Review articleOn the nature of the BOLD fMRI contrast mechanism
Introduction
Our view of brain function has improved impressively in recent years as a result of intense efforts to understand the neural mechanisms underlying perception in humans and nonhuman primates. A large body of evidence regarding the processes through which sensory information at the biochemical, electrophysiological and systems levels contributes to the conscious experience of a stimulus has accrued. Our efforts to understand the organization and function of the sensory and perceptual systems have been greatly aided by the development of new techniques including novel and powerful methods of molecular biology, the refinement of recordings from single and multiple cells for short or long periods and noninvasive neuroimaging techniques allowing us to localize and study activity within the human brain while subjects perform a variety of cognitive tasks.
The contribution of neuroimaging cannot possibly be overemphasized. All our mental capacities, ranging from sensory representation and perception to reasoning and planning, rely on distributed, synergistic activities of large neural populations; therefore, understanding these not only requires a comprehension of the physiological workings of individual neurons and glia cells but also demands a detailed map of the brain's functional architecture, a description of the connections between populations of neurons and insights into the operations performed by the neural networks involved in the task at hand.
The present review deals with spatiotemporally resolved functional magnetic resonance imaging (fMRI) in monkeys and its combination with other invasive neuroscientific techniques. Emphasis will be placed on simultaneous imaging and electrophysiology experiments aiming to elucidate the neural basis of the blood oxygen level-dependent (BOLD) signal. We first review the basic organizational principles of the cortical system. Although many examples are drawn from the visual system, generality is hardly sacrificed as evidence over the last decades suggest a similar organization in any other sensory system studied. Our intention is not to provide the reader with an exhaustive review on sensory or perceptual processing; instead, we summarize examples of current work showing the increasing need of neuroimaging and integrative approaches in addressing many of the interesting questions raised by systems neuroscience. A brief discussion on the principles of energy metabolism and neurovascular coupling follows, subsequent to which the neural and BOLD signals are examined in some detail.
Sensory systems are characterized by topographical organization in the sense that receptors in the peripheral receptive sheet project to central neurons in an orderly manner preserving neighborhood relationships. Within each modality, there are multiple such maps representing the sensory surface. Examples of maps (considering columns as spatial units) are the primary cortical areas of the visual, auditory and somatosensory modalities. Topographical maps, among others, imply local processing. In the case of the visual system, images from the retina at the back of each eye are channeled first to a subdivision of the cerebral hemispheres deep in the brain called the lateral geniculate nucleus (LGN) and, from then on, to the primary visual cortex. In the primary visual cortex (V1), neurons were thought to analyze local spatial information within their small receptive fields (RF), basically ignoring processing carried out elsewhere.
It is notable that in the past few years this concept was radically modified by studies showing that V1 cells can actually integrate information over a much larger part of visual space than originally believed and may be an important part of the network underlying perceptual organization. Because their responses are not solely determined by the optimal stimulus with their classical RF [1], they can rather be considerably modulated by perceived brightness, perceptual “pop-out” or grouping of line segments, figure-ground segregation based on a variety of visual cues [2], [3], [4] and the attentional requirement of the task (for a review, see Ref. [5]).
What was previously thought of as a static RF with fixed physiological properties is now conceived as a context-dependent dynamic entity that may undergo shifts in preferred orientation [6], position or size [7] or may modulate activity as a function of viewing distance [8]. Modulation of contextual effects, possibly through the feedback connections, may also be the mechanism by which attention affects the response of V1 cells [9]. Many of these effects are, at least in part, attributable to horizontal or feedback connections. Disentangling the relative contributions of feedforward, lateral and feedback connections is of obvious importance for understanding both sensory and perceptual processing and is likely to instigate a great deal of future research; it is also likely to require an increasingly greater involvement of neuroimaging in a typical neuroscience laboratory.
The influence of context on cell responsiveness is one prominent example of dynamic neural behavior; plasticity of maps, another. Neurons in various cortical sensory areas, even in the earliest ones, are far more flexible and mutable in their functional properties than previously thought. Maps are not fixed but are instead continuously modified by experience [10], [11] and are likely involved in perceptual learning [10], [11], [12].
Finally, modular organization is a principle shared by most sensory systems. In the visual system, modularity was already well established in the 1970s with the discovery of multiple areas that are functionally specialized to undertake different tasks and have certain hierarchical relationships with each other (for a review, see Refs. [13], [14]). From a physiological viewpoint, the notion of hierarchical organization emerged from the pioneering experiments of David Hubel and Torsten Wiesel on the structure of RFs in visual cortex. Evidence from anatomy came through the observation that different types of corticocortical connections [15], initially called rostral and caudal and subsequently termed feedforward and feedback, can be distinguished by their neurons of origin as well as by the highly specific laminar distribution of their efferent terminal arborizations.
Further detailed physiological and anatomical studies yielded an elaborate route map describing the functional organization of the visual system [16], [17]. In addition, an influential proposal that two anatomically distinct and functionally specialized cortical streams of visual processing emanating from the primary visual cortex exist — a dorsal, occipitoparietal stream stretching through the visual association areas of the parietal lobe, which processes spatial information, and a ventral, occipitotemporal stream through the visual areas of the temporal lobe, which is involved in the representation of visual objects — was made based on anatomical, physiological and lesion studies [18], [19].
Recently, a series of long-awaited tract-tracing, electrophysiological and lesion experiments in monkeys as well as imaging experiments in humans revealed a similar hierarchical organization in the primate auditory system (see, for a review, Ref. [20]). The reported similarity in the organization of the visual and auditory systems is in good agreement with the parallel nature of the visual and auditory perceptual requirements; namely, the localization and identification of patterns. While the former system integrates information across space and time, the latter does so over frequency and time dimensions.
Perceptual organization, selective or attentive information processing, decision making and categorization are only a few of the cognitive capacities intensively investigated in today's neuroscience, which seeks to understand the relationship between the mind and the brain. All these capacities rely on distributed, synergistic activities of large neural populations; therefore, understanding these not only requires a comprehension of the physiological workings of individual neurons and glia cells but also demands a detailed map of the brain's functional architecture, a description of the connections between populations of neurons and insights into the operations performed by the neural networks involved in the task at hand.
Yet, until very recently, systems neuroscience almost exclusively has been relying on physiological studies utilizing the much celebrated single microelectrode technique, reporting the number of action potentials produced by an isolated neuron within a unit of time. Although it proved very useful in characterizing the response properties of different structures, the method clearly falls short of providing information on spatiotemporal cooperativeness and on the global, associational operations performed by neural networks.
Understanding perception or cognition will ultimately depend on the development and application of integrative approaches. Single cell recordings, large electrode or tetrode-array recordings, monitoring of action potentials and slow waves must be employed in combination with neuroimaging using calibrated BOLD signals, cerebral blood flow (CBF), volume (CBV) and MR spectroscopy (MRS) of cerebral metabolites and neurotransmitters to obtain the information required for studying the brain's capacity to generate various behaviors. The recent development of high-field MRI and functional CBF imaging as well as MR spectroscopic imaging [chemical shift imaging (CSI)] in our laboratory [see papers in this volume by Pfeuffer et al. [21], [22] (MRI and MRS at 7 T) and Juchem et al. [23] (CSI at 7 T)] directly reflects our conviction that such integrative approaches can and must be applied in systems neuroscience.
Integrative approaches, however, require the interdisciplinary education of researchers and a thorough understanding of, at least, the basics of closely associated research fields. Invasive neuroimaging in a typical neuroscience laboratory, for instance, requires acquaintance — to a certain extent — with the organization of the brain's metabolism and vascular system in addition to fathoming into the workings of the nerve and glia cells. The next paragraphs attempt to briefly review the essentials of these fields, introducing the commonly measured neural signals and continuing with the very basics of metabolism and hemodynamics, to tap the issue of structural and functional neurovascular coupling.
Section snippets
The compound neural signal
The signal measured by an electrode placed at a neural site represents the mean extracellular field potential (mEFP) from the weighted sum of all sinks and sources along multiple cells (Fig. 1). If a microelectrode with a small tip is placed close to the soma or axon of a neuron, then the measured mEFP directly reports the spike traffic of that neuron and frequently that of its immediate neighbors as well. Recent studies in rats, for instance, show that tetrodes placed close (within 50 μm) to
Neural and hemodynamic responses
Functional neuroimaging techniques are divided into two fundamentally different approaches: (a) electromagnetic approaches including EEG and magnetoencephalography, both providing a high temporal resolution but poor spatial information, and (b) hemodynamic–metabolic approaches based on the fact that neuronal activity is coupled to energy metabolism and the subsequent changes in CBF and volume. The functional MRI techniques discussed in this paper rely on hemodynamic–metabolic changes during
Magnetic resonance imaging of the monkey brain
The application of neuroimaging for both noninvasive and invasive research in a nonhuman primate requires high spatiotemporal resolution and good SNR. Only then can the fMRI signals be compared with signals obtained from optical or electrical recordings. To meet these requirements, dedicated vertical, large-bore monkey MR systems were used at high magnetic field (4,7 T) in combination with extensive optimization of both the hardware (gradient performance, RF coils) and the acquisition
Connectivity studies with paramagnetic tracers
Neuroanatomical corticocortical and corticosubcortical connections have been examined mainly by means of degeneration methods and anterograde and retrograde tracer techniques (e.g., Refs. [118], [119]). Although such studies have demonstrated the value of the information gained from the investigation of the topographic connections between different brain areas, they do require fixed, processed tissue for data analysis and therefore cannot be applied to an animal participating in longitudinal
Conclusions
The suitability of MRI for functional brain mapping is firmly established. BOLD fMRI has been successfully implemented in awake human subjects as well as in animals such as rats, cats and monkeys. The use of high magnetic fields increases functional signal changes and improves both signal specificity and spatial resolution. MRI studies, in which small voxels of microliter volumes may contain as few as 600–800 cortical neurons, can help us understand how neural networks are organized and how
References (125)
- et al.
Feedforward, horizontal, and feedback processing in the visual cortex
Curr. Opin. Neurobiol.
(1998) Plasticity in visual perception and physiology
Curr. Opin. Neurobiol.
(1996)- et al.
Laminar origins and terminations of cortical connections of the occipital lobe in the rhesus monkey
Brain Res.
(1979) - et al.
Anatomical and functional MR imaging in the macaque monkey using a vertical large-bore 7 Tesla setup
Magn. Reson. Imaging
(2004) - et al.
High-field localized 1H NMR spectroscopy in the anaesthetized and in the awake monkey
Magn. Reson. Imaging
(2004) - et al.
Region and volume dependencies in spectral linewidth assessed by 1H 2D MR chemical shift imaging in the monkey brain at 7T
Magn. Reson. Imaging
(2004) - et al.
Extracellular microelectrode sampling bias
Exp. Neurol.
(1970) Electrophysiology of a dendritic neuron
Biophys. J.
(1962)- et al.
Averaged multiple unit activity as an estimate of phasic changes in local neuronal activity: effects of volume-conducted potentials
J. Neurosci. Methods
(1980) - et al.
Tetrodes markedly improve the reliability and yield of multiple single-unit isolation from multi-unit recordings in cat striate cortex
J. Neurosci. Methods
(1995)