One-step generation of multiple gene knock-outs in the diatom Phaeodactylum tricornutum by DNA-free genome editing

Recently developed transgenic techniques to explore and exploit the metabolic potential of microalgae present several drawbacks associated with the delivery of exogenous DNA into the cells and its subsequent integration at random sites within the genome. Here, we report a highly efficient multiplex genome-editing method in the diatom Phaeodactylum tricornutum, relying on the biolistic delivery of CRISPR-Cas9 ribonucleoproteins coupled with the identification of two endogenous counter-selectable markers, PtUMPS and PtAPT. First, we demonstrate the functionality of RNP delivery by positively selecting the disruption of each of these genes. Then, we illustrate the potential of the approach for multiplexing by generating double-gene knock-out strains, with 65% to 100% efficiency, using RNPs targeting one of these markers and PtAureo1a, a photoreceptor-encoding gene. Finally, we created triple knock-out strains in one step by delivering six RNP complexes into Phaeodactylum cells. This approach could readily be applied to other hard-to-transfect organisms of biotechnological interest.

T he term microalgae encompasses a vast diversity of organisms, which make major contributions to biogeochemical cycles on a global scale 1 . A very attractive aspect concerns the ability of most microalgae to perform photosynthesis, allowing the generation of biological compounds from carbon dioxide (CO 2 ) as an inorganic carbon source and light as an energy source. Thus, these versatile cell factories have the potential to become key enablers for industrial biotechnology. Today, only a few microalgae species are commercially exploited 2 . They include the diatom Phaeodactylum tricornutum, which naturally synthesizes numerous marketable compounds 3 , such as pigments and long-chain omega-3 fatty acids 4,5 .
Genetic engineering is a promising approach to improve the metabolic potential of microalgae and obtain further insights into their metabolism and physiology. Custom molecular scissors have recently emerged as useful tools to induce site-specific modifications to the genome of diatoms [6][7][8][9][10][11][12] . These nucleases recognize and introduce a double strand break (DSB) at a target genomic sequence, which is repaired by non-homologous end joining (NHEJ) with potentially associated targeted mutagenesis (TM) or by homology-directed repair (HDR) in the presence of a DNA donor template 13 . To date, meganucleases, TALE nucleases, and the CRISPR-Cas9 system have all been successfully applied to inactivate single target genes in Phaeodactylum, either for functional analysis [6][7][8][9][10][11][12] or to redirect natural metabolism towards increased neutral lipid biosynthesis 6 . However, the construction of complex synthetic metabolic pathways requires the simultaneous introduction of multiple genetic modifications, as exemplified by the ten genome changes required to generate a hydrocortisone-producing yeast 14 . Such a challenge has not yet been addressed by the microalgal community.
Indeed, the "nuclease-driven" genetic engineering of diatoms is still in its infancy. Until now, it has been mediated by transforming plasmids encoding a nuclease and an antibiotic resistance cassette into the cells, both then stably integrated at random sites within the nuclear genome [6][7][8][9][10][11] . Disadvantages of this approach include: the low transformation efficiencies (less than 10 -6 ); the long-term expression of the nuclease can potentially induce offtarget cleavage 15; the random integration of all or part of the plasmid DNA into the genome can lead to undesired gene disruptions or uncontrolled effects on gene expression near the integration site(s); and the impossibility to eliminate background mutations or integrated transgenes through outcrossing in Phaeodactylum, as it is a diploid organism with no known sexual reproduction. Alternative genome editing methodologies must be developed to circumvent these issues. Recently, an episome-based protocol that avoids genomic integration has been described 12 that still elicits Cas9 expression over several generations, with potential toxicity or off-target effects 15 . Another possibility is to establish DNA-free genome-editing approaches, as illustrated in only one microalga thus far: the chlorophyte Chlamydomonas reinhardtii [16][17][18][19] . In Chlamydomonas, the direct transformation of Cas9/ single-guide RNA (sgRNA) ribonucleoprotein complexes (RNPs) can drive TM, but requires the use of cell-wall less mutants 18,19 or a plasmid carrying an antibiotic resistance cassette must be cotransformed with the RNP for selection 17 . The challenge in Phaeodactylum rests on both the introduction of RNP complexes through the complex cell wall and the development of an antibiotic-free selection method to enrich for transformants harboring TM events. We thus propose a strategy relying on the simultaneous co-delivery of multiple RNP complexes by biolistic, one targeting an endogenous gene for which inactivation confers positive selection, and the others targeting genes of interest.
Here, we identify two endogenous marker genes in Phaeodactylum, one homologous to URA3 20 , the other to APT 21 . We demonstrate that their inactivation, by either nucleases expressed from the genome or RNP complexes, leads to 5-fluoroorotic acid (5-FOA) and 2-fluoroadenine (2-FA) resistance, a property that can be used for the direct selection of the transformants. Combining RNPs targeting either of these marker genes and another gene of interest, PtAureo1a, encoding a blue-light photoreceptor/ transcription factor 22 , we generated multiple knock-outs in a single step at high frequency. Our results illustrate that "DNAfree" genome-editing can be achieved in diatoms. We anticipate that it can be transferred to any other hard-to-transfect organism, as it relies on well conserved endogenous selection markers.

Results
Inactivation of the UMP synthase gene using TALENs. We first evaluated the use of endogenous positive selectable markers for DNA-free genome editing by attempting to produce knock-out (KO) strains of the PtUMPS gene (Phatr3_J11740), the Phaeodactylum URA3 homolog of which knock-down increases 5-FOA tolerance while leading to uracil auxotrophy 20 . We designed a TALEN pair to target the orotidine-5′-phosphate decarboxylase domain-encoding sequence within PtUMPS exon 1 (Fig. 1a). The corresponding TALEN-encoding plasmids were assembled and co-delivered by biolistic bombardment into wild-type (WT) Phaeodactylum cells (NCMA strain), together with a plasmid conferring resistance to nourseothricin (NAT). Of the 30 NATresistant colonies that appeared, five showed successful genomic integration of the TALEN monomer genes, as revealed by PCR. These colonies were then spotted onto 5-FOA selective medium supplemented with uracil 20 ; two of five colonies, hereafter referred to as 8A2 and 12A1, grew in this condition. Genomic PCR amplification of PtUMPS from 8A2 and 12A1 revealed the presence of mixed cell populations, a phenomenon referred to as mosaicism 13 (Fig. 1b). Based on chromatogram deconvolutions, the TIDE calculation software 23 suggested a variety of INDELs at the PtUMPS locus for the 5-FOA resistant colonies (Fig. 1c). Therefore, we subcloned the 8A2 and 12A1 populations on medium containing 5-FOA and uracil and then sequenced the PtUMPS locus in 36 individual subclones for 8A2 and 20 individual subclones for 12A1. As expected, all showed mutagenic events (examples are depicted in Fig. 1d). In 19 of 20 12A1 subclones, both PtUMPS alleles showed mutagenic events based on polymorphism patterns present upstream and downstream of the target site ( Supplementary Fig. 1). Only one allele was amplified in 35 of 36 8A2 subclones and 1 of 20 12A1 subclones (Fig. 1d), suggesting large insertions or deletions in the other allele 6,9 , a phenomenon known as loss of heterozygosity (LOH) 24 . In the detected allele, INDELs were systematically present.
We further characterized two 8A2 subclones, 8A2_1 and 8A2_8, and two 12A1 subclones, 12A1_2 and 12A1_8, for their ability to grow in liquid F/2 medium with or without uracil and with or without 5-FOA ( Fig. 1e-g; Supplementary Fig. 2). The WT strain (NCMA) and a colony (2A3) only transformed with the NAT R plasmid were used as controls. All subclones carrying mutations in PtUMPS exhibited growth in the presence of 5-FOA and uracil (Fig. 1e), in contrast to the two controls. This confirms the expected phenotype for the PtUMPS KO concerning 5-FOA resistance. The quasi-absence of growth observed for the PtUMPS mutants in the absence of uracil confirmed their auxotrophy for uracil (Fig. 1f). Importantly, their growth kinetics in medium supplemented with uracil was equivalent to that of controls cultivated in the presence or absence of uracil (Fig. 1g). This suggests that the PtUMPS mutation has no impact on cell fitness under our culture conditions, a major prerequisite for further laboratory and industrial applications. We thus confirmed that PtUMPS KO leads to 5-FOA R and uracil auxotrophy, a phenotype that can be used for direct selection.
Antibiotic-free Cas9-driven inactivation of the PtUMPS gene. Other prerequisites for our proposed DNA-free genome editing strategy consist of demonstrating that (i) a PtUMPS KO can be achieved efficiently without introducing an exogenous antibiotic resistance cassette, solely relying on the 5-FOA R phenotype for selection and (ii) multiple genes can be efficiently inactivated in a single step, including PtUMPS, which serves as an endogenous co-selection marker. The versatility of the CRISPR-Cas9 system, associated with its potential for multiplexing, prompted us to evaluate its capacity to achieve this strategy. We designed three gRNAs directed towards sequences situated within PtUMPS exons 1 and 3 (gUMPS1 and gUMPS4, Fig. 2a), which encode the putative catalytic sites, and exon 2 (gUMPS3) using the CRISPOR web-based tool 25 .
We tested whether the selected gRNAs can efficiently target the Cas9 protein towards the PtUMPS locus by first performing in vitro assays. gRNAs were complexed to the recombinant Cas9 protein ( Supplementary Fig. 3), thereby forming ribonucleoproteins (RNPs). The complexes were then mixed with a PtUMPS amplicon comprising the target sites. All RNP complexes efficiently catalyzed double strand DNA cleavage of the PtUMPS fragment, whereas Cas9 or gRNAs alone did not ( Supplementary  Fig. 3).
Three separate sgRNA expression vectors were assembled for gUMPS1, gUMPS3, and gUMPS4, as previously described 10 , to examine the in vivo functionality of these gRNAs. Individual sgRNA expression plasmids were co-bombarded into WT cells together with a plasmid encoding the Cas9 nuclease and another carrying the NAT R cassette. NAT R colonies appeared under all transformation conditions after 2-3 weeks. The presence of the Cas9-encoding gene was verified in all transformants by PCR before the PtUMPS locus was sequenced. From 29 to 100% of the screened colonies showed TM at the PtUMPS locus, depending on the gRNA used (Fig. 2b). Examples of mutagenic events are  Next we examined whether it was possible to obtain 5-FOAresistant colonies after direct selection of the cells transformed with plasmids encoding the Cas9 protein and either gUMPS1 or gUMPS4. Cells were transformed with either of these gRNA encoding plasmids and directly spread onto selective medium containing uracil and 5-FOA. Twelve 5-FOA R clones appeared. Their genotypic characterization confirmed the presence of mutations adjacent to the PAM sequence in 10 out of 12 cases (Supplementary Table 1), demonstrating that direct selection on 5-FOA can lead to the identification of mutagenic events. The two remaining clones were not mutated at the PtUMPS loci, suggesting that they were false positives.

Demonstration of RNP delivery in Phaeodactylumtricornutum.
We established the proof of concept of DNA-free genome editing in diatoms by adapting a proteolistic protocol established by Martin-Ortigosa and Wang 26 to deliver Cas9-gRNA RNPs into Phaeodactylum cells (Fig. 3a). We first evaluated the frequency of targeted mutagenesis induced by the Cas9-gUMPS1 RNP complex in the absence of selection. To achieve that, P. tricornutum cells were bombarded with a dose response of 0, 2, 4, and 8 µg of Cas9-gUMPS1 complex. We did not expect a RNP transformation efficiency higher than 10 -6 in the absence of selection, as in the case of DNA delivery it is necessary to co-transform 100 million cells with a plasmid encoding an antibiotic resistance to obtain 30 antibiotic-resistant clones 6,9,10 . To test this hypothesis, we collected the cells growing in the absence of selection at four days post-bombardment and quantified the mutagenesis at the PtUMPS locus using locus-specific PCR followed by deep sequencing (Supplementary Fig. 7). As positive controls, different amounts of a monoallelic mutant strain carrying a 1nt deletion at the gUMPS1 target site were mixed with WT cells to get cell-tocell ratios of 100%, 10%, 1%, 0.1% and 0%. Whereas mutagenic events were detected in the positive controls at the expected frequencies, no induced mutagenic event was detected in the samples corresponding to cells bombarded with Cas9-gUMPS1, which reveals a TM frequency of less than 0.4%. This result suggests that in the absence of selection, it is quasi impossible to produce DNA-free transgenic strains. We decided to evaluate our ability to generate such mutant strains using the PtUMPS counter-selectable marker.
The gUMPS1, gUMPS3, and gUMPS4 guides were individually mixed with the Cas9 protein to form the RNP complexes. Four micrograms of individual Cas9-gRNA RNP complex was coated onto gold particles and bombarded into WT Phaeodactylum cells using a helium gene gun. Cultures were plated onto 5-FOA plus uracil containing medium two or four days post-bombardment. After three to four weeks, 5-FOA resistant colonies appeared for all conditions, except when cells were bombarded with Cas9 only as a negative control (Supplementary Table 2). A second round of selection confirmed that four of four colonies obtained with the gUMPS4 RNP complex, one of three with the gUMPS1 RNP complex, and one of two with the gUMPS3 RNP complex could grow on 5-FOA uracil medium. These 5-FOA R strains were then analyzed for TM at the PtUMPS locus (Supplemental Table 1). We failed to amplify the PtUMPS locus in the gUMPS1-derived and gUMPS3-derived strains, suggesting that a large deletion may have occurred. However, we observed TM close to the PAM in all gUMPS4-derived strains. This shows that transient exposure to the RNPs is sufficient to induce TM at an endogenous locus. Thus, we generated knock-out transgenic strain without introducing exogenous DNA into the cells.
Optimizing multiplex DNA-free genome editing. Our main objective was to inactivate multiple genes of interest in a single step, relying on the PtUMPS KO 5-FOA-resistance phenotype for positive selection. We started with the assumption that double KO strains could be generated by bombarding cells with a combination of RNP complexes, targeting the PtUMPS gene and another gene of interest (Fig. 3a). We selected the gene encoding Aureochrome 1a as an example (PtAureo1a, Phatr3_J8113); PtAureo1a is a photoreceptor that participates in blue light perception 22 and whose TALEN-mediated biallelic genetic KO was previously reported 9,11 . The gRNAs targeting PtAureo1a (gAur-eo1a2 and gAureo1a3) were selected following the same in silico criteria as for PtUMPS (Supplementary Table 3). Their activities were first verified in vitro using the RNP-based assay (Supplementary Fig. 8). Next, WT Phaeodactylum cells were simultaneously bombarded with multiple RNP complexes, targeting both PtUMPS and PtAureo1a (Fig. 3a).We used two gRNAs per target (gUMPS1 and gUMPS3, targeting the PtUMPS locus; gAureo1a2 and gAureo1a3, targeting the PtAureo1a locus) and mixed them independently with recombinant Cas9 protein to increase our chances of obtaining TM at both loci. The four RNP complexes were then combined (Fig. 3a). Cells were then co-bombarded with 8 µg of RNP complexes (corresponding to 2 µg of each RNP) and transferred onto 5-FOA uracil medium two days postbombardment. Clones were first analyzed for TM at the PtUMPS locus (Fig. 3b). Among the 19 5-FOA R colonies that appeared, 17 (89%) showed TM at the PtUMPS locus: 4/17 (23%) with mutagenic events on both alleles and 13/17 (76%) with mutagenic   events on one allele only, the second carrying the WT sequence (Fig. 3c). Amplification of PtUMPS resulted in a clearly shorter PCR product for 11/17 (65%) of the clones (Fig. 3d, Supplementary Fig. 9). Sequencing of these PCR products revealed deletion of the fragment between gUMPS1 and gUMPS3, indicating that DSBs were created at both target sites (Fig. 3d, Supplementary Fig. 9). The PtAureo1a locus of the 19 5-FOA R colonies was amplified by PCR and sequenced. The two strains that did not show TM of PtUMPS also showed no TM of PtAureo1a. Further analysis revealed that 100% of the PtUMPSmutated clones also displayed a mutagenic event at the PtAureo1a locus; 5 of 17 carried mutations on both alleles (Fig. 3d, e). Thus we achieved simultaneous multiple gene knock-outs in diatoms.
Extrapolation to another selectable marker. We evaluated the transferability of this approach by investigating another putative selectable marker, the Adenine Phosphoribosyl Transferase (APT) gene, for which inactivation confers 2-FA resistance in several organisms, including plants 21,27 . APT normally contributes to adenine recycling as part of the nucleotide salvage pathway 27 . We identified a single potential APT encoding gene in the Phaeodactylum genome (Phatr3_J6834, hereafter referred to as PtAPT; UniProt B7G7K1), based on sequence homology of its translation product with the Physcomitrella patens APT protein (PpAPT, UniProt: Q45RT2) 21 . Overall, PtAPT shares 41.7% amino-acid identity with PpAPT and 28.6% with the Arabidopsis thaliana APT1 protein (AtAPT1, UniProt: P31166) (Fig. 4a). The phosphoribosyl transferase (PRT) type I domain, spanning amino-acid positions 28 to 132 in PtAPT, is well conserved. Before proceeding further, we determined whether 2-FA was toxic to Phaeodactylum cells and at what concentrations. We observed a dose-response effect of 2-FA on growth, with complete inhibition at 10 µM (Fig. 4b).
We designed gRNAs directed against PtAPT (gAPT1 and gAPT3) to target regions within exons 1 and 2, respectively (Fig. 4c). Both the gAPT1 RNP and gAPT3 RNP complexes efficiently catalyzed DSB DNA cleavage of a PtAPT amplicon in vitro (Supplementary Fig. 10). Either the gAPT1 or the gAPT3 RNP complex was then introduced into WT Phaeodactylum cells by proteolistic transformation. Cultures were plated onto selective medium containing 10 µM 2-FA and 5 mg.L −1 adenine, two days post-bombardment (Fig. 4d). Eleven and five 2-FA R colonies appeared after two weeks for the gAPT1 and gAPT3 RNPs, respectively, whereas no colonies appeared in the negative control bombarded with the Cas9 protein alone. Among the 16 2-FA R colonies, 15 (93%) showed TM at the PtAPT locus (Fig. 4d), all of which carried mutations on both alleles, suggesting that biallelic KO is required to trigger resistance to 2-FA. Examples of mutagenic events are shown in the Supplementary Figs. 11 and 12. We thus demonstrated that inactivation of PtAPT confers 2-FA R , allowing the use of PtAPT as an endogenous positive selectable marker in Phaeodactylum.
We next evaluated the impact of knocking out PtAPT on growth in F/2 medium, F/2 medium supplemented with adenine, and F/2 medium supplemented with both 2-FA and adenine (Fig. 4e). As expected, WT cells were unable to grow in the presence of 10 µM 2-FA. The two tested APT KO strains were unaffected by the addition of 2-FA and grew normally in F/2 or F/ 2 plus adenine. Inactivating PtAPT did not negatively affect culture fitness, an important requirement for downstream applications.
We verified that the PtAPT KO can be used as a co-selection marker for multiplexing experiments by simultaneously bombarding WT Phaeodactylum cells with RNP complexes targeting the PtAPT and PtAureo1a loci (Fig. 4f). Two gRNAs were used per target (gAPT1 and gAPT3 targeting the PtAPT locus; gAureo1a2 and gAureo1a3 targeting the PtAureo1a locus). Thus, cells were bombarded with 8 µg of an RNP mixture consisting of all four complexes and spread onto selective medium containing 2-FA and adenine two days post-bombardment. Thirty-four 2-FA R clones appeared within two weeks. All showed TM at the PtAPT locus. Amplification of PtAPT resulted in shortened PCR fragments for 47% of the colonies (16/34), indicating that the deletion occurred between the gAPT1 and gAPT3 target sites ( Supplementary Fig. 13). These results were confirmed by sequencing 29 of these 34 clones. We also investigated the presence of mutagenic events at the PtAureo1a locus. Sixty-five percent (19/29) of the colonies harboring PtAPT mutations also showed mutagenic events at the PtAureo1a locus and 52% (10/19) of them exhibited a deletion of the fragment between the two target sites. Interestingly, 52% (10/19) of the PtAureo1a mutants were mutated on both alleles, leading to the generation of PtAureo1a KO strains (Fig. 4f).
We next determined whether three genes can be simultaneously knocked-out. Cells were bombarded with a mixture of six RNP complexes, two RNPs targeting PtAPT, two PtUMPS, and two PtAureo1a, and selected on 2-FA medium supplemented with adenine and uracil. Fourteen 2-FA-resistant colonies appeared. Thirteen of the fourteen colonies (93%) showed a mutagenic event at the PtAPT locus (Fig. 5). Among the 13 clones, five were mutated at both PtAPT and PtAureo1a, two at both PtAPT and PtUMPS, and two at all the three loci. The genotype of these clones is summarized in Fig. 5.
We should point out that the selection of 2-FA resistant clones was achieved in all of our independent trials (n = 7) whereas the selection of 5-FOA-resistant clones failed in three experiments out of eight. Such a disparity has already been reported in other organisms where pH, light, and temperature have been shown to influence 5-FOA selection strength. The major point is that we have succeeded in generating multiple PtAureo1a knock-out strains in the three independent multiplexed experiments described in Supplementary Table 4. Our results show that this genome-editing approach is a powerful tool to simultaneously inactivate multiple genes without the use of DNA and antibiotics.
To go one step further, we phenotypically characterized eleven RNP-derived transformants from several independent experiments that were mutated in PtUMPS and/or PtAPT and/or PtAureo1a (Supplementary Fig. 14a). First we tested their resistance to 5-FOA, as well as 2-FA ( Supplementary Fig. 14b). For all these strains, sensitivity or resistance to different media reflected genotypes well. Next, we quantified the amount of PtAureo1a in the strains by Western Blotting ( Supplementary  Fig. 15). Whereas we did not detect any PtAureo1a protein in the samples derived from bi-allelic PtAureo1a mutant strains, we clearly detected it in protein samples derived from the tested monoallelic PtAureo1a strain and the positive controls. We additionally performed growth experiments and observed that the cells mutated in PtAureo1a had lower (20% maximum) growth rates compared to the parental strain or the PtUMPS and PtAPT single mutants (Supplementary Fig. 16). Finally, we evaluated the photo-physiological impact of the PtAureo1a mutation. This gene encodes for a blue-light photoreceptor of which knock-out strongly affects cells ability to dissipate excess light energy as heat upon a shift from low to high irradiances 9 . This can be quantified by measuring the Non-Photochemical Quenching (NPQ) of chlorophyll fluorescence 9 . As previously reported, biallelic P. tricornutum PtAureo1a mutant strains generated by TALE nucleases display reduced NPQ capacities compared to wildtype 9 . We performed similar NPQ capacity measurements. Whereas no NPQ phenotype was observed in the monoallelic PtAureo1a mutant strain tested here and in the PtAPT and PtUMPS knock-out strains, we measured a 50-73% reduction of NPQ in all RNP-derived biallelic PtAureo1a mutants tested (Fig. 6). Therefore, we confirmed the loss of function of the generated PtAureo1a knock-out strains, which reveals the power of our approach to study gene function.

Discussion
Here, we established a DNA-free approach that solves the important issues of random DNA integration into the diatom genome and long-term nuclease expression. This was achieved by combining RNPs targeting one endogenous marker gene with RNPs targeting another gene of interest, successfully generating dozens of transgenic strains free of foreign DNA, without the use of antibiotics. Moreover, the same strategy allowed the creation of triple-gene knock-out strains in a one-step procedure. This is a valuable research tool for algae genome engineering that should open the way to the study of putative redundant functions of multiple gene family members. For example, it could be suitable to investigate the role of specific light-harvesting complexes of the LHCX family during exposure to excess light 28 or the contribution of single aureochromes to photoacclimation 11 .
Our approach also represents a technical breakthrough for industrial biotechnology, as it will permit the simultaneous KO of multiple endogenous pathways to redirect metabolism towards the production of biomolecules of interest. Moreover, the success of our strategy was associated with the identification of the PtUMPS and PtAPT genes and the validation that their inactivation confers resistance to specific molecules. The use of such markers represents an attractive alternative to conventional antibiotic-resistance genes and solves the environmental issue posed by the dissemination of such resistance genes into other organisms through horizontal transfer.
The methodology we have developed addresses the global effort to exploit non-integrative expression of the CRISPR-Cas9 system. A recent report based on episome delivery provided a significant technological advance [12]; however, this system may not be sufficient to solve the challenge of limiting the amount and duration of nuclease exposure. Indeed, cells are exposed to nucleases for several weeks, the necessary time for the appearance of antibiotic-resistant colonies. To date, no studies have investigated a potential deleterious impact of long-term Cas9 expression in microalgae. If that were the case, the transient expression of the CRISPR-Cas9 system described here could decrease undesired mutations compared to long-term expression 29 . A major benefit of RNP delivery is its simplicity and ease of implementation, as it eliminates all subcloning steps, saving considerable time relative to traditional genome-editing approaches. Another attractive perspective relies on the fact that the UMPS and APT markers are well conserved within the microalgae phylogenetic tree and among other eukaryotic groups, making it highly likely that the method is extendable to other organisms. This is particularly important for hard-to-transfect species or those for which no biobricks (promoters, terminators) are available.

Methods
Culture conditions. The Phaeodactylum tricornutum strain CCMP2561 (NCMA) was grown axenically at 20°C in vented cap flasks containing silica-free F/2 medium (Sigma G0154) with 40% Sea Salts (Sigma S9883). Incubators were equipped with white neon light tubes providing an illumination of approximately 120 µmol photons m −2 s −1 and a photoperiod of 12 h light/12 h dark.
CRISPR-Cas9 design and constructs. Guide RNAs (gRNAs) were selected using the CRISPOR tool (http://crispor.tefor.net/) 25 , based on the Moreno-Mateos score and the absence of predicable off targets. Corresponding crRNA sequences are listed in Supplementary Table 3. The gRNAs were cloned under control of the U6 promoter in a vector derived from pKSdiaCas9 10 . This vector was co-delivered with the Cas9 encoding plasmid pKSdiaCas9.
Biolistic transformation of TALEN or Cas9 and gRNA vectors. Cells (1.5 × 10 8 total) were collected from exponentially growing cultures and spread onto 1% agar plates containing F/2 medium with 20 g L −1 sea salt. Transformations were carried out 24 h later using the microparticle bombardment method adapted from Apt et al. 32 , with minor modifications. Gold particles (0.6 µm diameter, BioRad) were coated with DNA using 1.25 M CaCl 2 and 20 mM spermidine. As a negative control, beads were coated with 3 µg NAT selection plasmid and 3 µg empty vector. For the CRISPR-Cas9 experiment, the DNA mixture contained 3 µg Cas9 expression vector (pKSdiaCas9) 10 , 3 µg U6-gRNA encoding plasmid, and 3 µg NAT selection plasmid. As a negative control, beads were coated with a DNA mixture consisting of 3 µg NAT selection plasmid, 3 µg Cas9 expression vector, and 3 µg empty vector. A burst pressure of 1550 psi and a vacuum of 25 Hg were used.
Preparation of the crRNA-tracrRNA duplex. Purified crRNAs and tracrRNA were purchased from IDT. Duplexes were prepared following the manufacturer's instructions.
In vitro CRISPR-Cas9 RNP cleavage assay. DNA fragments containing the target site were amplified, purified, and eluted with RNase-free water. CRISPR-Cas9 RNP complexes were assembled using crRNA::tracrRNA duplexes and HiFi Cas9 (IDT #1074181). Each RNP complex was combined with the corresponding DNA amplicon and the in vitro cleavage reaction allowed to proceed, following the manufacturer's instructions. Sequential RNase A and proteinase K treatments were performed before gel separation and visualization.
Biolistic bombardment of Cas9 RNP complexes. RNP complexes were assembled following the instructions of IDT and delivered to WT Phaeodactylum cells by particle bombardment. Briefly, 24 h before the experiment, cells (1.5 × 10 8 ) were spread onto 1% agar F/2 plates with 20 g L −1 sea salt and 50 µg mL −1 uracil, if targeting PtUMPS. For each shot, the equivalent of 4 or 8 µg Cas9 protein (for multiple targets, the total amount was split equally between the different RNPs) in a total volume of 8 µl Cas9 reaction buffer (NEB, B0386A) was mixed with 10 μl gold nanoparticles (3 mg, 0.6 μm in diameter, Bio-Rad) washed twice with Cas9 reaction buffer. The coated particles were distributed slowly in a circle directly onto the macrocarrier, which was allowed to dry for 2 h before biolistic bombardment. A burst pressure of 1550 psi and a vacuum of 25 Hg were used.
Genotypic characterization. Cell lysates served as the PCR templates 6 . Primers listed in Supplementary Table 5 were used to amplify the various loci of interest. The Q5 High Fidelity DNA polymerase (New England BioLabs, USA) was used for general screening purposes. The highly selective HiDi DNA polymerase (myPOLS, Germany) was used for allele-specific PCR, together with primers ending with single nucleotide polymorphisms.
Deep sequencing. To quantify the frequency of mutagenic events, locus-specific PCR products were sequenced using NGS technology (S5 Thermofisher). In all conditions, 13,000-72,000 reads per sample were analyzed. The sequencing and the processing of NGS data was done by the GeT-Biopuces platform (INSA Toulouse) using the Ion Torrent Suite software analysis. The percentage of mutagenic event was calculated at position 145 (3nt upstream of the gUMPS1 PAM) and at +/− 5 bases surrounding this position. The frequency of total INDELs (including 1 base INDEL) was calculated, with a background noise level lower than 0.4%. The frequency of INDELs larger than 1 base was also analyzed and in this condition the background noise level was extremely low <0.06%.
Cell growth experiments. For the growth experiments reported in Fig. 1e-g, cells were cultivated in F/2 medium for five days and then diluted to 1.10 5 cells mL −1 in F/2 medium, F/2 medium supplemented with 50 µg mL −1 uracil, or F/2 medium supplemented with 50 µg mL −1 uracil and 100 µg mL −1 5-FOA. For the growth experiment reported in Fig. 4, a similar procedure was followed, except that cells were diluted into F/2 medium, F/2 medium with 5 mg L −1 adenine, or F/2 medium with 5 mg L −1 adenine and 10 µM 2-FA after five days.
Measurement of non-photochemical quenching (NPQ). Cell suspensions in mid-exponential phase were adjusted to a chlorophyll a content of 1 µg mL −1 and NPQ was measured with an AquaPen-C AP 100 (Photon Systems Instruments, Brno, Czech Republic) using light pulses with an intensity of 2100 µmol photons m −2 s −1 applied every 20 s to induce maximal fluorescence and 700 µmol photons m −2 s −1 of actinic light to induce NPQ.