High-throughput Raman-activated cell sorting of microalgal genome-wide edited library revealed a regulatory pathway for carotenoid synthesis Authors: Qintao Wang, Yanhai Gong, Lianhong Wang, Nana Lv, Xuefeng Du, Jiashun Zhang, Yi Xin, Zoran Nikoloski, Yonghua Li-Beisson, Luyang Sun, Bo Ma, Xixian Wang, Jian Xu Published: 2026 ABSTRACT Functional genomics have been hampered by the paucity of efficient methods that connect genotype and metabolic phenotype at single-cell resolution. Using the industrial microalga Nannochloropsis oceanica as a model, we introduced a platform that comprises a genome-wide single-gene-edited mutant library and high-throughput Raman-activated cell sorting (RACS). The CRISPR/Cas-generated library consisted of 3567 microalgal mutants derived from 2397 effective guide RNAs. Label-free sorting of the library for high carotenoid content by RACS unraveled mutations in the violaxanthin de-epoxidase (noVDE) or in the proteasome assembly chaperone 4 (noPAC4) genes. Knocking out all five known noVDEs revealed that the high carotenoid content is due to violaxanthin increase, whilst noPAC4 knockout boosted carotenoid content with elevations in violaxanthin, zeaxanthin, and β-carotene. Genetic and transcriptomic evidence suggested two previously unknown modes of carotenogenesis regulation mediated by noPAC4: epigenetic mechanisms via histone deacetylase (HDAC) and post-translational controls by the 26S proteasome. Therefore, by label-freely sorting single-cell metabolic phenotype and rapidly yet unambiguously tracing it to a genotype, this forward-genetics approach can greatly accelerate the discovery of genes and pathways. Introduction Microalgae are unicellular photosynthetic organisms that generate over 45% of the planet’s annual net primary biomass from carbon dioxide and sunlight1. Moreover, they harbor diverse metabolic capabilities for synthesis of valuable compounds such as proteins, lipids, carbohydrates, vitamins, antioxidants, polyunsaturated fatty acids (PUFAs), and pigments2,3. Therefore, microalgae are promising cell factories for carbon-neutral production of nutraceuticals, pharmaceuticals and biofuels4–6. To fully harness their metabolic capabilities, hundreds of microalgal genomes have been sequenced7; yet, one striking finding is the large number of functionally unknown genes. For example, in the model oleaginous microalga of N. oceanica, over 50% of all predicted genes lack functional annotations, due to a lack of sufficient sequence similarity to other annotated genes8,9. Moreover, even in the case of high sequence similarity to genes with experimentally validated functions, direct transfer of functional annotation in the absence of experimental verification is often problematic, since distinguishing the functions among homologous genes is difficult10–14. Such lack of functional validation for a vast fraction of genes has long hindered mechanistic understanding and metabolic engineering of microalgae. To identify genes that underlie a particular metabolic trait, forward genetics approaches generally start with mutagenesis, whereby mutations are induced in the genomes, followed by screening for specific phenotypic changes due to the mutations, and then mapping of the genes responsible for the observed phenotype. Such approaches have been widely applied in microalgae to identify genes implicated in lipid metabolism and photosynthesis15,16, however limitations are profound. (i) Mutant library construction has traditionally relied on genetic mutagenesis induced by physical or chemical mutagens, or via random insertion15, and identifying genotypes with the desired traits is usually extremely laborious and time-consuming17–19. Moreover, mutants in so-generated mutant libraries generally have multiple mutation sites, which impedes an unambiguous assignment of mutant phenotype to a specific genetic locus19–21. Whilst random insertion can yield single-locus mutations, many of these mutations may occur in intergenic or non-coding sequences, which may generate very subtle phenotypes that are difficult to measure17. (ii) Metabolic-phenotype-based screening poses a major challenge for microalgae. Whilst fluorescence-activated cell sorting (FACS) offers a valuable tool to address this challenge, its reliance on tedious fluorescent labeling for a specific protein or metabolite has hindered broad application and may introduce biases due to alteration of the natural state of cells22,23. Although autofluorescence of certain metabolites (e.g. chlorophylls or carotenoids) suggests the potential of label-free sorting via FACS, sorting of carotenoid-hyperproducing microalgae cells have failed: fluorescence intensity of the large amount of chlorophyll in microalgal cells masks that of carotenoids24, which compromises sorting accuracy. Meanwhile, single-cell Raman spectrum (SCRS)-based Raman-activated cell sorting (RACS) holds great promise for metabolic-phenotype-based screening, as it is label-free, high-throughput, information-rich, and broadly applicable to all kinds of cells25,26. Recently, a series of RACS technologies and devices have emerged, which can be broadly categorized into three types (Supplementary Table 1): (i) coherent Raman sorting systems27–29, which provide high signal intensity but rely on nonlinear excitation and are prone to spectral distortions that compromise quantitative metabolic phenotyping, (ii) static-mode or optical tweezers-based spontaneous Raman sorting systems such as Raman-activated cell ejection (RACE)30–33, Raman-activated gravity-driven encapsulation (RAGE)34 and optofluidic RACS35,36, which offer high spectral fidelity but suffer from relatively low throughput, and (iii) trap-free or positive dielectrophoresis (pDEP)-based spontaneous Raman sorting systems such as Raman-activated microfluidic sorting (RAMS)37–39 and Raman-activated droplet sorting (RADS)40,41, which achieve single-cell-resolved sorting of higher throughput but are often constrained by long-term operational stability and sensitivity for rare-event enrichment. These trade-offs have so far prevented any single RACS modality from being fully compatible with genome-scale mutant library screens, which require the simultaneous satisfaction of three constraints: (i) quantitative metabolic readout at the single-cell level, (ii) high-throughput enrichment of rare phenotypic variants from large mutant libraries, and (iii) stable long-duration operation without signal drift or loss of sorting fidelity. The pDEP-DLD-RACS platform integrates spontaneous Raman spectroscopy with pDEP-induced deterministic lateral displacement (pDEP-DLD)42,43, thereby overcoming these coupled limitations. Specifically, spontaneous Raman provides information-rich and quantitatively reliable SCRS for metabolic phenotyping, while pDEP-DLD enables continuous, stable and high-throughput focusing, trapping and sorting of cells over extended operation times. Therefore, this design should allow pDEP-DLD-RACS to sort rare phenotypic variants from large mutant libraries and to preserve their metabolic signatures for downstream sequencing, making it particularly suitable for high-throughput genotype–phenotype mapping at single-cell resolution (Supplementary Table 1). However, such potential has not yet been explored. Nannochloropsis spp. are unicellular algae known for their rapid growth, adaptability to various environmental conditions, high lipid content, and production of high-value metabolites such as triacylglycerols (TAG), eicosapentaenoic acid (EPA), and carotenoids44. Employing the industrial oleaginous microalga of N. oceanica as a model, we introduce a forward genetics platform that consists of a genome-wide single-gene-edited mutant library and high-throughput RACS. By generating the first such library for microalgae and then sorting it via RACS, we rapidly uncover dozens of genes whose mutations reduce carotenoid contents, despite the lack of previous evidence linking them to carotenogenesis. In particular, we propose two previously unknown modes of carotenogenesis regulation mediated by proteasome assembly chaperone 4 (noPAC4), which involve epigenetic mechanisms via histone deacetylase (HDAC) and post-translational controls by the 26S proteasome. By tracing single-cell metabolic phenotype to single-gene genotype in a label-free, high-throughput manner, the platform can greatly accelerate the discovery of functional genes and pathways in microalgae and other organisms. Results Construction of a CRISPR/Cas system-based genome-wide mutant library of N. oceanica In contrast to the classical approaches (Fig. 1A), we proposed a strategy to establish the genotype-phenotype links by sequencing the RACS sorted cells of genome-wide, programmable, and targeted single-gene edited mutants (Fig. 1B). The platform consisted of two main modules. (i) A genome-wide single-gene targeted mutagenesis library for N. oceanica was created by the CRISPR/Cas system, which greatly simplified large-scale single-gene editing (and thus mutagenesis) due to its targeted-mutagenesis nature, as well as its efficiency, simplicity, and programmability45. Moreover, mutation sites can be tracked using the gRNA sequence, which is easily detectable. No such libraries have previously been reported for microalgae46, despite the successes in various higher plants such as rice, tomato, soybean, and maize47. (ii) A metabolism-driven label-free single-cell screening platform was established for N. oceanica based on high-throughput RACS. A spontaneous single-cell Raman spectrum (SCRS) provides the intrinsic biochemical profile of a cell at a given state and thus can be considered as a function-based snapshot of the molecular state of the cell. Importantly, this strategy is label-free, noninvasive, rich in metabolic information, and broadly applicable26.Fig. 1High-throughput mining of functional genes from CRISPR/Cas mutant library via Raman-activated cell sorting.A The conventional approach. The construction of mutant libraries is commonly achieved through physical or chemical mutagens, or via random insertion. Subsequent screening of these libraries typically relies on FACS or microtiter plate-based assays. B The approach here that couples genome-wide mutagenesis library with high-throughput RACS. The CRISPR/Cas system is capable of generating targeted single-gene mutated libraries at a global scale, with mutation sites readily identifiable through gRNA. This feature greatly facilitates the identification of functional genes based on the mutated genotypes following the SCRS-based high-throughput cell sorting. RACS enables label-free, culture-free, non-invasive screening of mutants at single-cell level, based on an intrinsic biochemical profile of the cell that captures its metabolic phenome. This figure was created in BioRender. Liu, Y. (2026) https://BioRender.com/owwivd8. High-throughput mining of functional genes from CRISPR/Cas mutant library via Raman-activated cell sorting. A The conventional approach. The construction of mutant libraries is commonly achieved through physical or chemical mutagens, or via random insertion. Subsequent screening of these libraries typically relies on FACS or microtiter plate-based assays. B The approach here that couples genome-wide mutagenesis library with high-throughput RACS. The CRISPR/Cas system is capable of generating targeted single-gene mutated libraries at a global scale, with mutation sites readily identifiable through gRNA. This feature greatly facilitates the identification of functional genes based on the mutated genotypes following the SCRS-based high-throughput cell sorting. RACS enables label-free, culture-free, non-invasive screening of mutants at single-cell level, based on an intrinsic biochemical profile of the cell that captures its metabolic phenome. This figure was created in BioRender. Liu, Y. (2026) https://BioRender.com/owwivd8. To generate a CRISPR/Cas-based mutant library of N. oceanica (Fig. 2A; Methods), a gRNA library was designed, synthesized and subsequently ligated to plasmids containing the CRISPR/Cas system. This plasmid library was transformed into N. oceanica to construct the mutant library, which was then sequenced using next-generation sequencing (NGS) to identify the mutation sites. Precise generation of gRNAs was ensured by engineering them with hammerhead (HH) self-cleaving ribozymes at the 5’ end and hepatitis delta virus (HDV) self-cleaving ribozymes at the 3’ end in the episome-based CRISPR/Cas system in N. oceanica48. The HH structure dictates that its first six bases must correspond to the reverse complement of the first six bases of the gRNA sequence (Supplementary Fig. 1A, purple). After the ligation of the gRNA target sequence library to the plasmid containing the CRISPR/Cas system via infusion, we sequenced 15 colonies and identified just one accurately ligated plasmid (Supplementary Fig. 1A, green): such observed low accuracy impeded the knockout efficiency in constructing the mutant library. It seemed that the loop structure necessary for HH self-assembly obstructed the correct ligation of the gRNA PCR library to the plasmid that harbors the CRISPR/Cas system.Fig. 2Construction and evaluation of the plasmid library for genome-wide single-gene mutations via the CRISPR/Cas system.A The overall screening process from a genome-wide CRISPR/Cas mutant library of N. oceanica based on SCRS. The gRNA library was designed using CHOPCHOP and synthesized by custom arrays, and amplified and ligated to tRNA-based CRISPR/Cas system to construct the plasmid library. The plasmid library was then transformed into N. oceanica, resulting in 5760 colonies, with target site mutations identified through NGS. Genes that underlie the phenotype of carotenoid overproduction were then identified from the CRISPR/Cas mutant library using RACS and NGS, at single-cell resolution. B The process of constructing a genome-wide CRISPR/Cas plasmid mutant library for transforming N. oceanica. The synthesized gRNA mutant library was amplified and ligated to the digested plasmid with the CRISPR/Cas system. Approx. 30,000 colonies were picked, and the extracted plasmids were sequenced by NGS to evaluate the quality of the plasmid library. C Accuracy and coverage of the gRNA library and the plasmid library. D The genome-wide distributions of relative abundance of gRNAs reveal uniformity of the gRNA library and the plasmid library. The y-axis represents the relative abundance of each gRNA, quantified by the log2-scaled read counts. The Gini index, indicating the inequality in gRNA-abundance distribution, is shown above the violin plots for the two libraries. A was created in BioRender. Liu, Y. (2026) https://BioRender.com/2e7m0v4. Construction and evaluation of the plasmid library for genome-wide single-gene mutations via the CRISPR/Cas system. A The overall screening process from a genome-wide CRISPR/Cas mutant library of N. oceanica based on SCRS. The gRNA library was designed using CHOPCHOP and synthesized by custom arrays, and amplified and ligated to tRNA-based CRISPR/Cas system to construct the plasmid library. The plasmid library was then transformed into N. oceanica, resulting in 5760 colonies, with target site mutations identified through NGS. Genes that underlie the phenotype of carotenoid overproduction were then identified from the CRISPR/Cas mutant library using RACS and NGS, at single-cell resolution. B The process of constructing a genome-wide CRISPR/Cas plasmid mutant library for transforming N. oceanica. The synthesized gRNA mutant library was amplified and ligated to the digested plasmid with the CRISPR/Cas system. Approx. 30,000 colonies were picked, and the extracted plasmids were sequenced by NGS to evaluate the quality of the plasmid library. C Accuracy and coverage of the gRNA library and the plasmid library. D The genome-wide distributions of relative abundance of gRNAs reveal uniformity of the gRNA library and the plasmid library. The y-axis represents the relative abundance of each gRNA, quantified by the log2-scaled read counts. The Gini index, indicating the inequality in gRNA-abundance distribution, is shown above the violin plots for the two libraries. A was created in BioRender. Liu, Y. (2026) https://BioRender.com/2e7m0v4. To streamline the design and ligation of the gRNA pool to the plasmid and to ensure accurate gRNA expression, we employed glycine tRNA in N. oceanica instead of HH self-cleaving ribozymes (Supplementary Fig. 1B; Supplementary Table 2). To evaluate the knock-out efficiency, a light-harvesting complex protein (LHC; NO13G02860) was chosen as the target gene. Sequencing of the target sites in twelve transformants revealed mutations in each of them (Supplementary Fig. 1B), indicating a high-efficiency of tRNA-based CRISPR/Cas system in N. oceanica. Moreover, the accuracy of gRNA target sequence ligation to the system was 85%, as 17 among the 20 sequenced colonies were correct (Supplementary Fig. 1C, blue). Next, we assessed the coverage, uniformity, and accuracy of the plasmid library which underlie the quality of the eventual tRNA-based CRISPR/Cas mutant library. These factors depend on the design, synthesis, and ligation (to the CRISPR/Cas-encoded plasmid) of the gRNA library. To maximize the coverage, a total of 9709 gRNAs were designed to span ~94% of the entire protein-coding sequences (Supplementary Data 1; Methods). The remaining protein-coding genes were not targeted for knocked out, to minimize off-target effects and to avoid the persistence of activity from resultant truncated proteins. The gRNA target sites from both libraries were amplified (Supplementary Table 2) and sequenced (Fig. 2B), indicating 92% accuracy of the gRNAs in the gRNA PCR library, and 84% in the plasmid library (Fig. 2C). Furthermore, in the plasmid library, all designed gRNAs were represented (Fig. 2C), with a high degree of uniformity in the distribution of gRNA abundance (Gini index of 0.04 for the plasmid library and 0.02 for the gRNA PCR library; Fig. 2D), thus the library was well-balanced and each gRNA was represented at a similar level. These findings confirmed the high quality of the plasmid library for a CRISPR/Cas system-based genome-wide single-gene mutated library of N. oceanica. High mutational efficiency and low bias in mutated genes in the N. oceanica mutant library After transformation, a total of 5760 transformants, representing all colonies recovered across approximately 30 transformation plates (15 cm diameter each), were distributed among 240 24-well plates, and mutation sites in each well were identified via NGS (Fig. 3). Specifically, the 24 transformants from each 24-well plate were pooled for whole-genome sequencing to map the mutations present within these genomes (Fig. 3A). When pooling mutants for genome sequencing, the observed frequency of mutation sites can be skewed due to the growth-rate variations among colonies on a plate (leading to low sequencing coverage for specific genomes), and the inherent variations in sequencing depth across different genomic regions. To maximize the chance of detecting mutation sites, we initially set the detection threshold to a minimum of one sequencing read. However, this approach also captured inherent single-nucleotide polymorphisms (SNPs), sequencing artifacts (e.g., those introduced during PCR amplification prior to sequencing), and potential DNA damage induced by zeocin. Therefore, we chose to employ the specific gRNA sequence present in each well to identify the mutation that was caused by the designed gRNA. Microalgal transformants in the 240 24-well plates were organized into ten pools, with each pool (24 plates) carefully collected (Fig. 3B). To confirm the gRNA composition within each well, the gRNAs were amplified using PCR (primers in Supplementary Table 2) and sequenced by NGS. Then the gRNAs in each well were verified based on their probability of occurrence. To validate the NGS-predicted mutations, we performed Sanger sequencing on the first three 24-well plates (Supplementary Note 1, Supplementary Data 2 and Supplementary Table 3). The detection of gRNAs in 68 out of 72 colonies indicated a high transformation efficiency of 94%. Thus, the number of actual transformants among the 5760 selected colonies was estimated to be 5440. Sanger sequencing identified mutations in 37 sites, while NGS detected 41. Among these, 35 sites were consistently identified by both methods, demonstrating the high accuracy of NGS-based mutation detection. Notably, among the 47 mutants identified with identical gRNA by Sanger sequencing and by NGS, 37 (78.7%) harbored mutations at the target sites. Thus, gRNA presence is a robust predictor of successful gene editing and facilitates the isolation of mutated clones through gRNA-based enrichment.Fig. 3Construction and evaluation of the CRISPR/Cas based genome-wide N. oceanica mutant library.A Detection of mutation sites in the mutant genomes. The mutants from 24 colonies on a single plate were pooled together, and their genomic DNA were extracted and sequenced via NGS to identify all mutations surrounding the gRNA target sites. B The identification of gRNA in each mutant. The mutants of 24 plates were pooled as shown, and their genomes were extracted and amplified using distinct primers. Subsequently, the PCR products were sequenced through NGS to determine the gRNA of each mutant. C Chromosomal distribution of mutation sites in the mutant library across the reference genome of wild-type strain. The numbers of mutated target genes, designed target genes and total genes on each chromosome are shown for the 30 chromosomes. D Statistical analysis of mutation sites and types in the N. oceanica genome-wide edited library. (A, B) were created in BioRender. Liu, Y. (2026) https://BioRender.com/gw3pz30. Construction and evaluation of the CRISPR/Cas based genome-wide N. oceanica mutant library. A Detection of mutation sites in the mutant genomes. The mutants from 24 colonies on a single plate were pooled together, and their genomic DNA were extracted and sequenced via NGS to identify all mutations surrounding the gRNA target sites. B The identification of gRNA in each mutant. The mutants of 24 plates were pooled as shown, and their genomes were extracted and amplified using distinct primers. Subsequently, the PCR products were sequenced through NGS to determine the gRNA of each mutant. C Chromosomal distribution of mutation sites in the mutant library across the reference genome of wild-type strain. The numbers of mutated target genes, designed target genes and total genes on each chromosome are shown for the 30 chromosomes. D Statistical analysis of mutation sites and types in the N. oceanica genome-wide edited library. (A, B) were created in BioRender. Liu, Y. (2026) https://BioRender.com/gw3pz30. By aligning the gRNAs against the genome sequence at the designated target sites in each well by NGS, we found that 2851 of the 5760 colonies exhibited exactly one mutated gene, 716 colonies displayed two or more mutated genes, and 2193 colonies showed no mutated genes (Supplementary Data 2; Fig. 3C). The mutations were categorized into nine distinct classes: 3663 frameshift-variants, 214 missense-variants, 115 disruptive-inframe-deletions, 78 synonymous-variants, 54 conservative-inframe-deletions, 27 disruptive-inframe-insertions, 13 conservative-inframe-insertions, 10 stop-gained, and 117 variants of multiple types (Fig. 3D). A substantial majority of the mutations (85.4%) were classified as frameshift-variants, which significantly contribute to the generation of loss-of-function mutants for reverse genetics in N. oceanica. In addition, we established a collection of 2397 experimentally validated, effective gRNAs, defined by their success in inducing target‑gene mutations (“one gene, one gRNA”). This curated set, despite limitation in its yield due to uncertainties in computational design and cellular lethality, constitutes a verified resource that can obviate the need for extensive future validation, thereby accelerating CRISPR‑based workflows (Fig. 3C; Supplementary Note 2; https://nandesyn.single-cell.cn/CRISPR_library). Notably, only ~1.5% of the mutant colonies exhibited chromosomal integration of the plasmid (Supplementary Data 3), which is logical as the plasmid used here is derived from an episomal CRISPR/Cas9 system48 that can be stably maintained as circular DNA. These results validated the successful construction of a CRISPR/Cas-based N. oceanica mutant library, providing 3567 mutated colonies and 2397 efficient gRNAs for subsequent genome editing and mutant screening endeavors. Quantitative analyses of the library revealed 78.1% of its mutated genes as unique among the mutant strains (Fig. 4A), slightly lower than the theoretically expected probability of 84%. This discrepancy may be attributed to the presence of lethal genes that cannot yield viable mutants. The average proximity between adjacent mutated genes was 12.5 ± 12.1 kb, showing no significant deviation from random sampling (Wilcoxon test, p > 0.05; Fig. 4B). Based on the Clusters of Orthologous Genes (COG) database, only the COG category of “translation, ribosomal structure, and biogenesis” exhibited significant difference (in this case, lower-than-expected representation) from the target proteome (Chi-square test, adjusted p < 0.01; Fig. 4C), which may be due to a higher prevalence of lethal genes within this category. Overall, the mutant library exhibited high quality, characterized by a uniform distribution of mutated genes that was genome wide and among the functional categories of proteome.Fig. 4Characterization of the genome-wide edited mutant library of N. oceanica.A The frequency of mutated genes across various colony counts. In the genotype-verified colonies, the number of colonies varies among the different mutated genes, with 99.0% of the mutated genes exhibiting a maximum of three colonies. B Comparison of the spacing intervals between neighboring pairs of validated mutated genes along the chromosome with those of a randomly selected gene set of the same size. Data are expressed as mean ± SD (n = 2367 adjacent gene pairs) for both groups, with each dot representing one adjacent gene pair. No significant differences are observed (Wilcoxon test, p = 0.667), indicating the uniformity of the library. C The proportion of mutated genes validated from this mutant library within each COG category relative to all validated mutated genes. The COG category J: translation, ribosomal structure and biogenesis. These ratios were compared to the values for all genes using a Chi-squared test (p = 0.005). D The nucleotide variations between validated gRNAs and designed gRNAs. The color bar represents the nucleotide frequency (%) difference in validated gRNAs versus designed gRNAs. E Relationship between the number of colonies and that of validated mutated genes. The linear regression line (in light blue) reveals a strong correlation between the number of colonies and the number of validated mutated genes (red dot; Pearson’s r = 0.999, t-test for significance: n = 7 colony numbers, p = 1.99e−8), suggesting that proportionally more mutated genes would be discovered with the increase in the number of colonies picked. The light gray shaded band represents 95% confidence level for the linear regression line. Significance was assessed using unpaired two-tailed Wilcoxon test in (B), two-tailed Chi-squared test with Benjamini–Hochberg correction in (C), permutation test using unpaired two-tailed Z-statistics with Benjamini–Hochberg correction in (D). *p < 0.05 and **p < 0.01. Exact p values, n numbers, mean and SD/SEM values can be found in the Source Data file. Characterization of the genome-wide edited mutant library of N. oceanica. A The frequency of mutated genes across various colony counts. In the genotype-verified colonies, the number of colonies varies among the different mutated genes, with 99.0% of the mutated genes exhibiting a maximum of three colonies. B Comparison of the spacing intervals between neighboring pairs of validated mutated genes along the chromosome with those of a randomly selected gene set of the same size. Data are expressed as mean ± SD (n = 2367 adjacent gene pairs) for both groups, with each dot representing one adjacent gene pair. No significant differences are observed (Wilcoxon test, p = 0.667), indicating the uniformity of the library. C The proportion of mutated genes validated from this mutant library within each COG category relative to all validated mutated genes. The COG category J: translation, ribosomal structure and biogenesis. These ratios were compared to the values for all genes using a Chi-squared test (p = 0.005). D The nucleotide variations between validated gRNAs and designed gRNAs. The color bar represents the nucleotide frequency (%) difference in validated gRNAs versus designed gRNAs. E Relationship between the number of colonies and that of validated mutated genes. The linear regression line (in light blue) reveals a strong correlation between the number of colonies and the number of validated mutated genes (red dot; Pearson’s r = 0.999, t-test for significance: n = 7 colony numbers, p = 1.99e−8), suggesting that proportionally more mutated genes would be discovered with the increase in the number of colonies picked. The light gray shaded band represents 95% confidence level for the linear regression line. Significance was assessed using unpaired two-tailed Wilcoxon test in (B), two-tailed Chi-squared test with Benjamini–Hochberg correction in (C), permutation test using unpaired two-tailed Z-statistics with Benjamini–Hochberg correction in (D). *p < 0.05 and **p < 0.01. Exact p values, n numbers, mean and SD/SEM values can be found in the Source Data file. Furthermore, to evaluate the effect of gRNA base preference on editing efficacy in N. oceanica, we investigated the nucleotide distribution at different positions within the designed and effective gRNAs (Supplementary Fig. 2). No discernible preference for specific nucleotides was found at any position when compared with the designed and effective gRNAs (Z-test, p > 0.01; Fig. 4D), indicating the efficient design of gRNA library. Given the high uniformity of the mutant library, we observed significant positive Pearson correlation between the number of mutated genes and the number of transformants (Fig. 4E), suggesting a saturated mutant library can be constructed by collecting additional mutant clones. Thus, a high-quality genome-wide knockout mutant library of N. oceanica was constructed with high efficiency and high uniformity. High-throughput SCRS-based sorting identified mutated genes underlying carotenoid overproduction Carotenoids, a class of terpenoid pigments with C40 backbones found in microalgae and plants, are pivotal to photosynthesis and widely used in many industries due to their health benefits49. Traditional screening for carotenogenesis genes typically involves time-consuming, labor-intensive, and low-throughput culturing of mutants and measuring changes in carotenoid levels from bulk biomass, before identifying target genes (Fig. 5A)50. In contrast, we propose that SCRS-based sorting of the genome-wide knockout mutant library enables single-cell-resolved, high-throughput screening of mutants with high carotenoid content, thus greatly accelerating function-directed gene mining (Fig. 5A).Fig. 5Label-free high-throughput screening for genes underlying carotenoid overproduction from the genome-wide edited N. oceanica mutant library based on single-cell Raman spectrum.A Comparison of high-throughput RACS and traditional methods for screening carotenoid-overproducing genes. Carotenoid-overproducing mutants were screened by RACS with Raman intensity >99%. The plasmids harboring CRISPR/Cas system in the screened mutants were subsequently amplified by MDA. The gRNAs from the plasmids were then amplified by PCR and sequenced with NGS to identify changes in abundance after the sorting, which offers much higher efficiency yet lower costs than traditional methods. The average SCRS (top panels) and the Pearson correlation between carotenoid concentration and peak Raman intensity (bottom panels) are shown for β-carotene (B), violaxanthin (C), and zeaxanthin (D). Boxplots illustrate the distribution of peak Raman intensity at each carotenoid concentration. Data are presented as box plots showing the median (center line), interquartile range (box; 25th–75th percentiles), and whiskers extending to 1.5× IQR. The bands shaded in light color represent the 95% confidence level for the linear regression lines. For each linear regression, the significance of Pearson’s r was assessed using an unpaired two-tailed t-test (n = 6 concentration levels; r = 0.989 and p < 2.2e−16 for β-carotene, r = 0.996 and p < 2.2e−16 for violaxanthin, r = 0.996 and p < 2.2e−16 for zeaxanthin). E The average SCRS of the mutant library (LIB) and wild-type (WT) are presented, with the variation range (mean ± SD) of Raman intensity at each Raman shift depicted as shaded ribbons. Characteristic Raman peaks are highlighted. F The standard deviations of Raman intensity at each Raman shift are compared between LIB (n = 2651 Raman shifts) and WT (n = 2651 Raman shifts) (Wilcoxon test, p < 2.2e−16). Raman shift regions of 500–3150 cm−1 are considered for the comparison. G The Raman intensities of the v3 peak are compared between LIB (n = 4714 cells) and WT (n = 6231 cells) (Wilcoxon test, p < 2.2e−16). H The Raman intensities of the v3 peak are displayed for single cells derived from LIB and WT. For each sample, a random subsample of 4000 SCRS was selected for comparative analysis. Only Raman intensities exceeding 6000 are included in the comparison. I The lollipop charts that illustrate the level of enrichment for each mutated gene enriched in a RACS-sorted sample, with each sphere color-coded based on the corresponding gene’s pathway or function. Enriched genes shared by at least two replicates are shown. J The overlap of enriched genes among the three replicates. Significance was assessed using paired two-tailed Wilcoxon test in (F) and unpaired two-tailed Wilcoxon test in (G). ****: p < 0.0001. Exact p values, n numbers, mean and SEM values can be found in the Source Data file. A was created in BioRender. Liu, Y. (2026) https://BioRender.com/fh5sj9g. Label-free high-throughput screening for genes underlying carotenoid overproduction from the genome-wide edited N. oceanica mutant library based on single-cell Raman spectrum. A Comparison of high-throughput RACS and traditional methods for screening carotenoid-overproducing genes. Carotenoid-overproducing mutants were screened by RACS with Raman intensity >99%. The plasmids harboring CRISPR/Cas system in the screened mutants were subsequently amplified by MDA. The gRNAs from the plasmids were then amplified by PCR and sequenced with NGS to identify changes in abundance after the sorting, which offers much higher efficiency yet lower costs than traditional methods. The average SCRS (top panels) and the Pearson correlation between carotenoid concentration and peak Raman intensity (bottom panels) are shown for β-carotene (B), violaxanthin (C), and zeaxanthin (D). Boxplots illustrate the distribution of peak Raman intensity at each carotenoid concentration. Data are presented as box plots showing the median (center line), interquartile range (box; 25th–75th percentiles), and whiskers extending to 1.5× IQR. The bands shaded in light color represent the 95% confidence level for the linear regression lines. For each linear regression, the significance of Pearson’s r was assessed using an unpaired two-tailed t-test (n = 6 concentration levels; r = 0.989 and p < 2.2e−16 for β-carotene, r = 0.996 and p < 2.2e−16 for violaxanthin, r = 0.996 and p < 2.2e−16 for zeaxanthin). E The average SCRS of the mutant library (LIB) and wild-type (WT) are presented, with the variation range (mean ± SD) of Raman intensity at each Raman shift depicted as shaded ribbons. Characteristic Raman peaks are highlighted. F The standard deviations of Raman intensity at each Raman shift are compared between LIB (n = 2651 Raman shifts) and WT (n = 2651 Raman shifts) (Wilcoxon test, p < 2.2e−16). Raman shift regions of 500–3150 cm−1 are considered for the comparison. G The Raman intensities of the v3 peak are compared between LIB (n = 4714 cells) and WT (n = 6231 cells) (Wilcoxon test, p < 2.2e−16). H The Raman intensities of the v3 peak are displayed for single cells derived from LIB and WT. For each sample, a random subsample of 4000 SCRS was selected for comparative analysis. Only Raman intensities exceeding 6000 are included in the comparison. I The lollipop charts that illustrate the level of enrichment for each mutated gene enriched in a RACS-sorted sample, with each sphere color-coded based on the corresponding gene’s pathway or function. Enriched genes shared by at least two replicates are shown. J The overlap of enriched genes among the three replicates. Significance was assessed using paired two-tailed Wilcoxon test in (F) and unpaired two-tailed Wilcoxon test in (G). ****: p < 0.0001. Exact p values, n numbers, mean and SEM values can be found in the Source Data file. A was created in BioRender. Liu, Y. (2026) https://BioRender.com/fh5sj9g. Carotenoids exhibit characteristic resonance Raman peaks near ~1520 cm−1, ~1156 cm−1, and ~1006 cm−1, whereas chlorophyll shows no peaks in these regions51,52, thus intracellular carotenoids can be detected without interference from chlorophyll signals in photosynthetic cells such as Chlamydomonas reinhardtii and Haematococcus pluvialis42,51,53,54. In N. oceanica, zeaxanthin, violaxanthin, and β-carotene are the three major carotenoids, collectively accounting for ~75% of the total carotenoids55. Raman spectra of a gradient of these individual purified carotenoids revealed the concentrations of β-carotene, violaxanthin, and zeaxanthin are linearly dependent (r = 0.989, 0.996, and 0.996, respectively) on normalized intensities at 1511 cm−1 (I1511 − I1800), 1522 cm−1 (I1522 − I1800), and 1517 cm−1 (I1517 − I1800; 1800 cm−1 used as baseline due to the absence of biological molecules vibrations from 1800 to 1840 cm−1; Fig. 5B–D). We then profiled the SCRS of the N. oceanica cells from the CRISPR/Cas mutant library (after five-day culture) via our previously established pDEP-DLD-RACS43 (Fig. 5A). The microalgal SCRS showed three most prominent bands at 1509 cm−1, 1154 cm−1 and 996 cm−1 (Fig. 5E), which agree with the carotenoid standards and support using these peaks for carotenoid quantification in N. oceanica. Among these peaks, 1509 cm−1 (C=C bond stretching vibrations in carotenoids) showed the highest intensity, thus the “I1509 − I1800 > x” criterion was applied to sort carotenoid-overproducing cells from the mutant library. We have demonstrated the high performance of pDEP-DLD-RACS in sorting yeast and Aurantiochytrium sp. cells43. Here, we validated its capability in sorting N. oceanica cells. We assessed the stability in SCRS acquisition, which is crucial for high-accuracy sorting. SCRS of N. oceanica cells were acquired in pDEP-DLD-RACS, and the peak position around 1509 cm−1 (defined as the position with the highest intensity) was evaluated. Among the individual cells, the peak was found at from 1500 to 1520 cm−1, due to the inherent cell-to-cell heterogeneity in dynamic synthesis of metabolites (e.g., pigments). Notably, the distribution pattern at the single-cell level and averaged peak position at the population level were both stable for at least 5 h (Supplementary Fig. 3A). Moreover, the intensity at 1509 cm−1 showed consistent distribution and no significant difference among each hour (Supplementary Fig. 3B), suggesting temporal stability in SCRS acquisition of N. oceanica cells. For sorting, a shorter sorting-trigger time means higher throughput, while a longer sorting-trigger time ensures higher success-sorting rate. To balance these two concerns, the trigger time for sorting of 100 ms was applied, achieving a success-sorting rate of >82% (Supplementary Fig. 3C). Finally, to assess the sorting stability and accuracy, the N. oceanica cells (featuring strong resonance Raman signal derived from pigment and red autofluorescence derived from chlorophyll) and S. cerevisiae cells (featuring weak nonresonance Raman signal and non-autofluorescence) were 1:1 mixed and underwent pDEP-DLD-RACS for 5 h to sort N. oceanica cells based on the sorting criteria of “I1509 − I1800 >100” (Supplementary Fig. 3D). To facilitate the counting of both target (N. oceanica) and nontarget (S. cerevisiae) cells, the cells in “Collect” and “Waste” were trapped and concentrated in the chip via pDEP force (Supplementary Fig. 3E) and then fluorescence imaged (Supplementary Fig. 3F), demonstrating a stable 5-h sorting (Supplementary Fig. 3G) with a sorting accuracy of 95.97% and yield of 73.70% (Supplementary Fig. 3F and H). Here, the mis-sorted events were mainly caused by the few multi-trapped (i.e., a nontarget cell was trapped with a target cell and sorted simultaneously) and untrapped nontarget cells which would flow into collect channel randomly. The missed-sorted cells were mainly caused by the unsuccess sorting rate of ~18% under 100 ms sorting duration (Supplementary Fig. 3C), few untrapped target cells which would flow into waste channel randomly, and few misaligned cells, trapped on the electrodes but not at the laser point, which would also flow into waste channel. To sort the N. oceanica genome-wide mutant library, over 1000 cells were first measured in a single pDEP-DLD-RACS run, while the value of I1509 − I1800 in SCRS that separates the top 1% of cells in intracellular carotenoid content from the remaining 99% of cells was chosen as the “x”. Given that direct detection of target sites was challenging due to natural variations and sequencing errors, we relied on gRNAs as markers for identifying mutated genes. Considering a throughput of ~600 cells min−1 while retaining only the top 1% of cells, the sorted cells during five hours of screening were subjected to DNA amplification by Multiple Displacement Amplification (MDA) with primers targeting the plasmid within the CRISPR/Cas system (Supplementary Table 2; Methods), then the MDA products were amplified by PCR and sequenced by NGS to analyze the gRNA abundance. In addition, the gRNA sequences in the unsorted mutant library were amplified by PCR and sequenced by NGS as a control. The screening experiment was divided into two steps. First, to determine whether there were mutant strains with the target phenotype in the mutant library, over 10,000 microalgal cells from the library and a wild-type sample (WT) were subjected to pDEP-DLD-RACS. The SCRS obtained from the mutant library exhibited significantly higher heterogeneity than the WT, as evidenced by the broader variations in Raman intensities across almost all Raman shifts (Fig. 5E). Indeed, the standard deviations of Raman intensities at each Raman shift (among Raman shifts of 500–3150 cm−1) were on average 1.69-fold higher in the mutant library than in the WT (Wilcoxon test, p < 0.0001; Fig. 5F). These variations encompass, but are not limited to, those characteristic carotenoid Raman peaks (v1: ~996 cm−1; v2: ~1154 cm−1; v3: ~1509 cm−1). For example, the v3 peak intensity in the mutant library was on average 1.61-fold higher than the WT (Wilcoxon test, p < 0.0001; Fig. 5G), suggesting the increase in carotenoid content in some of the cells from the mutant library. Notably, following subsampling of 4000 cells for both the mutant library and the WT sample, the screening identified 26 mutants with intensity exceeding 6000 at 1509 cm−1, whereas only four cells in the WT sample exhibited similarly high Raman intensities at the v3 peak (Fig. 5H). These results suggested the presence of many carotenoid-overproducing mutants in the library, and supported the suitability of pDEP-DLD-RACS for screening such metabolic phenotypes in microalgae. Second, we sorted carotenoid-overproducing mutants through three independent pDEP-DLD-RACS runs based on v3 peak intensities. Based on the fold-change of gRNA abundances before- and post-sorting (Methods), 40, 44, and 56 genes were found enriched respectively (Supplementary Data 4). This number, however, does not directly reflect the number of isolated mutant lines, as a single gRNA can give rise to multiple independent mutant lines carrying distinct genotypes that result in similar phenotypes. These mutated genes were related to over 20 pathways including carotenoid biosynthesis (e. g., noVDEs) and photosynthesis (Fig. 5I; Supplementary Data 4), suggesting a direct or pleiotropic link between carotenoid biosynthesis and other metabolic pathways. Moreover, among the collectively 109 unique mutated genes (Supplementary Data 4), 33 were of unknown function, nine of which were consistently identified across all three independent runs, while 13 were detected in two of the three runs (Fig. 5I, J). Most of the shared genes seem not directly linked to carotenoid biosynthesis, with eight of them classified as unknown genes, i.e., suggesting yet-to-appreciate mechanisms. Among the genes shared by at least two of the three pDEP-DLD-RACS runs are two violaxanthin de-epoxidases (VDEs), which may be responsible for the de-epoxidation of violaxanthin within the thylakoid membrane in the carotenoid biosynthesis pathway. Specifically, in higher plants, VDE is activated under high light and converts violaxanthin to antheraxanthin and zeaxanthin, resulting in a decrease in violaxanthin content56; thus VDE knock-out may result in the reduction of violaxanthin (and thus of carotenoids). In N. oceanica, violaxanthin content can be regulated by carotenoid biosynthetic genes such as carotenoid isomerase (from Phaeodactylum tricornutum57) and zeaxanthin epoxidase (ZEP)58, yet functional roles of the noVDEs are still unclear. Moreover, many of the genes represent entirely unexpected links to carotenoid metabolism. For instance, NO02G03540, the proteasome assembly chaperone 4 (PAC4), was among those with the highest enrichment (with log-fold changes of 16.8, 6.8, and 16.4 in the three replicates; Fig. 5I). This is a protein that facilitates biogenesis of the 26S proteasome59, a large protein complex responsible for degrading ubiquitin-dependent proteins in eukaryotic cells60. PAC4 and other proteasome assembly chaperones are known as potential targets for inhibiting accurate proteasome assembly in cancer cells61,62. In plants, chloroplast-targeted preproteins may undergo ubiquitin-dependent degradation by the 26S proteasome to prevent their accumulation in the cytosol and help shaping the chloroplast proteome63. However, no evidence has ever implicated PAC4 in pigment metabolism. Collectively, direct phenotypic screening of the N. oceanica genome-wide targeted mutant library by pDEP-DLD-RACS rapidly uncovered many unexpected genes associated with elevation in single-cell carotenoid content. To dissect the underlying mechanisms, we selected VDEs and PAC4 as targets for validating in vivo activities via metabolic engineering. VDEs contribute to accumulation of carotenoids in N. oceanica To probe the in vivo function of noVDEs, we started from the time-series transcriptomes of N. oceanica under six conditions, including: very low carbon levels (VLC, 100 ppm CO2), high carbon levels (HC, 5% CO2), nitrogen-depleted (N-), nitrogen-replete (N+), exposure to red or blue light (Supplementary Fig. 4A; Supplementary Table 4; https://nandesyn.single-cell.cn)64,65. Under VLC, noVDE1-5 transcript levels (noVDE1: NO24G00840, noVDE2: NO20G02350, noVDE3: NO11G00920, noVDE4: NO03G00410, noVDE5: NO08G01830) were 30.9%, 7.5%, 33.7%, 62.3%, and 136% higher than HC, respectively. This may be attributed to insufficient carbon supply for the dark reactions of photosynthesis under VLC. The excess absorption of light by photosynthetic pigments, relative to the cellular metabolic capacity, drives the overreduction of the photosynthetic electron transport chain, resulting in the generation of reactive oxygen species (ROS)66. Thus, the activation of noVDEs can facilitate the conversion of violaxanthin to zeaxanthin, which enhances non-photochemical quenching (NPQ) and aids in the scavenging of chlorophyll triplets and ROS67. These results supported the roles of noVDEs in carotenoid biosynthesis in N. oceanica. In addition, the putative NoVDE protein sequences were compared with VDE sequences from higher plants. Multiple sequence alignment (Supplementary Fig. 5) revealed high similarity within the catalytic domain between NoVDEs and the other VDEs analyzed. The N-terminal domain of NoVDEs is enriched in cysteine residues, a feature conserved across VDE sequences from all organisms examined. Within the lipocalin domain, the alignment also demonstrated conservation in NoVDEs of key residues essential for catalytic activity (e.g., Asp177 in NoVDE1 and NoVDE2, and Tyr198 in all NoVDEs), as previously reported for higher plants68. However, notable variations were observed in regions associated with structural organization and pH sensitivity69,70. Collectively, NoVDEs in N. oceanica may either carry a distinct enzymatic activity or operate via a mechanism distinct from their higher plant counterparts. To avoid compensatory effects of genes with similar function, a tRNA-based CRISPR/Cas system was employed to simultaneously mutate all predicted noVDEs using a single plasmid, resulting in two mutants, termed novde-1 and novde-2 (Fig. 6A), each of which carries frameshift mutations in all five noVDEs (except noVDE4 in novde-1 which carries a 18-base deletion). These mutants showed significant alterations in photosynthetic performance, plus in the levels of the violaxanthin and β-carotene (Fig. 6B–D). Under low light conditions (LL), the Fv/Fm ratios for novde-1 and novde-2 increased by 4.0% and 3.4%, respectively, compared to the WT (Fig. 6B). In addition, the violaxanthin levels were increased by 30.3% and 24.8% (Fig. 6C), while β-carotene levels increased by 29.8% and 30.9% (Fig. 6D), respectively. Furthermore, the total content of violaxanthin, β-carotene and zeaxanthin increased by 16.4% and 11.4%, respectively (Fig. 6F). Under high light conditions (HL), compared to the WT, the Fv/Fm ratios in novde-1 and novde-2 increased by 8.3% and 8.1%, respectively (Fig. 6B), while the β-carotene levels increased by 35.6% and 34.2%, respectively (Fig. 6D). Moreover, the changes were not accompanied by significant alterations in growth rate, biomass, chlorophyll a, or zeaxanthin content, with the exception of novde-2, which showed a 7.7% decrease in growth rate and a 7.8% decrease in biomass under LL and a 15.0% decrease in zeaxanthin content under HL (Fig. 6E; Supplementary Fig. 6A–C). The elevated violaxanthin levels observed under low light conditions in mutants harboring five noVDE knockouts confirmed the functional role of noVDE enzymes in the conversion of violaxanthin to zeaxanthin. This finding explains the accumulation of violaxanthin and the overall rise in carotenoid content observed in the quintuple noVDE-knockout mutants. This outcome contrasts with the earlier report by Perin et al.66 which reported no significant carotenoid accumulation in N. oceanica mutants harboring only a single noVDE knockout (Supplementary Note 3).Fig. 6The genotype and phenotype of the noVDE-knockout N. oceanica mutants.A Mutated loci of the five noVDEs. B The Fv/Fm ratio in the mutants and WT. The contents of violaxanthin (C), β-carotene (D) and zeaxanthin (E) in the mutants and WT are shown. F The contents of violaxanthin, β-carotene and zeaxanthin in the mutants and WT. Data are represented as mean ± SD (n = 3 biological replicates). Significant changes (*p ≤ 0.05; **p ≤ 0.01; one-sided Student’s t test) versus the WT. Source data for this figure is available in the Source Data file. The genotype and phenotype of the noVDE-knockout N. oceanica mutants. A Mutated loci of the five noVDEs. B The Fv/Fm ratio in the mutants and WT. The contents of violaxanthin (C), β-carotene (D) and zeaxanthin (E) in the mutants and WT are shown. F The contents of violaxanthin, β-carotene and zeaxanthin in the mutants and WT. Data are represented as mean ± SD (n = 3 biological replicates). Significant changes (*p ≤ 0.05; **p ≤ 0.01; one-sided Student’s t test) versus the WT. Source data for this figure is available in the Source Data file. Under high light conditions (300 μmol photons m−2 s−1), increases in Fv/Fm and β-carotene content were observed, while violaxanthin and zeaxanthin levels remained largely unchanged. As a conserved structural component of the photosynthetic core in both photosystems I and II across all oxygenic photosynthetic organisms71,72, β-carotene plays an essential role in photosynthesis, thus it is possible that the rise in β-carotene contributes to the elevated Fv/Fm ratios in N. oceanica. On the other hand, the conversion from violaxanthin to zeaxanthin is quite rapid (typically occurring within minutes under high light66,67), and our measurements, which were conducted after three days of cultivation, are unlikely to capture such events. We propose that violaxanthin, zeaxanthin, and β-carotene had reached a new dynamic equilibrium by this time point, leading to similar steady-state levels of violaxanthin and zeaxanthin, as well as similar total content of these three carotenoids in both the mutant and wild-type strains. In addition, the observed decline in carotenoid content under HL relative to LL points to a regulatory framework for carotenoid metabolism that is more extensive than the xanthophyll cycle alone56,73 and remains to be fully elucidated. Discovery of a role of PAC4 that enhances carotenoid synthesis in N. oceanica NO02G03540 encodes noPAC4, putatively a proteasome assembly chaperone involved in assembly of the 26S proteasome. noPAC4 shows 37% sequence identity to human PAC4 (NCBI Reference Sequence: NP_001129222.1), yet no significant similarity to yeast PBA4 (GenBank: DAA11291.1). Given the lack of evidence linking noPAC4 to carotenoid biosynthesis in any organisms, we analyzed the rich N. oceanica transcriptomes collected under diverse conditions in the NanDeSyn database8,9, to probe mechanisms of the high-carotenoid phenotype in noPAC4-edited mutants (Methods). Under HC and VLC conditions, at 24 h, noPAC4 expression was 51% lower under VLC than HC (Supplementary Fig. 4A), while ~75% of the 52 genes underlying 26S proteasome were downregulated (Supplementary Fig. 4B). Notably, the noPAC4 transcript showed a strong positive correlation (median Pearson’s r = 0.88) with those of 26S proteasome genes over 3, 6, 12, and 24 h (Supplementary Fig. 4B), suggesting functional connections between noPAC4 regulation, 26S proteasome activity and carotenoid biosynthesis. We propose that enhanced protein biosynthesis under HC conditions raises 26S proteasome activity, which then alters carotenoid biosynthesis. Phytoene synthase (PSY) catalyzes carotenogenesis by condensing two molecules of geranylgeranyl diphosphate (GGPP) to produce the C40 hydrocarbon 15-cis-phytoene. Acting as the first committed step in carotenogenesis, PSY directs the metabolic flux into the carotenoid biosynthetic pathway. The precursor protein of PSY can be ubiquitinated by the plastid protein-sensing ring E3 ligase 1 (PPSR1) and subsequently degraded by the 26S proteasome74. We hypothesize that the knockout of noPAC4 decreases the efficacy of the 26S proteasome, resulting in an increased transport of precursor proteins of PSY into the chloroplast, which in turn leads to enhanced carotenoid biosynthesis. To test these hypotheses, we independently knocked out noPAC4 (Fig. 7A). The mutants of nopac4-1 and nopac4-2 exhibited significant changes in growth rate, photosynthesis, and carotenoid content (Fig. 7B–F; Supplementary Fig. 6D, E). Specifically, under low light (50 μmol photons m−2 s−1) conditions, growth rate increased by 4.3% and 8.6%, respectively, compared to WT, while biomass decreased by 13.5% and 9.9%, respectively (Supplementary Fig. 6D, E). The Fv/Fm ratios for nopac4-1 and nopac4-2 increased by 5.2% and 6.1%, respectively (Fig. 7B). Moreover, violaxanthin levels increased by 11.6% and 13.3% (Fig. 7C); β-carotene levels increased by 18.7% and 30.2% (Fig. 7D); and zeaxanthin levels increased by 16.6% and 22.1% (Fig. 7E), respectively. Furthermore, the total content of violaxanthin, β-carotene and zeaxanthin increased by 14.9% and 19.9%, respectively (Fig. 7F). However, under high light (300 μmol photons m−2 s−1), no such difference was observed between WT and mutants, except for the 8.4% and 7.6% reduction in zeaxanthin levels in the mutants and 4.5% Fv/Fm-increase in nopac4-2 (Fig. 7B–F). The chlorophyll a content decreased by 8.7% and 8.3%, respectively (Supplementary Fig. 6F). The observed increase in carotenoid content, coupled with stable or decreased levels of chlorophyll a, suggested that the SCRS-based mutant screening relied specifically on carotenoid accumulation, independent of chlorophyll fluctuations. Therefore, the substantial increase in carotenoid levels in noPAC4-knockout mutants under low light is consistent with the single-cell Raman-based phenotypes as identified and sorted. Moreover, the distinct physiological responses to noPAC4 knock-out from noVDEs knock-outs suggest pDEP-DLD-RACS can identify mutants with very different mechanisms that underlie a target phenotype.Fig. 7The genotype, phenotype and transcriptomic dynamics of the noPAC4-knockout N. oceanica mutants.A The mutated loci of noPAC4. B The Fv/Fm ratio for the mutants and WT. The contents of violaxanthin (C), β-carotene (D) and zeaxanthin (E) in mutants and WT are shown. F The collective content of violaxanthin, β-carotene and zeaxanthin in the mutants and WT. G The average transcriptional levels (in TPM) of selected genes for both WT and mutant samples. In the heatmap, cells are color-coded based on the fold change in gene expression for the corresponding genes, with red indicating up-regulation and green denoting down-regulation. Enrichment plots for the upregulated proteasome pathway (ko03050) are presented for the mutants nopac4-1 (H) and nopac4-2 (I). Differentially enriched pathways between WT and mutants, which were identified by gene set enrichment analysis against the KEGG database. Relative positions of the genes are represented by the straight lines (line plot) in each graph, with lines clustered to the left indicating higher-ranked genes with significant fold changes in expression in the ranked list. Enrichment Score (ES) refers to the extent to which a gene set is overrepresented at the top or bottom of a ranked list of genes. The Normalized Enrichment Score (NES) adjusts for variations in ES among gene sets that are resulted from differences in gene set sizes. The P-value denotes the statistical significance of the enrichment score for an individual gene set, which is derived from the null distribution. The adjusted P-value corrects for multiple hypothesis testing, thus facilitates more accurate comparison among the various tested gene sets. Data are represented as mean ± SD (n = 3 biological replicates). Significant changes (*p ≤ 0.05; **p ≤ 0.01; one-sided Student’s t test) versus the WT. Source data for this figure is available in the Source Data file. The genotype, phenotype and transcriptomic dynamics of the noPAC4-knockout N. oceanica mutants. A The mutated loci of noPAC4. B The Fv/Fm ratio for the mutants and WT. The contents of violaxanthin (C), β-carotene (D) and zeaxanthin (E) in mutants and WT are shown. F The collective content of violaxanthin, β-carotene and zeaxanthin in the mutants and WT. G The average transcriptional levels (in TPM) of selected genes for both WT and mutant samples. In the heatmap, cells are color-coded based on the fold change in gene expression for the corresponding genes, with red indicating up-regulation and green denoting down-regulation. Enrichment plots for the upregulated proteasome pathway (ko03050) are presented for the mutants nopac4-1 (H) and nopac4-2 (I). Differentially enriched pathways between WT and mutants, which were identified by gene set enrichment analysis against the KEGG database. Relative positions of the genes are represented by the straight lines (line plot) in each graph, with lines clustered to the left indicating higher-ranked genes with significant fold changes in expression in the ranked list. Enrichment Score (ES) refers to the extent to which a gene set is overrepresented at the top or bottom of a ranked list of genes. The Normalized Enrichment Score (NES) adjusts for variations in ES among gene sets that are resulted from differences in gene set sizes. The P-value denotes the statistical significance of the enrichment score for an individual gene set, which is derived from the null distribution. The adjusted P-value corrects for multiple hypothesis testing, thus facilitates more accurate comparison among the various tested gene sets. Data are represented as mean ± SD (n = 3 biological replicates). Significant changes (*p ≤ 0.05; **p ≤ 0.01; one-sided Student’s t test) versus the WT. Source data for this figure is available in the Source Data file. Furthermore, we profiled the transcriptomes of noPAC4-knockout mutants and WT under low light via RNA-Seq (in triplicate; Supplementary Fig. 7). In the mutants (Supplementary Data 5), 73 differentially expressed genes (DEGs) were identified compared to the WT. For example, a histone deacetylase (HDAC, NO11G01600) was much lower in transcript level than the WT, by 43.0% and 54.8% in nopac4-1 and nopac4-2, respectively (Fig. 7G). HDAC interference accelerates the ripening process and increases β-carotene accumulation in tomato75, raising the possibility that the reduced expression of HDAC may also promote carotenoid biosynthesis in N. oceanica. In addition, transcripts of retinol dehydrogenase (RDH, NO04G04240) were 64.5% and 64.4% lower than that of WT, respectively (Fig. 7G). RDH β-carotene 15,15’-dioxygenase (BCO) cleaves β-carotene into retinal, which is then reduced to retinol by the retinol dehydrogenase76. The decreased expression of RDH may reduce the cleavage of β-carotene, resulting in increased carotenoid accumulation. In addition, the expression of proteasome-associated genes was up-regulated in both nopac4-1 and nopac4-2, as indicated by the KEGG gene set enrichment analysis (Fig. 7H, I; Supplementary Fig. 8; Methods), suggesting noPAC4-mediated regulation of 26S proteasome activity. It is possible that compensatory biosynthesis of additional 26S proteasome subunits has taken place to partially restore the attenuated 26S proteasome activity caused by noPAC4 knockout. Collectively, these results suggested the attenuation of 26S proteasome function by noPAC4 knockout, which may enhance the import of precursor proteins (e.g., PSY) into the chloroplast, thereby stimulating carotenoid biosynthesis. Discussion In N. oceanica, carotenoid biosynthesis starts with the condensation of isopentenyl diphosphate (IPP) and dimethylallyl diphosphate (DMAPP), derived from the methylerythritol phosphate (MEP) pathway, to form the 20-carbon geranylgeranyl diphosphate (GGPP). This reaction is catalyzed by GGPP synthase. Subsequently, GGPP is converted to the 40-carbon phytoene by the action of phytoene synthase. The conversion of phytoene to β-carotene involves a series of enzymatic steps, including desaturation, isomerization, and cyclization, mediated by phytoene desaturase (PDS), ζ-carotene isomerase, ζ-carotene desaturase, carotenoid isomerase, and lycopene β-cyclase (LCYB). Following the formation of β-carotene, hydroxylation reactions catalyzed by heme-containing cytochrome P450 enzymes lead to the synthesis of zeaxanthin. Zeaxanthin can be further modified to violaxanthin by zeaxanthin epoxidase (ZEP). Additionally, violaxanthin can be transformed into neoxanthin by violaxanthin de-epoxidase-like (VDL) and further metabolized to vaucheriaxanthin through enzymatic steps that remain to be fully elucidated. To date, only a few enzymes in the carotenoid biosynthesis pathway of N. oceanica have been characterized in vivo, including PDS, LCYB, ZEP, and VDL55,58,77,78. Here, by establishing a platform that comprises a genome-wide targeted knockout library and high-throughput RACS, we demonstrated an efficient strategy to mine functional genes by efficiently tracing a metabolic trait to its genotypes. From the CRISPR/Cas-generated library of 3567 N. oceanica mutants derived from 2397 effective guide RNAs, label-free sorting for high single-cell carotenoid contents via SCRS rapidly unraveled its underlying genes, including both expected (e.g., noVDEs) and unexpected (e.g., noPAC4) ones, which were then validated. These allowed us to propose a multi-level regulation model that provides insights into carotenoid biosynthesis in this microalga (Fig. 8). (i) At the transcriptional level, noPAC4 may reduce the expression of HDAC, thereby influencing transcript level of carotenoid-synthetic genes through epigenetic mechanisms. (ii) At the post-translational level, noPAC4 regulates the 26S proteasome, which in turn maintains chloroplast proteostasis63 and influences degradation of carotenogenic enzymes such as PSY74, thereby affecting carotenoid accumulation. (iii) Carotenoid catabolic enzymes such as noVDEs convert the carotenoid violaxanthin into antheraxanthin and zeaxanthin, as evidenced by the noVDE-knockout induced 43.3% increase in violaxanthin content. Moreover, reducing retinol synthesis by inhibiting RDH can also raise carotenoid level.Fig. 8An expanded, multi-level regulation model for carotenoid biosynthesis in N. oceanica.The regulation of carotenoid biosynthesis involves the transcriptomic level through epigenetics (such as HDAC), and the post-translational level via the 26S proteasome. It is also regulated by reducing carotenoid degradation via noVDEs or RDH. BHY β-carotene 3,3’-hydroxylase, HDAC histone deacetylase, GGPP geranylgeranyl diphosphate, PSY phytoene synthase, RDH retinol dehydrogenase, ZEP zeaxanthin epoxidase, VDE violaxanthin de-epoxidase. This figure was created in BioRender. Liu, Y. (2026) https://BioRender.com/m3fj5pr. An expanded, multi-level regulation model for carotenoid biosynthesis in N. oceanica. The regulation of carotenoid biosynthesis involves the transcriptomic level through epigenetics (such as HDAC), and the post-translational level via the 26S proteasome. It is also regulated by reducing carotenoid degradation via noVDEs or RDH. BHY β-carotene 3,3’-hydroxylase, HDAC histone deacetylase, GGPP geranylgeranyl diphosphate, PSY phytoene synthase, RDH retinol dehydrogenase, ZEP zeaxanthin epoxidase, VDE violaxanthin de-epoxidase. This figure was created in BioRender. Liu, Y. (2026) https://BioRender.com/m3fj5pr. Studies of microalgal carotenoid metabolism have mainly focused on key enzymes such as PDS, LCYB, ZEP, and VDL58,79. Here we identified noVDEs as an important player, and showed that knockout of five noVDE genes in a single transformation resulted in 43.3% increase of violaxanthin content in N. oceanica, which provides an effective strategy for violaxanthin overproduction. Under low light (50 μmol photons m−2 s−1), noVDEs knockout prevents the conversion of violaxanthin to zeaxanthin, whereas the epoxidation of zeaxanthin back to violaxanthin remains active, resulting in the accumulation of violaxanthin. Given the functional importance of zeaxanthin in photosynthesis for the light absorption and photoprotection67,72,80,81, de novo synthesis via β-carotene can help to maintain zeaxanthin content in the mutant at levels similar to those in the WT. Under high light conditions (300 μmol photons m−2 s−1), increases in Fv/Fm and β-carotene content were observed, while violaxanthin and zeaxanthin levels remained largely unchanged. As a conserved structural component of the photosynthetic core in both photosystems I and II across all oxygenic photosynthetic organisms71,72, β-carotene plays an essential role in photosynthesis, thus it is possible that the rise in β-carotene contributes to the elevated Fv/Fm ratios in N. oceanica. On the other hand, the conversion from violaxanthin to zeaxanthin is quite rapid (typically occurring within minutes under high light66,67), and our measurements, which were conducted after three days of cultivation, are unlikely to capture such events. We propose that violaxanthin, zeaxanthin, and β-carotene had reached a new dynamic equilibrium by this time point, leading to similar steady-state levels of violaxanthin and zeaxanthin, as well as similar total content of these carotenoids (β-carotene, violaxanthin, and zeaxanthin), in both mutant and wild-type strains. This observation is largely consistent with the phenotypes of a single-noVDE-gene (NO24G00840) knockout mutant reported by Perin et al.66. As for regulatory mechanisms, reactive oxygen species (ROS), nitric oxide (NO), and calcium ions (Ca2+) were found as signaling molecules under salinity or light stresses42,82. In contrast, we proposed a post-translational mechanism that modulates carotenoid levels involving proteasome assembly control through PAC. These findings suggested a strategy for engineering carotenoid production, while extended the impact of the 26S proteasome-mediated protein degradation from protein misfolding-related disorders such as neurodegenerative diseases83 to the unexpected realms of plant pigment synthesis and photosynthesis. These discoveries underscore the power of screening a genome-wide targeted knockout mutant library, which is not dependent on a priori knowledge or hypotheses about the underlying genotype. Notably, such mutant libraries have not been reported for any microalgae before, due to their relatively low transformation or genome-editing efficiency84. To tackle this challenge, we devised a tRNA-based CRISPR/Cas system for N. oceanica that simplifies the gRNA design and increases the accuracy of ligation for plasmid library construction. Compared to the hammerhead-based CRISPR/Cas system48, the tRNA-based system eliminates the complex design of the hammerhead structure as well as the spontaneous formation of the stem loop structure, thereby streamlining plasmid construction. This library offers key advantages over those generated by random insertion, chemical mutagens or physical mutagenesis85. (i) It enabled rapid and low-cost identification of mutation sites by employing gRNAs as markers to analyze target site mutations. (ii) It elevates the proportion of exon mutations: all gRNAs were targeted to such coding regions, in contrast to that only 40% of genes were of exon mutation in a random insertion based C. reinhardtii mutant library15. (iii) Its genome-wide targeted knockout feature allowed unambiguous and efficient mining of genotypes that underlie a target phenotype by gRNA enrichment, which however is particularly challenging for random mutagenesis libraries due to the presence of multiple mutation sites in a mutant genome21. Notably, the CRISPR/Cas9 system employed here operates via site-specific double-strand breaks. While Cas9’s tolerance for up to three sgRNA mismatches can theoretically cause off-target effects86,87, actual off-target mutations are infrequent in CRISPR/Cas9-edited plants (grapevine88) and animals (mice89 and human90). Furthermore, the gRNA design in this study followed stringent specificity criteria, requiring the absence of potential off-target sites with ≤3 nucleotide mismatches. Thus, the risk of off-target effects in this N. oceanica genome-wide editing library are likely low. Notably, at the colony level, although those with no detectable mutation (38.07%) and multiple mutations (12.43%) among the totally 5760 colonies are found, they do not pose a significant challenge to our screening strategy, because our Raman-based phenotype sorting takes place at the single-cell level from the resuspended cellular mixtures of all colonies (Supplementary Note 4). In screening such large mutant libraries for intracellular metabolites, throughput has long been the primary bottleneck, because the classical yet time-consuming, tedious and costly strategy of agar plating, colony picking, culture for sufficient clonal biomass, biomass lysis and profiling cell extracts for metabolite content still dominates both large-scale bio-foundry facilities and small labs. This process is particularly painful for autotrophic eukaryotic microalgae due to their typically slower growth than bacterial and fungal chasses15,17. Moreover, many mutants may fail to proliferate during the culture, leading to the loss of crucial genes19,22. FACS can greatly accelerate this process by sorting individual cells16,yet its application is limited by the narrow scope of metabolites that can be labeled, the interference of autofluorescence, and the often very demanding process of introducing the label either biochemically or genetically. For example, as the basic structural units of photosynthetic apparatus that harvest light for photosynthesis, quench the energy excess under high light stress and prevent the formation of highly reactive singlet oxygen91, carotenoids cannot be directly screened by FACS due to the interference of chlorophyll fluorescence. In fact, current quantification methods for carotenoids including near-infrared reflectance spectroscopy (NIRS), visible spectrometry, laser photoacoustic spectroscopy (PAS), UV-visible spectrometry, high-performance liquid chromatography (HPLC) and mass spectrometry (MS) all require carotenoid extraction from a significant amount of bulk biomass92. To tackle this challenge, we established a microfluidics-based, label-free, non-destructive, high-throughput approach for quantifying and then sorting of single-cell carotenoid contents directly from mutant mixtures, based on RACS. We showed that a spontaneous SCRS can accurately quantify contents of total carotenoids at single-cell resolution for a small-size autotrophic microalga such as N. oceanica with our previously developed pDEP-DLD-RACS43. Moreover, such contents were sorted at a high throughput of ~600 cells min−1, so the top-performing 1% of cells can be genotyped for enriched gRNA sequences to trace the phenotype to underlying genotypes. For instance, noPAC4 mutants with increased violaxanthin, zeaxanthin, and β-carotene levels were obtained in just five hours of pDEP-DLD-RACS run. By label-freely and non-destructively sorting individual cells for single-cell carotenoid content, the Raman cytometry avoids the tedious fluorescence labeling, saves the arduous time and efforts for agar plating, colony picking and biomass extraction, and minimizes the potential loss of slow-growing mutants during the various stages of cultivation. Besides carotenoids, a spontaneous SCRS can depict other high-value pigments (e.g., astaxanthin), lipids (e.g., TAG and DHA) and polysaccharides (e.g., starch), as well as their interconversion in the cell26,41–43. Moreover, key traits of cell factories, such as viability, vitality, assimilation rate of specific substrates, susceptibility and response to stresses can be profiled by inserting a stable-isotope probing step before SCRS acquisition and sorting26,93. Therefore, one direction is to exploit such rich and diverse single-cell metabolic phenome encoded in a SCRS for single-trait or multi-trait gene mining at a systematic fashion, from the genome-wide N. oceanica mutant library here or even directly from an environmental sample (e.g., sorting gut microbiota based on a flow-mode Raman tweezer platform35,36). On the other hand, a key limitation of spontaneous Raman-based sorting lies in spectral overlap among chemically related metabolites, which restricts the ability to resolve individual molecular species within a metabolite class. For carotenoids, closely related compounds often exhibit highly overlapping Raman features, making it difficult to distinguish specific carotenoid species within a single cell. In this study, we therefore focused on total carotenoid content rather than individual carotenoid identities. Importantly, for CRISPR-based genetic screens, robust and quantitative phenotypic ranking across large mutant populations is more critical than fine-grained chemical specificity. Thus, while SCRS-based sorting is not optimal for resolving carotenoid composition, it provides a sufficiently sensitive and scalable phenotype for discovering genes that regulate overall carotenoid accumulation. To deconvolute SCRS, a RACS-coupled mass spectrometry workflow that supports machine learning of SCRS and the matching population-level metabolomes can be established. Notably, the randomness of gene expression in individual cells can lead to change in metabolic state of the cells94, therefore, to reduce the likelihood of sorting for false positive cells due to such effects, additional strategies such as taking the intersection among multiple carefully-selected time points or culture conditions (in addition to the multiple rounds of sorting from the same sample) will be developed. Moreover, the potential bias introduced by the post-RACS multiple MDA reaction can distort or mask the identity of mutations that underlie the Raman-sorted phenotypes, thus MDA enzymes and reaction conditions that reduce some bias will be adopted and optimized95. Furthermore, to fully explore the power of this single-cell phenotyping and sorting approach, re-transforming the library into selected genetically engineered strains presents a promising strategy for interrogating pathway interactions as well as the combinatorial impact of mutations on the phenotype. Notably, in addition to this work, technological variants of RACS such as coherent RACS28, RAMS37,38, and RADS40 have been developed and shown promises in sorting microalgae, cyanobacteria, and yeast cells with high efficiency, which suggests the broad applications of RACS for many chassis of microbial cell factories. In summary, by label-free sorting genome-wide targeted mutant libraries for single-cell metabolic features, the forward-genetics approach established in this work would greatly accelerate function-based discovery of genes and pathways, in not just microalgae but bacterial, fungal, plant and mammalian cells. Methods Strains and growth conditions The Nannochloropsis oceanica strains were cultivated in modified f/2 liquid medium65, which contains 35 g L−1 sea salt, 1000 mg L−1 NaNO3, 66.6 mg L−1 NaH2PO4•H2O, 3.65 mg L−1 FeCl3•6H2O, 4.37 mg L−1 Na2EDTA•2H2O, 0.0196 mg L−1 CuSO4•5H2O, 0.0126 mg L−1 Na2MoO4•2H2O, 0.044 mg L−1 ZnSO4•7H2O, 0.0109 mg L−1 CoCl2•6H2O, 0.036 mg L−1 MnCl2•4H2O, 5 mg L−1 VB12, 5 mg L−1 biotin, and 0.1 mg L−1 thiamine HCl. Strains were cultured under low light (50 μmol photons m−2 s−1) at 25 °C in flasks shaken at 200 rpm. For high light conditions (300 μmol photons m−2 s−1), the cells were harvested after a 5-day culture period by centrifugation (3500 × g for 5 min), and then resuspended in fresh f/2 medium to achieve a cell density of OD750 = 1 and then cultured for three days. Cell growth was determined based on biomass dry weight and optical density (OD750). Genetic transformation of the wild-type N. oceanica The tRNA-based CRISPR/Cas system was derived from the plasmid pNSA-ARS-CRISPR-BlastR (Addgene plasmid # 10100948), which can be stably maintained as circular DNA in cells under selective pressure. The sequence of this base vector is exactly identical to that of pNSA-ARS-CRISPR-BlastR, except that for the purpose of gene knockout construction, the HH site was replaced with a glycine tRNA sequence (Supplementary Table 2; Supplementary Fig. 1A, C; Source data). To construct the tRNA-based CRISPR/Cas system, the tRNA sequence was amplified with primers (Supplementary Table 2) and ligated to the BspQI digested plasmid ARS-CRISPR-BlastR by In-Fusion. CRISPR/Cas plasmids were introduced into N. oceanica cells as circular DNA by electroporation, as described96. Briefly, wild-type N. oceanica cells were cultured in 100 mL to the logarithmic growth phase and then harvested by centrifugation at 5000 × g for 5 min at 4 °C. The cells underwent two rounds of wash with 25 mL of 375 mM sorbitol at 4 °C. For each transformation, 200 μL of cells were combined with 2 μg of the cassette in an electroporation cuvette with a 2 mm gap before electroporation. The electroporation was carried out using a BTX ECM 630 system with a 1100 V cm−1 field strength, 50 μF capacitance, and 600 Ohm shunt resistance. The cells were promptly mixed with 5 mL of f/2 medium and incubated under low light conditions at 120 rpm for 48 h. Subsequently, the cells were plated on f/2 agar plates supplemented with 5 mg L−1 zeocin and 1.6 g L–1 NaHCO3. Colonies emerged around three weeks later and were typically transferred after 25 days. CRISPR/Cas library construction and genetic screening strategy All gRNA candidates for each protein-coding gene were predicted by a local version of CHOPCHOP. The parameters used were “-t CODING -T 1 -g 20 -BB AGGCTAGTCCGT -M NGG -P -J -3 ‘PRODUCT_SIZE_MIN = 250, PRODUCT_SIZE_MAX = 350, PRIMER_MIN_SIZE = 18, PRIMER_MAX_SIZE = 25, PRIMER_OPT_SIZE = 22, PRIMER_MIN_TM = 57, PRIMER_MAX_TM = 63, PRIMER_OPT_TM = 60’ -a 60 -scoringMethod DOENCH_2016 -filterGCmin 30 -filterGCmax 70 -filterSelfCompMax 1 -p 0”. To reduce off-target effects and avoid sustained functional activity of the resulting truncated protein, the following selection criteria were applied: (i) lack of self-complementarity, (ii) absence of gRNA off-targets with zero to three base mismatches, (iii) predicted efficiencies of editing exceeding 50%, and (iv) target sites located within the first third of the CDS region. In cases where no gRNAs that satisfy these criteria can be identified, the final criterion was relaxed to include target sites within the first half of the CDS region in order to enhance coverage. Subsequently, the gRNA library was synthesized by custom microarrays, amplified (Supplementary Table 2), and then ligated to the digested plasmid with the CRISPR/Cas system by infusion. The ligated plasmids were then transformed into E. coli, and over 30,000 colonies were selected to establish a plasmid library. The gRNA PCR library and the plasmid library were sequenced by NGS to evaluate their quality. The plasmid library was then transformed into N. oceanica, with the resulting transformants being transferred into a 24-well plate for subsequent analysis. High-throughput detection of mutation sites in the transformants by NGS To detect mutations induced by the CRISPR/Cas system, transformants in each 24-well plate were combined to perform shotgun genome sequencing. A dedicated pipeline (scmli; https://github.com/gongyh/scmli) was developed to call the variants around gRNA-targeted sites. Briefly, raw reads were quality controlled using Trim Galore (https://github.com/FelixKrueger/TrimGalore) with parameters “--paired -q 20 --trim-n --max_n 0 --length 70”. Then, a modified version of snippy (https://github.com/gongyh/scmli/blob/main/libs/snippy) was used to call preliminary SNPs and InDels with parameters “--basequal 30 --minqual 0.0 --minfrac 0.0”. Quality-filtered reads were aligned to the reference genome employing bwa97 with the options “-Y -M”; alignments containing clipped bases exceeding a length of 10 were discarded using samclip (https://github.com/tseemann/samclip); duplicate mapped reads were removed via samtools98; variant calling was performed using freebayes99 with parameters “-p 1 -P 0 -C 1 -F 0.00000001 --min-coverage 10 --min-repeat-entropy 1.0 --min-base-quality 30 --min-mapping-quality 60 --strict-vcf” to enable detection of low-frequency mutations while ensuring reliance solely on high-quality mapped reads and supporting bases with high base quality scores. Based on detailed comparison between Sanger sequencing results and NGS data for three 24-well plates (Supplementary Data 2, Supplementary Table 3 and Supplementary Note 1), following filtering criteria were applied: mutation types categorized as “snps”, “mnps”, “ref”, “bnd”, and “other” were filtered to retain variants supported by at least two alternate observations, whereas “indels” were retained if supported by a minimum of one alternate observation. Finally, variants located within or overlapping with the gRNA target regions, defined as -11 bp to +5 bp relative to the theoretical cutting site, were retained for subsequent analysis. The dual filtering, read support plus gRNA-target localization ensure that the retained variants are both statistically and biologically consistent with CRISPR editing. As in other high-throughput genotyping methods (e.g., ChlaMmeSeq100 and LEAP-Seq17), a small residual false-positive rate is unavoidable. Therefore, to support quality control and assessments, all the genotyping information has been made publicly available via the NanDeSyn database at http://nandesyn.org/CRISPR_library. Additionally, all mapping files can be visually examined as tracks in JBrowse at https://nandesyn.single-cell.cn/browser. To identify potential plasmid integration into the chromosome from whole-genome sequencing (WGS) data, the bioinformatics tool Isling101 was utilized. The circular plasmid sequence was linearized, with the BleR gene designated as both the start and end. The detection was based on mapping reads that are either chimeric or discordant. To ensure reliable results, only evidence supported by at least five reads was taken into account. Single-cell sorting of carotenoid-overproducing mutants by RACS The RACS process was performed on a FlowRACS 3.0 instrument (Qingdao Single-cell Biotech, CN) using the pDEP-DLD-RACS chip, for both N. oceanica mutant libraries and wild-type controls42,43. The FlowRACS instrument was equipped with a Nd:YAG 532 nm laser emitter as the excitation light source (20 mW was used in all the experiments), a 50× objective (NA = 0.7, Sunny Group Co., Ltd., CN) to focus the laser beam on the sample, and an electron-multiplying charge-coupled device (EMCCD) to collect SCRS. The cells were centrifuged at 6000 × g and washed three times with sterile deionized water, collected and resuspended in 1% Pluronic F-127 (Sigma Aldrich, USA) for RACS. An alternating current (ac) of 18 volts peak to peak (Vp-p) at 1 MHz frequencies was applied to generate positive dielectrophoresis (pDEP) force to trap the cells, the 532 nm laser power of 20 mW and acquisition time of 100 ms was used for obtaining SCRS, and the flow velocity of 40 μL min−1 was used to load cells. To assess the stability of Raman acquisition, pure cultures of wild type N. oceanica cells were measured, and the peak position (Supplementary Fig. 3A) and intensity (Supplementary Fig. 3B) at 1509 cm−1 were directly extracted and analyzed via the SCRS acquired during the 5-h measurement. To optimize the sorting duration (Supplementary Fig. 3C), pure cultures of wild type N. oceanica cells were sorted. The sorting process was video recorded and set to be always-trigger, under which all the cells would be sorted, and the percentage of cells sorted successfully was manually counted via the video. To assess the sorting accuracy and yield (Supplementary Fig. 3G, H), concentration of target cells (N. oceanica, featuring strong resonance Raman signal derived from pigment and red autofluorescence derived from chlorophyll) and nontarget cells (S. cerevisiae, featuring a weak nonresonance Raman signal and non-autofluorescence) were measured using a cell counting plate and adjusted to 1.5 × 104 cells mL−1 (~600 cells in 40 μL loading buffer), then were 1:1 mixed and underwent pDEP-DLD-RACS based on the sorting criteria of “I1509 - I1800 > 100” (Supplementary Fig. 3D). After sorting, the same amount of liquid from “Collect” and “Waste” was loaded into a pDEP chip separately, and the cells were trapped and concentrated in the chip via pDEP force (Supplementary Fig. 3E) and then fluorescence imaged (Supplementary Fig. 3F) with a laser-scanning confocal microscope, FluoView FV1000 (Olympus, Japan). Chlorophyll autofluorescence was excited at 559 nm and detected at a bandwidth of 650–750 nm. We denoted the numbers of the target and nontarget cells in “Collect” as Ntarget, collect and Nuntarget, collect, respectively, and those in “Waste” as Ntarget, waste and Nuntarget, waste, respectively. From the enumeration results, the sorting accuracy was calculated as Ntarget, collect/(Ntarget, collect + Nuntarget, collect) and the yield was calculated as Ntarget, collect/(Ntarget, collect + Ntarget, waste). To sort the carotenoid-overproducing cells from the genome-wide mutant library, the cells cultured for 5 days were centrifuged at 6000 × g and washed three times with sterile deionized water, collected and resuspended in 1% Pluronic F-127 (Sigma Aldrich, USA). Cell concentration was measured using a cell counting plate and adjusted to 1.5 × 104 cells mL−1 for RACS. Over 1000 cells were first measured in a single pDEP-DLD-RACS run, while the values of “I1509 − I1800” in SCRS revealed that “5475” separates the top 1% of cells in intracellular carotenoid content from the remaining 99% of cells (Supplementary Fig. 3I). Therefore, the sorting criterion was set as “I1509 − I1800 > 5475” to sort carotenoid-overproducing cells from the mutant library. MDA amplification of CRISPR/Cas plasmids in the SCRS-based sorted cells The CRISPR/Cas plasmids in RACS-screened N. oceanica mutants underwent amplification (MDA) with phi29 DNA polymerase95. Approx. 800 cells were sorted using FlowRACS 3.0 with a pDEP-DLD-RACS chip (SCB-D002). The sorted cells were lysed with 1 μL lysis buffer (Buffer A) at 65 °C for 10 min, followed by the addition of 1 μL stop solution (Buffer B). Subsequently, a 10× HotJa-reaction buffer (0.4 M Tris-HC1 (pH 7.6), 0.5 M KCl, 50 mM (NH4)2SO4, 100 mM MgCl2, 40 mM DTT), dNTP (0.6 mM each), MDA-F1/F2/F3 primers (3 mM each), and HotJa Phi29 DNA Polymerase were added to each tube. The mixtures were thoroughly combined and incubated at 40 °C for 8 h with a heated lid at 70 °C on a thermocycler. The amplified DNA served as the PCR template for amplifying gRNA target sites, and the resulting PCR products were sequenced using NGS. Enrichment analysis of gRNA sequences in the sorted genome-edited N. oceanica mutants Genomic gRNAs were detected and quantified using the scmli package (https://github.com/gongyh/scmli). Significantly enriched gRNAs for each mutant library were identified using edgeR102, with a false discovery rate (FDR) of <0.05 and a fold change of greater than 2. Gene Ontology (GO) and KEGG Orthology (KO) terms for the genes targeted by these gRNAs were retrieved from the NanDeSyn database (https://nandesyn.single-cell.cn), and enrichment analysis was performed using the clusterProfiler package103,104. Construction of the tRNA-based CRISPR/Cas plasmids for mutating the five noVDEs in N. oceanica To construct five knockout CRISPR/Cas systems for noVDEs, gRNA-HDV was amplified from pNOC-ARS-CRISPR-v248, and glycine tRNA was amplified from the genome of N. oceanica to create a gRNA-HDV-tRNA structure by infusion. Four pairs of primers (Supplementary Table 2) were used to amplify the gRNA-HDV-tRNA structure, and the PCR products were ligated to the BspQ I-digested tRNA-based CRISPR/Cas backbone by infusion to generate the plasmid for transformation. Measurement of photosynthesis parameters for the wild-type and mutant N. oceanica Cells were cultured in flasks to reach the logarithmic phase at 23 °C under continuous light (50 μmol photons m−2 s−1). The maximum quantum yield of photosystem II (Fv/Fm) of WT and mutants was measured using a pulse-amplitude modulated fluorometer (Image PAM, Walz, Effeltrich, Germany) after 20 min dark treatment of the cells. Extraction and determination of pigments for the wild-type and mutant N. oceanica N. oceanica cells were collected via centrifugation, rinsed with deionized water, and extracted using ethanol. The ethanol extracts were then separated using a Thermo Scientific Vanquish high-performance liquid chromatography (HPLC) system equipped with a variable wavelength detector set at 450 nm for carotenoid detection and an Agilent 5 μm Eclipse XDB-C18 4.6 × 250 mm analytical column. Pigments were eluted at a flow rate of 1 mL min−1 with a linear gradient from 100% Solvent A (85% methanol) to 70% Solvent B (ethyl acetate) over a 16 min period, followed by 2 min 70% Solvent B. Then pigments were eluted using a linear gradient from 30% Solvent A to 100% Solvent A over 2 min, followed by 10 min of Solvent A. Individual carotenoids were identified by their absorption spectra, and their typical retention times were compared to standard samples of pure carotenoids. RNA-Seq library construction for the wild-type and mutant N. oceanica Total RNA of noPAC4 mutants and WT cultured for five days was extracted using Trizol reagents (Tiangen, China). Concentration and purity of the RNA were determined spectrophotometrically (IMPLEN, USA), and RNA integrity was assessed with the RNA Nano 6000 Assay Kit on an Agilent Bioanalyzer 2100. A total of 2 μg of RNA per sample was used as input material for RNA sample preparation. Sequencing libraries were generated using the NEBNext Ultra™ RNA Library Prep Kit for Illumina (NEB, USA), and indices were added to associate sequences with each sample. The clustering of indexed samples was performed on a cBot Cluster Generation System using the HiSeq 3000/4000 PE Cluster Kit Box 1 (Illumina, USA). After cluster generation, the library preparations were sequenced on an Illumina HiSeq 4000, generating 150 bp paired-end reads. Computational analysis of transcriptome data from the wild-type and mutant N. oceanica RNA-seq datasets underwent quality control and then generation of gene expression profiles, based on the nfcore/rnaseq pipeline (https://nf-co.re/rnaseq). Specifically, reads were aligned with STAR, employing modified parameters to limit intron lengths (“--alignIntronMin 20, --alignIntronMax 3000”). Subsequently, Trinity105 was utilized to generate a gene expression matrix using the RSEM2 method, with TMM normalization implemented to account for library size variations among samples. Then transcripts per million (TPM) values were averaged across replicates. For the mutant samples, DEGs were identified by edgeR102, via an FDR threshold of ≤0.001 and an FC of >2. Statistical significance of TPMs across different categories was assessed by the Wilcoxon test. Gene set enrichment for GO terms and KEGG pathways was analyzed by clusterProfiler103,104, with a Benjamini–Hochberg adjusted p value of 0.05 set as the significance cutoff. Statistical analysis All experiments were conducted with at least three independent biological replicates. Data are presented as the mean ± standard deviation (SD). Statistical analyses, including the assessment of growth rates, carotenoid production, and Fv/Fm, were performed using a two-sided Student’s t test in Microsoft Excel to identify significant differences between groups. For other comparisons, significance was assessed using Student’s t test, Wilcoxon test, Chi-squared test, permutation test, or linear regression analysis in R, as specified in the figure legends. p < 0.05 was considered statistically significant. Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Supplementary information Supplementary Information Description of Additional Supplementary Files Supplementary Data 1 Supplementary Data 2 Supplementary Data 3 Supplementary Data 4 Supplementary Data 5 Reporting Summary Transparent Peer Review file Supplementary Information Description of Additional Supplementary Files Supplementary Data 1 Supplementary Data 2 Supplementary Data 3 Supplementary Data 4 Supplementary Data 5 Reporting Summary Transparent Peer Review file Source data Source data Source data