Neural Mechanisms Giving Rise to Diffuse-to-Focal and Local-to-Distributed Developmental Shifts

Two seemingly contradictory trends characterize brain development during childhood and adolescence:

Diffuse to focal: a shift from relatively diffuse recruitment of neural regions to more focal and specific patterns of activity, whether in terms of the number of regions recruited, or the magnitude or spatial extent of that recruitment
Local to distributed: a shift in the way this activity correlates across the brain, from being more locally arranged to showing more long-distance correlations.

In this post I will describe some of the most definitive evidence for each of these developmental shifts, and will then conclude with a discussion of how they may relate to one another as informed through computational modeling.

Evidence for the Diffuse to Focal Shift

Nice evidence for the diffuse-to-focal shift comes from Durston, Davidson, Tottenham, Galvan, Spicer, Fossella & Casey (2006), who asked 21 children aged 9-11 to complete a Go/NoGo task (in which a stimulus presented 75% of the time requires a response, but a different stimulus presented 25% of the time requires no response whatsoever) while they underwent functional magnetic resonance imaging (fMRI).

Numerous areas in the prefrontal cortex were strongly recruited by 9-year-olds during this task; by 11 years of age, most of these areas were not recruited as strongly, except a single area showing an increase in activation with age (the right inferior frontal gyrus; rIFG). Importantly, these results were stronger in analyses that took a longitudinal approach (comparing each 9-year-old with themselves 2 years later) than those taking a cross-sectional approach (comparing 9-year-olds with a distinct set of 11-year-olds), suggesting that the results cannot be explained as group averaging artifacts (in which older children show more focal activation simply because of improved registration). The results also cannot be easily explained as a practice effect, since no differences were observed between the group of 11-year-olds who had experienced the task 2 years previously vs. those 11-year-olds who performed it for the first time, nor as a performance effect, since 9-year-olds and 11-year-olds showed similar levels of performance on the critical "NoGo" trials (with a trend towards worse performance among the 11-year-olds).

Although one may quibble with a few aspects of this paper (fixed-effects analysis; failure to statistically control for performance differences; use of an outdated multiple comparisons correction), the weight of the evidence from this study and numerous others indicate that a diffuse-to-focal shift is very likely to occur during late childhood, and to be a real effect that theoretical developmental cognitive neuroscience should explain.

Evidence for the Local-to-Distributed Shift

Clues to the appropriate explanation for a developmental diffuse-to-focal shift in neural recruitment might be found in the developmental local-to-distributed shift in correlations of neural recruitment. A 2009 paper by Fair, Cohen, Power, Dosenbach, Church, Miezin, Schlaggar & Peterson analyzed correlations in neural recruitment among 210 subjects aged 7-31 while they completed no task whatsoever - a so-called resting-state functional connectivity MRI (rs-fcMRI) study. To do this they analyzed the fluctuations in average signal intensity occurring with a low temporal frequency (.009 to .08 Hz) that were unique to gray matter (relative to white matter or the ventricles) in 4 distinct networks, including a cingulo-opercular network, a fronto-parietal network, a cerebellar network, and the so-called default network (the network most commonly recruited during resting-state studies). (While Fair et al argue for a particular function for each of these networks, it's tough to say if they're right; besides, the putative functions of these networks are somewhat irrelevant to our current question).

The results demonstrated that fluctuations in neural activity tended to correlate among regions that were spatial proximal in the youngest subjects, and tended to correlate among regions that were more distal in the older subjects. This effect was particularly pronounced in the frontal cortex, as can be observed in the below video, which uses a spring-embedded visualization to demonstrate how temporal correlations across neural areas changes with development: areas closer together manifest a stronger correlation. Pay special attention to the nodes with a light blue border, which are those located in the prefrontal cortex, and how they actually segregate by the end of the video, indicating a type of organization that is not strictly due to spatial proximity.

A slightly less confusing, if less flashy, visualization of these changes can be observed in the following graph, which illustrates correlations among areas that are spatially proximal (91mm) in children (x-axis) and adults (y-axis). In the absence of any developmental change the scatterplots should all lie along the line of identity, such that correlations across the brain are roughly the same in children and adults. Instead, scatterplots are below the line of identity for 91mm, indicating that short range correlations decrease with age whereas long-range correlations increase, respectively.

i-780435052ff352bbe0489998e13d3169-Fairetalcorrelations.jpg

Towards an Integrative Account of the Diffuse-to-Focal and Local-to-Distributed Shifts

At first glance, the two developmental shifts described above may seem to be in conflict. After all, the beginning state of one shift ("diffuse") sounds like the end state of the other ("distributed"), so one could be forgiven for thinking they're contradictory. But this conflict is more apparent than real: the diffuse-to-focal shift is taking place on a smaller spatial scale than the local-to-distributed shift. Thus, when individual regions become more functionally specialized from those immediately surrounding them (as in the diffuse-to-focal shift), they may also tend to show fewer local correlations in activity (as in the local-to-distributed shift).

Nonetheless, there are numerous underlying biological mechanisms that might give rise to these developmental shifts, and it may be that distinct mechanisms support both. Here's some speculation to prove the point. Increases in synaptic conduction velocity (perhaps as a result of the myelination processes that unfold across development) could enable more effective long-range connectivity; this, in turn, might cause information coming from long-range connections to more strongly dominate local processing, effectively shifting the balance towards more focal activation patterns in an area through reduced local-to-local processing. Alternatively, synaptic pruning might reduce the spatial extent of local clusters of activity, but lead to more coherent patterns within each cluster, ultimately enabling more stable long-range connections to develop between areas which are each more functionally coherent. The point is that even if we are confident that diffuse-to-focal and local-to-distributed patterns emerge, there are multiple candidates for the underlying mechanism(s) that give rise to these systems-level phenomena.

Informative constraints for this kind of speculation comes from a computational model developed by Edin, Macoveanu, Olesen, Tegner & Klingberg, 2007. The authors used a neural network model to simulate excitatory and inhibitory neurons in the prefrontal and parietal cortex; excitatory neurons in both areas were "tuned" to respond maximally to lines with a preferred orientation, and inhibitory neurons provided lateral competitive inhibition across the excitatory neurons. Between prefrontal and parietal areas, excitatory neurons were interconnected with a particular conduction delay. In addition, both intra- and interregional connections among excitatory neurons were preferentially stronger across neurons with similar tuning curves; in other words, the connection strengths between excitatory neurons was determined as a function of the discrepancy between the preferred orientation of excitatory neurons in the prefrontal and parietal layers.

The authors simulated development in this network according to five basic hypotheses: Development is associated with ...

1) increases in the strength of local connectivity, i.e., within the prefrontal and parietal layers (local synaptic strengthening)
2) increases in the strength of long-distance connectivity, i.e., across the prefrontal and parietal layers (long-distance synaptic strengthening)
3) increased specificity in the tuning curves of each neuron (a combined pattern of synaptic strengthening and synaptic pruning at the local level)
4) increased specificity in the connection curves (a combined pattern of synaptic strengthening and synaptic pruning at a more distributed level)
5) increases in signal conduction velocity in long-distance connectivity (myelination)

To determine which of these mechanisms might give rise to a developmental shift in activation magnitude (as observed by Durston et al) as well as to a developmental shift in inter-regional correlations (as observed by Fair et al), the authors calculated a proxy BOLD response from the model, as the sum of the ion channels simulated in each neuron convolved with a standard hemodynamic response function.

The results indicated that only hypotheses 1-3 lead to the increased activation magnitude observed by Durston et al., and that only hypotheses 2 & 3 also explained the observed increases in frontoparietal coherence observed by Fair et al. Because hypotheses 2 & 3 both effectively give rise to long-range synaptic strengthening, and only hypothesis 3 involves synaptic pruning, the authors conclude that it is long-range synaptic strengthening that gives rise to the patterns observed in developmental neuroimaging studies.

Note, however, that Edin et al did not directly analyze the "diffuseness" of the activations resulting from hypotheses 2 & 3. It seems likely to me that only hypothesis 3 would show that effect, but this awaits direct analysis.

Summary: Diffuse-to-Focal and Local-to-Distributed Shifts are Not Incompatible, and Could Be Driven by the Same Underlying Biological Mechanisms

In this post I've described evidence supporting two types of changes in neural recruitment: going from a relatively diffuse pattern of activation to more focal patterns with larger signal magnitude, and going from a relatively local pattern of functional connectivity to a more distributed pattern. Evidence from a computational model suggests that both these patterns may be captured by either (or both) of two underlying neurobiological changes: increases in the synaptic strength of long-range connections, or increases in the specificity of neuronal tuning within each area.

It remains to be seen whether existing learning algorithms, such as Hebbian learning normalized with Oja's rule, might cause these patterns to emerge over the course of learning, or whether additional mechanisms must be included to account for these results (such as those involving neuromodulatory change during development). Nonetheless, it is clear that the diffuse-to-focal and local-to-distributed shifts need not be considered contradictory, and it is entirely possible that the same underlying biological mechanisms may give rise to both.

More like this

How can we enhance perception, learning, memory, and cognitive control? Any answer to this question will require a better understanding of the way they are best enhanced: through cognitive change in early development. But we can't stop there. We also want to know more about the neural substrates…
Most computational models of working memory do not explicitly specify the role of the parietal cortex, despite an increasing number of observations that the parietal cortex is particularly important for working memory. A new paper in PNAS by Edin et al remedies this state of affairs by developing…
In last week's Science, Dosenbach et al describe a set of sophisticated machine learning techniques they've used to predict age from the way that hemodynamics correlate both within and across various functional networks in the brain. As described over at the BungeLab Blog, and at Neuroskeptic, the…
In their wonderful Neuroimage article, Braun & Mattia present a comprehensive introduction to the possible neuronal implementations and cognitive sequelae of a particular dynamical phenomenon: the attractor state. In another excellent paper, just recently out in Frontiers, Itskov, Hansel and…

Some time before, I really needed to buy a good house for my corporation but I didn't have enough cash and could not purchase anything. Thank God my sister proposed to try to take the personal loans at creditors. Hence, I did that and used to be happy with my financial loan.

The heart of your writing whilst appearing agreeable at first, did not sit well with me personally after some time. Someplace within the paragraphs you were able to make me a believer but just for a short while. I nevertheless have got a problem with your leaps in logic and one might do nicely to fill in all those gaps. If you can accomplish that, I will certainly end up being amazed.