Neuronal Recording Studies Of Perception And Cognition Research Paper

Single and multiple neuron recordings are particularly powerful when applied to experiments in which the brain must form some neural ‘calculation’ in order to make the appropriate response. This can be manifested on many levels, from the purely mechanical, such as integration of multiple inputs in a simple neuronal circuit, to complex behavioral responses dependent on an animal’s ability to ‘re- member’ prior stimuli. Inasmuch as the recorded neural activity reveals a pattern that correlates to the cognitive process, the neural activity can therefore be considered to ‘encode’ information related to the process. This does not necessarily mean that the neural pattern is essential to cognition, but that the neural activity suggests information, which is available to the brain during cognition.

1. Single Vs. Multiple Neuron Recording

Single neuron recording has revealed the function of many brain areas in terms of stimulus and response. For example, Hubel and Wiesel (1962) mapped columns of neurons in the occipital cortex that responded to specific visual stimuli. The identification of these ‘ocular dominance columns’—in which successive columns responded to incremental rotations of a bar of light, and alternating regions mapped the left vs. right visual fields—were critical to understanding how visual information was encoded by the brain. Similarly, the identification of ‘place cells’ in hippocampus—neurons that fire only when the subject is in fairly limited regions of its environment (O’Keefe and Dostrovsky 1971), provided a correlation between neural activity and behavior, even if the exact purpose of that activity is still being debated. Single neuron recording has thus been used to identify the type of information processed by different brain areas, not just the five senses, but also control of muscles, attention, and memory. To this end, the brain has been mapped anatomically and, to a certain extent, neurophysiologically.

Broader questions of function and cognitive processing are difficult to answer with only single neuron recordings. In essence, all neurophysiological experiments use multiple neuron recording, since the activity of a single neuron is not necessarily representative of that brain area. Instead, many single neurons are recorded, then analyzed to see if a pattern is detected which is representative of neural activity in that region. As long as single neurons are recorded under conditions that control variability between recording sessions, it is possible (within limits) to treat the collection of neural recordings as if they had been recorded simultaneously. Thus, single neurons are assembled in an ‘ensemble’ that represents the activity of multiple neurons from a single brain area. In this manner, Georgopoulos et al. (1986) observed that the activity of ensembles of neurons recorded from primate motor cortex predicted the direction of limb movement prior to actual movement. In this study, many neurons were recorded that exhibited different ‘preferred’ directions for limb movement. When the pattern of firing across neurons was analyzed, it was found that the direction of movement was represented by specific firing patterns prior to the actual movement. However, since the conditions under which two single neurons are recorded cannot be identical, the effect of those subtle variations on neural firing cannot be controlled. In addition, the activity of single neurons may be controlled by multiple stimuli, in which case a behavioral correlate may be shared or distributed across many neurons.

Recordings of multiple simultaneous neurons have played an important role in many exciting recent discoveries. Deadwyler and Hampson (1997) showed that patterns of neural activity in hippocampus predicted behavioral error in a short-term memory task. Laubach et al. (2000) similarly demonstrated that activity patterns in neocortical neurons predicted behavior as animals learned a motor task. Skaggs and McNaughton (1996) showed that during sleep, hippocampal ensembles ‘played back’ temporal firing patterns between neurons. The replayed patterns were identical to patterns generated by the animal’s exploration of its environment during the waking state. In addition, Chapin et al. (1999) recently demonstrated direct neural control of a robotic arm by neural signals recorded from rats. The researchers detected patterns of neural activity in motor cortex that preceded limb movements required to operate the robotic arm. By programming a computer to detect the pattern in real-time, the computer was able to operate the arm without requiring the animal to move its limbs. These recent results indicate rapid progress toward goals of better understanding the relationship between the activity of neurons and the cognitive functions of the brain.

2. Recording Many Neurons

Multiple neuron recordings not only rapidly increase the total number of neurons characterized, but also allow examination of interactions between neurons that may be critical to cognitive processing. Information may simply be encoded as the mean amplitude of firing across neurons at a given point in time, or as correlation between neurons (von der Malsburg 1994), or as part of coherent oscillations between neurons (Singer and Gray 1995). No matter how the conditions are controlled during single neuron recording, interactions between neurons cannot be determined unless neurons are recorded simultaneously. The major advantage, therefore, of recording from multiple neurons in any brain area, is the potential for deciphering the codes that a brain area generates during cognitive or behavioral responding. The techniques used to record and analyze the multineuron data are critical, and not all types of multiple neuron analyses are appropriate in all cases. Misapplication of an ensemble technique can at best misrepresent the data, or at the worst lead to unworkable models of brain action. Several common uses of multiple neuron or ensemble analyses are examined below.

2.1 Techniques For Multineuron Recording

The ionic currents and action potentials generated by neurons produce changes in the electrical potential of the environment surrounding the neuron. These extracellular potentials are volume-conducted through the electrically-conductive extracellular fluid, and can be detected by electrodes that are either in contact with tissue or the cerebrospinal fluid. Skull surface or brain surface electrodes detect summed electrical activity resulting from the action potentials produced by millions of neurons throughout the brain, with no differentiation into single neurons. A glass pipette electrode will record neural activity from a region hundreds of microns in diameter; however, as above, the recorded neural activity is difficult to resolve into single neuron action potentials. Isolation of single neuron activity is possible with metal microelectrodes made from electrically conductive, but nonreactive material, such as tungsten, platinum-iridium, or stainless steel. Such electrodes are constructed with tip diameters approximate to the soma diameter of the neurons to be recorded and are insulated everywhere except at that tip. Sharp electrode tips (1 µm) allow the electrode to be inserted with minimal tissue damage and allow the tip to be insinuated close to one neuron. In addition, the high impedance (resistance) of the tip ( > 1 MΣ) ensures that electrical potentials from more distal neurons are attenuated compared to potentials generated by proximal neurons. More recent developments have shown that stainless-steel wires with tips up to 25 µm and tip impedances of 100 kΣ are capable of isolating single neurons if left implanted for days or weeks (McNaughton et al. 1983, Deadwyler and Hampson 1997). Single neuron action potentials are identified by biphasic waveform (typically ± 10– 100 µV amplitude, 500–1000 µs duration). As long as the electrode tip is not moving relative to the neuron being recorded, each action potential generated by a single neuron should be the same. Therefore, the activity of a single neuron is isolated by identifying and recording all action potential ‘spikes’ with the same characteristic duration and amplitude. Multiple single neurons can be identified from a single microelectrode by further identifying similar spikes from all those rejected, or by using multiple microelectrodes.

2.2 Types Of Multiple Neuron Recording Electrodes

The simplest multiple electrode is a bundle of metal microelectrodes, of similar size and shape, with tips positioned to record from within a restricted area. These electrodes can contact each other without compromising recording ability, since they are insulated everywhere except at the tip. The advantage of microelectrode bundles is the large number of recording sites in a small area, the disadvantage is that there is no means of identifying where any one electrode tip is located. When electrode location is desired, arrays of microelectrodes can be used. The electrode tips are spaced in a pattern, so that each tip records from a separate, identified site. Another form of bundle electrode is the stereotrode or tetrode consisting of two or four wires twisted to keep all electrode tips in close proximity. The close proximity of recording tips on stereotrodes and tetrodes allows the recording of neurons on more than one wire, improving the ability to isolate single neurons. Finally, as stereotrodes and tetrodes can be constructed as arrays, to provide all of the benefits of both electrode types.

2.3 Discrimination And Identification Of Multiple Neurons

As stated above, the essential principle of single neuron isolation and identification is the fact that the action potential from a given neuron should always be the same amplitude and duration. Thus, action potentials with the same waveform characteristics, likely were generated by the same neuron. The first technique for identifying single neurons recorded from microelectrodes is to sort the extracellular action potentials by waveform. Under some circumstances, notably when recorded from dense monolayers of neurons, two neurons may be equidistant from the electrode tip, and hence produce action potentials of similar sizes and shapes. Stereotrodes and tetrodes were developed to identify single neurons under these conditions (McNaughton et al. 1983), since it is unlikely that both neurons would be equidistant from both electrodes in a stereotrode, or from all four electrodes in a tetrode. By comparing the action potential waveforms simultaneously recorded at all electrode tips, it is possible to ‘triangulate’ and separate neurons that may have appeared the same to a single microelectrode.

Statistical techniques are also used to ensure that the action potentials recorded on single or multiple microelectrodes originate from a single cell. The intervals between action potentials of a randomly firing single neuron (the neuron’s ‘spike train’) should be a Poisson function. A plot of the inters pike-interval (i.e., between action potentials) should be similar to a graph of the Poisson probability distribution. In addition, autocorrelation of the spike train should reveal a symmetric distribution, with no firing within the time interval equal to the action potential refractory period for that neuron. Uniform interspikeinterval distributions, or spikes occurring during the refractory period indicate a failure to isolate recordings into single neuron data (see Nicolelis 1999). These analyses, plus consistency in behavioral correlation, cross-correlation between neurons, and stability of action potential waveforms over an extended time (or multiple days) all indicate satisfactory identification of the single neuron (whether recorded with single wires, stereotrodes or tetrodes) and thus support the recording of multiple single neurons from bundled or arrayed microelectrodes.

3. Analysis Of Multiple Neuron Recordings

The basic method of multiple neuron recordings is to analyze each neuron’s spike train independently. This typically involves constructing a plot of average neural activity correlated to a specific cognitive event. For instance, perievent histograms depict mean firing rate synchronized to (peri-) a behavioral stimulus (event); similarly, a ‘place field’ recorded for hippocampal neurons consists of the correlation of neural firing rate with the animals instantaneous location in it’s environment. Multiple neuron activity is then used to examine to overall average or range of responses for neurons in a given brain area. More complex analyses of multiple spike trains were first described by Gerstein and Perkel (1969) prior to development of many of the above multiple neuron recording techniques. These techniques concentrated on determining cross-correlations between neurons, and statistically extracting repeated patterns of activity across many neurons with a view toward determining the connections between neurons. Many later analyses derived from these earlier techniques.

3.1 Population Vectors

A popular type of analysis is a direct examination of ensemble firing rate patterns corresponding to specific motor movements or behavioral events. This analysis has been termed the ‘population vector’ and is potentially quite powerful, especially in terms of encoding directional and velocity parameters. Population vectors consist of a set of measurements of the mean firing rates for each neuron in the ensemble corresponding to discrete times when unique events, or stimuli occur. The utility of a population vector analysis depends on identifying discrete conditions or ‘states’ (such as tracking visual images, or degrees of rotation of a monkey forearm), which correspond to a unique set of mean firing rates across the ensemble for each state. As applied by Georgopoulos et al. (1986), the mean firing of each neuron in an ensemble recorded in non-human primate motor cortex was computed in correlation to forearm movements. A set of population vectors was obtained which corresponded to specific angles of arm rotation, then successfully compared to new recordings to test the accuracy of the population vector which ‘predicted’ the forearm angle prior to movement.

3.2 Cross-Correlations

Multiple neuron recordings can also be analyzed simply by looking for repeating patterns of firing across populations of recorded neurons. These patterns may consist of spatial combinations of two or more neurons within the ensemble, a temporal pattern within single neurons, a particular sequence of firing between several neurons, or combinations of spatial and temporal oscillations. Ensemble spike trains can therefore be scrutinized using cross-correlation techniques and the construction of cross-correlation histograms (CCHs; Gerstein and Perkel 1969) for this purpose. Detection of specific spatiotemporal firing patterns (i.e., correlations between neurons) may suggest ways in which those neurons may be connected, or may simply result from common ‘driving’ influences on several different cells. The discovery that hippocampal neurons ‘replay’ recent experiences during sleep (Skaggs and McNaughton 1996) relied on the observation that temporal offsets in CCHs between pairs of hippocampal place cells corresponded to the sequence in which those place cells were activated. During the waking state, the temporal offset reflected the sequential movement of an animal through different places in the environment. During the following sleep cycle, the hippocampal neural pairs were again active with the same temporal offsets, suggesting a memory of the exploring the environment. One drawback in cross-correlation studies, however, is that while it is appealing to use cross-correlations to map temporal connectivity and patterned firing between neurons within an ensemble, such procedures are not well suited to circumstances where nonstationary processes (i.e., behavior) are likely to alter temporal firing patterns while the data is being recorded (see Hampson and Deadwyler 1999). On the other hand, the demonstration that cross-correlations between hippocampal neurons were altered provided important confirmation of the perturbations in place cell firing and spatial navigation observed in transgenic mice with deficient NMDA-receptors on CA1 hippocampal neurons (McHugh et al. 1996).

3.3 Multivariate Analyses

More elaborate analyses of ensemble spike trains include multivariate statistical treatments such as linear discriminant analysis and principal components analysis (Nicolelis 1999). The differences and advantages of these analyses compared to those cited above are many. For instance, multivariate analyses simultaneously take account of the activity of all neurons in the ensemble, compared to cross-correlations which operate on pairs of neurons. Multivariate analyses incorporate both spatial (i.e., relationship between neurons within the ensemble) and temporal (time before and after events) information in the analyses, unlike population vectors which incorporate the firing of all neurons, but within an ‘instantaneous’ average of ensemble activity at discrete points in time. Finally, instead of producing one measure of correlation with behavioral and cognitive events for a given neuron, the multivariate components derived independent sources of variance within and between neurons comprising an ensemble.

Multivariate analyses provide the greatest potential for detecting neural patterns that correlate with perceptual, behavioral or cognitive event. Recordings of inferior temporal cortex (Miller et al. 1993), revealed many different patterns of encoding, including stimulus-specific and task (i.e., memory)-specific neural patterns. Likewise, analysis of ensembles of neurons from hippocampal (Deadwyler and Hampson 1997), parahippocampal regions (Young et al. 1997), and orbitofrontal cortex (Schoenbaum and Eichenbaum 1995) have demonstrated a broad spectrum of encoding that encompasses. Stimulus representation, memory storage, and formation of decision appropriate to the behavioral task.

4. Applications

Perhaps the best example of the utility of multiple neuron recording and analysis described above is the control of a robotic arm cited earlier (Chapin et al. 1999). In the study, ensembles of motor cortex neurons were recorded while a rat pressed a lever to cause a robotic arm to deliver water. A search for patterns of ensemble firing corresponding to the behavioral response would have required a massive search algorithm and intensive computation (as originally described Gerstein and Perkel 1969). However, principal components analysis of the ensemble spike trains revealed a pattern of neural firing that closely correlated to (and preceded) the limb movements required to press the lever and move the arm. Detection of the ensemble firing pattern while the animal performed the task allowed a computer to control the robotic arm, eventually allowing the rat to obtain water on the basis of the multiple neuron activity, without the requisite limb movement. Hence the principal components analysis allowed derivation of the brain function (behavioral response) from multiple neuron recording.

Single and multiple neuron recording have been applied to many purposes. Examination of the firing characteristics of single neurons provides a means of correlating the activity of each neuron with the behavioral or cognitive activity of the animal. Multiple neuron recording is used simply to increase the yield of single neuron experiments, or for a more in-depth approach to correlating brain activity with external events. Analyses of the connectivity between neurons provides a framework for modeling neural circuits. Finally, extracting patterns of neural activity correlating to specific behavioral responses can result in a more detailed understanding of the role of specific brain areas in the process of cognition.

Bibliography:

1. Chapin J K, Moxon K A, Markowitz R S, Nicolelis M A 1999 Real-time control of a robot arm using simultaneously recorded neurons in the motor cortex. Nature Neuroscience 2: 664–70
2. Deadwyler S A, Hampson R E 1997 The significance of neural ensemble codes during behavior and cognition. In: Cowan W M, Shooter E M, Stevens C F, Thompson R F (eds.) Annual Review of Neuroscience. Annual Reviews Inc., Palo Alto, CA, Vol. 20, pp. 217–44
3. Georgopoulos A P, Schwartz A B, Kettner R E 1986 Neuronal population encoding of movement direction. Science 233: 1416–19
4. Gerstein G L, Perkel D H 1969 Simultaneously recorded trains of action potentials: analysis and functional interpretation. Science 164: 828–30
5. Hampson R E, Deadwyler S A 1999 Pitfalls and problems in the analysis of neuronal ensemble recordings during performance of a behavioral task. In: Nicolelis M (ed.) Methods for Simultaneous Neuronal Ensemble Recordings. Academic Press, New York, pp. 229–48
6. Hubel D H, Wiesel T N 1962 Receptive fields, binocular interaction and functional architecture in the cat’s visual cortex. Journal of Physiology 160: 106–54
7. Laubach M, Wessberg J, Nicolelis M A 2000 Cortical ensemble activity increasingly predicts behaviour outcomes during learning of a motor task. Nature 405: 567–71
8. McHugh T J, Blum K I, Tsien J Z, Tonegawa S, Wilson M A 1996 Impaired hippocampal representation of space in CA 1 specific NMDAR1 knockout mice. Cell 87: 1339–49
9. McNaughton B L, O’Keefe J, Barnes C A 1983 The stereotrode: A new technique for simultaneous isolation of several single units in the central nervous system from multiple unit records. Journal of Neuroscience Methods 8: 391–7
10. Miller E K, Li L, Desimone R 1993 Activity of neurons in anterior inferior temporal cortex during a short-term memory task. Journal of Neuroscience 13: 1460–78
11. Nicolelis M 1999 Methods for Simultaneous Neuronal Ensemble Recordings. Academic Press, New York
12. O’Keefe J, Dostrovsky J 1971 The hippocampus as a spatial map. Preliminary evidence from unit activity in the freely moving rat. Brain Research 34: 171–5
13. Schoenbaum G, Eichenbaum H 1995 Information coding in the rodent prefrontal cortex. II. Ensemble activity in orbitofrontal cortex. Journal of Neurophysiology 74: 751–62
14. Singer W 2000 Why use more than one electrode at a time? In: New Technologies for the Life Sciences—A Trends Guide. Elsevier, pp. 12–17
15. Singer W, Gray C M 1995 Visual feature integration and the temporal correlation hypothesis. Annual Review of Neuroscience 18: 555–86
16. Skaggs W E, McNaughton B L 1996 Replay of neuronal firing sequences in rat hippocampus during sleep following spatial exposure. Science 271: 1870–3
17. Von der Malsburg C 1994 The correlation theory of brain function. In: Domani E et al (ed.) Models of Neural Networks II. Springer, Berlin
18. Young B J, Otto T, Fox G D, Eichenbaum H 1997 Memory representation within the parahippocampal region. Journal of Neuroscience 17: 5183–95