The evolution and maintenance of cooperation for direct fitness benefits is a topic of great interdisciplinary interest. Existing game theoretic models assume either that genes code directly for a strategy, or that gene-environment interactions lead to conditional strategies. However, in organisms with more complex central nervous systems, gene-environment interactions can affect brain development, allowing individuals to make decisions based on learning, memory and experience. Decision-making in the brain is also influenced by information sent by the body: for instance, through the production of hormones. Such processes may cause ubiquitously observed inter-species and inter-individual variation in levels of cooperation that is poorly explained by current models. In order to advance evolutionary questions about cooperation, we need to first understand how differences in personal experience, cognition, and physiology influence variation in cooperative interactions. I propose to address these questions using marine cleaning mutualisms as a model system. In these mutualisms, small fish called ‘cleaners’ remove ectoparasites from so-called ‘client’ fish species. Conflict arises when some cleaner fish species prefer to eat the protective mucus layer of their clients, which constitutes cheating. The major focus of this proposal is to study factors influencing decisions during cooperative interactions, and their link to variable behaviour in the cleaner fish L. dimidiatus, our main study system. Individual cleaners differ with respect to levels of cooperation and expression of cognitive abilities; in turn, client species differ with respect to the quality of the service they receive. Recent evidence suggests that the complexity of the cleaner’s interspecific social environment and/or cleaner-cleaner competition over access to clients may cause variation in cognition and cooperation due to ontogenetic effects. Therefore, one aim is to explicitly test for ontogenetic effects by translocating cleaners between sites that differ in social complexity and by experimentally reducing the cleaner-to-client ratio for 12 months in a large area in the field. We will investigate the consequences of cleaner reduction (i.e., increased cleaning demand) on service quality and client strategies via cleaning observations, individual follows of clients, transects and remote monitoring of home range use in various client species. Lab experiments testing the swimming performance and mucus quality of 18 client species belonging to 14 families will provide insights into the physiological costs that may limit partner control mechanisms like punishment and partner switching, which may explain variation among client species in received service quality. Furthermore, experiments on translocated cleaners and cleaners from the experimentally reduced density area will allow us to test whether these manipulations changed individual levels of cooperation and cognition. Finally, we will use a comparative approach to test in how far the physiology underlying a cleaner’s decisions in cooperative interactions varies with strategic options. Hormone and neurotransmitter levels offer the brain information about the individual’s current physiological state or life history stage, often causing pleiotrophic effects on behaviour. We have established that arginine-vasotocin, serotonin, and cortisol affect levels of cooperation in L. dimidiatus. This species prefers ectoparasites over client mucus. Our working hypothesis is that the same effects exist in other cleaner species with such food preference, while service quality will not be affected in species that prefer to eat ectoparasites and hence do not have to solve a conflict with clients over what to eat. In parallel with empirical research, we aim to develop models that explicitly link parameters affecting social complexity with costs and benefits of more or less sophisticated decision rules, i.e. rules that use different amounts of information about the social environment. Our variables of interest will affect the efficiency of different updating rules that would lead to more or less knowledgeable / cognitively enabled individuals. The updating rules should take into account likely learning mechanisms used by animals, like reinforcement learning. We will explore the effects of social complexity and learning on both ecological and evolutionary time scales. The powerful combination of modelling and empirical data, especially the explicit links between mechanisms and function, should provide us with novel insights concerning the conditions promoting stable cooperation.