We examine the time accuracy of cortical activity by jittering the spike trains and examining the effect on the statistics of neuronal firing.

Activity of several single units was recorded in parallel from frontal areas of behaving monkeys. Recording time was divided into epochs of quasi-stationary activity by the hidden Markov modeling technique. All the data from the epochs in each Markov state was analyzed together by constructing three fold correlations among triplets of single units. These correlations were constructed as histograms comprising 100x100 bins of 3 ms each.

For each bin we computed the probability of seeing so many counts in the bin, given the counts in its neighboring bins. Two statistics were extracted from these correlations: 1. The density of these probabilities as derived from the correlations of all the available data in each state. 2. The probability of the least likely event in each correlation histogram. For both types of statistics we get values which are far beyond what is expected by chance.

Validation of the significance of these statistics is confirmed by the method of Bienenstock, Date, and Geman. According to their approach, the null hypothesis, to be tested, is that the spike trains do not have any accuracy beyond some value W. If so, then jittering each spike within a window of W ms should not affect the statistic derived from the spike trains. The data is then jittered for 100 times and a histogram of the derived statistic is constructed. If the statistic of the real data is well outside this histogram then the null hypothesis may be rejected.

According to this test, the accuracy of cortical firing was well below 6 ms.

How can such accuracy be generated in the noisy cortical mesh of week connections? This issue and its implications for information processing in the cortex will be discussed.

Supported in part by grants from the BSF and GIF.

The task of organizing perception and behavior in meaningful interaction with the external world prompts the brain to recruit its resources in a properly orchestrated manner. Our principal research goal is to understand how this organization is dynamically brought about, and how the brain uses such coordinated activity. Studies of cortical function on the basis of multiple single-neuron recordings revealed neuronal interactions which depend on stimulus- and behavioral context, exhibiting dynamics on different time scales, with time constants down to the millisecond range [1-4]. Mechanisms underlying such dynamic organization of the cortical network are investigated by experimental and theoretical approaches. Recent work focuses on the occurrence of precise joint-spiking events in cortical activity [5-9]. Specifically, we tested the hypothesis that precise synchronization of action potentials among groups of neurons is supported by cortical network activity, in spite of the fluctuating background [10]. Thus, we found evidence [11-13] that volleys of precisely synchronized spikes can propagate through the cortical network in stable fashion with a temporal precision down to ±1 ms, consistent with experimental observations. These findings suggest that a combinatorial neural code, based on rapid assocofiations of groups of neurons coordinating their activity at the single spike level, is biologically feasible.

1. Aertsen AMHJ, Gerstein GL, Habib MK, Palm G (1989) Dynamics of neuronal firing correlation: modulation of "effective connectivity". J Neurophysiol 61: 900-917
2. Aertsen A, Arndt M (1993) Response synchronization in the visual cortex. Current Opinion in Neurobiology 3: 586-594
3. Vaadia E, Haalman I, Abeles M, Bergman H, Prut Y, Slovin H, Aertsen A (1995) Dynamics of neuronal interactions in monkey cortex in relation to behavioral events. Nature 373: 515-518
4. Nawrot M, Aertsen A, Rotter S (1999) Single-trial estimation of neuronal firing rates - From single-neuron spike trains to population activity. J Neurosci Meth 94: 82-92 back
5. Abeles M., Bergman H., Margalit E., & Vaadia E. (1993) Spatio-temporal firing patterns in the frontal cortex of behaving monkeys. J. Neurophysiol. 70: 1629-1643
6. Abeles M, Prut Y, Bergman H, Vaadia E, Aertsen A (1993) Integration, synchronicity and periodicity. In: Brain Theory: Spatio-Temporal Aspects of Brain Function, pp 149-181. Aertsen A (ed). Amsterdam, New York: Elsevier Science Publ
7. Riehle A, Grün S, Diesmann M, Aertsen A (1997) Spike synchronization and rate modulation differentially involved in motor cortical function. Science 278:1950-1953
8. Prut Y.,Vaadia E., Bergman H., Haalman I., Slovin H. & Abeles, M. (1998) Spatiotemporal structure of cortical activity: properties and behavioral relevance. J. Neurophysiol.79: 2857-2874
9. Grün S, Diesmann M, Grammont F, Riehle A, Aertsen A (1999) Detecting unitary events without discretization of time. J Neurosci Meth 94: 67-79 back
10. Arieli A, Sterkin A, Grinvald A, Aertsen A (1996) Dynamics of ongoing activity: Explanation of the large variability in evoked cortical responses. Science 273:1868-1871 back
11. Diesmann M, Gewaltig M-O, Aertsen A (1995) Characterization of synfire activity by propagating pulse packets. In: Computational Neuroscience–Trends in Research 1995, PP 59-64. Bower J (Ed). San Diego: Academic Press
12. Aertsen A, Diesmann M, Gewaltig MO (1996) Propagation of synchronous spiking activity in feedforward neural networks. J Physiol (Paris) 90:243-247
13. Diesmann M, Gewaltig MO, Aertsen A (1999) Stable propagation of synchronous spiking in cortical neural networks. Nature 402: 529-533 back

*Funded by DFG, GIF and HFSP. Further information at http://www.brainworks.uni-freiburg.de

In previous work (Alonso, Usrey and Reid. Nature 383, 815), we have found that neighboring neurons in the lateral geniculate nucleus (LGN) fire many of their spikes in a highly correlated fashion (within 1.0 msec of each other). Pairs of neurons with the strongest correlations are also very likely to connect to the same simple cells in striate cortex (Reid and Alonso, Nature, 378, 281).

In order to examine the source and the potential implications of these synchronous events, we have made intra-ocular recordings from single retinal ganglion cells simultaneously with recordings from up to eight neurons in the LGN. We have recorded simultaneously from pairs of LGN neurons along with a retinal ganglion cell that provides strong input to these cells. We found that the LGN cells were clearly being synchronized by the common input from the single retinal ganglion cell. Further, it appeared that the retinal influence on each geniculate cell was itself correlated; if a retinal spike produced a spike in one LGN cell, it was more likely to do so in the other. Thus, the pattern of divergent inputs from retina to LGN causes the very high degree of synchrony in the LGN. This synchrony is likely to be "read off" by the convergent input from several LGN cells onto single cortical neurons.

Two types of information-theoretic analyses of LGN spikes trains will be presented. First, the pattern of firing of a single neuron will be shown to carry more information than could be transmitted if single spikes were considered in isolation. Second, the synchronous firing of two LGN cells is shown to carry more information than if the two spike trains were considered separately.

The olfactory system conveys information about the environment contained in the local concentrations of volatile organic molecules ('odorants') emitted by plants and animals or produced by events such as combustion or decomposition. Olfactory signal transduction begins with binding of odorants to subsets of ~1000 different types olfactory receptor in the nasal epithelium. In the rat olfactory bulb, input from each receptor type is targeted to pairs of glomeruli. Together, the glomeruli form a two dimensional map of receptor types across the surface of the bulb.

Using intrinsic optical imaging in the anesthetized rat, it is possible to simultaneously visualize the relative activation of a significant portion of the glomerular map. Unlike cortical intrinsic imaging, robust and consistent signals can be obtained using single 5-20 sec trials without averaging. By imaging glomerular activity vectors evoked by different odorants, we can study how odorants are encoded across a population of neuronal functional units.

Pure single-chemical odorants produce sparse population activity across multiple glomeruli. Activation patterns are typically distributed in the sense that they include glomeruli in multiple regions of the bulb, but are local in the sense that functional specialization is seen in different regions of the bulb. Responses to chemical mixtures can under some conditions can be precisely predicted by linear summation of responses to the individual components. Similarity of the activation vectors corresponding to different odorants can be computed using a Euclidean distance metric. This analysis shows greater similarity of responses for odorants judged (by humans) to belong to similar perceptual classes. Thus, it is possible the perceptual as well as the chemical properties of odorants are represented in patterns of glomerular activation in the olfactory bulb.

My talk will focus on two closely interrelated issues: activity-instructed or "correlation-based" development of the layer 4 circuit of cat V1; and the structure of the mature circuit.

Models of development of the geniculocortical connections have demonstrated conditions on LGN spontaneous activity correlations that seem required for correlation-based development to account for the development of orientation-selective simple cells and ocular dominance segregation with orientations matched in the two eyes. These conditions are thus far consistent with experimental data, although they are not yet strongly tested. In addition, several prominent recent experiments that have been claimed to call into question the activity-instructed scenario in fact are more or less exactly what is expected under this scenario.

To address the structure of the mature circuit, we begin with the assumption of "Hubel-Wiesel" connectivity from LGN to simple cells (ON-subregions of simple cells receive input from ON-center cells whose centers overly the ON-subregion, and similarly for OFF-center cell), as demonstrated by Reid and Alonso and as results from the above developmental models. Based on the observations of "push-pull" in simple cell receptive fields (inhibition evoked by dark where excitation is evoked by light, and vice versa), as well as the observation that the excitation and inhibition received by a cat layer 4 simple cell have similar or identical orientation tuning, we have proposed a "correlation-based" model of intracortical connectivity in layer 4. Here, "correlation-based" refers to the pattern of connections rather than the developmental mechanism (though the latter was a motivation): the probability of an excitatory cell making a connecting to another cell increases strongly with the degree of correlation between the LGN inputs received by the two cells, while the probability of an inhibitory cell making a connection to another cell increases strongly with the degree of anticorrelation between the LGN inputs received by the two cells. In addition, we require that the inhibition received by a simple cell be significantly stronger than the LGN excitation it receives. This model reproduces a variety of intracellularly and extracellularly measured simple cell response properties, including the contrast-invariance of orientation tuning, lowpass temporal frequency tuning, the effects of cortical cooling, and a number of "nonlinear" response properties that have previously been argued to require untuned "normalizing" inhibition.

Finally, we have found that this full layer 4 circuit—both the geniculocortical and intracortical connectivity—will arise from simple "correlation-based" developmental mechanisms, given only the same LGN activity correlation structures we had previously identified. This leads us to tentatively generalize to layer 4 of cortex more generally: if layer 4 develops in a correlation-based way based on the activity of its inputs, with inhibition dominant, it will develop a magnitude-invariant representation of "form" (coactive structures) in its input; in V1 this manifests as contrast-invariant orientation tuning.

I will address this overall picture, but will not be able to cover all of the points mentioned above; perhaps inclusion of the extra detail in the abstract will help spur discussion.

We consciously perceive the visual world around us as a composition of distinct objects. This information is used to guide behavior where salient visual objects result in faster and better detection than less evident stimuli. Moreover, visual information needs to be maintained for short time periods since in the constant and rapidly changing visual world a stimulus may have disappeared whereas a behavior response still has to be made. Thus, the perception of an object and the response to it are closely related operations. However, the way the visual system manages to produce perception is still unclear.

In the primary visual cortex, neuronal activity is modulated by the way the stimulus is embedded in the visual scene. This activity -contextual modulation- is a late component of neuronal activity and originates partly from feedback projections from extra-striate cortical areas. Contextual modulation in the primary visual cortex thus combines retinal information with activity already processed in higher visual areas.

We recorded from neurons in the primary visual cortex of macaque monkeys engaged in a figure-ground discrimination task. According to the behavioral report of the monkeys, contextual modulation parallels the perception of the visual stimulus. A high contrast stimulus results in a better performance than a low contrast stimulus where detection is poorer. Similarly, modulation belonging to the high contrast stimulus is stronger than the modulation for low contrast stimulus and, under some conditions, modulation is absent when the stimulus in not detected. Furthermore, when the reaction time is short, modulation is stronger than when reaction times are longer. Thus, the strength of contextual modulation reflects the perception of the visual stimulus. It is even the case that after a brief, transient presentation of a visual stimulus contextual modulation is retained until a response is made. The strength of this modulation is weakened in case the stimulus has become irrelevant or when the stimulus position has been misjudged i.e. forgotten. So, modulation may serve as a neuronal substrate for working memory.

Thus, the properties of contextual modulation correlate with the behaviorally relevant aspects of visual perception and is thus a suitable candidate for the neuronal substrate that is used as an internal representation of the visual world around us.

In the primate visual cortex, a series of anatomically connected cortical areas, collectively referred to as the "motion system", contain high proportions of directionally selective cells, and have been shown to be intimately involved in motion perception. Receptive fields (RFs) become larger and selectivities become more complex as one ascends this hierarchy of motion processing. This trend is exemplified forcefully by the difference in RF properties in areas MT and MST. MST cells' RFs are much larger than those in MT, and are often selective for complex motion patterns such as rotations, expansion or contraction. It is unclear how these complex RFs are assembled from the simpler RFs of the areas (such as MT) which supply afferents to MST.

To explore how the RFs of these extrastriate areas are constructed, we have been using combinations of simple "motion impulse" stimuli to probe mechanisms of spatial summation. The stimuli are small, moving spatial Gabor functions, briefly and locally presented. In our experiments, these are used either individually or in various combinations to investigate how responses depend on the locations, directions, and contrasts of the local component stimuli within the RF. We found that when presented with multiple stimuli within their RFs, both MT and MST cells responded well below what would have been predicted on the basis of linear summation. The reduction in response in both areas ("normalization") was contrast-dependent in MT, with a high contrast sensitivity. It was not strongly dependent on either stimulus location or direction, either in MT or MST. In both MT and MST, reasonable account of the response to combinations of stimuli could be made using a simple linear model incorporating this contrast normalization. Therefore, despite the profound differences in visual responses between the two areas, their basic mechanisms of spatial summation appear largely similar.

Single-cell recordings in monkey cerebral cortex have shown that MST neurons respond selectively to optic flow, the patterned visual motion that is seen during observer self-movement. These responses reflect heading direction by integrating visual and vestibular signals about self-movement.

Pursuit eye movements complicate optic flow perception by adding rotation to the retinal image and displacing the retinal location of the simulated heading. We find that pursuit alters the optic flow responses of almost all MST neurons. But a vector representation of the neuronal responses, summed across a number of neurons, correctly indicates heading during fixation and pursuit. In three-dimensionally structured environments, MST combines heading and speed selectivity to maintain a veridical heading representation.

The importance of visual mechanisms for spatial orientation is revealed by their impairment in the spatial disorientation of Alzheimer's disease (AD). AD patients show a selectively elevated perceptual threshold for the patterned visual motion of optic flow. This perceptual deficit is correlated with spatial navigation deficits. Spatial disorientation in AD has been linked to posterior association cortex lesions that involve the human homologue of area MST.

Together, these findings suggest that extrastriate visual areas integrate visual and nonvisual cues about spatial orientation. The failure of these mechanisms may result in debilitating spatial disorientation.