Neural networks generate complex behaviors. The relationships between neural networks and behaviors aren’t understood well. Optogenetics is a powerful technique that enables on-demand photo-regulation of neural activity and exploration of the mechanism of neural computation. Conventional optogenetics presents three difficult challenges: 1. a hypothesis is necessary to conduct experiments; 2. optogenetics is low-throughput because different transformants are required to test each hypothesis: and 3. single-cell analysis is difficult due to the lack of single-cell-specific promoters. To overcome these difficulties, we have developed a new methodology named “functional cellomics,” which can annotate the function of neural networks in a hypothesis-free, high-throughput, and single-cell resolution manner. The key factor is the stochastic labeling of opsin by applying the Cre-lox system. We designed a genetic circuit in which two sets of lox variants, lox2272 and loxP sequences, are inserted alternately. A transcription factor, QF2w, is interposed between loxP sequences. Induced Cre excises DNA exclusively either between loxP sequences or between lox2272 sequences in every single cell. When Cre is allowed to act on lox2272 sequences, QF2w is expressed. Then, QF2w induces expression of opsin. We succeeded in the stochastic labeling of opsin dependent on the induction of Cre. However, the labeling rate of opsin was about 30% . If opsin is labeled in many neurons, it becomes difficult to distinguish the function of individual neurons. In this study, we developed a new methodology to enable precise control of the labeling rate of opsin to conduct an in-depth analysis of the function of the nervous system.
In our genetic circuit, the opsin is expressed in neurons in which the Cre-lox recombination event occurs between the lox2272 sequences. Thus, we hypothesized that the labeling rate of opsin reduces by using lox2272 variants which is difficult to be excised by Cre. We generated a library of randomized lox sequences by PCR and introduced them in S. cerevisiae. Then, we induced Cre production in the S. cerevisiae library and evaluated the excision rates of lox2272 variants by using NGS. To develop a machine learning model to predict the cleavage rate of lox2272 variants, we trained a Gaussian process (GP) model using data containing lox2272 variant sequences and their cleavage rates by Cre. We randomly chose 1000 lox2272 variants as training data and 100 variants to test prediction accuracy.
Results and Discussion
We quantified the excision rates of more than 2000 lox2272 variants by NGS and successfully found variants that have excision efficiencies ranging from 0.05% to 100%. We confirmed the results of NGS analysis by an in vivo assay using S. cerevisiae and qPCR. Then, we build a GP model and observed a high correlation between the actual cleavage rates and predicted cleavage rates of the test data (R = 0.93). We used the GP model to predict the efficiency of unevaluated lox2272 sequences and succeeded in the identification of lox2272 variants that had various cleavage rates. These results indicate the feasibility of precise control of the labeling rate of opsin in functional cellomics .