The most well-known molecular mechanism of learning involves coincidence detection. In post-synaptic LTP, the NMDA receptor only opens fully if a postsynaptic depolarization has removed the magnesium block by the time glutamate arrives at the receptor. In pre-synaptic facilitation, adenylyl cycase only generates large amounts of cAMP when stimulated both by transmitter and by coincident Ca2+ influx. Thus, in both cases, you need neural activity (i.e., action potentials or spikes) to coincide onto the synapse in question. Insect learning, specifically clasical olfactory conditioning, has been instrumental in developing this model of "spike timing-dependent plasticity" (STDP). For delay conditioning in which the conditioned stimulus (CS) overlaps with the unconditioned stimulus (US), this is not a problem: the spikes of the CS are still arriving at the convergence point when the spikes from the US start to come in. However, in trace conditioning, when there is a delay between the end of the CS and the onset of th US (of up to 24h in the case of conditioned taste aversion), it is difficult to imagine how the well-known mechanisms of STDP could occur. What happens during the interval between CS offset and US onset for trace conditioning to occur?
To say it right away, we just discussed the new paper in Nature Neuroscience from the lab of Mark Stopfer (NNeuro preview) in our journal club and it doesn't answer this question either. What it does show is that in the paradigm which was so instrumental in develping STDP (classical olfactory learning in insects), STDP appears not to be able to explain trace conditioning either. The authors recorded from projection neurons (projecting from the antennal lobes to the mushroom bodies) and from Kenyon cells (intrinsic mushroom-body neurons) in moths (Manduca sexta). They showed that after odor alone presentations (no conditioning) no more spikes are fired in the Kenyon cells at a time point where they had demonstrated a US presentation to lead to maximum learning behaviorally. This is remarkable, because the Kenyon cells are considered to be the site where the associative memory is stored in this paradigm. The really new aspect of this work was that electrophysiological recordings (albeit not during conditioning) were combined with a behavioral approach analyzing optimal inter-stimulus intervals for classical conditioning. What the authors found was basically a negative result: STDP in the Kenyon cells cannot account for the learning exhibited by the insects. This is reminiscent of trace and delay conditioning in mammals: "In delay eyeblink conditioning, the CS overlaps with the US and only a brainstem-cerebellar circuit is necessary for learning. In trace eyeblink conditioning, the CS ends before the US is delivered and several forebrain structures, including the hippocampus, are required for learning, in addition to a brainstem-cerebellar circuit." (source).
Maybe also in insect trace conditioning, both Kenyon cells and some other structure are required? Maybe this other structure works as a buffer to store the eligibility trace of the CS until the US arrives? Another option could be residual calcium (or some second-messenger) lingering for a few seconds until the US spikes arrive in the Kenyon cells. Only Kenyon cell recordings during conditioning can show the behavior of the Kenyon cells when the US arrives (to fully rule out STDP). I also think a trace conditioning paradigm for Drosophila needs to be developed in order to harness the genetic power also for this type of learning (this would address the calcium or second messenger hypothesis). This paper didn't really answer any questions, but it was so thoroughly done and well-designed that it threw up a lot of interesting ones which will hopefully lead to a completely new line of learning research in insects.

Citation:
Iori Ito, Rose Chik-ying Ong, Baranidharan Raman, Mark Stopfer (2008). Sparse odor representation and olfactory learning Nature Neuroscience, 11 (10), 1177-1184 DOI: 10.1038/nn.2192