Introduction
Biological psychiatry entails the study of how the brain functions to manage cognition, behavior, and emotion in health and disease. In recent years, magnetic resonance imaging (MRI) has emerged as a mainstream tool for developing models of brain structure and function. In large part, this shift toward MRI usage reflects the increasing ability of MRI techniques to non-invasively characterize inter-individual differences in the organization of human brain circuitry. Crucially, MRI measurements can be related to individual variations in cognition, behavior, and emotion. Already, MRI studies of clinical populations have offered several novel perspectives on pathologic processes. Perhaps most exciting, there is growing evidence that MRI may yield clinically useful brain-based biomarkers (Buckholtz & Meyer-Lindenberg, Reference Buckholtz and Meyer-Lindenberg2012; Castellanos, Di Martino, Craddock, Mehta, & Milham, Reference Castellanos, Di Martino, Craddock, Mehta and Milham2013; Di Martino, Fair, et al., Reference Di Martino, Fair, Kelly, Satterthwaite, Castellanos, Thomason and Milham2014; Kaiser, 2013; Kelly, Biswal, Craddock, Castellanos, & Milham, Reference Kelly, Biswal, Craddock, Castellanos and Milham2012; Watanabe, Kessler, Scott, Angstadt, & Sripada, Reference Watanabe, Kessler, Scott, Angstadt and Sripada2014). To realize these ambitions, researchers are endeavoring to increase the validity and reproducibility of imaging findings via leveraging multicenter designs (Van Essen et al., Reference Van Essen, Smith, Barch, Behrens, Yacoub and Ugurbil2013), as well as large-scale aggregate samples generated through data sharing initiatives (Di Martino, Yan, et al., Reference Di Martino, Yan, Li, Denio, Castellanos, Alaerts and Milham2014; Mennes, Biswal, Castellanos, & Milham, Reference Mennes, Biswal, Castellanos and Milham2013; Nooner et al., Reference Nooner, Colcombe, Tobe, Mennes, Benedict, Moreno and Milham2012). However, increasing sample size alone will not be sufficient to address the broader range of neuroscientific and clinical questions, as most datasets collected to date are cross-sectional. While such datasets have value for efforts to differentiate populations from one another (e.g., establishing the neural correlates of illness, carrying out brain-based diagnostic classification), they are limited in their ability to support a mechanistic understanding of the developmental processes underlying psychiatric illness, as well as to offer predictions regarding risk or prognosis.
The importance and value of developing cohorts in which individuals are imaged repeatedly over the course of time (i.e., longitudinal designs) is reinforced by the growing consensus that most psychiatric illness is neurodevelopmental in origin. Some of the most compelling support for this notion are epidemiologic findings suggesting that at least 50% of mental disorders arise before age 14, and 75% before age 24 (Kessler, Avenevoli, & Merikangas, Reference Kessler, Avenevoli and Merikangas2001). Furthermore, genetic research has documented the high heritability of many forms of severe psychiatric illness (e.g., autism, bipolar disorder, schizophrenia) (Burmeister, McInnis, & Zollner, Reference Burmeister, McInnis and Zollner2008; Colvert et al., Reference Colvert, Tick, McEwen, Stewart, Curran, Woodhouse and Bolton2015; Cross-Disorder Group of the Psychiatric Genomics et al., Reference Lee, Ripke, Neale, Faraone, Purcell and Wray2013; Muhle, Trentacoste, & Rapin, Reference Muhle, Trentacoste and Rapin2004). These findings are bolstered by the growing number of associations between psychiatric illness and genetic variants (i.e., common and rare) (Caspi & Moffitt, Reference Caspi and Moffitt2006; Gaugler et al., Reference Gaugler, Klei, Sanders, Bodea, Goldberg, Lee and Buxbaum2014; Martin, O’Donovan, Thapar, Langley, & Williams, Reference Martin, O’Donovan, Thapar, Langley and Williams2015; Sandin et al., Reference Sandin, Lichtenstein, Kuja-Halkola, Larsson, Hultman and Reichenberg2014), as well as undeniable familial associations with psychiatric illness (e.g., 19% risk for autism spectrum disorder [ASD] in younger siblings of an individual with ASD) (Ozonoff et al., Reference Ozonoff, Young, Carter, Messinger, Yirmiya, Zwaigenbaum and Stone2011).
Arguably, the neurodevelopmental perspective of psychiatric illness suggests not only the need for longitudinal research, but also the need to begin research as early in development as possible. In support of these needs, imaging researchers are rapidly developing the capabilities needed for functional and structural imaging in prenatal and early postnatal life (e.g., neonatal imaging, infant imaging, toddler imaging) (e.g., Almli, Rivkin, McKinstry, & Brain Development Cooperative, Reference Almli, Rivkin and McKinstry2007; Ramenghi et al., Reference Ramenghi, Rutherford, Fumagalli, Bassi, Messner, Counsell and Mosca2009; Reddy, Abuhamad, Levine, Saade, & Fetal Imaging Workshop Invited, Reference Reddy, Abuhamad, Levine and Saade2014; Smyser, Snyder, & Neil, Reference Smyser, Snyder and Neil2011). Undeniably, several technical challenges still exist, as exemplified by head motion (Power, Barnes, Snyder, Schlaggar, & Petersen, Reference Power, Barnes, Snyder, Schlaggar and Petersen2012; Satterthwaite et al., Reference Satterthwaite, Elliott, Gerraty, Ruparel, Loughead, Calkins and Wolf2013; Yan, Craddock, He, & Milham, Reference Yan, Craddock, He and Milham2013) and the various logistical hurdles of imaging during the prenatal period, or during natural sleep (Di Martino, Fair, et al., Reference Di Martino, Fair, Kelly, Satterthwaite, Castellanos, Thomason and Milham2014). Nevertheless, pioneering studies have demonstrated the feasibility of longitudinally mapping developmental trajectories in clinical and non-clinical populations alike.
Beyond neuroscientific ambitions, the growing interest of the medical community in “personalized” health care and stratified (or “precision”) interventions, is further motivating the generation of longitudinal datasets. In “stratified psychiatry” (Kapur, Phillips, & Insel, Reference Kapur, Phillips and Insel2012), it is crucial to identify and validate objective measures or “stratifiers” capable of enhancing our ability to inform predictions of risk, prognosis, or treatment outcome, beyond that afforded by conventional diagnostic systems (e.g., DSM-5) or historical data. Such measures have the potential to guide clinical efforts focused on prevention, early intervention, and the selection of optimal treatments (Costa e Silva, Reference Costa e Silva2013; Schumann et al., Reference Schumann, Binder, Holte, de Kloet, Oedegaard, Robbins and Wittchen2014). The genetics community has led the way in such pursuits, although marked gaps remain. While imaging has the potential to meaningfully contribute to these efforts, the prerequisite longitudinal data are yet to be generated.
Despite their potential benefits, the scale with which longitudinal studies can be deployed in imaging research remains limited by their associated costs and time-requirements. Realizing the need to balance promise with an appreciation of logistical challenges, it is our tenet that targeting high-yield “at-risk” clinical populations and age-ranges is essential for efficacious and impactful implementations of longitudinal imaging studies. As will be elucidated below, risk can be defined using a range of variables, including genetics, familial factors, fetal exposures, environmental exposures, and current psychiatric symptomatology. Our discussion is organized into four topic areas: (1) a special role of connectomics in psychiatric research; (2) practical considerations for establishing temporal precedence in psychiatric research; (3) readiness of the field for conducting longitudinal MRI, particularly for neurodevelopmental questions; and (4) illustrations of high-yield populations and time windows for examination that can be used to rapidly generate meaningful and useful data.
A Special Role of Connectomics
The principles and frameworks discussed in the present work are broadly applicable to various brain imaging methodologies. In recent years, the human connectome has taken a central focus in efforts to map brain function and structure (e.g., Van Essen et al., Reference Van Essen, Smith, Barch, Behrens, Yacoub and Ugurbil2013). A growing number of studies are working to map changes in the connectome across the lifespan (Nooner et al., Reference Nooner, Colcombe, Tobe, Mennes, Benedict, Moreno and Milham2012), as well as to identify connectome-based markers of illness, risk and prognosis (see Castellanos et al., Reference Castellanos, Di Martino, Craddock, Mehta and Milham2013 for review; Di Martino, Yan, et al., 2014). The increased focus on the connectome allows for more direct appreciation of interactions among brain regions, particularly between cortical and subcortical regions. Importantly, such complex interactions contribute to the occurrence of psychiatric illness, a concept originally espoused by Norman Geschwind in his 1965 papers entitled “Disconnexion syndromes in animals and man” (Geschwind, Reference Geschwind1965a, Reference Geschwind1965b). Recent advances in MRI imaging modalities are providing researchers with the critical tools needed for in vivo mapping of structural and functional brain connectivity underlying pathological symptoms. Although still in their infancy, studies of connectomes are rapidly extending models of psychiatric disorders, to integrate the concepts of hypo- and hyper-connectivity, as well as more complex disturbances embedded in the connectome (e.g., autism: Di Martino, Yan, et al., Reference Di Martino, Yan, Li, Denio, Castellanos, Alaerts and Milham2014; ADHD: Fair et al., Reference Fair, Posner, Nagel, Bathula, Dias, Mills and Nigg2010; schizophrenia: Whitfield-Gabrieli et al., Reference Whitfield-Gabrieli, Thermenos, Milanovic, Tsuang, Faraone, McCarley and Seidman2009).
Given its role in psychiatry, there has been an increased interest in connectome-based analyses in individuals at risk of a disorder. Initial longitudinal efforts are emerging, a few of which will be introduced in this review. These efforts have capitalized on the increasing capabilities of diffusion imaging (Ameis & Catani, Reference Ameis and Catani2015; White, Nelson, & Lim, Reference White, Nelson and Lim2008), white matter volumetrics (Radua, Via, Catani, & Mataix-Cols, Reference Radua, Via, Catani and Mataix-Cols2011), and resting-state fMRI (R-fMRI) (Castellanos et al., Reference Castellanos, Di Martino, Craddock, Mehta and Milham2013) approaches. Looking forward, advances in quantitative MRI (see Alexander et al., Reference Alexander, Hurley, Samsonov, Adluru, Hosseinbor, Mossahebi and Field2011, for review), particularly relaxometry (e.g., myelin water fraction) and magnetization transfer (myelin volume fraction, axon volume fraction, g-ratio) methods, can provide researchers with tools capable of providing in vivo characterizations of myelin in healthy and disordered brains. In light of recent models suggesting the contributions of dysregulated myelination to early-life psychiatric disorders (Bartzokis, Reference Bartzokis2004), these quantitative MRI techniques, which focus on characterization of myelin properties (e.g., sheath thickness), are particularly important in imaging the “at-risk” brain.
Establishing Temporal Precedence
Optimally, the study of a given illness will lead to an understanding of its etiology and pathophysiological mechanisms; in turn, this information can be used to identify modifiable targets for intervention. For example, phenylketonuria (PKU) is an autosomal recessive disorder, for which the causal mechanisms, downstream effects, and the timing and optimal treatments are well understood. PKU is caused by mutations in the phenylalanine hydroxylase gene; the optimal intervention is to identify neonates soon after birth using a screening blood test and promptly begin a low phenylalanine diet. Such early identification and interventions work well for diseases with a clear etiology that can be validated through objective means such as a blood test. In contrast, the origins of most psychiatric disorders are poorly understood; they are presumed to be complex and the result of interactions between genetic and environmental factors over time, rather than a single causal factor occurring at a single time point.
Longitudinal designs can help address causality by demonstrating temporal precedence (i.e., “the cause” precedes “the effect”). Given the presumed multifactorial nature of psychiatric illness, imaging studies are unlikely to identify most etiologic agent(s) or processes. However, imaging can identify relationships between particular neurophenotypes (i.e., prototypic brain organizations) and later behavioral manifestations of illness. In some instances, a given neurophenotype can either indicate an increased risk for the later onset of a disorder or serve as a predictor of prognosis, without any assertions regarding causality. In either the risk or prognostic scenario, such information has the potential to guide clinicians’ decision-making. As demonstrated in recent studies of depression (Liston et al., Reference Liston, Chen, Zebley, Drysdale, Gordon, Leuchter and Dubin2014; McGrath et al., Reference McGrath, Kelley, Dunlop, Holtzheimer, Craighead and Mayberg2014), researchers can attempt to explain variations in treatment response (i.e., responder vs. non-responders) based on neurophenotypic features assessed before treatment. Such distinctions are central to stratified medicine and can meaningfully impact clinical decision-making.
Research Readiness of Longitudinal MRI
While few would argue against the potential scientific value of longitudinal MRI studies of the developing brain, the many logistical challenges and substantial expenses prompt concerns regarding the field’s readiness to proceed. The recently awarded NIH Adolescent Brain Cognitive Development (ABCD) Study signals increased confidence in this regard. This effort is designed to determine the impact of substance use on the trajectories of brain, behavior, and cognitive development, while also identifying potential brain-based markers for risk. The cost of the proposed longitudinal examination of 10,000 nine-to-ten year old children over the next 10 years is projected to be US $300,000,000; its implementation will bring together more than a dozen imaging sites from around the United States and its success will depend on a combination of careful coordination and state-of-the-art neuroinformatics. Despite the welcomed endorsement of the imaging community’s progress provided by the inception of the NIH ABCD Study, many questions remain as the community considers similar initiatives focused on other areas of psychiatry, or more technically challenging efforts focused on even earlier ages.
Logistical Challenges Related to Imaging: Reliability and Validity
While longitudinal imaging studies to date have primarily used T1-weighted morphometry, a growing array of imaging modalities has been suggested (e.g., R-fMRI, diffusion weighted MRI, arterial spin labeling, quantitative T1/T2 mapping, and myelin water fraction (Deoni, Dean, Remer, Dirks, & O’Muircheartaigh, Reference Deoni, Dean, Remer, Dirks and O’Muircheartaigh2015; Dubois et al., Reference Dubois, Dehaene-Lambertz, Kulikova, Poupon, Huppi and Hertz-Pannier2014; Gao et al., Reference Gao, Alcauter, Elton, Hernandez-Castillo, Smith, Ramirez and Lin2015; Krogsrud et al., Reference Krogsrud, Fjell, Tamnes, Grydeland, Mork, Due-Tonnessen and Walhovd2015; Sadeghi et al., Reference Sadeghi, Prastawa, Fletcher, Wolff, Gilmore and Gerig2013; Smyser et al., Reference Smyser, Inder, Shimony, Hill, Degnan, Snyder and Neil2010; Walker et al., Reference Walker, Chang, Nayak, Irfanoglu, Botteron and McCracken2016; Wang et al., Reference Wang, Fernandez-Seara, Alsop, Liu, Flax, Benasich and Detre2008). Given that scan session durations are limited by the tolerability of the scanner environment (particularly for developing and clinical populations) and cost (~$500–$700 per hr + manpower costs), investigators must determine which imaging modalities to include, and how to prioritize them, based upon readiness of these modalities for study at the present time. Central to assessments of the readiness for longitudinal examination are determinations of their reliability and validity of data collected for these imaging modalities (e.g., Finn et al., Reference Finn, Shen, Scheinost, Rosenberg, Huang, Chun and Constable2015; Jovicich et al., Reference Jovicich, Marizzoni, Bosch, Bartres-Faz, Arnold and Benninghoff2014; Kuhn et al., Reference Kuhn, Gullett, Nguyen, Boutzoukas, Ford, Colon-Perez and Bauer2015; Mueller et al., Reference Mueller, Wang, Fox, Pan, Lu, Li and Liu2015; Zuo et al., Reference Zuo, Anderson, Bellec, Birn, Biswal, Blautzik and Milham2014). In particular, test–retest reliability is an essential prerequisite for longitudinal studies, as experimental error in measurements from one time point to the next inherently limits the detectability of meaningful time-related changes. Despite the best of intentions, hardware and software upgrades commonly occur during the conduct of studies; any such changes should be carefully denoted in datasets and considered at the analytic level when possible. Phantom measurements and comprehensive quality control assessments (NessAiver, NessAiver, Harms, & Xu, Reference NessAiver, NessAiver, Harms and Xu2015) can be used to alleviate potential concerns related to these events.
Although arguably less essential for single-site studies, inter-scanner reliability should also be considered for longitudinal examinations. It is generally accepted that the best solution to the maximization of inter-scanner reliability is standardization to a single scanner type and protocol (see Brain Genomics Superstruct Project, Holmes et al., Reference Holmes, Hollinshead, O’Keefe, Petrov, Fariello, Wald and Buckner2015, for an example). However, efforts, such as the Pediatric Imaging, Neurocognition, and Genetics (PING) (Jernigan et al., Reference Jernigan, Brown, Hagler, Akshoomoff, Bartsch and Newman2015), have demonstrated it is possible to minimize protocol differences across scanner types (e.g., Siemens, GE, Philips). Additionally, a growing number of post hoc statistical correction approaches is becoming capable of minimizing the impact of scanner differences (Yan, Craddock, Zuo, Zang, & Milham, Reference Yan, Craddock, Zuo, Zang and Milham2013). Importantly, for any multicenter design, it is critical to ensure that variables of interest do not vary across scanners (e.g., all ADHD participants imaged on scanner A but all typical comparisons on scanner B).
Putting the Pieces Together from Fetus to Childhood
Enthusiasm has been building for imaging brain development from the fetus to childhood and beyond. However, we must acknowledge the challenges inherent to early life imaging methodologies. For example, the state of wakefulness during which children are scanned varies with age for task-independent functional MRI. With sufficient pre-scan habituation and training, children ages 5 years and older can usually cooperate with rest scans while awake. However, younger children (e.g., infants or toddlers) are unable to remain sufficiently still while awake, but can be scanned during natural sleep (see Di Martino, Fair, et al., Reference Di Martino, Fair, Kelly, Satterthwaite, Castellanos, Thomason and Milham2014 for a review). While functional connectivity is known to vary systematically between awake and sleep states (Barttfeld et al., Reference Barttfeld, Uhrig, Sitt, Sigman, Jarraya and Dehaene2015; Fukunaga et al., Reference Fukunaga, Horovitz, van Gelderen, de Zwart, Jansma, Ikonomidou and Duyn2006; Uehara et al., Reference Uehara, Yamasaki, Okamoto, Koike, Kan, Miyauchi and Tobimatsu2014), as well as between eyes-open and eyes-closed states (Wang, Li, Xu, & Ding, Reference Wang, Li, Xu and Ding2015; Zuo et al., Reference Zuo, Anderson, Bellec, Birn, Biswal, Blautzik and Milham2014), it is not clear how these differences impact the reliability of intra- and inter-individual differences during a developmental time course. This is one of the challenging but important questions that can be addressed by longitudinal MRI studies.
Non-imaging Logistical Challenges
Discussions of the readiness for large-scale longitudinal studies tend to focus most heavily on the reliability and validity of MRI measurements. However, it is equally important to take into account the behavioral, cognitive, psychiatric and physical characterizations to which MRI measures will be linked. Similar to the genetics community, imaging researchers are increasingly highlighting the need for “deep” (i.e., precise and comprehensive) phenotyping protocols (Loeffler et al., Reference Loeffler, Engel, Ahnert, Alfermann, Arelin, Baber and Thiery2015; Rafii et al., Reference Rafii, Wishnek, Brewer, Donohue, Ness, Mobley and Rissman2015; Stepniak et al., Reference Stepniak, Papiol, Hammer, Ramin, Everts, Hennig and Ehrenreich2014). Simultaneously, there is increasing pressure to obtain broader assessments, which provide a more holistic characterization of an individual (Nooner et al., Reference Nooner, Colcombe, Tobe, Mennes, Benedict, Moreno and Milham2012), although often at the cost of additional burden on participants and experimenters. Additional trends are an increased focus on the adoption of dimensional characterizations of behaviors and symptomatology, which can be used to transcend DSM-5 diagnostic classifications. A multitude of measurements are emerging for all of these domains, ranging from self-administered, clinician-administered, informant-based assessments, to more objective measures (e.g., device, computer, or laboratory measurements).
Selection of High-Yield Target Populations and Appropriate Time Windows for Longitudinal Examination
How can we best design a longitudinal study? Should all longitudinal studies begin at birth, since many psychiatric disorders have early onsets and strong genetic influences? The answer may be “Yes” if a research question is etiological (i.e., causal links). However, from an implementation and timeliness perspective, this would likely be unrealistic and impractical for the vast majority of MRI studies. Critically, the ability of longitudinal MRI examinations to advance the stratified psychiatry agenda will depend upon (1) strategic identification of high-yield targets (e.g., risk populations), and (2) the selection of the appropriate time windows for examination. Here, we provide a targeted review of psychiatric imaging and epidemiologic studies that can help to build longitudinal MRI research models, with a particular focus on how to determine the appropriate time window(s) to answer specific research questions.
Autistic Spectrum Disorder
ASD poses a range of challenges for the healthcare professions, which exemplify the complexities that can arise in the detection and treatment of neurodevelopmental disorders. The first challenge is the task of diagnosing autism. Beyond the core domains (i.e., social-communicative, repetitive and restricted interests/behaviors), ASD is characterized by marked clinical heterogeneity/comorbidity (Georgiades et al., Reference Georgiades, Szatmari, Boyle, Hanna, Duku and Zwaigenbaum2013; Geschwind, Reference Geschwind2009; Jeste & Geschwind, Reference Jeste and Geschwind2014; Matson & Williams, Reference Matson and Williams2013; Wiggins, Robins, Adamson, Bakeman, & Henrich, Reference Wiggins, Robins, Adamson, Bakeman and Henrich2012), which can lead to confusion and delays in diagnosis. The second challenge is the need to identify those most at risk for developing the disorder as early as possible, to create opportunities for early intervention. The third challenge is determining prognosis, which often cannot be determined until years after the initial diagnosis, except for the most severely affected cases. This is particularly critical for autism, for which long-term outcomes vary widely; some individuals experience substantial improvements as they mature, while others face significant, lifelong impairment, particularly those who are non-verbal (Bal, Kim, Cheong, & Lord, Reference Bal, Kim, Cheong and Lord2015; Pickles, Anderson, & Lord, Reference Pickles, Anderson and Lord2014). It is the latter two challenges, focused on the establishment of risk and prognosis, which are best positioned to benefit from longitudinal research designs. Efforts aiming to improve the diagnostic process (e.g., identification of distinct neurophenotypic subtypes, identification of biomarkers of illness that can inform complex presentations) may also benefit from longitudinal data, although less directly.
MRI Predictors of Risk
As with other neurodevelopmental disorders in psychiatry, researchers studying ASD are seeking a range of predictive biomarkers, capable of informing assessments of risk. Initial work with brain imaging suggests its potential utility in studying infants at risk (i.e., younger sibling of a child with ASD) (Elison et al., Reference Elison, Paterson, Wolff, Reznick, Sasson, Gu and Network2013; Wolff et al., Reference Wolff, Gerig, Lewis, Soda, Styner and Vachet2015, Reference Wolff, Gu, Gerig, Elison, Styner and Gouttard2012). ASD is highly heritable; for example, monozygotic twin concordance ranges between 0.77 and 0.99 (Colvert et al., Reference Colvert, Tick, McEwen, Stewart, Curran, Woodhouse and Bolton2015), and 18.7% of younger siblings develop ASD (Ozonoff et al., Reference Ozonoff, Young, Carter, Messinger, Yirmiya, Zwaigenbaum and Stone2011). Appreciating the strong genetic contributions to ASD, longitudinal efforts have begun to scan infants with a familial risk of ASD to identify early predictors of later ASD development, as well as early mechanisms of disease, compensation and protective factors. Typically these studies are designed to follow children at both high- and low-risk (i.e., siblings of children with and without an older sibling affected by ASD, respectively) from the first few months of life to at least age 36 months, when the diagnosis of ASD is considered stable. Initial findings have suggested that early-life MRI indices can predict later ASD diagnosis by age 2 or later.
Example MRI indices include larger total cerebral volume at 12–15 months (Shen et al., Reference Shen, Nordahl, Young, Wootton-Gorges, Lee, Liston and Amaral2013), excessive frontal cerebrospinal fluid at 6–9 and 12–15 months (Shen et al., Reference Shen, Nordahl, Young, Wootton-Gorges, Lee, Liston and Amaral2013), and higher white matter fractional anisotropy (FA) values at 6 months (Wolff et al., Reference Wolff, Gu, Gerig, Elison, Styner and Gouttard2012), and increased cortical thickness in the corpus callosum in the first year of life (Wolff et al., Reference Wolff, Gerig, Lewis, Soda, Styner and Vachet2015). Most recently, Blasi et al. (Reference Blasi, Lloyd-Fox, Sethna, Brammer, Mercure, Murray and Johnson2015) quantified differences in fMRI activation in response to human voice sounds and non-voice sounds between high- and low-risk ASD groups during natural sleep: 4- to 7-month-old infants at high risk of ASD exhibited reduced sensitivity to human voices, relative to those at low risk of ASD, during natural sleep. This result suggests the presence of early disturbances in perceptual processing of human voices, which are the most basic and crucial acoustic stimuli in the social environment, will impact later social and emotional development. These neuroimaging findings suggest that longitudinal MRI may serve not only to provide clues for potential biomarkers but also to reveal pathologic mechanisms.
MRI Predictors of Prognosis
Another potentially important application of longitudinal MRI is the prediction of prognosis (i.e., to predict later behavioral and functional outcomes). One example would be the ability to predict whether a child diagnosed with ASD at age 24 months will develop more than minimal verbal skills by age 60 months, which in turn is an important predictor of later-life behavioral and social outcomes (Baghdadli et al., Reference Baghdadli, Picot, Michelon, Bodet, Pernon, Burstezjn and Aussilloux2007; Billstedt, Gillberg, & Gillberg, Reference Billstedt, Gillberg and Gillberg2005; Howlin, Goode, Hutton, & Rutter, Reference Howlin, Goode, Hutton and Rutter2004; Szatmari et al., Reference Szatmari, Merette, Bryson, Thivierge, Roy, Cayer and Maziade2002). At present, the predictive value of behavioral assessment during early toddlerhood (e.g., joint attention, repetitive behavior, and nonverbal IQ at 2 years) for later language outcomes is limited to identifying ASD children with severe impairments (Anderson et al., Reference Anderson, Lord, Risi, DiLavore, Shulman, Thurm and Pickles2007; Pickles et al., Reference Pickles, Anderson and Lord2014; Wolff et al., Reference Wolff, Botteron, Dager, Elison, Estes, Gu and Network2014). This leaves a gap in the ability to predict subsequent verbal development for less severely affected children with ASD; this gap might be addressable by longitudinal MRI studies of toddlers.
A recent longitudinal examination demonstrated that pre-diagnosis fMRI activity at ages 12–48 months differentiated language deficits 1 year later in children with ASD (Lombardo et al., Reference Lombardo, Pierce, Eyler, Barnes, Ahrens-Barbeau, Solso and Courchesne2015). More specifically, ASD children with poor language outcomes exhibited reduced activation in language-sensitive superior temporal regions in response to speech when compared to ASD toddlers with relatively good language outcomes; activation in this ASD group was also found to be reduced when compared to typically developing children and language-delayed non-ASD children. This finding suggests the potential value of MRI assessment during toddlerhood for identifying objective brain-based prognostic markers of language development in ASD. If successfully obtained, such markers would have the potential to inform decisions regarding the deployment of early intervention efforts, as well as to assess brain responses to intervention.
Pediatric Anxiety Disorders
Anxiety disorders are the most prevalent forms of psychiatric illness in children and adolescents. From a stratified psychiatry point of view, pediatric anxiety disorders are of interest because: (1) they predict risk for later onset of a second psychiatric illness (e.g., depressive disorder); and (2) their origins appear to lie in the fetal and/or perinatal period. Thus, pediatric anxiety disorders are another example of a potentially rich target for longitudinal evaluation in imaging studies.
Pediatric anxiety disorders as predictors of later depression
Pediatric anxiety disorders are believed to contribute to, or at a minimum signal the likelihood of, the development of secondary psychiatric complications later in life (Woodward & Fergusson, Reference Woodward and Fergusson2001). Cross-sectional and longitudinal studies have consistently identified the presence of anxiety disorders in childhood and early adolescence as a strong predictor of adult depressive symptoms (Beesdo, Pine, Lieb, & Wittchen, Reference Beesdo, Pine, Lieb and Wittchen2010; Kessler et al., Reference Kessler, Avenevoli and Merikangas2001; Pine, Cohen, Gurley, Brook, & Ma, Reference Pine, Cohen, Gurley, Brook and Ma1998; Silberg, Rutter, Neale, & Eaves, Reference Silberg, Rutter, Neale and Eaves2001; Silk, Davis, McMakin, Dahl, & Forbes, Reference Silk, Davis, McMakin, Dahl and Forbes2012; Weissman, Fendrich, Warner, & Wickramaratne, Reference Weissman, Fendrich, Warner and Wickramaratne1992; Weissman et al., Reference Weissman, Wickramaratne, Nomura, Warner, Verdeli, Pilowsky and Bruder2005). While studies looking at pubertal/adolescent populations have found links between subclinical depressive symptoms (e.g., fatigue, poor concentration) and the later development of major depression in adulthood (Hill, Pettit, Lewinsohn, Seeley, & Klein, Reference Hill, Pettit, Lewinsohn, Seeley and Klein2014; Pine, Cohen, Cohen, & Brook, Reference Pine, Cohen, Cohen and Brook1999), findings from pre-pubertal studies are less clear in such links. In fact, some have found that previously observed relationships between pre-pubertal-onset depression and recurrence of depression in adulthood do not survive after taking into account the presence of anxiety and externalizing disorders (Weissman et al., Reference Weissman, Wolk, Wickramaratne, Goldstein, Adams, Greenwald and Steinberg1999).
Consistent with proposed commonalities in symptoms, shared structural and functional correlates of both anxiety and depressive disorders have been reported (Bremner, Reference Bremner2004; Bruhl, Delsignore, Komossa, & Weidt, Reference Bruhl, Delsignore, Komossa and Weidt2014; Etkin & Wager, Reference Etkin and Wager2007; Kaiser, Andrews-Hanna, Wager, & Pizzagalli, Reference Kaiser, Andrews-Hanna, Wager and Pizzagalli2015; Miller, Hamilton, Sacchet, & Gotlib, Reference Miller, Hamilton, Sacchet and Gotlib2015; Schmaal et al., Reference Schmaal, Veltman, van Erp, Samann, Frodl, Jahanshad and Hibar2015). These mostly include limbic structures (e.g., amygdala) involved in emotion and its regulation, as well as anterior cingulate cortex involved in cognitive control (see Ray & Zald, Reference Ray and Zald2012 for a review). For example, anxiety disorders are associated with abnormalities in MRI metrics of amygdala (e.g., gray matter volume, functional connectivity) not only in adults (Basser & Pierpaoli, Reference Basser and Pierpaoli1996; Robinson et al., Reference Robinson, Krimsky, Lieberman, Allen, Vytal and Grillon2014; Roy et al., Reference Roy, Fudge, Kelly, Perry, Daniele, Carlisi and Ernst2013), but also in young children (i.e., 7–9 and 12 years old) (Milham et al., Reference Milham, Nugent, Drevets, Dickstein, Leibenluft, Ernst and Pine2005; Qin et al., Reference Qin, Young, Duan, Chen, Supekar and Menon2014). For depressive disorders, nearly all MRI studies that focus on adults also highlight alterations (e.g., gray matter volume, functional connectivity) in the ventral anterior cingulate cortex (Bora, Harrison, Davey, Yucel, & Pantelis, Reference Bora, Harrison, Davey, Yucel and Pantelis2012; Botteron, Raichle, Drevets, Heath, & Todd, Reference Botteron, Raichle, Drevets, Heath and Todd2002; Carballedo et al., Reference Carballedo, Scheuerecker, Meisenzahl, Schoepf, Bokde, Moller and Frodl2011; Drevets, Savitz, & Trimble, Reference Drevets, Savitz and Trimble2008; Hastings, Parsey, Oquendo, Arango, & Mann, Reference Hastings, Parsey, Oquendo, Arango and Mann2004; Ho et al., Reference Ho, Yang, Wu, Cassey, Brown, Hoang and Yang2014; Price & Drevets, Reference Price and Drevets2012; Yan, Cheung, et al., Reference Yan, Cheung, Kelly, Colcombe, Craddock, Di Martino and Milham2013). This region is tightly coupled to amygdala function (Pezawas et al., Reference Pezawas, Meyer-Lindenberg, Drabant, Verchinski, Munoz, Kolachana and Weinberger2005; Roy et al., Reference Roy, Shehzad, Margulies, Kelly, Uddin, Gotimer and Milham2009) and commonly shows abnormalities in studies of anxiety as well (Bruhl et al., Reference Bruhl, Delsignore, Komossa and Weidt2014; Drevets et al., Reference Drevets, Savitz and Trimble2008).
From the viewpoint of longitudinal study design, individuals affected by pediatric anxiety disorders represent a valuable target for imaging studies aiming to identify biomarkers to predict adult depressive disorders. Additionally, when combined with intervention trials (e.g., cognitive behavioral therapy, serotonin selective inhibitors), longitudinal studies have the ability to address questions regarding the impact of interventions on the neural and behavioral trajectories of those at risk for later development of depression.
To date, there are no published longitudinal MRI studies mapping neural trajectories of individuals diagnosed with pediatric anxiety disorders, in an effort to study later life depression. However, using an alternative approach to studying the development of depression, recent work comparing neural development in children of depressed parents (high-risk) with those of non-depressed parents (low-risk), has provided initial insights. The high-risk group in this work is at increased risk of developing anxiety and depressive disorders relative to the low-risk group (Foland-Ross, Hardin, & Gotlib, Reference Foland-Ross, Hardin and Gotlib2013; Tau & Peterson, Reference Tau and Peterson2010; Weissman et al., Reference Weissman, Wickramaratne, Nomura, Warner, Verdeli, Pilowsky and Bruder2005). For example, a cross-sectional study conducted by Foland-Ross, Gilbert, et al. (Reference Foland-Ross, Gilbert, Joormann and Gotlib2015) demonstrated that diagnosis-free girls (9–17 years old) in a high-risk group (born to mothers with recurrent depression) had cortical thinning in the anterior cingulate cortex, which was in turn associated with greater difficulty managing sadness. In a longitudinal extension (Foland-Ross, Sacchet, et al., Reference Foland-Ross, Sacchet, Prasad, Gilbert, Thompson and Gotlib2015), alterations in cortical thickness, particularly cortical thinning in the orbitofrontal cortex, at ages 10–15 reliably predicted the subsequent onset of depression (i.e., 5 years later). This pioneering work illustrates that longitudinal MRI research can play a role in identifying biomarkers that can better predict individuals at high risk for developing depression.
Focus on fetal origins
A growing literature is documenting associations between early neurodevelopmental stressors and the eventual development of psychiatric illness—particularly pediatric anxiety disorders (Dawson, Ashman, & Carver, Reference Dawson, Ashman and Carver2000; Glover, Reference Glover2014). Accumulating evidence from animal and human studies shows that prenatal maternal psychological stress (e.g., anxiety, depression) is a risk factor for anxiety disorders in offspring (Buss, Entringer, Swanson, & Wadhwa, Reference Buss, Entringer, Swanson and Wadhwa2012; Glover, Reference Glover2014; Hay, Pawlby, Waters, Perra, & Sharp, Reference Hay, Pawlby, Waters, Perra and Sharp2010; Kingston, Tough, & Whitfield, Reference Kingston, Tough and Whitfield2012; Monk, Spicer, & Champagne, Reference Monk, Spicer and Champagne2012; O’Connor, Monk, & Fitelson, Reference O’Connor, Monk and Fitelson2014; Van den Bergh & Marcoen, Reference Van den Bergh and Marcoen2004). For example, prospective longitudinal studies show that levels of self-reported maternal stress and anxiety during pregnancy predict atypical behavioral/emotional development in infants, such as greater cry responses to novelty (Petzoldt et al., Reference Petzoldt, Wittchen, Wittich, Einsle, Hofler and Martini2014; Werner et al., Reference Werner, Myers, Fifer, Cheng, Fang, Allen and Monk2007) and cognitive delay (Davis & Sandman, Reference Davis and Sandman2010; Huizink, de Medina, Mulder, Visser, & Buitelaar, Reference Huizink, de Medina, Mulder, Visser and Buitelaar2002). In turn, such atypical behavioral/emotional responses during infancy predict the development of anxiety disorders in childhood and adolescent (Perez-Edgar & Fox, Reference Perez-Edgar and Fox2005).
Retrospective longitudinal examinations have begun linking prenatal maternal stress to MRI-based brain outcomes in childhood. Sandman, Buss, Head, & Davis (Reference Sandman, Buss, Head and Davis2015) demonstrated significant cortical thinning in children (ages: 6–9 years) with a history of prenatal exposure to maternal depressive symptoms. In particular, maternal depression at 25-weeks gestational age was strongly associated with cortical thinning in the prefrontal cortex, which is coupled with childhood externalizing symptoms. In parallel, prospective longitudinal efforts linking prenatal maternal stress to brain and behavioral developments in offspring of stressed mothers during pregnancy are beginning to emerge. Initial work has successfully linked the increased level of prenatal maternal anxiety to lower values of white matter structural metrics (e.g., FA) at birth (Rifkin-Graboi et al., Reference Rifkin-Graboi, Bai, Chen, Hameed, Sim, Tint and Qiu2013), as well as slower volumetric growth of the hippocampus over the first 6 months (Qiu et al., Reference Qiu, Rifkin-Graboi, Chen, Chong, Kwek, Gluckman and Meaney2013). Furthermore, Buss, Davis, et al. (Reference Buss, Davis, Shahbaba, Pruessner, Head and Sandman2012) report adverse effects of prenatal exposure to maternal stress hormone on amygdala volume and affective behaviors at the age of 7.
The aforementioned retrospective and prospective longitudinal studies point to the potential benefit of prenatal/fetal MRI studies for understanding the impact of fetal programming on the development of later life illness. Related to prenatal anxiety, researchers are actively working to clarify the effect of prenatal exposure to selective serotonin reuptake inhibitors (SSRIs)—the most commonly prescribed antidepressants for anxiety disorders and depressive disorders. While epidemiologic studies alone remain a source of confusion due to inconsistences in findings (Eriksen, Kesmodel, Pedersen, & Mortensen, Reference Eriksen, Kesmodel, Pedersen and Mortensen2015; Odsbu et al., Reference Odsbu, Skurtveit, Selmer, Roth, Hernandez-Diaz and Handal2015; Ornoy & Koren, Reference Ornoy and Koren2014), future longitudinal imaging studies focused on the prenatal and neonatal periods may be able to help inform this controversy. In particular, they can provide a mechanistic understanding of possible disturbances associated with SSRIs, if present. Arguably, imaging efforts focused on non-human samples have the greatest potential to inform our understanding of causality, given the ability to experimentally manipulate exposures.
A crucial question for any efforts focused on the prenatal period, is whether fetal MRI is reliably applicable to longitudinal examination (Anderson & Thomason, Reference Anderson and Thomason2013; Limperopoulos & Clouchoux, Reference Limperopoulos and Clouchoux2009; Saleem, Reference Saleem2014; Welsh, Nemec, & Thomason, Reference Welsh, Nemec and Thomason2011). Fetal MRI requires interactive scanning to account for a moving fetus (Ferrazzi et al., Reference Ferrazzi, Kuklisova Murgasova, Arichi, Malamateniou, Fox, Makropoulos and Hajnal2014; Seshamani, Cheng, Fogtmann, Thomason, & Studholme, Reference Seshamani, Cheng, Fogtmann, Thomason and Studholme2014) and the respiratory motion of the mother (Liu, Glenn, & Xu, Reference Liu, Glenn and Xu2014). Although fetal MRI remains practically challenging and limited in its availability, recent advances are beginning to reveal the structural (Jakab et al., Reference Jakab, Kasprian, Schwartz, Gruber, Mitter, Prayer and Langs2015; Kasprian et al., Reference Kasprian, Langs, Brugger, Bittner, Weber, Arantes and Prayer2011) and functional (Jakab et al., Reference Jakab, Schwartz, Kasprian, Gruber, Prayer, Schopf and Langs2014; Schopf, Kasprian, Brugger, & Prayer, Reference Schopf, Kasprian, Brugger and Prayer2012; Thomason et al., Reference Thomason, Brown, Dassanayake, Shastri, Marusak, Hernandez-Andrade and Romero2014, Reference Thomason, Dassanayake, Shen, Katkuri, Alexis, Anderson and Romero2013, Reference Thomason, Grove, Lozon, Vila, Ye, Nye and Romero2015) organization of the brain during this highly dynamic period. Understanding typical fetal brain development is vital for identifying sensitive periods of vulnerability, in which environmental factors (e.g., exposure to maternal stress) impact most on later psychiatric and neurological disorders. Obtaining such knowledge would be valuable when determining the timing of optimal/effective intervention for these disorders.
Preterm Birth
Complementing the growing associations between pediatric anxiety disorders and fetal stress is the MRI examination of individuals born preterm. Although life-preserving technologies have raised survival rates up to 85% (Smith, Draper, & Field, Reference Smith, Draper and Field2014), the preservation of life does not assure typical, healthy development. Individuals with a history of preterm birth have three to five times increased risk of neurodevelopmental disorders in childhood (Behrman & Butler, Reference Behrman and Butler2007; Glass et al., Reference Glass, Costarino, Stayer, Brett, Cladis and Davis2015), such as ASD, anxiety disorders, and attention deficit hyperactivity disorder (Johnson & Marlow, Reference Johnson and Marlow2011; Treyvaud et al., Reference Treyvaud, Ure, Doyle, Lee, Rogers, Kidokoro and Anderson2013). This elevated risk mostly continues into adolescence and adulthood (Aarnoudse-Moens, Weisglas-Kuperus, van Goudoever, & Oosterlaan, Reference Aarnoudse-Moens, Weisglas-Kuperus, van Goudoever and Oosterlaan2009; Allin et al., Reference Allin, Walshe, Fern, Nosarti, Cuddy, Rifkin and Wyatt2008; Barre, Morgan, Doyle, & Anderson, Reference Barre, Morgan, Doyle and Anderson2011).
Based on clinical vulnerability of this population, an increasing number of MRI studies has retrospectively examined individuals with a history of preterm birth (i.e., toddlers, children, adolescents, and adults), revealing alterations in multiple MRI metrics (Lubsen et al., Reference Lubsen, Vohr, Myers, Hampson, Lacadie, Schneider and Ment2011), even in individuals born late preterm (Degnan et al., Reference Degnan, Wisnowski, Choi, Ceschin, Bhushan, Leahy and Panigrahy2015). These include gray matter volume (Ment, Kesler, et al., Reference Ment, Kesler, Vohr, Katz, Baumgartner, Schneider and Reiss2009); FA white matter microstructure assessed by DTI (Duerden, Card, Lax, Donner, & Taylor, Reference Duerden, Card, Lax, Donner and Taylor2013); intrinsic functional networks assessed by R-fMRI (Constable et al., Reference Constable, Ment, Vohr, Kesler, Fulbright, Lacadie and Reiss2008; Damaraju et al., Reference Damaraju, Phillips, Lowe, Ohls, Calhoun and Caprihan2010), and task-based fMRI activity/connectivity (Carmody et al., Reference Carmody, Bendersky, Dunn, DeMarco, Hegyi, Hiatt and Lewis2006; Frye et al., Reference Frye, Malmberg, Desouza, Swank, Smith and Landry2009; Myers et al., Reference Myers, Hampson, Vohr, Lacadie, Frost, Pugh and Ment2010 ). Some of these MRI studies highlight that the reported abnormalities in MRI metrics, particularly white matter microstructure (e.g., FA), are concurrently associated with neurodevelopmental deficits in children and adolescents born preterm. Examples include mental development at the age of 2 (Counsell et al., Reference Counsell, Edwards, Chew, Anjari, Dyet, Srinivasan and Cowan2008), IQ at the age of 12 (Constable et al., Reference Constable, Ment, Vohr, Kesler, Fulbright, Lacadie and Reiss2008), language-related skills at the age of 9–16 years (Feldman, Lee, Yeatman, & Yeom, Reference Feldman, Lee, Yeatman and Yeom2012; Mullen et al., Reference Mullen, Vohr, Katz, Schneider, Lacadie, Hampson and Ment2011) and 16 years, and short-term memory in adulthood (Finke et al., Reference Finke, Neitzel, Bauml, Redel, Muller, Meng and Sorg2015).
Most relevant to the present work, prospective MRI longitudinal studies following preterm infants from soon after birth have begun, providing valuable insights into developmental trajectories of this vulnerable population (see Kwon, Vasung, Ment, & Huppi, Reference Kwon, Vasung, Ment and Huppi2014; Ment, Hirtz, & Huppi, Reference Ment, Hirtz and Huppi2009). The vast majority of studies have reported the presence of structural (Brown et al., Reference Brown, Miller, Booth, Andrews, Chau, Poskitt and Hamarneh2014; Kersbergen et al., Reference Kersbergen, Leemans, Groenendaal, van der Aa, Viergever, de Vries and Benders2014; Young et al., Reference Young, Powell, Morgan, Card, Lee, Smith and Taylor2015) and functional (Smyser et al., Reference Smyser, Snyder and Neil2011) MRI aberrancies at birth or during infancy, which are associated with subsequent neurodevelopmental outcomes (Rose et al., Reference Rose, Butler, Lamont, Barnes, Atlas and Stevenson2009; Young et al., Reference Young, Powell, Morgan, Card, Lee, Smith and Taylor2015). Future longitudinal MRI research can target this high-yield “at-risk” clinical population, scanning from birth, and over the course of time, for providing more precise information regarding prediction of risk and prognosis for neurodevelopmental disorders.
Specific Learning Disabilities (SLDs)
Specific learning disabilities (SLDs) are uniquely positioned among the various DSM-5 disorders, as they can be readily quantified using a broad range of standardized tests. That is, SLDs are reliably defined by impaired performance, relative to norms, in reading, writing, and/or mathematics. Despite the availability of objective assessments, similar to other psychiatric disorders, SLDs are characterized by marked heterogeneity in clinical presentations (McArthur et al., Reference McArthur, Kohnen, Larsen, Jones, Anandakumar, Banales and Castles2013; Ramus et al., Reference Ramus, Rosen, Dakin, Day, Castellote, White and Frith2003), as well as a high degree of comorbid illness (Germano, Gagliano, & Curatolo, Reference Germano, Gagliano and Curatolo2010; Willcutt et al., Reference Willcutt, Betjemann, McGrath, Chhabildas, Olson, DeFries and Pennington2010, Reference Willcutt, Petrill, Wu, Boada, Defries, Olson and Pennington2013). Among SLDs, reading disability (i.e., dyslexia) has received the most attention in the MRI literature, with a range of studies examining the neural bases of risk (Raschle, Chang, & Gaab, Reference Raschle, Chang and Gaab2011; Raschle, Stering, Meissner, & Gaab, Reference Raschle, Stering, Meissner and Gaab2014; Saygin et al., Reference Saygin, Norton, Osher, Beach, Cyr, Ozernov-Palchik and Gabrieli2013), current impairments (Eckert et al., Reference Eckert, Leonard, Richards, Aylward, Thomson and Berninger2003; Richlan, Kronbichler, & Wimmer, Reference Richlan, Kronbichler and Wimmer2011, Reference Richlan, Kronbichler and Wimmer2013), and responses to intervention (e.g., Barquero, Davis, & Cutting, Reference Barquero, Davis and Cutting2014). The detection of neural changes associated with widely accepted phonology-based treatments has gained particular attention; convergent results suggest the normalization of dyslexia-associated abnormalities in functional activity within the known left-hemisphere reading network in children and adults (Aylward et al., Reference Aylward, Richards, Berninger, Nagy, Field, Grimme and Cramer2003; Eden et al., Reference Eden, Jones, Cappell, Gareau, Wood, Zeffiro and Flowers2004; Shaywitz et al., Reference Shaywitz, Shaywitz, Blachman, Pugh, Fulbright, Skudlarski and Gore2004; Simos et al., Reference Simos, Fletcher, Sarkari, Billingsley-Marshall, Denton and Papanicolaou2007; Temple et al., Reference Temple, Deutsch, Poldrack, Miller, Tallal, Merzenich and Gabrieli2003). Complementing these findings is work suggesting a compensatory reliance on alternative circuits outside the reading network in remediated dyslexic individuals (Eden et al., Reference Eden, Jones, Cappell, Gareau, Wood, Zeffiro and Flowers2004; Koyama et al., Reference Koyama, Di Martino, Kelly, Jutagir, Sunshine, Schwartz and Milham2013).
Identification of non-responders
Of particular relevance to the present work are findings that the interventions are not equally effective across individuals: 20–30% of children with (or at risk of) dyslexia do not respond adequately to generally effective interventions (see Brown & Felton, Reference Brown and Felton1999; Torgesen, Reference Torgesen2000 for a review; Torgesen et al., Reference Torgesen, Wagner, Rashotte, Rose, Lindamood, Conway and Garvan1999; Vellutino et al., Reference Vellutino, Scanlon, Sipay, Small, Pratt, Chen and Dencla1996). This gap could be addressed through the establishment of neurophenotypes at baseline that are predictive of treatment effectiveness, or demonstration of differential patterns of treatment-related brain changes associated with behavioral outcomes. Examples come from recent cross-sectional work suggesting that treatment responders and non-responders (ages 10–14 years) differ with respect to left inferior parietal lobule activation following a 4-week reading intervention (Odegard, Ring, Smith, Biggan, & Black, Reference Odegard, Ring, Smith, Biggan and Black2008).
Similarly, despite the lack of a single experimentally specified intervention, Hoeft et al. (2011) demonstrated that initial MRI-based profiles at age 14 years (i.e., right prefrontal activation during phonological processing and FA of right superior longitudinal fasciculus) predict later reading improvement in children with dyslexia at age 16.5 years. The next necessary step is longitudinal MRI examination to identify biomarkers predictive of treatment response (i.e., who will be responders vs. non-responders), which can be used to individualize intervention. For this, it is ideal to implement systematic interventions, ideally different types (e.g., attention-based) beyond phonology-based ones.
Similar to other neurodevelopmental disorders, the age of diagnosis in reading disorders should be reduced. At present, although commonly delayed, the signs and symptoms of dyslexia can be readily diagnosed soon after the official start of schooling (6–7 years old). However, as emphasized by a growing literature, behavioral signs and symptoms are detectable as early as in toddlerhood. Early detection of endophenotypes that are most predictive of dyslexia (e.g., phonological awareness, attention, and motor skills) (Gooch, Hulme, Nash, & Snowling, Reference Gooch, Hulme, Nash and Snowling2014; Thompson et al., Reference Thompson, Hulme, Nash, Gooch, Hayiou-Thomas and Snowling2015), and determination of optimal interventions based upon this information (e.g., phonological training vs. attention training), could rapidly change the trajectory of illness in children affected by dyslexia, before their school performance is negatively impacted.
Concluding Remarks
The goal of delineating trajectories for typical and atypical brain development across the first 2 decades of life is increasingly in our grasp. As highlighted in the present work, strategic selection of at-risk populations for examination can help to overcome the various logistical challenges inherent to longitudinal imaging examinations, which can otherwise hinder the pace of progress. Focusing on the potential of stratified psychiatry, we have provided illustrative examples of MRI examinations in autism spectrum disorder, pediatric anxiety disorders, and specific learning disorders. For each, we highlighted the scientific rationale and appropriate time windows for developmentally informed longitudinal designs to be implemented. We anticipate that biomarkers predictive of risk and prognosis derived from longitudinal examinations, as well as those capable of monitoring treatment outcomes, will advance the goals of early intervention and even prevention of psychiatric illness.
Acknowledgments
This work was supported in part by NIH U01MH099059 to M.P.M. and NIH 1R21MH102660-01 to A.D.M., along with gifts from Phyllis Green and Randolph Cowen to F.X.C. and M.P.M. and from Joseph P. Healey to M.P.M. The authors declare that there are no conflicts of interest.
