Targeted mutagenesis permits selective manipulation of genes in mice and recent developments make it even possible to design conditional and reversible mutations. In neuroscience, gene targeting studies often aim at understanding the molecular and cellular basis of higher brain functions and behavior. In order to reconstruct the functional chain from a gene to behavior, the effects of genetic manipulations must be studied at the level of cellular engines and signaling pathways, single cells, cell ensembles and brain systems, and finally at the level of behavior itself. Due to confounding factors from both the environment and the genetic background, it becomes increasingly difficult to attribute behavioral alterations to a single genetic change as one moves up this hierarchy. Our laboratory has begun to study the genetics of mouse behavior in relation to the genetic variability of hippocampal structure long before we became involved in gene targeting studies. Because all behavioral analyses were conducted using standardized protocols, a large database is now available, on which we have conducted a meta-analysis using factor analytical techniques.

Behavioral factors in the water-maze
With some modifications, our swimming navigation protocol follows the original description given by Richard G. M. Morris for the rat. The entire procedure takes five days, each animal doing a total of 30 trials, six per day. The position of the hidden platform remains fixed for the first three days (18 trials, acquisition phase). Then the platform is placed in the opposite quadrant for two more days (12 trials, reversal phase). The first reversal trial serves as probe trial to analyze spatial retention and search behavior of the animals. Factor analysis extracts four factors that account for >80% of the variation observed in the behavioral measures. Factor 1 explains 48% of the behavioral variation. We term it "thigmotaxis" because it is closely associated with frequent swimming near the wall as well as with prolonged escape latencies. Thigmotaxis is an innate behavior and is displayed to some degree by almost all mice at the beginning of the test. Factor 2 accounts for another 16% of the behavioral variation. One might term it "retention" because it reflects primarily the precision and intensity of searching for the former goal during the probe trial. Factor 3 which we call "passivity" explains further 11% of the measured variation. It correlates with frequent passive floating and reduced swimming speed and is loosely associated with prolonged escape latencies. Factor 4, "chaining", finally accounts for 7% of behavioral variation and reflects primarily the prevalence of chaining responses, a strategy where animals use the wall of the pool as a local cued to locate the goal.
Two observations are significant. First, in our hands escape performance during acquisition appears largely determined by the ease with which the mice give up "non-cognitive" strategies such as thigmotaxis and passivity. Second, the successful development of a spatial strategy becomes evident only during the probe trial and scores earned during the probe trial cannot be predicted by any parameter measured during learning. Even parameters that were specifically designed to assess the goal directedness of the swim path, such as the Gallagher cumulative search error, Whishaw’s error, and % time spent in the goal quadrant, mainly measure thigmotaxis and have little predictive value for the probe trial.

Mutation and background effects
An overview of our database of 2605 mice with respect to a large number of parameters shows that scores earned by mice with various symptomatic targeted mutations or randomly inserted transgenes largely fall within the range of variation already observed between samples of wildtype mice with inbred, mixed or F1 hybrid genetic background. Thus, one cannot rely on the assumption that designed mutations have larger effects than variations of genetic background. Moreover, factor analyses conducted on database subsets containing only symptomatic or wild-type and asymptomatic mice, respectively, extract the same factors with virtually identical loading patterns. Taken together, the overview of individual parameters and the extended factor analysis indicate that there are no safe criteria, neither quantitative nor qualitative, permitting to distinguish mutation and genetic background effects. 
This has important consequences for the design and interpretation of experiments involving transgenic mouse models. First, systematic background differences between test and control groups must be strictly avoided, else interpretation of behavioral effects is no longer possible. Second, for results to be comparable across laboratories, experiments should be carried out in a defined and reproducible genetic background. Finally, inhomogeneous genetic background within an experimental group are likely to increase behavioral variation.

Controlling genetic background
Systematic background differences between test and control groups are best avoided if one mates heterozygote animals and compares wildtype and homozygous offspring within the same litters. In order to minimize within group variation, one may want to use samples with homogeneous genetic background. Even though backcrossing into an inbred strain is mandatory for long-term maintenance of mutations, inbred mice are themselves not suitable for behavioral analysis. Most inbred mouse strains have biochemical, anatomical, and behavioral peculiarities that complicate mutation analysis. Also, many inbred mice tend to show particularly unstable and variable behavior. In theory, F1 hybrid generations are the ideal test samples with respect to their genetic background situation, but their practical use for mutation analysis is complicated by two problems. First, their production is time and cost intensive, especially for conditional mutations. Second, due to residual environment driven variation and to unwanted effects of hybrid vigor, the gain of sensitivity in detecting mutation effects may not be as large as expected.
For practical and economical reasons, one will be often be forced to work with F2 or even F3 generations derived from crosses between two or more strains. Provided that systematic between group differences of genetic background are avoided, increasing the sample size and contrasting mutation effects statistically against the behavioral variation within the samples will permit to detect even minute mutation effects. However, it must absolutely be avoided to maintain such a population by brother sister mating beyond the F3 generation, because this will eventually result in a new and uncharacterized recombinant inbred strain.