Factor analysis of water maze learning in the Morris maze (see David P. Wolfer) reveals that a major source of behavioral variation in this task is thigmotaxis. Although statistically independent of probe trial measures, strong thigmotaxis can confound measures of spatial retention. But what means variations of thigmotaxis?
Many knockout mutations entail differential phenotypes: this is known as pleiotropy and means that a group of mutants may contain animals in which thigmotactic behavior is different, plus another group in which the mutation affects spatial memory as assessed by probe trial scores.
Our data show that there are many mice which learn the platform in the single first trial which is sufficient to establish a spatial map. Subsequent variations in learning performance and probe trials must then reflect the influence of confounding factors, e.g., of thigmotaxis. This is often considered as a simple fear/stress response. However, our data from 3000 mice indicate that this holds only for the very first trials. Prolonged thigmotaxis, however, probably reflects reduced behavioral flexibility or another cause of inability to switch to an appropriate classical search strategy. There will be examples to show that even excessive thigmotaxis does not prevent the formation of spatial memory. 
Analyzing the roots of thigmotaxis in genetically modified mice is extremely difficult, and it is even more difficult to decide whether the hippocampal formation is affected or involved. Obviously, one approaches the limits of the classical bottom-up approach in analyzing the genotype-to-phenotype pathway.
Thus, I shall present a tentative solution to the riddle by presenting our classical top-to-bottom approach based on the analysis of the intra/infrapyramidal mossy fiber projection (IIP-MF) and their relations to hippocampus-dependent (and hippocampus-mediated) learning and innate behavior of mice. In brief, the extent of the IIP-MF correlates often with hippocampus-dependent learning tasks (negatively with two-way avoidance, positively with radial maze learning and open-field habituation, and positively with place reversal learning in the water maze). The general picture is that mice with scanty IIP-MF projections behave as they would suffer from a very mild hippocampal lesion syndrome. Using the genetic variation of the IIP-MF system as a marker of hippocampal involvement, we and others investigated the role of IIP-MF variations in more naturalistic behavior such as attack latency towards intruders, and in non-learned behavior such as paw preference. Somewhat surprisingly, animals with small IIP-MF projections showed reduced paw lateralization and shortened attack latencies.
These findings caused us to study the effects of natural selection on the IIP-MF projection in mice exposed to 4 years live in outdoor pens in western Russia. This resulted, after 3-5 generations of outdoor life, in a significant reduction of the IIP-MF projection as compared to animals housed conventionally. Latest results show that this difference is genetic. In addition, we studied IIP-MF projections in a variety of small mammals. Results imply that the IIP-MF co-vary positively with encephalisation, complexity of environment, and a hunting lifestyle. Taken together, the data imply that one basic function of the hippocampus (and perhaps associated neocortex) is to maintain complex behavior patterns under conditions of stress and multiple sources of interference. Some specific environments appear to select a behavioral phenotype characterized by hyper-reactivity and poorer performance in complex tasks, for example in root voles and in the house mice living in a ghetto situation. This conclusion was then tested in the water maze by comparing two vole species with differential IIP-MF projections and complex versus simple environments (bank voles versus root voles). 
Predictions were derived from re-analyzing our old IIP-MF data and water maze learning. This showed a significant negative correlation between thigmotaxis, and a positive correlation with the ability to find a relocated platform. Predictions were confirmed as the bank voles with larger IIP-MF showed perfect water maze learning, including better ability of reversal learning. Root voles (with almost missing IIP-MF) showed excessive thigmotaxis and less efficient reversal learning. Probe trial scores were at chance level (as in the worst KO-mice), but the analysis showed that they had nonetheless developed a spatial memory.
Hence, by combining a bottom-up approach based on analysis of KO-mice with a top-to-bottom approach in the laboratory and the field, we can conclude that variations of thigmotaxis reflect, indeed, a basic hippocampal function, namely stabilization of ongoing complex behavioral patterns. This property is probably not hippocampus-specific, but is behaviorally sensitive to a variety of genetic and epigenetic factors because of the critical position of the hippocampus in linking proximally connected neocortical association areas with hypothalamus, mesolimbic structures, reticular formation and associative thalamus.