Do Fruit Flies Dream of Piña Coladas?

The olfactory learning circuitry of the fly brain.

Our brains didn't come from nowhere, but rather out of hundreds of millions of years of evolutionary work that developed mechanisms for dealing rapidly and intelligently with the environment. Primitive organisms are fascinating to study as waystations along that long road, and their simplicity can clarify what in humans may still be beyond our intellectual or technical reach to study.

Fruit flies have only ~ 1-200,000 cells in their brains, compared with the maybe 100 billion in humans. Moreover, most of these neurons are laid down in totally genetically determined fashion, easing study, but also reducing the ability to learn. But one area of the fly brain appears to break that mold- the mushroom body, which is intermediate in the path from olfactory reception to behavior. It is sort of the fly's glimmer of higher intelligence. Indeed, flies can be trained in the traditional Pavlovian manner to associate otherwise meaningless or pleasant odors with torture like electric shocks, and thus avoid them. Researchers have spent decades teaching them other tricks like avoiding certain odors, heading towards sounds, lights, manipulating social status and fighting behavior, etc. Along the way they have given some of their mutants names like dunce and rutabaga.

Location of the mushroom body in the fly head/brain. Signals arrive through the antennal lobes, from the various hairs and other nerves on the mouth and antenna.

Fluorescence image of two labels, one on an annal lobe (AL) neuron (red), and another on a mushroom body (MB) neuron (green, left side), which labels more cells of the same type on the right side. The lateral horn (LH) is one destination for signals as they start going out again to muscles and behavior downstream.

A recent paper conducted a tour de force of anatomy, tracing every single neuron going to and from the mushroom body. The technique they used to do this is interesting in itself, called an "enhancer trap". Fly researchers have been generating a vast number of "lines", or inbred fly mutants, by inserting a two bits of DNA from yeast cells. The first is the gene encoding a transcription activator, GAL4. This is induced to jump randomly in the fly genome, hoping that lands downstream of the regulatory region of an endogenous gene, i.e. its enhancer or promoter. The second bit is a binding site for this GAL4 protein, linked to a gene that expresses some useful marker, typically a fluorescent protein like GFP. Since the yeast GAL4 protein works just fine to activate RNA transcription and gene expression in flies, the end result is that GFP gets expressed in reponse to a single enhancer somewhere else in the genome. Indeed researchers try to "saturate the genome", generating a huge number of lines with such mutations, hoping for mutants at every single enhancer in the fly genome, and even enslave their undergraduate students to that end.

Enhancer trap schematic. An introduced regulatory gene, GAL4 is hopped within the fly genome to random locations, some of which are downstream of enhancers (E1). The signal is received by another introduced gene, which expresses a marker (green fluorescent protein X in this case) in response to the production of GAL4. UAS = upstream activating sequence.

Screening with a fluorescence microscope, one will see the cells where GFP is expressed, and thus where the particular enhancer of that line is active. This might be anywhere and any time, in the egg or in the adult, in the brain or in the leg, or everywhere and all the time, or nowhere. This stage of the process tends to be very tedious, as is all the fly breeding leading up to it. The current researchers used 7000 such lines that had been built previously to screen for those showing GFP expression in neurons in or connected to the mushroom body.

Anatomical details of neurons leading through, and from the mushroom body, drawn from studies of many flies with many individually labelled neurons. Video here. Note the wide distribution of MBON (mushroom body output) neuron connections, and also the density of DANs, (dopaminergic neurons), which feedback to the MBONs from sensations elsewhere in the body.

The mushroom body is made up of ~ 2,000 of what are called Kenyon cells. Inputs come from the antennal lobe, where olfactory receptors from the fly's "nose" and face link to ~200 projecting neurons (diagrammed below). The Kenyon cells synapse to what are called the mushroom body output neurons (MBONs), and these same output axons get inputs both from other MBONs (making some recurrent loops) and from other neurons called DANs, which seem to be crucial for the feedback that leads to learning.

Wiring diagram of the olfactory learning system. Projection neurons (left) come from the nose and face, to the Kenyon cells of the mushroom body (KC). Then MBONs of various types (using different neurotransmitters in their synapses) come out (gray lines) to innervate the lateral horn and other downstream neurons. DAN neurons from other sensations are labelled for their valence. The colored boxes correspond to various anatomical bodies or sub-areas.

The key part of the organization is that while the layout and connections of the olfactory neurons and projection neurons are genetically determined, those of the Kenyon cells are not. They are a generic type of cell whose sporadic connections on both ends are made on the fly (sorry!) during development, and are not the same from one fly to the next, as are all the other cells. These connections are also plastic during adulthood, and the volume of the mushroom body on a gross level expands in response to usage in honey bees. The mushroom body is not essential to actions of the fly, really- the built-in programs for sensations and stereotypical behaviors lie elsewhere. The mushroom body system only biases those behaviors based on learned feedback, which arrives via the DAN neurons from positive and negative shocks / experiences.

This arrangement is pretty much what has long been called a "neural network", which is a computer science tool developed over several decades by analogy to how people thought neural systems work. They offer the unique capacity to learn from sample data, and to solve problems that are not well-specified or are complex. These feature a hidden array (or black box) of many "neurons", connecting inputs with outputs. The box is trained (iteratively) by providing it with (known) sample input and output data. Feedback of error values from the output vs input comparison is sent back through the network and adjusts the weights of those connections which then evolve through time, in their connections and connection strength, leading the network to arrive (slowly) at an approximation of the behavior desired, emitting the correct output for given inputs.

For instance, an image shape recognition program might be made as a neural network, with many examples fed in, and the outputs judged and signals sent back into the network to either reinforce connections that improve the match over previous trials, or weaken connections that reduce it. If a sufficiently broad range of input images are used, then the network stands a good chance of identifying those same shapes in images it has never "seen" before in training.

"... MBs are a paradigmatic case of reused neural networks in action."

"This model was built in a connectionist manner, obeying, although in a scaled version, the neurobiological topology. The model was initially built for showing basic learning and conditioning capabilities; subsequently it was found able to show other interesting behaviors, like attention, expectation, sequence learning, consolidation during sleep and delayed-matching-to sample tasks."

Likewise, in the mushroom body system, the DANs provide the essential feedback in real-life situations as the fly buzzes and walks about, smelling the wonderful world. If they can dampen neural connections among the Kenyon cells and their downstream targets that lead to painful results, and juice up those that lead to pleasure, we have a smarter, and more successful fly.

"The identification of the full complement of 21 MBON types highlights the extensive convergence of 2000 KCs onto just 34 MBONs, a number even smaller than the number of glomeruli in the AL. Thus, the high-dimensional KC representation of odor identity is transformed into a low-dimensional MB output. This suggests that the MBONs do not represent odor identity but instead provide a representation that may bias behavioral responses."

So, what is it like to be a fly? Are flies conscious? They are clearly responsive to their environment, and have what we would call "experiences", such as hunger, searches for food, mating, etc., and one would assume these experiences can be very intense. So I think it would be hard to count them entirely out of the consciousness department. But it would have to be an extremely small consciousness, with little association to past, let alone to future events, metaphorical, or conceptual abstractions. But feeling- that is likely to be there in some form.

• One problem with the rich ... they are well-hidden, even when they exercise their "free" speech.
• Martin Wolf on the injustice of housing shortages and zoning control by incumbents. It is the price of overpopulation, really.
• Everyone's a critic!
• Atheism- a little more dangerous than you might think.
• The economics of knowledge & organization.
• Exploitation and income.
• Want to know about a government boondoggle?
• California raises the bar on environmental responsibility.
• On the tension between science and humanity in economics.
• The sanctimonious disaster that is Jeffrey Sachs.
• On the power of satire & truth: "... the Chinese government decided to ban puns."
• In France, there is a culture war. It was not about "freedom", but about cultural dominance.
• Dirty jobs- better than clean jobs?