure to be strong enough that there is reason to believe it might be a useful way of thinking about individual-level network organization.

Relationship Between Brain System Segregation and Cognitive Ability

Features of the Dallas Lifespan Brain Study dataset were also helpful for exploring how measures of the brain relate to cognitive ability. The data show that increasing segregation is related to long-term memory (Chan et al., 2014). When the age effect is accounted for, people that have more segregated systems have better memory ability independent of age (Chan et al., 2014). In other words, some older adults with very high system segregation have relatively higher memory scores, while some younger adults with lower system segregation have poorer memory scores (Chan et al., 2014). The trajectory of these individuals over time is a research direction of interest.

Modifiers of System Segregation Across the Adult Life Span

Wig’s group has also looked for other modifiers of system segregation across the adult life span, such as the role of the individual’s environment, by using a bigger sample from the Dallas Lifespan Brain Study (Chan et al., 2018).¹¹ In adults, lower socioeconomic status (SES) is associated with worse cognition (Koster et al., 2005) and greater risk of Alzheimer’s disease (Stern et al., 1994), so Wig’s group explored whether SES relates to this measure of brain organization. SES is a crude construct, but it does give a sense of access to resources, nutrition, health care, cognitive stimulation, and levels of stress.

In this case, Wig’s group defined SES by education and occupational status. The analysis, using continuous measures of age and SES, found that lower SES was associated with reduced system segregation in middle-age adulthood, but not in younger or older adults (see Figure 7-6). The relationship persists when controlling for demographics, physical health, mental health, cognitive ability, and a measure of childhood SES based on parental education. Because the measure does not differ for older adults, Wig suggested that this indicates a survivor bias in lower SES older adults (Chan et al., 2018). For younger adults, he suggested that the commonly used SES measure is inappropriate, because they may have evolving educational and occupational states.

___________________

¹¹ A second round of data collection was conducted with relaxed exclusion criteria to include people with lower education and chronic health conditions, providing a more population-representative sample.

Future Research Directions

Wig concluded by outlining a set of his group’s future research directions. Rather than looking at cross-sectional data, his group is focused on examining networks within an individual by bringing together multiple large longitudinal datasets to explore whether brain networks change as an individual grows older in both health and disease, whether brain network organization is modified by changes in brain degeneration, and if brain network patterns can be used to predict who may be vulnerable to brain disorders. The group is also looking at which aspects of an individual’s environment mediate the SES–brain network relationships in a study focused on community-based, middle-age adults who are at or near the household poverty line. Through longitudinal assessment of

changes in brain network organization and cognition, Wig’s group plans to extensively characterize changes in health and lifestyle over time using biological measures, survey-based measures, and other techniques. Work focused on interventions is also under way to explore whether the system segregation of an individual’s brain network can be changed.

One of the studies involves engaged learning of new skills by older adults, based on the idea that maintaining brain health and cognition involves continuous learning. Another study is looking at the potential for precision brain stimulation to alter network organization. Participants are receiving extensive transcranial magnetic stimulation at specific parcels (brain regions) based on the individual’s specific brain topography to determine whether there is a change in network segregation as a function of stimulation location.

PANEL DISCUSSION ON THE WAY FORWARD IN MEASUREMENT AND RESEARCH

Damien Fair, associate professor of behavioral neuroscience, associate professor of psychiatry, and associate scientist at the Advanced Imaging Research Center at the Oregon Health & Science University, asked the panelists to comment on the types of research needed to make the leap from the conceptual to the actionable in long-term brain health.

Brain Health Growth Charts

Barch highlighted a challenge in measuring the development and decline of brain health throughout the life span. It is possible to observe group differences and identify individuals, but it is not yet possible to determine which people need extra clinical intervention. This will require developing the appropriate psychometrics, measurement tools, databases, and study designs. She suggested tracking brain health through a “growth chart” to capture individual changes over time, as is already done for physical growth measurements. Tottenham remarked that creating brain health growth charts would be very useful, but a potential complication is the need to integrate context-specific environmental factors into the system.

Behavioral outcomes have an additional layer of complexity in that whether or not certain outcomes are healthy is contextualized by the specific environment. Manly noted that physical growth charts assume universality and variance in their predictions of important outcomes; they are generally intended to be screening tools rather than diagnostic tools. Brain health growth charts have the potential to be used as a screening tool for inappropriate purposes that might actually widen disparities

rather than narrowing them, Manly cautioned. A participant added that body type and physiological factors such as insulin resistance can substantially affect brain development, so brain–body interactions would be important to capture.

Leveraging Electronic Health Records Systems

Health systems that have integrated electronic health record systems are at an advantage in this regard, said Satterthwaite. Such systems are already capturing important biologically based measures that could be easily scaled—through mobile platforms or otherwise—and then linked to meaningful health outcomes. Currently, differential treatment response to measures of brain health is a major gap in the literature that will need to be addressed to make this research clinically actionable. Building systems of informatics could also help to link biological measures to health outcomes. Fair highlighted this suggestion as a potentially actionable item that could have a relatively large effect.

Identifying Practical Measures with Clinical Usefulness

Lis Nielsen, chief of the individual behavioral processes branch of the Division of Behavioral and Social Research at the National Institute on Aging, commented about the work on network differentiations and patterns over development. Currently, these kinds of measures cannot feasibly be captured on a broad scale for use as diagnostic or screening tools. She was interested in how mapping network development over time relates to the evolution of the entire range of functions that those networks represent. Mapping those functions onto simple cognitive assessments, in order to capture measures in larger populations or in clinics, could potentially shed light on the drivers of network differentiation or interaction that enable the network functions.

Nielsen asked what research would be needed to facilitate the ability to capture this simply by assessing cognition and other functions. Wig replied that network functions correspond to systems of the brain that are functionally distinct. Data show that increasing differences among measures are associated with changes in activity of the specific regions that tie the networks together. His group is conducting a newer study that includes multiple features to identify relationships to other measures that could feasibly be collected on a larger scale. It is not practical to collect neuroimaging data from everyone, so the aim is to identify measures or sets of features that relate to and mediate the relationships observed in imaging studies, but which can be collected simply and quickly for use in health care settings.

Balancing Conceptual Research with Practical Interventions

Barch commented that the focus on individual differences in brain health that are unrelated to environmental factors may distract from efforts to address brain health issues that are known consequences of well-established factors—such as early adversity, SES, poverty, and stress—that also influence physical health (e.g., insulin resistance). Research on the biological bases of brain development and developing measures of brain health outcomes would be helpful in many ways, but there will always be individual variation, and this work should not preclude efforts to intervene on environmental factors that are known to influence brain development. For example, relatively simple interventions, such as a modest income transfer to families living with housing or nutritional instability, would likely have positive effects on the brain health as well as physical health of children in those families. She emphasized that the focus on measuring outcomes and identifying people with brain health issues should not come at the expense of implementing strategies to address the environmental factors that are driving those brain issues.

Tottenham agreed, noting that measures such as insulin resistance are actually outcome measures of a number of different factors. While it may theoretically be easier to change external environmental factors than genetic influences, for example, these types of interventions are constrained by a range of political, economic, and social forces on a practical level. She reiterated that at the very least, 30 percent of mental illness is attributable to childhood adversities. These childhood adversity factors are entirely modifiable, but doing so requires a society to collectively decide to improve them.

Geographic Diversity in Brain Health

A participant noted that because brain health outcomes have wide geographic diversity, research should seek to better understand the mechanisms that underlie this diversity and how factors such as place of birth versus place of current residence interact with each other and predict outcomes. Barch noted that much of that geographic diversity reflects variation in different facets of SES, such as quality of schools, median income, and other metrics. Urbanicity also has both positive and negative effects on mental health outcomes that depend on factors such as minority status. She added that there is a dearth of neuroimaging data from rural areas, so it is unclear if the same types of relationships hold as in urban areas.

Next Steps

Fair suggested focusing on next steps in identifying interventions that would have the most “bang for the buck.” For instance, delaying school start times to allow adolescents to get a little more sleep is a relatively small change that would have a great effect.

Guidelines for Improving Brain Health

Stephanie Cacioppo, director of the brain dynamics laboratory, assistant professor of psychiatry and behavioral neuroscience, and assistant professor at the Grossman Institute for Neuroscience at the University of Chicago Pritzker School of Medicine, suggested developing specific, practical guidelines for improving brain health that would be analogous to existing guidelines on how to improve physical health. Barch commented that there is a range of commonsense good practices that can improve brain health, such as exercise and sleep. However, she warned against framing that kind of information to imply incorrectly that it will prevent the occurrence of Alzheimer’s disease or other brain disorders with underlying genetic factors.

Tottenham suggested adding Maslow’s hierarchy of needs to the list of good practices for brain health, but that these needs should be developmentally tailored. Notably, for infants and young children, availability of a reliable caregiver should be added to the list of other survival needs like food, water, and shelter at the most fundamental level of the hierarchy. Adequate, stable caregiving would be a basic need early on, but it is not necessarily a fundamental basic survival need for an older individual.

Clinical Assessment of Environmental Factors

Nielsen suggested that importing the assessment of environmental variables into the clinical context may be low-hanging fruit. There is evidence that individuals with a history of early life adversity may be differentially responsive to treatments for depression and other mental disorders, but clinics do not typically ask about those factors. There may be value in collecting such data to examine implications for preventive and treatment interventions addressing a range of cognitive and emotional disorders. Clinical assessment of personality, which can predict a portion of a person’s risk for Alzheimer’s disease, could also be useful in informing treatment decisions, and awareness of risk could potentially encourage people to mitigate other risk factors for the disease. Similarly, evidence about differential response to depression treatment as a function of early-life adversity could inform interventions and treatment decisions in a way that does not require any kind of new brain measure.

__________________