Alexander Borst

Information theory and neural coding

Information theory quantifies how much information a neural response carries about the stimulus. This can be compared to the information transferred in particular models of the stimulus–response function and to maximum possible information transfer. Such comparisons are crucial because they validate assumptions present in any neurophysiological analysis. Here we review information-theory basics before demonstrating its use in neural coding. We show how to use information theory to validate simple stimulus–response models of neural coding of dynamic stimuli. Because these models require specification of spike timing precision, they can reveal which time scales contain information in neural coding. This approach shows that dynamic stimuli can be encoded efficiently by single neurons and that each spike contributes to information transmission. We argue, however, that the data obtained so far do not suggest a temporal code, in which the placement of spikes relative to each other yields additional information.

Drosophila Mushroom Body Mutants are Deficient in Olfactory Learning

Two Drosophila mutants are described in which the connections between the input to and the output from the mushroom bodies is largely interrupted. In all forms of the flies (larva, imago, male, female) showing the structural defect, olfactory conditioning is impaired. Learning is completely abolished when electroshock is used as reinforcement and partially suppressed in reward learning with sucrose. No influence of the mushroom body defect on the perception of the conditioning stimuli or on spontaneous olfactory behavior is observed. The defect seems not to impair learning of color discrimination tasks or operant learning involving visual cues.

Principles of visual motion detection

Motion information is required for the solution of many complex tasks of the visual system such as depth perception by motion parallax and figure/ground discrimination by relative motion. However, motion information is not explicitly encoded at the level of the retinal input. Instead, it has to be computed from the time-dependent brightness patterns of the retinal image as sensed by the two-dimensional array of photoreceptors. Different models have been proposed which describe the neural computations underlying motion detection in various ways. To what extent do biological motion detectors approximate any of these models? As will be argued here, there is increasing evidence from the different disciplines studying biological motion vision, that, throughout the animal kingdom ranging from invertebrates to vertebrates including man, the mechanisms underlying motion detection can be attributed to only a few, essentially equivalent computational principles. Motion detection may, therefore, be one of the first examples in computational neurosciences where common principles can be found not only at the cellular level (e.g., dendritic integration, spike propagation, synaptic transmission) but also at the level of computations performed by small neural networks.

A genetically encoded calcium indicator for chronic in vivo two-photon imaging

Neurons in the nervous system can change their functional properties over time. At present, there are no techniques that allow reliable monitoring of changes within identified neurons over repeated experimental sessions. We increased the signal strength of troponin C–based calcium biosensors in the low-calcium regime by mutagenesis and domain rearrangement within the troponin C calcium binding moiety to generate the indicator TN-XXL. Using in vivo two-photon ratiometric imaging, we show that TN-XXL exhibits enhanced fluorescence changes in neurons of flies and mice. TN-XXL could be used to obtain tuning curves of orientation-selective neurons in mouse visual cortex measured repeatedly over days and weeks. Thus, the genetically encoded calcium indicator TN-XXL allows repeated imaging of response properties from individual, identified neurons in vivo, which will be crucial for gaining new insights into cellular mechanisms of plasticity, regeneration and disease.

How does Nature Program Neuron Types

Neurons come in different flavors. Once a recording is established, the electrophysiological properties, either in response to current injection or in response to sensory stimulation, are found to be characteristic for each neuron type, very much the same as their typical dendritic anatomy or axonal projection. In invertebrates, these statements even account for individual neurons occurring only once in each hemisphere of a ganglion. While these response characteristics have long been thought to be due to a typical composition of membrane currents, characteristic for each neuron type and constant across the different members of this type, this view has been challenged by a series of investigations performed on various neurons (beautifully reviewed in Marder and Goaillard, 2006). One of these studies was performed on the stomatogastric ganglion of the crab. This ganglion consists of only a handful of neurons, together forming two oscillatory networks one controlling the gastric mill and the other the pyloric rhythm of the animal. Each of the neurons can be named according to its invariant electrophysiological response properties: some of them are endogenously bursting, while others are passively following the rhythm of the others. Looking at the different membrane currents underlying the bursting behavior in the same neurons in different individuals, the surprising finding was that every neuron seemed to be bursting by a different mechanism: while in all of them the same basic set of membrane currents was present, their pattern of maximal ion conductances was found to widely differ in each case. Yet: all of them were bursting in an almost indistinguishable mode. Not too astonishingly, the amount of mRNA encoding for the membrane channels underlying the different currents was found to vary along with the maximum conductance values measured electrophysiologically (Schulz et al, 2006). Along similar lines, the work of Turrgiano, Nelson and colleagues (Turrigiano et al, 1998) has shown that synaptic strength is another parameter that is regulated such as to remain in an operating range appropriate to neuronal function. What does this mean? Although never stated explicitly in this way, everyone in the field would probably think that a neurons functional identity is the result of the expression of some sort of cell-specific transcription factor which in turn controls the expression of a set of ion channels in a cell-specific relative amount to each other. This way, the neuron would reveal its typical shape of action potential, its typical current-spike frequency relationship and other characteristics like plateau potential or bursting mode. This would also be the way that every biophysical neuron-model is set up (e.g. Wicher et al, 2006) with the maximum conductances of all the different currents being defined in the parameter file executed upon initializing the simulation software. However, this does not seem to be the way that nature programs it. Instead, in real neurons, something else seems to be defined as characteristic for each neuron, and this is its overall behavior (its electrical “character”). From the studies mentioned above, one can conclude that neurons are given a goal like being ‘bursting’ and each neuron has to find a combination of maximum conductances that does the job. Once this point (the ‘set point’) is reached, the neuron is stable and keeps on expressing the ion channels in this fixed relative amount. When thinking about it in programming terms, this would be identical to parameter fitting in a model simulation: the task is to find the minimum of the error function between the desired output (the ‘target’) and the actual output of the model (e.g. Borst and Haag, 1996; Druckmann et al, 2007). This immediately poses a lot of questions: how on earth is the target function defined genetically? How is the actual output function measured by the neuron, and how is the error between target and the actual firing pattern measured? How is the minimum recognized? Obviously, no one today has an idea. Most neuroscientists will readily agree that ‘homeostasis’ is at the heart of all biological phenomena. But, somehow it has been largely ignored in most of the studies in cellular neuroscience. On the other hand, the relevance of this design principle of a fixed and cell-specific set-point can hardly be overestimated. How else could it be that drugs prescribed to make up for the loss of transmitter in a locally restricted area of the brain are delivered systemically, affecting all receptors in all neurons that express them in the brain? Just think of what would happen if, in a network simulation with biophysically realistic neuron models, one altered the baseline activation of all GABA receptors by a factor of 5 or 10? Would the network still be functional? Of course it will not. But this is the difference between the real nervous system and the way we understand and model it at the moment: in the real nervous system, those neurons lacking GABAergic input due to e.g. neurodegeneration, will take the opportunity of the additional GABA to happily return to their set-point, while all others driven away from their set point by the medication, will find some means to homeostatically go back to their appropriate set-point after a while. Understanding these phenomena at the cellular level, thus, will lead not only to a deep understanding of how a neuronal identity is genetically programmed, it will also lead to an understanding of plasticity in a much more general way as the basic way that neurons are built and guaranteed to robustly perform in a largely unpredictable environment.

Computational structure of a biological motion-detection system as revealed by local detector analysis in the fly's nervous system

The computations performed by individual movement detectors are analyzed by intracellularly recording from an identified direction-selective motion-sensitive interneuron in the flys brain and by comparing these results with model predictions based on movement detectors of the correlation type. Three main conclusions were drawn with respect to the movement-detection system of the fly: (1) The essential nonlinear interaction between the two movement-detector input channels can be characterized formally by a mathematically almost perfect multiplication process. (2) Even at high contrasts no significant nonlinearities seem to distort the time course of the movement-detector input signals. (3) The movement detectors of the fly are not perfectly antisymmetrical; i.e., they respond with different time courses and amplitudes to motion in their preferred and null directions. As a consequence of this property, the motion detectors can respond to some degree to stationary patterns whose brightness is modulated in time. Moreover, the direction selectivity, i.e., the relative difference of the responses to motion in the preferred and null directions, depends on the contrast and on the spatial-frequency content of the stimulus pattern.

Fly Motion Vision

Fly motion vision and resultant compensatory optomotor responses are a classic example for neural computation. Here we review our current understanding of processing of optic flow as generated by an animals self-motion. Optic flow processing is accomplished in a series of steps: First, the time-varying photoreceptor signals are fed into a two-dimensional array of Reichardt-type elementary motion detectors (EMDs). EMDs compute, in parallel, local motion vectors at each sampling point in space. Second, the output signals of many EMDs are spatially integrated on the dendrites of large-field tangential cells in the lobula plate. In the third step, tangential cells form extensive interactions with each other, giving rise to their large and complex receptive fields. Thus, tangential cells can act as matched filters tuned to optic flow during particular flight maneuvers. They finally distribute their information onto postsynaptic descending neurons, which either instruct the motor centers of the thoracic ganglion for flight and locomotion control or act themselves as motor neurons that control neck muscles for head movements.

Transient and steady-state response properties of movement detectors

The transient and steady-state responses of movement detectors are studied at various pattern contrasts (i) by intracellularly recording from an identified movement-sensitive interneuron in the flys brain and (ii) by comparing these results with computer simulations of an array of movement detectors of the correlation type. At the onset of stimulus motion, the membrane potential oscillates with a frequency corresponding to the temporal frequency of the stimulus pattern before it settles at its steady-state level. Both the transient and the steady-state response amplitudes show a characteristic contrast dependence. As is shown by computer modeling, the transient behavior that we found in the experiments reflects an intrinsic property of the general scheme of movement detectors of the correlation type. To account for the contrast dependence, however, this general scheme has to be elaborated by (i) a subtraction stage, which eliminates the background light intensity from the detector input signal, and (ii) saturation characteristics in both branches of each movement-detector subunit.

In Vivo Performance of Genetically Encoded Indicators of Neural Activity in Flies

Genetically encoded fluorescent probes of neural activity represent new promising tools for systems neuroscience. Here, we present a comparative in vivo analysis of 10 different genetically encoded calcium indicators, as well as the pH-sensitive synapto-pHluorin. We analyzed their fluorescence changes in presynaptic boutons of the Drosophila larval neuromuscular junction. Robust neural activity did not result in any or noteworthy fluorescence changes when Flash-Pericam, Camgaroo-1, and Camgaroo-2 were expressed. However, calculated on the raw data, fractional fluorescence changes up to 18% were reported by synapto-pHluorin, Yellow Cameleon 2.0, 2.3, and 3.3, Inverse-Pericam, GCaMP1.3, GCaMP1.6, and the troponin C-based calcium sensor TN-L15. The response characteristics of all of these indicators differed considerably from each other, with GCaMP1.6 reporting high rates of neural activity with the largest and fastest fluorescence changes. However, GCaMP1.6 suffered from photobleaching, whereas the fluorescence signals of the double-chromophore indicators were in general smaller but more photostable and reproducible, with TN-L15 showing the fastest rise of the signals at lower activity rates. We show for GCaMP1.3 and YC3.3 that an expanded range of neural activity evoked fairly linear fluorescence changes and a corresponding linear increase in the signal-to-noise ratio (SNR). The expression level of the indicator biased the signal kinetics and SNR, whereas the signal amplitude was independent. The presented data will be useful for in vivo experiments with respect to the selection of an appropriate indicator, as well as for the correct interpretation of the optical signals.

ON and OFF pathways in Drosophila motion vision

Motion vision is a major function of all visual systems, yet the underlying neural mechanisms and circuits are still elusive. In the lamina, the first optic neuropile of Drosophila melanogaster, photoreceptor signals split into five parallel pathways, L1–L5. Here we examine how these pathways contribute to visual motion detection by combining genetic block and reconstitution of neural activity in different lamina cell types with whole-cell recordings from downstream motion-sensitive neurons. We find reduced responses to moving gratings if L1 or L2 is blocked; however, reconstitution of photoreceptor input to only L1 or L2 results in wild-type responses. Thus, the first experiment indicates the necessity of both pathways, whereas the second indicates sufficiency of each single pathway. This contradiction can be explained by electrical coupling between L1 and L2, allowing for activation of both pathways even when only one of them receives photoreceptor input. A fundamental difference between the L1 pathway and the L2 pathway is uncovered when blocking L1 or L2 output while presenting moving edges of positive (ON) or negative (OFF) contrast polarity: blocking L1 eliminates the response to moving ON edges, whereas blocking L2 eliminates the response to moving OFF edges. Thus, similar to the segregation of photoreceptor signals in ON and OFF bipolar cell pathways in the vertebrate retina, photoreceptor signals segregate into ON-L1 and OFF-L2 channels in the lamina of Drosophila.