Dynamic imaging of brain function

Abstract

In recent years, there have been unprecedented methodological advances in the dynamic imaging of brain activities. Electrophysiological, optical, and magnetic resonance methods now allow mapping of functional activation (or deactivation) by measurement of neuronal activity (e.g., membrane potential, ion flux, neurotransmitter flux), energy metabolism (e.g., glucose consumption, oxygen consumption, creatine kinase flux), and functional hyperemia (e.g., blood oxygenation, blood flow, blood volume). Properties of the glutamatergic synapse are used as a model to reveal activities at the nerve terminal and their associated changes in energy demand and blood flow. This approach reveals that each method measures different tissue- and/or cell-specific components with specified spatiotemporal resolution. While advantages and disadvantages of different methods are apparent and often used to supersede one another in terms of specificity and/or sensitivity, no particular technique is the optimal dynamic brain imaging method because each method is unique in some respect. Because the demand for energy substrates is a fundamental requirement for function, energy-based methods may allow quantitative dynamic imaging in vivo. However there are exclusive neurobiological insights gained by combining some of these different dynamic imaging techniques.

1. Introduction

The brain is a highly complex organ, both anatomically [1] and physiologically [2], requiring an impressive arsenal of technological tools to study it. In recent years, neuroscientists and neurophysiologists have benefited from the emergence of several dynamic imaging techniques. Currently a variety of electrophysiological, optical, and magnetic resonance methods allow in vivo probing of brain activities in terms of transients in neuronal activity as well as their fundamentally associated energetic and hyperemic events [3]. Since these functional imaging methods either measure directly or exploit relationships between electrical, metabolic, and/or hemodynamic changes, it is important to classify the underlying basic processes with appropriate spatial and temporal scales.

Functional integrity of the working brain is maintained by electrical communication amongst an enormous number of neurons with active partnership provided by astrocytes [4,5]. Cytological association of neurons and astrocytes with the microvasculature (Fig. 1) provides the framework that links activities at the nerve terminal to energy demand [6,7] and blood flow [8,9]. In the mammalian cerebral cortex, glutamate is the major excitatory neurotransmitter whereas γ-amino butyric acid (GABA) is its conjugate inhibitory partner. Together they constitute nearly 90% of cortical neurons [10]. Glutamate metabolism plays a central role in both glutamatergic and GABAergic synapses because glutamate is a precursor of GABA and it is a constituent of both neurons and astrocytes [11,12]. Thus featuring properties of the glutamatergic system seem to be an appropriate starting point for this discourse.

Figure 1.

Cytological association between microvasculature with neurons and astrocytes in the glutamatergic synapse.

2. Activities at the nerve terminal

Electrical communication between cells at the glutamatergic nerve terminal is characterized by 1–2 ms epochs of cellular discharge (or depolarization) which are followed by quiescent charging periods lasting a few ms [2]. Cellular excitability (or signaling) depends on ionic distribution across the cell membrane. The cell maintains a large difference in ionic distribution (~10:1) between the intra- and extracellular compartments: higher concentrations of sodium (Na⁺), calcium (Ca²⁺), and chloride (Cl⁻) ions outside and higher but unbalanced concentration of potassium (K⁺) ions inside. Since the resting cell membrane is only marginally leaky to ions, passage of ions between intra- and extracellular compartments is mediated predominantly by ion channels and pumps [13]. Both voltage- and ligand-gated ion channels allow ion movement down large chemical gradients, whereas ion pumps (e.g., Na⁺/ K⁺ ATPase) help restore the ionic distributions (back to that in the resting cell) by moving ions against large chemical gradients. Restoration of ion gradients (requiring energy input) is needed to keep neurons ready to discharge (or fire) whenever needed.

At rest, the cell membrane can be considered to be nearly impermeable to Na⁺ ions with a membrane potential of approximately −70 mV (resting potential). Upon depolarization, however, the membrane becomes almost freely permeable to Na⁺ ions and the potential is nearly reversed in an action potential (i.e., when the neuron has fired). Therefore complete depolarization of the pre-synaptic axon (>1 mm long) can be approximated by increased permeability of Na⁺ ions across the cell membrane. However complimentary roles of K⁺ and Ca²⁺ voltage-gated ion channels are important. Delayed increase in K⁺ conductance (in conjunction with delayed decrease in Na⁺ conductance) causes the membrane potential of the pre-synaptic neuron to ultimately move back towards the resting value following depolarization [14]. Rapid pre-synaptic Ca²⁺ influx (~200 µM) within 1–2 ms after onset of depolarization triggers vesicles to release glutamate into the extracellular space [15].

Glutamate discharge into the synaptic cleft (10–100 nm wide) is a vital step in transmission of the pre-synaptic signal onto post-synaptic elements [2]. Depolarization of the post-synaptic neuron is initiated by raising extracellular glutamate concentration (from nM to µM). Glutamate activation, mediated by ligand-gated ion channels on post-synaptic elements, occurs within 10 ms [16]. During transient binding of glutamate with these ionotropic receptors, Na⁺ ions enter post-synaptic dendrites. The pyramidal dendritic branching patterns (spanning several mm) are thought to play a role in amplification of post-synaptic depolarization [1,2].

What is the fate of extracellular glutamate after pre-synaptic release? There is a large glutamate concentration gradient between pre-synaptic and extracellular compartments (mM vs. nM) [17]. Also there is a high density of glutamate receptors on astrocytic end feet [13]. Together these favor extracellular glutamate release followed by glial uptake of glutamate (which is co-transported with Na⁺ ions). Astrocytic glutamate uptake prevents excitotoxicity, and in doing so, helps maintain glutamate homeostasis [11,12]. Glutamate and glutamine are more concentrated respectively in neurons and astrocytes. However glutamate is used in astrocytes to synthesize glutamine which in turn is a precursor of glutamate synthesis in neurons. Reestablishment of glutamate reserves (requiring energy input) is essential for continuous signaling at the glutamatergic nerve terminal. The glutamate-glutamine cycle, therefore, is a necessary part of overall cellular excitability [17], but more importantly, links membrane depolarization to synaptic activity [11,12]. A prudent advisory, however, to the highlighted events at the nerve terminal is that synaptic activity is not exclusively mediated by digital (i.e., all or none) signaling alone [2]. Membrane potential changes can also be quite graded (and slow) to produce analog signals which may also influence activities at the nerve terminal [18].

3. Energy demand and blood flow

Signaling at the glutamatergic nerve terminal, spanning wide bandwidths [19], depends on synchronized electrical as well as chemical events where neurons and astrocytes play complimentary roles [6,7]. Removal of any one step from the rest compromises function of the entire system [11,12]. While most of the energy is expended for moving Na⁺ and K⁺ ions against large chemical gradients, a small but non-negligible fraction is needed for intracellular Ca²⁺ homeostasis and neurotransmitter recycling/synthesis/repackaging [6,7]. Because activities at the nerve terminal are in continuous need of energy, demand for it is a fundamental requirement [20].

In no other organ is the continuous energy supply more imperative than in the brain. In humans, the brain is merely 2% of the body’s weight but it consumes more than 20% of the oxidative fuels in the entire body [21] and receives nearly 15% of the cardiac output [22] to supply nutrients (i.e., glucose and oxygen). Furthermore endogenous energy reserves in the brain – glucose (1–3 mM [23]), oxygen (50–100 µM [24]), glycogen (2–4 mM [25]), and creatine (8–10 mM [26]) – are minimal [27]. Normal function needs, therefore, blood circulation to efficiently provide nutrients (and remove waste) throughout the brain [28]. Energy demand and blood flow are well correlated over different brain regions [29].

The energetic costs for brain work are mainly met by ATP derived exclusively from glucose oxidation [30]. Glucose metabolism is limited by phosphorylation, not transport [31]. Glucose is rapidly delivered [32] by a variety of transporters [33] located at the blood-brain barrier and plasma membranes and then broken down by non-oxidative (~5% ATP) and oxidative (~95% ATP) pathways to maintain ATP concentration (2–4 mM [27]). Some glucose is stored by glial cells as glycogen [34]. Because astrocytes lack the enzyme to generate glucose from glycogen, glucose-6-phosphate generated in glia may be transferred to neurons to serve as an additional energy buffer [35]. Total creatine, which represents the quantities of phosphocreatine (PCr) and creatine (Cr), can undergo phosphorylation-dephosphorylation reaction catalyzed by creatine kinase (i.e., ADP + PCr ↔ ATP + Cr). Therefore PCr may provide an additional energy reserve when oxidative phosphorylation cannot maintain constant ATP supply [36]. These alternate energy reserves together can provide energy support for a short time (a couple of minutes) under ischemic conditions [27]. It is possible, however, that these extra non-oxidative pathways may provide faster ATP (in ms range) than from mitochondrial respiration [35,36].

To date, in spite of several decades of research, it remains unclear by what mechanism(s) nutrient supply (of glucose and oxygen) adjusts to changing energy demanded by neurons and astrocytes [37]. Many vasoactive agents have been implicated in mechanism(s) leading to functional hyperemia. Given the extremely low and high oxygen contents, respectively, in brain and blood (µM vs. mM [24]) but at the same time its ubiquitous need for oxidation of glucose (and other carbohydrates), should blood flow be tightly coupled to oxidative energy use, and thereby, suggest potential mechanism(s) for functional hyperemia [38]? Although theoretical [39] and experimental [40] studies show that these parameters are indeed well coupled, oxygen is not a candidate for a vasoactive agent because its excess has no impact on the hemodynamic response (within 500 ms) during sensory stimulation [41]. In other words, the system’s use of energy substrates is only based on demand [20], not availability [42]. Since suggestions of other agents (e.g., H⁺, K⁺, Ca²⁺, adenosine) have lacked clear evidence of an impartial role with functional hyperemia [37], new proposals [43] have shifted some attention to astrocytes as a key participant, thereby revising the century-old neurovascular coupling idea [44] to include glia (i.e., “neurogliovascular” coupling).

Recent opinions [5,9,12] suggest that neuronal glutamate release not only induces metabolic responses in glial cells but it may also even trigger hemodynamic events through them. This seems plausible given that astrocytes are proximal to both cerebral microvasculature [4,5] as well as glutamatergic and GABAergic neurons [45,46]. Past experiments [47] suggested that astrocytes are activated by glutamate uptake (and increase glial glucose utilization as well as glial lactate production; see refs. 11 and 12 for details). Recent immunohistochemical and molecular studies [48–51] have identified receptors for a variety of neurotransmitters (including glutamate) on specific cellular components of the microvasculature. Furthermore small increases (in nM range) in cytosolic Ca²⁺, both in the post-synaptic neuron and astrocytes mediated respectively by activation of ionotropic [52,53] and metabotropic [54,55] glutamate receptors, may stimulate enzymes like nitric oxide synthase (and others) to generate strong vasoactive agents like nitric oxide (and others). Together these new findings shed some preliminary [56,57] but not always complete [58] understanding about complexities of “neurogliovascular” coupling.

4. Functional brain imaging

The goal of the ideal functional imaging technique is to map task-induced changes in activity in the brain in vivo. The task may be sensory or cognitive and the extent of the activated region may be localized (as generally observed with sensory paradigms [59]) or delocalized (as often found with cognitive paradigms [60]). To what extent do different methods allow dynamic identification of task activated brain region(s)? The answer depends on the spatiotemporal resolution of the method. There are at least three relevant factors. Spatial coverage of the brain by the method is a practical limitation. Specificity depends on the physiological basis of the measurement. Sensitivity if enhanced can improve either the spatial or temporal resolution, but rarely both. Together these factors describe the spatiotemporal resolution of the method (Fig. 2). Other (e.g., hardware) factors that affect spatiotemporal resolution of a given method are beyond the scope of discussion here.

Figure 2.

Comparison of spatiotemporal resolutions of current methods used in neuroscience.

Schematic illustration of spatial and temporal resolution ranges from a few experimental techniques is shown in Fig. 2. Since it is impractical to refer to the several dozen studies that were used to develop this figure, three reports [61–63] with similar comparisons are highlighted. But these may be questioned for several reasons. Techniques which did not reflect dynamic activity changes in their measurements (i.e., static or steady-state) were included with methods that were strictly designed for measuring transient signal changes. Methods specifically designed for use in humans were not separated from those applied in small animals. Similarly differences between in vivo and in vitro methods were not distinguished. Thus the apparent superiority of one method vs. another is debatable. Because there have also been technological advancements that influence this type of cross comparison across methods since then, up-to-date results were compiled from in vivo dynamic imaging studies of primarily small animals. Unfortunately these criteria excluded two dominant techniques in functional brain imaging: ¹⁴C-deoxy-glucose (¹⁴C-DG) autoradiography and positron emission tomography (PET).

¹⁴C-DG, a glucose analog, measures glucose consumption by the amount of metabolically trapped ¹⁴C-DG-6-phosphate throughout the brain [64]. However about an hour is required for radioactivity to build up in the tissue. ¹⁴C-DG autoradiography provides superior spatial resolution (50–100 µm) but it is a terminal experiment and therefore used only in animals.

PET, fashioned after the ¹⁴C-DG method but designed for application to humans [65], has several methods for functional imaging. Glucose consumption measurement is analogous to the ¹⁴C-DG method except that ¹⁸F-fluoro-deoxy-glucose (¹⁸F-FDG) is the glucose analog for PET. ¹⁵O-water is used to measure blood flow. This is done by injecting ¹⁵O-water into the blood stream and then detecting the rate of appearance of ¹⁵O-water in the brain tissue. Carbon monoxide with ¹¹C or ¹⁵O, which mimics binding of oxygen to hemoglobin in red blood cells, is used to measure blood volume. The principle is that carbon monoxide is blood borne and therefore the detected radioactivity is only representative of the vascular compartment. The blood flow and blood volume methods in combination with inhaled ¹⁵O-oxygen can be used to measure oxygen consumption by detecting the rate of metabolized ¹⁵O-water in brain tissue. In all of these studies the radioactive labels are at trace levels (µM or less). Most PET methods use radioactive isotopes that are relatively short-lived (<30 min), but as in ¹⁴C-DG autoradiography, require long accumulation times and/or repeated scans (5–30 min) for radioactivity to build up. PET provides spatial resolution appropriate for human brain imaging (3–5 mm) and is well suited for clinical use. Recent advancements have improved PET resolution further (1–3 mm) to image brains of smaller animals [66].

The techniques shown in Fig. 2 include a variety of electrophysiological, optical, and magnetic resonance methods which are now widely used in many neuroscience laboratories, both in animals and in humans. These methods dynamically measure everything from changes in neuronal activity (e.g., membrane potential, ion flux, neurotransmitter flux) to the coupled alterations in energy metabolism (e.g., glucose consumption, oxygen consumption, creatine kinase flux) and hemodynamics (e.g., blood oxygenation, blood flow, blood volume).

4.1 Electrophysiology

Electrophysiology deals with the study of the electrical properties of cells and tissues [2]. Classical electrophysiological methods, using different types of microelectrodes, measure changes in voltage (or current). Microelectrodes are inherently designed to cover a few specific regions because it is impractical to perforate very large areas of the brain. Microelectrode arrays cover slightly larger (e.g., 4×4 mm²) superficial regions of the brain.

Intracellular recordings involve direct access into the intracellular milieu so that membrane potential can be measured with microelectrodes. The obvious advantage with intracellular recording is the unrivaled information about the individual cell, but the disadvantage is the difficulty in sampling large number of neurons. Extracellular recordings are made with the microelectrode tip (1–3 µm) situated next to cell bodies so that the extracellular voltage can be measured. Depending on the impedance and location of the electrode tip, extracellular recordings can capture the activity from either a single neuron (i.e. single-unit activity (SUA) recording) or from several neurons (i.e., multi-unit activity (MUA) recording). The clear benefit of extracellular recording is that several individual neurons can be assayed simultaneously. However only the largest signals from neurons closest to the microelectrode tip can be reliably identified individually.

Since microelectrodes are designed to measure the high frequency action potentials (or spikes), the data are collected with high temporal resolution (< 100 µs). The high signal-to-noise ratio of action potentials can be sorted out from the lower amplitude signals (by spike sorting). The extracellular signal, however, is also susceptible to the slower waves representing the local field potential (LFP) which may arise from graded events at the nerve terminal as shown in Fig. 1. Thus appropriate filtering is applied to separate the MUA (10²–10⁴ Hz) from the LFP (<10² Hz) signals. SUA and MUA signals are typically representative of signals from neuron(s) in the microelectrode’s vicinity (spanning 10–100 µm). LFP signals can integrate potentials over much larger distances (µm to mm) to represent the aggregate cellular activity in that region.

Both electroencephalography (EEG; [67]) and magnetoencephalography (MEG; [68]) allow mapping with much better brain coverage because the measurement devices nearly surround the entire head. EEG measures summed activity of post-synaptic currents, whereas MEG measures tiny magnetic fields (in fT range) produced by synchronized ionic currents flowing in dendrites. Since EEG and MEG signals originate from slightly different cortical locations and are acquired with different temporal resolutions (ms vs. µs), the types of information obtained vary. Because both EEG and MEG suffer from the inverse problem (i.e., difficulty localizing origin of signal), the spatial resolution (in cm range) is compromised. EEG signals are susceptible to body movements and MEG signals from the brain sometimes compete with higher magnitude environmental noise. However non-invasive use of EEG and MEG in humans is invaluable for basic science and clinical research.

4.2 Optical imaging

Optical imaging techniques are based on the discovery [69] that optical properties of cells, and therefore tissue, change during activity. Sensitivity of all optical techniques is limited by the magnitude of the changed optical properties of the system. At present, four dynamic optical techniques dominate: optical intrinsic signal (OIS), voltage sensitive dye (VSD), near infra red spectroscopy (NIRS), and laser-Doppler flowmetry (LDF). These methods can be used to measure a variety of dynamic events in cells (e.g., ion flux) and vessels (e.g., blood oxygenation). OIS and VSD are primarily applied in animals because of the need for a window (e.g., 4×4 mm²) on the skull and typically have high spatial resolution (10–50 µm). The data are limited to superficial regions of the brain because of limited light penetration in tissue (mm range). NIRS is a non-invasive technique and suited for use in humans. The spatial resolution of NIRS and LDF is limited by the distance between the emitting and detecting probes. Most of these techniques allow two dimensional images of brain activity with relatively high temporal resolution (10–100 ms).

OIS is perhaps the most widely used optical technique in neuroscience [70], in part, because it does not require exogenous optical probes. However the origin of the task-dependent changes in the intrinsic signal is complex. Changes in blood oxygenation (increase or decrease in deoxyhemoglobin), blood volume (vessel dilation, capillary recruitment, etc.), and light-scattering that accompany cell swelling with activation, all can affect the intrinsic signal in different ways and to various extents. Because these optical properties of the tissue vary in time, multiple illumination wavelengths (e.g., 600–630 nm for oxymetry) are necessary to calculate the desired parameters. The intrinsic signals are based on small reflectance signal changes (0.1–1%) which peak shortly (2–3 sec) after stimulus onset.

VSD uses small exogenous probes (weight <500 Da; length < 200 nm) that are designed to preferentially bind to the periphery of the cell membrane, and therefore, act as molecular transducers [71]. Changes in the membrane potential or ion concentration can then be transformed into an optical signal. There are two key factors to a successful dye used in VSD imaging. The optical yield (or contrast-to-noise ratio) when the molecule binds to the membrane is obviously critical. Another factor, which partly limits its application more widely, is successful delivery of the dye to the site of interest. Since the blood-brain barrier limits most foreign molecules to enter the extracellular milieu, direct injection of the dye into the site of interest is preferred. Because different events at the nerve terminal are linked to intracellular Ca²⁺ ion, dyes sensitive to Ca²⁺ voltage-gated ion channels have wide use in imaging activities both neurons and astrocytes with good temporal resolution (<10² ms). The VSD signals changes can be quite large (2–4%) which peak quite rapidly (<1 sec) after stimulus onset.

NIRS is a spectroscopic method which utilizes the near infra-red region (650–950 nm) of the electromagnetic spectrum to sense changes in the hemoglobin oxygenation state [72]. Because NIRS is non-invasive and has better tissue penetration (cm instead of mm) than visible light, the method has found wide use in both clinical monitoring of tissue oxygenation and functional imaging in humans and small animals [73]. Since near-infrared absorption spectra of oxyhemoglobin and deoxyhemoglobin allow separation of these two species, NIRS measurements commonly utilize three separate wavelengths to measure these components for blood oxygenation, and later sum them to get blood volume. The spatial resolution (in mm range) is lower than other optical methods because of the diffuse nature of photon migration through tissue which can affect sensitivity and quantitation of oxyhemoglobin and deoxyhemoglobin [74]. The NIRS signals changes larger than intrinsic signal changes (>2%) and peak slowly (within 5 sec) after stimulus onset.

LDF is an optical technique for measuring movement of red blood cells in the microvascular bed [75]. The principle is based on the “Doppler” frequency shift that arises in light that is scattered by moving red blood cells. By illuminating the tissue with single-frequency (low power) laser light and detecting the frequency distribution of the back-scattered light, red blood cell velocity and/or flux can be estimated. Since the LDF device is usually standardized with suspensions (e.g., latex spheres undergoing Brownian motion), it is a relative method and cannot easily provide absolute units of blood flow. LDF is a widely used technique in laboratories because of its minimal invasiveness and relatively high sensitivity for stimulation-induced blood flow/volume changes [76]. LDF has found popular applications in the clinical environment (e.g., neonatal care [77]) because it is quite simple to use. The LDF signals changes can be quite large (10–100%) and peak faster than NIRS signals (within 4 sec) after stimulus onset.

4.3 Magnetic resonance

Magnetic resonance (MR) is based upon the quantum mechanical properties of an atom’s nucleus [78]. Nuclei that contain odd numbers of protons or neutrons have an intrinsic magnetic moment. Some MR visible and biologically relevant non-radioactive isotopes are ¹H, ¹³C, ¹⁵N, ¹⁷O, ¹⁹F, ²³Na, and ³¹P. Sensitivity of MR detection for a given isotope depends on the physical intrinsic property of the isotope (cannot be changed, easily) and the abundance of the isotope prescribed by nature (can be enhanced by enrichment). The most sensitive and weak isotopes that are commonly used for in vivo MR experiments are ¹H and ¹³C, respectively. ¹H has four times stronger intrinsic sensitivity than ¹³C. Also ¹H is ~91% more naturally abundant than ¹³C. Together, this makes ¹H orders of magnitude more sensitive than ¹³C.

MR imaging (MRI) is primarily used in diagnostic imaging to map structure, but lately also function. MRI is predominantly based on ¹H signal from water. MRI data have good spatial resolution and the temporal resolution depends on the specific type of function being measured (see below). MR spectroscopy (MRS) is a powerful technique used to obtain rich chemical information about a wide range of biomolecules (other than water) based on their variable chemical shifts. MRS detection of an isotope depends on the experimental purpose (see below). MRS data can also be viewed in two or three dimensions but the images do not appear as crisp as MRI data because the MRS voxels are much larger. The lower spatial resolution of MRS (mm to cm) compared to MRI (µm to mm) is mainly because of the orders of magnitude lower concentration of biomolecules (mM range) compared to water (>10 M) being detected. Since a wide range of MR methods can be used for neurobiological studies [79], the few that meet the criteria of dynamic functional imaging in Fig. 2 is discussed.

4.3.1 MRS methods

In the past, ³¹P MRS had been mainly used to detect high energy phosphates (e.g., ATP, ADP) which lead to assessment of tissue pH [80]. Since these signals do not change significantly with modest functional challenges, ³¹P MRS found little use in functional brain imaging. Recently, due to better sensitivity at high magnetic fields, ³¹P MRS has been able to measure forward and backward fluxes of creatine kinase [81]. Although the temporal resolution of this approach is still low (seconds to min), there is great potential of its use because comparable techniques do not exist yet.

Similar to ³¹P MRS, ¹⁷O MRS was also demonstrated nearly two decades ago [82] and due to recent advancements, the method has been revitalized [83]. The ¹⁷O MRS method for oxygen consumption follows almost the same approach as ¹⁵O PET (i.e., inhale labeled-oxygen and detect labeled-water). However a key difference between the two methods is that ¹⁷O MRS can distinguish between labeled-water (very strong MRS signal) and labeled-oxygen (exceptionally weak MRS signal), whereas PET cannot. This allows for removal of a few assumptions for modeling oxygen consumption with ¹⁷O MRS. The ¹⁷O MRS method can also measure blood flow and follows the same principles as ¹⁵O PET (i.e., inject labeled-water in blood and detect appearance of labeled-water in brain).

Although ¹⁹F is intrinsically almost as sensitive as ¹H and it is naturally 100% abundant [78], there are no biomolecules in vivo (of relevance) with fluorine. Therefore FDG, as in PET, has been applied with ¹⁹F MRS for glucose consumption measurement [84], but the MRS tracer is not radioactive. Furthermore ¹⁹F MRS can distinguish between FDG-6-phosphate created by metabolism of FDG from FDG itself, whereas ¹⁸F PET cannot. Dynamic separation between the amount of FDG-6-phosphate and FDG helps quantify the glucose consumption rate. For ¹⁹F MRS detection of these signals with modest signal-to-noise ratio, either very high FDG dose (>10 mM) have been used with acceptable acquisition times or very long acquisition times (minutes to hours) have been used with low FDG dose (<1 mM). More recently, due to superior sensitivity at high magnetic fields, ¹⁹F MRS with reasonable spatial (mm range) and temporal resolution (<10 min) has been demonstrated using a low FDG dose [85].

Glucose consumption measured by the FDG method (MRS or PET) has a non-oxidative and oxidative component, where the latter can be measured by ¹³C MRS. Although ¹³C detection sensitivity is quite low (see above), using ¹³C-labeled exogenous biomolecules provides a viable alternative because of compartment/substrate specific information. ¹³C MRS is a very powerful tool to study metabolism when it is used in combination with a ¹³C-labeled substrate, such as [1-¹³C]glucose [86]. Infusion of [1-¹³C]glucose (into the blood stream) results in ¹³C turnover into glucose, glutamate, glutamine, GABA, lactate, aspartate, etc, in brain tissue [25,42]. Since ¹³C turnover into these pools is time dependent, coupled differential equations can be used to estimate a range of fluxes (including neuronal/glial glucose oxidation, glutamate-GABA-glutamine cycling; see refs. 11, 17, and 79 for details). But this is a steady-state method where the measured turnover times are on the order of about half an hour (or slightly more). Recently, however, by dynamic nuclear polarization (DNP) the MR detection sensitivity can be artificially improved by orders of magnitude [87]. To date, this method (using injection of hyperpolarized ¹³C or ¹⁵N biomolecules) has primarily been used to demonstrate perfusion through organs with quite high temporal (within seconds) and spatial (below mm range) resolutions. Although technical challenges remain [88], study of metabolism with ¹³C and/or ¹⁹F MRS in conjunction with DNP is quite promising.

4.3.2 MRI methods

There are several hemodynamic-based functional MRI (fMRI) methods that are sensitive to functional hyperemic events. The blood oxygenation level dependent (BOLD) method [89] is the most popular, in part, due to its simplicity. Both BOLD and NIRS methods are sensitive to changing concentrations of oxyhemoglobin and deoxyhemoglobin (i.e., diamagnetic or red blood, and paramagnetic or black blood, respectively). With NIRS both moieties are detected (by different absorption bands), with BOLD the magnetic properties of blood changing with oxygenation influences the tissue water signal (i.e., less paramagnetic deoxyhemoglobin, higher MRI signal).

Just as hemoglobin is nature’s endogenous MRI contrast agent to measure blood oxygenation changes, blood borne exogenous MRI contrast agents are used to measure alterations in blood volume with almost the same assumptions (i.e., more paramagnetic agent, lower MRI signal) [90]. Principles of blood flow measured by MRI are similar to that in PET. Following magnetic labeling of arterial blood (water), subsequent dynamic MRI maps reflect the degree to which the label has decayed (by mixing with unlabeled water), thereby reflecting tissue perfusion [91].

Since its discovery, the biophysical understanding of BOLD has improved significantly. Better specificity of the BOLD effect at higher magnetic fields [92] allows changes in oxygen consumption to be titrated out by combining the contributing changes in blood flow and blood volume with the BOLD signal [93]. At steady-state, the oxygen consumption predicted by BOLD calibration has been shown to be in good agreement with oxygen consumption measured by ¹³C MRS [94] and which in turn concurs with ensemble neuronal activity measured by extracellular recordings [95]. Since these multi-modal MRI methods can be applied dynamically [96], there is potential for transient energetics with higher temporal resolution (200–500 ms) [97].

Recently there has been a surge in fMRI that are based on non-hemodynamic events. Two prominent directions are effects of electrical currents directly or indirectly on the MRI signal [98] and task-related structural changes which may lead to alteration of diffusion of water molecules [99]. Both of these emerging fMRI approaches have potential to improve temporal resolution because the signals are not delayed by hemodynamics. However the greater challenge is to enhance the magnitude of the non-hemodynamic effects (<1%) which are much smaller than hemodynamic-based fMRI (>3%).

5. Future

The brain is mapped by several dynamic imaging methods sensitive to a variety of events at the glutamatergic nerve terminal (Fig. 1). These methods span wide ranges of spatial and temporal resolutions (Fig. 2) where each technique has obvious limitations, but also definite advantages. No one method can cover the several orders of magnitude in temporal and spatial resolutions and at the same time capture the many cellular and vascular events. Any such promise [100] discounts potential reachable insights, even at present, gained by combining different techniques which compliment each other [101]. The recent trend to combine fMRI [102–105] or optical imaging [106–109] with electrophysiology is in accord with this suggestion. Since energy is the currency of trade between cellular events and the substrate that those activities demand, dynamic functional imaging with energetics may be a crucial direction to pursue in the future [79]. In functional imaging, the qualitative way by which the baseline signal is conveniently differenced away to reveal stimulation-induced focal area(s) of interest ignored the high energy utilization in the resting brain [6] for a long time. Recent awareness [6,110] of the high baseline activity, which slowly varies in time, has been investigated by multi-modal methods like fMRI [111], optical imaging [112], and electrophysiology [113]. The anatomical [1] and physiological [2] complexities of the brain, both at rest and during activation, reward an extraordinary set of dynamic imaging [2,3,79] and analytical [114,115] tools.