Essays on Science and SocietyEppendorf

Brains Don't Play Dice—or Do They?

See allHide authors and affiliations

Science  01 Nov 2013:
Vol. 342, Issue 6158, pp. 574
DOI: 10.1126/science.1245982

Life is often unpredictable. The brain thus needs to draw on past experience to prepare for future events. It does so by associating sensory stimuli, such as an odor, with desirable or undesirable outcomes. The smell of burning wood, for example, can evoke the soothing recollection of a cozy campfire or the traumatic memory of a burning house. It is through such associations that sensory stimuli that are a priori neutral become endowed with meaning. But life is full of possibilities. So how are associative brain centers able to account for an infinite number of possible associations?

Using the olfactory system of the fruit fly Drosophila melanogaster, I tackled this question in the laboratory of Richard Axel at Columbia University (1). The Drosophila olfactory circuit is similar to that of vertebrates in terms of its design logic, but consists of far fewer neurons. Fruit flies can learn to associate an odor with an electric shock or a sugar reward. On the basis of these associations, they will adjust their behavior when encountering that odor again. Such olfactory associative memories are formed in the mushroom body, a brain center comprising about 2000 neurons (known as Kenyon cells) and receiving mainly olfactory input (2, 3).

Mapping mushroom body connections.

(A) Schematic of the D. melanogaster olfactory circuit. All olfactory sensory neurons expressing the same odorant receptor (like colors) converge in one of the 51 antennal lobe glomeruli (dotted circles). One or more projection neurons connect each glomerulus to the lateral horn (not shown) and the mushroom body, where associative memories are formed. How projection neurons are wired to Kenyon cells was unknown. (B and C) Mapping Kenyon cell inputs (Scale bars, 10 µm). (C) Starting with a single photolabeled Kenyon cell (white arrowhead), a projection neuron (red) connected to one of the dendritic endings of that Kenyon cell (red arrowhead) was filled with dye. (B) The innervation pattern of the red-labeled projection neuron in the antennal lobe allows for the identification of the corresponding glomerulus (arrowhead). (D) Matrix of 654 connections between 51 glomeruli and 200 Kenyon cells. Each row corresponds to the glomerular inputs of a single Kenyon cell, and each column represents one of the antennal lobe glomeruli. Red bars indicate a connection between a Kenyon cell and a glomerulus. Yellow bars indicate two connections to the same glomerulus. Differences in the number of connections per glomerulus are a function of the number of synaptic endings of their projection neurons. Statistical analyses revealed a random pattern of connectivity.

Fruit flies detect odors by means of olfactory sensory neurons located on their antennae and maxillary palps. Most of these neurons express only one of about 50 odorant receptor proteins (47). The olfactory sensory neurons project to the antennal lobe, where they are sorted such that neurons expressing the same receptor converge onto one of the 51 glomeruli (6, 7). This wiring diagram (see the figure), in which each odorant receptor is mapped onto one glomerulus, prefigures how odors are represented in the antennal lobe. When an odorant molecule binds to its cognate receptors, the sensory neurons expressing these receptors fire, and an odor-specific pattern of glomerular activity emerges, a so-called “odor-evoked map” (8). Projection neurons dedicated to each glomerulus then relay olfactory information to two higher brain centers: the lateral horn, which mediates innate behaviors, and the mushroom body, where associative olfactory memories are formed (9, 10).

Eppendorf and Science are pleased to present the essay by Sophie Caron, a 2013 finalist for the Eppendorf and Science Prize for Neurobiology.

How does the mushroom body represent odors such that olfactory associations can emerge? We reasoned that, much like in the antennal lobe, the wiring diagram might hold the key. Together with my colleagues at Columbia University, I developed a technique to identify the input of individual Kenyon cells. Kenyon cells were labeled in order to visualize their synaptic endings, revealing that each Kenyon cell is connected to, on average, seven projection neurons. Subsequently, between two to seven of the connected projection neurons were marked with a neural tracer such that their cognate glomeruli could be identified. We charted 654 connections of 200 Kenyon cells and assigned them to the 51 antennal lobe glomeruli (11). Some glomeruli are represented more often than others, but this skewed frequency is simply a consequence of the number of presynaptic endings specific to the different projection neurons.

Within this set of connections, we then searched for any discernible pattern that might predict how glomeruli connect to the Kenyon cells. We first tested whether there are certain pairs of glomeruli that preferentially converge on a Kenyon cell, but could not find any preferred pairings. Likewise, there are no Kenyon cells dedicated to a specific glomerulus. Next, we searched for any biological criteria, such as glomerulus position, odor-response profile, or developmental origin, by which one input on a Kenyon cell would predict its other inputs. We statistically tested more than 10 different criteria but could not detect any such predictive pattern. Instead, the connections appeared to be completely random (11).

To confirm this notion, we created a data set in which the relative frequency of glomerular connections remained the same, but the Kenyon cells that they connect to were randomly shuffled. This shuffled data set behaved the same as our observational data set in all statistical tests. The lack of any discernable pattern in the connections between glomeruli and Kenyon cells means that each Kenyon cell receives its input from an essentially random set of glomeruli (11), which is in line with the odor responses of an individual Kenyon cell varying between individual flies (12).

This result might seem puzzling: Why would the ordered, odor-evoked map established in the antennal lobe be randomized in the mushroom body? The answer may lie in the way odors are represented there: Any given odor activates only about 5% of the Kenyon cells, regardless of the number of active glomeruli (13). To understand how random connectivity affects the formation of these sparse representations, we devised a computational model of the fly olfactory circuit in collaboration with Larry Abbott. The model reveals that randomization of Kenyon cell input allows the mushroom body to generate activity patterns that overlap minimally between different odors. In fact, as soon as any order is imposed on the connections between projection neurons and Kenyon cells, the resulting activity patterns begin to overlap and render the system less able to form odor-specific associations. Counterintuitive as it may seem, randomization appears to be an advantageous strategy for forming associations, because it minimizes the similarity between different sensory representations.

The randomization of inputs we discovered in the Drosophila mushroom body may be a general principle of neural circuit architecture in associative brain centers. In the vertebrate cerebellum, which processes sensorimotor information, granule cells, like Kenyon cells, also receive a limited number of inputs. David Marr and James Albus postulated that, owing to their limited inputs, the granule cells encode different motor patterns as sparse representations; and to accommodate more representations in the coding space defined by granule cells, they proposed that granule cell input would have to be randomly organized (14).

Our characterization of Kenyon cell connections in Drosophila provides the first anatomical evidence for randomized sensory input in an associative brain center. Randomization may be a fundamental strategy to represent the vast number of possible associations with a limited number of neurons, it might be the very embodiment of the unpredictability of life.


Sophie Caron is currently a postdoctoral fellow in the Department of Neuroscience at Columbia University. Sophie grew up in St-Blaise-sur-Richelieu in Canada and earned a B.Sc. in Biochemistry at the Université de Montréal. She moved to New York City to study the developmental mechanisms behind the diversification of sensory neurons in the laboratory of Dr. Alexander Schier at New York University and, later, Harvard. Having completed her Ph.D., Sophie joined the laboratory of Dr. Richard Axel at Columbia University, where she studies how the information gathered through the senses is represented in higher brain centers, in particular those involved in memory.


For the full text of finalist essays and for information about applying for next year's awards, see Science Online at

References and Notes

  1. Acknowledgments: I thank V. Ruta, L. F. Abbott, and R. Axel for our collaboration.

Navigate This Article