Close neighbors can often influence each other’s behavior. The same seems true for neuronal synapses. In this week’s Nature, Howard Hughes Medical Institute investigator Karel Svoboda and colleagues at Cold Spring Harbor Laboratory, New York, show that induction of long-term potentiation (LTP) at one synapse reduces the threshold for LTP at neighboring synapses. The discovery may have important implications for understanding synaptic plasticity, a key requirement for learning and memory, and it is likely to lead to several new avenues of research, suggests Bernardo Sabatini, Harvard Medical School, in an accompanying News & Views. The finding is a basic neuroscience advance. It is also relevant to Alzheimer disease (AD) research since compromised synaptic plasticity is a central part of AD pathology, and the right way of influencing the threshold for LTP could eventually have therapeutic implications. Also in this week’s Nature, a second paper from Svoboda’s group reports that mice can learn to detect brief stimulatory pulses applied to as few as 60 neurons, while Arthur Houweling and Michael Brecht at the University of Berlin, Germany, report that stimulation of just a single neuron can sometimes alter an animal’s behavior. Together, these two papers strengthen the case that activating small numbers of neurons is often sufficient to elicit profound behavioral effects.
In the first paper, Svoboda and coauthor Christopher Harvey tackle a much-debated question, namely, the extent to which LTP, or the ability of synapses to increase in size and strength in response to repeated stimuli, is synapse specific. Previous work, notably by Daniel Madison at Stanford University (Schuman and Madison, 1994), suggests that LTP at one synapse may influence that at others, but the results have been controversial. Reports indicate that nearby synapses are depressed, potentiated, or just plain unmoved by neighboring LTP, noted Sabatini. “The problem with most previous studies on the effects of LTP on unstimulated synapses is that the experimenter did not choose which synapses to study, and instead analyzed those that happened to be stimulated by an electrode,” he writes. The shortcoming with that approach is that in a densely packed dendritic arbor, such as that found in the hippocampus, electrode stimulation may affect remote synapses, making it difficult to determine whether synaptic activation is due to a direct effect or to the influence of a neighboring synapse. Harvey and Svoboda circumvented this problem by using two-photon light pulses to release the neurotransmitter glutamate from a chemical “cage.” Because the photons can be focused through deep tissue to coincide at a miniscule spot, uncaging stimulates only individual synapses, which means any impact on nearby synapses must be due to crosstalk.
Harvey and Svoboda applied this approach to acute hippocampal slices from mice. Pulsing a single synapse repeatedly with light induced increases in spine volume and also in the amplitude of microexcitatory post-synaptic currents (μEPSCs) in the targeted spine only. Nearby spines were unaffected. However, the authors found that the threshold for LTP induction dropped significantly at those nearby spines. While a subthreshold LTP protocol of shorter pulses failed to induce LTP when given alone, it did elicit increased spine volume and higher μEPSCs amplitudes when given shortly after LTP had been induced in a nearby spine. “LTP induction at one spine therefore lowered the threshold for potentiation at nearby spines while maintaining input specificity,” the authors write. They found the same effect by stimulating synapses electrically. In that case they monitored the initial LTP by measuring spine enlargement microscopically; then they applied a subthreshold protocol to a nearby, unenlarged spine. This experiment controlled for any artifact that might be introduced by the two-photon procedure.
The researchers characterized the crosstalk, finding that it carries as far as 8 μm away but at 10 μm, the effect wore off. It seems to depend on some kind of intracellular rather than extracellular signal, because nearby spines on different dendrites do not crosstalk. It does not seem to rely on intracellular calcium, however, because depleting stores did not eliminate the subthreshold effect; neither did protein synthesis inhibitors, suggesting that the phenomenon is not related to “synaptic-tagging,” a form of synaptic plasticity that relies on new proteins.
What does mediate this synaptic crosstalk? “Our results indicate that the intersynaptic spread of intracellular signaling factors probably has a key role. The timescale and spatial scale of crosstalk are consistent with a diffusing cytoplasmic factor,” write the authors. It is unclear what that factor might be, but this is bound to be one of those avenues of research that Sabatini predicts. In addition, Sabatini suggests that researchers will be interested in any new behavioral and learning capabilities this phenomenon bestows on neural networks. Rich questions also include the mechanisms underlying the “priming” of neighboring synapses for LTP induction, whether the spatial and temporal limits depend on the number of synapses stimulated, and how the whole system is related to long-term plasticities, such as synaptic capture and tagging.
The other two papers also examine the role of neighbors but at a bigger scale. Svoboda and colleagues used microstimulation to see how many neurons need to be activated to drive learning. Again they used photons, but this time to activate whole neurons. First author Daniel Huber and colleagues transfected animals in utero with a transgene for channelrhodopsin-2(ChR2). This light-activated calcium channel was originally found in algae but has since been adapted as a cool new imaging technology to map functional connectivity in mice (see Wang et al., 2007). Huber found ChR2 expression was restricted to pyramidal cells in layer 2/3 of the somatosensory cortex—mainly in the barrel cortex. The authors showed that light stimulated action potentials in these neurons (distinguished by their coexpression of both red and green fluorescent proteins) and that the action potentials depended on light intensity (as had Arenkiel et al., 2007). Next, Huber used light to train freely moving mice in a reward protocol. A window placed in the cranium allowed the researchers to match light pulses with delivery of water to one of two ports. After four to seven training sessions, all the mice had learned to associate the light pulse with the correct port for reward.
Knowing that the mice were trained to the light pulses, the researchers next asked how little light is needed to guide the mice to the correct port. By a combination of light intensity experiments and mathematically figuring out how many neurons might be photoactivated, they conclude that for trains of five induced action potentials, as few as 60 neurons need be activated to drive reliable performance. For a single induced action potential, around 300 neurons have to be activated. It appeared that the total number of action potentials was the limiting factor, rather than the pattern. But the authors claim that these numbers are maxima, not minima. First, the scientists did not account for the deterioration of the light signal as it penetrates the dura, and second, the measured spatial distribution of the light may be falsely large, they note. Thus, fewer neurons are probably required for the conditioned response.
Houweling and Brecht would probably agree. The German scientists also trained animals, this time rats, with a water reward in response to electrical microstimulation of the barrel cortex. Then they used juxtacellular stimulation, a technique developed to evoke action potentials in single cells (see Pinault, 1996) to test if individual neurons can elicit the correct response. They found that rats given microstimulation (affecting multiple neurons) responded appropriately 71 percent of the time. Animals receiving no or subthreshold stimulation responded correctly only 13 percent of the time. But rats given single-cell stimulation responded correctly 47 percent of the time. “Quantification of the responses suggests that the animals reported single pyramidal cell activity,” write the authors.
Together, the two papers suggest that scientists may one day be able to identify individual circuits among the many that form the basis for perception and cognition.—Tom Fagan