Developmental Cognitive Neuroscience 11 (2015) 12–17

Contents lists available at ScienceDirect

Developmental Cognitive Neuroscience

journal homepage: http://www.elsevier.com/locate/dcn

Is “efficiency” a useful concept in cognitive neuroscience?

Russell A. Poldrack ∗
University of Texas, Imaging Research Center, 100 East 24th Street, R9975, Austin, TX 78712, United States

a r t i c l e i n f o a b s t r a c t

Article history: It is common in the cognitive neuroscience literature to explain differences in activation
Received 28 February 2014 in terms of differences in the “efficiency” of neural function. I argue here that this usage
Received in revised form 31 May 2014 of the concept of efficiency is empty and simply redescribes activation differences rather
Accepted 3 June 2014
than providing a useful explanation of them. I examine a number of possible explanations
Available online 13 June 2014
for differential activation in terms of task performance, neuronal computation, neuronal
energetics, and network organization. While the concept of “efficiency” is vacuous as it
Keyword:
is commonly employed in the neuroimaging literature, an examination of brain develop-
Neuroimaging
ment in the context of neural coding, neuroenergetics, and network structure provides a
Neural Energetics
Metabolism roadmap for future investigation, which is fundamental to an improved understanding of
fMRI developmental effects and group differences in neuroimaging signals.
Response Time © 2014 The Author. Published by Elsevier Ltd. This is an open access article under the CC
Networks BY-NC-SA license (http://creativecommons.org/licenses/by-nc-sa/3.0/).

1. Introduction (through the study of amnesic patients like H.M.) to a deep

understanding of the circuitry of the medial temporal lobe
As scientists we aim to help understand how the and the computational role of different subregions (Milner
world works, which generally means providing mechanis- et al., 1998). However, in many other domains of cognitive
tic explanations of natural phenomena. For example, the neuroscience, including developmental cognitive neuro-
synaptic plasticity theory of memory proposes that mem- science, some very commonly used explanations are far less
ories are created through the modulation of the strength satisfying.
of synapses between neurons via specific mechanisms In the present paper I will focus on the concept of
such as NMDA-dependent long-term potentiation. Such an “efficiency” which is commonly used as an explanation in
explanation tells us something about the putative causal cognitive neuroscience. Perhaps the best known usage of
structure of the mechanisms that generate the relevant the term comes from the “neural efficiency theory” of intel-
data, which supports the generation of predictions about ligence proposed by Haier and colleagues. For example:
the effects of particular experimental interventions. For
A series of investigations in normal subjects indi-
example, this theory predicts that manipulations that block
cate an inverse relationship between brain glucose
NMDA receptor function should reduce the ability to form
metabolic rate (GMR) and psychometric measures of
new memories (and indeed they do). In cognitive neuro-
intelligence. . .These studies have been interpreted as
science, we possess only a small number of theories with
evidence for a brain efficiency model of intelligence:
similar explanatory power. For example, in the last 50
Intelligence is not a function of how hard the brain
years memory research has moved from the initial estab-
works but rather how efficiently it works. . . This effi-
lishment of the medial temporal lobe’s role in memory
ciency may derive from the disuse of many brain areas
irrelevant for good task performance as well as the more
∗ Tel.: +1 512 232 9504. focused use of specific task-relevant areas. (Haier et al.,
E-mail address: [email protected] 1992)

http://dx.doi.org/10.1016/j.dcn.2014.06.001
1878-9293/© 2014 The Author. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-SA license (http://creativecommons.org/
licenses/by-nc-sa/3.0/).
 R.A. Poldrack / Developmental Cognitive Neuroscience 11 (2015) 12–17 13

Other examples are found throughout the literature None of us would accept the latter as a suitable expla-
across studies of development, aging, individual differ- nation for the difference in fuel consumption; in fact, we
ences and learning. would likely recognize immediately that the second “expla-
nation” is not an explanation at all, but rather simply a
• “. . .improved handwriting is associated with increased redescription of the data.
I propose that the common usage of the concept of
computational efficiency or neural coding and hence
“efficiency” in cognitive neuroscience is equally vacuous.
reduced BOLD signal increase in the right IFG for reading-
Looking back at the list of quotes above, in each case it is
related functions.” (Gimenez et al., 2014).
• “We show that increased neural efficiency and capacity, apparent that the term “efficiency” is simply a redescrip-
tion of the phenomenon of reduced activation; although
as reflected by more “youth-like” brain response pat-
it sounds like it is explaining the result, it does not tell us
terns in regions of interest of the frontoparietal WM
any more about the mechanism, and implies no additional
network, were associated with better behavioral training
experiments that one might do to test the explanation.
outcome” (Heinzel et al., 2014).
• “These results suggest that lifelong bilingualism offsets To the degree that we judge a scientific explanation with
regard to its production of new testable hypotheses, the
age-related declines in the neural efficiency for cognitive
efficiency explanation fails completely when used in this
control processes.” (Gold et al., 2013)
• “. . .activity was significantly reduced for trained items way.
This is not to argue that concept of efficiency is intrin-
so that a further increase from two to three items was
sically useless; to the contrary, a better understanding of
observed. We interpret this difference as a correlate of
the relation between energy expenditure and neural com-
a gain in neural efficiency. . .We assume that training
putations is fundamental to understanding developmental
causes a more efficient neural representation of trained
differences in neuroimaging signals. Take any developmen-
items supported by long-term memory and this allows
tal experiment where groups are compared and differences
holding more items in working memory.” (Zimmer et al.,
in activation are observed on a particular comparison of
2012)
• “These results indicate that a short regimen of [work- task conditions. These differences could reflect any of the
following biological differences:
ing memory] training is associated with lower prefrontal
activation – a marker of neural efficiency – in divergent
thinking.” (Vartanian et al., 2013)
• A different set of cognitive processes is being performed. For
example, a child with Tourette syndrome participating in
In general, the term is used to describe situations where
an fMRI study is effectively in a dual task situation, per-
performance appears similar but activation is greater for
forming the specified cognitive task while also actively
one group (which is taken to be “less efficient”). It should
attempting to suppress their tics, whereas a healthy child
be noted that there is a very different notion of efficiency
would not experience this secondary response inhibition
that has arisen from the network analyses of brain connec-
demand.
tivity; I discuss this further below, but here I am focused on • A different neural computation is performed to complete
the use of the concept to describe differences in univariate
the same task. For example, in an pseudoword pronun-
activation.
ciation task, one group pronounces the stimulus through
grapheme-to-phoneme conversion, while a more skilled
2. Is “efficiency” really an explanation? group pronounces it using analogy to known words.
Alternatively, one group represents a stimulus domain
Upon closer examination it is clear that “efficiency” using a dense coding scheme, whereas another repre-
generally fails as a scientific explanation when used in sents it using sparse coding.
this way. An analogy is instructive. Let’s say that we are
interested in understanding individual differences in gas
mileage between different makes of automobiles. We per- Neither of these captures what is generally meant by the
form an experiment in which we drive two cars (a hybrid concept of efficiency, i.e. reduced energy expenditure for
Toyota Prius and a gas-only Porsche Carrera) from Los the same work. However, there are several other potential
Angeles to San Francisco along exactly the same route, explanations that come closer to this concept:
and we measure their fuel consumption. The results of the
experiment show that the Prius uses 1/2 as much fuel as the
Porsche to travel the same distance. Here are two possible • The same neural computation is being performed for the
explanations of this difference in fuel consumption:
same amount of time, but with different intensity. For exam-
ple, in one group neurons in the relevant region fire at
1. The Prius has a gas-electric hybrid engine (which uses 20 Hz for 100 ms, while in the other group neurons fire at
surplus engine power to generate electricity which is 10 Hz for the same amount of time.
then turned back into drive power) and regenerative • The same neural computation is performed at the same
braking (which captures energy that would otherwise intensity, but for different lengths of time. For example, in
be lost as heat). one group neurons fire at 20 Hz for 100 ms, while in the
2. The Prius is more efficient. other group the same neurons fire at 20 Hz for 50 ms.
14 R.A. Poldrack / Developmental Cognitive Neuroscience 11 (2015) 12–17

Finally, there is one explanation that comes closest 3.2. Is the computation the same?
to capturing the essence of the concept of neural cost-
efficiency: The next major question regarding the interpretation of
activation differences is whether the same neural compu-
tations are being performed. There are numerous ways in
• The same neural computation is performed with identical
which the computations underlying task performance can
time and intensity, but the metabolic expenditure differs
change with learning and development. Most notable is the
between the groups. For example, the groups could dif-
fact that as experience accrues, it is often possible to per-
fer in the amount of transmitter release, the nature of
form a task based on memory for prior experiences rather
neurovascular coupling (e.g. due to differences in con-
than through application of rules or brute force computa-
tact between astrocytes and neurons), or the degree to
tion (Logan, 1988). For example, there is a hypothesis in
which they rely upon oxidative versus non-oxidative
reading development that initial application of grapheme-
metabolism.
to-phoneme conversion rules is supplanted by direct access
to phonological word forms from visual forms for famil-
The point of laying out each of these other possibili- iar words (Grainger et al., 2012). Similarly, acquisition of
ties is to highlight that while there is a way in which the mirror-reading skill is thought to progress from initial
concept of efficiency can be well-specified as a plausible visuospatial transformation to later use of direct visual
explanation of activation differences that makes non-trivial recognition (Poldrack et al., 1998). If the computations
predictions, doing so requires confronting a number of dif- being performed differ between groups, then labeling the
ficult questions which are rarely if ever addressed in the differences as using “efficiency” is invalid because the work
cognitive neuroscience literature. being performed differs; an analogy would be compar-
ing the fuel expended to travel from Los Angeles to San
Francisco, where one car travels via the interstate and the
3. Explaining differences in activation other follows the Pacific Coast Highway.
Determining empirically whether the same computa-
Because many researchers in cognitive neuroscience tion is being performed is challenging because it rests on
(including but not limited to developmental cognitive accepting a null hypothesis. Strategies could include the
neuroscientists) are fundamentally interested in under- use of behavioral manipulations to assess the effects of
standing differences in brain activity between individuals relevant variables on behavior. For example, in a mental
at different points in time, we desperately need better rotation task where the subject compares two block fig-
explanations for such differences (cf. Poldrack, 2000). Here ures (Shepard and Metzler, 1971), one could measure the
I will outline in greater depth the potential explanations slope of response times as a function of the angle of rota-
for differential activation mentioned above. tion; major changes in this slope would suggest that the
task is being performed in a different manner. Similarly
one might ask whether the same factors cause behavioral
3.1. Is the set of cognitive processes the same? interference between the groups. Using fMRI one might
attempt to establish whether different networks are active,
We nearly always assume that the cognitive processes or whether the same network is active to different degrees,
being performed by the subject are strictly defined by the though this will be statistically challenging. Overall, it is
experimental paradigm that the subject is presented with, straightforward to show that different computations are
but it is clear that this is often an invalid assumption. In being used but fundamentally challenging to show that the
particular, the requirements of the fMRI acquisition envi- computation has not changed.
ronment (i.e. lying very still and quiet in a small tube for an Changes in the nature of neuronal information coding
extended period of time) represent a “meta-task” that the could also result in differences in activation. In particular,
subject must perform in order to comply with the experi- the concept of “efficient coding” (Barlow, 1961) suggests
menter’s demands. Whereas most adults have little trouble that energy usage is minimized when information is coded
exerting the executive control necessary to accomplish this in a way that reduces the redundancy of information across
meta-task, for children or individuals with executive con- neurons, which leads to the development of sparse codes
trol disorders one might consider the meta-task demands (cf. Olshausen and Field, 1996). There has been little study
to essentially be a demanding secondary task. If the pri- of how the sparseness of neuronal coding changes with
mary task is one that engages cognitive processes that development in mammalian cortex, but recent work has
overlap with the meta-task (e.g. response inhibition, work- demonstrated that sparseness of coding in ferret visual
ing memory), then one might expect interactions with the cortex decreases rather than increases across develop-
experimental task that could result in activation differ- ment (Berkes et al., 2011), which is inconsistent with the
ences (e.g. enhanced activation due to overload, or reduced decreased activation generally observed in the context of
activation due to ceiling effects on activation). Similar con- developmental neuroimaging. Nonetheless, this suggests
cerns arise regarding differential task difficulty; if a task is that in order to fully identify whether the same computa-
much easier for one group than another, then differences tion is being performed, it would be necessary to examine
could reflect the fact that subjects in the easy group are the way in which neural coding may have changed. This is
engaging in additional task-independent thought during rarely possible in humans, but could be examined in animal
performance of the experimental task. models.
 R.A. Poldrack / Developmental Cognitive Neuroscience 11 (2015) 12–17 15

in activation intensity. Conversely, if one shows activation

differences after having properly removed the effects of
time on task across groups, then this provides some degree
of confidence that the results do indeed reflect differences
in the level of activation intensity. However, even when RT
has been modeled as a nuisance factor, it will be very dif-
ficult to fully remove its effects (e.g. due to nonlinearities)
and thus it will always be difficult to interpret activation
differences that are accompanied by behavioral differences
in RT.

4. Neural energetics and activation differences

As noted above, another possible explanation of activa-

tion differences is that the computation remains the same
in quality, time, and intensity, but the energy used to per-
Fig. 1. Differences in processing time are very difficult to distinguish from form the computation differs. This most closely approaches
differences in intensity of activation. The thick black line reflects data
the usual meaning of the concept of efficiency, i.e. the
generated for four trials that vary in processing times (as listed in the
figure) through convolution with a standard SPM hemodynamic response amount of energy needed to perform a given amount of
function. The dashed gray line reflects the results of fitting a model to work. There is a large number of biophysical mechanisms
those data in which differences between trials are based on the intensity by which the same neuronal firing pattern could consume
of the activation, using a fixed duration of 400 ms (with the fitted intensity
a differential amount of energy.
for each trial presented as “beta” in the figure). The black and dashed gray
lines are almost indistinguishable (difference between these functions is
The energy “budget” for neuronal signaling in the
plotted in blue), highlighting the fact that the effects of processing time cerebral cortex involves action potentials (21%), synap-
and activation intensity are for practical purposes indistinguishable, both tic processes (59%), and resting potentials (20%) (Howarth
resulting in similarly increased activation. Code to generate this figure is et al., 2012). The energetic efficiency of action poten-
available from https://github.com/poldrack/rtmodel. (For interpretation
tials lies largely in the overlap of the inward (Na+)
of the references to color in this figure legend, the reader is referred to the
web version of the article.) and outward (K+) currents, and varies greatly between
classes of neurons; for example, squid giant axons are
highly inefficient, whereas cortical and thalamic neurons
in rodents are highly efficient (Sengupta et al., 2010). Fur-
3.3. Intensity and timing are (nearly) indistinguishable ther, metabolic efficiency varies across different parts of
with fMRI the neuron (Hallermann et al., 2012). These data suggest
that any changes in neuronal structure could potentially
We can next ask how differences in intensity and tim- result in changes in energy usage; the time course of any
ing of neural signaling affect fMRI signals, and whether such changes appears to be currently unknown. Similarly,
they can be distinguished. This is particularly pertinent changes in the nature of synaptic signaling, such as the
given that experience and maturation both lead to the abil- well known changes in synaptic density (i.e. “pruning”) that
ity to perform tasks more quickly, and it is rare to see a accompany brain development along with changes in the
neuroimaging finding in these domains that is not accom- relative abundance of NMDA receptors and metabotropic
panied by changes in response times, even in cases where signaling, could impact the relative efficiency of signaling
one has good reason to believe that the same computa- with development. These suggestions only scratch the
tion is being performed. Further, these differences in time surface of the potential biophysical causes for differen-
on task are associated with regionally specific differences tial energy efficiency of neuronal signaling; my point is
in activation regardless of the specific task. For example, simply to highlight how little we currently understand
(Yarkoni et al., 2009) found that differences in response about the potential causes of true differences in neural
time (RT) between task conditions were associated with efficiency.
differences in activation in prefrontal and other regions in Using a combination of neuronal recordings and 2DG,
five different studies using different tasks. Unfortunately, Picard et al. (2013) recently demonstrated the first evi-
it is almost impossible to distinguish changes in activa- dence of experience-dependent changes in the energy
tion due to differential intensity from changes due to time usage of neural activity. Monkeys were trained to perform
on task. Fig. 1 shows that differences in activation timing a internally generated motor task over an extended period
can be mimicked almost exactly by differences in intensity, (up to 6 years), and then compared performance of this task
which leads to a fundamental difficulty in the interpreta- with an untrained (visually guided) motor task. Neuronal
tion of activation changes. Fortunately, RT can be measured recordings in motor cortex revealed no differences in spike
independently and used in the statistical model to correct rates between the two tasks, but 2-deoxy-glucose uptake
for the effects of time on task (cf. Grinband et al., 2008), and differed significantly between the conditions, suggesting a
the foregoing demonstration highlights the absolute neces- difference in energy usage in the face of seemingly iden-
sity of such corrections. If RT is not modeled, then there tical neuronal computations. While this kind of approach
is no way to know whether differences in activation are is likely to remain unfeasible with human subjects (par-
due to differences in time on task versus true differences ticularly with children), it does highlight the fact that
16 R.A. Poldrack / Developmental Cognitive Neuroscience 11 (2015) 12–17

specific hypotheses about the relation between neuronal found no relation between age and the overall efficiency of
computation and energetics can be addressed using exist- brain networks, whereas there was an inverted-u shaped
ing methods in nonhuman animal models. It may also relation between age and local efficiency (which is the aver-
be possible to address some of these questions through age efficiency of the subnetwork formed by each node).
combinations of imaging and stimulation techniques in Similarly, a study of children from 5 to 18 years of age
humans. showed no relation between global efficiency and age but a
Another potentially interesting source of developmen- positive relationship between local efficiency and age (Wu
tal differences in energetic efficiency is the degree to which et al., 2013) (cf. also Supekar et al., 2009). Similar results
brain activity relies upon different energetic mechanisms have appeared for structural connectivity. For example,
across development. Neurons can rely upon a number of Dennis et al. (2013) examined global and local efficiency
different energetic pathways, including oxidative phos- of structural networks identified using diffusion tensor
phorylation (Kety, 1957; Sibson et al., 1998; Sokoloff, imaging in a large sample ranging from 12 to 30 years of
1960), aerobic glycolysis (Fox et al., 1988), and metabolism age, and found no relation between global efficiency and
of ketone bodies (Sokoloff, 1973), which vary greatly in age but a significant relation between local efficiency in
their energetic efficiency (e.g. oxidative phosphorylation a number of regions (including hub-like regions such as
can generate 15 times more ATP than glycolysis using the posterior cingulate and medial prefrontal cortex). A study
same amount of glucose; Fox et al. (1988)). The brain’s of the very early development of brain networks based on
use of these different energetic pathways is known to vary gray matter volume showed that global efficiency increases
with development. In particular, a recent meta-analysis by gradually from one month of age to two years (Fan et al.,
Goyal et al. (2014) showed that aerobic glycolysis is greatly 2011). Together these data begin to suggest that while
increased during childhood, and that regional increases in global efficiency reaches adult-like levels fairly early in
glycolysis are associated with expression of genes related development, local efficiency continues to change across
to synaptic development. There are also known changes the span of development. One current unknown is how
in the expression of genes relevant to energetics, such as these changes in local network efficiency might related to
the glucose transporter (Vannucci and Vannucci, 2000). A changes in task activation; this is an important question for
better understanding of developmental changes in neural future research. In addition, developmental differences in
energetics across development and their relation to func- resting state connectivity may suffer from the same prob-
tional imaging signals could provide important calibration lems noted above regarding task activation; in particular,
to help understand the degree to which development children may require greater effort to engage the meta-task
effects in imaging studies reflect neuronal versus energetic of remaining still during the scan, which would presumably
differences, in the same way that studies of neurovas- affect network structure.
cular coupling in aging have helped provide guidance in
the interpretation of aging-related imaging signals (e.g.,
Gazzaley and D’Esposito, 2004)
6. Conclusion
5. Network efficiency
“Efficiency” is such a facile explanation that it is easy to
miss the emptiness that is inherent in its usual application
The total amount of energy consumed by neuronal com-
to neuroimaging data. This is not uncommon when every-
putations depends not just upon the function of individual
day terms are used to describe psychological phenomena.
neurons, but also on how those neurons are connected
For example, Navon (1984) made a very compelling argu-
to one another. With the advent of graph-theoretical
ment that the concept of “mental resources” (which was
approaches to the analysis of neuroimaging data (Bullmore
commonly used by cognitive psychologists in the 1970s
and Sporns, 2009), there is increasing evidence regarding
and remains in common usage amongst cognitive neuro-
the cost-efficiency of large-scale brain network structure
scientists) was similarly non-explanatory, a “soup stone”
(Bullmore and Sporns, 2012), where “efficiency” in this
in his parlance. Here I have laid out the problems with
case is defined in terms of cost of transmitting informa-
the concept of efficiency as it is usually applied within the
tion within the network. In particular, it appears that brain
neuroimaging literature, and examined a number of alter-
networks are organized in a way that approaches the max-
native explanations for differences in activation signals.
imum possible cost-efficiency, though the embedding of
This analysis uncovers deep problems for the interpreta-
complex topological structure that maximizes complexity
tion of developmental changes in neuroimaging signals,
while minimizing transmission costs (Bassett et al., 2010).
and highlights the need for additional work using animal
This approach provides an alternative but well-formulated
models to examine the ways in which neuronal energet-
conception of “efficiency” that can be tested using both
ics change in concert with developmental experience and
structural and functional MRI data.
brain maturation.
Recent work has begun to examine individual differ-
ences in the efficiency of brain network structure and
function. Cao et al. (2014) measured the topological effi-
ciency of resting state networks in a group of participants Conflict of interest statement
ranging from 7 to 87 years old, where efficiency is defined
as the inverse of the mean path length between each pair The author has no conflicts of interest to declare for this
of nodes in the network (Latora and Marchiori, 2001). They manuscript.
 R.A. Poldrack / Developmental Cognitive Neuroscience 11 (2015) 12–17 17

Acknowledgments Hallermann, S., de Kock, C.P.J., Stuart, G.J., Kole, M.H.P., 2012. State and
location dependence of action potential metabolic cost in cortical
pyramidal neurons. Nat. Neurosci. 15 (7), 1007–1014.
Thanks to Jessica Church-Lang, Sarah Helfinstein, Tom Heinzel, S., Lorenz, R.C., Brockhaus, W.-R., Wüstenberg, T., Kathmann, N.,
Schonberg, and Tal Yarkoni for helpful comments on an Heinz, A., Rapp, M.A., 2014. Working memory load-dependent brain
earlier draft. response predicts behavioral training gains in older adults. J. Neurosci.
34 (4), 1224–1233.
Howarth, C., Gleeson, P., Attwell, D., 2012. Updated energy budgets for
References neural computation in the neocortex and cerebellum. J. Cereb. Blood
Flow Metab. 32 (7), 1222–1232.
Barlow, H., 1961. Possible principles underlying the transformation Kety, S.S., 1957. The general metabolism of the brain in vivo. In:
of sensory messages. In: Sensory Communication., pp. 217–234, Metabolism of the Nervous System., pp. 221–237.
http://redwood.berkeley.edu/w/images/f/fd/02-barlow-pr-1954.pdf Latora, V., Marchiori, M., 2001. Efficient behavior of small-world networks.
Bassett, D.S., Greenfield, D.L., Meyer-Lindenberg, A., Weinberger, D.R., Phys. Rev. Lett. 87 (19), 198701.
Moore, S.W., Bullmore, E.T., 2010. Efficient physical embedding of Logan, G.D., 1988. Toward an instance theory of automatization. Psychol.
topologically complex information processing networks in brains and Rev. 95, 492–527.
computer circuits. PLoS Comput. Biol. 6 (4), e1000748. Milner, B., Squire, L.R., Kandel, E.R., 1998. Cognitive neuroscience and the
Berkes, P., Orb’an, G., Lengyel, M., Fiser, J., 2011. Spontaneous cortical activ- study of memory. Neuron 20 (3), 445–468.
ity reveals hallmarks of an optimal internal model of the environment. Navon, D., 1984. Resources: a theoretical soup stone? Psychol. Rev. 91,
Science 331 (6013), 83–87. 16–22.
Bullmore, E., Sporns, O., 2009. Complex brain networks: graph theoretical Olshausen, B.A., Field, D.J., 1996. Natural image statistics and efficient
analysis of structural and functional systems. Nat. Rev. Neurosci. 10 coding. Network 7 (2), 333–339.
(3), 186–198. Picard, N., Matsuzaka, Y., Strick, P.L., 2013. Extended practice of a motor
Bullmore, E., Sporns, O., 2012. The economy of brain network organization. skill is associated with reduced metabolic activity in M1. Nat. Neurosci.
Nat. Rev. Neurosci. 13 (5), 336–349. 16 (9), 1340–1347.
Cao, M., Wang, J.-H., Dai, Z.-J., Cao, X.-Y., Jiang, L.-L., Fan, F.-M., Song, X.-W., Poldrack, R.A., 2000. Imaging brain plasticity: conceptual and method-
Xia, M.-R., Shu, N., Dong, Q., Milham, M.P., Castellanos, F.X., Zuo, X.-N., ological issues – a theoretical review. Neuroimage 12 (1), 1–13.
He, Y., 2014. Topological organization of the human brain functional Poldrack, R.A., Desmond, J.E., Glover, G.H., Gabrieli, J.D., 1998. The neural
connectome across the lifespan. Dev. Cogn. Neurosci. 7, 76–93. basis of visual skill learning: an fMRI study of mirror reading. Cereb.
Dennis, E.L., Jahanshad, N., McMahon, K.L., de Zubicaray, G.I., Martin, N.G., Cortex 8 (1), 1–10.
Hickie, I.B., Toga, A.W., Wright, M.J., Thompson, P.M., 2013. Devel- Sengupta, B., Stemmler, M., Laughlin, S.B., Niven, J.E., 2010. Action poten-
opment of brain structural connectivity between ages 12 and 30: a tial energy efficiency varies among neuron types in vertebrates and
4-Tesla diffusion imaging study in 439 adolescents and adults. Neu- invertebrates. PLoS Comput. Biol. 6, e1000840.
roimage 64, 671–684. Shepard, R.N., Metzler, J., 1971. Mental rotation of three-dimensional
Fan, Y., Shi, F., Smith, J.K., Lin, W., Gilmore, J.H., Shen, D., 2011. Brain objects. Science 171 (3972), 701–703.
anatomical networks in early human brain development. Neuroimage Sibson, N.R., Dhankhar, A., Mason, G.F., Rothman, D.L., Behar, K.L., Shulman,
54 (3), 1862–1871. R.G., 1998. Stoichiometric coupling of brain glucose metabolism and
Fox, P.T., Raichle, M.E., Mintun, M.A., Dence, C., 1988. Nonoxidative glu- glutamatergic neuronal activity. Proc. Natl. Acad. Sci. U.S.A. 95 (1),
cose consumption during focal physiologic neural activity. Science 316–321.
241 (4864), 462–464. Sokoloff, L., 1960. The metabolism of the central nervous system in
Gazzaley, A., D’Esposito, M., 2005. BOLD functional MRI and cognitive vivo. Handbook of Physiology – Neurophysiology, vol. 3., pp. 1843–
aging. In: Cabeza, R., Nyberg, L., Park, D. (Eds.), Cognitive Neuroscience 1864.
of Aging: Linking Cognitive and Cerebral Aging. Oxford University Sokoloff, L., 1973. Metabolism of ketone bodies by the brain. Annu. Rev.
Press, New York, NY, US, pp. 107–131. Med. 24, 271–280.
Gimenez, P., Bugescu, N., Black, J.M., Hancock, R., Pugh, K., Nagamine, M., Supekar, K., Musen, M., Menon, V., 2009. Development of large-scale func-
Kutner, E., Mazaika, P., Hendren, R., McCandliss, B.D., Hoeft, F., 2014. tional brain networks in children. PLoS Biol. 7 (7), e1000157.
Neuroimaging correlates of handwriting quality as children learn to Vannucci, R.C., Vannucci, S.J., 2000. Glucose metabolism in the developing
read and write. Front. Hum. Neurosci. 8, 155. brain. Semin. Perinatol. 24 (2), 107–115.
Gold, B.T., Kim, C., Johnson, N.F., Kryscio, R.J., Smith, C.D., 2013. Lifelong Vartanian, O., Jobidon, M.-E., Bouak, F., Nakashima, A., Smith, I., Lam, Q.,
bilingualism maintains neural efficiency for cognitive control in aging. Cheung, B., 2013. Working memory training is associated with lower
J. Neurosci. 33 (2), 387–396. prefrontal cortex activation in a divergent thinking task. Neuroscience
Goyal, M.S., Hawrylycz, M., Miller, J.A., Snyder, A.Z., Raichle, M.E., 2014. 236, 186–194.
Aerobic glycolysis in the human brain is associated with development Wu, K., Taki, Y., Sato, K., Hashizume, H., Sassa, Y., Takeuchi, H., Thyreau,
and neotenous gene expression. Cell Metab. 19 (1), 49–57. B., He, Y., Evans, A.C., Li, X., Kawashima, R., Fukuda, H., 2013. Topo-
Grainger, J., Lété, B., Bertand, D., Dufau, S., Ziegler, J.C., 2012. Evidence for logical organization of functional brain networks in healthy children:
multiple routes in learning to read. Cognition 123 (2), 280–292. differences in relation to age, sex, and intelligence. PLoS ONE 8 (2),
Grinband, J., Wager, T.D., Lindquist, M., Ferrera, V.P., Hirsch, J., 2008. e55347.
Detection of time-varying signals in event-related fMRI designs. Neu- Yarkoni, T., Barch, D.M., Gray, J.R., Conturo, T.E., Braver, T.S., 2009. Bold
roimage 43 (3), 509–520. correlates of trial-by-trial reaction time variability in gray and white
Haier, R.J., Siegel, B., Tang, C., Abel, L., Buchsbaum, M.S., 1992. Intel- matter: a multistudy fMRI analysis. PLoS ONE 4 (1), e4257.
ligence and changes in regional cerebral glucose metabolic rate Zimmer, H.D., Popp, C., Reith, W., Krick, C., 2012. Gains of item-specific
following learning. Intelligence 16 (3–4), 415–426 (Special Issue: Biol- training in visual working memory and their neural correlates. Brain
ogy and Intelligence) http://www.sciencedirect.com/science/article/ Res. 1466, 44–55.
pii/016028969290018M