CRISPR/Cas9‐RNA interference system for combinatorial metabolic engineering of Saccharomyces cerevisiae

Abstract The yeast Saccharomyces cerevisiae is widely used in industrial biotechnology for the production of fuels, chemicals, food ingredients, food and beverages, and pharmaceuticals. To obtain high‐performing strains for such bioprocesses, it is often necessary to test tens or even hundreds of metabolic engineering targets, preferably in combinations, to account for synergistic and antagonistic effects. Here, we present a method that allows simultaneous perturbation of multiple selected genetic targets by combining the advantage of CRISPR/Cas9, in vivo recombination, USER assembly and RNA interference. CRISPR/Cas9 introduces a double‐strand break in a specific genomic region, where multiexpression constructs combined with the knockdown constructs are simultaneously integrated by homologous recombination. We show the applicability of the method by improving cis,cis‐muconic acid production in S. cerevisiae through simultaneous manipulation of several metabolic engineering targets. The method can accelerate metabolic engineering efforts for the construction of future cell factories.


| INTRODUCTION
Industrial biotechnology uses cell factories to produce therapeutical proteins, antibiotics, enzymes, fuels, and chemicals. To achieve favorable process economics, one needs to optimize the cell factories, where performance metrics as titer, rate, and yield are improved.
Strain development programs for the products that are not native to the host are very costly and take a long time. The required investment in biotechnology companies that develop novel strains and processes is typically above $50 Mio. During the strain development, hundreds to thousands of strain variants are engineered in iterative design-build-test cycles. High-throughput strain construction and screening in the range of 10 5 -10 6 variants are possible when a biosensor indicating the product presence is available (Zhang, Jensen, & Keasling, 2015); however, this is seldom the case. construction via polymerase chain reaction (PCR), cloning, and transformations. The cloning and strain construction is typically performed at 10-50-μl scale, where the high cost of specialized reagents also contributes to the high price of the strain development.
Controlled downregulation of gene expression, however, remains a challenge. Gene downregulation is often a more desirable metabolic engineering strategy than complete gene inactivation, and, in case of essential genes, the only option. Catalytically inactivated dCas9, also in a variant coupled to a transcriptional repressor, has been applied for downregulation, but typically multiple gRNA binding sites need to be tested to obtain the desired repression level (Deaner & Alper, 2017;Jensen et al., 2017;Zalatan et al., 2015). Alternatively, RNA interference (RNAi) has been demonstrated to allow more precise control of gene downregulation (Crook, Schmitz, & Alper, 2013;Drinnenberg et al., 2009;Si, Luo, Bao, & Zhao, 2014;Suk et al., 2011).
In this study, we aimed to develop a method that would allow multiplex upregulation and downregulation of several genes by combining the advantages of the CRISPR/Cas9 system and RNAi.

| Biobricks amplification and plasmids construction
The oligonucleotides, biobricks, and plasmids used in this study are listed in Tables S2, S3,   . DNA manipulations in E. coli were carried out according to standard procedures. The clones with correct inserts were identified by colony PCR, and the plasmids were isolated from overnight E. coli cultures and confirmed by sequencing. The list of the constructed vectors can be found in Table S5.
For the construction of overexpression cassettes for in vivo assembly, there are five part types in our assembly standard (promoters, genes, terminators, upstream homology arm, and downstream homology arm). The specific overhangs flanking individual parts were designed and introduced at 5′ end of the forward and reverse primers as described in Table S3. All DNA parts were PCR amplified using Phusion U DNA polymerase according to the manufacturer's instructions. DNA fragments were gel purified and were assembled by consecutive procedures of USER reaction, T4 ligation, and PCR amplification of the assembled expression cassettes as follows: 17 μl of gel-purified DNA fragments containing similar molar ratio of all parts was mixed with 2 μl of CutSmartTM buffer and 1 μl of USER enzyme (New England BioLabs). The mixes were incubated for 25 min at 37°C followed by 10 min at 25°C. After USER reaction was complete, 1 μl of T4 ligase, 3 μl of ligase buffer, and 6 μl of water were added. The mix was incubated for 5 min at room temperature.
Two to three microliter of this ligation mix were used as a template for the final PCR reaction in order to amplify the whole expression cassette. The fragments were purified from the gel and used for yeast transformation (0.7 pmoles per transformation). For fragments smaller than 500 BP, ca. 2 pmoles of the fragment were used per transformation.

| Construction of shRNAs
The small hairpin RNA (shRNA) constructs were composed of two DNA fragments. The first fragment contained approximately 250 BP sense sequence of the target gene under the control of the constitutive promoter and an 81-BP sequence spanning intron 1 from Schizosaccharomyces pombe rad9. The second fragment contained the antisense sequence of the target gene together with terminator and an 81-BP sequence of intron 1 from S. pombe rad9. Sense, antisense, promoter, and terminator fragments were amplified by PCR.
The corresponding fragments for generating sense and antisense cassettes were assembled via USER-ligation-PCR as described above. The intron sequence was implemented in the primer overhang.
Sense and antisense DNA fragments were introduced together with UP-and DW-fragments for CAN-1 and were assembled into the genome of S. cerevisiae at CAN-1 locus via homologous recombination.

| Construction of dsRNA
To generate double-stranded RNA (dsRNA) constructs, the target gene was PCR amplified and assembled with PGK1p and TEF1p promoters, and ADH1t and RPM9t terminators in convergent direction via USER-ligation-PCR as above.

| Yeast strains construction
All strains used in this study are listed in Table S1. The integrative plasmids were NotI-linearized and transformed into S. cerevisiae cells using the lithium acetate protocol (Gietz & Woods, 2002). The cells were selected on SD medium selecting for URA, HIS, LEU and TRP markers.
For the selection of strains carrying KanMXsyn and CloNatMXsyn, the ammonium sulfate in the SD medium was replaced with 1 g L −1 monosodium glutamate. The medium was supplemented with 200 μg ml −1 G418 sulfate and 100 μg ml −1 nourseothricin. The correct transformants were confirmed by PCR using primers described in Supplementary Table S2. Flow cytometry data were analyzed and interpreted using FlowJo software.

| Muconic acid production in S. cerevisiae
At least 12 single colonies of each transformant were cultivated in 24well plate with air-penetrable lids (EnzyScreen, NL) to test for the production of CCM. The colonies were inoculated in 1-ml SD medium without uracil, histidine, and leucine and grown at 30°C with 250 rpm agitation at 5-cm orbit cast for 24 hr; 300 μl of the overnight culture was used to inoculate 3 ml of defined mineral medium (pH 6.0) in 24-deep well plate and incubated for 72 hr at the same conditions as above. Experiments were done in triplicates. At the end of the cultivation, OD 600 was measured in microplate reader BioTek Synergy MX (BioTek). The culture broth was spun down at 3,500 g, and the supernatant was analyzed for CCM concentration using High-performance liquid chromatography (HPLC).

| Quantification of CCM and its intermediates by HPLC
The samples were diluted five times with water and then analyzed for 45 min using Aminex HPX-87H ion exclusion column with eluent 1-mM H 2 SO 4 flow of 0.6 ml min −1 . The temperature of the column was 60°C. The UV detector (Dionex) was used for detection of CCM (250 nm), PCA (220 nm), and catechol (220 nm). CCM, PCA, and catechol concentrations were quantified by comparison with the standard calibration curve.

| qRT-PCR analysis
The expression level of ZWF1 in recombinant yeast strains was determined by quantitative real-time PCR (qRT-PCR). Samples for RNA isolation were taken from the cells grown in the mineral medium for 24 hr in triplicates. Sampling procedure and total RNA extraction were performed as previously described  We recombined 20 BP-UPTAG and DNTAG sequences from yeast knock-out library (Giaever and Nislow 2014) to obtain final sequences of 60 BP. These sequences were BLASTed against S. cerevisiae genome to select the sequences with low homology that were used as overhang sequences for assembly.
The gene BioBricks included standard 6-8 BP USER overhangs for easy assembly with promoters and terminators (Figure 1a) Figure 1b).
In the past few years, several CRISPR/Cas9 mediated multiplex genome engineering approaches were demonstrated. Mans et al.
(2015) explored the potential of CRISPR/Cas9 to combine gene

| Validation of the method for downregulation of gene expression using RNAi
RNAi machinery is present in multiple eukaryotes, including some yeast species, such as Naumovozyma castellii (Crook et al., 2013;Drinnenberg et al., 2009;Suk et al., 2011). Although S. cerevisiae does not harbor an active RNAi pathway, this pathway can be restored by introducing Argonaute (AGO1) and Dicer (DCR1) genes from Naumovozyma castellii into the genome of S. cerevisiae. In this study, we sought to reconstitute the RNAi machinery in S. cerevisiae to allow controlled downregulation of multiple target genes. We first implemented AGO1 and DCR1 from Naumovozyma castellii into S. cerevisiae In our study, the hairpin length of approximately 250 BP was used. It should also be noted that in vivo assembly of sense and antisense fragments provides a more straightforward approach to introduce shRNA compared with the cloning of inverted repeats via restriction-ligation cloning in E. coli as in Yoshimatsu and Nagawa (1989).

| Engineering CCM production through multiplex engineering
In the previous study, we have constructed a S. cerevisiae CCM producing strain ST3058 (Skjoedt et al., 2016).  Figure 3a). It has been reported that PCA-DC was a rate-limiting step for the CCM flux (Curran, Leavitt, Karim, & Alper, 2013;Weber et al., 2012). For this reason, we integrated KpAroY.B and KpAroY.Ciso genes in multiple copies into long 113 terminal repeats (LTRs) of retrotransposon of the TY4 family (Maury et al., 2016). As the transformants were expected to have different copy numbers of the expression vector, we screened 12 randomly selected clones to test for CCM production. The best isolate of ST3058 produced 400 mg L −1 CCM in defined mineral medium and was chosen for evaluating the CRISPR/Cas9-RNAi method. We implemented Cas9, AGO1, and DCR1 into the best isolate of ST3058, resulting in strain ST3639 that was suitable for testing our method.
For the test, we designed to vary the expression of four native genes that could influence the CCM flux: was investigated by qRT-PCR (Figure 3e). In the strain, where the only implemented modification was ZWF1 downregulation, the expression level decreased by 80% or 95% when weak and strong promoters were driving shRNA expression, respectively. In the strain, where additional three genes were overexpressed, the downregulation of ZWF1 was at 35% or 55%, again depending on the promoter for shRNA. The positive effects of TKL1 overexpression and ZWF1 downregulation on CCM production are in agreement with a previous report, where ZWF1 was though deleted rather than downregulated (Curran et al., 2013;Weber et al., 2012). Both genes are involved in the pentose phosphate pathway, and the modification of their expression possibly improved the supply of the aromatic amino acids precursor-erythrose 4-phosphate. The positive effects of these modifications need to be further confirmed in fed-batch fermentations in controlled bioreactors.
In the past few years, there has been a growing interest in applying CRISPR methods for combinatorial metabolic engineering. Vanegas, Lehka, and Mortensen (2017)  for CRISPRd. As a proof-of-concept, the trifunctional CRISPR system was used to increase β-carotene production via simultaneous upregulation of HMG1, downregulation of ERG9, and deletion of ROX1. Furthermore, 2.5-fold improvement in the display of an endoglucanase on the yeast surface was obtained by combinatorial optimization of several metabolic targets. At this point, the selection of efficient gRNA for CRISPRi remains a challenge and multiple variants need to be tested. This increases the number of strains that need to be constructed for testing downregulation targets or combinations of downregulation targets with overexpression targets.
During this work, a study was published by Si et al. (2017)  shDNA were shown to be more effective for downregulating gene expression.
Our method combines the advantages of RNAi for precise downregulation, of CRISPR/Cas9 for efficient genomic integration and of yeast homologous recombination for the multiple fragment assembly.
The method is convenient for testing defined combinations of multiple upregulation and downregulation targets for metabolic engineering.
The method can facilitate the strain development efforts by increasing the throughput and decreasing the cost of strain construction. In the future, it can be further applied for generating combinatorial libraries of strain variants by using mixes of BioBricks rather than specific BioBricks. The library approach is particularly attractive if a highthroughput method for screening the strain libraries is available, as is the case with muconic acid, where a biosensor has been reported (Skjoedt et al., 2016).