Combinatorial CRISPR Interference Library for Enhancing 2,3-BDO Production and Elucidating Key Genes in Cyanobacteria

Cyanobacteria can convert CO2 to chemicals such as 2,3-butanediol (2,3-BDO), rendering them promising for renewable production and carbon neutralization, but their applications are limited by low titers. To enhance cyanobacterial 2,3-BDO production, we developed a combinatorial CRISPR interference (CRISPRi) library strategy. We integrated the 2,3-BDO pathway genes and a CRISPRi library into the cyanobacterium PCC7942 using the orthogonal CRISPR system to overexpress pathway genes and attenuate genes that inhibit 2,3-BDO formation. The combinatorial CRISPRi library strategy allowed us to inhibit fbp, pdh, ppc, and sps (which catalyzes the synthesis of fructose-6-phosphate, acetyl-coenzyme A, oxaloacetate, and sucrose, respectively) at different levels, thereby allowing for rapid screening of a strain that enhances 2,3-BDO production by almost 2-fold to 1583.8 mg/L. Coupled with a statistical model, we elucidated that differentially inhibiting all the four genes enhances 2,3-BDO synthesis to varying degrees. fbp and pdh suppression exerted more profound effects on 2,3-BDO production than ppc and sps suppression, and these four genes can be repressed simultaneously without mutual interference. The CRISPRi library approach paves a new avenue to combinatorial metabolic engineering of cyanobacteria.


INTRODUCTION
Bio-based production of chemicals from renewable resources draws increasing attention due to growing concerns on environmental sustainability and global warming (Chae et al., 2020). Cyanobacteria are photoautotrophic prokaryotes capable of converting CO 2 to organic compounds via photosynthesis. These useful traits have led to genetic manipulation of cyanobacteria including Synechococcus elongatus PCC7942 and Synechocystis sp PCC6803 for production of industrially relevant chemicals such as 2,3-butanediol [2,3-BDO), ethanol, acetone, and isopropanol (for review see (Zhou et al., 2016;Stephens et al., 2021)]. However, the product titers are usually much lower than those from other microbial hosts (Zhao et al., 2021). In the case of 2,3-BDO, which is used in the food, fine chemical, cosmetics, and pharmaceutical industries (Liao et al., 2016), metabolically engineered PCC6803 produces 0.43 g/L of 2,3-BDO (Savakis et al., 2013), while PCC7942 confers 2,3-BDO titers from 0.496 g/L to 2.38 g/L (Oliver et al., 2013;Oliver et al., 2014). More recently, the carbon metabolism of PCC7942 was engineered to improve glucose utilization, enhance CO 2 fixation, and increase the 2,3-BDO titer to 12.6 g/L (Kanno et al., 2017). Yet, this titer is still lower than those using other metabolically engineered organisms (Kim et al., 2013;Li et al., 2015). Production efficiency poses a challenge in developing cyanobacteria for chemical production (Shabestary et al., 2018;Liu et al., 2022).
Synthetic biology and metabolic engineering have converged to allow the construction of cell factories for the synthesis of highvalue-added biochemicals (Zhao et al., 2021). In recent years, various strategies have been developed to modulate the expression of pathway genes and regulate metabolic networks (Zhao et al., 2021), such as MAGE (multiplex automated genome engineering), CoS-MAGE (co-selection MAGE) and COMPACTER (customized optimization of metabolic pathways by combinatorial transcriptional engineering). Apart from these methods, the RNA-guided genome editing tool CRISPR has been used to insert or delete pathway genes for microbial metabolic engineering (Chung et al., 2017;Pham et al., 2020;Zhao et al., 2021). CRISPR requires expression of Cas nuclease such as Cas9 and chimeric single-guide RNA (sgRNA), comprising the spacer and scaffold motif (Jinek et al., 2012). sgRNA coordinates with Cas9 to recognize the protospaceradjacent motif (PAM) on the target DNA, with the guide by the spacer sequence. By changing the spacer sequence, one may harness the Cas9/sgRNA complex to specifically recognize the target gene for programmable gene editing Doudna, 2020). We have previously shown that CRISPR improves gene integration into cyanobacterium PCC7942 with a shorter homology arm , which demonstrates the potential of CRISPR for genome engineering of PCC7942.
CRISPR was also repurposed for CRISPR interference (CRISPRi) by using a catalytically inactive SpCas9 (SpdCas9), which orchestrates with sgRNA to inhibit target gene transcription (Qi et al., 2013;Behler et al., 2018;Hsu et al., 2020). SpdCas9-based CRISPRi can repress endogenous and heterologous gene expression in PCC6803 (Yao et al., 2016) and PCC7942  in a reversible and titratable manner (Gordon et al., 2016;Yao et al., 2016). In PCC7942, integration of the SpdCas9-expressing cassette into the neutral site I (NSI) and sgRNA-expressing cassette into the neutral site II (NSII) enables repression of three endogenous genes in competing pathways at efficiencies up to 95%, resulting in a ≈12.5-fold enhancement of succinate production . However, CRISPR and CRISPRi are yet to be coupled to improve bio-derived product synthesis in cyanobacteria.
To further enhance the cyanobacterial production of 2,3-BDO, here, we first explored orthogonal CRISPR systems to integrate 2,3-BDO pathway genes and the SpdCas9-based CRISPRi system into different NS sites in PCC7942 to simultaneously overexpress pathway genes and suppress potential genes that inhibit 2,3-BDO formation. We next developed a CRISPRi library approach to inhibit fbp, pdh, ppc, and sps at different levels in a combinatorial way. The CRISPRi library approach opens a new avenue to combinatorial metabolic engineering of cyanobacteria.

Assessment of Orthogonal CRISPR Systems for Gene Integration Into PCC7942
The CRISPR system based on SpCas9 (Cas9 from Streptococcus pyogenes) can induce PCC7942 death and was harnessed for programmable gene integration into PCC7942 . There are other Cas9 orthologs derived from different microorganisms such as Streptococcus pyogenes (SpCas9) (Cong et al., 2013;Mali et al., 2013), Staphylococcus aureus (SaCas9) (Ran et al., 2015), and Streptococcus thermophilus (St1Cas9) (Kleinstiver et al., 2015). These Cas9 variants can mediate genome editing with higher specificity and are orthogonal to one another in E. coli (Sung et al., 2019). However, whether they can mediate gene integration into PCC7942 is yet to be explored. To integrate the CRISPRi library into the cyanobacterial genome using CRISPR, while avoiding mutual interference, we evaluated SpCas9, SaCas9, and St1Cas9 and used pSpCas9, pSaCas9, and pSt1Cas9 to express these three proteins ( Figure 1A). Because SpCas9, SaCas9, and St1Cas9 recognize different PAM sequences (NGG, NGGRRT, and NNAGAAW, respectively) and require different sgRNA handle regions, we also constructed plasmids (e.g. psgRNA-NSI-Sp) expressing their corresponding sgRNA handle regions and identical spacers to target the non-essential NSI site ( Figure 1A).
We first transformed the Cas9-expressing plasmid alone at different doses and observed no apparent cell death when compared with cells without the Cas9-expressing plasmid ( Figure 1B), verifying that all the three Cas9 proteins barely induced cytotoxicity in PCC7942. We next co-transformed the three cognate plasmid pairs and observed >90% cell death rate ( Figure 1C) at 250~1,000 ng, indicating that all the three CRISPR/Cas9 systems induced double strand break (DSB) and cell death at high frequencies. Importantly, SpCas9, SaCas9, and St1Cas9 systems were orthogonal to each other in PCC7942, as judged from complete cell death only when Cas9 was paired with its cognate sgRNA (e.g. SaCas9/sgRNA-Sa, Figure 1D). These data demonstrated the feasibility of using SaCas9 and St1Cas9 systems to integrate the SpdCas9-based CRISPRi system or vice versa.
After selection and segregation, the Sa group conferred significantly higher colony-forming units (CFU) than the Sp and control groups ( Figure 2C). Therefore, we randomly picked five colonies from the Sa group for PCR analyses using two primer pairs targeting the left and right junctions at the integration site ( Figure 2B). The data confirmed successful integration in colonies 1 and 4 ( Figure 2D). After re-streaking of colony 4 and culture in shake flasks, the final engineered strain (designated as 7942-BDO, Figure 2E) produced 792.8 mg/L of 2,3-BDO in 11 days ( Figure 2F).
For CRISPRi integration, we first integrated the SpdCas9 expression cassette into the NSI site of the 7942-BDO strain with the SaCas9 system to avoid mutual interference, yielding the 7942-BDOdCas strain ( Figure 3B). 7942-BDOdCas produced similar levels of BDO (818.4 mg/L) as compared with 7942-BDO (792.8 mg/L, Supplementary Figure S1), indicating that insertion of the SpdCas9 cassette did not disturb 2,3-BDO production. We next inserted the sgRNA-expressing cassette with spacers that target fbp, pdh, ppc, or sps and confirmed that separately knocking down these four genes enhanced 2,3-BDO production without apparent cell growth inhibition (Supplementary Figure S1).
To build the combinatorial CRISPRi library, for each gene, we designed sgRNAs with the spacers that have the highest on-target scores (sg1 or sg2, Figure 3C; Supplementary Table S1) for sequences located downstream the transcription start site (TSS) of fbp, pdh, ppc, or sps. We also designed a control sgRNA containing an identical handle region but with a sham spacer that targets no gene in cyanobacteria (sg0). Because gene suppression efficiency is inversely proportional to the distance from TSS (Hsu et al., 2020), sg0 would not repress gene expression; sg1 would confer weaker suppression, while sg2 would confer stronger repression. We labeled each sgRNA with a barcode (A: fbp; B: pdh; C: ppc; and D: sps; Supplementary Table S1) and constructed the combinatorial CRISPRi plasmid library ( Figure 3D, Supplementary Figure S2). Each gene has three possible suppression levels (0: none; 1, weak; and 2: strong); thus, there are 3 4 (=81) possible combinations. The combinatorial sgRNA library was integrated into the NSIII site of 7942-BDOdCas using SaCas9 to yield 7942-BDOdCas-library ( Figure 3E). After antibiotic selection, 210 colonies were picked and cultured in shaker flasks. Almost all clones conferred higher 2,3-BDO titer than the parent 7942-BDOdCas strain, with an average titer of 1226.1 mg/L ( Figure 3F).

Effects of Combinatorial Gene Suppression on 2,3-BDO Production
With the barcode, we sequenced all the clones and verified the presence of 81 sgRNA combinations (Supplementary Table S2). Figures 4A,B illustrate the relationship between the extracellular 2,3-BDO titer and sgRNA targeting site. BDO2222 (inhibiting all four genes at sg2) conferred the highest titer (1583.8 mg/L), which was almost 2-fold that of the parent 7942-BDOdCas strain (818.4 mg/L). BDO2202 (sg2 for fbp, pdh, and sps; sg0 for ppc) yielded the second highest titer, while BDO0000 (inhibiting all four genes with sg0) did not improve 2,3-BDO synthesis.

Effects of Gene Suppression on Intracellular Metabolite Levels
Since the combinatorial CRISPRi library blocked different pathways, we also analyzed the concentrations of intracellular FIGURE 2 | right junctions at the integration site. The expected PCR amplicon size is 2.7 kb. (E) Illustration of the 7942-BDO strain, which was picked after Km selection and re-streaking of colony 4 as shown in (D). (F) 2,3-BDO titer after shake flask culture of 7942-BDO strain for 11 days. The quantitative data represent the mean ± SD of at least three independent culture experiments and were analyzed using Student's t-test.

Effects of Multiple Genes and Two-Way Interactions
The aforementioned data revealed the effects of single genes on the responses (e.g. 2,3-BDO production). To gain more insight in the context of complex metabolic networks, we used response surface methodology to analyze how repression of multiple genes and their two-way interactions correlate with different responses (e.g. 2,3-BDO, F6P, AcCoA, OAA, and sucrose titers). We used a second-degree polynomial model (Table 1) for regression analysis of the experimental data from all 81 combinations as shown in Figures 4, 5. In this statistical model, Y n indicates the response, X n indicates the suppression level of each gene (fbp, pdh, ppc, or sps), X m X n represents the two-way interactions of two genes, and C n represents the effect coefficient.
For F6P (Y 2 ), C 1 has a significantly higher negative value (-104.9) than C 2 , C 3 , and C 4 , indicating that suppressing fbp reduces the F6P titer but repressing pdh, fbp, and ppc does not. For AcCoA and OAA, C 2 (−5.0) and C 3 (−0.6) values are small because AcCoA and OAA titers are low ( Figures 5B,C). Nonetheless, their values are statistically significant, indicating that suppressing pdh and ppc inhibits the production of AcCoA and OAA, respectively. Other coefficients are negligible, suggesting that two-way interactions barely influence the titers of AcCoA and OAA. For sucrose, C 4 (−24.6) is statistically significant and virtually equal to C 0 (+24.6), while other constants are mostly negligible. This echoes the finding that strong inhibition of sps completely suppresses sucrose production ( Figure 5D). These data collectively indicated that suppressing fbp, pdh, ppc, and sps significantly (p < 0.05) reduce F6P, AcCoA, OAA, and sucrose synthesis, respectively, hence promoting the carbon flux rewiring toward 2,3-BDO synthesis. However, the levels of metabolites (F6P, AcCoA, OAA, and sucrose) are not markedly influenced by two-way interactions.

DISCUSSION
Despite the promise of photoautotrophic production of chemicals by cyanobacteria, the low volumetric product titer of 2,3-BDO (at levels of mg/l) impedes the application of cyanobacteria for industrial purposes and inspires the need to enhance the product titer (Shabestary et al., 2018). Although CRISPR and CRISPRi have emerged as promising tools to modulate the metabolic networks in cyanobacteria (Gordon et al., 2016;Huang et al., 2016;Li et al., 2016;Yao et al., 2016;Xiao et al., 2018), they are yet to be coupled to improve product synthesis in cyanobacteria.
Here, we first confirmed that SpCas9, SaCas9, and St1Cas9 can be used as orthogonal CRISPR systems for genome engineering of PCC7942 (Figure 1), and we utilized SaCas9 to integrate the synthetic pathway genes for 2,3-BDO synthesis (Figure 2). In addition to Cas9, in recent years, Cas12a has gained popularity as a genome editing tool. In contrast to Cas9, Cas12a only requires a single short crRNA to program target specificity and cleaves the DNA strands into staggered ends [instead of blunt ends cut by Cas9) (for review see ]. Therefore, Cas12a may also be used in lieu of SaCas9 as an orthogonal system for integrating the CRISPRi system into the genome. Despite the successful integration of the 2,3-BDO synthesis pathway, the 2,3-BDO titer was merely mediocre at 792.8 mg/L ( Figure 2). Nonetheless, the orthogonality between SpCas9-and SaCas9-based CRISPR allows for integration of the SpdCas9based CRISPRi system using SaCas9 to specifically knockdown endogenous genes and enhance 2,3-BDO production without overt cell growth inhibition (Supplementary Figure S1). Since CRISPRi can be used to fine-tune endogenous gene expression by the sgRNA target site design, such orthogonal CRISPR/Cas systems render simultaneous heterologous gene overexpression and endogenous gene knockdown in a multiplex and intricate fashion without mutual interference and may serve as a new toolbox for the combinatorial metabolic engineering of cyanobacteria.
To increase and redirect carbon flux toward 2,3-BDO production, glucose utilization and CO 2 fixation in PCC7942 were recently enhanced together by engineering the glycolytic pathways and CBB cycle (Kanno et al., 2017). This study, similar to many other reports, attempts to improve bio-derived chemical synthesis (Wang et al., 2017;Liu et al., 2018;Yu et al., 2018;Cheah Table  S1) and was expressed from the same J23119 promoter. Each gene has three possible suppression levels (0: none; 1, weak; and 2: strong); thus, there are 3 4 (=81) possible combinations. The combinatorial sgRNA library contained the homology arms for NSIII sites (NSIII L and NSIII R) The sequences were confirmed by PCR analyses of the unique barcode using primers P1/P2. (E) 7942-BDOdCas-library which was constructed by integrating the sgRNA library into the NSIII site of 7942-BDOdCas using SaCas9. (F) 2,3-BDO titer from 7942-BDOdCas-library. After antibiotic selection, 210 colonies were picked from 7942-BDOdCas-library and cultured in shaker flasks for 11 days for extracellular 2,3-BDO analyses using GC-BID. The quantitative data represent the mean ± SD of at least three independent culture experiments and were statistically analyzed by one-way ANOVA.
Frontiers in Bioengineering and Biotechnology | www.frontiersin.org June 2022 | Volume 10 | Article 913820  Table S2). All the 210 clones were sequenced and designated based on the sgRNA combination. (B) Illustration of the targeting site (sg0, sg1, or sg2) at each gene (fbp, pdh, ppc, and sps) and the designated sgRNA combination. For instance, BDO2222 indicates that the sgRNA targeted four genes at sg2. BDO2202 indicates that the sgRNA targeted fbp, pdh, and sps at sg2 and ppc with sg0. (C-F) Effects of suppression of fbp (C), pdh (D), ppc (E), or sps (F) on the 2,3-BDO titer. The gene expression levels were measured by qRT-PCR, and suppression levels were calculated for each clone. The data represent the mean ± SD of at least three independent experiments and were statistically analyzed by one-way ANOVA. p < 0.05 was considered statistically significant.
Frontiers in Bioengineering and Biotechnology | www.frontiersin.org June 2022 | Volume 10 | Article 913820 8 et al., 2020; Lee et al., 2020;Santos-Merino et al., 2021) and adopted a design-build-test-learn cycle by deleting and inserting genes step-by-step. However, this strategy is very timeconsuming, and gene deletion often results in failure or slow cell growth during the strain improvement process (Kanno et al., 2017). This is not uncommon because deleting multiple genes often gives rise to unpredictable outcomes such as accumulation of toxic intermediates in the complex cellular environment (Zhao et al., 2021;Liu et al., 2022). For other products, several strategies, such as reducing the loss of intermediates to competing pathways (Liu et al., 2022), genome-scale modeling (Broddrick et al., 2016), and modular engineering (Liu et al., 2019), have been used to assess or tune cellular metabolism. Moreover, metabolomic techniques such as mass spectrometry (MS) and nuclear magnetic resonance (NMR) are used to identify key metabolites that contribute to product synthesis (Kato et al., 2022). For instance, isotopically nonstationary metabolic flux analysis (INST-MFA) is used to identify gene targets for increasing aldehyde production in PCC7942 (Cheah et al., 2020). Conversely, pooled CRISPRi screening is applied to mammalian cells (Du et al., 2017;Liu et al., 2017) and bacteria Peters et al., 2019) to screen gene essentiality and diverse phenotypes. Recently, the pooled CRISPRi strategy was used in cyanobacterium PCC6803 to screen genes essential for cell growth (Yao et al., 2020).
However, this methodology is yet to be combined with the orthogonal CRISPR system to enhance bio-derived chemical production in PCC7942.
In contrast to the aforementioned strategies and owing to the ability of orthogonal SaCas9 to integrate the SpdCas9-based CRISPRi library for multiplex suppression, we developed a combinatorial CRISPRi library approach for rapid screening and improvement of 2,3-BDO production in PCC7942. This methodology allowed us to inhibit fbp, pdh, ppc, and sps at different levels in a combinatorial way to enhance 2,3-BDO production ( Figure 3) and confirm that differentially inhibiting all the four genes enhances the 2,3-BDO titer to varying degrees (Figure 4). Particularly, repression of all the four genes at the strongest levels imparted additive effects to improve 2,3-BDO titers by 93.5%-1583.8 mg/L ( Figures 4A,B).
Furthermore, we combined the CRISPRi library data and regression analysis to quantify the influences of each gene and gene pairs on the titers of desired 2,3-BDO and undesired intracellular metabolites (Figures 4, 5; Table 1). In addition to verifying that simultaneously inhibiting all the four genes (fbp, pdh, ppc, and sps) enhances the 2,3-BDO titer and decreases the respective titers of metabolites downstream of the gene, we found more significant effects of fbp and pdh suppression on 2,3-BDO production as their effect coefficients are larger than those of ppc and sps (Table 1). These effects may be attributed to the roles of these Frontiers in Bioengineering and Biotechnology | www.frontiersin.org June 2022 | Volume 10 | Article 913820 9 genes on carbon flux partition. fbp regulates F6P synthesis in the CBB cycle, which can be converted to sucrose ( Figure 3A) or glycogen (Hickman et al., 2013) as storage compounds. Since the carbon for 2,3-BDO synthesis can be provided directly by G3P without F6P ( Figure 3A), inhibiting fbp reduces unnecessary F6P formation without significant impairment of 2,3-BDO synthesis, thus exerting profound effects on 2,3-BDO titer increase. Conversely, PYR is an important branching point for cyanobacterial production of 2,3-BDO (Oliver et al., 2013;Savakis et al., 2013;Oliver et al., 2014); thus, increasing the intracellular PYR pool and redirecting the carbon flux toward 2,3-BDO production is critical. pdh is responsible for converting PYR to AcCoA to enter TCA cycle or produce other byproducts such as acetate (Hirokawa et al., 2020); thus, inhibiting pdh markedly blocks the carbon flux toward AcCoA, TCA cycle, and other byproducts, hence augmenting 2,3-BDO production. Our finding concurs with the recent reports that overexpressing pdh increases the carbon partition into AcCoA and acetate/isopropanol production (Hirokawa et al., 2020), while pdh knockdown through antisense RNA elevates PYR partition (Cheah et al., 2020).
In contrast, inhibiting sps only suppresses sucrose synthesis, hence imparting weaker effects. In addition, ppc is helpful to OAA replenishment and ppc inhibition, indeed, improves 2,3-BDO formation. However, OAA can be recycled by TCA cycle itself, which might offset the effect of inhibiting the ppc gene and result in less significant effect on 2,3 BDO production. We also unveiled that the two-way interactions between fbp/pdh and fbp/ppc slightly decreases the 2,3-BDO titer, but their effects are overshadowed by suppressing individual genes ( Table 1). By and large, the effect of two-way interactions between gene pairs on metabolite titers are less significant than that of single genes (Table 1), indicating that these four genes can be repressed simultaneously to enhance 2,3-BDO production without mutual interference.
In summary, we identified the orthogonal CRISPR/Cas9 systems for the integration of synthetic pathway genes and the CRISPRi system to regulate the gene expression/suppression levels in PCC7942 to intricately modulate the metabolic flux and product synthesis. We also developed a combinatorial CRISPRi library strategy that allows for systematic and rational analysis to elucidate the effects of individual genes and gene pairs on 2,3-BDO production. The CRISPRi library approach enables us to yield a strain that improves 2,3-BDO production by 93.5% and is less time-consuming and laborious than traditional approaches. Since re-engineering glucose utilization and reinforcing CO 2 assimilation substantially increase 2,3-BDO synthesis (Kanno et al., 2017), our approach may be exploited to investigate the roles of genes in glycolytic pathways and CBB cycle to further increase and redirect the carbon flux to 2,3-BDO production.
Although fast-growing cyanobacteria strains, such as UTEX 2973 (Yu et al., 2015), have been identified, their genetic backgrounds remain poorly understood. Nonetheless, our CRISPRi library strategy may be extrapolated to unravel genes critical for bioproduct synthesis in these fast-growing strains. Furthermore, our combinatorial CRISPRi library may be combined with metabolomics (Kato et al., 2022) to unveil the roles of genes essential for production of other chemical products and accelerate the application of cyanobacteria for biosynthesis.

MATERIALS AND METHODS
Culture and Transformation of S. elongatus PCC7942 PCC7942 cells were cultured using BG-11 medium in a 30°C incubator (600SR, Hipoint), with illumination from continuous cool white fluorescent light (intensity≈70 μmol//m 2 ·s). For suspension culture, PCC7942 cells were cultivated using 40 ml medium in a 250-ml shaker flask (gyratory shaking at 100 rpm) until OD 730 reached ≈0.6 (for transformation) or until 11 days (for 2,3-BDO production). For plate culture, PCC7942 cells were streaked onto 90-mm plates containing 40 cm 3 BG-11/agar medium supplemented with 1 mM sodium thiosulfate and cultured for 7-9 days until the colonies developed. Plasmid transformation into cells was performed as described previously .

Construction of Plasmids and Engineered 7,942 Strains
All the cloning protocols were performed with standard methods using E. coli DH5α strain as the host. pSpCas9, pSaCas9, and pStCas9 that expressed SpCas9, SaCa9, and St1Cas9, respectively, were constructed previously (Sung et al., 2019). For sgRNA expression, we used psgLacZ-Sp, psgLacZ-Sa, and psgLacZ-St (Sung et al., 2019) as the backbone but swapped the spacer sequence to target the NSI site (yielding psgRNA-NSI-Sp, psgRNA-NSI-Sa, or psgRNA-NSI-St) or the NSII site (yielding psgRNA-NSII-Sp or psgRNA-NSII-Sa). The plasmids for integration into the three neutral sites (NSI, NSII, or NSIII) were constructed based on pSyn-1, pSyn-2, or pSyn-3 (Invitrogen). These plasmids contained the spectinomycin resistance gene (Spc R ), a multiple cloning site (MCS), downstream of sc promoter, a rrnB transcription terminator, and flanking sequences homologous to the NSI, NSII, or NSIII site. We PCR-amplified the inducible smt promoter and subcloned it into pSyn-1 to replace the sc promoter and yielded psmt-NSI. For pNSII, we replaced Spc R and sc promoter by the kanamycin resistance gene (Km R ) and LacO1 promoter to yield pLacO1-NSII. For NSIII, Spc R was replaced by the chloramphenicol resistance gene (Cm R ) to yield pCm-NSIII.
To clone the synthetic 2,3-BDO pathway, we chemically synthesized alsS from Bacillus subtilis, alsD from Enterobacter cloacae, and adh from Clostridium beijerinckii (Genomics BioSci & Tech, Taiwan) with flanking AvrII and XhoI. The three genes were subcloned into the MCS of pLacO1-NSII to yield pNSII-23BDO. We transformed the wild-type PCC7942 strain with only 2000 ng of pNSII-23BDO as the control group. In parallel, we co-transformed 2000 ng of pNSII-23BDO with 500 ng of pSpCas9 and 500 ng of pSgRNA-NSII-Sp (or 2000 ng of pNSII-23BDO with 500 ng of pSaCas9 and 500 ng of pSgRNA-NSII-Sa). The cells were streaked to plates containing kanamycin (Km), and colonies were picked and segregated as described . The colony numbers were counted using a colony counter, and colony-forming units (CFUs) were obtained. After segregation, five colonies were picked for PCR analyses using two primer pairs targeting the left and right junctions at the integration site. The colonies with correct integration without contaminating bands were subcultured, stored at -80°C, and designated as the 7942-BDO strain.

Construction of CRISPRi Library and Engineered 7,942 Strain
To build the combinatorial CRISPRi library, we first PCR-amplified SpdCas9 together with the trc promoter from pBac-SpdCas9 (Hsu et al., 2020). The PCR amplicon was subcloned into psmt-NSI to yield pNSI-dCas9. pNSI-dCas9 (2000 ng) was transformed into the 7942-BDO strain, and SpdCas9 integration into the NSI site was verified as described previously. The resultant strain was subcultured, stored at -80°C, and designated as 7942-BDOdCas.
For the sgRNA library, we chose to inhibit four genes (fbp, pdh, ppc, and sps) and calculated the on-target scores (Supplementary Table  S1) using online software benchling (https://www.benchling.com/). We designed and chemically synthesized the partially overlapping oligonucleotides 1 to 8 to encode the sgRNA cassette, unique barcode, and Cm R (Supplementary Figure S2). The sequences for oligonucleotides 3 and 4 were designed to encode a non-targeting spacer (sg0) or the spacer that has the highest on-target score (sg1 or sg2). Oligonucleotide 8 was designed to encode the unique barcode (Supplementary Table S1). The sequences for oligonucleotides 3, 4, and 8 are different for different gene targets. Using these oligonucleotides for gene assembly, we synthesized DNA amplicons encoding sgRNA cassettes with unique barcodes and Cm R by overlapping PCR. These sgRNA target sg0, sg1, and sg2 for fbp (A), pdh (B), ppc (C), and sps (D) as shown in Supplementary  Table S1. As shown in Supplementary Figure S2, the three DNA amplicons for the fbp sgRNA (A) were cloned into pCm-NSIII to yield three different psgRNA(A): psgRNA (A2), psgRNA (A1), and psgRNA (A0). In step 2, we cloned the three DNA amplicons for pdh (B) into three psgRNA(A) to yield nine different psgRNA (AB). In step 3, we cloned the three DNA amplicons for ppc (C) into nine different psgRNA (AB) to yield 27 unique psgRNA (ABC). In step 4, we cloned the three DNA amplicons for sps (D) into the 27 different psgRNA (ABC) to yield 81 different psgRNA (ABCD). The 81 different plasmids were verified using primers P1 and P2 (Supplementary Figure S2), owing to the unique barcode sequences. The 81 plasmids were pooled and transformed into 7942-BDOdCas for integration into the NSIII site using the SaCas9 system as described previously to yield the 7942-BDOdCas-library. The cells were spread to plates, and colonies were picked for analyses.

2,3-BDO Analysis and Intracellular Metabolite Analyses
The extracellular 2,3-BDO titer was analyzed by GC-BID (Shimatsu, Japan) using the TG-WAXMS column (Shimatsu). The cells were cultured in 40 ml medium in the shaker flask for 11 days. After harvesting and centrifugation (6,000×g, 15 min), the supernatant was diluted 10-fold and analyzed using GC-BID. Meanwhile, 2,3-BDO standard (Sigma) was diluted to 200, 100, 50, 25, and 10 mg/L and analyzed similarly to establish the standard curve. The 2,3-BDO titers were determined based on the area of the signal and the standard curve.
For the analysis of intracellular metabolites F6P, AcCoA, OAA, and sucrose, the cells were cultured in 40 ml BG-11 medium for 11 days and 1 ml of the supernatant was withdrawn. After centrifugation, the cells were lysed using 100 μL TE buffer containing 2 mg/ml lysozyme in a 37°C water bath for 20 min, followed by centrifugation (12,000×g, 5 min). Subsequently, 500 μL supernatant was analyzed by HPLC-MS (Shimadzu) using Hypersil ™ BDS C18 HPLC Column (5 μm, 10 × 250 mm, Thermo Fisher) and 5% acetonitrile as the mobile phase. The standard F6P, AcCoA, OAA, and sucrose (Sigma) were diluted to 1,000, 500, 100, 10, and 1 mg/L and analyzed similarly to generate the standard curve.

Quantitative Real-Time Reverse Transcription PCR
For gene expression, a single clone was picked and subcultured in 40 ml medium in a 250-ml shaker flask for 11 days, followed by centrifugation and lysis with the lysozyme as described previously. Approximately 3-5 μg total RNA was extracted using TRIzol ® and mixed with SYBR ® Green Master Mix and gene-specific primers. The mixture was subjected to real-time PCR as described . The gene expression levels were calculated by software (LightCycler ® 96 SW 1.1, Roche) using the glgc gene as the internal control. The data were normalized with the expression levels in the 7942-BDOdCas strain as the reference. The suppression levels were calculated as (1-expression level)×100%.

Statistical Analysis
All the quantitative data represent the mean ± SD of at least three independent culture experiments and were analyzed using Student's t-test or one-way ANOVA. p < 0.05 was considered significant. The CRISPRi library data, including gene suppression levels, titers of 2,3-BDO, F6P, AcCoA, OAA, and sucrose, were analyzed by the statistical software Design-Expert ® 12 using the response surface methodology as the statistical model. We used a second-order polynomial equation for regression analyses of how inhibition of multiple genes and their two-way interactions correlate with different responses (e.g., 2,3-BDO, F6P, AcCoA, OAA, and sucrose titers) from the experimental data of all 81 combinations of the CRISPRi library (See Results).

DATA AVAILABILITY STATEMENT
The original contributions presented in the study are included in the article/Supplementary Material. Further inquiries can be directed to the corresponding author.