Multiple memory mechanisms? The long and the short of it.

Synaptic plasticity is known to be intrinsically linked to cytoskeletal remodelling. In a new study published in this issue of The EMBO Journal, Rust et al have therefore generated and analysed a forebrain-specific mouse knockout of the actin-severing factor n-cofilin. Intriguingly, these mice exhibit a particular form of learning impairment-intact working memory but impaired spatial reference memory, correlating with a deficit in long-lasting long-term potentiation (LTP). Rust et al (2010) have generated genetically modified mice that lack the actin filament depolymerizing protein n-cofilin (n-cof) in the adult forebrain, a protein important for remodelling of the cytoskeleton in neurons and thus for both regulating spine morphology and receptor trafficking. The authors show that these cytoskeletal changes are important for altering the efficacy of the synaptic connections between neurons, and for certain kinds of learning. Notably, the n-cof mice were impaired on the classic Morris watermaze task, in which they are required to form a long-term association between a particular spatial location and a hidden escape platform (a spatial reference memory task). In contrast, the n-cof mice were unimpaired on two tests of spatial, short-term working memory (spontaneous and rewarded alternation), which required the animals to choose arms on the basis of their relative familiarity (i.e. choose the arm that has been visited less recently). Strikingly, the n-cof mice performed as well as the controls on the spatial working memory tasks but showed no evidence of any learning whatsoever on the associative, spatial memory watermaze task. What is equally striking is that the pattern of results obtained with n-cof mice is an inverse mirror image of data obtained from GluA1 AMPA glutamate receptor subunit knockout mice. GluA1 / mice show normal acquisition of associative long-term spatial reference memory tasks like the Morris watermaze but are impaired on short-term spatial memory tasks, including the same spontaneous and rewarded alternation tasks as used by Rust et al (Zamanillo et al, 1999; Reisel et al, 2002; Sanderson et al, 2007). Furthermore, on a six-arm radial maze task, GluA1 / mice are as good as wild-type controls in learning to discriminate between which three arms contain food rewards and which three arms are never baited (normal spatial reference memory), but are unable to discriminate between arms based on recent familiarity, and thus are unable to avoid going back into arms that they have already been into on that visit to the maze (impaired spatial working memory; Schmitt et al, 2003). Thus, the complementary learning deficits in n-cof and GluA1 / mice provide independent genetic proof for a double dissociation of shortand long-term spatial memory processes, thus confirming the existence of two distinct memory mechanisms. It has been argued that spatial working memory tasks are simply more sensitive to hippocampal disruption than spatial reference memory tasks. The double dissociation argues against this. Furthermore, whereas other dissociations between different kinds of memory are based on the processing of different kinds of information (e.g. spatial versus non-spatial), the distinction between n-cof and GluA1 / mice is based on different processing of the same information type (i.e. using spatial cues). Rust et al also investigated putative neurobiological substrates that might underlie associative long-term memory. According to the authors, the n-cof mice fail to exhibit a robust, long-lasting form of LTP. The correlated inability to form long-term memories supports the idea that the neurobiological substrates underlying associative memory is longlasting LTP in the forebrain. It is also tempting to speculate that the short-term memory deficits in GluA1 / mice could reflect the absence of a short-lasting plasticity in these animals (Hoffman et al, 2002; Romberg et al, 2009). However, caution is required when equating shortand long-term forms of LTP to short and long-term memory. Earlier studies with gene-targeted mice disprove a simple correlation between LTP and learning. For example, GluA2deficient mouse lines exhibit normal or even enhanced LTP but dramatic spatial learning impairments (Shimshek et al, 2006). These results are also potentially at odds with Rust et al (2010), who suggest that the trafficking of GluA2 containing AMPA receptors during synaptic remodelling might underlie LTP. Unfortunately, the involvement of GluA1 and GluA3 during synaptic remodelling by n-cofilin was not investigated. It may be that GluA1 actually interferes with the n-cofilin-dependent mobilization of GluA2 containing receptors. Interestingly, we recently showed that GluA1 / mice, despite exhibiting impaired short-term spatial memory, actually display enhanced long-term spatial memory under certain conditions (Sanderson et al, 2009). This reflects the fact that in normal animals, short-term memory can be The EMBO Journal (2010) 29, 1790–1791 | & 2010 European Molecular Biology Organization | All Rights Reserved 0261-4189/10 www.embojournal.org