the channel. We did not use a-GTPYS subunits at concentrations higher than 1 nM because GTPYS alone began to turn on channels in the 1-10 nM range. In cases where a40 or a39 activated the channel, subsequent addition of By or GTPYS did not increase activity further as measured by the normalized integral of iK.ACh over baseline (Npo). Likewise, addition of a39-GTPYS after By activation did not increase net iK.ACh (Logothetis et al. 1988). Thus, it is likely that ẞy and a-GTPYS gate the same population of channels. Since By may be common in liganding several a subunits, one would expect that release of By from Gs via ẞy receptor binding would result in iK-ACh activation. We tested this idea by applying 10 μM isoproterenol to the bath and monitoring cell-attached channel activity (the cell remains intact). In some cases, iK.ACh activity increases. Outside-out patches can also be exposed to isoproterenol and ACh and activities compared (figure 4). Although isoproterenol occasionally activates iK.ACh, it has not increased activity beyond 10 percent of that induced by ACh. Also, isoproterenol does not increase activity at all in many cases. Similarly, cholera toxin treatment produces variable results. As expected, isoproterenol should not fully induce the K current. Thus, there may be specificity among the By subunits; By subunits may be geographically constrained, or there may be fewer Gs heterotrimers than pertussis toxin-sensitive G proteins linked to the muscarinic receptor. Both By and a-GTPYS may activate the iK.ACh channel. These data may be interpreted in many ways, since reconstitution techniques do not uniquely address the mechanisms by which G proteins work in vivo. For example, it is possible that either a or By or both activate iK.ACh indirectly. Splitting the G protein in the membranes leaves two possible mediators of activation, one of which may bind directly to the protein and use of which may activate enzymes or other intermediates that increase channel activity. Alternatively, By may remove a tonic a inhibitor (although this seems less likely in view of a-GDP reversal of By activation). Finally, a- and By might bind to different sites on the same channel protein to produce more complicated gating kinetics. This could explain, for example, the biexponential decay during ACh-induced desensitization. At the moment, we favor the first hypothesis, but more experiments are needed to resolve the issue. REFERENCES Birnbaumer, L., and Brown, A.M. G protein opening of K+ channels. Nature 327(6117):21-22, 1987. Breitwieser, G.E., and Szabo, G. Uncoupling of cardiac muscarinic and ẞ-adrenergic receptors from ion channels by a guanine nucleotide analogue. Nature 317(6037):538-540, 1985. Codina, J.; Yatani, A.; Grenet, D.; Brown, A.M.; and Birnbaumer, L. The a subunit of the GTP binding protein Gk opens atrial potassium channels. Science 236(4800):442-445, 1987. Gaskell, W.H. The electrical changes in the quiescent cardiac muscle which accompany stimulation of the vagus nerve. Journal of Physiology 7:451-452, 1886. Gilman, A.G. G proteins: Transducers of receptor-generated signals. Annual Review of Biochemistry 56:615-649, 1987. Kirsch, G.E.; Yatani, A.; Codina, J.; Birnbaumer, L.; and Brown, A.M. a-Subunit of Gk activities atrial K + channels of chick, rat, and guinea pig. American Journal of Physiology 254(6):H1200-H1205, 1988. Kurachi, Y.; Nakajima, T.; and Sugimoto, T. On the mechanism of activation of muscarinic K+ channels by adenosine in isolated atrial cells: Involvement of GTP-binding proteins. Pflugers Archiv. European Journal of Physiology (Berlin) 407:264-274, 1986. Loewi, 0. Uber humorale Ubertragbarkeit der Herznervenwirkung. I. Mitteilung. Pflugers Archiv. European Journal of Physiology (Berlin) 189:239-242, 1921. Logothetis, D.E.; Kim, D.H.; Northup, J.K.; Neer, E.J.; and Clapham, D.E. Specificity of action of guanine nucleotide-binding regulatory protein subunits on the cardiac muscarinic K+ channel. Proceedings of the National Academy of Sciences of the United States of America 85(16):5814-5818, 1988. Logothetis, D.E.; Kurachi, Y.; Galper, J.; Neer, E.J.; and Clapham, D.E. The ẞy subunits of GTP-binding proteins activate the muscarinic K* channel in heart. Nature 325(6102):321-326, 1987. Nargeot, J.; Nerbonne, J.M.; Engels, J.; and Lester, H.A. Time course of the increase in the myocardial slow inward current after a photochemically generated concentration jump of intracellular cAMP. Proceedings of the National Academy of Sciences of the United States of America 80(8):2395-2399, 1983. Pfaffinger, P.J.; Martin, J.M.; Hunter, D.D.; Nathanson, N.M.; and Hille, B. GTPbinding proteins couple cardiac muscarinic receptors to a K channel. Nature 317(6037):536-538, 1985. Soejima, M., and Noma, A. Mode of regulation of the ACh-sensitive K-channel by the muscarinic receptor in rabbit atrial cells. Pflugers Archiv. European Journal of Physiology (Berlin) 400(4):424-431, 1984. Stryer, L., and Bourne, H.R. G proteins: A family of signal transducers. Annual Review of Cell Biology 2:391-419, 1986. Yatani, A.; Codina, J.; Brown, A.M.; and Birnbaumer, L. Direct activation of mammalian atrial muscarinic potassium channels by GTP regulatory protein Gk. Science 235(4785):207-211, 1987. The Biochemistry of Memory D.E. Koshland, Jr. Department of Biochemistry, University of California One of the major cognitive functions now becoming amenable to experimental approaches is memory. To understand memory involves many aspects, but they can be considered as two extremes: The change in properties within a neuron and the change in the network of neurons. The work of our laboratory is concentrated on the first aspect and is focused on both shortterm memory and long-term memory. The model for short-term memory is the bacterial chemotaxis system, in which a memory is required for the organism to select an optimal environment. The elucidation of the mechanism of this short-term memory has identified, in turn, those components that would be appropriate for mediating the long-term memory within a cell. Experiments on mammalian cells, both in our laboratories and in others, indicate repeatedly the close analogies between bacterial systems and those of higher species (Macnab 1985). The Short-Term Memory for Gradient Sensing The bacterial system has developed a short-term memory with information processing devices of widespread applicability. In essence, as described in equation 1, The concentration of a response regulator, X, controls the behavior of the output system (Macnab and Koshland 1972). The level of this response regulator can then be controlled by the velocities of the enzymes that produce X (vf) and degrade it (va). In the bacterial system, the production of X is controlled by a very fast response to one stimulus- in this case, the concentration of chemoeffector in the medium. The degradation of X is also controlled by the chemoeffector, but only after a delayed slow response. This means that the production of X is a constant reflection of the present concentration of chemoeffector, and the degradation of X is a reflection of the past concentration (Koshland 1983). If the bacterium is swimming around at a constant concentration, the two rates will be equal. If, however, it is swimming up or down a gradient, the present concentration will differ from the past concentration, and the level of X will reflect that difference. Thus, an extremely simple and remarkably efficient comparator of the past and the present has been devised. That such a logic pattern is used in other sensory systems seems beyond dispute, although the degree of understanding is far less than in the bacterial system. One of the major components of the bacterial system is the covalent modification of the receptor. Since covalent modification is seen in almost all receptors studied so far, the utilization of two rate processes on different time scales as a memory device seems likely to be an important feature of many neurological systems. One of the important features of this short-term memory device is that the rates of the forward and reverse processes can be adjusted by changing the amount (numbers per cell) of receptors, the amount of synthesizing enzymes, or the amount of degrading enzymes. Thus, precisely the same biochemical machines can be used for neurons that must vary in response characteristics from milliseconds to minutes or days, simply by adjusting the amount of one or two proteins. Integration and Feedback in the Signaling System When a stimulus binds to the receptor, it initiates two types of reactions, an excitation, which essentially involves the fast reaction, and an adaptation, which appears to be a device that includes both the slow reaction and a resetting of the system to zero (Koshland et al. 1982). Beginning at the receptor, a cascade generates phosphorylation and is ultimately responsible for the phosphorylation of the CheY protein, which appears to be the response regulator (Hess et al. 1987; Bollag and Koshland 1988; Wylie et al. 1988). That protein, in its phosphorated form, appears to bind to the motor and provide the mechanism for signaling a change in direction. This activation occurs within a fraction of a second. When the stimulus activates the excitation pathway, it also triggers a change in methylation of the receptor. This occurs through a transmembrane conformational change that changes the accessibility of various glutamic acid residues toward methylation and demethylation (Springer et al. 1977; Koshland 1981). If, for example, an attractant binds to the receptor, it generates a signal toward smooth swimming, i.e., a signal to keep going in the same direction, and at the same time an increase in methylation. The increase in methylation over a time scale of several seconds desensitizes the receptor and returns it to the nonexcitation state. In this way, the receptor can respond to a gradient and the bacterium can swim to a higher outside level of stimulus, but the methylation will reset the system to zero to offset the increased stimulus. A decrease in attractant will generate the opposite response, an immediate signal to change direction and a signal to adapt to the new environment, which again resets the system to zero so that it can respond to future environmental stimuli. An interesting feature that may be common to many signaling systems is the interaction between the two second-messenger systems. This interaction between the phosphorylation and methylation pathways generates an asymmetry in the responses. When the bacterium is going up a gradient of attractant, it is going in a favorable direction and should keep going as long as possible. When it goes in the wrong direction, as soon as it has received a signal to change direction, it need continue no further. It should reset to zero as quickly as possible once it has changed direction. In fact, the excitation system (phosphorylation) interacts with the adaptation system (methylation) to do just that. The response to a wrong direction is much briefer than a response to a correct direction, thus optimizing the survival behavior (Koshland 1988). The reset to zero occurs largely through the conformation of the receptor. The protein has evolved so that the conformational change generated by methylation compensates for the conformational change generated by the binding of the stimulus. This is the basis of adaptation, and the existence of the Weber-Fedner law for other sensory phenomena shows that the same adaptation is used in higher species to desensitize incessant signals. Adaptation to light over many orders of magnitude apparently occurs by phosphorylation of rhodopsin. Taste, smell, hearing, and many other receptors behave similarly. Most of the receptors isolated so far undergo covalent modification. If many of these modifications turn out to be related to desensitization, either through internalization or by modification of the receptor, they will have been selected to extend the range of receptors and increase the sophistication of their feedback systems. The reset to zero mechanism is the device of a biological system to amplify the response to small changes in the environment while retaining the ability to respond over a wide range. Potential Devices for Long-Term Memory To achieve long-term memory or, in fact, any kind of learning, a neuron must be permanently modified. The signaling system must be changed so that a stimulus after learning elicits a different response from that prior to such behavioral modification. In bacteria, models for such a process have been shown in a number of ways: • The galactose receptor can be induced by growing the bacterium in a |