A primer for brain imaging: a tool for evidence-based studies of nutrition?
Rotman Research Institute, University of Toronto, Toronto, Ontario, Canada; School of Psychology, University of Nottingham, United Kingdom; and Montreal Neurological Institute, McGill University, Montreal, Quebec, Canada.
ILSI Europe a.i.s.b.l, Av. E. Mounier 83, box 6, 1200 Brussels, Belgium. E-mail: firstname.lastname@example.org, Phone: +32-2-771-00-14, Fax: +32-2-762-00-44.
Nutrition affects brain structure and function throughout life. Nutritional scientists and practitioners are interested in gathering evidence clarifying which of the micro- or macronutrients may affect particular aspects of brain functioning at a given period of the life cycle and in identifying possible brain mechanisms underlying such effects. This article provides a primer on brain imaging techniques suitable for the assessment of the structure and function of the human brain, focusing on noninvasive techniques such as structural and functional magnetic resonance imaging, electroencephalography, and magnetoencephalography. The article concludes with a few suggestions regarding the choice of a particular imaging tool in observational studies and randomized controlled trials investigating nutritional effects on the human brain.
Adequate nutrition is essential for sustaining optimal brain structure and function throughout life.1,2 Reviewed here are the basic facts about brain growth, from conception to adulthood, followed by a brief description of issues motivating the use of brain imaging in randomized controlled trials (RCTs) of nutrition.
The most dramatic phase of brain growth happens during pregnancy. During the short period of 9 months, the initial “mother” cell gives rise to more than 100 billion nerve cells and a brain that weighs approximately 400 g when the child is born. During the first 4 years of life, the brain continues to grow, reaching the size of 1,200 g, which is only approximately 200 g less than that of the an adult's brain.3
Over the next 10–15 years, brain growth continues; it now involves different brain compartments in a slightly different way. For example, the thickness of the different regions of the cerebral cortex changes between the ages of 5 and 18 years at different paces, with the regions important for reasoning, planning, and social communication maturing last. The white matter (WM), which contains pathways connecting different brain regions, also continues to mature during this period. In boys, the volume of WM increases sharply during adolescence, perhaps under the influence of rising levels of the sex hormone testosterone. In girls, changes in WM seem subtler and may reflect a process called myelination, by which axons gain additional layers of a fatty substance called myelin, enhancing the speed of nerve conduction.4
The brain of an adult does not stop growing; experiences continue to shape the brain even after a person leaves college. For example, if a person decides to learn how to juggle three balls up in the air and then practices every day for 2 months, growth occurs in the parts of the cerebral cortex that are tracking the moving balls.5 Although it is not known which cellular components (e.g., dendrites, glia, vessels) are growing, it is likely that all of the additional brain activity in this brain module, specialized for tracking movement of visual stimuli, triggers a cascade of events leading to a structural change that may be seen with a structural magnetic resonance imaging (MRI) scan.
Finally, what about the “aging” brain? Does it grow or shrink? This seems to depend on the specific context. For example, older professional musicians playing in an orchestra are possibly gaining, and certainly not losing, grey matter (GM) in cortical regions that may be engaged repeatedly during their work, such as frequent sight reading of musical scores.6 This observation suggests that brain structure continues to be plastic and amenable to experience, even later in life.
Without a doubt, nutrition plays an important role in supporting structural and functional “growth” of the human brain from conception, through childhood and adolescence, and into adulthood. The brain is an energy-expensive organ; in an adult, it accounts for only 2% of body weight but consumes 20% of the resting metabolism. Most of the energy is spent on supporting the propagation of action potentials, followed by postsynaptic signaling and maintenance of resting potentials.7 A steady supply of macro- and micronutrients is important also for the synthesis of neurotransmitters (and their receptors and transporters), for the renewal and maintenance of the axonal cytoskeleton and its myelin sheath, for the growth of synaptic spines and, as such, neural plasticity, and for neuronal survival.
Although many epidemiological studies have uncovered associations between various macro- and micronutrients and cognitive performance and mental health (as reviewed by Benton8), the results of RCTs (as reviewed by McCracken9) less often support causal effects of nutrition on the brain and cognition. It has been argued that most RCTs are too short (lasting only a few months), too focused on a single micronutrient, or too small to detect modest effect sizes against the heterogeneous genetic and environmental background of the participating individuals (see McCracken9). Furthermore, the primary outcome measures, such as rating scales of cognition and mental health or even some of the cognitive tests, may be too insensitive or may have low test–retest reliability. In this context, the use of various approaches to directly measure brain structure and function seems appealing. The main goal of this review is to provide researchers in the nutrition field with the necessary background information to make an informed choice about a particular imaging modality for observational studies and RCTs. Table 1 provides an overview of the readily available techniques reviewed here, together with information on set-up time, scanning time, brain coverage, and the approximate cost of data acquisition. (For a detailed description of various brain-mapping methods, see Paus10).
Table 1. Techniques for structural (aMRI) and functional (EEG, MEG, fMRI, PET) imaging of the human brain
Set-up time (min)
Scanning time (min)
per type of imaging sequence.
aMRI, anatomical magnetic resonance imaging; EEG, electro-encephalography; fMRI, functional magnetic resonance imaging; MEG, magneto-encephalography; PET, positron emission tomography.
Interindividual variations in brain structure reflect an individual's genetic background and unique life experience and environment. As mentioned above, several recent studies have suggested that even short (months) variations in an individual's experience may induce changes in brain structure that are detectable with MRI (e.g., in musicians,6,11 London taxi drivers,12 bilingual subjects,13 initially naïve jugglers5). Furthermore, given the high reliability of the acquisition and analysis of structural magnetic resonance images in large numbers of individuals (>100), it is an ideal tool for use in population-based studies of genetic and environmental factors shaping the human brain.14
Briefly, MRI is a noninvasive technique that creates detailed three-dimensional “pictures” of the brain, each consisting of thousands of three-dimensional image elements named voxels. The most common structural MRI sequence produces T1-weighted images, which are characterized by a high contrast between grey matter and white matter. These images can be analyzed using various computational algorithms that quantify, automatically and precisely, many different features, such as thickness of the cerebral cortex, volume of grey and white matter, and tissue densities (Fig. 1).15 In addition, other MRI sequences allow one to assess the microstructure of white matter (diffusion tensor imaging, magnetization transfer imaging) or the chemical composition of brain tissue (magnetic resonance spectroscopy).
In the 1990s, several authors introduced voxel-based approaches to analyze regional differences in grey-matter and white-matter density in adult patients with schizophrenia,16,17 and in children and adolescents.18 The voxel-based analyses of these densities involve several steps: 1) non-linear registration of T1-weighted images to the template brain (e.g., MNI305); 2) classification of brain tissue into grey-matter and white-matter; 3) blurring the binary white-matter or grey-matter maps to create density maps with values of white-matter/grey-matter densities varying on a continuum between 0 and 1; and 4) testing the statistical relationship between a variable of interest (e.g., age or sex) and white-matter/grey-matter density in a voxel-based manner using a general linear model (with appropriate corrections for multiple comparisons).
To calculate global or regional (e.g., lobes) volumes of grey matter and white matter, three-dimensional maps of classified tissue are merged with a labeled template brain registered nonlinearly with the individual's brain. Specific brain structures, such as the hippocampus, can be segmented automatically using algorithms developed for this purpose.19
Using T1-weighted images, cortical thickness can be measured using FreeSurfer, a set of automated tools for the reconstruction of the cortical surface of the brain.20 For every subject, FreeSurfer segments the cerebral cortex, the white matter, and other subcortical structures and then computes triangular meshes that recover the geometry and topology of the pial surface and the grey and white interface of the left and right hemispheres. Local cortical thickness is measured based on the difference between the positions of equivalent vertices in the pial and grey and white surfaces. Correspondence between the cortical surfaces between subjects is established using a nonlinear alignment of the principal sulci in each subject's brain with those of an average brain.21
White matter microstructure
The introduction of diffusion tensor imaging in the mid 1990s opened up new avenues for in vivo studies of white-matter microstructure.22 This imaging technique allows one to estimate several parameters of water diffusion in live tissue, such as mean diffusivity and fractional anisotropy (FA). The latter parameter reflects the degree of directionality of water diffusion; voxels containing water moving predominantly in a single direction have higher FA. In white matter, FA is believed to depend on the microstructural features of fiber tracts, including the spatial alignment of individual axons, their packing “density” (which affects the amount of interstitial water), and myelin content, but the hypothesized differences in myelination are perhaps too hastily considered as an explanation for age-related differences in FA, to the exclusion of other possible factors. Using the short T2-component signal to estimate myelin-water fraction, a recent study carried out in adult participants found a significant correlation between this measure of myelin content and FA between but not within various brain regions.23 This finding raises the possibility that the interindividual variations in FA values observed in developmental studies in a particular brain region, such as the corpus callosum, may not be related to myelination. (See Paus4 for more details.)
After acquiring diffusion tensor imaging data throughout the brain, values of FA and mean diffusivity can be assessed in a number of ways. For example, one can average FA and mean diffusivity values across all voxels constituting white matter in the four cerebral lobes or of major fiber tracts.24 It is also possible to evaluate variations in FA and mean diffusivity throughout the brain on a voxel-based approach using, for example, tract-based spatial statistics.25
Diffusion tensor imaging is not the only method that allows the study of white-matter microstructure. As mentioned above, the short T2-component signal fraction provides highly specific estimates of myelin-associated water.23 MTI is another MRI technique employed in studies of the structural properties of white matter. Contrast in MTI reflects the interaction between free water and water bound to macromolecules26; the macromolecules of myelin are the dominant source of the MTI signal in white matter.27 Postmortem data that reveal a significant positive correlation between myelin content and MTR support this interpretation of magnetization transfer ratio (MTR).28,29 Magnetization transfer data are acquired using a dual acquisition with and without an MTI saturation pulse; MTR images are calculated as the percentage of signal change between the two acquisitions.30 Mean MTR values can subsequently be summed across all WM voxels constituting, for example, the four lobes of the brain.
FUNCTIONAL MAGNETIC RESONANCE IMAGING
For imaging brain function, the most commonly measured MRI parameter is the blood-oxygenation-level-dependent (BOLD) signal. The BOLD signal reflects the proportion of oxygenated and deoxygenated blood in a particular brain region at a given moment. A strong correlation between the amount of synaptic activity and regional cerebral blood flow is why the BOLD signal is a good, albeit indirect, measure of brain function.31 In the majority of functional MRI (fMRI) studies, changes in BOLD signal are measured in response to various sensory, motor, or cognitive stimuli. Therefore, only brain regions that respond to a particular set of stimuli can be interrogated using the given paradigm.
In a typical BOLD-based fMRI study, a set of T2*-weighted images is acquired, with the size of a voxel being 4 × 4 × 4 mm. To cover most of the cerebrum, approximately 32 contiguous 4-mm-thick slices must be acquired. By virtue of collecting a set of images over a relatively brief period of time (seconds), fMRI affords a great deal of flexibility in designing brain-mapping studies. Two types of fMRI designs are in use: block-design fMRI and event-related fMRI.
Using block-design fMRI, a number of similar events are presented over a block of time and compared with different events administered during a different block. For example, the subject may be moving a finger during one 20-second block and resting during another block; typically, the two conditions would alternate for several cycles. The images (volumes) are acquired in rapid succession (e.g., every 3 seconds) throughout the session, so that, for example, 120 volumes (e.g., 1 volume = 32 4-mm-thick axial slices) would be collected in 6 minutes. The volumes collected during the two respective conditions are pooled together (e.g., 60 volumes/condition) and analyzed without the use of information about the exact temporal relationship between an event (e.g., onset of the finger movement) and the volume acquisition.
Event-related design, on the other hand, takes advantage of the discrete sampling of brain activity made possible using echo-planar imaging. Thus, the acquisition of a volume is time-locked to the onset of an event of interest. Because of the delay of the hemodynamic response to the onset of event-related neural activity, the acquisition is typically delayed by 4–6 seconds. In slow event-related fMRI, successive events are spaced 10–12 seconds apart to allow for the return of the signal to baseline before the next acquisition. This arrangement also makes it possible to present stimuli (e.g., tones) and record verbal output in a quiet environment, which minimizes the effect of speech-related head movements on image acquisition. In rapid event-related fMRI, the events occur quickly (<2 seconds), one after another, resulting in overlapping hemodynamic responses generated by the successive events; this overlap is deconvolved at the analysis stage. In both types of event-related fMRI, the relative timing of the events and acquisitions can be systematically varied (jittered) to allow for a detailed temporal analysis of the hemodynamic response function. Although the latter approach virtually guarantees that the “peak” response will not be missed because of suboptimal event-acquisition timing, it also reduces the proportion of samples with a high-enough signal. The event-related design is particularly useful when various types of events are interwoven (e.g., verbs and nouns in a sentence) or when the events are classified post hoc according to the subject's response during scanning (e.g., errors, onset of hallucinations).
POSITRON EMISSION TOMOGRAPHY
In adults, positron emission tomography (PET) is one of the in vivo techniques used to assess the state of neurotransmitter systems, such as the activity of enzymes involved in the synthesis or metabolism of a given neurotransmitter or the number of receptors present. To assess the distribution of a particular receptor or any other physiological process, a small amount of a specific radioactive tracer is injected into the bloodstream. (See Table 2 for an overview of common PET tracers.) Provided that radiochemists develop a suitable PET tracer, a large variety of physiological processes can be measured.32 Mainly for reasons of radiation safety, this approach is not used in studies of healthy children and adolescents.
Table 2. Common positron emission tomography tracers (radioligands), their half-life, and physiological processes measured
ELECTRO- AND MAGNETOENCEPHALOGRAPHY
Electroencephalography (EEG) and magnetoencephalography MEG allow more-direct and real-time measurements of brain activity than can be achieved with fMRI; they do this by recording the electrical and magnetic signals, respectively, generated by brain tissue.33,34
The main sources of these signals are intra- and extracellular currents, or field potentials, elicited by the activation of excitatory and inhibitory synapses. Spatial and temporal summation is necessary to generate signals strong enough to be detected from outside the head (Fig. 2).34 Such a summation occurs most often during simultaneous excitatory inputs onto apical dendrites of pyramidal cells; the apical dendrites are for the most part oriented in parallel with each other and are perpendicular to the cortical surface. Release of excitatory neurotransmitters causes a local change in membrane potential, namely an excitatory postsynaptic potential. Each excitatory postsynaptic potential generates a potential gradient along the membrane that, in turn, gives rise to the field potential.
EEG detects field potentials, regardless of their orientation, relative to the skull. In contrast, MEG can measure only magnetic fields perpendicular to the skull. Such fields are generated by tangential current dipoles and, because of the orientation of pyramidal cells and their apical dendrites in the cortex, reflect primarily postsynaptic activity occurring in the cerebral sulci (folds).
EEG is the oldest of all brain-mapping techniques, having been used to measure activity of the human brain for more than 70 years. The wealth of knowledge accumulated over this period is contained in many handbooks that can be consulted for detailed explanations of issues briefly discussed below.35–37 Three elements are essential for signal acquisition: electrodes, amplifiers, and a computer. The use of multiple (64, 128, or 256) electrodes necessitates the use of different types of electrode caps or sensor nets. The EEG signal picked up by the electrodes is fed to a set of EEG amplifiers. The bandwidth of the amplifiers must be such as to allow undistorted measurement of the expected signal frequencies; in most studies, the adequate range is between 0.16 Hz and 100 Hz and can be changed using high-pass and low-pass filters, respectively. Finally, the continuous analog outputs of the amplifiers are sampled with analog-to-digital converters and processed online (display) and offline (analysis) by a computer. The sampling rate is important because of the possible signal distortions it can introduce; the highest frequency that can be represented in the digitized signal is half the sampling rate – the so-called Nyquist frequency or limit. The most common sampling rate sufficient for the majority of brain-mapping applications is 256 samples/s. In addition to the EEG signal, the timing of different events such as the onset of stimuli and participants' responses is recorded in the same file.
Magnetic fields generated by the brain are on the order of tens of femtotesla (fT; 1 fT = 10−15 T); in comparison, the Earth's magnetic field is 10−3 T, and magnetic fields generated by electric equipment (e.g., elevators) are on the order of 10−9 T. (See Table 60.1 in Hari.34) To record such a weak signal, sensitive measuring instruments are operated in magnetically shielded rooms. The basic element of all commercially available MEG systems is the Superconducting QUantum Interference Device (SQUID). The SQUID-based magnetometer measures the brain's magnetic field using a superconducting flux transformer containing a single pick-up loop. To minimize the effect of distal environmental noise, another coil is added above (axial first-order gradiometer) or within (planar gradiometer) the plane of the pick-up coil. Thus, gradiometers and magnetometers allow investigators to measure the relative and absolute magnitudes, respectively, of the brain's magnetic fields. Current MEG systems contain between 122 and 306 channels, covering the entire convexity of the two cerebral hemispheres.
The design of EEG and MEG studies is conceptually similar to that of event-related fMRI. A variable number of trials are presented during a session while brain electric or magnetic signals are continuously recorded together with the stimulus and response onset. Depending on the task, the number of trials necessary to obtain a clear EEG or MEG response to a given event varies widely (40–200 trials). The length of an intertrial interval depends on the expected latency of a given component of the event-related potentials (ERPs) and can vary between 0.2 seconds and 5 seconds. If two types of events were studied, the recording session would last for about 20 minutes (2 events × 100 trials × 5-second intertrial interval). In the case of multichannel EEG, a significant amount of time is required for electrode placement and an impedance check (∼30 minutes) and for sampling of the exact electrode positions with frameless stereotaxy (15 minutes). The main advantage of multichannel EEG is the minimal demand on the subject vis-à-vis head restraint and low (if any) discomfort during the recording session; this is why EEG is still the most popular technique in developmental studies.38–40
The analysis of EEG data for brain-mapping purposes focuses typically on two domains: event-related potentials and event-related synchronization (ERS) and desynchronization (ERD) of brain activity. Event-related potentials are stimulus-locked, small (2–20 µV) potential changes extracted from the ongoing “background” EEG activity by averaging tens of brief EEG epochs surrounding the stimulus. The latency and amplitude of individual ERP components can be quantified in the individual averages; this is usually done at an electrode yielding a prototypical waveform. Location of the brain sources of a given ERP component, or any other feature of the EEG signal, from the scalp-recorded potentials is a complex undertaking. (For an overview of different methods, see Lagerlund.41) Model-dependent methods approach the so-called inverse problem (e.g., identification of the intracranial source that generated a particular distribution of scalp potentials) by making a priori assumptions about the number, type, and location of the dipole. Model-independent methods do not require any assumptions to be made about the number and configuration of the sources in the brain. A simple topographic display of the scalp potentials provides spatial information about the possible sources of the signal with minimal manipulation of the raw data; interpolation of potentials to intermediate points between the scalp electrodes is the only computation required. (See Lagerlund41 for different interpolation algorithms.) Owing to the blurring properties of the tissue located between the assumed (cortical) source and the recording electrodes (e.g., the skull and scalp), maps of scalp potentials are rather broad even when recorded with a high number of scalp electrodes. Spatial resolution of the topographic maps can be increased using a variety of methods (reviewed in Lagerlund41), including the calculation of the Laplacian of the scalp potential42 and minimum norm estimates,43,44 as well as the use of deblurring algorithms.45 Acquisition of a structural magnetic resonance image of the subject's brain together with the recording of the scalp position of the electrodes greatly facilitates location of the neural generators.
ERS and ERD allow the study of changes in spontaneous rhythmic activity of the brain elicited by different events. (For an overview, see Pfurtscheller.46) Unlike ERPs, ERDs and ERSs are typically not phase locked to the event onset; the temporal relationship between an event and EDR and ERS is less sharp, developing over a period of tens of milliseconds. The basic steps involved in the analysis of EDR and EDS include the following: 1) bandpass filtering of raw EEG data obtained in each trial epoch (e.g., 4 seconds before and 3 seconds after event onset); 2) squaring (rectifying) the amplitude of the filtered signal to obtain power measurements; 3) averaging over all trials; and 4) averaging over a small number of consecutive power samples to reduce the variance.46 Decreases and increases in power relative to a reference interval are referred to as ERDs and ERSs, respectively.
The main advantage of MEG over EEG lies in the virtual transparency of the skull and other extracerebral tissue to magnetic fields and the reference-free recording of local magnetic fields. These two features facilitate the location of brain sources of signals recorded outside the skull. The analysis of MEG signals employs virtually the same set of techniques described above; ERPs as well as ERDs and ERSs are of primary interest in brain-mapping studies performed with MEG. (See Hari34 for an overview.)
A number of techniques are available for the assessment of nutrition-related variations in brain structure and function. With the exception of PET, all of the methods reviewed here can be readily applied from childhood onward. Structural MRI and resting EEG procedures have been used in healthy, unsedated infants as young as 7 days old. (See Leppert et al.47 for MRI and John et al.48 for EEG). fMRI and event-related EEG and MEG require a great degree of cooperation from the patient and, without extensive training, are feasible on subjects aged approximately 5 years or older. As expected, test–retest reliability is higher for structural than functional assessment (Fig. 3).14
What is the optimal imaging modality when setting up observational studies or RCTs? In large (N > 100) studies and RCTs, structural MRI is the best imaging modality for evaluating the long-term effects of nutrition on the brain. It has high test–retest reliability, it is versatile vis-à-vis the type of sequences (imaging modalities) available, and it allows one to derive a wealth of quantitative information from each modality (e.g., from a T1-weighted image). In RCTs with modest sample sizes (N < 100), a combination of structural MRI and fMRI with EEG would provide the most-comprehensive assessment of brain structure and function and, hence, offer insights into possible brain mechanisms underlying the effect of nutrients on cognition and mental well-being. Finally, RCTs testing the specific effects of nutrients on a particular neurotransmitter system need to consider the use of PET, limiting this type of work to a small number (N < 20) of adult participants. Overall, brain imaging offers a rich armamentarium of acquisition and analysis tools for the quantitative in vivo assessment of the effects of nutrition on the human brain.
The author thanks Rosanne Aleong for reviewing the manuscript.
Funding. The author's research is supported by the Canadian Institutes of Health Research, the Royal Society (United Kingdom), the National Institutes of Health (United States), the Sixth Framework Program of the European Union and Unilever. The author received a small honorarium for writing this article. The coordinator for this supplement was Ms Agnes Meheust, ILSI Europe.
Declaration of interest. This work was commissioned by the Nutrition and Mental Performance Task Force of the European branch of the International Life Sciences Institute (ILSI Europe). Industry members of this task force are Abbott Nutrition, Barilla G. & R. Fratelli, Coca-Cola Europe, Danone, Dr Willmar Schwabe, DSM, FrieslandCampina, Kellogg Europe, Kraft Foods, Martek Biosciences Corporation, Naturex, Nestlé, PepsiCo International, Pfizer, Roquette, Soremartec – Ferrero Group, Südzucker/BENEO Group, Unilever. For further information about ILSI Europe, please call +32-2-771-00-14 or email: email@example.com. The opinions expressed herein are those of the authors and do not necessarily represent the views of ILSI Europe.