While great strides have been made in understanding age-related changes in brain processes underlying cognitive development, there are still important challenges that we need to address regarding methods and the interpretation of data. A crucial feature of developmental neuroimaging studies is that adults are considered the model system. Therefore, any deviation from adult brain activity is interpreted as an immaturity. This particular aspect of interpreting developmental results can be viewed by non-developmental investigators as inconsistent, and even opportunistic, in that both increases and decreases in brain activity in adolescents compared to adults are considered a reflection of immaturity. This is an aspect of cognitive neuroimaging that is present in all studies where comparisons are made between groups, be it developmental groups or aging and patient groups. In all these cases, it is important to recognize that the nature of the group differences may not be uniform across brain systems and that results must be interpreted in the context of a model and described in a manner where clear testable hypotheses can be made. We now discuss this issue and provide suggestions for interpretation.
The dependent measure in fMRI studies -- the blood oxygen level dependent (BOLD) response -- is the microvascular response in blood flow resulting from fluctuations in the metabolic needs of a population of neurons as they become involved in a task. Developmental studies investigate age-related differences in the magnitude and time evolution of this BOLD response. However, interpreting what these differences mean in the context of development is not well-established. We know there are maturational changes in brain structure during development that impact information processing, such as synaptic pruning and myelination, as well as other changes including development of neurotransmitter processing (Geier & Luna, 2009
) and hormonal influences (see other contributions to this issue). However, it is not immediately clear what the predicted change in BOLD response would be when considering these (i.e, pruning and myelination) and other (e.g., neurotransmitters, hormones) brain maturational processes. Synaptic pruning, the elimination of unused excess neuronal connections, may support more direct and less noisy computations allowing more efficient regional neural processing. This improved efficiency (decreased, but more direct neural processing) would presumably sustain more complex computations and lead to improved performance. However, how this age-related synaptic pruning would affect the BOLD response is not clear. On the one hand, fewer synaptic connections could have a lower metabolic need, resulting in a lower BOLD response in adults. This is consistent with the proposal that developmental changes follow a diffuse to focal trajectory, such as has been seen in PFC during some cognitive tasks (Casey, Trainor, Orendi, Schubert, ystrom et al., 1997
; Casey, Tottenham, Liston, & Durston, 2005
; Durston, Davidson, Tottenham, Galvan, Spicer et al., 2006
). On the other hand, synaptic pruning may support more complex
computations, and this may allow pruned regions to be recruited for a specific task that they would not support in the immature system. In this case, a region which is not involved in a task early in development may be evident in the adult system, resulting in increased activity in this region (Bunge, Wallis, Parker, Brass, Crone et al., 2005
; Tamm, Menon, & Reiss, 2002
Myelination, the thickening of the myelin sheath surrounding axons, speeds the neuronal transmission at both the regional and systems level. While myelination per se would not have a direct effect on the BOLD response, increased speed of neuronal transmission in local circuitry would enhance regional efficiency and support more complex computations locally, thus affecting the BOLD response as was suggested for synaptic pruning. Importantly, however, myelination can impact transmission at the systems level, allowing for the integration of widely distributed circuitries supporting top-down modulation of behavior. This top-down modulation may be essential for executive function (Goldman-Rakic, Chafee, & Friedman, 1993b
). A more distributed network in the adult could result in a lower BOLD response of executive regions such as PFC, while other regions, more specialized for the task at hand, are recruited. A related concern is that age-related changes in brain structure can undermine the ability to assess true developmental change. However the gross morphology of the brain is in place by mid-childhood (Goldman-Rakic, Chafee, & Friedman, 1993a
; Caviness, Jr., Kennedy, Richelme, Rademacher, & Filipek, 1996
; Giedd, Snell, Lange, Rajapakse, Casey et al., 1996
) so that size and organization of brain anatomy are roughly equivalent in adolescents and adults (Schlaggar, Brown, Lugar, Visscher, Miezin et al., 2002b
). There are also concerns that age-related vascular changes may undermine the ability to discern true differences in the BOLD response. However, studies comparing the bold response across the brain at different ages indicate that this is not a problem (Kang, Burgund, Lugar, Petersen, & Schlagger, 2003
). While gross structural changes of the brain are complete early in development, there are continued refinements in brain structure that are precisely the changes that many studies seek to identify.
It is not generally clear if changes in the BOLD response reflect age-related changes in brain processing, or differences in the use of strategies which recruit a distinct circuitry. For instance, adults may use verbal strategies to enhance cognitive performance. Therefore, it is possible that BOLD increases and decreases associated with performance are not related to age-related changes in brain processes per se, which may be mature, but instead to age-related differences in psychological processes. This is particularly a concern when performance differs by age. Differences in performance could reflect that the younger group uses a different strategy, with a distinct circuitry, from adults or that they utilize a comparable strategy as adults but use a suboptimal (immature) version of the mature circuitry. The former does not show developmental differences in brain function per se, but instead in psychological function, while the latter probably does reveal true developmental differences in brain function. However, while characterizing both changes is crucial for understanding development, distinguishing between these possibilities is not straightforward. While changes in behavior have long led to hypotheses about differences in brain function, it is now evident that changes in the pattern of brain function can provide insight into what psychological changes co-occur with development. In other words, the network of areas used in adolescence and how closely related it is to that used in adulthood provides insight to what psychological processes are changing.
Several approaches have been used to control for the effects of strategy use, which can confound the ability to characterize developmental changes. One approach is to equate performance across ages. This can provide important information about specific differences in brain circuitry that support similar performance at different ages (Schlaggar, Brown, Lugar, Visscher, Miezin et al., 2002b
). Developmental differences in brain function supporting equivalent performance can reflect that greater effort
is being exerted by the immature group, compensatory brain processes are being used due to limitations in accessing the correct circuitry, or that distinct strategies are being used. Characterizing differences in effort is important because it suggests that the basic circuitry is available but has not yet reached the mature processing level evident in the adult system. That is, the immature system may use more neural tissue or recruit this system for a longer period, to process the same neural computations as the mature system. This approach can also show alternate circuitries that are used as a compensatory mechanism as well as elucidate what parts of the circuitry are not being accessed. For example, young subjects’ IFG activation, associated with inhibitory processing, may not be able to support quick inhibitory responses. Instead parietal cortex, which supports control of attention and visuospatial processing, may be used by adolescents (Schlaggar, Brown, Lugar, Visscher, Miezin et al., 2002a
). However, care must be taken that a bias in sampling does not occur when performance is equated -- that is, that exceptional children are not being compared to adults with unusually poor performance.
Alternatively, evidence of age-related differences in performance can also elucidate how each age is performing the task, including the use of different strategies. If adolescents do not use the brain systems evident in adults, this may reflect limitations in accessing the mature circuitry, which leads to using a different compensatory circuitry or strategy. One way to reconcile such a difference (strategy use vs. true inability to access a given circuitry) is to use a parametric approach where cognitive load is manipulated, and the regions sensitive to these manipulations can be examined. On the one hand, younger subjects could show evidence of recruiting a given circuitry during low cognitive load but then use a different circuitry when load increases, indicating that the circuitry is in place but not mature enough to support tasks with high cognitive loads. On the other hand, younger subjects could show an inability to access a given circuitry regardless of cognitive load or performance differences, suggesting that they are using an alternative approach regardless of task difficulty, possibly due to limitations in accessing the optimal strategies and circuitry (Rubia, Smith, Taylor, & Brammer, 2007
Another approach to control for performance is to investigate differences in activation at the trial level. Even if there are age-related differences in performance across a task, examining correct and error trials separately can be used to equate performance and elucidate differences in brain function (Velanova, Wheeler, & Luna, 2008
). It is still possible that different strategies are being used to achieve similar overall performance. However, at the trial level, the same response is being generated and developmental differences in the circuitry supporting the same behavior can elucidate true age-related differences in brain processing. Training to a criterion can also be used to control for performance differences and, if a strategy is provided, it can also control for strategy use (Rubia, Smith, Taylor, & Brammer, 2007
). However, this technique may decrease the sensitivity to discriminate developmental differences important in every-day life. Differences in strategy use can also be seen as akin to learning. Neuroimaging studies of learning in adults indicate that the circuitry that is initially recruited when a task is novel is qualitatively different from the one that is recruited once the behavior has been learned (Ungerleider, Doyon, & Karni, 2002
). Similarly, cognitive control in the adolescent could approximate the “pre-learned” state evident in the adult. That is, the adult has more ease in exerting cognitive control than the adolescent, since some aspects of cognitive control have been “learned” and can be processed more automatically.
Direction of Developmental Differences
When age groups differ in activation, qualitatively different processes may be involved, each with distinct interpretations. One common result is that the younger group demonstrates an increased magnitude of activity compared to adults in an equivalent region. When the same circuitry as adults is being accessed by younger subjects, but the younger subjects show higher activity, this is often interpreted as indicating that greater effort
is required for younger subjects to do the task (Tamm, Menon, & Reiss, 2002
; Luna, Garver, Urban, Lazar, & Sweeney, 2004
). This interpretation is based on adult studies showing that activation increases with cognitive load (Keller, Carpenter, & Just, 2001
). On the other hand, results indicating that younger subjects show decreased activity in an equivalent region along with poor performance compared to adults, suggests that younger subjects are unable to access the mature and presumably optimal circuitry. Both results could indicate that the brain circuitry is available but immature, with the neural mechanisms not processing as efficiently
as they do in the mature system. Increased efficiency, on the one hand, can refer to adults being able to perform the same computations as younger subjects with less neural processing. On the other hand, increased efficiency could be interpreted as also underlying increased activity in adulthood, compared to younger subjects, reflecting that mature neural processing is able to support computations that the immature system can not yet support. That is, the adult may recruit neural systems that may not be accessed by younger subjects because their local circuitry is immature (for example, less-pruned) or connectivity is immature and cannot quickly access a region to support complex behavior. The former possibility implies that the basic circuitry is available to children but still immature and inefficient, hence children need to drive it to higher magnitudes, which is possibly related to effort. The latter result suggests that there are actual limitations prior to adulthood in the computational ability of the system to support higher levels of activity, undermining its use. Additionally, how to interpret increases and/or decreases in activation draws attention to the importance of the baseline task. Developmental differences in the baseline comparison task can undermine the ability to assess developmental changes in the experimental task (Marsh, Zhu, Schultz, Quackenbush, Royal et al., 2006b
Many of these concerns can be addressed, and insight into the meaning of developmental differences can be gained, by comparing the BOLD time courses generated from the regions of interest across age groups. Age-related differences in magnitude could be due to distinct differences in the BOLD response that would appear equivalent in an activation map but are qualitatively different, such as when both groups have a comparable timecourse but one has a higher magnitude of activity () or deactivation of the same region by one group () or a failure of one group to recruit the region (), or one group fails to exhibit a double peak (); or display a sustained response (). These different patterns of group effects could have distinct implications, for instance either a simple difference in effort () or a difference in the specific task component that is supported by that region, such as maintenance or response preparation (). In this manner, “increases” in activity can be quantified in relation to whether there is a decrease, lack of recruitment, or difference in the shape of the time courses between groups (Marsh, Zhu, Schultz, Quackenbush, Royal et al., 2006a
). Several statistical approaches can be used to test for age-related differences, including repeated measures analyses to test for magnitude differences (for each TR of the experiment), the magnitude of peak activity, the time of peak (younger subjects may have a delay in processing information in a region delaying the peak of the BOLD response), or the response shape (double peaks, or prolonged peaks reflecting sustained processes (Geier, Garver, & Luna, 2007a
; Geier, Garver, & Luna, 2007b
). Double peaks are still not well understood and are usually not considered. However, if these appear consistently across subjects within an age group and if there are typical time courses evident in other regions of the brain, the meaning of this response type should be addressed.
Figure 7 Idealized depiction of developmental changes in BOLD timecourses: a) similar time courses but one group shows lower magnitude; b) deactivation of the same region by one group; c) Only one group recruits a region; d) one group fails to exhibit a double (more ...)
Another methodological issue that affects how we interpret neuroimaging results is block vs. event-related designs. fMRI studies can be performed using a blocked design, where brain activity represents the collective activity of a block of trials, or an event-related design where brain activity is assessed at the single-trial level. The blocked design offers an optimal signal to identify the brain regions participating in a task and may help characterize a ‘response state’. However, correct and incorrect trials are grouped together. Event-related designs allow the characterization of brain function that underlies trial types (e.g., correct vs. incorrect; different cognitive loads). While this necessitates more trials, it allows for only correct trials (or incorrect, if there are enough) to be assessed, assuring that the comparisons between ages are more appropriate since the same behavior is being examined. A particularly fruitful approach is the mixed block event related fMRI design that permits the assessment of correct and incorrect trials as well as the block level processes that can reflect the status of response state (Velanova, Wheeler, & Luna, 2008
). As mentioned previously, differences in processes supporting the ability to retain a response state may be crucial to understanding developmental improvements in cognitive control during adolescence.
It is also important to balance the benefits of a theory-driven ROI analysis with the need for more exploratory analyses of development to identify distinct regions or increased variability in children compared to adults. Limiting studies to only investigating regions already implicated in the literature, such as PFC, undermines the ability to characterize system level changes that may affect integration of the PFC with other regions. Studies usually provide both approaches, allowing confirmation with the previous literature as well as extension to new areas. Finally, the ages chosen to represent different developmental stages can influence outcomes. Some studies group children and adolescents into a homogenous group (Rubia, Overmeyer, Taylor, Brammer, Williams et al., 2000
; Rubia, Smith, Woolley, Nosarti, Heyman et al., 2006
), limiting the ability to see stage-like differences in development that may be especially pertinent to the transition through adolescence (e.g., see below for “U” shaped function of executive regions in adolescence) and developmental differences in general. Following a large sample in a longitudinal fashion provides the most powerful and sensitive approach to characterizing different profiles of development, including stage-like phases and the nature of individual differences. However, there are drawbacks to these types of designs, since they are expensive and time-consuming, and take longer to produce results.