Targeting redundant gene families: A multiplexed, tissue-specific CRISPR toolbox for Arabidopsis genetic screens

Summary

Genome-scale targeted CRISPR libraries for forward genetic screens in plants are powerful tools for functional analysis, but they suffer from limited spatial control, single sgRNA design, and poor handling of genetic redundancy. We develop multiplexed CRISPR libraries in which each construct contains two sgRNAs that simultaneously target multiple members of a gene family. The libraries can also function at the cell-type-specific and tissue levels. A double-barcoding strategy enables efficient tracking and identification of sgRNA combinations at the plant level without individually sequencing each line. Using this platform, we generate over 1,000 Arabidopsis lines that express sgRNAs targeting 707 transporter genes across 114 gene families involved in nutrient uptake. The multiplexed design increases gene coverage and editing efficiency, underscoring its improved targeting capability to reveal hidden phenotypes. This toolbox provides a scalable resource for multi-targeted genome editing and spatially precise forward genetic screens in plants.

Graphical abstract

Highlights

• •Large multiplexed CRISPR libraries overcome redundancy, improving coverage and efficiency
• •A tissue-specific, multiplexed CRISPR library enables spatially precise editing in plants
• •Barcode-based “CRISPR-GuideMap” links sgRNAs to each plant, unmasks hidden phenotypes
• •SWEET13/14 transports trans-zeatin and functions redundantly with SWEET11/12

Anfang et al. present a tissue-specific, multiplexed CRISPR toolbox for plants that uses paired sgRNAs and barcoded libraries to target redundant gene families. Applied to Arabidopsis nutrient transporters, it boosts editing efficiency and gene coverage, enabling spatially precise screens that uncover phenotypes hidden by functional redundancy.

Introduction

The ability to perform genetic screens has tremendously impacted plant research. The creation of genetic variability via induced mutagenesis increases phenotypic variation and serves as the cornerstone for plant breeding programs, yielding a rich pool of phenotypic diversity for selection and improvement.¹ Over the past few decades, genetic screens have revealed novel phenotypes and their associated genes, but the majority of plant genomes remain uncharacterized.^{2,3,4,5,6,7,8} One of the main reasons for this is genetic redundancy. Whole-genome analyses have revealed that the majority of genes in plant genomes belong to gene families.^9,10,11,12 Therefore, a single-gene loss-of-function mutant is not likely to yield an observable phenotype because other family members, which have similar sequences and functions, compensate for the lost activity, a phenomenon termed functional redundancy.^{5,10,12,13,14} Partial, conditional, or complete functional redundancy among family members explains much of the absence of visible phenotypes in single-gene deletion mutants in a variety of plants.^2,15,16 This redundancy can occur in various ways, such as when multiple effectors modulate the same component, influence different components within the same pathway, or participate in redundant pathways contributing to a single process.¹⁷ In many cases of functional redundancy, a loss-of-function phenotype is only noticeable when multiple genes are knocked out.^{13,16,18,19,20,21}

Functional redundancy is vital in plants as it ensures adaptability and resilience in response to a changing environment, thus significantly influencing evolutionary outcomes.²² Redundant genes may be derived from convergent evolution, but the primary source of functional redundancy is the presence of paralogous gene groups.²³ These gene families originate from duplication events that have resulted in multiple copies of genes with overlapping functions.¹¹ Such duplications can occur at the genome-scale through large segmental duplications, including whole-genome duplications, or locally through single-gene duplications. The total duplicate gene content in Arabidopsis, for example, ranges from 63% to 78%, with variation in estimates due to differences in gene models, methodology, and parameters of the computational pipelines employed.^24,25

Forward genetics strategies, which involve the application of mutagens such as ethyl methanesulfonate, involve the use of unbiased methods to identify genes that play essential roles in specific biological phenomena.^26,27 However, these methods have limitations, including the completely random nature of mutations and the inability to overcome functional redundancy. One approach to overcome redundancy and reveal visible phenotypes is to simultaneously target sequences conserved in genes from the same family. In the context of large-scale tools, microRNAs have been used to silence multiple genes simultaneously with conserved sequences, even when mismatches are present in the targeted region.^28,29 However, microRNA-based silencing achieves only partial and variable gene suppression rather than complete knockout.

In recent years, CRISPR-Cas9 technology has been applied to overcome these challenges, allowing the development of precise gene editing methods that can also work at large scales.^3,6,30,31,32 The CRISPR Multi-Knock approach addresses the challenge of redundancy by targeting multiple genes within a family using one single-guide RNA (sgRNA).³ This method has uncovered previously hidden phenotypes and revealed novel biological functions in Arabidopsis. However, the approach was constrained by sequence requirements. For many gene families, the lack of sufficiently conserved sequences between family members meant that only two-thirds of the genes in the genome that belong to families could be targeted. Furthermore, when targeting multiple genes simultaneously, the increasing number of mismatches between the sgRNA and some of its targets often resulted in reduced editing efficiency. The Multi-Knock library, similar to all genome-scale libraries developed thus far in Arabidopsis and other crops, has employed one sgRNA per vector and plant.^30,33,34 sgRNA multiplexing allows targeting several genes per plant at high efficiency and has been used to explore gene function in multiple plant species^7,35; however, multiplexing has not been applied yet at large scales. In addition, current genome-scale CRISPR tools have not been utilized yet to screen at a library scale for cell-type-, tissue-, or organ-specific activities.

Another limitation of large-scale CRISPR library screens is the difficulty in tracking which sgRNAs are present in each transformed plant. Since library transformation is done in bulk, researchers typically sequence only plants showing the phenotypes of interest, leaving all other plants uncharacterized. This limited approach misses potentially valuable information about genotype-phenotype relationships and makes it difficult to validate results through independent lines carrying the same genetic modifications.

Ion transport-related genes in plants play essential roles in nutrient uptake, signaling, and stress responses. Their expression is often tightly regulated in a tissue-specific manner; for example, phosphate and nitrate transporters are highly active in root epidermal cells under nutrient deficiency stress.^36,37,38 Despite the molecular-level understanding of nutrient transport, there are significant gaps in the knowledge of how nutrients travel from the soil to various tissues and organs within the plant. For example, it is currently unclear how iron (Fe) is exported from the epidermis cells, bypasses the endodermis cell file with the Casparian strip and suberin barriers, exits the pericycle cells to translocate into the xylem, and how it enters the chloroplasts’ outer envelope.^39,40

In this study, we present the development of genetic CRISPR tools with a number of new capabilities. We selected an improved algorithm for designing CRISPR libraries, allowing higher CRISPR efficiency, manifested by fewer mismatches. We constructed a multiplexed library in which each vector contains two sgRNAs that target genes from the same gene family. The resulting multiplexed library significantly expands the gene coverage, accuracy, and efficiency of the screen (Figure 1). To validate the approach, we designed libraries to target genes associated with nutrient transporters in Arabidopsis, allowing trait-specific screens. We generated over 1,000 independent plants from the library and revealed multiple novel phenotypes. We developed a double barcode tagging strategy—a technology that we term CRISPR-GuideMap—to enable the identification of sgRNAs present in every plant in the library, allowing the comprehensive tracking of the genetic modifications present in the entire transformed population. The designed libraries can function at the cell-type-specific level, thus allowing spatial genetics screens, expanding our ability to identify signaling molecules produced in one cell type that act in a distal tissue (Figures 1 and S1). In summary, the tools we developed significantly expand the genetic toolbox for scientists and breeders and are expected to drive the next generation of plant genetics.

Figure 1.

Overview of the next-generation CRISPR genetic toolbox

Level 1: the algorithm includes improved scoring functions for the design of efficient sgRNAs. Level 2: design and construction of multiplexed CRISPR libraries. Level 3: trait-oriented CRISPR libraries—nutrient uptake sub-libraries as a proof of concept. Level 4: tissue-specific CRISPR vectors for spatial genetics screens. Level 5: CRISPR-GuideMap: plant barcoded screens of ∼1,000 plants, allowing large-scale reverse genetics multi-knockouts. Outcome: a next-generation tool for plant genetics.

Results

Trait-oriented CRISPR design: Seven nutrient uptake sub-libraries as a proof of concept

The multi-targeted CRISPR library genetic approach has the potential to transform random forward genetic screens into a large-scale, targeted genetic application, supporting basic science and breeding of crops for specific traits. As a proof of concept, we tackled nutrient uptake, as it is a rate-limiting step in plant biology and agriculture. Fertilizer usage is expensive, but in many cases, only 30% of the fertilizer applied is taken up by the plant, and the rest contaminates underground water and the environment.⁴¹ To identify genes that participate in these processes, we generated a CRISPR library that specifically targeted genes putatively associated with ion transport. To this end, we first generated a list of all transporter genes differentially expressed in response to seven different nutrient deficiencies: nitrate (N), potassium (K), phosphate (P), calcium (Ca), Fe, magnesium (Mg), and sulfur (S).⁴² Genes encoding 452 transporters were differentially expressed under at least one of the deficiencies, with 40.9% of them exclusively differentially expressed under nitrogen-deficient conditions (Figure S2).

Next, we constructed a large CRISPR library directed toward the 452 differentially expressed nutrient-associated genes,⁴² together with their closely related homologs, for a total of 707 targeted genes. Each construct incorporated a single sgRNA designed to target several genes from the same family, thus allowing their simultaneous editing in the same plant to overcome genetic redundancy. For each gene family, multiple sgRNAs were designed to target different combinations of family members. The design process accounted for the differential sequence similarity between family members, such that genes that are more closely related are targeted together by more sgRNAs. For each gene combination, we applied the CRISPys algorithm⁴³ to design sgRNAs that would optimally target the gene set. The design process incorporated two main improvements compared with the original Multi-Knock approach.³ First, the scoring function, which predicts the differential impact of the mismatch type and position across the targeted regions, was previously based on the outdated CFD (cutting frequency determination) score.⁴⁴ The libraries presented here are based on the MOFF (a model-based predictor of Cas9-mediated off-target effects) deep learning-based scoring function,⁴⁵ which was found superior among several more recent alternatives: CFD,⁴⁴ gold-off,⁴⁶ and uCRISPR⁴⁷ (Figure S3). For example, the MOFF-based sgRNA design resulted in fewer mismatches per target compared to the CFD score (Figure 2A). Second, we filtered the sgRNAs that target sparsely related members of a given gene family (detailed in STAR Methods). This increased the probability that the designed sgRNAs effectively target functionally redundant genes (Figure 2B).

Figure 2.

Trait-oriented CRISPR design results in seven nutrient uptake sub-libraries

(A) Relative frequencies of the number of mismatches per target for each scoring function: CFD,⁴⁴ MOFF,⁴⁵ gold-off,⁴⁶ and uCRISPR.⁴⁷

(B) Examples of the distance rule for determining sgRNA targeting sparsity. The leaf nodes represent genes from the same family, and genes and nodes marked with blue circles represent the most recent common ancestors of genes clustered and targeted by the same sgRNA. Green clusters represent included gene clusters (a minimal cluster distance of three); pink clusters were not included (a minimal distance of four).

(C) Number of sgRNAs in the library.

(D) Number of genes targeted by individual sgRNAs.

(E) Number of mismatches per sgRNA.

(F) Total number of sgRNAs in each nutrient sub-library.

(G) sgRNA distribution in each of the seven sub-libraries based on deep-sequencing results: nitrate (N), potassium (K), phosphate (P), calcium (Ca), iron (Fe), magnesium (Mg), and sulfur (S). Coverage indicates the percentage of sgRNAs detected relative to the total number of synthesized sgRNAs.

The resulting library consisted of a total of 2,859 individual sgRNAs, each expressing a unique sgRNA designed to target 2 to 8 genes, with an average of 2.32 genes (Figures 2C and 2D). In cases where the targeted gene lacked a direct homologous family member (20.7% of the sgRNAs, targeting 13 singleton genes), an sgRNA was designed to target only that gene (Figure S4A). Each gene was targeted by an average of 8.2 sgRNAs (Figure S4B). The number of mismatches between the sgRNAs and their targets ranged from 0 to 3, with an average of 0.71 mismatches (Figure 2E), with fewer mismatches closer to the PAM (protospacer adjacent motif) position (Figure S4C).

To increase the flexibility of the trait-specific forward genetics tool, the sgRNAs were classified into seven subgroups based on the nutrients associated with the differentially expressed genes. Each sgRNA was associated with one or more nutrient groups, meaning the same sgRNA could appear in more than one subgroup. This classification resulted in seven sub-libraries: N, 1,805 sgRNAs; K, 1,314 sgRNAs; P, 457 sgRNAs; Ca, 476 sgRNAs; Fe, 187 sgRNAs; Mg, 165 sgRNAs; and S, 137 sgRNAs (Figure 2F; Table S1). The complete library design is available in Data S1.

Subsequently, we synthesized and cloned all sgRNAs into a vector that contains the RPS5 meristematic enriched promoter (which generates stable and inherited mutations),⁴⁸ yielding seven sets of vectors pRPS5a:zCas9i:U6:Specific nutrient libraries: pRPS5:K, pRPS5:N, pRPS5:P, pRPS5:S, pRPS5:Mg, pRPS5:Ca, and pRPS5:Fe. To verify if the cloning step was successful, we deep-sequenced all libraries. The deep sequencing results showed narrow bell-shaped distribution of sgRNAs with very low skew values (across the libraries, skewness values ranged from −2.88 to 0.628) (Figure 2G). Across the seven libraries, the coverage (i.e., the number of sgRNAs present in the deep-sequencing [deep-seq] out of the number of sgRNAs bioinformatically designed) ranged from 99.48% to 100% (Figure 2G; Table S2). The libraries allow comprehensive targeting of nutrient transporter genes and can be used as a versatile resource to study nutrient transporter genes in various contexts.

Construction of cell-type-specific CRISPR libraries

The use of a constitutive, meristematic-enriched, or egg-cell promoter is a common and straightforward method for the expression of CRISPR components in plants. However, this approach can have undesired consequences, such as lethality, sterility, or general pleiotropic effects (a plethora of seemingly unrelated phenotypes in a single plant/organ). To address these shortcomings and allow the design of spatial genetic screens, we aimed to generate a generic cell-type-specific CRISPR system. This will enable, for example, the identification of signaling effectors produced in one cell type that act in a distal tissue or the creation of a system that addresses a specific function, such as nutrient uptake from the root epidermis. To this end, we chose eight known tissue-specific promoters: root-specific pARSK1,⁴⁹ guard-cell-specific pKST1,⁵⁰ endodermis- and bundle-sheath-specific pSCR,⁵¹ shoot-specific pSIG6,⁵² phloem-companion-cell-specific pSUC2,⁵³ spongy-mesophyll-specific pCORI3,⁵⁴ root-cortex- and epidermis-specific pPGP4,⁵⁵ and shoot-epidermis-specific pML1⁵⁶ (Figure 3A). Using these promoters should result in cell-type-specific DNA mutations rather than inherited mutations in the entire plant.

Figure 3.

Cell-type-specific multi-targeted CRISPR library screen identifies putative iron transporters

(A) Images demonstrating tissue-specific localization of genes expressed from promoters used in this study: pSCR, endodermis and bundle sheath; pCORI3, spongy mesophyll; pSIG6, shoot; pSUC2, phloem companion cells; pKST1, guard cells; pPGP4, root epidermis and cortex; pML1, shoot epidermis; and pARSK1, root. For pSIG6:NLS-YFP, scale bars, 200 μM; for pSUC2:YFP and pARSK1:NLS-YFP, scale bars, 100 μM; for pSCR:YFP, pCORI3:GUS:mCitrine, pPGP4:NLS-YFP, and pML1:H2B-GFP, scale bars, 50 μM; for pKST1:GFP, scale bars, 20 μM; and for pSIG6:GUS and pARSK1:GUS, scale bars, 2,000 μm.

(B) Schematic of the cell-type-specific CRISPR vector that was generated in this study. The vector includes tissue-specific promoter cloning sites, an intronized Cas9 coding sequence, and library cloning sites.

(C) Schematic of the pPGP4:Fe library vector.

(D) sgRNA distribution in the pPGP4:Fe library based on deep-sequencing results. Coverage indicates the percentage of sgRNAs detected relative to the total theoretical number of sgRNAs theoretically synthesized.

(E) Representative images of 21-day-old pPGP4:sg-IRT1,2 with respective controls. Scale bars, 1 cm. The bottom illustrations represent expression patterns (red color) extracted from single-cell sequencing databases. The red color in pRPS5 is presented throughout as it generates stable inherited mutations throughout the plant.

(F) Quantification of plant color shade of 21-day-old pPGP4:sg-IRT1,2 and control plants. Statistical analysis was determined for RGB (90, 98, and 58) using two-way ANOVA followed by Tukey’s HSD (honestly significant difference) test.

(G) Representative images of 29-day-old pPGP4:sg-SULTR1.1,1.2 with respective controls. Scale bars, 1 cm. The bottom illustrations represent the expression patterns (red color) extracted from single-cell sequencing databases.

(H) Rosette fresh weight (Fw) of 32-day-old pPGP4:sg-SULTR1;1,1;2 and respective controls. n ≥ 7. Statistical analysis was performed using one-way ANOVA followed by Tukey’s HSD test.

(I) Total sulfate content (mg/g fresh weight) was measured in 43-day-old pPGP4:sg-SULTR1.1,1.2 and respective controls. Significance was evaluated using two-way ANOVA followed by Tukey’s HSD test. n = 4.

(J) Chromatograms of sequences of the targeted genes in shoots and roots. Regions within dashed orange lines indicate sequences with cell-type-specific mutations. sgRNA binding regions are indicated in gray above the chromatograms.

To generate tissue-specific vectors for the CRISPR-Cas9 libraries, we constructed new cloning vectors (Figure 3B). First, we introduced Golden Gate cloning sites to facilitate easy insertion of promoters of interest. Second, downstream to those sites, we incorporated the sequence encoding an intronized Cas9. This construct contains 13 introns integrated into the maize codon-optimized Cas9. These introns were shown to significantly enhance Cas9 genome editing efficiency in Arabidopsis.⁵⁷ Third, downstream to the intronized Cas9 sequence, we included BsaI sites for cloning the CRISPR library (or any sgRNA of interest).⁵⁸ The eight plasmids we generated may provide the scientific community with tools to perform targeted mutagenesis in specific cell types, whether for CRISPR libraries or individual sgRNAs of interest. As a proof of concept, we cloned the iron sub-library sgRNAs under the tissue-specific promoter PGP4⁵⁵ (Figure 3C), which should result in mutagenesis in the root epidermis and cortex. This approach enabled us to target iron-associated transporters responsible for iron uptake from the soil into the root, together with their homologous family members. Deep-seq showed a narrow distribution of the sgRNAs and nearly full coverage (Figure 3D; Table S2).

We transformed the pPGP4:Fe library into the background of Col-0 plants as one bulk transformation, and 500 T1-resistant plants were selected using Basta. Several lines exhibited nutrient-mediated physiological phenotypes of defective shoot growth. For example, we isolated an independent line with pale and small shoots. DNA was extracted, and the sgRNA cassette was amplified. Sequencing revealed that this sgRNA putatively targets IRT1 and IRT2 (Figures 3E and 3F). The single irt1 mutant has a strong iron-deficient phenotype,⁵⁰ whereas the irt2 single mutant did not have a visible loss of function phenotype.⁵⁹ The double mutant irt1/irt2 has not been reported so far. Therefore, we generated pRPS5:sg-IRT1,2, a stable CRISPR line mutated in these two genes. Quantification of the color-shade phenotypes showed that pPGP4:sg-IRT1,2 resembles the single irt1 mutant, suggesting that the activities of IRT1 and IRT2 are not redundant. The pPGP4:sg-IRT1,2 line resembled pRPS5:sg-IRT1,2 and the single irt1 mutant, suggesting that IRT1 acts predominantly from the root epidermis. This result is in line with single-cell transcriptomics data, which showed that IRT1 and IRT2 are exclusively expressed in the root epidermis (https://rootcellatlas.org/) (Figure 3E).

In addition, the screen recovered a line that exhibits a shoot growth inhibition phenotype (Figures 3G and S5). Sequencing the sgRNA insertion showed that it is designed to target the SULTR genes. SULTR1;1 and SULTR1;2 encode sulfate transporters responsible for sulfate uptake from the soil.^60,61 Neither single mutant has a phenotype, but the pPGP4:sg-SULTR1;1,1;2 and double knockout line pRPS5:sg-SULTR1;1,1;2 have significant shoot growth inhibition phenotypes,⁶⁰ suggesting that these two genes encode proteins with redundant activity. In addition, we quantified rosette fresh weight and sulfate levels in the leaves, and found that while the single mutant sultr1;1 showed a mild reduction compared to wild type, targeting both SULTR1;1 and SULTR1;2, either using the cell-type-specific or the inherited pRPS5 promoter, resulted in significantly enhanced reduction in fresh weight and sulfate levels compared to both the wild type and the respective single mutants (Figures 3G and 3H). These results highlight the effectiveness of tissue-specific knockout approaches in altering nutrient accumulation.

To verify the tissue-specific mutagenesis, we sequenced DNA that was separately extracted from multiple roots and shoots of pPGP4:sg-IRT1,2 and pPGP4:sg-SULTR1;1,1;2. The results showed mixed chromatograms in the root, indicating cell-type-specific gene editing in the root but not the shoot (Figure 3J). The results demonstrate that our strategy enables ionome-scale gene family knockout in a tissue-specific manner, demonstrating the functionality of the tissue-specific CRISPR library screening system. The cell-type-specific plasmids generated in this study will be available in Addgene for community use.

A multiplexed, multi-targeted CRISPR library efficiently targets nutrient transporter genes

The Multi-Knock library,³ as well as libraries developed by others,^6,30 contain a single sgRNA per vector. To enhance targeting capabilities, we developed a library in which each vector encodes two sgRNAs, thereby significantly improving the tool capabilities, particularly in terms of gene coverage, diversity of gene combinations, and targeting efficiency (i.e., allowing the targeting of the same gene combination with fewer mismatches per gene). To this end, we developed an algorithm that optimizes the design of multiple sgRNAs (in this case, two) to edit multiple members of a gene family simultaneously while accounting for the sequence similarity among family members. Briefly, for each gene family (in this case, as a proof of concept, nutrient transporter genes), its phylogeny was reconstructed. Internal nodes in this tree represent subgroups of genes that are more closely related to each other than to the rest of the family (Figure 2D). For each such subgroup, we designed three multiplexes (i.e., three pairs of sgRNAs) that efficiently target different combinations of genes within that subgroup.

Using this approach, we generated a library encompassing 1,599 multiplex vectors that encoded 1,826 unique sgRNAs targeting a total of 694 genes belonging to the nutrient-associated genes and their family members (Figure 4A). Each multiplex vector holds a pair of sgRNAs that target genes from the same family. Compared to the sgRNA library with a single sgRNA per vector, the multiplex design achieved higher gene coverage, more genes per vector were targeted (an average of 3.03 genes per vector versus 2.32; Figure 4B), and improved targeting efficiency by reducing the average number of mismatches from 0.706 to 0.347 per target (Figure 4C). In addition, in this library, each gene was targeted by 6.9 multiplex vectors on average (Figure S6A), with fewer mismatches positioned proximal to the PAM (Figure S6B). The large library generates a robust coverage network relationship of the multiplexes and targeting genes. For example, a multiplex design for a single gene family composed of five genes is targeted by 12 multiplexes, collectively generating 41 unique gene-multiplex combinations (Figures 4D and 4E), demonstrating how the multiplexed strategy in silico addresses functional redundancy. The complete library design is available in Data S1.

Figure 4.

Construction and validation of nutrient, multi-targeted, multiplexed CRISPR library

(A) Number of multiplexes in the multiplexed library. The multiplex holds tandem sgRNAs targeting the same gene family.

(B) Number of genes targeted by sgRNA pairs within each multiplex in the multiplexed library.

(C) Number of mismatches per multiplex in the multiplexed library.

(D) Representative family network design in the nutrient transporter-targeted, multiplexed library. Genes can be targeted by multiple multiplexes, and each multiplex can target multiple genes at once. The blue triangle represents a single multiplex; the orange rectangle represents a single gene.

(E) Overview of all gene target networks in the multiplexed CRISPR library. The blue triangle represents a single multiplex; the orange rectangle represents a single gene.

(F) Coverage and distribution of multiplex vectors (pairs of sgRNAs) within the E. coli pRPS5:Multiplex library. Coverage indicates the percentage of sgRNAs detected relative to the total number of synthesized sgRNAs. Skew indicates an equal distribution demonstration.

(G) Comparison of deep sequencing analyses of the library after cloning into Agrobacterium versus E. coli.

(H and I) Gene editing efficiencies (cleavage efficiencies) of the tandem sgRNAs (multiplexed library) compared to the single sgRNA library previously published.³ Editing efficiencies representing the entire library (H) and editing efficiency divided by the mismatch average numbers (I). Gene editing efficiencies (cleavage efficiencies) were calculated as the average of the editing percentages of multiple genes targeted by the sgRNA. n ≥ 11 (H) and n ≥ 3. (I). Statistical significance was evaluated by Student’s t test.

(J) Comparison of the mismatch average number between the single sgRNA Multi-Knock library³ and the new multiplexed library. The outer pie chart represents the single sgRNA library, while the inner pie chart, highlighted in green, represents the multiplexed library. For the multiplexed library, 1.2% of sgRNAs have up to 1.5 mismatches, 0.3% have up to 2 mismatches, and 0% have more than 2 mismatches. The number of mismatches per sgRNA is calculated as the average number of mismatches per gene targeted by the sgRNA.

(K) Gene editing efficiencies (cleavage efficiencies) per target gene, between genes targeted by only a single sgRNA within each multiplex, and genes targeted by both of the sgRNAs within each multiplex. n ≥ 11. Statistical significance was evaluated by Student’s t test.

(L) Comparison of sgRNA scores between cut and uncut genes. n ≥ 11. Statistical significance was evaluated by Student’s t test.

Next, we synthesized and cloned the multiplex library, driven by the RPS5 meristematic-enriched promoter, which generates stable heritable mutations.³⁹ To verify the cloning step, we deep-sequenced the library. We observed a narrow bell-shaped distribution of the multiplexes (sgRNA pairs) in the library, with a very low skew and high coverage (Figure 4F). To verify that the transition to Agrobacterium does not affect the library coverage and distribution, we sequenced the library in Agrobacterium and compared it to E. coli. We obtained highly correlated, similar results (R² = 0.93) (Figure 4G), suggesting a robust cloning system and verifying that the tool is ready for plant transformation.

To assess the functional impact of the library on plant phenotypes, we transformed the nutrient transporter-targeted, multiplexed library into Col-0 as a bulk, generated approximately 1,000 plants, and screened the T2 for shoot growth and color phenotypic changes compared to wild-type plants. Several lines exhibited significant phenotypic alterations, indicative of successful knockout of the targeted genes. To validate our ability to sequence the pairs of sgRNAs within the transformed plants, we analyzed 74 individual plants. We successfully determined the multiplex sequence in 56 plants (75.7%). In 3 plants (4.1%), the multiplex contained two distinct sgRNAs designed in the library, meaning the first sgRNA was incorrectly paired with a second sgRNA, likely due to an issue in synthesis or cloning. In the remaining 15 plants (20.2%), we were either unable to sequence the full multiplex (12.2%) or detected more than one sequence (8.1%).

We then compared the multiplexed library to the previously published single sgRNA Multi-Knock library³ to benchmark the improvements in editing efficiency and flexibility. In both cases, Cas9 expression was driven by the RPS5a promoter, and the same intronized Cas9 was used.^3,57 We genotyped representative plants from the single sgRNA Multi-Knock library and from the multiplex library and then calculated the gene editing efficiency from each library. Editing efficiency was calculated as the editing percentage of the single or tandem sgRNAs. For example, if the single or tandem sgRNAs were able to create mutations in 3 out of the 4 targeted genes, the gene editing efficiency for this sgRNA was 75%. Genotyping the targeted genes in representative plants underscored the enhanced efficiency of using a multiplexed pair of sgRNAs compared to a single sgRNA per vector. Examining the editing efficiency of sgRNAs from the single sgRNA library showed an editing efficiency of 48.4% compared to the significantly higher editing efficiency of 84.2% of the new multiplexed library (Figure 4H). Furthermore, looking into the gene editing efficiency by the number of mismatches in the single sgRNA library revealed that the editing efficiency was highest when there were no mismatches, with a significant decrease observed as the number of mismatches increased. However, for the multiplexed sgRNAs, the presence of mismatches did not impact the overall editing efficiency (Figure 4I). Notably, the multiplexed library was designed with a lower average number of mismatches compared to the single sgRNA Multi-Knock library. This is reflected in the mismatch distribution: 98.5% of sgRNAs in the multiplexed library had up to 1 mismatch, compared to 56.7% in the single sgRNA library, indicating high target specificity of the multiplexed library (Figure 4J). Moreover, the maximal number of mismatches in the multiplexed library was limited to 2 (0.3% of the entire library had 2 mismatches), with an overall average of 0.35 mismatches (Figure 4C). In contrast, the maximal number of mismatches in the single sgRNA Multi-Knock library was 5 (33.7% of the sgRNAs had 2 or more mismatches) with an overall average number of 1.26 mismatches (Figures 4J and S7). Furthermore, the gene editing efficiency was 100% when both sgRNAs within the multiplex targeted the same genes, unlike targeting a gene with only 1 sgRNA within the pair (68.4%) (Figure 4K). We also found that the sgRNA score showed a higher trend in successfully edited genes (0.87) compared to genes that were not edited (0.70) (Figure 4L). The results indicate that the multiplexed system consistently outperformed single sgRNAs and is a flexible system that can be used for large-scale multi-targeted forward and reverse genetics.

Plant-barcoded, multiplexed, large-scale genetic screens

The CRISPR library approach we developed for large-scale bulk transformation allows robust and comprehensive screening of a targeted population of gene families. However, bulk library transformation is a major drawback, as one does not know the exact sgRNA encoded by each plant in the screen. To tackle this shortcoming, we developed a robust barcoding system that enables efficient identification of the sgRNAs within each plant in a large-scale manner, termed CRISPR-GuideMap (Figure S8). The approach is based on barcoding the tandem sgRNAs inserted in each plant and deep sequencing to generate a catalog and map the multiplexed content of the entire population. To this end, we collected DNA from 1,014 plants at T1 generation (no transfer DNA line segregation). Each line was amplified using a unique set of barcoded primers and deep-sequenced as a pool to reveal the expressed sgRNAs. The multiplexed sequence (pair of sgRNAs) was determined in 73% of the plants, 17% contained two multiplexes (meaning two transfer DNA insertions), and in 10%, noise in the deep-seq data precluded analysis (Figure 5A).

Figure 5.

CRIPSR-GuideMap plant barcoding allows mapping of the plant population after library transformation

(A) The number of multiplexes (sgRNA pairs) per plant across transformed barcoded plants (n = 1,014). The multiplex holds 2 tandem sgRNAs targeting the same gene family.

(B) Distribution of unique multiplexes in plants and their frequency of occurrence within barcoded plants (n = 742).

(C) Distribution of unique targeted genes in plants and their frequency of occurrence within 1,014 barcoded plants (n = 742).

(D) The number of genes targeted per plant in 742 barcoded plants that contain one multiplex.

(E) Number of plants where the barcoded sgRNAs are associated with specific nutrients. The sgRNAs multiplexing is designed to target genes differentially expressed in the deficiency of the indicated nutrient.

(F) Number of families and genes in the designed library and in the 1,014 barcoded plant population.

(G) Network relationship between targeted genes and barcoded plant population lines for the PUP gene family. The phylogenetic tree on the left represents a subset of the PUP family. On the right, the gray numbers indicate individual plant lines identified through the barcoding data. Multiple plants listed at the same connection point suggest that the same multiplex is present in independent plant transformation lines.

(H) Shown are shoot phenotypes of 3 independent 30-day-old pRPS5:Multiplex-ACA4,11 lines recovered using the CRISPR-GuideMap tool, scale bars, 1 cm.

(I) Sanger sequencing results are displayed using a color code: blue indicates a mutated gene and gray indicates a wild-type form (no mutation). Associated sequence chromatograms are presented in Figure S10.

To validate the CRISPR-GuideMap method, we compared the analysis to Sanger sequencing of selected lines. We found that for 18 of the 20 (90%) selected plants, the barcoding-based analysis matched the Sanger sequencing, indicating that the barcoding tool is reliable. In addition, we analyzed the distribution of each tandem sgRNA within the multiplex in the plant population and found that, on average, each multiplex vector was present in 2.09 plants (Figure 5B) and each gene was, on average, targeted in 4.16 plants (Figure 5C). We discovered that, on average, 2.96 genes were targeted per barcoded plant (Figure 5D), and we could map the exact lines expressing various sgRNA multiplexes, which are associated with different nutrients (Figure 5E). We detected sgRNAs targeting 86 of the 114 gene families targeted by the designed library, and these sgRNAs targeted 528 of the target 694 genes (Figure 5F). The unique resource allows a reverse genetics data source to be used for the first time, where one can order seeds of a family gene knockout. For example, if one is interested in the PUP cytokinin transporter family,^3,62,63 we identified a network of plants expressing sgRNAs that target this family using a large combinatorial matrix. This matrix consists of 10 diverse multiplexes across 22 independent lines, collectively targeting 9 PUP genes, allowing multiple genes from the same family to be targeted simultaneously (Figure 5G). The 1,014 lines genotyped here may serve as a valuable resource for the community, functioning as a seed bank for reverse genetics research (Data S2). This tool also enables the identification of multiple alleles for the targeted genes. By leveraging the seed stock list, we can access additional independent lines targeting the same genes, providing more substantial support that observed phenotypes are indeed linked to disruption of the candidate genes. For instance, we recovered the vacuolar calcium transporters aca4 aca11 double mutant.⁶⁴ Using CRISPR-GuideMap, we identified three distinct alleles corresponding to this line (Figures 5H, 5I, and S9). Going forward, this resource will facilitate faster and more reliable screening, ensuring that phenotypes arise from on-target gene modifications.

Hidden phenotypes recovered by nutrient transporter library screen

The screen successfully recovered multiple previously reported phenotypes caused by a mutation in the targeted transporter genes. For example, we isolated a line with a significantly smaller shoot compared to the wild type (Figure 6A). This plant expressed a pair of sgRNAs that target HMA2, HMA3, and HMA4 genes that encode transporters known to be involved in zinc and cadmium homeostasis.^60,61,62 The previously reported double mutant hma2 hma4 has a small shoot phenotype,⁶⁵ similar to the phenotype of our triple-mutant line. Sanger sequencing of this line revealed that all 3 genes were knocked out (Figures 6A and S10). Another plant identified in the screen had pale, small shoots (Figure 6B). Evaluation of shoot morphology and plant color shade showed that this line has a visibly different color shade and a significantly smaller shoot area compared to wild-type plants. DNA extraction, amplification, and sequencing of the multiplex cassette revealed that the pair of sgRNAs targeted AAC1, AAC2, AAC3, and ER-ANT1. ER-ANT1 encodes an adenine transporter, and its loss-of-function mutant has a similar phenotype.⁶⁶ Sanger sequencing revealed that AAC2 and ER-ANT1 were mutated (Figures 6B and S10). In another example, a dark, small plant contained a multiplex targeting PIN3, PIN4, and PIN7 (Figure 6C), which encode auxin transporters.^67,68,69 Shoot morphology and color evaluations revealed distinct changes in shade, reduced shoot size, and altered leaf shape compared to the wild-type plant, consistent with the previously described phenotypes.⁷⁰ Sanger sequencing showed that all 3 genes were knocked out (Figures 6C and S10). To test whether the phenotypes arose from the simultaneous disruption of multiple gene family members, we compared the multiplexed CRISPR lines (Figures 6A–6C) with their corresponding transfer DNA single mutants. This comparison demonstrated the robustness of our multiplexed tool in efficiently targeting gene families and overcoming functional redundancy. In most cases, the phenotypes observed in our CRISPR lines were more pronounced than in the respective single mutants (Figures 6A–6C and S11). For the pRPS5:Multiplex-HMA2,3,4 line (Figure 6A), which did not produce seeds, the comparison was made to the previously characterized hma2 hma4 double mutant.⁶⁵

Figure 6.

The multiplex nutrient library recovers key proof-of-concept genes and reveals hidden phenotypes

Shoot phenotypes (scale bars, 1 cm), leaf area measurements, plant color shade (ordered by RGB score), Sanger sequencing mutation summary, and comparison to transfer DNA single mutant lines (scale bars, 1 cm). (A) 32-day-old Col-0 and pRPS5:Multiplex-HMA2,3,4 plants, n ≥ 7; (B) 28-day-old Col-0 and pRPS5:Multiplex-AAC1,2,3,ER-ANT1 (Multiplex-AACs,ER-ANT1) plants, n ≥ 8; and (C) 20-day-old Col-0 and pRPS5:Multiplex-PIN3,4,7 plants, n ≥ 8. Sanger sequencing results are displayed using a color code: blue indicates a mutated gene; gray indicates a wild-type form (no mutation); and white represents genes not targeted by the sgRNA. Associated sequence chromatograms are presented in Figure S6. For leaf area measurements, statistical significance was evaluated by Student’s t test; ^∗p value < 0.05, ^∗∗p value < 0.01, and ^∗∗∗p value < 0.005.

Several phenotypes from the screen mapped to genes without previously reported knockout lines. For example, line #762 develops smaller shoots (Figures 7A and 7B). Sequencing of the multiplex cassette showed sgRNAs targeting four SWEET genes, SWEET11–14. All four are clade III SWEET sucrose uniporters that mediate bidirectional facilitated diffusion according to the transmembrane sucrose gradient. SWEET11 and SWEET12 are plasma-membrane transporters in phloem parenchyma that export sucrose into the apoplast for apoplastic loading; double mutants accumulate sugars in leaves and exhibit impaired long-distance translocation.^71,72 SWEET13 and SWEET14 are required for pollen nutrition and male fertility through sucrose transport. Although they have been reported to transport gibberellins, genetic complementation indicates that sucrose transport, rather than GA transport, restores fertility in the sweet13/14 double mutant.^73,74 Target-gene sequencing confirmed that #762 carries loss-of-function mutations in all four SWEET genes. CRISPR-GuideMap indicated that line #807 harbors a multiplex cassette targeting the same four genes, with an sgRNA overlapping the #762 guide for SWEET13/14 and a distinct guide for SWEET11/12. Sequencing showed mutations in SWEET12, SWEET13, and SWEET14, while SWEET11 remained wild type. Thus, #762 is a quadruple mutant and #807 a triple mutant (Figure S10). In both cases, the leaf area of the triple and quadruple SWEET mutants was smaller than that of the single and double mutants, indicating that our multiplex approach successfully disrupted multiple members of the same gene family within one construct (Figures 7A and 7B).

Figure 7.

Redundant and specialized activities of clade III SWEET uniporters

(A) Shoot phenotypes of 26-day-old Arabidopsis plants of the indicated genotypes, including multiplex lines targeting SWEET11, 12, 13, and 14 (scale bars, 1 cm).

(B) Leaf area for the plants in (A). Bars show mean ± SD; dots are individual plants (n ≥ 3). Statistical analysis was performed using one-way ANOVA followed by Tukey’s HSD test.

(C) Sanger sequencing results are displayed using a color code: blue indicates a mutated gene and gray indicates a wild-type form (no mutation). Associated sequence chromatograms are presented in Figure S10.

(D and E) Xenopus oocyte uptake assays. (D) Cytokinins: trans-zeatin (tZ), trans-zeatin riboside (tZR), and N6-isopentenyladenosine (iPR). (E) Gibberellin A3 (GA₃) and abscisic acid (ABA). Rates are expressed as μM oocyte⁻¹ h⁻¹; points are biological replicates. Statistics: one-way ANOVA with Tukey’s HSD. n ≥ 5. Different letters indicate significant differences at p < 0.05.

Because SWEET transporters have been shown previously to transport various phytohormones, gibberellic acid 3 (GA₃),^73,74 and cytokinins (CKs),⁷⁵ we tested whether the redundant phenotype may reflect substrate specialization within clade III. None of the four proteins transported GA₃ or abscisic acid (ABA). The inability to transport GA₃ stands in contrast to prior reports where uptake was shown using yeast-based transport assays.^73,74 This indicates that transport activity may depend on the expression and detection systems used. Notably, we observed a significant uptake of the cytokinin derivative trans-zeatin (tZ) by SWEET13 and SWEET14 but not by SWEET11 or SWEET12 (Figures 7D and 7E). The multi-target CRISPR approach revealed a phenotype that remained hidden due to the redundancy of at least three of the four clade III SWEET members (SWEET12, SWEET13, and SWEET14). However, biochemical characterization showing specialization of only SWEET13 and SWEET14 toward CKs speaks against this phenotype being solely attributed to impaired CK transport. Thus, identifying a phenotype in different orders of mutants and combining the observations with biochemical activity represents an essential first step toward dissecting the molecular mechanism underlying redundant phenotypes in genetically complex, multi-mechanism systems.

To verify that the screen-derived phenotypes were caused by on-target gene disruption, we sequenced the two highest-scoring predicted off-target sites for each of the four phenotypes described in this study. We found no evidence of Cas9-mediated cleavage at these loci, supporting the conclusion that the observed phenotypes result from specific editing of the intended target genes (Figure S12). Together, the results highlight the diverse phenotypic outcomes and the ability of the plant-barcoded, multiplexed, large-scale genetic screens to overcome functional redundancy and reveal hidden phenotypic variation.

Discussion

Advancements in CRISPR technology have transformed plant functional genomics, but most large-scale knockout libraries still suffer from limited spatial control, single-sgRNA designs, and poor handling of genetic redundancy.^3,47 Here, we address these constraints by combining improved sgRNA design, multiplexed constructs, and cell-type-specific expression into a single, scalable platform for forward genetic screens (Figure 1).

To overcome genetic redundancy and increase coverage, we implemented a tandem two-sgRNA multiplexing strategy. Previous plant studies used multiple sgRNAs per vector only in relatively small-scale applications with labor-intensive cloning,^7,35 whereas most CRISPR libraries in Arabidopsis and other crops relied on single sgRNAs.^30,33,34 By contrast, our transportome-scale library multiplexes two sgRNAs per construct and is explicitly designed to target several closely related genes in the same family in a single plant. This design increases gene coverage, reduces average mismatches per sgRNA, and improves editing efficiency at the library level compared with the previous Multi-Knock tool,³ enabling a shift from testing a small number of candidates to a broadly multi-targeted, forward-genetic framework (Figures 4H–4L; Table S8).

The improved design pipeline also allows the construction of trait-focused libraries. As a proof of concept, we created nutrient transporter-oriented libraries that target 707 Arabidopsis genes involved in nutrient uptake and transport, including seven element-specific sub-libraries. Such trait-specific libraries can be readily extended to genes enriched in particular tissues or expression states, for example, root-enriched transporters, genes associated with stomatal control, or loci strongly induced by biotic and abiotic stresses.^76,77 In this context, the shoot phenotypes recovered in our screen are consistent with known loss-of-function mutants affecting Fe transport (IRT1^78,79), sulfate uptake (SULTR1;1, SULTR1;2^60,61), metal homeostasis (HMA2/HMA4⁶⁵), adenine transport (ER-ANT1⁶⁶), and auxin transport and nitrate-auxin crosstalk (PIN3/4/7^{67,68,69,70,80,81,82,83}) (Figure 6). This concordance supports the conclusion that the phenotypes we observe arise from perturbations in nutrient and hormone transport rather than nonspecific pleiotropic effects.

A central advance of this work is the implementation of cell-type-specific CRISPR libraries. Tissue-restricted Cas9 systems have been described,⁸⁴ but to our knowledge, this is the first application of cell-type-specific libraries at the transportome scale. Expressing multiplexed sgRNAs from promoters active in defined tissues (e.g., the root epidermis) allows spatially refined interrogation of gene function and facilitates the study of processes that are difficult to dissect with whole-plant knockouts, such as the multi-step movement of Fe from soil to shoots and seeds.^39,40 At the same time, restricting editing to specific tissues helps to mitigate lethality and reduce systemic pleiotropy, enabling the recovery of viable plants for genes that would otherwise be essential (Figure 3).

We also developed CRISPR-GuideMap, a double-barcoding strategy that enables efficient identification of sgRNA multiplexes in individual plants. This approach provides a plant-level map of sgRNA content, transforming the library into a reverse-genetics resource in which 742 lines with known multiplexes can be used to interrogate specific gene families. Seed stocks can facilitate both forward screens and targeted follow-up analyses (Figure 5).

Finally, the multiplex architecture not only addresses redundancy within gene families but also lays the groundwork for pathway-level targeting, in which multiple genes in the same biological pathway can be edited simultaneously. Coupling such pathway-oriented designs with high-content readouts, including reporter-based or single-cell-resolved assays,^85,86 should further increase the power and resolution of plant CRISPR screens. Together, these developments establish a broadly applicable toolbox for cell-type-specific, multi-targeted genome editing and forward genetics in plants.

Limitations of the study

Although the multiplex, cell-type-specific CRISPR libraries presented here provide a powerful framework for transportome-scale genetic screens, several limitations remain. Some degree of chimerism and incomplete editing is unavoidable in T1 plants, and residual wild-type alleles may persist at certain loci. Despite stringent sgRNA design and barcoding-based genotyping, we cannot fully exclude contributions from off-target events or transfer DNA insertions to individual phenotypes; therefore, one must verify the phenotypes across independent lines targeting the same set of genes. Finally, this platform is primarily a high-throughput discovery tool, and detailed mechanistic dissection of individual genes will still require conventional genetics, targeted editing, and follow-up experiments.

Resource availability

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Eilon Shani (eilonsh@tauex.tau.ac.il).

Materials availability

All materials are available upon request. The cell-type-specific plasmids will be available on Addgene for community use.

Data and code availability

• •Data: amplicon deep sequencing data have been deposited in the NCBI Sequence Read Archive (SRA) database under accession numbers PRJNA1397588 and PRJNA1397745. Additional datasets generated in this study are provided as Data S1 and Data S2. See the key resources table for details.
• •Code: this paper does not report original code.
• •Other items: All other reagents and resources generated or used in this study are listed in the key resources table or are available from the lead contact.

Acknowledgments

We thank Sylvestre Marillonnet for sharing pAGM47523, pICH47742, pAGT5472, pAGM55145, pICH41780, pAGM35171, pICH49344, and pAGM4723 plasmids that allowed us to construct the cell-type-specific CRISPR vectors; Ari Pekka for sharing the p1R4-pSUC2:XVE construct; Lucia Strader for sharing the pSCR GW CD3-1945 plasmid; Gilor Kelly and Nir Sade for sharing pKST1:GFP seeds; Sigal Savaldi-Goldstein for sharing BJ36/ML1-BAS1-YFP seeds; and Adrienne Anfang for sharing pML1:H2B-GFP seeds. Funding: this work was supported by the Israel Science Foundation (1462/24, 2712/24, and 1346/25 to E.S.), the Zimin Institute (to E.S. and I.M.), and the European Research Council (101113412-Multi-Crop and 101118769-HYDROSENSING to E.S.).

Author contributions

M.A., H.H.N.-E., I.M., and E.S. designed the research; M.A., R.H.Y., O.C., S.B.Y., U.L., Y.H., Z.M.B., and C.C. performed the research; A.B. and D.X. contributed analytic tools, seeds, and constructs; M.A., R.H.Y., O.C., S.B.Y., and Z.M.B. analyzed the data; and M.A., I.M., and E.S. wrote the manuscript.

Declaration of interests

US provisional patent applications: Multi-Knock (no. 63/329,506), cell-type-specific multi-targeted CRISPR libraries (no. 63/589,255), and CRISPR-GuideMap (no. 63/674,807) systems described in this study have been filed. E.S. and I.M. are co-founders of NetaGenomiX, a biotech company that is not associated with this study.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Bacterial and virus strains
Escherichia coli DH5α	Thermo fisher	EC0112
Agrobacterium tumefaciens GV3101	Gold Bio	CC-207-5x50
Chemicals, peptides, and recombinant proteins
Propidium iodide	Sigma-Aldrich	P4864-10ML
Calcofluor white	Sigma-Aldrich	18909-100ML-F
Murashige and Skoog medium	Getter	YM-M0222-0050
Plant agar	Getter	YM-P1001-1000
HEPES-based kulori buffer	Sigma-Aldrich	H4034-25G
Critical commercial assays
Plasmid Miniprep Kit	Sigma-Aldrich	ZYMD4068
Plasmid Maxiprep Kit	Qiagen	20–12162
mMessage mMachine T7 Kit	Invitrogen	N/A
Drummond NANOJECT II	Drummond Scientific Company	N/A
Deposited data
PRJNA1397588	https://www.ncbi.nlm.nih.gov/sra/PRJNA1397588	Arabidopsis Sequencing of sgRNA multiplex insertion. SubAccessions: SRX31678850, SRX31678849, SRX31678848, SRX31678847, SRX31678846
PRJNA1397745	https://www.ncbi.nlm.nih.gov/sra/PRJNA1397745	Bacteria Agrobacterium - Sequencing transformant colonies for multiplex insertion. SubAccession: SRX31678951 E.coli - Sequencing transformant colonies for multiplex insertion. SubAccession: SRX31678950
Experimental models: Organisms/strains
Arabidopsis transfer DNA lines (Figure 6): hma2, hma3, hma4, er-ant1, aac2, pin3, pin4, pin7	Salk Institute	SALK_073511C, SALK_034393, SALK_050924, SALK_043626, SALK_207505C, SALK_113246C, CS9368, SALK_048791C
Arabidopsis transfer DNA lines (Figure 7): sweer11, sweet12, sweet13, sweet14, sweet11/12, sweet 13/14	Salk Institute	SALK_073269C, SALK_031696C, SALK_087791C, SALK_010224C, CS68845, CS2110217
Arabidopsis transfer DNA lines: sultr1;1, sultr1;2	Rouached et al.61www.plantphysiol.org/cgi https://doi.org/10.1104/pp.108.118612	N/A
Arabidopsis transfer DNA lines: irt1, irt2	Vert et al.59https://doi.org/10.1007/s00425-009-0904-8	N/A
pCORI3:GUS:mCitrine	Procko et al.54https://doi.org/10.1093/plcell/koac167	N/A
pML1:H2B-GFP	Savaldi-Goldstein et al.87https://doi.org/10.1038/nature05618	N/A
pKST1:GFP	Kelly et al.50 https://doi.org/10.1093/jxb/erx159	N/A
pSIG6:GUS, pPGP4:GUS	Anfang et al.88https://doi.org/10.1093/plphys/kiaf682	N/A
pSCR:YFP, pSUC2:YFP	Anfang et al.88https://doi.org/10.1093/plphys/kiaf682	N/A
Oligonucleotides
Multiplex and single CRISPR libraries sgRNAs	This paper	Data S1
Recombinant DNA
pRPS5a:zCas9i CRISPR vectors	https://www.addgene.org/153210/	pAGM55621
Cell-type-specific Cas9 vectors	Grützner et al.57; https://doi.org/10.1016/j.xplc.2020.100135	pAGM47523 (Level 0 zCas9i vector), pICH47742 (Empty Backbone), pAGT5472 (L1, LacZ), pAGM55145 (L1, ole citrine), pICH41780 (L1 vector), pAGM35171 (Level 1 position 1 part containing the Nos promoter, the BAR gene, and the Ocs terminator), pICH49344 (for L1 construction, nosT), and pAGM4723 (L2 acceptor vector)
Software and algorithms
ZEN software (Zeiss)	https://www.zeiss.com/microscopy/en/products/software/zeiss-zen.html	Zen Blue
ImageJ	https://imagej.net/ij/	N/A
SnapGene (for alignment and sequence cloning)	https://www.snapgene.com/	SnapGene 8.2
Python	Used for data analysis, no custom code generated	https://www.python.org/
CRISPys	Hyams et al.43; https://doi.org/10.1016/j.jmb.2018.03.019	N/A
GraphPad-PRISM	https://www.graphpad.com/features	N/A
Other
CRISPR-Guide-map for all barcoded plants	This paper	Data S1

Experimental model and study participant details

All Arabidopsis thaliana lines used in this work are in Colombia background (Col-0 ecotype, Salk Institute). Sterilized seeds were plated on Murashige & Skoog (MS) x 0.5 (Duchefa Biochemic) medium containing 1% sucrose and 0.8% plant agar (Duchefa Biochemic) with pH adjusted to 5.6–5.8 with 1 M KOH. Plates with seeds were stratified for 48 h at 4°C, then transferred to growth chambers (Percival CU41L5) at 21°C, 100–120 μEm⁻²S⁻¹ light intensity under long-day conditions (16 h light/8 h dark). For seed production, plant transformation, crossing, and soil pot assays, seeds were sown on wet soil. Plants were grown in growth rooms under long-day conditions at 21°C.

Method details

Seed sterilization

Seeds were sterilized by vapor-phase sterilization (chlorine fumes) for 2.5 h in Eppendorf tubes in the presence of 100 mL of 11% sodium hypochlorite and 5 mL of 32% hydrochloric acid in a sealed desiccator.

Bacterial material and growth conditions

All bacteria were grown on LB agar media prepared by adding 20 g of LB and 10 g bacteriological agar (DIFCO) to 1 L of doubly distilled water and autoclaving for 20 min at 120°C. Antibiotics were added according to the specific resistances of bacteria at final concentrations of 50 μg/mL kanamycin, 25 μg/mL gentamycin, and 25 μg/mL rifampicin. Plasmids were multiplied in chemically competent E. coli strain DH5α and extracted with a GenElute plasmid mini extraction kit (Sigma-Aldrich) following the manufacturer’s protocol.

Construction of CRISPR libraries

General design principles: The following steps were conducted for all libraries generated in this study. The gene annotations and assignments to gene families in A. thaliana were obtained from the PLAZA Plant Comparative Genomics Database (version 5.06;²⁴), retaining only the canonical transcript with the longest coding sequence for each gene. Coding sequences for each gene family were extracted using BEDTools.⁸⁹ For each gene family, multiple sequence alignment was performed using MAFFT,⁹⁰ followed by phylogenetic tree reconstruction using UPGMA.⁹¹ Large gene families were partitioned into subfamilies, each having at most eight genes. For each family/subfamily, we iterated over all internal nodes of the phylogeny, designing sgRNAs targeting the respective subgroup of genes (i.e., the genes descended from that internal node). In all libraries, potential sgRNA targets were confined to the first two-thirds of the coding sequence. In case the same sgRNAs were generated twice (for different subgroups of homologous genes, with one subgroup being a subset of another), only one occurrence was retained.

Trait-oriented single sgRNA library: To identify genes associated with nutrient transport, we analyzed RNA-sequencing data from Arabidopsis roots under nutrient deficiency conditions for specific nutrients (Fe, N, P, Ca, Mg, K, and S), as obtained from.⁴² From 33,558 analyzed genes, 5,609 showed significant differential expression under one or more nutrient deficiencies (q-value <0.05). These differentially expressed genes were cross-referenced with the ARAMEMNON transporter database,⁹² identifying 452 nutrient-responsive transporters. This set, along with their closely related homologs (total of 707 genes, belonging to 201 homologous gene families and subfamilies), formed the basis for our CRISPR library design. The CRISPys algorithm⁴³ was recursively applied to each internal node in the gene tree to determine the optimal sgRNAs for targeting the induced genes descendant from that node. The efficiency of the sgRNAs was assessed using MOFF,⁴⁵ with a targeting efficiency threshold of Ω = 0.15. We then applied the following filtration steps to remove: (1) sgRNAs targeting sparsely-related genes within the subgroup of targeted genes [i.e., filtering sgRNAs that target a monophyletic cluster of genes whose most recent common ancestor (MRCA) is separated by more than three branches from the MRCA of another cluster]; (2) sgRNAs containing restriction sites; (3) sgRNAs with potential off target effect, as identified using crispritz⁹³ searching the Arabidopsis genome for off-targets with up to four mismatches (candidate sgRNAs were filtered in case one of their off-targets had a MOFF score above 0.15); (4) To prevent multiple sgRNAs from targeting the same genomic region, we allowed a maximum of 2 bp overlap between sgRNAs. Finally, for each subgroup of genes, the top eight sgRNAs with the highest predicted targeting efficiency were retained.

For 74 of the differentially-expressed transporters, we could not design potential sgRNAs using the above procedure (21 belonging to single-gene families and 53 for which CRISPys could not generate designs). For these genes, we designed singleton sgRNAs (targeting only one gene) by searching their coding sequences for potential CRISPR targets and selecting the top eight according to the uCRISPR⁴⁷ score, applying the same filtering criteria as detailed above. In total, we consolidated the designs from all 201 subfamilies and singletons into a library of 2,859 sgRNAs targeting 707 genes, encompassing all 452 differentially-expressed transporters and their closely-related homologs.

Multiplex library

The design process for each of the 201 families/subfamilies followed the same general steps as the design for the single sgRNA library, but with the goal of identifying optimal pairs of sgRNAs. For each subgroup in the gene family phylogeny, sgRNA candidates targeting multiple genes were generated using CRISPys. These were supplemented with five sgRNAs targeting each individual gene (i.e., singletons). All sgRNA candidates were filtered using the same criteria detailed above. We then exhaustively evaluated each possible sgRNA pair {sg₁, sg₂}, computing their editing potential to target the set of genes using the following function:

$$f\left(s_1,s_2|G\right)=\sum _{g_i\in G}1-\left(1-\delta \times {\phi }_{s_1}\left(g_i\right)\right)\times \left(1-\delta \times {\phi }_{s_2}\left(g_i\right)\right)$$

where G is the set of genes in the given subgroup, φ_s(g_i) is the targeting potential of sgRNA s to target gene g, as assessed using the MOFF function,⁴⁵ and δ is a correction factor (set here to 0.9) that prevents he targeting probability from reaching 1 when the editing potential of one or both of the sgRNAs is maximal for a given gene. At each internal node, the top three sgRNA pairs were selected. Across all 201 families/subfamilies, this procedure resulted in 1,599 multiplexes.

Determining sgRNA targeting sparsity

Given the assumption that genes with high sequence similarity are more likely to exhibit functional redundancy, there is a need to design sgRNAs that effectively target the homologous genes most closely related within the same family, which we termed monophyletic clusters. To accomplish monophyletic cluster targeting, we first classified gene families into clusters, where each cluster forms a monophyletic group exclusively comprising targeted genes. For each targeted monophyletic group, the shortest distance to all other clusters was computed, where the distance between two clusters was defined as the minimal number of branches separating their respective most recent common ancestor. sgRNAs were then removed if the minimum distance for at least one of the obtained clusters exceeded a predefined threshold (Figure 2B). A threshold value of 3 effectively retained instances where there is only one untargeted cluster located between two targeted clusters while excluding instances where two or more untargeted clusters are situated between targeted clusters (Figure 2B). The restricted monophyletic clusters, combined with the high stringency of the MOFF scoring function, are expected to enhance overall efficiency of the multi-targeted CRISPR library approach.

Agrobacterium transformation

Electro-competent Agrobacterium tumefaciens strain GV3101 was incubated on ice with 100 ng plasmids for 2 min, then electroporated in a MicroPulser (BIO-RAD) (2.2 Kv, 5.8 ms). Bacteria were transferred immediately to 1 mL liquid LB and shaken for 2 h at 28°C. Subsequently, bacteria were plated on LB agar plates containing the relevant antibiotics for 2 days at 28°C. For all libraries, we used a total of >10,000 Agrobacterium colonies.

Plant DNA extraction and PCR

Young leaves/roots from each plant (about 100 mg) were placed in a 2-mL round-tip Eppendorf tube and frozen in liquid nitrogen. The plant matter was crushed using a tissue-lyser to a thin powder and homogenized with 400 μL DNA extraction buffer (200 mM Tris-HCL, pH 7.5–8.0, 25 mM EDTA, 250 mM NaCl, 0.5% SDS). The tubes were vortexed and centrifuged for 1 min at 13,000 rpm in an Eppendorf mini centrifuge. The supernatants were transferred to new tubes, and DNA was precipitated with 300 μL isopropanol and incubated for 2 min at room temperature, followed by centrifugation at 13,000 rpm for 5 min. The pellets were washed with 400 μL 70% EtOH and centrifuged for 1 min at 13,000 rpm, then dried and resuspended in 100 μL doubly distilled water. DNA amplification for sequencing and cloning was done by PCR in a Sensoquest labcycler using the Taq Ready Mix (HyLabs) following the manufacturer’s protocol.

Plant phenotyping

For leaf area measurements, plants were grown on soil, one plant per pot, in a growth chamber under long-day conditions. The surface area was measured using ImageJ software (http://rsbweb.nih.gov/ij/index.html).

Phenomics

Morphological and color shade parameters were analyzed with the PlantScreen Phenotyping System (Photon Systems Instruments (PSI)). Plants were sowed in PSI standard pots and imaged after 20–30 days.

Cross-sections

Leaf cross-section shown in Figure 3A was performed as previously described.^94,95 Leaves from 3-week-old plants were fixed in 4% paraformaldehyde for 1 h, rinsed twice in 1× PBS, embedded in 8% agarose, and sectioned to 100-μm slices using a Leica VT1000S vibratome. Sections were counterstained with 0.1% calcofluor white in 1× PBS solution for 30 minutes. Next, the sections were washed in 1× PBS for 30 minutes with gentle shaking. For imaging, sections were mounted directly in 1× PBS and imaged using a Zeiss LSM 780 inverted microscope.

Confocal imaging

Seedlings were stained in 10 mg/L propidium iodide for 1 min, rinsed, and mounted in water. Seedlings were imaged on a Zeiss LSM 780 laser scanning confocal microscope with the laser set at 514 nm for YFP, 488 nm for GFP, and 493 nm for propidium iodide excitation. Emission filters used were 517–570 nm. Image analysis and signal quantification were done using ZEN lite 2012 software.

Histochemical GUS staining

For histochemical detection of GUS activity, plant tissues were incubated for 16 hours at 37°C in 100 mM sodium phosphate buffer (pH 7.0) containing 0.1% Triton X-100, 1 mM 5-bromo-4-chloro-3-indolyl-β-D-glucuronic acid cyclohexylammonium salt (Sigma-Aldrich), 2 mM potassium ferricyanide, and 2 mM potassium ferrocyanide.⁹⁶ The tissues were then immersed in 70% ethanol until they became transparent. GUS-stained tissues were imaged using a Zeiss Stemi 2000-C stereomicroscope, and images were captured with ZEN software (Zeiss).

Sulfate extraction and quantification

For sulfate quantification, rosette leaves of Arabidopsis (∼100–200 mg fresh weight) were harvested and immediately frozen in liquid nitrogen. Frozen samples were ground to a fine powder in a pre-chilled mortar and pestle, transferred into pre-weighed tubes, and the fresh weight was recorded. Sulfate was extracted by adding 1 mL of deionized water (approximately 10 μL per mg FW), followed by vortexing and incubation at 70°C for 20 min. After cooling on ice, samples were centrifuged at 13,000 rpm for 10 min at 4°C, and the resulting supernatant was filtered through a 0.22 μm syringe filter. Sulfate content was analyzed using 930 Compact IC Flex with an anion-exchange column and quantified against a standard curve of sodium sulfate.

Cloning of cell-type-specific CRISPR vectors

To generate cell type-specific cloning vectors, Golden Gate restriction sites were added upstream to the sequence of intronized Cas9. Two SapI sites were cloned into a pICH41295 vector⁵⁷ to generate the final vector (including two SapI sites for promoter insertion, intronized Cas9 coding sequence, and two BsaI sites for library cloning). Moclo cloning with the following vectors was performed as described in Grutzner et el.⁵⁷: pAGM47523 (Level 0 zCas9i vector), pICH47742 (Empty Backbone), pAGT5472 (L1, LacZ), pAGM55145 (L1, ole citrine), pICH41780 (L1 vector), pAGM35171 (Level 1 position 1 part containing the Nos promoter, the BAR gene, and the Ocs terminator), pICH49344 (for L1 construction, nosT), and pAGM4723 (L2 acceptor vector).

Tissue-specific promoters were amplified from genomic DNA or from plasmids containing the specific promoters using Phusion high-fidelity Taq polymerase (New England Biolabs) following the manufacturer’s protocol. The amplification was carried out using primers containing SapI restriction sites (Table S3) necessary for cloning into the SapI sites in the final destination vector. Promoters that contain SapI sites in their sequence (pKST1 and pSIG6) were cloned into the destination vector in a similar way to the insertion of the SapI sites into the pICH41295 vector using primers that contain BbsI sites (Table S4). The promoters used in this study for library construction were previously described and characterized: pPGP4⁹⁷, pARSK1,⁴⁹ pSIG6,⁵² pKST1,⁵⁰ pSUC2,⁹⁸ pCORI3,⁵⁴ pML1⁸⁷, pSCR,⁹⁹ and pAPL.^88,100

Library cloning

A plasmid library expressing a pool of sgRNAs or tandem sgRNA (for the multiplexed library) was designed and synthesized as described in Hu et al.³ The libraries were cloned directly from the amplified oligo pools into vectors (tissue-specific as indicated above, or pAGM55261 for using the meristematic RPS5 promoter) using the Golden Gate¹⁰¹ method. An aliquot of 2 μL of Golden Gate products was transformed into 50 μL E. coli DH5α competent cells by heat shock reaction at 42°C for 30–60 s. For each construct, we performed 12–18 transformations. After 1 h of shaking at 37°C, samples from three tubes were pooled, and the bacteria were plated on 145/20 mm LB plates containing 50 mg/mL kanamycin, resulting in four to six plates for each construct. Cells were grown overnight in a 37°C incubator. Bacteria from each plate were scraped into 1 L of sterile LB media with kanamycin, and cultures were shaken at 37°C overnight. For plasmid extraction, we used QIAprep Spin Midiprep (Qiagen 20–12143) or Maxiprep (Qiagen 20–12162) kits to extract bulk DNA from >10,000 E. coli colonies. Next, aliquots of 1 μL of DNA were transformed into Agrobacterium by electroporation. For each promoter, 12–18 transformations were performed to generate at least 10,000 Agrobacterium colonies. After 2 h of shaking at 30°C, the Agrobacterium were plated on 145/20 mm LB plates containing 50 μg/mL kanamycin, 25 μg/mL gentamycin, and 25 μg/mL rifampicin, resulting in four to six plates for each construct. Bacteria were grown for 48 h at 30°C. Next, each construct was transformed into six trays of Col-0 plants as described above. T₁ seeds were collected and sown in soil. When the plants were two weeks old, they were sprayed with 0.1% BASTA every 3 days, 3 times in total.

Constructs cloned and used in this study are listed in Table S8.

Deep sequencing

For deep sequencing analysis of the multiplexed transporter library, we created an amplicon using PCR for the sgRNA using the following primers: forward, ctactagaattcgagctcggag; reverse, atgggaattcgtaaagcgaaaaaaaa. Following amplification, the PCR product was purified using the NucleoSpin Gel and PCR Clean-up system (MACHEREY-NAGEL), and samples were sequenced by Novogene using a paired-end 150-bp or 250-bp read length. Deep sequencing data were analyzed using Python. Numbers of reads per sgRNA sequence were determined using the Biopython package. Following deep sequencing, the coverage of represented sgRNAs or tandem sgRNAs in the library was calculated. A skew value was also calculated: 3 ^∗ (Mean – Median)/Standard Deviation, as calculated in previous studies.^3,102

Network construction

The network graphs were created using Cytoscape (http://www.cytoscape.org/).

Bioinformatics

DNA sequences alignment was performed using the SnapGene alignment software (http://www.snapgene.com/) and the NCBI BLAST tool (https://blast.ncbi.nlm.nih.gov/Blast.cgi/). Gene sequences and relevant information were obtained from the Arabidopsis Information Resource (https://www.arabidopsis.org/).

Barcoding

First, to confirm successful amplification of a single sgRNA from the multiplex cassette in transgenic T1 plants, primers without overhangs were tested (forward: 5′- CACATCGCTTAGATAAGAAAACG-3′, reverse: 5′- GTTGATAACGGACTAGCCTTA-3′). For barcoding, 64 sequences, each 8 base pairs long, were chosen from Hamady et al.⁹⁸ The barcodes were split: 32 were added to the 5′ end of the forward primer, and the other 32 to the 5′ end of the reverse primer. All primer sequences and their combinations are detailed in the Table S5. Each sample was uniquely amplified using a distinct set of barcode primers. PCR was run with an annealing phase of 55°C for 5 cycles to bind the primers to the DNA, followed by 25 cycles at 61°C to enable barcode creation. The elongation step lasted 5 s using Vazyme’s X2 Rapid Taq Master Mix. The resulting fragment was 149-bp long, including the two 8-bp barcodes. Since the multiplex cassette contained two sgRNAs, two amplifications were required per plant (due to the size limitation of the barcoded region). PCR products were then pooled, run via gel electrophoresis, and the relevant bands were excised, and DNA was purified using the Macherey and Nagel Nucleospin Gel and PCR Clean-up Kit. Samples were deep-sequenced by Novogene on the Illumina NovaSeq X Plus, PE−150.

Next, quality checks were carried out on the paired-end sequencing reads for data analysis to ensure high reliability. The overlapping region, which includes the sgRNA and barcodes, was examined for each read pair. Discrepancies in this region led to reads being discarded. Similarly, if barcode sequences at the 5′ end of the reads did not match the preselected list, those reads were excluded. Any mismatches, insertions, or deletions in non-variable regions were also grounds for removal. We assigned the remaining read pairs to their corresponding plant numbers using barcode combinations listed in Table S5. The multiplex sequences were then linked to each plant, with two sgRNAs amplified for each plant, using identical barcode combinations to determine multiplexes.

Even after the initial filtering, some residual noise was still present, prompting the need for additional analysis to accurately capture the relevant data. We organized the reads for each plant according to their frequency and evaluated their distributions. Plants where the most frequent read of a multiplex was at least 1.5 times more common than the next were classified as having a single multiplex. If this ratio was not met, we compared the second-highest read with the third. If the second read was 1.5 times more prevalent than the third, we determined that two multiplexes were present. If neither of these conditions was satisfied, the plant was marked as having an undefined multiplex.

Data S1 contains data on 742 plants, each containing a single multiplex targeting nutrient-related transporter genes.

Genotyping

To identify the sgRNAs of transgenic plants, we PCR amplified the vectors using forward primer: tactagatcgacgctactag and the reverse primer: tgttccgcgacattctagaac. The sgRNAs and multiplexes used in this study are listed in Table S6. To identify the mutation in the targeted genes, primers were designed to amplify the sgRNA region in each targeted gene. The sequence for amplification for each gene was ∼300 bp long. The amplified sequence was then Sanger sequenced to determine if there is a mutation (chromatograms presented in Figure S6). The mutations in the targeted genes in the pRPS5:Multiplex-PIN3,4,7 are homozygous, while the mutations in the other lines are heterozygous or biallelic.

Calculating plant CRISPR population size and probabilities

To estimate the plant population size required to represent all vectors in a given library, we calculated the expected recovery probability of each vector as a function of the number of transformed plants (x) and the total number of sgRNAs (n). The computation is based on the assumption of equal transformation probability among vectors and that each transformed plant receives one construct at random. The probability that a specific vector appears at least once in a population of x plants is given by:

$$f\left(x\right)=1-{\left(1-\frac{1}{n}\right)}^x$$

This relationship provides an estimate of the expected representation of constructs in a given population. For example, screening 1,425 T₁ plants from a library of 1,599 unique multiplex vectors is predicted to recover approximately 89% of all constructs, closely matching the empirical recovery observed (Table S8, top row).

Phytohormone transport assays in Xenopus laevis oocytes

Coding DNA sequences of SWEET11, SWEET12, SWEET13, and SWEET14 were cloned into Xenopus expression vector pNB1u using the USER cloning technique.¹⁰³ Linear DNA templates for in vitro transcription were generated by PCR with pNB1u plasmid-specific primers (5′ – AATTAACCCTCACTAAAGGGTTGTAATACGACTCACTATAGGG – 3′) and reverse primer (5’ –TTTTTTTTTTTTTTTTTTTTTTTTTTTTTATACTCAAGCTAGCCTCGAG – 3′). Capped cRNA was synthesized in vitro using the mMessage mMachine T7 Kit (Invitrogen), and the concentration of each cRNA was normalized to 250 ng μL⁻¹. Defolliculated stage V–VI X. laevis oocytes were microinjected with 50.6 nL cRNA (or nuclease-free water as mock control) using a Drummond NANOJECT II (Drummond Scientific Company). Injected oocytes were incubated for three days at 16°C in HEPES-based kulori buffer (90 mM NaCl, 1 mM KCl, 1 mM MgCl₂, 1 mM CaCl₂, 10 mM HEPES pH 7.4) supplemented with 100 μg mL⁻¹ amikacin.

Phytohormone uptake assays were conducted as described previously⁷⁵ with minor modifications. Equimolar mixtures (100 μM each) of cytokinins (trans-zeatin (tZ), trans-zeatin riboside (tZR), and isopentenyl adenosine (iPR)) or a combination of gibberellic acid 3 (GA₃) and abscisic acid (ABA) were prepared in HEPES-based kulori buffer (90 mM NaCl, 1 mM KCl, 1 mM MgCl₂, 1 mM CaCl₂, 10 mM HEPES pH 7.4). Three days after cRNA injection, oocytes were incubated for 1 h in either the cytokinin mixture or the GA₃/ABA solution. Following incubation, oocytes were washed four times with kulori buffer (pH 7.4), homogenized in 50% methanol, and stored overnight at −20°C. Subsequently, the extracts were spun down at 15,000 g for 15 min at 4°C, and the supernatant was diluted with water and analyzed by liquid chromatography–tandem mass spectrometry (LC–MS/MS) as described previously.⁷⁵

Quantification and statistical analysis

Statistical details of the experiments, including the statistical tests, the exact value of n, and dispersion and precision measures (e.g., mean, SD), can be found in the figure legends.

Published: March 9, 2026

Supplemental information

Data S1. CRISPR library design for functional gene targeting

The dataset contains the design and annotation of a single and multiplex CRISPR-Cas9 sgRNA libraries, organized into sub-libraries based on nutrients. Each row in the table represents a single sgRNA or a multiplex of 2 sgRNAs, annotated with the gene ID, number of genes targeted, and predicted on-target gene IDs and scores. Functional group classification into sub-library membership is included in the “nutrient” column

Data S2. Barcode-based identification of sgRNAs in plants

The dataset contains the results of high-throughput barcode sequencing used to identify sgRNAs integrated into independent T1 plants. Each plant was uniquely indexed using a pair of 8-nucleotide barcoded primers (forward and reverse) to create a double-barcode signature, enabling demultiplexing of pooled samples following deep sequencing. Each row in the table represents an independent plant from the library, showing the respective multiplex of 2 sgRNAs and target gene ID numbers