Sophisticated genetic tools that make brain cells responsive to light have now been used in mice to trigger a memory connected with a particular place, and to switch its association from negative to positive, or vice versa.

We often believe that our memories are accurate, but in fact they can be malleable, changing over time as recollections become less precise or as events that never happened are falsely remembered1. There is also another way in which memory can change. The memory of a romantic first meal out with a partner may take on a different mood when the relationship falters. That of a favourite family beach in summer may be destroyed by witnessing a swimming tragedy there. In these cases, memory of the place remains accurate, but the positive associations with that place are lost. In a paper published on Nature's website , Redondo et al.2 investigate the neural basis of this selective change.

Our memories are representations of past experiences that are believed to be encoded in networks of neurons that fire together or in sequence. The representation of a particular place - a 'where' memory - is encoded in a brain structure called the hippocampal formation, which is embedded within the medial temporal lobe. A separate representation in the amygdala of the brain encodes a 'what' memory, which recalls whether one feels good about a place (a positive valence) or has marked it off as dangerous (a negative valence). These two representations are thought to become connected during learning. The amygdala also has direct downstream connections to the action and endocrine systems that are involved in approach and avoidance3.

Redondo and colleagues investigated the separate representations of 'where' and 'what' memories, and looked at whether the associations between them could be changed. To do this, they used several molecular-engineering tools, including optogenetics, which enables the manipulation of neurons in response to light. The authors genetically engineered male mice such that, when the antibiotic doxycycline was removed from their daily diet, a light-sensitive protein called channelrhodopsin-2 (ChR2) was able to be expressed. Depending on the genetic tools used, this took place in neurons of either the hippocampus or the amygdala in which the gene c-fos was active - a response to neural activity and learning. When doxycycline was briefly removed from the diet and the animals were given either contextual fear conditioning (a small electric shock) or reward conditioning (interaction with a female mouse), memory encoding caused c-fos activation in these brain areas and resulted in labelling of the 'where' or 'what' memory neurons with ChR2.

After training, the authors added doxycycline to the diet of the mice once again, preventing any further ChR2 labelling and ensuring that only the training memory representations could be activated by blue light. Redondo and co-workers then performed a place-preference test, in which blue light was turned on whenever the mice entered a designated target zone. This selectively activated the ChR2-labelled neurons and the networks to which they had been associatively connected. Fear-conditioned mice duly moved away from the target zone, because the negative memory was optogenetically reactivated when they were in this area, whereas reward-conditioned mice stayed longer in the target zone, recalling the positive memory.

The key aim of the study was to determine whether it is possible to change the 'what' association linked to a 'where' representation. The authors were able to do this, but the additional conditioning did not involve returning the animals to the training arena. Instead, the researchers optogenetically reactivated the appropriate hippocampal neurons of fear-conditioned mice while allowing them access to a female. The outcome was a successful switch of the original aversive association with the target zone into an attractive association. A switch from attraction to fear could also be achieved. However, the amygdala representations of aversion or attraction stayed as they were - their downstream connections were unchanged.

These data imply that functional connectivity between memory representations in the hippocampus and the amygdala can be altered. Analysing these connections under the microscope, Redondo et al. found that additional conditioning led to decreases in the proportion of hippocampal-activated neurons in the amygdala that represented the original conditioning. Finally, they observed that if a hippocampal representation was formed by fear conditioning, and then had its valence changed through subsequent conditioning with reward, the animals would later display less fear when returned to the original fear-conditioning box.

It has long been apparent that memories can be changed from bad to good, or vice versa. What is so intriguing about this study is that the memory representations associated with a place are dissected into their network components and, rather than re-exposing the animals to the training situation to achieve a change, light is used to selectively reactivate the representation of the 'where' component of a memory and then change its 'what' association.

Contemporary theories of learning are less about stimuli and responses than about the internal representation of events. For example, work done in the past 10 years has focused on associatively activated event representations in learning; that is, on the acquisition of memories that can later be evoked by a reminder cue of a specific stimulus or by the act of returning to a particular place4. A key finding of this work was that associatively activated event representations can successfully substitute for the events that created them, with the possibility that new learning will successfully associate new information to the event memory evoked by the reminder cue. This may result in an altered response to that cue.

Optogenetic techniques, used so ingeniously by Redondo and colleagues, complement and expand on this previous work4 and on the classical 'disconnection' approach, which involves unilaterally damaging two structures on opposite sides of the brain to establish the importance of their anatomical connectivity for learning5, 6. The use of ChR2 cell labelling in a way that is temporally controlled and dependent on neural activity, followed by optogenetic reactivation of the representation, takes us closer to identifying the networks that underlie certain forms of memory.

There are limitations to this optogenetic strategy, notably when sequences of neural firing are the essence of the memory7 (for instance, the memory of a musical tune or a sequence of actions). This is because the resulting labelling will represent the sum of all the neurons that upregulate the activity-regulated genes, such as c-fos, rather than any explicit representation of sequence. Through optogenetics and exceptionally careful design of behavioural studies, molecular engineering is nonetheless shedding light on our understanding of the underlying physiological networks of memory