I want to let you in on a little secret. We neuroscientists are actually quite jealous of the physicists. They may lament the fact that their unified theory of everything hasn’t turned up yet, but they’re sitting pretty with a bevy of universal laws, forces, constants and equations that do a bang up job explaining the universe. We neuroscientists are still on the hunt for some all-encompassing laws and principles that would explain brain function at a larger scale than the operation of single neurons (on which, I must say, we’ve done a pretty awesome job). However, a paper from Randy Bruno's lab at Columbia, published recently in the journal Science (Constantinople and Bruno, 2013), reminds us why it is so important not to get too comfortable with attractive unifying models of brain organization. Such ideas are compelling, because we certainly could use some simplifying principles to cut down the seemingly intractable complexity of neural circuits. They are dangerous, however, in their attractive simplicity, because they lead us to analyze data in the context of the simplifying model and (often unconsciously) ignore data that doesn't fit. As usual, Cajal said it best: “Unfortunately, nature seems unaware of our intellectual need for convenience and unity, and very often takes delight in complication and diversity” (Ramon y Cajal, 1906).
For many years, neuroscientists have worked to describe the "canonical" circuit for the neocortex, that outermost wrinkled layer of the cerebrum that has long been an enduring mystery of neuroscience. Cortex is a signature of our mammalian heritage, and a strong candidate for the neural structure most responsible for us humans being so damn clever. It is responsible for an artist's inspiration, a critic's discernment, and a scientist's reason. As might be expected of such an important brain structure, the organization of cortex is, well, complicated. Ramon y Cajal, who inked many beautiful inkings of the six-layered cortical structure in his day, once called it an "impenetrable jungle," and so it has proved for many generations of researchers.
In the hundred years since Cajal, neuroscientists have continued to feel their uncertain way around the dense undergrowth of the cortex with compass and machete, searching for patterns that might help explain how this entanglement of dozens of types of neurons stores your earliest childhood memories, appreciates the difference between a Degas and a Rembrandt, and enables you to understand this sentence.
One leading idea about the organization of the cortex is that it is composed of discrete, modular columns of cells, each of which is wired up in approximately the same way, and is responsible for a particular type of computation. This columnar hypothesis was introduced in the late fifties by Vernon Mountcastle (reviewed in Mountcastle, 1997), based on physiological recordings of neurons in monkey somatosensory (ie. touch-sensitive) cortex, where he noted that cells arranged in vertical columns relative to the cortical surface were sensitive to a particular type of tactile stimulation, while their horizontal neighbors were sensitive to different types of touch. In the 1960s, following Mountcastle's lead, Hubel and Wiesel made similar observations in the visual cortex of the cat, measuring neuronal responses to images of simple bars of light with different orientations (eg. horizontal, vertical, or tilted 45 degrees) projected on a screen. When they inserted their electrodes perpendicular to the surface of the cortex, the cells they recorded mostly responded to bars with the same orientation. However, if the angle of their electrode was off even a little, they observed cells with very different orientation preferences. In fact, the orientation preferences seemed to rotate as they moved their electrodes laterally in the cortex. This supported the idea that cells aligned in vertical columns are processing similar information about visual stimuli, while different types of stimuli are arranged in horizontal maps across the surface of the brain.
The notion of the column is deeply appealing to neuroscientists because it suggests that we could reduce the seemingly insurmountable complexity of cortical wiring to a regular array of identical columnar circuits arranged in parallel. Perhaps all of these columns have approximately the same circuit function – extracting some meaningful statistical structure from a set of input information – but each is responsible for processing a slightly different set of inputs. This makes the problem of understanding cortical wiring much easier: understand the circuit of one column and you understand them all. Therefore this idea has been embraced by many neuroscientists, particularly those with computational proclivities (eg. see George & Hawkins, 2009, Kaschube et al., 2010, or Henry Markram’s Blue Brain Project).
Of course, just because an idea is appealing doesn't mean it is true. The columnar hypothesis – at least in its strong form: that all of cortex is comprised of identical columnar modules – is still controversial. Some species, such as rodents, lack the clearly arranged orientation columns Hubel and Wiesel observed in cat visual cortex, and anatomical differences between brain regions and between species have cast some doubt on whether the proposed columnar modules are necessary for cortical function (eg. see Horton & Adams, 2005; Douglas & Martin, 2007; da Costa & Martin, 2010).
But whether or not all of cortex is organized into discrete columns, an important concept that has come out of – and to some extent replaced – the study of the anatomical column is the "canonical circuit" (da Costa & Martin, 2010) in which different cell types within the six-layered neocortex pass sensory information back and forth between them in a highly stereotyped fashion. These circuits may not be strictly organized into well-defined columns, but (as with the messier organization of orientation tuning in rodent visual cortex) intermingled and mixed together. (The neuroanatomist needs convenient maps, after all, not the brain itself.)
The organization of this canonical circuit has been most carefully studied in the rodent whisker cortex, where true columnar organization is perhaps clearest of any other brain region. Rodents, who spend much of their time in dark, cramped spaces, use their whiskers as you or I might use our fingertips if we found ourselves blindfolded in an unfamiliar room. Each whisker can detect important information about the locations, shapes and textures of objects in the environment. This information passes from sensory nerves in the whisker follicles through the animal’s brainstem to a deep brain structure called the thalamus, which is responsible for routing the information to somatosensory cortex, which handles all kinds of touch information from across the body.
Thalamic axons carrying information from individual whiskers form great clusters of synapses in layer 4 of a particular part of somatosensory cortex. These clusters are called "barrels" for their distinctive shape, and have given this region its nickname: "barrel cortex." Because of the specificity of thalamic inputs, barrels contain neurons which mainly respond to sensory information from individual whiskers, and form a map of the organization of whiskers on the animal's face.
These clearly organized columns have helped researchers pick apart the flow of information through the local neuronal circuit that processes information from a given whisker. A canonical circuit for the barrel cortex has been built up by scores of papers tracing neuronal axons and dendrites in fixed tissue, measuring physiological connection probabilities between living neurons of different types in brain slices, and probing functional neuronal responses to whisker stimulation in anesthetized and awake, behaving animals.
According to this model, the whisker-specific information arriving in layer 4 barrels spreads mostly vertically within the column to cells in superficial layers 2 and 3, where it is processed a bit and may be shared by horizontal transmission to neighboring cortical regions. The information then flows down from superficial to deep cortical layers to activate output neurons in layer 5, which send out columnar "conclusions" about the sensory information being processed to the rest of the brain.
Alright, now that I have carefully explained our delight in having formulated this very sensible and functional canonical circuit for barrel cortex, I would like to share a recent paper which casts doubt on the whole thing.
In the course of a separate study on experience-dependent changes in the anatomy of the thalamic axons that feed into barrel cortex (Oberlaender et al, 2012), Columbia University’s Randy Bruno noticed that the thalamic axons he and his students had labeled histologically actually seemed to make connections in two locations within the cortical column. There was the large cluster of connections in the layer 4 barrels, of course, but also a layer of fine processes that branched off and seemed likely to contact cells deep in layer 5, the cortex's output layer.
Curious how these deep-layer thalamic axons might influence the activity of layer 5 cells, Bruno’s student Christine Constantinople followed up on these observations by recording from neurons in different layers of the barrel cortex of sedated rats while simultaneously stimulating the whiskers. She and her colleagues were puzzled to note that many layer 5 neurons responded to whisker stimulation just as quickly as the layer 4 neurons that are supposed to be the gateway for inputs to the cortex. Contrary to the standard "canonical" circuit, this timing data implied that layer 5 neurons were responding directly to thalamic input, rather than waiting for the information to pass all the way through the circuit from L4 to L2/3 and back to L5.
In order to test how much of layer 5 activity was actually coming directly from thalamic sensory inputs and how much was dependent on the rest of the columnar circuit, Constantinople et al. used a venerable physiologist's trick. They injected lidocaine (the stuff the dentist uses to numb your gums) into layer 4, where it not only shut down the activity of layer 4 neurons, but blocked all transmission of information from the upper layers of cortex. (This is because lidocaine blocks the voltage-sensitive sodium channels that are required for transmission of neuronal signals along axons.)
To the shock (I suspect) of the researchers, blocking the entire upper half of the column hardly affected the layer 5 responses to whisker stimulation at all. This meant (following a number of control experiments & tests of alternative hypotheses) that many layer 5 "output" neurons were in fact not dependent on the canonical circuit at all for their sensory responses, but seemed to be perfectly capable of detecting whisker stimulation through their own private contacts from the thalamus, without relying on inputs from the other layers.
I have heard Bruno speak about these results, and his interpretation, which seemed partly designed to distress and agitate his audience, but also largely in earnest, was that we have been wrong about the canonical circuit all along. There are two circuits in cortex, he claimed, a superficial one driven by inputs to layer 4, and a deep one driven by direct inputs to layer 5. In the Science paper, he states his conclusions quite clearly:
"Neocortical columns may contain two separate processing systems or “strata”: an upper stratum (L4 and L2/3) and a lower stratum (L5/6) possibly subserving different functions. …. L2/3 targets other neocortical regions, whereas L5/6 targets both cortical and subcortical structures, [many of which], especially those of L5, are action-related (striatum and spinal cord) or high-order (secondary thalamic nuclei, which innervate high-order cortical regions). Both strata therefore have direct access to the same sensory information and can alter behavior via different anatomical pathways." (Constantinople and Bruno, 2013)
These new findings do not overturn what we knew about cortical circuits. They merely add another wrinkle to that wrinkliest of questions – how does the cortex work? For instance, does it matter that the animals were sedated and not actively using their whiskers to explore their environment? Perhaps the layer 5 cells are attuned to detect surprising whisker contacts quickly and directly when the animal is quiescent, but require the superficial layers for more complex processing of objects and their shapes and textures. As always, much to be done, many new questions to be asked.
However, to my mind, one of the most remarkable things about this whole story is that these layer 5 inputs from thalamus were actually old news. They have been observed many times by histological staining and are even included (usually as a dashed line labeled "modulatory") in many circuit diagrams. You can see them yourself in the images of thalamic axons entering barrel cortex:
In fact, thalamic axons were already known to make synaptic contacts with deep layer 5 neurons (Petreanu et al., 2008, Meyer et al., 2010). Indeed, Bruno was even an author of an earlier paper that had already made the observation that some of these deep 5 pyramidal neurons respond to whisker stimulation just as quickly as layer 4 neurons (de Kock et al. 2007).
Constantinople and Bruno's experiments demonstrating that layer 5 neurons can function decoupled from the rest of the canonical circuit were extremely elegant, but in retrospect not as conceptually shocking as they might seem. In fact, what is surprising is not that we missed these connections, but rather that we saw them and, for the most part, ignored them – they did not fit into the simplifying model that we had built to help us understand the circuit.
In addition to requiring some textbook revisions, this paper should remind neuroscientists to take the time to step back and revel in cortical complexity. There might be something in that chaos that pulls off the blindfold of the everyday and reveals some new order.
Ramón y Cajal, S. The structure and connexions of neurons. Nobelprize.org [online], http://nobelprize.org/ medicine/laureates/1906/cajal-lecture.pdf (1906).
Constantinople, Christine M., and Randy M. Bruno. “Deep Cortical Layers Are Activated Directly by Thalamus.” Science 340, no. 6140 (June 27, 2013): 1591–1594. doi:10.1126/science.1236425.
Da Costa, Nuno Maçarico, and Kevan A C Martin. “Whose Cortical Column Would That Be?” Frontiers in Neuroanatomy 4, no. May (January 2010): 16. doi:10.3389/fnana.2010.00016.
De Kock, C P J, R M Bruno, H Spors, and B Sakmann. “Layer- and Cell-type-specific Suprathreshold Stimulus Representation in Rat Primary Somatosensory Cortex.” The Journal of Physiology 581, no. Pt 1 (May 15, 2007): 139–54. doi:10.1113/jphysiol.2006.124321.
Douglas, Rodney J, and Kevan a C Martin. “Mapping the Matrix: The Ways of Neocortex.” Neuron 56, no. 2 (October 25, 2007): 226–38. doi:10.1016/j.neuron.2007.10.017.
George D, Hawkins J (2009) Towards a Mathematical Theory of Cortical Micro-circuits. PLoS Comput Biol 5(10): e1000532. doi:10.1371/journal.pcbi.1000532
Horton, Jonathan C, and Daniel L Adams. “The Cortical Column: a Structure Without a Function.” Philosophical Transactions of the Royal Society of London. Series B, Biological Sciences 360, no. 1456 (April 29, 2005): 837–62. doi:10.1098/rstb.2005.1623.
Kaschube, Matteo, Michael Schnabel, Siegrid Löwel, David M. Coppola, Leonard E. White,and Fred Wolf. Universality in the Evolution of Orientation Columns in the Visual Cortex.Science 19 November 2010: 330(6007), 1113-1116. Published online 4 November 2010[DOI:10.1126/science.1194869]
Meyer, Hanno S, Verena C Wimmer, Mike Hemberger, Randy M Bruno, Christiaan P J de Kock, Andreas Frick, Bert Sakmann, and Moritz Helmstaedter. “Cell Type-Specific Thalamic Innervation in a Column of Rat Vibrissal Cortex.” Cerebral Cortex (New York, N.Y. : 1991) (June 2010): 1–17. doi:10.1093/cercor/bhq069.
Mountcastle, V B. “The Columnar Organization of the Neocortex.” Brain : a Journal of Neurology 120 ( Pt 4) (April 1997): 701–22. http://www.ncbi.nlm.nih.gov/pubmed/9153131.
Oberlaender, Marcel, Alejandro Ramirez, and Randy M Bruno. “Sensory Experience Restructures Thalamocortical Axons During Adulthood.” Neuron 74, no. 4 (May 24, 2012): 648–55. doi:10.1016/j.neuron.2012.03.022.
Petreanu, Leopoldo, Tianyi Mao, Scott M Sternson, and Karel Svoboda. “The Subcellular Organization of Neocortical Excitatory Connections.” Nature 457, no. 7233 (February 2009): 1142–5. doi:10.1038/nature07709.
Wimmer, Verena C, Randy M Bruno, Christiaan P J de Kock, Thomas Kuner, and Bert Sakmann. “Dimensions of a Projection Column and Architecture of VPM and POm Axons in Rat Vibrissal Cortex.” Cerebral Cortex (New York, N.Y. : 1991) (May 2010). doi:10.1093/cercor/bhq068.