Bidirectional titration of yeast gene expression using a pooled CRISPR guide RNA approach

Significance

Traditional genetic modulation approaches are typically restricted to binary perturbations, single-sided titrations (either graded up- or down-regulation), or individual gene expression modulation. Here, we utilize a library approach that is able to simultaneously modulate gene expression in a metabolism-wide manner. This library, when coupled with next-generation sequencing, allows for the identification of gene perturbations that would have been missed by classic approaches. For the examples tested, this library identified targets that improved growth on alternative carbon sources as well as improved production of betaxanthins.

Abstract

Most classic genetic approaches utilize binary modifications that preclude the identification of key knockdowns for essential genes or other targets that only require moderate modulation. As a complementary approach to these classic genetic methods, we describe a plasmid-based library methodology that affords bidirectional, graded modulation of gene expression enabled by tiling the promoter regions of all 969 genes that comprise the ito977 model of Saccharomyces cerevisiae’s metabolic network. When coupled with a CRISPR-dCas9–based modulation and next-generation sequencing, this method affords a library-based, bidirection titration of gene expression across all major metabolic genes. We utilized this approach in two case studies: growth enrichment on alternative sugars, glycerol and galactose, and chemical overproduction of betaxanthins, leading to the identification of unique gene targets. In particular, we identify essential genes and other targets that were missed by classic genetic approaches.

Classic genetic approaches for identifying gene targets have traditionally been limited to binary modifications consisting of either deletion or strong overexpression (1–3). This approach is not ideal for identifying all targets, including essential genes that need to be repressed or any genes whose expression only requires modest modulation. Moreover, even most emerging genetic tools including CRISPR/CRISPRi, TALEN, and RNAi still mainly invoke binary gene expression modulation (4–6). While these approaches have been successful in ascribing functional annotation and identifying dominant gene targets, they are often insufficient when seeking to optimize cellular metabolic function, or when attempting to identify other salient genetic targets. For example, deletion library screens are unable to identify essential genes whose knockdown may in fact be positively correlated with phenotype. Moreover, it has long been recognized that optimal expression of genes resides at intermediate expression levels between the extremes of complete gene deletion and strong overexpression (7). Traditional approaches fail to identify these targets that are only effective at intermediate expression levels. Thus, new approaches are needed to both enable high-throughput identification of targets and complement the limitations of most genetic screens.

Transcriptional element libraries [such as promoter libraries (8)] have traditionally been used to enable graded gene expression for individual pathway applications for phenotypes such as small-molecule production (9) and the consumption of alternative carbon sources (10, 11). However, these techniques are often not accessible to genome-wide, high-throughput implementation. Recent advances to establish graded expression-level libraries have utilized RNAi approaches with either micro-RNAs or full-length complementary mRNAs to afford two levels of gene knockdown (12). More recently, a CRISPR interference (CRISPRi) based single-guide RNA (sgRNA) library was implemented in mammalian cells to create a broader spectrum of down-regulation to identify causative genetic targets for fitness phenotypes (13). In addition, one group utilized dCas9 fused to VP64-p65-Rta (VPR) to identify finely tuned, ideal levels of gene expression for 168 different genes, by tiling each one 21 times (14). While these approaches are all working toward an accessible high-throughput library for graded expression ranges, they fall short of achieving both knockdown and overexpression capacity and thus lack the full range of possible gene expression perturbations.

Here, we showcase the implementation and utility of a plasmid-based library methodology that affords bidirectional titration of yeast gene expression in a manner that is complementary to traditional genetic approaches (Fig. 1A). We leverage our previously reported STEPS approach to design a panel of single-guide RNAs that tile promoter regions for all 969 genes represented in the ito977 genome-scale metabolic model (15). These sgRNAs were synthesized in a pooled format through collaboration with the US Department of Energy (DOE) Joint Genome Institute (JGI) Synthesis Science Program. By coupling this sgRNA library with a previously established CRISPR-dCas9 system, we can take advantage of dCas9 fusions with Mxi1 for heterochromatin formation or VPR for recruitment of the mediator complex, thus creating graded up and down regulations, respectively (16).

Fig. 1.

Enrichment of a bidirectionally titrated metabolism-wide library. (A) A metabolism-wide, bidirectional titration panel of sgRNAs was synthesized based on the ito977 model of metabolism and enabled via dCas9 fused to either Mxi1 or VPR. These fusions allow for graded knockdown or overexpression of targeted genes of interest, respectively. (B) Enrichments were performed on the alternative carbon sources of galactose and glycerol using the Mxi1 and VPR libraries independently to allow for ease of deep sequencing. Final populations from each enrichment were deep-sequenced to identify statistically significantly enriched guides. (C) Volcano plots of guide enrichment for each condition. These plots show that the majority of guides in the library were depleted, indicating that most perturbations were outcompeted in growth on alternative carbon sources.

Results and Discussion

Construction of an sgRNA Library for Metabolism-Wide, Graded Expression.

The aforementioned sgRNA library was designed in a pooled format to target broad classes of metabolic function including carbohydrate and lipid metabolism (pool 1), energy and cofactor metabolism (pool 2), amino acid and nucleotide metabolism (pool 3), and housekeeping/other (pool 4) (Dataset S1 A–D). More specifically, oligos containing unique sgRNAs were designed to tile upstream of the promoter region of 969 metabolic enzyme genes in Saccharomyces cerevisiae. This DNA was synthesized in collaboration with the JGI and was then amplified and cloned into both the dCas9-Mxi1 and -VPR backbone (SI Appendix, Fig. S1). These libraries were then transformed and either propagated in broth or on plates in order to amplify the libraries. Plasmids were then harvested to obtain 100 micrograms of DNA and subsequently sequenced via polymerase chain reaction (PCR) amplification of the guide RNA region followed by next-generation sequencing (NGS) from both Broth-based and Agar plate-based transformations (SI Appendix, Fig. S2 and Table S1) to validate coverage. Plate-based transformations were originally theorized to allow for an equal representation of all sgRNAs, as this process does not have the same outgrowth step as in liquid-broth propagation that could bias representation through potential competitive growth. However, these experiments demonstrated that broth-based transformations actually resulted in the best representation of the individual library members, and thus purified DNA from this condition was used for subsequently screening experiments (SI Appendix, Fig. S1). To demonstrate the utility of this approach for the identification of unique targets as well as their optimal expression levels, we evaluated several growth and production phenotypes as proof-of-concept experiments.

Library Enrichments on Alternative Carbon Sources Identify Novel Targets.

As in the first set of case studies, we utilized growth-based enrichments on alternative carbon sources using the common laboratory strain of S. cerevisiae, BY4741. Here, we chose to select enrichments on both glycerol and galactose due to their industrial relevance (galactose comprises a significant portion of marine biomass, and glycerol is produced in large quantities as a byproduct of biodiesel transesterification (17, 18)) as well as extensive prior studies on these carbon sources using classic approaches (19). Prior to conducting a deep-sequencing analysis of enrichment using these pools, we first validated the enrichment capacity of these libraries using a repeated subculturing, colony isolation, and sequence analysis approach (Fig. 1B and SI Appendix, Fig. S3). In these trials, individual beneficial guide RNAs were isolated and confirmed to not only show enrichment over time, but also confer a growth advantage when retransformed.

Following these validation tests, a full-scale growth enrichment process accompanied by deep-sequencing analysis was used to globally identify targets along with their optimal expression levels (Fig. 1B). To do so, we chose a partial subculture condition to detect both enrichment and depletion as well as preventing overenrichment by a few dominant targets. Macroscopic analysis of statistical enrichment and depletion of guide RNAs within the library illustrates that the majority of guides were depleted in the postenrichment pools, indicating that most perturbations to gene expression are outcompeted in this assay when growing on glycerol and/or galactose (Fig. 1C and SI Appendix, Sheet 2).

Given the high-resolution aspect of this dataset (i.e., having both target identifications along with their optimal expression levels), multiple modes of analysis are possible (20). For example, a cluster analysis allows for a full mapping of gene expression-level phenotype enhancement for both carbon sources (Figs. 1C and 2A and SI Appendix, Fig. S4). Each major cluster links together targets whose optimal expression profile and patterns are similar. Initial evaluation of these trends and patterns indicates an overrepresentation of guide enrichment at moderate levels of expression (both for knockdown and overexpression) (Figs. 1B and 2 A and B and SI Appendix, Fig. S4). At the onset, these data illustrate the complementary nature of this approach to coarse-level, binary modification of gene expression. Several gene targets emerge whereby moderate knockdown greatly enhances growth relative to the complete knockout. As examples, knockdown of IPK1 and TPS2 resulted in improved growth on glycerol, but the complete deletion of these targets results in substantially reduced growth (Fig. 2B). Additional examples include identifying guides targeting FUN26 whose growth showed improvement far over the counterpart deletion of this gene (SI Appendix, Fig. S5). On the opposite end of the spectrum, glycerol-growth–enhancing targets like GRS2 overexpression are only optimal/functional when targeted at modest overexpression sgRNA localization regions. This point is especially poignant when comparing growth with both lower and higher levels of expression for this target (Fig. 2C). Examples such as these illustrate the depth of new targets identifiable with this approach.

Fig. 2.

Cluster analysis and subsequent ideal guide confirmation of glycerol and galactose enrichments. (A) A representative set of clusters for enriched guides are illustrated for both galactose and glycerol. In this representation, gene ID is on the vertical axis, and the bidirectional titration is on the horizontal axis going from strong knockdown to strong overexpression. The predominant level of expression in each cluster is highlighted by blue or orange. Results illustrate a strong enrichment of moderate expression levels across these conditions. (B) An example of intermediate knockdown targets identified from glycerol enrichment is compared with the knockout, demonstrating that moderate regulation was optimal. (C) As an example of a moderate expression target, GRS2 overexpression is highlighted wherein the guide furthest upstream (i.e., slightest up-regulation) was significantly enriched following serial culturing and provided a far improved growth over the strong expression guide and the wild type. This ability to tune gene expression to specific levels of overexpression is not seen in traditional overexpression libraries. (D) Essential genes were uniquely identifying with this method as down-regulation targets. As examples, gpi18 was identified from the glycerol enrichment and dim1 from the galactose enrichment. Both guides were recloned and showed improved growth, whereas the deletion is lethal, and thus these represent new targets unseen with traditional approaches.

Beyond visually confirming the premise that different expression levels are required for different subsets of genes, these clusters can be analyzed to determine underlying metabolic trends for these growth phenotypes. For example, through Gene Ontology (GO) analysis, a significant number of phosphate-related metabolic genes were seen to be enriched for the medium knockdown level in galactose selection (utilizing the SGD GO analysis tool) (SI Appendix, Fig. S6). More specifically, the most represented genes within this knockdown level were associated with phosphorous (44.4%) metabolic processes. Gene expression clusters for glycerol consumption indicated significant (P value < 0.05) enrichment of genes associated with organic acid synthesis (42.9%) in the case of the highly up-regulated cluster (VPR −150). Phosphorous metabolism is also important for glycerol catabolism, wherein we observed the Mxi1 -500 cluster enriched with genes significantly (P value < 0.05) associated with phosphorous metabolic processes (50%) and a further enrichment of genes associated with carbohydrate phosphorylation (28.6%) in the low-knockdown cluster, VPR +80. Thus, large-scale trends of metabolism can be extracted through this analysis.

The approaches described here are indeed complementary to more classic genetic approaches such as gene deletion libraries, especially in the capacity to identify essential genes as critical targets for gene knockdowns. The sgRNA library synthesized here contained 86 essential genes. Within this set, 8 of these targets are identified as down-regulation targets in galactose, and 10 of these genes we identified as down-regulation targets in glycerol (with an additional 10 of these genes enriched as overexpression targets) (SI Appendix, Fig. S7 and Dataset S2). These examples in particular are poignant demonstrations of the benefit of an expression titration library, as these targets would not be identified in a full knockout collection. As examples, knockdowns were identified for both DIM1 and GPI18, both of which are essential genes. The knockdown condition substantially improved growth rates in their respective alternative carbon sources, whereas the complete deletion of these genes is lethal (Fig. 2D). Thus, this approach identified a unique set of targets for which galactose and glycerol-improvement phenotypes were not previously ascribed.

Finally, overlap of the targets identified here with previously identified literature targets that serve to improve or hinder growth on alternative carbon sources was evaluated and found to be limited (SI Appendix, Fig. S7). Of particular note, this study utilized a competitive growth enrichment assay, whereas most targets in literature were isolated from knockout collections and utilized individual target analysis. Nevertheless, there is some overlap of targets. For example, HXT17 is a known overexpression target for improving growth on galactose (21), and this same target was identified through this screen. In some cases, other well-known individual overexpression targets for improved growth on galactose, including PGM2, were not identified as statistically enriched, likely due to being outcompeted in this assay. However, sgRNAs associated with down-regulation of PGM2 were in fact significantly depleted, thus demonstrating the implication of this enzyme on galactose consumption in general (SI Appendix, Table S2). Additional known targets such as the detrimental effect of aim10 and gal10 deletion growth on galactose (22) are instead complemented by the enrichment for only moderate levels of down-regulation in our library. Collectively, the coupling of our sgRNA library with a competitive growth enrichment adds a dimension of enriching for fitness and optimal expression levels. Thus, while we have identified some known targets, we predominately highlight here the knowledge of optimal expression levels. As with any screening process, the mode of library screening is important for target identification, and this library is likewise suitable for individual candidate analysis as has been conducted with the gene deletion libraries.

Identifying Novel Targets for Improved Betaxanthin Production.

As our second overall case study, we sought to utilize a production-based screen rather than growth selection to demonstrate the versatility of this library. In this regard, we evaluated targets that would improve product secretion of betaxanthins in a S. cerevisiae strain background (BX3) overexpressing the CYP76AD5 tyrosine hydroxylase from Beta vulgaris and l-Dopa dioxygenase (DOD) from Mirabilis jalapa (23) in addition to the negative-feedback–resistant mutant of DAHP synthase (ARO4_K229L) (24) (SI Appendix, Fig. S8). Using this strain and the fluorescent nature of this product, we were able to employ a parallel microplate-based approach suitable for quantifying total (rather than simply intracellular) betaxanthin—a divergence of work in the literature. With this approach, our roughly 10³ screening capacity demanded a two-layer screening approach that first looked for high variability in each of the four synthesized sgRNA pools followed by a deeper analysis of the most variable pools (Fig. 3A). For example, an initial screening of the Mxi1 library highlighted that Mxi1 pool 1, pool 2, and pool 4 produced similar variation in fluorescence, whereas pool 3 (amino acid and nucleotide metabolism) displayed minimal variation above the pool median (Fig. 3A). As a result, we performed a deeper screen across pool 1 by evaluating over 400–500 colonies (equivalent to roughly 70–80% of the coverage).

Fig. 3.

Betaxanthin production phenotype improvement utilizing a pooled screening approach. (A) Initial, pool-based screening of Mxi1 pools was conducted using a plate-based assay followed by deep screening, as shown with pool 1. Characterization of identified sgRNAs imparting an improved fluorescence was identified via Sanger sequencing and recloned into the parent strain, BX3. (B) Identified targets when analyzed and in multiplex showcase several unique targets achieving up to 30% increase in fluorescence over the parent strain. In addition, the knockouts of the two, best performing, single guides were made in BX3 and their fluorescence measured, indicating that these targets would not have been identified through a traditional approach. (C) Tiling of the ZWF1 promoter region with guide RNAs showed an ideal level of expression (i.e., more fluorescence) at the medium level of knockdown. This ideal midlevel of expression is uniquely identified in a titratable expression library.

A total of four unique targets (in either expression level or identity) were identified from the analysis of pool 1. To investigate interactions between these isolated targets, we constructed combinations of these sgRNAs and tested them against their individual sgRNA targets (Fig. 3B). While almost all combinations produced fluorescence values that approached the best sgRNA in the pair, the combination of the ZWF1-RKI1 sgRNAs showed positive epistasis and produced the highest fluorescence of any single or double sgRNA combination and values that exceed improvements seen in the literature (25).

As with the growth selections, the expression titration afforded by this pooled approach was able to identify novel targets not previously ascribed to this phenotype by more traditional approaches. For example, as RKI1 is an essential gene, its utility as a perturbation target would be missed under classic approaches. Likewise, intermediate expression is often optimal as can be seen in the direct comparison of individual guides targeting PDR11 and LAC1 that resulted in increases in fluorescence compared to the parent strain, while their respective knockouts actually decreased fluorescence when compared to the parent strain (Fig. 3B). These targets would have been missed in a traditional knockout screen. Perhaps the target that best exemplified this optimality is ZWF1, a gene identified as a slight knockdown target. Indeed, by establishing a panel of sgRNAs associated with repression of ZWF1, the intermediate knockdown (sgRNA targeting the region −371 basepairs upstream of the open reading frame) resulted in the greatest improvement, whereas strong knockdown had a strongly negative phenotype (Fig. 3C). In this instance, we hypothesized that an intermediate level of ZWF1 expression would properly balance nicotinamide adenine dinucleotide phosphate (NADPH) and precursor availability for l-DOPA production, whereas too strong of a knockdown would simply improve 3-dehydroshikimic acid (DHS) levels, as previously validated (16). The isolated ZWF1 gene has been previously characterized as an important target to control precursor flux into the shikimate pathway (26). While zwf1 deletion properly balances the ratio of phosphoenolpyruvate (PEP) and erythrose-4-phosphate (E4P) needed for enhanced flux into the shikimate pathway, it also compromises flux by limiting the amount of NADPH available for the conversion of 3-DHS to shikimate. Our library identified an intermediate level of ZWF1 expression that led to an increase in betaxanthin production. This level of intermediate expression would have been missed in a knockout screen. This example showcases the importance of evaluating multiple expression levels, as many beneficial targets can be missed otherwise.

Finally, deep screening of 384 clones from VPR pools 1 and 4 indicated several novel targets requiring modest overexpression, including DIT2, FBA1, and the alpha-1,2 mannosyltransferase, GPI10. These nonpathway-specific targets provide additional key insight into pathway regulation. Therefore, this library not only can identify novel targets and levels of expression, but could be used as a discovery tool to learn more about pathway regulation and metabolic flux.

Conclusion.

Taken together, these two case studies highlight the importance of utilizing bidirectional titration of gene expression for identification of novel gene targets as well as their optimal level of expression. This approach is highly complementary to classic genetic approaches that tend to only create binary changes in gene expression. In many cases, we find gene knockdowns of essential genes that greatly impact phenotypes of interest yet have been missed thus far due to a reliance on binary genetic changes. Likewise, we identify targets for which only moderate levels of modulation are optimal. The unique ability to screen for novel targets as well as their optimal level of expression using this library provides a powerful tool for studying genotype–phenotype relationships. We foresee that the knowledge gained using this tool will push forward a second wave of genetic analysis in the yeast S. cerevisiae.

Materials and Methods

Data Availability.

All NGS data can be found at National Center for Biotechnology Information (NCBI) under accession number PRJNA625119. In addition, all code used for analysis can be found at the following link: https://github.com/emkbowman/Bi-directional-Titration-NGS-Analysis. Analyzed NGS data are available under Dataset S2. Transformed or purified sgRNA library is available upon request.

Strain Design.

DH10β was used to propagate all yeast expression vectors including those described by Mumberg (27) and pJED103-based (28). To amplify plasmids, Escherichia coli strains were cultivated in Luria broth (LB) or super optimal broth (SOB) media (Teknova) supplemented with 50 µg/mL ampicillin or kanamycin (Sigma) with 225 rpm orbital shaking at 37 °C. Yeast strain BY4741(EUROSCARF) was cultured in yeast synthetic complete (YSC) medium containing roughly 6.7 g/L of yeast nitrogen base (Difco), 20 g/L glucose (MP Biomedicals) and 1× CSM-URA, CSM-URA-LEU, or CSM-URA-LEU-HIS (MP Biomedicals) depending on the required auxotrophic selection. The S. cerevisiae strain BX3 (URA3::pCCW12-MjDOD-tADH1-pTDH3- BvCYP76AD5-tTDH1-pTEF1-ScARO4K229L-tENO2) was constructed from BY4741 (Mat a; his3Δ1; leu2Δ0; met15Δ0; ura3Δ0) by digestion of the pCMC0759 integration vector, provided as a gift from the Dueber Laboratory, University of California, Berkeley (25).

Transformations.

Transformations were performed as in ref. 28. Briefly, in order to transform Gibson cloning reactions, 3 µL of threefold diluted Gibson reaction (NEB, 254 ng backbone) was mixed with 50 µL of electrocompetent E. coli DH10β and electroporated (2 mm Electroporation Cuvettes) with a BioRad Genepulser Xcell at 2.5 kV. For transformation of ligations, 3 µL ligation mix (NEB) was directly added to 50 µL competent cells and transformed as above. For ampicillin-marked plasmids, transformants were resuspended in 500 µL of SOB, plated on LB agar supplemented with 50 µg/mL ampicillin, and incubated at 37 °C overnight. Transformants of kanamycin-marked plasmids were recovered in 500 µL SOB for 30 min at 37 °C and then plated. Individual clones were picked into SOB media containing 50 µg/mL antibiotics and incubated at 37 °C overnight. Plasmids were then miniprepped (GeneJET Plasmid Miniprep Kit, Thermo Scientific) and sequence-verified via Sanger sequencing. The Frozen EZ Yeast Transformation II Kit (Zymo Research) was used to transform plasmids into yeast. Briefly, between 200 ng and 1 µg of plasmid was mixed with 20 µL chemically competent cells prepared by manufacturer’s instructions and 200 µL EZ Solution III followed by incubation at 30 °C for 45 min. Transformations were then plated on YSC+agar plates containing either CSMURA, CSM-URA-LEU, or CSM-URA-LEU-HIS and incubated at 30 °C for 2 d. Individual colonies were randomly picked in triplicate into 1 mL of YSC media and incubated at 30 °C for another 2 d. For long-term storage, all yeast strains with the exception of transformed libraries were stocked in 15% glycerol and kept at −80 °C in sterile flat-bottomed microtiter plates (Corning) covered with an adhesive aluminum foil seal (Thermo Scientific) and plastic lid.

Cloning Procedures.

Oligonucleotides were purchased from Integrated DNA Technologies. Sequences and annotations can be found in SI Appendix, Table S3. PCR and anneal/extend double-stranding reactions were performed with Q5 DNA Polymerase from New England Biolabs according to the manufacturer specifications. Digestions were performed according to manufacturer’s (NEB) instructions. PCR products and digestions were cleaned with a QIAquick PCR Purification Kit (Qiagen). All vectors were dephosphorylated with Antarctic Phosphatase (NEB) according to the manufacturer’s instructions and heat-inactivated for 15 min at 65 °C.

All plasmids for expression of dCas9 were derived from the pJED103 vector series acquired from AddGene catalog #46921 (28). All RGR plasmids were derived from the dCas9-Mxi1and dCas9-VPR plasmids previously reported (16). To construct new RGRs, the dCas9-Mxi1 and dCas9-VPR RGR cloning vectors were linearized at the 5′ end of the sgRNA scaffold using the SpeI enzyme, and then a 100-bp fragment containing a variable HH-sgRNA sequence was inserted via Gibson assembly to create the full TEF1p-HH-sgRNA-HDV-TKC27t cassette. The 100-bp insert fragments were constructed by an anneal/extend PCR using two 60-bp oligos (IDT) with 20-bp overlaps at their 3′ ends. Multiplexed sgRNAs were constructed using PCR with primers MD1522/MD1523 followed by Gibson Assembly into Mxi1 sgRNA vectors linearized with EcoRI.

Library Enrichment on Alternative Carbon Sources.

BY4741 was transformed in a pooled format utilizing a scaled-up Geitz transformation. These libraries were cultivated in 2% Raffinose, CSM-L, YNB to maintain the plasmid and prevent any errant glucose-based enrichment. Libraries were stocked in 1-mL aliquots, mixed 1:1 with 40% glycerol, and stored at −80 °C . For use, these aliquots were thawed and grown in a 50-mL flask containing 2% Raffinose YSD-L until the OD₆₀₀ was over 1.0. Libraries were then diluted down to an optical density (OD) of 0.1 and resuspended into 50 mL of media comprised of CSM-L, YNB, and either 6% Glycerol or 4% Galactose. Biological duplicates were used for library enrichment (i.e., two separate enrichments were performed for each condition). The 6% Glycerol media was adjusted to a pH of 4.0. These cultures were monitored for growth, and once the OD₆₀₀ reached 1.0, they were serially diluted into a fresh flask containing new media with a starting OD₆₀₀ of 0.1 for three times before conducting a yeast mini prep followed by NGS analysis.

NGS Sample Prep and Analysis.

Yeast mini preps were used as a template and amplified utilizing Q5 polymerase and following manufacturer’s instructions (Primers in SI Appendix, Table S3). To avoid sequence bias, only 20 amplification cycles of PCR were used. The product was purified via gel electrophoresis and extracted utilizing a gel purification kit from Qiagen prior to being submitted for library construction and analysis at the UT FBS Sequencing Core. All raw NGS data can be found at NCBI under accession number PRJNA625119. Analysis of sequencing reads was performed by obtaining individual read counts and aligning to the sgRNA library, utilizing code that can be found at https://github.com/emkbowman/Bi-directional-Titration-NGS-Analysis (SI Appendix, Table S1). These were normalized to the total number of reads and then averaged. A two-tailed t test was run on each guide to confer significance of enrichment, determined by a P value of < 0.05. Finally, the log fold change was calculated for each guide utilizing Excel and clustered using Cluster 3.0, with city-block distance and complete linkage settings. This analysis was visualized in Java TreeView, with pixel settings centered at log 1.0.

Growth Analysis of Select Guides.

Guides identified via either Sanger sequencing or NGS were either retransformed (if isolated via Sanger sequencing to confirm enrichment of beneficial guides) or recloned utilizing cloning procedures mentioned above and transformed utilizing the EZ yeast transformation according to manufacturer’s instructions and plating on synthetic selective media (CSM-L, 2% Raffinose). Three clones were picked from each transformation and grown in 2% Raffinose synthetic defined media and glycerol stocked. Glycerol stocks were taken out and grown overnight in 2% Raffinose synthetic defined media and then diluted down to an OD₆₀₀ of 0.1 in the same 6% Glycerol or 4% Galactose enrichment media from which they were identified. OD measurements were collected roughly every 24 h.

Plate Reader Measurement of Betaxanthin Fluorescence.

All betaxanthin-producing strains were characterized using the Cytation 3 MicroPlate Reader (BioTek Instruments). At minimum, biological triplicates were used for each strain and each condition, with up to six replicates used for more sensitive fluorescence assays. Strains were grown from glycerol stock for 30 °C at 72 h in CSM-URA-LEU media and then back-diluted 100× into CSM-URA-LEU supplemented with 10 mM ascorbic acid and 50 mM beta-alanine. Strains were grown until fluorescence no longer increased from day-to-day (typically around 72–120 h). To measure bulk fluorescence, cultures were diluted either 20× or 40× in phosphate-buffered saline (PBS) (to ensure the detector was not saturated). To measure supernatant fluorescence, cultures were spun down at 1,600 g for 4 min, and the supernatant was diluted 20× in PBS. To measure intracellular fluorescence, cells were spun down and washed once with PBS, resuspended in 100 μL PBS, and diluted 20×. Fluorescence was normalized to OD₆₀₀ for intracellular fluorescence measurements.