Fine-Tuning Gene Expression For Model Microorganisms Based On N-Terminal Coding Sequences

N-terminal coding sequences（NCS）generally refer to gene coding sequences starting after start codon and having a length of 30 to 96 nt,which regulate gene expression from the translation level.Modifying NCS of target genes controls expression level of these genes,which has the advantages of simple molecular manipulation and significant regulatory effect.NCS also coordinate with other gene regulatory elements,such as promoters,to achieve fine-tuing of gene expression.However,due to complex regulatory mechanism of NCS and difficulty of rational design,large-scale de novo design of NCS in model microorganisms has not been achieved,which limits the application of NCS in synthetic biology fields,such as fine-tuning of gene expression and optimization of genetic circuits.For the de novo design of NCS in three model microorganisms,including Escherichia coli（Gram-negative bacteria）,Bacillus subtilis（Gram-positive bacteria）and Saccharomyces cerevisiae（eukaryotic microorganisms）,synthetic 30-nt NCS libraries were firstly constructed using green fluorescent protein（GFP）as reporter protein,and then divided into high or low bins according to different fluorescences based on fluorescence-activated cell sorting（FACS）,thus obtaining DNA sequences of NCS in each bin by high-throughput sequencing.Secondly,data analysis and modeling of the obtained NCS were performed using statistical and machine-learning techniques.Based on machine learning model,model-driven design of NCS could be achieved.Finally,NCS was used to weaken the expression of acetic acid in E.coli and to enhance the production of ovalbumin in B.subtilis and D-limonene in S.cerevisiae,in order to verify the effectiveness of NCS in synthetic biology.The main results of this study are as follows:（1）In three model microorganisms,NCS libraries of approximately 250,000 clones in size were constructed by degenerate primers,and continuously spanned five orders of magnitude of gene expression level regulation.Each NCS library was divided into two sub-libraries according to up-or down-regulated fluorescence intensity by flow cytometry sorting.After high-throughput sequencing,17,134,5,521 and 15,579 valid sequences were obtained in E.coli,B.subtilis and S.cerevisiae,respectively.Statistical analysis was performed using minimum free energy（MFE）,t RNA adaption index（t AI）and different amino acid classification characteristics as indicators,showing that these above factors were significantly different between high-and low-expression libraries（P<0.05）.（2）Machine learning was used to construct 5 different compared NCS prediction models in E.coli,B.subtilis and S.cerevisiae,respectively,to predict the effect of NCS on gene expression levels in different microorganisms.Combining machine learning evaluation indicators and biological experiments of randomly-generated NCS to measure the performance of different comparative models,the model integrating deep learning and multi-view learning performs the best,of which prediction accuracy in E.coli,B.subtilis and S.cerevisiae was 66.9%,69.9%and 65.5%,respetively.Besides,the model-driven design of de novo NCS further up-regulated the expression of GFP to 3.65 times,4.21 times and 18.25times the relative fluorescence intensity of that without NCS.（3）According to the different production characteristics of three model microorganisms,different genes of interest were chosen to verify the effective application of NCS.Attenuating the expression of acetate kinase（Ack A）by NCS reduced the production of by-product acetate in E.coli BL21（DE3）strain by 94%.By adding NCS to the 5’end of ovalbumin gene,the expression of ovalbumin in B.subtilis 168 showed a 1.71-fold increase.The expression of limonene synthase in S.cerevisiae was also enriched by NCS,and therefore the titer of D-limonene in the engineered S.cerevisiae strain increased 48%,reaching 541.45 mg·L^-1.