(last updated Sept 2007)
A page from: Eric Hargreaves'Page O'Neuroplasticity
Hippocampal "place" cells are presumably the principal cells in each of the layers that fire in complex bursts when an animal moves through a specific location in an environment. The region in which a cell fires the most is that cell's "firing-field" or "place-field" (O'Keefe and Dostrovsky, 1971; O'Keefe, 1976). Inside its field, a place cell may have a maximum rate of 20Hz or more, whereas outside its field, a place cell may fire less than 1 spike every 10 seconds (.1Hz). Given a sufficient number, place cells and their fields are able to cover or "map" any given environment. Thus, evidence from place cells offers strong support for the hippocampus' involvement in spatial mapping (O'Keefe and Nadel, 1978).
A Who's Who and What's What of Place Cell Research
John O'Keefe and Lynn Nadel and the Cognitive MapThe discovery of "Place-cells" by O'Keefe and Dostrovsky (1971) was the key phenomena that later led John O'Keefe and Lynn Nadel, both graduates of McGill University, to propose that the hippocampus was the locus of a "Cognitive Map" (1978). O'Keefe and Nadel, pictured at right, further posited in their book The Hippocampus as a Cognitive Map, that the function of the map was not only to permit spatial navigation throughout an environment, but to further act as a memory framework upon which the significant items and episodes of experience could be superimposed. The book laid the groundwork for the theory of the same name, which remains one of the predominant and enduring conceptualizations of rodent hippocampal function. Although, long out of press, the book used to have a small but dedicated following of researchers and students clamouring for copies from used book sellers, or passing around complete photocopies. After receiving numerous requests for the book themselves, the authors sought to regain the copyright from Oxford Universsity Press, who willingly relinquished their hold. Consequuently, the authors have converted the book to PDF files and posted it on the web for free downloading. (How cool is that?) Thus, one of the main peices of literature that is still driving research today is free for the taking at: Cognitivemap.net. Of course, John O'Keefe is still doing place cell work, and has trained a number of others that have passed through his lab, most notable of which are Kate Jeffery and Patrick Martin.
The State University of New York (SUNY), Brooklyn groupHowever, much of the supporting evidence of the Cognitive Map theory has come from the "Downstate SUNY" group in Brooklyn, which next to O'Keefe, is the founding group of place cell research. Initially, James Ranck Jr., alone, characterized hippocampal cells in freely behaving rats as either complex-spike or theta units (Ranck, 1973). Joined by then postdoc Steve Fox, and then grad student John Kubie they began to focus more on theta units. However, with the addition of then new faculty member Bob Muller, the SUNY group during the mid-80s both revolutionized and legitmized place cell research by quantifying place field characteristics with statistical measures derived from software developed "in house" (Muller et al., 1987; Muller and Kubie, 1987). Since that time the SUNY group has been one of the pre-eminant place cell research groups, and have produced a number of similarly eminant single unit researchers, in several fields... ...Jeffery Taube: head direction cells, Pat Sharp: subicular cell characteristics, Greg Quirk: neural basis of fear conditioning and development of Neuroscience in Puerto Rico, Cliff Kentros: place cells in mice, and one of the most recent additions at SUNY Andre Fenton.
The Barnes and McNaughton labs
Data from place cells are also consistent with newer versions of "path-integration", which are related to the cognitive map, but have earlier and independent origins. In his book, The Organization of Learning Galistel spends a chapter discussing the navigation methods used by old wooden sailing ships during transcontinental crossings. This discussion provides a strong analogy for how animals may construct maps using "deduced" or "dead" reckoning. More important however, is the identification of the error types and the distortions that this type of navigation allows to become incorporated into the maps that are subsequently constructed for getting to locations important for an animal's survival.
Thus, the main difference between the cognitive map and path-integration, is path-integration's focus on the animal's ability to synthesize its motor activity (direction, velocity, and time spent in specific movement vectors) into a navigational position (Gothard, et al., 1996, Knierim et al., 1996, McNaughton et al., 1996, Whishaw, Cassel & Jarrad, 1995, Whishaw & Jarrad, 1996, Whishaw, McKenna & Maaswinkel, 1997). As can be seen, many of the articles and reviews come from the group headed by Bruce McNaughton and Carol A. Barnes, pictured above and to the left, one of the big American dynasties of hippocampal research, which I like to characterize as "more- bigger, better, faster". Consequently, Bruce and Carol have managed to influence much of the next generation of place cell researchers that passed through their lab including (alphabetically) Jim Knierim, Sheri Mizumori, Dave Redish, Pat Sharp, Bill Skaggs, and Matt Wilson. Of course, Bruce and Carol come with their own strong lineage that includes both Per Andersen and Donald O. Hebb. The path-integration review that covers the most recent research, including entorhinal grid cells can be found in (McNaughton, Battaglia, Jenson, Moser and Moser (2006).
The Eichenbaum lab
A third option argues that place cells are a specialized operation that fit into the more general "episodic" memory function of hippocampus (Eichenbaum et al., 1999). This idea has largely been drawn from human neuropsychology, and the non-human primate literature, but has been implemented for rats by Howard Eichenbaum and colleagues (Wood, Dudchenko and Eichenbaum, 1999), and in a separate fear conditioning paradigm by Joe LeDoux's group ( Moita, et. al., 2003). This has led the Eichenbaum group to question the memory function of the rodent hippocampus as a "spatial memory" system, and instead, try to characterize it as a "memory space", able to encode the many different types of infomation that could occur as "episodes". These episodes are then linked together in sequences, and inter-related, eventually building a semantically based network (Eichenbaum et al., 1999; Shapiro and Eichenbaum, 1999). Both the episodic and semantic memory functions are part of the larger "Declarative" memory system, again originating from the clinical human and non-human primate experimental literature ( Mishkin et al., 1982; Schacter and Tulving, 1994; Squire et al., 2004; Suzuki and Eichenbaum, 2000). The most recent of the Eichenbaum group's prodigous number of reviews weave together evidence from many areas ( Eichenbaum, 2004; Eichenbaum and Fortin, 2005; Eichenbaum, Yonelinas, and Ranganath, 2007). As such, Howard Eichenbaum's group is one of the other big American Dynasties, and since Matthew Shapiro, during a sabbatical in the mid 90s, working with Heikki Tanila, then a postdoc himself, brought the single unit place cell recording techniques to the lab, they have also generated a small cadre of place cell researchers such as Paul Dudchenko, and Emma Wood. Thus, somewhat like the McNaughton/Barnes group the Eichenbaum group is a more general hippocampal research group that have delved into the place cell phenomena, as opposed to the SUNY group, which for the most part have focused on everything "spatial" via single unit recording.
The Buzsaki lab
The final hippocampal dynasty in the states that demands mentioning, centers on Gyorgy Buzsaki's lab at Rutgers NJ. Somewhat like Jim Ranck, Gyorgy's (Yuri) origins deal with theta, which led to single unit recordings as a way to get at the underlying mechanisms of the endogenous hippocampal EEG rhythm (for review see Buzsaki, 2002). Consequently, much of the focus of Gyorgy's work has therefore been on hippocampal interneurons, managing to fill a whole issue of the journal "Hippocampus" with everything physiological and anatomical about interneurons (Freund and Buzsaki, 1996). Further contributing to one of Gyorgy's collaborators Peter Somogyi's, projects that has to be the "coolest" theta article in existence (Klausberger et al., 2003). Here, individual interneurons were recorded in vivo, during theta and non-theta states, relating them to the local field potential (LFP) polarizations and depolarizations, and then labelled and identified by anatomical type afterwards. Separately each feat is no big deal, recording from different types of interneurons and relating them to the LFP patterns, or labelling and identifying individual interneuronal types anatomically. However, doing it all together, anatomically labelling the very neurons being recorded from in vivo, is a very big deal, in fact, extremely huge! On the flip side Gyorgy's lab has also focused on hippocampal sharp waves (SPWs), a field he almost single handedly opened up and promoted to its present level of importance, where everyone else has caught on. This is also where complex-spike cells (or pyramidal cells) re-enter the picture and how their temporal firing patterns contribute to different aspects of place fields ( Harris et al., 2001) , and why Gyorgy's lab also deserves extensive discussion on my place cell page (among other locations on my website).
Thus, the rodent hippocampus' exclusive role in spatial mapping has been a topic of some dispute (Hippocampus Forum, 1991; various authors), which almost a decade later, continues as an ongoing discussion/debate (Hippocampus Forum, 1999; various authors). Still the few studies that indicate the presence of non-spatial signals that they are often dominated or carried by spatial signals ( Knierim, 2003; Knierim, 2004).
Single Unit Recording
For one to record place cells, one must be able to record individual cells (or units) in a behaving animal. Although this may appear to be stating the obvious, much single unit recording is performed in anesthetized whole animals, or slice preparations. In order to record single cells, one needs suitable wire. Suitable wire is that which is the same size or smaller in diameter than the cell bodies being recorded. Yet, suitable wire is also that which is of low enough impedance to detect the microvoltage discharges of the cell bodies being recorded. These two requirements are at odds with each other, since the more fine the wire, the greater the impedance. How fine is fine? Well, typically less than a hair's width, which brings us to another difficulty. How does one implant and keep such a fine electrode in a stable position nestled up close to the cells bodies of a moving animal?
Multitrode recording technologyOne of the solutions has been to use multitrode technology, which is simply the use of multiple recording wires. Typically, multitrode recording either consists of stereotrodes (2 wires; McNaughton, O'Keefe, & Barnes, 1983) , or tetrode (4 wires; Recce & O'Keefe, 1989 Gray et al., 1995). The data presented on this site come from a variety of techniques, dependent upon the labs and projects on which I've worked. In Matthew Shapiro's lab at McGill we used 30 micron wire twisted into tetrodes. In Bob Muller's lab at SUNY Brooklyn my primary project required the use of single wire electrodes bundled into a brush of 10 wires each of which were 25microns in diameter. Of course others like Alex Rottenberg, had used tetrodes in the Muller lab, while students there, like Bruno Rivard, continue to use tetrodes. Currently in Jim Knierim's lab here in Texas, I have returned to using tetrodes as well, but using really fine wire; just 12microns in diameter. Since the diameter is so small, we decrease the impedance of the tetrodes by gold plating the tips. We gold plate our tips in exactly the same way electroplating jewelery is done. Apart from the multiple twisted wires adding strength, and therefore stability, they also add a large advantage in discriminating units through cluster cutting discussed later (McNaughton, O'Keefe, & Barnes, 1983; Recce & O'Keefe, 1989; Gray et al., 1995). The electrodes are then implanted in the brain, and attempts are made to snuggle up against the cell bodies one wishes to record from, or at least get as close as possible. In the case of acute recordings in anesthetized animals this is done through stereotaxic manipulators or drivers, which can offer 100th of a mm movement precision for placing electrodes next to the target cells.
MicrodrivesIn the case of chronically implanted, freely behaving animals (as when recording place cells), the multitrode is attached to a small device mounted on the headstage of the animal that allows fine movement of the multitrode. These devices are called microdrives. Microdrives are typically manufactured "in house". The figure on the left is the drive system traditionally used in the Muller lab, developed by John Kubie, pictured above, whose lab along with those of Bob Muller's, Steve Fox's and Jim Ranck Jr.'s made up the SUNY group at Brooklyn (Kubie, 1984). Both Steve Fox and John Kubie were a former student and postdoc respectively, of Jim Ranck. Additionally, newest to this group as faculty is Andre Fenton, who was a student of Bob's, before going abroad to Postdoc. So all in all, there's now three academic generations within the SUNY group, but more on these guys later.
The wires are stripped of their insulation at the top and attached to the pins of an augat connector by either solder or silver paint to ensure good electrical contact. The wires are then passed through a 26guage cannula and cut so that 1-2mm protrude from the tip, which will be implanted directly into the brain above the hippocampus. The augat connector is then embedded in a dental acrylic form along with 3 machine screws making a stable platform. The screws are mounted in a set of hollowed out teflon legs, which are directly anchored to the surface of the skull with grip cement or dental acrylic. By turning the screws the entire augat platform and therefore the cannula with the bundle of electrodes are lowered into the brain. Of course the whole microdrive is less than 1.5cm in diameter, as such, my Kubie drive shown above appears larger then it actually is. Smaller versions of the Kubie drive were manufactured by Alex Rotenberg for his work with mice in Bob's lab, and in turn, Alex's drives were remodelled by Cliff Kentros, a postdoc trained in the Muller lab, who implemented the "Mill-Max" connectors for his mouse work in Eric Kandel's lab at Columbia. As such, in house drive systems like this are always in a state of evolution, with slight variants being tried by virtually everyone who has to manufacture one. Similarly, the Kubie drive is not the only one out there, with the following published microdrives being but a mere smattering of what's out there (Bilkey and Muir, 1999;Korshunov, 1995;Bland et el., 1990).
The Recording setupAltogether the recording setup should look something like the pic at right, which was Matthew's unit recording setup when he was at McGill University. The leftmost apparatus was the recording chamber, covered on the outside with copper shielding much like a Faraday Cage, and cloaked in black cloth to seal the inside from light giving cues. I've left the door open, with the recording leads draped outside to be visible. The chamber was a uniform dark color inside, with removeable white cues by which the rats could navigate. The leads had unity gain preamplifiers at the end connected to the rat, similar in principle to a phonograph arm which also has preamps to boost the weak signal coming from the needle on the record (which is different then the signal picked up from a CD or cassette tape, which is much stronger to begin with). The boosted unit signal from the rats brain travelled through the leads to a set of amplifiers seen next to the computer monitor. Here the signals were truly amplified, going from the microvoltage of differential unit discharges to voltages readable and discernable by the analog to digital (A/D) system in the relay rack, situated between the recording chamber and the amplifiers. Also mounted in the relay rack were a couple of oscilloscopes for visual monitoring of the units and a "boombox" placed on top of the amplifiers for the audio monitoring of the units. Finally, a video monitor in the relay rack was hooked up to a camera mounted on the ceiling inside the recording chamber, which permitted the digital tracking of the rat's position.
Here's one of my guys in action at McGill. The shot shows the interior of the recording chamber. The white cue card to the right of the rat is the same cue visible through the open door in the larger figure of the recording setup. As can be seen both the floor and walls are covered in the same stone grey mica board making the interior largely uniform. Running horizontally along the walls was a strip of Velcro, which allowed the cues in the recording chamber to be easily attached, removed, or shifted to different walls, but more on the reasons for that later. You can also note how lightweight the recording heastage and leads are, with the rat totally unimpeded in its movements, as it runs around the chamber in search of rewards. Finally, note the "light emitting diodes" (LEDs) mounted on the headstage. Through the digital contrast video recording of these LEDs by the video camera, the location of the animal is monitored and recorded, such that its path and coverage of the environment can be reconstructed later.
So thats the standard hardware required for recording place cells, and a brief description of Matthew's unit recording setup, prior to his move to Mount Sinai in New York. Displayed at right was the setup I used in Bob Muller's lab in Brooklyn. It was fairly standard for the Muller lab except for the patch panel, stimulators and 2nd A/D board, which were required for the project I had inherited. The line of research was originally initiated at Bob's request by Matt Stead, an M.D./Ph.D. student, who reconfigured the system and designed the new patch panel.
On top of the relay rack were the amplifiers, beneath which was Matt's patch panel. The patch panel in this case split the amplifier output and redirected it to 2 separate A/D systems. Below the patch panel were the oscilliscopes, and stimulators. Beneath these were the 2 A/D connector boards. The first was the board setup by Matt, and the second was a commercial system from "datawave". Datawave, was also the A/D system used in Matthew Shapiro's lab at McGill, but now he's trying out Casey Stengel's "Neuralynx" system in his new Mount Sinai lab.
Returning to the Muller setup, one can just make out the cylinder through the open doorway of the recording room. Curtains typically enclose the recording chamber, but were parted so the cylinder could be seen in scale and relation to the rest of the setup. During the actual running of experimental sessions, both the curtains would be drawn, and the door closed.
The photo at left depicts the cylinder in more detail as "B17" (Feb'99) runs around in search of sucrose pellets. The food was dropped randomly from a pellet dispenser mounted in the ceiling. The cylinder itself was painted a homogenous grey, and the matching floor was simply a large sheet of photographic backdrop paper ripped from rolls that were always kept on hand. The sheets were replaced between behavioral recording sessions to prevent the use of odour cue trails that the rat may have left as it traversed the chamber. The recording chamber was adorned by a single white cue card, which extended from floor to rim, spanning some 1000 of arc. The 76cm diameter cylinder was one of the recording chambers used in the 1987 back to back Journal of Neuroscience papers discussed later. The SUNY group has published a detailed account of their recording techniques in Academic Press' Neuroscience Lab Fax series on neurobiology-methodology (Kubie, Muller, and Hawley, 1997).
Next up, are the various software techniques for discriminating single unit data into the different cells that have been recorded.
Recording of single unit activity typically starts by scanning the multiple unit activity (MUA) on the various electrodes for individual unit spikes. "Hunting" for single spike activity that is discernable from the background MUA takes up the majority of the recording time. Often weeks are taken to monitor the activity and slowly turn down the microdrives lowering the electrodes in search of the target cells. In contrast, very little time is spent running the manipulations that constitute the actual recording experiments.
Multiple unit activity (MUA) is much like slow wave brain activity or the electroencephalogram (EEG) recorded from the scalp, except one is recording MUA directly from within the brain. The frequency filter settings are also set to prevent the larger, more global slow wave activity patterns (low frequency) that make up traditional EEG data from being captured. Additionally, to aid in the selection of events proximal to the electrode tips, recordings are typically collected against "indifferent" wires, staggered slightly back from the electrode tips (0.5mm). Further, in order to prevent the indifferent or "reference" wires from picking up individual cells, sometimes, two or more indifferents are shorted together, such that any unit picked up by an individual wire becomes drowned out in the aggregate recording from the combined indifferents. Thus, only fast events, or individual unit activity very near the recording electrode tips are picked up.
The oscilloscope-like recording at right is a small 22msec snapshot from an original 16min recording session. The continuous trace is from a single electrode tip. During the 22msec epoch, a single unit spike waveform, two-three times the size of the background activity or "hash", discharges several times in a patterned or complex burst. During the burst each successive spike recorded from the cell is smaller than the last, such that the unit amplitude decrements across the complex burst. It is also worth noting that the "intraburst" frequency ranges between 5-4msec or 200-250Hz, which is quite fast when contrasted to the "interburst" frequency, which is typically in the range of seconds or thousands of msec.
The patterned burst is a complex-spike event. Complex-spikes, such as these, although rare, are one of the defining characteristics of "place cells" in the hippocampus and often the labels "complex-spike" and "place" are used interchangebly. Complex-spike bursts are comprised of anywhere from 2-10 spikes, with an intraburst frequency of up to 250Hz, and typically, but not always, each successive spike is smaller than the last. However it should be noted that whereas both "complex-spike" and "place" units are tentatively pyramidal cells, not all complex-spike cells are place cells.
Virtually all descriptions of complex-spike bursts are derived from one paper by James Ranck Jr., who also bestowed upon them the name of "complex-spike" cells (Ranck, 1973).
His descriptions of complex-spike cells along with the smaller, faster, and more ubiquitous firing "theta" units and their relations to behavior were one of the first papers to record individual unit activity in the hippocampus of behaving rats (Ranck, 1973;Fox and Ranck, 1975;1981).
As such everyone, who wishes to speak of complex-spikes, cites the original 1973 article, although of the many who do, few have actually taken the time to read it all, and it is well worth reading as a standard for behavioral electrophysiology.
Oddly, it has only been recently that complex-spike bursts have begun to be examined once more
(Quirk and Wilson, 1999;Quirk et al., 2001).
Regardless, portrayed at left is Jim executing hosting duties, with his usual "panache" at his wife's and his annual christmas party out of their Greenwich Village apartment in 1999.
Cluster CuttingCluster cutting is currently the main technique for discriminating different units recorded from the same electrode, whether recorded from a single wire, stereotrode or tetrode. Through analog (solid state) or digital (computer-like) window discriminators, unit waveform voltages cross an amplitude threshold and "trigger" the collection routines that capture and store the spike waveforms. Waveforms from the tips of the electrode are collected simultaneously, and when overlaid look something like whats pictured at right; a big mess of waveforms. However, from each waveform, parameters, such as "peak amplitude", "latency to peak", "area beneath the curve", and a whole host of others can be derived or calculated. These parameters from the simultaneous waveforms from the different wires of the same electrode are then graphed against each other in scatterplots. Supposedly, each unit will have similar waveforms to itself and therefore similar waveform parameters, but different waveforms to those of other units and therefore different waveform parameters. Thus, when the parameters are plotted they cluster in such a way as to differentiate or discriminate among the multiple units collected on a single electrode, stereotrode, or tetrode (McNaughton, O'Keefe, & Barnes, 1983; Recce & O'Keefe, 1989; Gray et al., 1995).
So given another set of undifferentiated waveforms such as those on the left, one can see that when the peak amplitudes of the waveforms are plotted against each other in scatterplots, the peak parameters form clusters.
These clusters are then "cut" or discriminated from each other by demarcating the upper and lower boundaries of the parameters with boxes
(although currently different software packages use a variety of shapes such as ellipses, or polygons).
The red green and blue parameter boundaries correspond to the same colored boxes bounding the different waveform peaks from whihc they were derived. Waveforms that cluster together in this way are thought to be from the same units or cells. Recently, this technique and the accompanying degree of accuracy and error were assessed by the Buzsaki lab at Rutgers, N.J. (Harris et al., 2000; Henze et al., 2000, who in anesthetized preparations simultaneously recorded extracellularly, and intracellularly. The intracelular recording acted as a 100% accuracy measure or "gold standard", by which to assess the extracellular cluster-cutting techniques. Depending upon the number of units and clarity of the clusters, errors could range from less than 1% up to a whopping 30%.
Earlier in my discussion of multitrodes, I indicated that one of the real advantages of stereotrodes and tetrodes over single wire recording techniques had to do with an increased power of discrimination. Apart from the obvious increase in the number of parameters one can use to "cut" the clusters, there's an increase in the accuracy of locating the 3D position of the cell by the added number of wires. This increase works very much the way having two eyes works in discriminating the depth of different objects. If an individual is limited to one eye, or a single point of view then the accuracy of depth vision goes down. This is easy enough to test by wandering around with one eye closed picking up different objects. It will soon be discovered that on occasion an individual will missjudge the depth. By having two eyes or two points of view one enables stereoscopic vision.
Assume, for the moment, that all of your target cells fire the same size spike with identical waveform parameters.
Using a single wire electrode one can only discriminate cells based on their distance away from the electrode tip.
Cells that are further away from the electrode tip have smaller waveforms, and cells closer to the electrode tip have larger waveforms.
Thus, if two cells were the same distance from the electrode and had the same temporal firing characteristics than those two cells could easily be confused with each other, or not discriminated well by cluster cutting techniques.
Potentially then, all cells an equal distance from the electrode tip in 3D space could be confounded with each other.
Thus, there would be a "sphere" of error surrounding the electrode tip.
Increase the number of wires to two, and the only cells in 3D space that could be an identical distance from both wires is dramatically reduced, generating a "plane" of error bisected by the electrode tips.
Add a third wire, and the "plane" is reduced to a "line".
Add a fourth wire and the "line" is reduced to a single "point".
Thus, the real power behind multitrode recordings is the ability to limit the number of confounding units in 3D space
(McNaughton, O'Keefe, & Barnes, 1983;
Recce & O'Keefe, 1989;
Gray et al., 1995).
Of course, what has been described above is an ideal setting where all the target cells are the same size, and have the same discharge characteristics.
Yet, by already adding in the decrementing nature of
the situation becomes more complicated.
Of course the strictness of the parameter limits can also affect the interpretation of the subsequent results. By tightly defining the cutting parameters there is a high probability that every waveform assigned to a unit belongs to that unit, and thus also, an increased probability of excluding waveforms that belong to that unit (false negative). Alternately, by loosely defining the cutting parameters, there is a high probability of assigning every waveform to a discriminated unit, and thus also, an increased probability of including waveforms that do not belong to discriminated units (false positive). Prior to the use of multitrodes, which give much stronger discriminative powers, whether an individual unit could have multiple fields or subfields was a debate often suspected of being contaminated by loose cutting parameters. Thus, earlier on it was thought that each unit may only have a single true field, and that previous work suggesting subfields may simply have not discriminated well between units (Muller et al., 1987). Conversely, work indicating that place fields are stable over long periods of time, by using overly strict cutting parameters, may have excluded subtle shifts and variations in the fields by excluding unit waveforms.
Displayed on the right is the discriminated result of the earlier tetrode recording. The separated wave forms (red, green, blue) potentially represent individual units, each with a unique waveform pattern across the four tetrode channels.
The "red" discriminated unit is recorded as a large spike on channel two, and relatively small spikes on the remaining channels. In contrast, the "blue" discriminated unit is recorded as a small spike on the channel two, but recorded as a large spike on the channels three and four. Finally, the "green" discriminated unit is only recorded as a large spike on channel three.
Cluster cutting is also used when recording single unit data on fewer electrode tips, such as two tips (stereotrode) and one tip (single-ended). Yet, as can be seen from the current example, had cluster cutting been used single-endedly with only waveforms on channel one available, then no cells would have been differentiated. Similarly, if stereotrode recording of channels one and three had been used, but only with peak amplitude as the discriminating feature, then the waveforms identifying the blue profile would not have been separated from those identifying the green profile.
The result of the example at right was cut manually (ie the drawing of the boxes was guided by visual inspection). However, automated statistical procedures, such as discriminant cluster analysis or cannonical correlation, can also perform similar functions.
Other Discrimination Techniques
Other techniques like "template matching" also exist for discriminating between cells. As with cluster cutting, template matching involves the selection of waveform features like peaks and valleys, the values of which are then used to generate mathematical functions fitted to the full shape of the waveforms. Subsequently, the waveform functions themselves are mathematically discriminated into separate clusters, as is done with parameter cluster cutting.
Apart from the complex bursts, most of what has been described so far is applicable to discriminating single unit activity throughout the brain.
Whether the full power of tetrode recording is necessary, depends upon the size of the cells and how densely they are packed.
Larger cells, and/or more sparsely spaced cells are easier to discriminate than smaller more densely packed cells.
In the CA1 pyramidal layer of the hippocampus, where the cells are densely packed into a thin layer, tetrode recording often enables one to record a number of cells simultaneously, as in the current example, and in what has been described as "ensembles"
(Wilson and McNaughton, 1993; Wilson and McNaughton , 1994;
Eichenbaum et al., 1989).
Currently, the published record is held by Matt Wilson, when in Bruce McNaughton's lab, simultaneously capturing some 150 cells
(Wilson and McNaughton, 1993).
Now of course, we all love to talk about recording 150 cells ourselves, and love to allude to the Wilson Science papers as though its an everyday occurrence, but honestly, for much of our ensemble work, we consider ourselves lucky to have 15-20 simultaneousy recorded units.
Below, you will find an example of my largest ensembles, utilizing three tetrode aimed at CA1.*********************************
Place Field Basics
Just to recap or alternately introduce... Hippocampal "place" cells are cells that fire in complex bursts when an animal moves through a specific location in an environment. The area in which the cell fires maximally is that cell's "place-field".
Displayed at the right are results from a discriminated tetrode sample recorded during a 10min session from the CA1 pyramidal layer of the dorsal hippocampus. The steps necessary to arrive at a discriminated tetrode recording have been described in the preceding section. In this sample there are three discriminated units, each assigned a different color (red, green, blue). Although aggregated together and overlaid on top of each other, the individual firing of each unit spike is marked in "time" and "2D-space". Thus, recording sessions can be reconstructed based the dimensions of time and/or space.
The display of the units' discharge over time is known as the rastor plot. Coupled with the location and specific activity of the animal, the rastor plot is the best way to determine the behavioral correlates of individual units' discharge patterns. The data below are from the same rat and session as the discriminated tetrode recording at right. The rastor plot displays a 5 second epoch of the full recording session. The horizontal gray traces represent the four zero lines of the tetrode recording channels, and the colored vertical slashes represent the discharge of the individual units, as well as the approximate size of the unit spikes on each of the four channels.
As described above, the "red" unit has the largest spikes on the second channel, the "green" unit has the largest spikes on the third channel, and the "blue" spikes are largest on both the third and fourth tetrode channels.
It is worth noting that the units do not fire continuously, or regularly, but fire in bursts or trains
as the animal passes through the various fields.
In order to know where a place field is... ...one needs to know where a place field isn't. In other words, the whole recording environment must be adequately covered in order to define the true boundaries of a place field. Currently in place cell work, the location of the animal is typically known through contrast video recording of light emitting diodes (LEDs), and thus, the path of the animal can be reconstructed and plotted on a cartesian map of the environment. Earlier work did not have the same exact monitoring methods available, and thus, environments that had well isolated areas, such as the radial maze or related T-mazes were popular. Theoretical "ties" also link the radial-maze to place cell work in that an intact hippocampus was required for normal learning and performance of the maze, and as such, was argued to test spatial memory.
In the figure at left, the perimeter of the enclosed recording chamber is outlined in green and the path of the rat is traced in orange. In order to show where a unit's place field exists, and where it does not exist, it is important to get full and even coverage of the environment during each recording session. Even so, one can see from the location of the raw spike plots at right that the least travelled area, at left, also had the fewest spikes and that the heavily travelled areas, typically the perimeter and corners tend to have more spikes plotted. Spike rates are currently used to compensate for these differences in coverage. Evenness in coverage was also one of the reasons why the radial maze was popular as a recording environment. Regardless of the coverage in the present example, one can still clearly see that the three discriminated units at right (red/green/blue) have specific locations within the recording environment that they fire in. These locations are place fields.
Naturally, a rat will not continually run around an environment for ten minutes or longer without some inducement to do so. Traditionally, the motivation has been food, so sucrose pellets or chocolate chips are distributed throughout the recording chamber, or alternately dispersed one at a time in a random location. In our example here, a newer technique has been employed, that of lateral hypothalamic stimulation. Intracranial stimulation of specific brain sites has been known for a number of decades to be rewarding (Olds and Milner, 1954). It is not surprising then, that rewarding brain stimulation can be used to motivate an animal to thoroughly explore the environment during place cell work (Fukada et al., 1992; Hetherington and Shapiro, 1997; Tanila, Shapiro, & Eichenbaum, 1997; Shapiro, Tanila & Eichenbaum, 1997).
In the above example at left, drawn from one of my sessions in Matthew's lab, the cyan colored dots represent locations where the rat received rewarding stimulation, and the yellow circle in the lower left of the figure is the next randomly selected region for the rat to receive rewarding stimulation. The randomly placed reward region is a computer generated area, "virtually" defined in the environment, and thus, exhibits no concrete landmarks by which the rat can identify its location. As such, the rat's optimal strategy for receiving the greatest degree of rewarding stimulation is to continually run around the environment covering all areas. In our example using LH reward, the total path length of the animal was 80.69 metres over a 10min session, and our recording chamber was 83cmX83cm (for details see Hetherington and Shapiro, 1997).
As a final comment, some experiments leave little choice, but to use brain stimulation, as in one of my favourites... ...the McNaughton group's attempt to ascertain the 3D nature of place cell's by sending up rats and recording place cells in space.
Yup, part of the Neurolab's Space Shuttle mission in April '98 was to take up 3 implanted rats and record place cells in zero gravity on what became known as the "Escher staircase track".
Naturally in zero gravity one cannot drop food pellet rewards, without them floating off into space, so rewarding brain stimulation was a natural.
Well OK, natural if one considers rats floating in zero gravity natural
(Knierim, McNaughton and Poe, 2000).
Place Field Computations and Firing Rate Maps
Given another example of a discriminated unit in the figure at right, how does one define what is and what is not part of the place field? ie what are the numerical procedures used to quantify the raw spike data and associated x-y coordinates that allow for statistical comparisons?
Originally, once the unit was discriminated and the general location determined, like the figure on the right, that was it.
For many of the early experiments that were attempting to determine what place fields responded to in terms of environmental manipulations, such analyses were sufficient, due to the large differences between a unit's "in-field" and "out-field" firing rate (Muller et al., 1987). However, the more refined and subtle questions that came later, required a similar refinement and subtlety in analyses, and the advent of personal computers with digital recording systems soon allowed such refinements.
Currently, a number of steps are used in computing place fields, the data from which are often then further filtered or smoothed according to known or assumed nature of the fields and firing properties of the units themselves.
First, the more exact x-y coordinates of each discriminated spike's location in the environment, recorded from the attached LEDs, are aggregated into larger blocks of space, along with the number of spikes in each of the resulting matrix locations. In the case of our example, the 83cmX83cm recording chamber was divided into 28 matrix locations with each location approximately 2.96cm a side.
Second, the raw spike count matrix above is then matched up with the two other behavioral matrices available from the LED data at right: 1) number of visits to each matrix location and 2) time spent in each matrix location.
Restricting criteria can be placed on the spike data, based upon these behaviors, such that if a minimum number of independent visits in any given matrix location was not made or a minimum amount of time was not spent, then that matrix location's spike data may be deemed as under-represented and unreliable and subsequently discounted
(Hetherington and Shapiro, 1997).
Third, the raw spike rate per matrix location is calculated as spikes/time. Although much of the environment was visited more than 3 times and greater than 0.25 sec was spent in each location, many of the spike rates, still zeroed out in our example above, indicating no unit firing for that location. To a certain extent this indicates the precision with which place fields represent the environment. Alternately, this degree of spike-rate drop off can also be modulated by the strictness of the upper and lower limits of the cutting parameters used to discriminate the units.
Often filtering and smoothing algorithms are applied to the spike rates at this stage, to make the spike rate data obtained from brief (10min) recording sessions resemble data obtained from much longer sessions (>30min) (Hetherington and Shapiro, 1997; Muller et al., 1987) and in fact, Gaussian distributions have been used to describe place field firing rates (Hetherington and Shapiro, 1997; O'Keefe and Burgess, 1996).
Thus, the data from our example have been filtered and smoothed, and then plotted, much like a geographic map would be displayed. In fact, these matrices are known as spike rate maps or firing rate maps, conceived initially by Bob Muller, John Kubie and Jim Ranck Jr. at SUNY Brooklyn, whose back to back publications in Journal of Neuroscience, set the numerical standard for analysing and presenting place field data. Thus, presented above are the data as a "contour plot", and presented to the right are the same data as a "3D-mesh plot". In both cases, the successive colors represent different firing rates from no firing at 0 to this unit's maximum firing rate of 10/sec.
Depicted at left is another place field, totally unrelated to any of the examples discussed so far. The depiction style is the one developed by Muller et al. (1987). As indicated at the start of the computation section, prior to the numerical standards set by the Brooklyn group, place field work was largely reported anecdotally. Consequently, the existence of place fields was not widely accepted until the Muller et al. (1987) papers provided objective numerical measurements of place fields, and the effect of various environmental manipulations upon these measurements.
The "Muller" firing rate map is much like the 2D cartesian map depicted above. However, instead of the different color bands delineating equal firing rate divisions, the six discrete color pixels (matrix cells) delineate proportions of the firing field area, such that each successive darker color (subsequent to yellow), is represented by 80% fewer pixels than the preceeding lighter color. Thus, the firing rate of the darkest pixel (or highest firing rate) is twice that of the lightest pixel (or lowest firing rate). In our example, in succession, the colors occupy 42 pixels (orange), 35 pixels (red), 27 pixels (green), 21 pixels (light blue), and finally 17 pixels (dark blue). As such, the "Muller" rate map depicts changes in firing rate tied to the area of the field, and not directly to the firing rate itself. This field in particular, is from "E10", who will be showing up in subsequent sections, and as such is one of my favourites.
Of course, since some of my best work was done in Texas in
Jim Knierim's lab, here is one of my guys, from there, running around the recording chamber foraging for randomly dropped chocolate sprinkles (Yep, jes' like you find on chocolate cakes).
The chamber itself had very similar dimensions to the "Muller" cylinder, and the color scheme of the walls and single lone cue was based on the cylinder as well.
The floor was covered with brown packaging paper, which was replaced every behavioral session to remove any lingering odour trails laid down, which the rat may use in subsequent behavioral sessions as orienting or landmark cues.
In this room, in particular, there appeared to be shifts in the venting, such that we would get asymmetries in the air movement, which sometimes carried very light scents.
Just underneath the table was a white noise generator (like the hissing sound your TV makes, when you wake up late at night after having fallen asleep during "Letterman" to find the station has gone off the air).
When turned on the generator masked any of the localized auditory cues that the rat could use to orient with, such as individuals walking down the outter hallways, or the nearby elevator.
The behavioral manipulations we were performing between recording sessions were very delicate and dictated that we control as many orienting cues as possible.
In the extreme, for some of the rats, towards the end, we would do a "quick spray" of air freshener around the room between trials, masking any subtle scents brought in by the venting.
The chamber was set on a small table in the center of the recording room, which itself was draped and darkened by a circular enclosure of heavy photographic curtains at its perimeter.
Above and to the right is the behavioral coverage (grey line) and the firing locations of individual spikes from a single place cell (blue dots) recorded in the square chamber above.
You can see from the density of the continuous path function that the animal tended to be a bit thigmotactic (wall hugging) and even where the spike firing density is thickest, the path density is fairly sparse.
The "smattering" of spikes firing outside of the place field, is also worth noting, such that not all of the cell's firing is within the confines of the place field.
Of course, this out of field smattering simply washes out and vanishes, when averaged into the smoothed firing rate map.
Although smoothed with a boxcar filter (originally implemented by Bill Skaggs), the firing rate map's color scheme is linearly scaled from the minimum firing rate to the maximum firing with no special emphasis on the "banding" of colors as there was with the "Muller/Kubie" firing rate map.
In fact, the color scheme was determined by the default for Matlab's "Colormap" function.
Place Field Manipulations
OK, so now one has what appears to be a place field, in that 1) the discriminated unit has been well isolated and the probability of it representing the discharge of more than a single unit is remote, 2) the discriminated unit appears to have a well defined firing field, where the in-field firing rate is several times greater than the out-field firing rate and 3) behaviorally, the environment of the recording chamber was well covered. Yet, even though it appears to be a place field, how can one be sure? Two standard criteria have been used, stability and more importantly environmental rotation, both of which are described in the next sections.
Beyond defining criteria, a number of other manipulations have been performed in attempts to determine what it is that place fields respond to exactly. These manipulations such as cue removal, lights out, and environmental distortions, are described in subsequent sections.
Place Field Stability
One of the first criterion for defining whether a unit's firing field is a "place field" is the firing field's reliability. Therefore, if an environment is stable over time, then a place field should exhibit similar stability. How stable is stable? Thompson and Best (1990) recorded place fields over a period of days to months, with the best unit reliably recorded for 153 days. It was possible to record the unit for longer, but the decision was made to advance the multitrode further through the cell layer in search of other units.
The animated field displayed below is one of the longest running fields I've recorded in the Muller lab, with a duration of 18 days. The unit generating this field, also happened to be the largest unit I've recorded, initially weighing in at over 400ÁVolt (Sept 04/99), and "maxing" out at just over 1mVolt (Sept 10/99). These data were recorded from E10, whom I've already mentioned in the preceeding section. The variability of the field is fairly striking, and is supported numerically by the descriptive data presented in the accompanying table.
Of course, many more sessions then those displayed were recorded, and after the first two days of the unit's appearance, environmental manipulations were initiated, including a variety of novel environments, cue removals, and rotations.
Regardless of the instability of the environment, all the sessions displayed were recorded as "standard" or baseline sessions, prior to which no manipulation had occurred for at least 12 hours.
However, the possibility remains that had the number of manipulations been fewer, the place fields may have had less day to day fluctuations.
Thus, one of the issues in place field stability is the degree of environmental constancy,and almost as if to voice its support of the environmental constancy idea was the the unit and accompanying field from E8 displayed below, which appeared within days of the end of the unit from E10.
Apart from environmental constancy, the stability of unit recording can be affected by solidness of the recording assembly, and by the strictness of the waveform parameters used. How solid, or physically stable the microdrive and its accompanying headstage assembly are when mounted on the skull, fixing the multitrode in position in the brain, is an obvious factor. Less obvious however, is the counterbalancing of the recording cable and the resultant changes in strain exerted on the connected headstage. How strict the waveform parameters are cut along also affects the day to day variability of a unit's place field. By restricting the upper and lower limits of the unit cutting parameters, one may very conservatively define a place field's characteristics, as did Thompson and Best (1990). Alternately, by loosely defining the upper and lower parameter limits one may capture a greater floridity of place field characteristics, but at the expense of a more stable core.
Place Field Rotation
Another criterion for defining whether a unit's firing field is a "place field" is the firing field's dependence upon the external environment. If a place field is controlled by a configuration of cues distal to the field itself, then rotating the environment should have a similar effect on the place field. This is one of the more traditional criteria for place fields and dates back to O'Keefe's (1976).
Systematic rotation manipulations however, were not carried out until the next set of experiments published two years later ( O'Keefe and Conway, 1978). In these experiments place field units were recorded on an elevated open platform with attachable arms, such that they formed the shape of a "T". The "T-maze" was enclosed on all four sides by curtains, each associated with a single distinct cue, such as a low wattage light or electric fan. Rats were trained to run from the end of a "start" arm to the end of a "goal" arm to receive either rat chow or sweet condensed milk, after which they would run to the non-goal arm and back to the start arm to receive a further reward, covering the whole maze during a trial. After the animals were successfully trained, the maze and curtains were rotated by 900, 1800, and 2700 degrees on different trials, such that the relationships between the cues and arms of the maze remained constant, but that the relationships between the external room and the enclosed maze and its cues changed from trial to trial. Additionally, the arms of the maze were not truly rotated, but interchanged among each other, preventing any local olfactory or sensory texture cues from having control over the firing fields. When the enclosed maze and its cues were rotated, the units maintained their fields in relation to the arm positions and cues. Thus it was the cue-controlled enclosure that the units relied upon for determining their location on the maze ( O'Keefe and Conway, 1978). Since that time environmental rotations have become standard manipulations for characterizing place fields, and often are performed as precursors to further manipulations involving controlling environmental cues (Muller and Kubie, 1987).
Above and to the right is an example of a rotation, this one performed in the Muller lab. During the two sessions reported here, the standard 76cm diameter cylinder and single 1000 cue card were employed as the recording environment and the rat was running around the cylinder chasing randomly dropped sucrose pellets (Muller and Kubie, 1987). The animation flips between the two-frames of the baseline recording session, and the clockwise 900 rotated recording session. In all, 10 separate fields were recorded simultaneously, although one of the isolated units had two fields. Yet, since single electrodes were employed to record these cells, it is possible that the two fields may have been from two separate units. However, it may also have been that the isolated unit had two fields, since even well discriminated tetrodes units have exhibited multiple fields.
The animation may convincingly indicate that all 10 fields rotated somewhat with the cue, but to what degree is determined by the analyses shown at left. The figure is a polar plot of the correlations between the standard baseline session of a field and its rotated session for all 10 fields, with each field's correlations plotted as a different color. Correlations are a statistical measure indicating the degree of relationship between two sets of numbers, and here, the sets making up the standard and rotated firing rate maps. A correlation of 0.0 shows no relationship between sets, and a correlation of 1.0 shows a perfect linear relationship between sets.
Focusing on a subset, the "zoomed" subplot at left depicts the correlational analyses for the first (red plot) and third (blue plot) place fields of the animation. Initially, a single correlation between the baseline field and its rotated counterpart is plotted. Since the fields in their original positions have no overlap, the correlation is close to zero and therefore plotted on the polar plot at 00 at the vertices of the two axes. The baseline field is then reoriented by computationally rotating it by one degree. The correlation between this new computed position and its rotated counterpart is subsequently plotted. This procedure is repeated until all 360 correlations have been calculated and plotted revealing at what angle the baseline field best matches the rotated field. As can be seen around 750 the correlations dramatically increase from 0.0 to 0.6, as the reoriented baseline field begins to overlap with its rotated counterpart. Once the fields overlap, the correlations steadily increase to become maximal at about 0.8, indicating a very strong relationship between the fields. Conversely, a dramatic decrease in the correlational values occurs around 1150. Thus, the strongest relationship or best match between the baseline field and its rotated counterpart occurs at close to 900. However, it is worth noting that for the two fields, the peak correlations occur at slightly different angles to each other, and further occur at slightly different angles to 900. This imprecesion has to do with the session to session variability of recording place fields, as mentioned previously in the section on stability. The variability may stem from differences in the behavioral coverage of the environment, as the rat becomes satiated with the sucrose pellets during the second rotational session. Alternately the variability may stem from the firing properties of the fields themselves as noted above, and found by Fenton and Muller (1998). Similarly, briefly returning to the full analyses, one may note that the correlation series for the yellow plot (the fourth field in the vertical animation) appears to be less pointed then the others. The broadness of the plot indicates that there is a much wider range of overlap between the original and rotated position of the 4th field. The reason that the optimal rotation is not so distinct is that the 4th field is larger and more centrally located than some of the other more punctate and peripherally located fields. Taken to an extreme, if a field were spherical in nature and exactly placed in the center then all correlations would be equal. Regardless, these analyses were part of the numerical standards first laid out by Muller et al. (1987), and the results presented here are well within the norm.
The fact that all 10 fields rotate simultaneously as a coherent ensemble indicates that under normal conditions, and without major perturbations in the recording environment, all recorded fields tend to function as a single unitary constellation. However, when the environment is majorly perturbed, such that location identifying information is placed into conflict, then the constellation of fields may break apart. Shapiro et al., (1997) recorded ensembles on a 4-arm radial-maze, with both distal (visual) and local (tactile, olfactory, and visual) cue sets available. Standard sessions (cue sets aligned) were run until well discriminated units could be detected. After baseline sessions identified and located units and their fields, the cue sets were doubly rotated, with the distal cues rotated 900 in one direction and local cues rotated 900 in the other direction. Results of the doubly rotated sessions indicated that when placed in conflict the place field constellations broke apart, with 28% of the fields rotating with the distal cue set, 15% rotating with the local cue set, 43% re-mapping, and 12% remaining stable in the baseline position. Further sessions were then run scrambling or removing individual cues within the cue set to which the units responded.
As was apparent from the previous section in which place fields shifted with the cue rotations in a controlled environment, place fields can be dependent upon distal visual stimuli. If this is the case, then removing cues in a controlled environment should provide some insight as to how they may determine place field location. Cue removals were also the subject of the O'Keefe and Conway (1978) paper, in which unit place fields were recorded atop of a "T-maze" enclosed on four sides by curtains, each with a distinct cue such as a card or buzzer. Removal of any individual cue or set of cues did not disrupt the majority of place field locations. Total cue removal however, resulted in the majority of units losing their fields, such that the units began firing more diffusely over the T-maze. Both increases and decreases in overall firing rates were observed in the different units that had lost their fields ( O'Keefe and Conway, 1978). The systematic examination of cue removal supported the anecdotal evidence from the original studies, and led to the understanding that place fields depended upon some set or constellation of cues, and not upon any single cue (O'Keefe and Dostrovsky, 1971; O'Keefe's, 1976).
Using the automated procedures developed to more fully characterize place fields
(Muller et al., 1987),
Muller and Kubie (1987) also examined the effects of cue removal among a variety of manipulations.
In these experiments the recording arena was enclosed by a circular wall which defined the perimeter of the recording chamber, unlike the more distal curtains in the
O'Keefe and Conway (1978) experiment.
In the circular environment the only determining cue was a card mounted on the chamber wall spanning 100 degrees of visual arc.
Removal of the card did not result in the disappearance of the place fields, as total cue removal had previously, but instead resulted a random rotation of the place fields, with their size, shape, and distance from the wall retained
(Muller and Kubie, 1987).
Although seemingly contradictory with the previous O'Keefe and Conway (1978)
results, the proportional height of the circular perimeter may have served as a partial, but continuous cue
(Muller and Kubie, 1987).
Given the arc and edge of the rim, the place fields may have had sufficient information to calculate the appropriate size, shape, and distance from the wall,
but have insufficient information to calculate the compass position that would allow the fields to become anchored in their location.
In the former O'Keefe and Conway (1978) experiment none of this information would be available once the four cues were removed, since the distally located curtains ran from ceiling to floor.
A final experiment on cue removal worth mentioning was done by Phil Hetherington and Matthew Shapiro (1997). This of course is where I did my previous postdoc, so I have referred to this study a number of times when describing the recording chamber and computational techniques used in my examples. However, since the main purpose of the Hetherington and Shapiro (1997) paper was to systematically examine the removal of individual cues, such that the salience and effect each cue had on the place fields could be determined, I thought I would now highlight it. Although this sounds like a mere replication of the original O'Keefe and Conway (1978) experiment, again by using the newer digitizing techniques and analyses more subtle changes were detected in the place fields beyond simple location. At left, is a depiction of the 83cm by 83cm recording chamber that was used, which is fully enclosed and described in previous sections on area coverage and environmental rotation. Rats were trained to thoroughly cover the area of the recording chamber in search of LH reward, while unit recordings with place fields were sought. Once such recordings were found, the units were "tuned in" and stablized, afterwhich standard rotation manipulations were performed, demonstating that the cue cards mounted on the walls did exert control over the place fields. Following these requisite manipulations, a single long recording session was performed, involving seven separate 12 min trials, in which baseline trials (with all the cues present) alternated with single cue removals (Hetherington and Shapiro, 1997). In agreement with previous findings, no single cue removal totally disrupted the place fields. However, subtle changes to the overall firing rate and size of the place fields occurred during individual cue removals, and were a function of the cue distance from the place field. Specifically, place fields decreased in area from a baseline average when a cue less than 30cm away from the centroid of the field was removed, but increased in area when a cue greater than 30cm away was removed. Further, in concert with alterations in place field area, were changes in firing rate, such that removal of cues less than 30cm away resulted in a firing rate decrease, while removal of cues greater than 30cm away resulted in a firing rate increase. Thus, the place fields appeared to have optimized the information available, such that removal of distal cues resulted in an increase of importance of nearby cues , while removal of nearby cues had the converse effect (Hetherington and Shapiro, 1997).
One of the early debates about place cells centred around whether the fields were mere "sensory correlates" of the stimuli in the environment. If this were true, and place cells did not really form a coherent map of the environment, then turning off the lights would make the apparent map dissipate. If, on the other hand, place cells truly represented a persistent cognitive map of the environment, then turning out the lights would not immediately affect its mnenonic representation. The situation is very analagous to getting around in the dark at home when attempting to change a blown fuse. For going about this task in a familiar environment, we typically have a map of the rooms we must pass through that will include the furniture locations and theh placement of the different doorways. However, overtime in the darkened environment, our estimate of the distances between objects weaken, and we still may unexpectedly bump into things.
An early study by
O'Keefe (1976) provided anecdotal evidence that place cells could persist in the dark.
Here, the rats were initially run on a radial-maze in the light and partway through the session the lights would be turned out and the rats' movements would be tracked thereafter by either dim red lights or by loosely holding their tails.
Not exactly, the way one envisions scientific technolgy, but it did provide an initial approximation of an affirmative answer.
The next advancement was
Hill and Best (1981), who manufactured tiny rat blindfolds for their rats.
They also found that place fields persist in the absence of visual cues.
However, not only did the fields persist, they had also developed in the absence of visual cues, since training never occurred without the blindfolds.
Accordingly, when the radial-maze was rotated (and left unwashed between the individual rat sessions), the majority of the fields rotated with the maze, suggesting that the majority of fields were developed in accordance with local olfactory cues (the rats were also deaf).
Of course, from the ealier O'Keefe and Conway (1978) rotation experiments, in which the arms of the T-maze were interchanged on each of the trials, we know that the fields followed the distal cues, and were not simply responding to local sensory information.
In this context the Hill and Best (1981) results indicate that in the absence of distal visual cues, rats can utilize other salient cues.
However, this did leave the minority of place fields (4) that did not rotate with the radial-maze unexplained (Hill and Best, 1981).
Of course, the proper way to do the experiment was to track the animals in the dark, and a decade later using infrared LEDs and camera, this is what Quirk, Muller, and Kubie (1990) did. In the Quirk, et al. (1990) experiment rats were trained to chase sucrose pellets in both a cylinder and square recording chambers. Over a two week period, rats were given 15 exposures to the environments including sessions in which the lights were turned out partway. Subsequent to training, the rats were implanted with single unit electrodes and after recovery from the surgery were run while the electrodes were turned down in search of units. Every time units could be well isolated, rats were run through a set protocol. During a single session the lights would be on for an initial 8 minutes, afterwhich the lights would be turned out for another 8 minutes, and then turned on for the final 8 minutes (LDL). This procedure was then repeated for the second apparatus. Following the LDL sessions, the rats would be introduced into the recording chambers again, but introduced in the dark. After the standard 8 minutes of chasing pellets, the lights would be turned on (DL). The DL session would then be repeated in the second apparatus. Thus, overall the experimental sequence was LDL LDL DL DL, with the floor of the recording chambers being changed between sessions.
Correlations for each place field across the sessions were determined to compare the persistence of the unit firing. As could be expected, the best correlations were found between the L segments of the different sessions. Significant correlations were also found between LD segments, and even DL segments. Although in many cases, even if the field persisted from one condition to the next there were reductions in the firing rate of the units. However, for the majority of cells that entered into the same chamber, from the last LDL to the first DL sessions, dramatic changes were found, including degradation, complete re-mapping, and shutting off. If the units shut off, the absence of firing would persist in the next L segment following the D segment, but then would return during subsequent L sessions. Similarly, if remapping occurred during the dark, the new remapping would often persist into the next L segment, only to be reset at the start of the next L session. So although disruptions can occur from dark to light and light to dark, overall the units showed a persistence in their fields, dependent upon their recent experience. Thus, although fields frequently changed from session to session, within the segments of a session, they were more than likely to be persistent. Finally, it is also worth noting that a little more than a third of the cells showed complete persistence in their firing throughout the LDL DL sessions Quirk, et al. (1990).
Scrolling down on the right you may have noticed the three sessions of an individual cell I recorded at McGill during a lights out sequence that was part of a larger experimental protocol. The first panel was the initial baseline session run for 10 minutes. Subsequently, and without removing the rat from the apparatus or unhooking it from the cable, the lights were turned off in the chamber and another 10 minutes of data were recorded. This session is depicted in the 2nd panel with the dimmed out walls indicating the "lights out" condition. Similar to Quirk, et al. (1990), the field persists, but has a lower firing rate than the baseline trial. Following this session the rat was unhooked and removed from the chamber, which was then washed down. The third panel shows the baseline session at the start of the next day.
Place field elasticity is actually a term I've made up, but I don't necessarily expect it to really catch on in the field. However, place field elasticity, as a term very accurately describes an important property or characteristic of place fields, which is their ability to stretch and distort in response to similar alterations of the controlling dimensions in their environments.
Two experiments illustrate this idea. The first is one of the manipulations done in the Muller and Kubie (1987) paper. As discussed before, the recording environments used by Muller and Kubie (1987) were chambers with a single cue card mounted on the wall. Although previously only manipulations involving the circular recording chamber have been discussed, there were in fact, two sets of two recording chambers. One set of chambers was square, while the other set, as described above, was round. Within each set, the two recording chambers were identical in all features, except that one chamber was proportionately twice the size of the other. The experiment was to find place fields within the smaller recording chamber and then record again in the larger chamber, to see if the fields scaled up in size. This is what occurred, such that the fields proportionately stretched to fit the enlarged dimensions of bigger recording chamber (Muller and Kubie, 1987). This scaling of the place fields occurred within both the square and circular sets of recording chambers, indicating that it was a relatively universal property and not just limited to a particular shaped recording environment. However, on average the place fields did not scale up to the expected ratio, but the fields recorded in the larger chamber may have been undersampled. Changes in the analyses attempting to compensate for undersampling did increase the scaling of the place fields, but still not up to the predicted size.
The second study to examine elasticity properties is O'Keefe and Burgess (1996). Again, much like Muller and Kubie (1987), a highly controlled, but minimalistic recording environment was employed. The recording chamber was constructed from four grey planks clamped at the top such that they could be slid back and forth to form a variety of squares and rectangles, as displayed in the figure on the right. No specific cues were mounted on the walls of the recording chambers, but with the walls only 61cm high certain cues within the recording room itself were accessible to the animals. Place fields, once found, were recorded in four different shaped chambers (small square, large square, horizontal rectangle, vertical rectangle).
Very much like the Muller and Kubie (1987) experiments, the idea was to present the different chambers as the same space, and not have them identified as different. However, where Muller and Kubie (1987) examined symmetrical enlargements or distortions, O'Keefe and Burgess (1996) were now additionally able to examine asymmetrical distortions.
The results indicated that many of the units were recordable in all four chambers, and were stretched or enlarged in the appropriate direction to match the distortion of the chamber walls. However, not all the cells stretched in every which way, or were recordable in all four chambers. By contrasting the changes in the place fields that were recorded in all four chambers to those that were not, O'Keefe and Burgess (1996) suggested that the distances of each of the walls were acting as determinants of the place fields' peak firing rates. They were able to model these distance determinants by fitting gaussian functions (like the normal bell curve) to the firing rates of the fields in the small square chamber, assuming that the peak firing rate at the centre of each field was composed of the sum of several curves, anchored to opposing walls at fixed distances.
Therefore, when these determinants and their gaussian distributions were pulled apart, the place field would stretch and the peak firing rate decrease as the centres of the gaussian distributions no longer overlapped to the same degree. If separated enough, theoretically, the single field would breakdown into two subfields, each firing at a subtraction of the summed rate, and thus, potentially beneath a recognizable threshold, which could lead to the loss of the place field altogether.
I have attempted to illustrate this idea in the figure above. In both the small square chamber on the left and on the horizontal rectangle shaped chamber on the right, the red and black curves represent the gaussian functions fit to the place field firing rate. The place field itself is designated by the yellow region encircled by the red and black determinants. As such, when the field is recorded from one chamber to the other, the two gaussian functions slide apart and stretch the place field in that plane.
As a final comment on these studies, although they provide a great deal of insight as to the nature of the computation that place cells may be involved with, in order to do so the units must be recorded in highly artificial and unrealistic environments.
Relations to theta/RSA
The complex-spike bursts are often linked to an EEG pattern known as Rhythmical Slow wave Activity (RSA) or Theta, a pattern which is endogenous to the hippocampus. Briefly, RSA is a sinusoidal wave form of regular amplitude with a modal frequency of about 6-9Hz in normal, awake, actively behaving rats. Thus, during most place work, a rat will exhibit RSA in its hippocampal electrocorticogram as it is running around the recording chamber.
At right is an RSA bearing hippocampal EEG epoch (green) with a rastor-style display of place cell firing from a single discriminated unit (red). The EEG and unit discharge amplitude are on different scales. If the data were properly scaled for amplitude, the single unit amplitude would be much smaller than the EEG waveform amplitude, which represents mass cell firing or oscillations. The EEG has also been heavily filtered through a 5-10Hz bandpass, such that waveforms with frequencies greater than 10Hz or less than 5Hz have had their amplitude severely diminished. The yellow dashed lines are pretty much set at the RSA amplitude limits. Thus, the portions of the EEG waveform that fall within the dashed limits, fall outside the defined RSA frequencies. Conversely, the portions of the EEG waveform that fall outside of the dashed limits, fall within the defined RSA frequencies.
There are three points to be made from this example.
First, the unit tends to fire when theta is present, and does not fire at the same rate when theta is absent.
Second, the complex bursts of the unit tend to discharge at a theta frequency, and is therefore entrained to the ongoing theta.
Finally, the unit discharges exclusively during the positive half of each theta cycle, an aspect of the complex-spike bursting and theta EEG's relationship known as "phase-locking" (
O'Keefe and Recce, 1993).
Relationships, such as phase-locking, are why continual attempts have been made to link theta to
higher cognitive functions, of which path-integration is the latest
(Knierim et al., 1996;
McNaughton et al.,1996).
Directional Tuning of Place Fields
Very much like complex-spike units can have a preference for firing in a specific place, they may further have a preference within that place for firing in a specific direction. McNaughton, Barnes and O'Keefe (1983) were one of the first to use a televideo tracking system to identify the rat's location while running around a radial maze. In doing so, they identified fields on individual arms that had different firing rates, dependent upon whether the rat was heading outward to the end of the arm or alternately heading inward to the center of the maze. As such, McNaughton et al (1983) concluded that place fields, apart from having location preferences, also had directional preferences or directional tunings. However, Muller et al. (1987) in their seminal work on quantifying place field characteristics, did not find a noticeable directional tuning of place fields recorded in their more open cylindrical chamber. Yet, tracking rats back and forth along a single runway O'Keefe and Recce (1993) did find the same polar directional differences in firing rate found earlier. In response, Muller et al. (1994) utilized the dual LED tracking system setup earlier that allowed a rat's directional "movement" to be separated from its directional "facing", as well as utilizing the earlier open circular recording chamber that allowed a rat's path greater variability when passing through fields than the earlier mazes and runways had allowed. Additionally, Muller et al. (1994) recorded from a radial maze as well.
Results indicated that in the open circular environment much of the directional tuning could be accounted for by the portion of the place field that the rat's path heading would take it through. In other words, certain path headings would tend to bias the rat to pass through portions of the place field with low firing rates, whereas other path headings would tend to bias the rat to pass through portions of the place field with high firing rates. When these biases were taken into account, expected firing rates did not differ greatly from observed firing rates. However, when the same procedures were performed on fields recorded on the radial-maze, expected firing rates differed sufficiently from observed firing rates to suggest that there was a true directional component to the place fields. Thus, Muller et al. (1994) supported both sets of findings.
In their model of gaussian distance determinants it is possible that O'Keefe and Burgess (1996) may have provided the explanation for these apparently contradictory results. Noting that when two opposing walls were separated enough to stretch a place field into having two firing peaks, each of those peaks had a polar directional preference that depended upon which wall determinant the rat was travelling away from. Conversely, when the walls were close enough together such that the two firing peaks were centered together, any directional preference was obscured. Therefore, in any open symmetrical recording chamber the potential for having equally compacted determinants is greater than on a long and drawn out rectangular runway. In the latter case, the critical set of determinants would be drawn out as to have a greater probability of having polar directional preferences.
The still undecided findings of hippocampal place field directional tunings is in sharp contrast to the well-defined directional tunings of nearby, and adjacent structures, of which some are directly downstream of hippocampal output
(Blair and Sharp, 1995;
Blair et al., 1997;
Mizumori, and Williams, 1993;
Sharp and Green, 1994;
Taube et al., 1990a;1990b;
Taube et al., 1996).
Cue Modality (not just vision)
Although much use has been made of visual cues in controlling place fields, place fields do not solely respond to visual cues alone. Early studies (O'Keefe and Dostrovsky, 1971; O'Keefe's, 1976) tended to use open recording rooms with their complex array of objects and recording equipment. Later studies attempting to identify critical cues, and their properties moved to highly simplified environments that emphasized peripheral or distal stimuli and de-empahsized local stimuli (Muller and Kubie, 1987; O'Keefe and Burgess, 1996; O'Keefe and Conway, 1978; Wiener et al., 1989). In fact, some studies have tried specifically to prevent the use of local non-visual cues, by making them unreliable markers, as when the detachable arms of the T-maze were continually swapped with each other (O'Keefe and Conway, 1978). Although much has been learned through the isolation and manipulation of distal visual stimuli, it has greatly diminshed what we understand of the of local and non-visual determinants of unit firing fields.
In response, Eichenbaum's laboratory has specifically sought after non-place correlates of complex-spike units, finding almost equal numbers of units that tended to have firing fields defined by local olfactory or texture cues, as defined by more distal stimuli (Weiner et al., 1989; Young et al., 1994; Wood et al., 1999). There appears to be no consistent, strict hierachy among the codings of complex-spike units, with all possible combination of individual cues, and cue constellations having a dynamic control over the firing fields (Shapiro et al., 1997 Tanila et al., 1997). As a final note to place fields' dependence upon distal visual cues, rats blind since birth appear to have normal place fields that can respond to cleaned out wine bottles placed on the periphery of a "Muller-type" recording cylinder (Save et al., 1998).
Place Field Mechanisms
Admittedly place fields are fascinating enough among themselves without even attempting to get at some of the underlying mechanisms. This is largely attested to by the fact that no real attempts were made for almost two decades after their discovery in 1970, and almost a decade after the critical discovery of the excitatory amino acid subreceptor n-methyl-d-aspartate (NMDA) as the critical underlying trigger mechanism for many forms of LTP
Links to Scientists who do Place Field and related Research
(last updated July 2007)
Place field References...
Hippocampus Forum 1991
Hippocampus Forum 1999
Blair HT., Liscombe, BW and Sharp, P.E. (1996). Anticipatory time intervals of head-direction cells in anterior thalamus of the rat: implications for path-integration in the heading direction circuit. Journal of Physiology, 15, 6260-6270.
Blair HT. and Sharp, P.E. (1995). Anticipatory head-direction cells in anterior thalamus: evidence for a thalamocortical circuit that integrates angular head motion to compute head direction. The Journal of Neuroscience, 15, 6260-6270.
Gray C.M., Maldonado P.E., Wilson M. and McNaughton B.L. (1995). Tetrodes markedly improve the reliability and yielf of multiple single-unit isolation from multi-unit recordings in cat striate cortex. Journal of Neuroscience Methods, 63, 43-54.
Fenton A. & Muller R.U. (1998). Place cell discharge is extremely variable during individual passes ofthe rat through the firing field. Proceedings of the National Academy of Scinces USA, 95, 3182-3187.
Fukada M., Kobayashi T., Bures J. and Ono T. (1992). Rat exploratory behavior controlled by intracranial self-stimulation improves the study of place cell activity. Journal of Neuroscience Methods, 44, 121-131.
Gothard KM. Skaggs WE. McNaughton BL. (1996). Dynamics of mismatch correction in the hippocampal ensemble code for space: interaction between path integration and environmental cues. Journal of Neuroscience. 16(24), 8027-40.
Harris, K.D., Henze, D.A., Csicsvari, J., Hirase, H. and Buzsaki, G. (2000). The accuracy of tetrode spike separation as determined by simultaneous intracellular and extracellular measurments. Journal of Neurophysiology, 84, 401-414.
Henze, D.A., Borhegyi, Z., Csicsvari, J., Harris, K.D. and Buzsaki, G. (2000). Intracellular features predicted by extracellular recordings in the hippocampus in vivo. Journal of Neurophysiology, 84, 390-400.
Klausberger, T., Magill, P.J., Marton, L.F., Roberts, J.D.B., Cobden P.M., Buzsaki, G. and Somogyi P. (2003). Brain-state and cell-type-specific firing of hippocampaal interneurons in vivo. Nature, 421 844-848.
Knierim JJ. Kudrimoti HS. McNaughton BL. (1996). Neuronal mechanisms underlying the interaction between visual landmarks and path integration in the rat. International Journal of Neural Systems. 7(2), 213-8.
Kubie, J.L., Muller, R.U., and Hawley, E.S. (1997). Single-cell recording in awake behaving rats. Chapter 18 In M.A Lynch and S.M. O'Mara (EDs), Neuroscience Labfax - Neurobiology Methodology, 269-281.
McNaughton BL. Barnes CA. Gerrard JL. Gothard K. Jung MW. Knierim JJ. Kudrimoti H. Qin Y. Skaggs WE. Suster M. Weaver KL. (1996). Deciphering the hippocampal polyglot: the hippocampus as a path integration system. Journal of Experimental Biology. 199 ( Pt 1), 173-85.
McNaughton, B.L., Barnes, C.A., and O'Keefe, J. (1983). The contributions of position direction and velocity to single unit activity in the hippocampus of freely-moving rats. Experimental Brain Research, 52, 41-49.
McNaughton BL, O'Keefe J, & Barnes CA, (1983). The stereotrode: a new technique for simultaneous isolation of several single units in the central nervous system. Journal of Neuroscience Methods, 8, 391-397.
Mishkin M., Speigler, B.J., Saunders, R.C., and Malamut, B.J. (1982). An animal model of global amnesia. In Alzheimer's disease: A Report of Progress Research, eds., S. Corkin, KL Davis, JH Growdon, EJ Usdin, RJ Wurtman, Raven Press, New York, 235-247.
Moita, et. al., 2003 Moita, M.A>P., Rosis, S., Zhou, Y., Ledouox, J.E., and Blair, H.T. (2003). Hippocampal place cells acquire location-specific responses to the conditioned stimulus during auditory fear conitioning. Neuron, 37, 485-497.
Olds, J. and Milner, P. (1954). Positive reinforcment produced by electrical stimulation of the septal area and other regions of the rat brain. Journal of comparative Physiological Psychology, 47,419-427.
Quirk, M.C., Blum, K.I. and Wilson, M.A. (2001). Experience-dependent changes in extracellular spike amplitude may reflect regulation of dendritic action potential back-propagation in rat hippocampal pyramidal cells. Journal of Neuroscience, 21, 240-248.
Tanila, H., Shapiro, M.L. and Eichenbaum, H. (1997). Discordance of spatial representation in ensembles of hippocampal place cells. Hippocampus, 7, 613-623.
Taube, J.S., Muller, R.U., and Ranck, J.B., Jr. (1990). Head-direction cells recorded from the postsubiculum in freely moving rats. I. description and quantitative analysis. The Journal of Neuroscience, 10, 420-435.
Taube, J.S., Muller, R.U., and Ranck, J.B., Jr. (1990). Head-direction cells recorded from the postsubiculum in freely moving rats. II. effects of environmental manipulations. The Journal of Neuroscience, 10, 436-447.
Whishaw IQ. Cassel JC. Jarrad LE. (1995). Rats with fimbria-fornix lesions display a place response in a swimming pool: a dissociation between getting there and knowing where. Journal of Neuroscience. 15(8), 5779-88.
Wilson M.A. and McNaughton B.L., (1994). Reactivation of hippocampal ensemble memories during sleep. Science, 265, 676-679.