|Home | About | Journals | Submit | Contact Us | Français|
In vivo optical imaging of cerebral blood flow (CBF) and metabolism did not exist 50 years ago. While point optical fluorescence and absorption measurements of cellular metabolism and hemoglobin concentrations had already been introduced by then, point blood flow measurements appeared only 40 years ago. The advent of digital cameras has significantly advanced two-dimensional optical imaging of neuronal, metabolic, vascular, and hemodynamic signals. More recently, advanced laser sources have enabled a variety of novel three-dimensional high-spatial-resolution imaging approaches. Combined, as we discuss here, these methods are permitting a multifaceted investigation of the local regulation of CBF and metabolism with unprecedented spatial and temporal resolution. Through multimodal combination of these optical techniques with genetic methods of encoding optical reporter and actuator proteins, the future is bright for solving the mysteries of neurometabolic and neurovascular coupling and translating them to clinical utility.
With this Special Issue we celebrate 50 years of dedicated symposia on Cerebral Blood Flow and Metabolism. During the past half century we, as a research community, have accumulated a considerable body of experimental and theoretical knowledge on cellular metabolic pathways in health and disease, identified a variety of vasoactive substances, established correlations between vascular, metabolic, and neuronal parameters, developed computational models and took aboard a broad suite of methodologies. Yet, a central piece of the cerebrovascular puzzle is missing: Despite a number of hypotheses (for recent reviews see Attwell et al, 2010; Cauli and Hamel, 2010; Hamilton et al, 2010; Iadecola and Nedergaard, 2007; Kleinfeld et al, 2011; Paulson et al, 2010), we still do not have a clear mechanistic understanding of local regulation of cerebral blood flow (CBF) and metabolism by neuronal activity. By ‘mechanistic' we mean determining causal relationships and identifying molecular messengers, which communicate a change in neuronal activity to the vasculature causing dilation or constriction. What makes the neurovascular signaling so difficult to grasp and what is required for a breakthrough? In this essay, we argue that further advancement in a mechanistic understanding of neurovascular communication and dynamic regulation of blood flow critically depends on the advent of new imaging technologies with microscopic resolution applicable to in vivo studies.
The most intuitive scenario for neurovascular coupling might be that in which consumption of energy by neuronal tissue provides a feedback signal to the feeding vasculature: Changes in neuronal activity drive changes in energy metabolism, which then drive vasodilation/constriction and the associated changes in blood flow. This idea, usually referred to as the ‘metabolic hypothesis,' comes in different flavors with relation to the putative molecular mediators, including lactate, NAD+/NADH (nicotinamide adenine dinucleotide) ratio, ATP/ADP ratio, adenosine, and an (unidentified) O2 sensor (Paulson et al, 2010; Raichle and Mintun, 2006). As an alternative hypothesis, changes in neuronal activity can drive vasodilation and vasoconstriction by feed-forward mechanisms releasing neurotransmitter and neuropeptide molecules related to neuronal signaling (Attwell et al, 2010; Cauli and Hamel, 2010). In this ‘neurogenic hypothesis,' blood flow and energy metabolism are driven in parallel by neuronal activity. Astrocytes, ‘more than a glue' of the central nervous system (Allaman et al, 2011; Fiacco et al, 2009; Giaume et al, 2010; Iadecola and Nedergaard, 2007; Koehler et al, 2009), can potentially have a role in both scenarios: via release of vasoactive metabolic biproducts (metabolic) or synthesis and release of vasoactive gliotransmitters in response to neurotransmitters and neuropeptides (neurogenic). Supportive evidence for both hypotheses has been derived from experiments in isolated tissue: brain slices, excised vessels, and even cell cultures. Of these, brain slice preparation produced a wealth of data in experiments with controlled perfusion, pharmacological manipulations, and excitation of single neurons with identified phenotypes (Cauli et al, 2004; Gordon et al, 2008; Zonta et al, 2003). However, homeostasis of brain slices departs from that in vivo in many ways (Huchzermeyer et al, 2008; Turner et al, 2007). Significantly, many of these departures are unknown or difficult to quantify, sometimes making extrapolation of the observed phenomena to the in vivo situation uncertain.
The ability to descend to the single-cell and single-capillary levels in vivo and observe firing of individual neurons, vasodilation, glucose uptake, and infusion of O2 into the tissue—all while directly controlling neuronal activity—has long been a dream of scientists interested in understanding the complex regulation of blood flow and metabolism as related to neuronal activity. However, in contrast to the detailed and elegant mechanistic studies in isolated tissue, in vivo reports have, in the main, focused simply on correlations between the ‘observables,' limited by the available methods. This ‘too hard to do' status quo for mechanistic studies in vivo is starting to change, due to rapid developments in optical microscopy. In fact, already today, a versatile suite of optical tools is available for high-resolution, high-sensitivity measurements of vascular, metabolic, and neuronal parameters in deep tissue and local, cell-type specific manipulations of neuronal activity. Below, we consider the current state of the art of a number of key optical microscopy technologies that will be critical in the effort of graduating from correlation driven to mechanistic approaches for studies in vivo. The technological requirements necessary for this endeavor include
The tools suitable for unraveling the mechanics of neurovascular and neurometabolic coupling will be complemented by other noninvasive optical technologies that will enable translation of the physiological findings from animal to human studies and clinical application. Importantly, these noninvasive optical technologies can be used in both animals and humans and thus can facilitate the connection of microscopic to macroscopic observables from animals to humans.
The use of novel optical technologies has been instrumental for a number of central discoveries in both basic and clinical neuroscience. Examples from basic neuroscience include the fine mapping of cortical functional organization (Grinvald et al, 1986) and the discovery of glial calcium excitability (Cornell-Bell et al, 1990; Nedergaard, 1994). Among the clinical applications, optical tools played an important role in the study of neurovascular and neurometabolic disregulation in animal models of stroke (Zhang and Murphy, 2007), epilepsy (Schwartz and Bonhoeffer, 2001), migraine (Bolay et al, 2002), and cancer (Barretto et al, 2011). Likewise, noninvasive optical technologies have started making inroads into bedside imaging of blood flow and oxygen consumption in human patients (Grant et al, 2009; Mesquita et al, 2011).
Below, we highlight many of the optical methods used for vascular, hemodynamic, metabolic, and neuronal imaging at different resolution scales—from cellular to macroscopic—with an emphasis on in vivo methodology (Figure 1). We apologize in advance for the less-than-comprehensive coverage of this exceedingly broad topic. We have had to cite the literature sparsely, but have strived to include sufficient citations to lead the reader to more detailed information. Subcellular optical imaging methods (e.g., fluorescence resonance energy transfer (FRET)-based fluorescent methods to track protein–protein interactions) are beyond the scope of the current review.
Optical imaging can utilize several endogenous contrast mechanisms for vascular imaging, including hemoglobin absorption, red blood cell (RBC) motion-induced Doppler shifts, and many exogenous fluorescent contrast agents for labeling the blood plasma or RBC. These different contrast mechanisms are used to image hemoglobin concentration and oxygenation changes, to image blood flow, and to obtain angiograms of the microvascular network (Table 1).
Since hemoglobin is a dominant absorber in the brain tissue in the visible and near-infrared spectrum, changes in hemoglobin concentration and oxygenation associated with neuronal activity can be monitored and imaged via optical intrinsic signal imaging (OISI) (Figure 2A). Imaging of cerebral function using these intrinsic absorption changes in vivo was demonstrated over 25 years ago (Grinvald et al, 1986).
Optical intrinsic signal imaging is limited to imaging of the cortical surface, generally requiring a cranial window or thinned skull to illuminate the cortex with light and image the reflected light with a camera. The amount of light reflected from the cortical surface is modulated by changes in the absorption coefficient of the tissue. These changes are related to changes in the concentrations of HbO (oxy-hemoglobin) and Hb (deoxy-hemoglobin). Estimating the hemoglobin concentration changes, therefore, requires measurements at two or more wavelengths and depends on knowledge of the path length of light through the tissue (Kohl et al, 2000). The tissue scattering of light degrades image contrast and spatial resolution, with lateral resolutions ranging from 1 to 100μm, increasing with depth. The depth sensitivity is limited to the top 500μm with exponential weighting toward the surface (Tian et al, 2010a).
The use of OISI was in the center of the discovery of fine details of functional mapping and plasticity in the cerebral cortex (Grinvald et al, 1991; Kalatsky et al, 2005; Shtoyerman et al, 2000; Vnek et al, 1999) and played a key role in temporal parsing of the hemodynamic response, with implications for the order of neuronal, metabolic, and vascular events underlying functional hyperemia (Mayhew et al, 2000; Sheth et al, 2005; Vanzetta and Grinvald, 1999). Optical intrinsic signal imaging also played a significant role in studying pathological departures in neurovascular coupling in disease, such as an increase in Hb associated with focal epileptic seizures (indicating inadequate increase in CBF) (reviewed in Schwartz et al, 2011).
Greater depth penetration and depth resolution than is possible with OISI can be achieved with photoacoustic tomography (PAT) (Figure 2B), up to several centimeters deep, with spatial resolution of 5/15/500μm at depths of 0.7/3/50mm, respectively (Wang, 2009). Photoacoustic tomography utilizes the photoacoustic effect (Bell, 1880), in which a pulse of light scattering through the tissue is absorbed by hemoglobin (or any other absorber), producing local heating and thermal expansion, resulting in an acoustic wave. The position of the optical absorption is then recovered based on the time-of-flight of detected ultrasound waves, yielding a three-dimensional (3D) image based on optical contrast and ultrasound resolution. Exploiting hemoglobin absorption, PAT has been used to image vascular structure and oxygenation in small rodents at various spatial scales, frequently through the intact skull (Hu et al, 2009; Laufer et al, 2009; Wang et al, 2003).
Since diffuse photons contribute to the signal as much as ballistic photons, generation of ultrasound waves can be achieved deep inside the sample: the low scattering of ultrasound in soft tissue further facilitates detection from increased depths. Two main PAT system types have been introduced (Wang, 2009): reconstruction-based PAT (Wang et al, 2003) and raster scan-based photoacoustic microscopy (Hu et al, 2009; Zhang et al, 2006). Application of ultrasound arrays and pulsed lasers that can rapidly change excitation wavelength will increase acquisition speeds. Photoacoustic tomography systems are now becoming commercially available.
Although PAT is a new optical imaging method and its potential for neurovascular research being explored, it is likely to have a broad impact in the next 5 to 10 years due to improved resolution and a greater depth penetration in comparison with OISI.
Noninvasive human measurements of hemoglobin concentrations are routinely performed with near-infrared spectroscopy (NIRS) (Figure 2C). The principle of NIRS is equivalent to that of OISI, described above, in that near-infrared light is used to monitor the changes in hemoglobin concentration via absorption spectroscopy. While OISI is generally performed through a cranial window or thinned skull to provide direct viewing of the cortical surface, NIRS measurements are usually obtained through the intact skull and scalp. Near-infrared spectroscopy measurements of hemoglobin concentration changes in the human brain through the intact scalp and skull are feasible because of the weak absorption of near-infrared light by hemoglobin and sufficiently small tissue scattering, as first demonstrated by Jobsis (1977). By 1993, NIRS was being used to measure hemoglobin responses to brain activation in humans (Hoshi and Tamura, 1993; Villringer et al, 1993) similar to blood oxygen level-dependent functional magnetic resonance imaging (fMRI). While fMRI has the advantage of spatial resolution and whole brain sensitivity, NIRS has the physiological advantage of sensitivity to HbO and Hb as well as the advantage of being a portable instrument suitable for continuous bedside measurements. While continuous-wave NIRS is used to measure changes in hemoglobin, frequency-domain and time-domain NIRS devices enable estimation of the absolute hemoglobin concentrations (Wolf et al, 2007).
Because of the ease in applying NIRS to infants relative to fMRI, it is enabling the identification of neural regions associated with and responsible for the emergence of various behavioral traits in developing human infants (Lloyd-Fox et al, 2010). Clinically, it has revealed a hypermetabolic response to acute brain injury in newborn infants, potentially providing a tool to guide therapies aimed at reducing metabolic burden (Grant et al, 2009). Further, NIRS is broadly important in studying language function in healthy and psychiatric populations, as it permits natural language production and conversation (Dieler et al, 2011).
Laser speckle contrast imaging utilizes RBC-induced Doppler shifts of laser light to measure blood flow (Figure 3A). A temporally coherent light source is required to have sensitivity to the Doppler shifts. Laser speckle contrast imaging is analogous to laser Doppler methods (Briers, 2001), but instead of resolving the CBF-induced temporal fluctuations of the light at a single point, it images the blurring of the speckle contrast that occurs over the integration time of the camera, thus enabling near-real-time imaging of blood flow. The relationship between speckle contrast and blood flow and the underlying assumptions were recently reviewed in Boas and Dunn (2010). A recent advance of note is the ability to measure the temporal speckle contrast instead of spatial contrast. Temporal contrast affords better spatial resolution at the expense of temporal resolution and is less sensitive to the contaminating effects of static scattering from, for example, the skull. The spatial resolution and depth sensitivity of laser speckle contrast imaging (LSCI) are comparable to those of OISI.
Our own work with LSCI has revealed that cortical spreading depression during the aura preceding a migraine headache, activates trigeminal afferents, which results in inflammation of the pain sensitive meninges generating the migraine (Bolay et al, 2002), and that cortical spreading depression results in a prolonged state of hypoperfusion in mouse stroke models that results in further growth of ischemic cortex (Shin et al, 2006). Others have used LSCI to show local retinal vasodilation in response to focal light stimulation, revealing a neurovascular coupling similar to that observed in cortex (Srienc et al, 2010).
The technical aspects of two-photon microscopy (TPM) (Denk et al, 1990) and the spectrum of its applications in neuroscience have been extensively covered elsewhere (Svoboda and Yasuda, 2006). In the context of blood flow, the technique has enabled depth-resolved measurements of RBC velocities (Kleinfeld et al, 1998) and vascular diameters (Devor et al, 2007; Tian et al, 2010b) routinely at depths of up to 500μm deep in the cortex and as deep as 1mm when utilizing advanced laser systems (Kobat et al, 2009). In contrast with LSCI, which utilizes Doppler contrast from RBC motion, TPM velocity measurements typically utilize a fluorescent dye to image the blood plasma and track the RBC ‘shadows' to estimate velocity in individual capillaries (Figure 3B). While LSCI is generally performed through a thin skull, TPM typically requires a cranial window, although thin skull measurements are now being conducted at the expense of depth penetration (Drew et al, 2010). Velocity and diameter measurements are performed on no more than a few vessel segments at a time. As a result, studies that require measurements throughout the vascular geometry to ascertain the collective behavior are not practical with TPM.
Two-photon microscopy vascular measurements were instrumental in studying the reorganization of blood flow following experimental disruption of the vascular network (targeted ‘microstrokes') (Nishimura et al, 2006, 2007, 2010; Schaffer et al, 2006) and of dilation of cortical microvasculature below the confocal reach (Stefanovic et al, 2008; Tian et al, 2010b).
Doppler optical coherence tomography (OCT) enables depth-resolved imaging of blood flow in individual diving arterioles and ascending venules (Figure 3C). The penetration depth of OCT in highly scattering media can exceed 1mm (Izatt et al, 1994). Full volumetric imaging of blood flow over a cortical surface area of 1mm2 is possible in ~1minute. Commercial systems are now available, facilitating widespread adoption of OCT. Doppler OCT promises to be an important tool for studying cerebrovascular pathology.
Optical coherence tomography is in many ways analogous to ultrasound, though instead of measuring the scattering of sound waves by tissue, it measures the scattering of light waves. Optical coherence tomography uses the principle of low-coherence interferometry to resolve the delay between different light scattering ‘echoes.' A low-temporal-coherence light source provides a coherence gate that rejects multiply scattered light (Izatt et al, 1994) to improve contrast of tissue structure at depths greater than can be achieved with confocal microscopy, with a typical depth resolution of ~5μm and diffraction-limited lateral resolution typically ranging from 1 to ~20μm. In common practice, OCT is implemented as a scanning method like confocal or TPM and forms images by moving the light beam over the surface of the tissue.
All scanning methods result in a tradeoff between field of view and image acquisition rate. While confocal and TPM form an image at a single depth, OCT and PAT form images over a range of depths simultaneously. Simultaneous measurements over this range of depths is achieved in OCT by implementing recently developed Fourier domain detection techniques (Choma et al, 2003; Leitgeb et al, 2003), which offer tremendous improvements in the volumetric image acquisition rate relative to time-domain OCT (Huang et al, 1991). Importantly, OCT is sensitive to Doppler shifts in the scattered light that arise from moving RBCs, enabling high-resolution measurements of RBC velocities (Chen et al, 1997) and blood flow (Srinivasan et al, 2011; Wang et al, 2007b).
While the spectrum of brain OCT applications is still being explored, its utility for minimally invasive quantitative measurements of blood flow in vivo (Srinivasan et al, 2011) will become an important tool because of its improved spatial and temporal resolution over conventional gold-standard methods and because it can be applied in longitudinal studies.
Diffuse correlation spectroscopy (DCS) offers the ability to noninvasively measure CBF in humans through the intact scalp and skull (Kim et al, 2010), albeit with only superficial cortical sensitivity, lateral resolution of 1 to 3cm, and no depth resolution. It has been extensively crossvalidated against other blood flow measures (Kim et al, 2010) (Figure 3D). Diffuse correlation spectroscopy measurements are generally obtained using a long-coherence-length laser at around 800nm to exploit the weak absorption of tissue and enable light to propagate through thick tissues. A long-coherence-length laser is used so the light paths that travel long distances through the tissue still interfere with the short light paths. The tissue spatial sensitivity profile is identical to that of NIRS measurements; therefore, the partial volume effects of the overlying scalp and skull must be considered when estimating CBF. Diffuse correlation spectroscopy is similar to arterial spin labeling fMRI in that it provides a measure of blood flow. Near-infrared spectroscopy and DCS both hold the same advantage over fMRI: that of being portable and suitable for continuous bedside measurements.
Diffuse correlation spectroscopy has been crossvalidated extensively against other blood flow measures (Mesquita et al, 2011). It has been shown in acute stroke patients that cerebral perfusion varies with the elevation of the patient's head in the bed and that while optimal perfusion is usually achieved with the patient lying down, that some patients are optimized in a more elevated position. Diffuse correlation spectroscopy has also demonstrated a flow—volume uncoupling in newborn infants during the first 8 weeks of life that is a result of dramatic reductions in hematocrit as fetal hemoglobin is replaced with adult hemoglobin (Roche-Labarbe et al, 2010).
Two-photon microscopy and OCT, in addition to providing measures of RBC velocity and blood flow, are able to provide high-resolution angiograms of the microvasculature. Two-photon microscopy angiograms are commonly performed (Kleinfeld et al, 1998) (Figure 3B); OCT angiograms have recently appeared in the literature (Wang et al, 2007a) (Figure 3C). In contrast with TPM, OCT angiography does not require the administration of dyes or extrinsic contrast agents. Also, OCT angiography performs 3D imaging on time scales of minutes, whereas TPM requires time scales of hours to achieve comparable fields of view (Vakoc et al, 2009). While OCT is able to penetrate deeper than TPM can, the limit is slightly >1mm. Photoacoustic tomography (Figure 2B) offers the ability to obtain angiograms at depths of several millimeters (Hu and Wang, 2010). In all cases, greater depth penetration requires a craniotomy.
Using TPM angiograms, Nishimura et al (2007) have demonstrated that while pial arteries provide a mesh network of redundant blood flow, penetrating arterioles are bottlenecks of flow to deeper levels, such that occlusion of a penetrating arteriole will result in downstream ischemic damage. Optical coherence tomography is playing an important role in longitudinally quantifying angiogenesis following brain injury and investigating the effect of different agents on promoting angiogenesis (Jia et al, 2011).
Optical methods are well suited to measure oxygen delivery by blood, and through multimodal approaches, estimating oxygen consumption by tissue. Because of the autofluorescence of the coenzymes NADH and flavin adenine dinucleotide (FAD), it is possible to monitor cellular energetics directly. Fluorescent analogs of glucose have recently become available and these will likely enable more direct measures of glycolysis with optical resolution.
Phosphorescence lifetime imaging of oxygen (PLIO2) measures oxygen-dependent phosphorescence lifetimes of an exogenous contrast agent (Rumsey et al, 1988; Vanderkooi et al, 1987). It can be used to image both cerebral intravascular and tissue oxygenation using widespread optical imaging systems such as widefield charge-coupled device (CCD) imaging (Sakadzic et al, 2009) or different microscopy modalities (Sakadzic et al, 2010; Yaseen et al, 2009) (Table 1). Combining PLIO2 with TPM (Finikova et al, 2008; Sakadzic et al, 2010) enables measurement of cortical oxygen delivery with subcapillary resolution in tissue and deep microvasculature (Devor et al, 2011; Lecoq et al, 2011; Sakadzic et al, 2010) (Figures 4A and 4B).
The phosphorescence lifetime of a probe depends on the partial pressure of oxygen (pO2) in the immediate vicinity of the probe, providing a spatially localized measurement of dissolved oxygen. Probe molecules were specially designed for either linear or two-photon excitation regimes, with a high degree of encapsulation that ensures stability of lifetime calibration in a complex biological environment (Finikova et al, 2008; Lebedev et al, 2009). Unlike spectroscopy-based hemoglobin saturation measurements, PLIO2 lifetime imaging is insensitive to changes in tissue optical properties during imaging. The acquisition speed is currently limited to 0.2 to 1second per measurement point by relatively long phosphorescence lifetimes and the number of decay averages required at each point.
Through combination of PLIO2 with TPM, we recently demonstrated that the increase in blood oxygenation during the hemodynamic response, which has been perceived as a paradox, may serve to prevent a sustained oxygenation drop at tissue locations remote from the vascular feeding sources (Devor et al, 2011). We have also observed that a significant amount of oxygen is delivered to the tissue from the arteries, and that venous intravascular pO2, surprisingly, is higher than that in the capillary bed on average (Sakadzic et al, 2010).
As noted above, OISI, PAT, and NIRS are all able to measure hemoglobin oxygenation through absorption spectroscopy, while PLIO2 can measure pO2 in blood and tissue. Tissue oxygen consumption is commonly estimated based on the difference between oxygen flowing into a region through arteries and out through veins. Thus, combining optical measures of oxygen with a measure of flow, it is possible to estimate oxygen consumption. This has been performed by combining OISI with LSCI (Dunn et al, 2005), by combining PLIO2 with LSCI (Sakadzic et al, 2009) and Doppler OCT (Yaseen et al, 2011), by using PAT (Yao et al, 2011), and by using NIRS in combination with DCS (Durduran et al, 2010).
Nicotinamide adenine dinucleotide and FAD are important coenzymes for energy metabolism and both are intrinsically fluorescent (for recent reviews see Heikal, 2011; Shuttleworth, 2010; Turner et al, 2007). In vivo studies generally focus on measuring changes in NADH or FAD fluorescence (Table 1). Nicotinamide adenine dinucleotide fluorescent changes can serve as an indicator of the balance between glycolysis and oxidative phosphorylation rate changes. Flavin adenine dinucleotide increases indicate an increase in oxidative phosphorylation. Fluorescent changes can be imaged using single- and two-photon excitation (Harbig et al, 1976; Huang et al, 2002; Kasischke et al, 2010; Weber et al, 2004) (Figure 4C). However, measuring FAD in the brain can be challenging, because of its low two-photon absorption cross-section, possible low concentration, and emission overlap with lipofuscin fluorescence and other flavin proteins (Heikal, 2011).
Single-photon NADH and FAD imaging was used to map cortical activity (Husson et al, 2007; Reinert et al, 2007; Sirotin and Das, 2010) and demonstrate that an increase in metabolism occurs faster than an increase in CBF (Weber et al, 2004). Two-photon imaging of NADH in brain slices indicated that the ratio and time course of oxidative and nonoxidative metabolism can differ in neurons and astrocytes (Kasischke et al, 2004). Two-photon imaging of NADH also has been applied in vivo in healthy cerebral cortex (Kasischke et al, 2010) and during experimentally induced cortical spreading depression (Takano et al, 2007). These studies showed a close spatial association between NADH fluorescence and arteriolar geometry, implying that arterioles serve as oxygen sources, and demonstrated vulnerability of tissue midway between capillaries for hypoxia caused by large-scale pathological increases in neuronal activity.
Glucose metabolism is routinely assessed with 2DG autoradiography postmortem. FDG positron emission tomography enables in vivo measurements. Both of these require radioactive glucose accumulation over tens of minutes. Fluorescent glucose analogs offer the exciting ability to estimate glucose transport into individual cells in vivo (Chuquet et al, 2010), but still require accumulation of the glucose analog over minutes (Table 1). In addition, the interpretation of the data may be complicated due to the difference in kinetic parameters of individual glucose transporters with respect to glucose and fluorescent glucose analogs (Barros et al, 2009), as well as the existence of different glucose transporters in neuronal and glial cells (Simpson et al, 2007). Further advances in the design of novel glucose probes (Lee et al, 2011) or glucose fluorescent sensors (Pickup et al, 2005) are needed to address these challenges.
The ability to both observe and experimentally manipulate neuronal activity is a prerequisite for conducting successful mechanistic neurovascular/neurometabolic studies with unambiguously interpretable results. A change in neuronal activity is associated with multiple processes that can be measured optically, including changes in transmembrane voltage (depolarization or hyperpolarization), intracellular changes in ionic concentration (e.g., increases in [Ca2+]), release of neurotransmitters, and changes in pH (Table 1). None of these processes alone can be considered an absolute measure of neuronal activity. Rather, each reflects a particular aspect of neuronal activity and various of these aspects (e.g., voltage changes and the amount of released neurotransmitter) can be nonlinearly related.
Well-controlled experimental manipulation of neuronal activity on a cellular scale provides a powerful tool for understanding the associated metabolism and testing the role of specific cell types in control of vasodilation/vasoconstriction.
Voltage-sensitive probes or ‘dyes' (VSD) (Cohen and Salzberg, 1978; Grinvald and Hildesheim, 2004) reside in the plasma membrane of neurons and act as molecular transducers that transform changes in membrane potential into optical signals: absorption, emitted fluorescence, a shift in the spectrum of the dye, or a change in its second-harmonic generation properties (for a recent review see Peterka et al, 2011). Since dendritic arborizations constitute a large percentage of the total membrane area, VSD signals are sensitive to subthreshold neuronal activity.
Synthetic VSD bind to all plasma membranes; targeting to particular neuronal cell types is not feasible. To overcome this problem, approaches for genetically encoded and ‘hybrid' voltage sensors have been explored (Akemann et al, 2010; Siegel and Isacoff, 1997; Wang et al, 2010). Although cell-type specific expression of genetically encoded voltage sensors has not yet been demonstrated, it is an active area of research (Homma et al, 2009). Another challenge is development of voltage probes suitable for two-photon excitation (Kuhn et al, 2008).
Neuronal changes in membrane voltage occur on a millisecond timescale, introducing a strict requirement for the temporal response properties of the sensor. In common practice, voltage sensors are excited in the single-photon regime, and images are generally acquired using a camera detector with no depth resolution (but see Kuhn et al, 2008).
Due to sensitivity of VSD signals to subthreshold changes in neuronal polarization, VSD imaging is very useful for detection of neuronal inhibition. By implementing VSD imaging, we previously demonstrated the occurrence of CBF decrease and vasoconstriction in inhibited cortical regions (Devor et al, 2008). Within the realm of pathophysiology, VSD imaging provided a significant insight in stroke-induced neuronal reorganization, including short- and long-term sensory remapping (Brown et al, 2009; Sigler et al, 2009) and the recent finding of the immediate neuronal disinhibition in the unaffected hemisphere (Mohajerani et al, 2011).
Fluorescent ionic-sensitive indicators are widely employed as measurements of neuroglial activity. Among them, calcium indicators have become an important tool in brain research due to the importance of calcium in neuronal and astrocytic physiology and the availability of bright and sensitive acetoxymethyl ester derivatives (Tsien, 1981) that can be delivered in vivo and excited in the two-photon regime (for recent reviews see Garaschuk et al, 2006; Grewe and Helmchen, 2009; Kerr and Denk, 2008).
In contrast to voltage sensors, where usable genetically encoded variants are just starting to arrive, genetic calcium probes are widely used (Hires et al, 2008; Mank and Griesbeck, 2008). Intracellular calcium concentration changes on a slower time scale than transmembrane voltage. Therefore, relatively slow kinetics of genetically encoded sensors—ultimately limited by the rate of conformational change of proteins—is less of an issue in the design of calcium probes.
When imaged with TPM, calcium increases within individual neuronal cell bodies can be used to reconstruct spike trains (Vogelstein et al, 2010). Different types of neurons vary in their expression level of particular calcium channels, and multiple types of calcium channels exist (Tsien et al, 1995). Although cytosolic calcium can fluctuate within the subthreshold range of membrane potentials (Ross et al, 2005), calcium imaging is believed to reflect spiking and not subthreshold (e.g., synaptic) activity (Cossart et al, 2005).
The use of calcium indicators enabled the discovery of astrocytic excitability (Cornell-Bell et al, 1990; Nedergaard, 1994), and, in combination with TPM, has truly revolutionized both basic and applied neuroscience allowing visualization of microscopic cortical functional organization (Ohki et al, 2005), abnormal waves of astrocytic activity in Alzheimer's disease (Kuchibhotla et al, 2009), and functional rewiring after a stroke (Winship and Murphy, 2008), just to name a few.
Synaptic release of (potentially vasoactive) neurotransmitters and neuropeptides involves depolarization of the presynaptic terminal, calcium entry through voltage-gated channels located on the plasma membrane in the vicinity of docked and ready-to-release synaptic vesicles, fusion of the vesicles with the plasma membrane, and loss of the acidic intravesicular environment. The change of pH experienced by the luminal side of a vesicle upon fusion provides the foundation for pHluorin-based reporters of synaptic release (Miesenbock et al, 1998). These reporters are genetically encoded and have been used extensively in the olfactory bulb. Another approach is to target genetically encoded calcium indicators to the cytoplasmic side of vesicular membranes (Dreosti et al, 2009). Imaging the release of specific transmitters requires specific optical probes. One such probe, suitable for TPM, has recently been developed for imaging of extracellular glutamate (Okubo et al, 2010). Genetically encoded probes have been applied in the olfactory bulb to demonstrate that functional hyperemia was highly correlated with glutamate release but not with postsynaptic activity (Petzold et al, 2008).
Genetically encoded optical actuators for manipulation of neuronal activity in vivo have recently become available due to the development of methods based on expression of light-activated ionic channels and pumps of bacterial origin (Miesenbock, 2011) (Figure 5A). These methods, together with genetic methods for reporting of neuronal activity—such as the genetically encoded voltage and calcium probes discussed above—comprise a new field called ‘optogenetics.' Current optogenetic tools for optical control of neuronal activity offer both excitation and inhibition (Miesenbock, 2011; Yizhar et al, 2011).
Genetically encoded photoactuators, expressed in cortical pyramidal cells, were used to drive the hemodynamic response in a recent fMRI study (Lee et al, 2010), providing a prove-of-principle for the utility of optogenetics in neurovascular and neurometabolic research.
Another approach for photoactivation (with a longer history of applications) is based on the use of synthetic derivatives of transmitters and second messengers in which addition of a chemical bond makes them biologically inert (Adams and Tsien, 1993). Photolysis (‘uncaging') breaks a bond, liberating active properties of the compound. Caged compounds designed for two-photon photolysis allow targeted manipulations on a cellular and even subcellular level (e.g., mimicking a synaptic input to a single spine) (Figure 5B). The large majority of uncaging studies have been performed in brain slices and cell cultures. This is because the need to deliver extrinsic caged molecules presents a challenge for in vivo studies (Noguchi et al, 2011). Nevertheless, future improved delivery strategies and technical advances in two-photon photostimulation (Nikolenko et al, 2007) are likely to promote two-photon uncaging to a valuable manipulation method in microscopic studies of the neurovascular unit physiology in vivo.
The spectrum of optical technologies outlined above offers a comprehensive set of tools for measurement of a wide range of physiological parameters. How can we use these optical tools to make progress toward a mechanistic understanding of the regulation of flow and metabolism? First, let us consider the following specific Questions, significant for achieving this understanding, which can be tackled with the use of optical imaging tools:
Below are a number of approaches, which, in our view, will be essential in addressing these questions.
Availability of high-resolution optical microscopy tools will have a central role in addressing Questions 1 to 6 in the list above. In the context of Question 1, recent reports in brain slices have demonstrated that stimulation of neurons of different types produces specific responses in the embedded vascular segments: dilation or constriction (Cauli et al, 2004; Rancillac et al, 2006). Can these experiments be translated in vivo? In other words, can we identify the microscopic in vivo vascular ‘signature' of activation in neurons with known phenotype and neurotransmitter content? While one-photon excitation of photoactuators currently used in optogenetics affects a population of cells, eliciting spikes in a single cell can be achieved by two-photon uncaging as has been demonstrated in vitro (Fino et al, 2009) (Figure 5B). Two-photon photoactivation of channelrhodopsin-2 (Rickgauer and Tank, 2009) and delivery of a transgene to a single cell in vivo (Kitamura et al, 2008) have also been demonstrated.
Another potential strategy for evaluating vascular diameter changes induced by the firing of specific neurons in vivo is based on triggering the diameter measurements on spikes in a particular cell. Spike timing can be estimated from calcium imaging (Vogelstein et al, 2010). Identification of cell types in vivo can also be achieved through the use of genetically encoded fluorescent markers (Tsien, 2005), postmortem immunolabeling (Kerlin et al, 2010), or single-cell PCR if activation is achieved by a targeted whole-cell patch recording (Margrie et al, 2003). This strategy requires that firing in the neuron of interest does not temporally coincide with firing in other neighboring neurons to avoid release of additional types of vasoactive substances. Therefore, the analysis would be limited to spontaneous firing, rather than firing in response to stimulation, and use of either awake animals or anesthesia that does not induce neuronal synchronization.
Combining novel and improved optical imaging technologies with the recent revolutionary advances in optogenetics (Miesenbock, 2011; Yizhar et al, 2011) will allow in vivo measurement of vascular and metabolic consequences of controlled experimental manipulation of neuronal activity, critical for addressing Questions 2 and 3. Already today, a number of transgenic mouse lines with cell-type specific expression of light-gated ionic channels have been developed; some are already commercially available. Photoactivation of these and other genetically encoded optical actuators results in activation (or inhibition) of cells expressing the transgene. Therefore, when expression is specific to a particular cell type, it might be well suited to addressing the neuronal cell-type specificity of vascular regulation: which cell types induce vasodilation/constriction upon depolarization (firing), and which vasoactive neurotransmitters are released. To ensure specificity in such an experiment, one has to prevent propagation of activation to other neuronal cell types. For example, unless synaptic communication is inhibited, photoactivation of excitatory pyramidal neurons is instructive in elucidating connectivity between the brain regions (Lee et al, 2010) but might not provide any advantage for the study of neurovascular coupling over a sensory stimulation: in both cases firing of many neuronal cell types results in release of a mixture of neurotransmitters and peptides.
Optogenetic activation of specific modulatory (e.g., cholinergic) projections will have an important role in addressing Question 4. In this respect, transgenic mouse lines with targeted expression of photoactuator proteins in cholinergic and serotonergic neurons have been developed (Zhao et al, 2011) and are available from Jackson Laboratory (http://jaxmice.jax.org/).
Our ability to grasp a physiological process and macroscopic ‘observables,' relevant for noninvasive imaging in humans, critically depends on understanding the behavior of the underlying microscopic parameters. The most recent example illustrating the importance of specific, selective, and sensitive optical probes for high-resolution microscopic measurement is a new phosphorescent probe to measure pO2 (Finikova et al, 2008), which will be instrumental is addressing Question 5. This probe, applicable for two-photon imaging, has already been utilized to study intravascular and extravascular oxygenation at baseline and during neuronal activation (Devor et al, 2011; Lecoq et al, 2011; Sakadzic et al, 2010). Other examples include a two-photon excitable probe that has also been recently developed for imaging of extracellular glutamate (Okubo et al, 2010) and novel cell-based fluorescent reporters for detection of acetylcholine release (Nguyen et al, 2010). Further progress in design of existing optical reporters (glucose) (Lee et al, 2011) and future development of new optical sensors—among them the ones for lactate and adenosine, important for addressing Question 6—will open unprecedented opportunities to visualize directly neurovascular and neurometabolic processes.
In the context of cortical function, even the most specific, sensitive and high-resolution measurement on its own provides only a descriptive view of a single aspect of the underlying multifaceted physiological processes. This being the case, integration of two or more imaging technologies, each sensitive to a different aspect of the physiological process under study (the ‘multimodal' imaging approach) will allow simultaneous measurements of multiple relevant physiological and biophysical parameters and computational inference of processes that cannot be measured directly. For example, a combination of PLIO2 (Sakadzic et al, 2010) with OCT (Srinivasan et al, 2011) may provide a microscopic measure of oxygen consumption. The multimodal principle is of course not limited to optical technologies—one can combine optical imaging with, for example, electrophysiological recordings (Berwick et al, 2008; Sheth et al, 2004) or MRI measurements (Kennerley et al, 2005).
Optical technological innovation continually strives to improve spatial resolution, image acquisition rate, and depth penetration. We are now seeing tremendous advances on these fronts for scanning microscopies. The heterogeneous structure of the tissue distorts the optical wave front, degrading the spatial resolution from the diffraction limit. Adaptive optics is developing as a robust approach to counterbalance the tissue distortions and restore diffraction-limited resolution for TPM (Ji et al, 2010) and OCT.
The image acquisition rate is dictated by the raster scanning rate divided by the image volume. Scanning rate can be increased by using acousto-optic deflectors (Grewe et al, 2010). However, one must consider the implementation of acousto-optic deflectors so as not to degrade spatial resolution (Kirkby et al, 2010). Two-photon microscopy imaging in 3D is slow because of the need to physically move the objective, although piezo-electric resonators now enable rapid z-translation (Gobel et al, 2007). Faster z-scanning is being demonstrated by using adaptive optic strategies to dynamically adjust the focal depth of an objective with no moving parts (Grewe et al, 2011). Scanning rate is ultimately limited by the number of photons detected. Photon collection efficiency can be improved using novel approaches to increasing the effective numerical aperture of the microscope collection optics (Engelbrecht et al, 2009). More efficient excitation of fluorophores can be achieved by increasing the repetition of the pulsed laser sources used in TPM (Ji et al, 2008; Li et al, 2010). Larger image volumes can be scanned faster by essentially dividing the larger volume into smaller subvolumes that are imaged in parallel (Bewersdorf et al, 1998).
Tissue scattering is wavelength dependent and limits depth penetration. Reducing tissue scatter would therefore enable greater penetration. While tissue-clearing methods can be used ex vivo, we can use longer excitation wavelengths in vivo where the optical scattering is reduced. The penetration depth has been shown to increase from ~600 to 1,000μm when moving from 800 to 1,300nm excitation with TPM (Kobat et al, 2009). Similar advantages have been shown for OCT. The challenge is to ensure that the fluorophore excitation extends out to these longer wavelengths. Even greater penetration depths are being achieved with TPM and OCT by using GRIN lenses and microprisms to invasively provide an optical window to deeper tissue structures (Barretto et al, 2011).
Several approaches not detailed here are worth mentioning because of their potential impact when applied in vivo. Intensity-based ionic imaging cannot be used to quantify absolute concentration of ions because the intensity depends on the ionic concentration and dye concentration. This limitation can be overcome by measuring fluorescent lifetime, as has been demonstrated for calcium (Wilms et al, 2006). Specifically, fluorescent lifetime provides an absolute measure when the ionic concentration alters the probe conformation, resulting in a change in lifetime that is not dependent on probe concentration. Fluorescent lifetime can also be used to distinguish different conformations of fluorescent molecules, such as bound and free NADH (Vishwasrao et al, 2005).
The fundamental diffraction limit of optical imaging has recently been shattered by a variety of clever solutions that achieved resolution of better than 20nm (Hell, 2009; Huang, 2010). These new nanoscopic methods reveal fine-scale subcellular structures in culture, and have the potential to be applied in vivo with depth penetrations likely comparable to confocal microscopy of 50 to 100μm.
We have described advances of optical imaging in obtaining microscopic measures of cellular and vascular functioning. For the foreseeable future, though, noninvasive human neuroimaging will only be able to access the collective behavior of large groups of cells and vessels. Thus, translation of our microscopic mechanistic understanding of neurovascular and neurometabolic coupling will depend on our understanding of how this microscopic behavior is reflected in macroscopic observables (Question 7). Because of their overlapping spatial and temporal resolutions, optical technologies enable us to perform experiments that step incrementally from the cell/capillary level to the few cell/capillary level and all the way to noninvasive imaging of cubic centimeters of tissue relevant to human studies. This transition can be made in an animal from super-resolution microscopy to NIRS/DCS to characterize the micro–macro relationship. Near-infrared spectroscopy/DCS, combined with noninvasive electrophysiological recordings, can then be used in humans to verify the physiological findings, and finally put into clinical utility. As an example, NIRS has been used in rodents to test whether neurovascular coupling is driven by cortico-cortical connections or thalamic inputs in the somatosensory cortex (Franceschini et al, 2010) and then translated to humans to confirm the findings (Ou et al, 2009). Confirming that the vascular response is driven by cortico-cortical processing rather than by thalamic inputs would have profound clinical importance because vascular-based neuroimaging methods (fMRI, NIRS, and DCS) could then be used to assess the integrity of the sensory system at the level of cortico-cortical circuits in injury (stroke and trauma) and disease, and when those systems become active in development or turn off due to aging.
The arsenal of available optical imaging technologies offers the ability to measure a spectrum of parameters related to vascular, metabolic, and neuronal activity at multiple scales. Merging these technologies with recent revolutionary methods in genetic labeling and remote control of neuronal activity, allowing targeted activation of identified neuronal cells and cellular populations, is going to be a hallmark of cerebrovascular research in the next decade, or maybe even the next ‘50 years of dedicated CBF and metabolism' research. In parallel, findings from animal imaging will be translated to human studies through application of the noninvasive optical methods NIRS, DCS, and PAT. Although optical technologies are already broad and versatile, their performance is going to improve. In combination with continuous efforts in the development of novel optical sensors, and in expanding the array of transgenic animals with genetically encoded structural and functional fluorescent labels, the future of the brain imaging division of biomedical optics is as bright and exciting as ever before.
The authors declare no conflict of interest.
The authors gratefully acknowledge support from the National Institute of Health: NS051188, NS057198, NS057476, NS055104, EB00790, EB009118, EB007279, and K99NS067050; American Heart Association: 11SDG7600037 and 11IRG5440002; and the Glaucoma Research Foundation.