Tailor-made memory: natural differences in associative olfactory learning in two closely related wasp species

Learning and memory formation are often seen as traits that are purely beneficial, but they are associated with metabolic costs as well. Since costs and gains of learning and memory are expected to vary between species, the ease and speed with which stable (consolidated) long-term memory (LTM) is formed, is expected to differ between species. For animals that occupy different ecological niches, ‘slow’ learning may be as adaptive as ‘fast’ learning. If an animal encounters a relatively predictable environment during its lifetime, fast learning is a good strategy. If the environment is relatively unpredictable, however, an animal may need more time and experiences to evaluate information before storing it as long-lasting memories. This concept is known as tailor-made memories: a species learns in the way that is most favourable, given the circumstances. In order to assess how such tailor-made memories evolve, I have used a multitrophic model system. This system consisted of (1) two closely related parasitic wasps (Cotesia glomerata and C. rubecula) that show a profound difference in learning, (2) the herbivorous cabbage white butterfly larvae Pieris brassicae and P. rapae, in which the parasitic wasps lay their eggs, and (3) the host plants Brussels sprouts (Brassica oleracea var. Gemmifera) and nasturtium (Tropaeolum majus). In my experiments, the wasps could learn to associate the odours of a plant with the presence of suitable hosts, by having one or more oviposition experiences (‘conditioning trials’) on that plant. Previous experiments showed that C. glomerata needs only one conditioning trial to form LTM, whereas C. rubecula needs three trials spaced in time to do so. In addition to LTM, another form of consolidated memory exists; anaesthesia-resistant memory (ARM). Both LTM and ARM are resistant to retrograde amnesia, which can be induced by cooling the wasps after conditioning. In contrast to LTM however, ARM is not protein synthesis-dependent. It can therefore be seen as a ‘cheap’ form of long lasting memory. Consolidated memory in C. glomerata is thought to consist exclusively of LTM, whereas in C. rubecula it appears to be a mixture of both ARM and LTM. LTM formation requires protein synthesis, a process in which the transcription factor cAMP response element-binding protein (CREB) plays a key role. As a result of alternative splicing of the CREB mRNA transcript, the CREB protein occurs in different forms called isoforms. In model organisms such as the fruit fly Drosophila melanogaster, the mollusc Aplysia californica, and also in mammals such as mice and men, CREB isoforms have been shown to activate or repress transcription. Therefore, it has been hypothesized that the ratio of activator and repressor isoforms acts as a molecular switch for LTM formation. Such a switch could be responsible for species-specific differences in learning and memory. In this study the CREB gene of C. glomerata and C. rubecula was cloned and sequenced, and nine isoforms were identified in the two Cotesia species. The abundance of two of the nine mRNA variants coding for these isoforms differs significantly between C. glomerata and C. rubecula; the other variants are expressed similarly in both species. A conditioning trial, however, seems to induce changes in the expression of some of the major isoforms, indicating that the learning process itself may establish a ratio between activators and repressors that determines whether LTM is consolidated or not. Although such molecular mechanisms can potentially act very quickly, it may sometimes take up to days or weeks before information is stored in long-lasting memories. To explain how and why such differences in memory dynamics occur, we need insight in what happens when selection acts on natural variation in learning rate. In order to investigate this, I applied a bidirectional selection regime and reared two lines of C. glomerata wasps that differed significantly in learning rate (the decreased-learning line (DLL) and the increased-learning line (ILL)). By applying the protein synthesis inhibitor anisomycin before conditioning and measuring memory retention after conditioning, I showed that the memory consolidation dynamics of the selection lines differed. The DLL did not consolidate LTM anymore, whereas the ILL still did. By combining this study with experiments in which I induced retrograde amnesia by cooling at certain time intervals after conditioning, I demonstrated that in C. glomerata, anaesthesia-sensitive short-term memory directly consolidates into LTM, without an intermediate ARM phase. ARM represents a low-cost form of long-lasting memory (since it is not protein synthesis-dependent) and its presence is assumed to be favourable in animals that need more time to evaluate information, before storing it in the form of consolidated memories (e.g., in C. rubecula). The inability of C. glomerata to form ARM is costly because it may lead to an expenditure of energy (i.e., protein synthesis) on the ‘premature’ storage of unreliable information. Comparison of my selection lines showed that a high learning rate has costs. Longevity appeared to be significantly higher in wasps from the DLL than in those from the ILL. Moreover, females of the ILL have significantly larger brains than females from the DLL, while retaining a similar body size. These exciting results show that trade-offs occur (i.e., brain size vs. longevity) as a result of the bidirectional selection pressure that we applied. Moreover, the costs associated with a high learning rate seem to be of a constitutive nature. This means that animals that are able to quickly form consolidated memory pay for it by maintaining a large, costly brain and having a decreased lifespan, even when they do not actually use their learning abilities. The results of my work show that comparative research involving a model system consisting of two closely related animals with a natural difference in learning rate yields unique information, and is preferred over the use of ‘traditional’ model organisms. It enables testing of various hypotheses with an ecologically relevant learning paradigm. Neuroscience (and biology in general) would benefit greatly from an increase in the use of model systems that consist of closely related species that show differences in the trait of interest. The work described in this thesis shows how fruitful such a comparative approach can be.