CRISPRi screen for enhancing heterologous α-amylase yield in Bacillus subtilis

Abstract

Enhancing yield during bacterial enzyme production could have positive economic and environmental impacts. For cell factories, such improvements in yields could potentially be obtained by fine-tuning the metabolic processes and their regulatory mechanisms for gene candidates. In pursuit of such candidates, we performed RNA-sequencing of two α-amylase producing Bacillus strains and predict hundreds of putative novel non-coding transcribed regions. Complex operons that are regulated by a wide variety of transcription factors, non-coding and structured RNAs add to the challenge of finding yield-affecting candidates. Surprisingly, we found that non-coding genomic regions are proportionally undergoing the highest changes in expression during fermentation (75% of novel RNA predictions had absolute logFC > 2). Since these classes of RNA are also understudied, we targeted the corresponding genomic regions with CRIPSRi knockdown to test for any potential impact on the yield. From differentially expressed annotations, including both novel candidate and prior annotated ncRNAs, we selected 53 non-coding candidates. The targeting with CRISPRi knockdowns transcription in a genomic region on both the sense and the antisense strand. Thus, the CRISPRi experiment cannot link causes for yield changes to the sense or antisense disruption. Nevertheless, we observed on several instances with strong changes in enzyme yield. The knockdown targeting the genomic region for a putative antisense RNA of the 3' UTR of the skfA-skfH operon led to a 21% increase in yield. In contrast, the knockdown targeting the genomic regions of putative antisense RNAs of the cytochrome c oxidase subunit 1 (ctaD), the sigma factor sigH, and the uncharacterized gene yhfT decreased yields by 31 to 43%.

The impact of PrsA over-expression on the Bacillus subtilis transcriptome during fed-batch fermentation of alpha-amylase production

“…The overall experimental setup is as previously described in (Geissler et al, 2022). In summary: B. subtilis strain 168 ΔspoIIAC ΔamyE Δapr ΔnprE ΔsrfAC was maintained at 4 °C on LBGG medium.…”

“…In summary: B. subtilis strain 168 ΔspoIIAC ΔamyE Δapr ΔnprE ΔsrfAC was maintained at 4 °C on LBGG medium. The AMY JE1 (sequence label je1zyn in (Geissler et al, 2022)) was inserted by Splicing by Overlapping Extension (SOE) linear recombinant transformation, together with the commercial sigA promoter sequence P4199 and chloramphenicol marker, in the pel locus. The PrsA over-expressing strain (referred to as ‘+prsA’ strain) had the insert by SOE of P4199, prsA , and spectinomycin marker in the amyE locus.…”

“…All samples were immediately mixed with 5 ml 100% ethanol and stored on dry ice. The RNA extraction and purification method is the identical phenol-chloroform protocol of (Geissler et al, 2022). RNA libraries and sequencing were conducted by BGI Hong Kong with DNBseq in single-ends of 50 bp length.…”

“…The 36 samples (triplicates, 2 strains, 6 timepoints) were sequenced in 3 batches with technical replicates for QC (Supplementary Table S1). The computational analyses were conducted in an adapted workflow of (Geissler et al, 2022) (doi: 10.5281/zenodo.4534403), which provides a pipeline in a Snakemake framework nested in computational reproducible Anaconda environments (Koster and Rahmann, 2012). In concordance with the read quality assessment of FastQC (version 0.11.8) (Simon Andrew), any adapter contaminations were removed with Trimmomatic (version 0.39) for up to 2 seed mismatches at a minimal 10 bp sequence overlap and 30 bp palindromic overlap (Bolger et al, 2014).…”

“…Differential expression for pairwise comparisons between the strains at each of the 6 time points and within each strain along the time axis (Fig 1 C) were assessed with DESeq2’s Wald test. Similar to the analysis presented in (Geissler et al, 2022), the pairwise tests were weighted in a stage-wise procedure to guarantee an overall relative to the number of annotations: Each annotation was screened for dynamic expression with a log-ratio test against a static expression model before confirming which of the pairwise tests had changes in expression. The screening and pairwise tests included a linear factor in the regression models to account for potential batch effects.…”