Optimized inducible shRNA and CRISPR/Cas9 platforms for in vitro studies of human development using hPSCs

Loss-of-function experiments in human pluripotent stem cells [hPSCs; comprising human embryonic stem cells (hESCs) or human induced pluripotent stem cells (hiPSCs)] provide a unique opportunity to study the mechanisms that regulate human development, physiology and disease (Avior et al., 2016; Pourquié et al., 2015; Zhu and Huangfu, 2013). However, functional genomic applications of hPSCs are currently limited by the lack of an easy and efficient method to conditionally manipulate gene expression in both hPSCs and hPSC-derived cells. Indeed, such a system is necessary both for the study of genes essential for hPSC self-renewal and for functional analyses at specific stages of differentiation.

Historically, the expression of inducible short hairpin RNAs (shRNAs) has been the most popular method to trigger gene knockdown in human cells. This has been achieved using a TET-ON system, which relies on a modified RNA polymerase (Pol) III promoter that is responsive to a tetracycline-sensitive repressor protein (tetR) to induce shRNA expression by simple tetracycline (TET) treatment (Lambeth and Smith, 2013). Nevertheless, application of this TET-ON system in hPSCs has proved challenging for two main reasons: (1) tight control of shRNA expression is difficult to achieve, thereby resulting in uncontrolled knockdown; (2) induction of shRNA rarely works in differentiated derivatives. Indeed, very high and homogenous expression of both the tetR and the inducible shRNA is required to obtain potent yet controlled knockdown. However, transgene silencing is a recurring problem in hPSCs (Ellis, 2005; Herbst et al., 2012; Yao et al., 2004), and randomly integrated promoters are often subject to positional effects that can strongly limit their activity (Zafarana et al., 2009). Differentiation further increases the chances of silencing, as transgenes can be located in regions where heterochromatin forms following cell fate choices (Herbst et al., 2012; Raya et al., 2009). As a consequence, inducible shRNA expression in both hPSCs and a wide variety of their differentiated progenies has never been reported.

More recently, CRISPR/Cas9-mediated gene knockout has emerged as a powerful method to interrogate gene function (Wright et al., 2016), and inducible manipulation of gene expression in hPSCs using this approach has been reported (Chen et al., 2015; González et al., 2014; Mandegar et al., 2016). However, these methods are either very complex and time consuming, as they involve multiple genome editing steps that need to be individually tailored for each gene to be examined (Chen et al., 2015), or are not widely applicable in multiple differentiated cell types as they rely on inducible promoters that are not stably and homogeneously expressed following hPSC differentiation (González et al., 2014; Haenebalcke et al., 2013; Mandegar et al., 2016; Ordovas et al., 2015). Overall, there are currently no methods for inducible gene knockout in hPSCs that fulfill all the criteria described above.

Here we describe novel platforms for single-step optimized inducible gene knockdown or knockout (sOPTiKD or sOPTiKO) that address all the limitations of current inducible shRNA or CRISPR/Cas9 systems, thus providing powerful and scalable platforms that have the potential to greatly simplify the study of human gene function.

We aimed to develop optimal conditional loss-of-function platforms using inducible shRNAs or guide RNAs (gRNAs) for CRISPR/Cas9. We reasoned that inserting each element of the TET-ON system into a different genomic safe harbor (GSH; Sadelain et al., 2012) would maximize expression in hPSCs and their differentiated progenies while avoiding potential promoter interference (Shearwin et al., 2005). The AAVS1 and ROSA26 loci appeared particularly suitable for this purpose as these GSHs have been suggested to allow strong expression of various transgenes in hPSCs, including constitutively expressed shRNAs (DeKelver et al., 2010; Hockemeyer et al., 2009; Irion et al., 2007). We first improved the targeting efficiency for both GSHs by developing a CRISPR/Cas9n-based gene-trap strategy to target the human ROSA26 locus (Fig. 1A,B, Fig. S1A) and by refining an existing zinc-finger nuclease (ZFN)-based targeting strategy for the AAVS1 locus (Hockemeyer et al., 2009) (Fig. 1A,B). In both cases, hPSC targeting occurred with very high efficiency (59-100%; Table S1), while neither ROSA26 nor AAVS1 modifications resulted in chromosomal abnormalities (data not shown).

We then sought to identify the most efficient promoter to drive constitutive transgene expression from GSHs. We tested the ability of different promoter configurations to express an enhanced green fluorescent protein (EGFP) transgene from the ROSA26 locus in hESCs (Fig. S1A,B). The highest and most homogenous EGFP expression (100%) was achieved with the artificial CAG promoter (Fig. S1C-E), which was stronger by an order of magnitude than the endogenous ROSA26 promoter (Irion et al., 2007). Interestingly, and in contrast to previous reports (Ramachandra et al., 2011), we observed that the EF1α (EEF1A1) promoter was strongly silenced, as shown by mosaic EGFP expression (Fig. S1C-E). Similar results were obtained after EGFP targeting into the AAVS1 locus (data not shown), thereby preventing the use of this promoter in subsequent experiments.

To further evaluate the robustness of the CAG promoter activity, we analyzed in detail hESCs with heterozygous or homozygous targeting of a CAG-EGFP transgene in the ROSA26 or AAVS1 loci (Fig. 1A,B). For both GSHs EGFP was homogeneously expressed at high and comparable levels for more than 30 passages (Fig. S1F), and similar results were obtained after differentiation of hESCs into the three primary germ layers (Fig. S1G-M). Importantly, targeting did not interfere with pluripotency or differentiation, as shown by appropriate expression of lineage markers (Fig. S1N,O). We further differentiated these EGFP-hESC lines into fifteen different cell types (Fig. 1A), and both GSHs allowed homogeneous and strong EGFP expression in all cell types analyzed (Fig. 1C,D, Fig. S2). Overall, these results validate the ROSA26 and AAVS1 loci as suitable for robust transgene expression in both hPSCs and their derivatives.

Having demonstrated the suitability of the ROSA26 and AAVS1 loci for transgene expression, we developed a TET-ON inducible knockdown system based on dual GSH targeting (Fig. 2A, Fig. S3A). To simplify knockdown evaluation and method optimization we generated hESC lines in which an EGFP transgene could be silenced in an inducible fashion (Fig. 2B). To achieve this we targeted: (1) a CAG-tetR expression cassette into the ROSA26 locus; and (2) a CAG-EGFP transgene plus an inducible EGFP shRNA cassette into the AAVS1 locus (Fig. 2A,B). Interestingly, we observed a strong and homogeneous decrease in EGFP fluorescence following tetracycline treatment for 5 days (>95%; Fig. 2C), thereby confirming efficient knockdown. However, a decrease in EGFP expression was also noticed in the absence of tetracycline (Fig. 2C), suggesting a significant leakiness in the expression of the shRNA and thus confirming previous reports (Henriksen et al., 2007).

We then hypothesized that this limitation could be bypassed by expressing higher levels of the tetR protein to more strongly repress shRNA expression in the absence of tetracycline. We performed a multi-parameter RNA and codon optimization of the bacterial tetR cDNA (Fath et al., 2011) and used the resulting codon-optimized tetR (OPTtetR) to generate new EGFP inducible knockdown hESC lines (Fig. 2B). This modification achieved a tenfold increase in tetR expression compared with the standard sequence (STDtetR; Fig. 2D). Furthermore, homozygous expression of OPTtetR was sufficient to completely prevent shRNA leakiness while fully preserving efficient knockdown induction (Fig. 2C, Fig. S3B). Of note, the inducible knockdown was rapid, reversible and dose responsive (Fig. 2E,F, Fig. S3C-E). Finally, inducible hESCs displayed a normal karyotype (data not shown), demonstrating that the genome engineering necessary to create these lines did not alter their genetic stability.

Based on these encouraging results, we further validated this method in the context of endogenous genes by generating hESCs carrying inducible shRNAs against OCT4 (POU5F1) or B2M (Fig. S3F). Remarkably, all the sublines analyzed (six for each gene) showed robust inducible knockdown with no significant shRNA leakiness (Fig. S3G,H). Tetracycline titration identified optimal concentrations to partially or fully knockdown OCT4 (Fig. 2G, Fig. S3I,J). As expected, a strong decrease in OCT4 specifically resulted in loss of pluripotency and induction of neuroectoderm and definitive endoderm markers (Fig. 2H, Fig. S3I,J) (Thomson et al., 2011; Wang et al., 2012). Similar results were obtained with 20 additional OCT4 inducible knockdown hESC sublines, confirming the robustness and reproducibility of this method (Fig. S3K). Importantly, the generation of hESCs with strong and tightly regulated knockdown was so efficient that phenotypic analyses could be performed immediately after antibiotic selection on a mixed population of cells, thereby entirely bypassing the need to pick individual colonies for clonal isolation (Fig. S3K).

Overall, these results establish that dual targeting of GSHs with an optimized inducible knockdown system is a powerful method to control gene expression in hPSCs. This approach is hereafter named optimized inducible knockdown, or OPTiKD (Fig. 2A, Fig. S3F).

We then sought to further improve the OPTiKD system by developing an all-in-one targeting approach that would facilitate the rapid and scalable generation of inducible knockdown hPSCs. We constructed a single AAVS1 targeting vector to carry both the inducible shRNA and the CAG-tetR expression cassette (Fig. 3A), and validated this approach by knocking down the expression of an EGFP transgene targeted in the ROSA26 locus (Fig. S3L). Remarkably, this method shared key properties with OPTiKD, such as both the absence of shRNA leakiness (Fig. S3L,M) and rapid, reversible and dose-responsive inducible knockdown (Fig. S3N,O). Thus, this all-in-one strategy, which we named single-step optimized inducible knockdown, or sOPTiKD (Fig. 3A), is as efficient as our original dual targeting approach.

To further demonstrate the versatility of sOPTiKD, we generated both hESC and hiPSC lines carrying an inducible shRNA against OCT4 or B2M. Generation of sOPTiKD hPSCs following lipofection was rapid (2 weeks) and extremely efficient, as all the sublines generated showed robust inducible knockdown (Fig. 3B,C). qPCR analyses confirmed that knockdown of OCT4 using sOPTiKD induced differentiation of both hESCs and hiPSCs, whereas knockdown of B2M had no effect (Fig. 3D). Overall, these experiments show that sOPTiKD provides an efficient system to knock down gene expression that can be easily applied to a large number of hPSC lines.

Finally, we explored whether sOPTiKD could enable simultaneous knockdown of multiple genes (Fig. 3E). We focused on the cyclin D family (CCND1, CCND2 and CCND3). These cell cycle regulators are functionally redundant, and thus their study in hESCs has previously required laborious multiple rounds of stable shRNA transfection in order to achieve double or triple knockdown (Pauklin and Vallier, 2013). We developed a method to easily combine multiple shRNAs into the same targeting vector using a one-step Gibson assembly, and generated sOPTiKD plasmids carrying one, two or three shRNAs against cyclin D genes or scrambled control shRNAs (Fig. 3E). These vectors were tested in hESCs without isolation of clonal sublines, and inducible knockdown proved highly efficient and comparable with single, double and triple shRNA constructs (Fig. 3F). Interestingly, prolonged knockdown of one or two cyclin Ds was compatible with hESC self-renewal (Fig. 3G), whereas triple knockdown resulted in definitive endoderm differentiation (Fig. 3G,H), as previously reported (Pauklin and Vallier, 2013; Pauklin et al., 2016). Collectively, these results demonstrate that sOPTiKD can be used to simultaneously decrease the expression of several genes with redundant functions.

The capacity to knock down genes in a variety of differentiated cells would represent a significant advance over existing systems for inducible gene knockdown. To thoroughly test this possibility, we analyzed the efficacy of the OPTiKD and sOPTiKD platforms to knock down an EGFP transgene in hPSCs differentiated into the three germ layers, as well as in a panel of 13 fully differentiated cell types (Fig. 1A). For both methods, qPCR analyses demonstrated strong and inducible knockdown of EGFP transcripts in all lineages tested (Fig. 4A). Microscopy observations confirmed a robust decrease in EGFP protein expression (Fig. 4B), and flow cytometry showed a decrease in EGFP fluorescence of more than 70% for most lineages (Fig. S4A-G).

Interestingly, EGFP was less reduced in cell types with slower proliferation rates (Fig. S4A). Since EGFP has an extended half-life of more than 24 h, protein loss upon transcriptional or post-transcriptional inhibition relies heavily upon its gradual dilution following cell division (Li, 1998). Considering the strong decrease in EGFP mRNA, we concluded that residual EGFP fluorescence was likely to be a consequence of the relatively short tetracycline treatment performed to trigger knockdown (5 days). To test this, we induced prolonged EGFP knockdown in postmitotic cardiomyocytes, and we indeed observed a slow but constant decrease in protein expression for up to 20 days, at which point the EGFP was decreased by more than 75% (Fig. S4H,I). To reinforce these observations, we performed similar experiments using a ROSA26-targeted EGFP reporter transgene fused to a destabilization domain (EGFPd2; Li et al., 1998), which avoids confounding effects due to the long half-life of standard EGFP (Fig. S4J-M) (Wahlers et al., 2001). Remarkably, EGFPd2 inducible knockdown in cardiomyocytes using the sOPTiKD method resulted in >90% protein knockdown after 5 days of tetracycline treatment (Fig. S4N,O). Considered together, these results establish that OPTiKD and sOPTiKD allow efficient manipulation of gene expression even after differentiation of hPSCs.

We then sought to exemplify the use of OPTiKD and sOPTiKD to rapidly and efficiently evaluate endogenous gene function in a variety of cell types and at different stages of hPSC differentiation related to embryonic development. First, we focused on the master developmental regulator T (brachyury), which plays an essential role in mesoderm formation and, in particular, during the development of posterior mesoderm, notochord and somites (Martin, 2015; Papaioannou, 2014). Indeed, mice carrying a heterozygous mutation in T exhibit a short tail phenotype, while homozygous mutations are embryonic lethal at around 9.5 dpc (Dobrovolskaia-Zavadskaia, 1927; Gluecksohn-Schoenheimer, 1944). T mutants also present severe cardiovascular and placental defects (David et al., 2011; Inman and Downs, 2006; King et al., 1998). Furthermore, T was recently shown to specifically regulate mesoderm but not endoderm differentiation in hPSCs (Faial et al., 2015).

To investigate the role of T during the differentiation of hPSCs we combined T sOPTiKD sublines with culture conditions known to drive the differentiation of hPSCs into subpopulations that recapitulate different portions of the primitive streak and their derived lineages (Fig. 5A) (Bernardo et al., 2011; Cheung et al., 2012; Mendjan et al., 2014; Touboul et al., 2010). Inducible knockdown of T was robust in all cell types analyzed (Fig. 5B,C, Fig. S5A,B), confirming the efficiency of sOPTiKD to knock down developmental genes. Decrease in T expression did not affect definitive endoderm specification, while differentiation into posterior primitive streak cells, cardiac mesoderm and lateral plate mesoderm was mildly impaired (Fig. 5D, Fig. S5C). By contrast, the generation of late primitive streak progenitors (recapitulating the onset of somitogenesis) and their further specification into presomitic mesoderm and chondrocytes were severely affected following inducible knockdown of T (Fig. 5D-G). In particular, induction of TBX6, MSGN1 and MEOX1 was nearly abolished, in agreement with the established role of T in the expression of such genes (Chapman et al., 1996; Faial et al., 2015; Martin, 2015).

Collectively, these results strikingly recapitulate the known role of T during early embryonic development, thereby demonstrating the versatility of OPTiKD platforms to study the mechanisms of human development in vitro.

We then aimed to demonstrate the suitability of the OPTiKD platforms to investigate the function of genes that are not only expressed during early development, but also in differentiated cells. We focused on DPY30, a ubiquitously expressed co-factor of the COMPASS histone methyltransferase complexes required for histone H3 lysine 4 trimethylation (H3K4me3) (Jiang et al., 2011). This epigenetic modifier is necessary for mouse early embryonic development, as its knockout leads to impaired gastrulation associated with ectopic neuralization of the post-implantation epiblast (Bertero et al., 2015). Similarly, DPY30 is required for hESC pluripotency (Bertero et al., 2015), and this early role had prevented further studies of its function during differentiation. Finally, Dpy30 has been implicated in mouse ESC differentiation and in the proliferation and differentiation of hematopoietic progenitors (Jiang et al., 2011; Yang et al., 2014). Consequently, we decided to employ our inducible knockdown platform to bypass the early function of DPY30 in hPSCs and specifically suppress its expression during differentiation (Fig. 6A).

First, we generated DPY30 OPTiKD hESC sublines (Fig. S6A), and confirmed that inducible DPY30 knockdown in undifferentiated hESCs impaired the expression of pluripotency genes and triggered neuroectoderm differentiation (Fig. 6B, Fig. S6B), as shown previously (Bertero et al., 2015). We then analyzed the function of DPY30 during lineage specification by differentiating DPY30 OPTiKD hESCs into five different cell types while inducing DPY30 knockdown from the induction, specification or maturation stages (Fig. 6A). qPCR confirmed the decrease in DPY30 expression in all the cells generated (Fig. S6C-H). Interestingly, phenotypic analyses demonstrated that DPY30 knockdown from the early induction of cardiac specification impaired cardiomyocyte differentiation, as shown by the decrease in contractile markers (Fig. 6C). However, knockdown at later stages had no significant effects (Fig. 6C). A similar result was observed for the hepatocyte lineage, since decrease of DPY30 expression in endoderm progenitors led to extensive cell death at the anterior foregut stage thereby preventing further differentiation (Fig. S6I). Similarly, specification of pancreatic endocrine cells was also impaired by knockdown of DPY30 in the initial stage of differentiation (Fig. 6D). However, neither hepatocyte nor pancreatic endocrine cell specification was significantly affected by knockdown of DPY30 in maturing progenitors or differentiated cells (Fig. 6D,E, Fig. S6I). By contrast, neuronal differentiation was promoted following DPY30 knockdown during the induction of neuroepithelial progenitors (Fig. S6J). Finally, DPY30 knockdown at any stage during smooth muscle cell differentiation had no effect on the expression of key lineage markers (Fig. S6K).

Considered together, these data confirm a key role for DPY30 during germ layer specification while suggesting that the requirement for DPY30 expression could vary during the differentiation and maturation of specific lineages (Fig. 6E). Overall, these experiments illustrate how the optimized inducible knockdown platform can be easily applied to acquire novel information about developmental mechanisms by performing functional studies at different steps of hPSC differentiation into multiple cell types.

Having established an optimized inducible knockdown platform, we turned our attention to developing a complementary inducible knockout approach. Current inducible CRISPR/Cas9 methods rely on conditional overexpression of Cas9 in the presence of a constitutively expressed gRNA (González et al., 2014; Mandegar et al., 2016). In this case, control of Cas9 overexpression is achieved by a TET-ON method in which, following doxycycline treatment, a tetracycline-controlled reverse transactivator (rtTA) activates a Pol II-dependent tetracycline-responsive element (TRE) promoter (a fusion between multiple TET operons and a minimal CMV promoter). Although this TET-ON platform has been successfully applied to certain human cell types (Qin et al., 2010), we observed that this inducible system is silenced during hPSC differentiation into multiple lineages (including cardiomyocytes, hepatocytes and smooth muscle cells), even after targeting into the AAVS1 GSH (Fig. S7). These observations reinforce previous reports (Haenebalcke et al., 2013; Mandegar et al., 2016; Ordovas et al., 2015) and demonstrate that recently described systems for inducible CRISPR/Cas9 (González et al., 2014; Mandegar et al., 2016) are unlikely to work in a diversity of hPSC-derived cell types. For this reason, we explored the possibility of developing an alternative and improved method by combining a constitutively expressed CAG promoter-driven Cas9 with an inducible gRNA cassette based on that developed for inducible shRNA expression in sOPTiKD (Fig. 7A,B).

We generated hESC lines in which a fluorescent reporter gene could be knocked out in an inducible fashion (Fig. 7C). For this, we targeted ROSA26-EGFPd2 reporter hESCs (Fig. S4J,K) with both an inducible EGFP gRNA and a constitutive Cas9 in the AAVS1 locus, each transgene being integrated into one of the two alleles (Fig. 7C,D). This dual targeting approach was rapid (<2 weeks) and efficient (>90% of lines containing both transgenes; Table S1). Remarkably, when individual clonal sublines were grown in the presence of tetracycline we observed decreased EGFPd2 expression in all of the targeted lines, and EGFPd2 homozygous cells showed near-homogeneous loss of at least one copy of the reporter gene as early as 5 days following tetracycline induction (as demonstrated by 50% reduction in EGFPd2 fluorescence, Fig. S8A). Prolonged treatment with tetracycline progressively led to the complete loss of EGFPd2 fluorescence in up to 75% of EGFPd2 homozygous cells (Fig. 7E,F, Fig. S8A,B). Interestingly, co-expression of either two or three copies of the same EGFP gRNA cassette from the same AAVS1 locus was sufficient to significantly increase the speed and efficiency of inducible EGFPd2 knockout in all the clonal sublines analyzed (Fig. 7H,I, Fig. S8A). For instance, simultaneous induction of three copies of the same gRNA resulted in a remarkable 95% knockout efficiency following tetracycline treatment (Fig. 7I). Importantly, inducible EGFPd2 knockout hESCs did not show any significant decrease in the proportion of EGFPd2-positive cells nor in their fluorescence after prolonged culture in the absence of tetracycline, even when several gRNA copies were used (Fig. 7G, Fig. S8C,D). This demonstrated that the inducible gRNA expression was tightly controlled. Finally, testing of additional gRNAs against EGFPd2 revealed that the speed and efficiency of the inducible knockout strongly relied on the gRNA. Indeed, an optimal sequence allowed up to 90% knockout after only 2 days of induction (Fig. S8E,F,K). Of note, the most efficient gRNA also resulted in uncontrolled EGFPd2 knockout (Fig. S8G), but this limitation was avoided by simply adding a second TET operon to the inducible H1 promoter to ensure even more stringent transcriptional control (Fig. S8H-K).

Collectively, these results show that the sOPTiKD system could be readily repurposed to support inducible gRNA expression and allow tightly controlled activity of CRISPR/Cas9 over a broad range of gRNA potency (Fig. S8L). To the best of our knowledge, this is the first conditional CRISPR/Cas9 approach based on inducible gRNA expression. We named this method single-step optimized inducible gene knockout, or sOPTiKO.

Having demonstrated that sOPTiKO allows efficient control of CRISPR/Cas9 activity in undifferentiated hPSCs, we thoroughly tested its performance following differentiation. We differentiated homozygous EGFPd2 inducible knockout cells carrying a single copy of inducible EGFP gRNA into the three primary germ layers and into five cell types of clinical interest (neurons, cardiomyocytes, smooth muscle cells, hepatocytes and endocrine pancreatic cells). Immunostaining for lineage-specific markers demonstrated that treatment with tetracycline resulted in strong loss of EGFPd2 expression (Fig. 8A-F, Fig. S9A,B) in all these cell types. Moreover, flow cytometry quantification confirmed that inducible knockout in differentiated cells was tightly controlled and efficient (Fig. S9C-H). For example, 85% of neuronal cells and 75% of smooth muscle cells completely lost EGFPd2 expression following tetracycline treatment (Fig. S9C,F). Considered together, these results validate that sOPTiKO allows efficient control of CRISPR/Cas9 activity not only in hPSCs, but also into a large panel of mature cell types (Fig. 8G).

This report describes sOPTiKD and sOPTiKO – two novel platforms for inducible knockdown or knockout of gene expression that address the limitations of previous methods. Compared with alternative approaches that rely on viral transduction or random integration of inducible shRNAs (Lambeth and Smith, 2013; Zafarana et al., 2009), sOPTiKD is simpler to use (plasmid based), quicker (2 weeks or less to generate stable lines following a single step of gene targeting by lipofection), more efficient (>95% of the resulting cells show inducible knockdown), more scalable (isolation of clonal sublines can be entirely bypassed) and significantly more robust (due to the use of GSHs and the lack of leakiness). Furthermore, sOPTiKO shares these same advantages, thus outperforming recent inducible CRISPR/Cas9 methods that rely on the conditional expression of Cas9 (González et al., 2014) or of a fusion protein between a catalytically inactive Cas9 and the transcriptional repressor KRAB (Mandegar et al., 2016). Indeed, these systems rely on the TRE promoter, which is heavily silenced upon hPSC differentiation into multiple lineages. Furthermore, these are lengthy two-step methods, and expression of the gRNA is achieved either by transient transfection, which can be poor in efficiency, or by random integration of the gRNA, which can result in mosaic expression. Finally, whereas CRISPR interference can only efficiently control gene promoter activity (Mandegar et al., 2016), sOPTiKO allows the deletion of a broader range of genomic targets, including regions outside of promoters that might not have a direct role in transcriptional regulation. Overall, sOPTiKD/KO are the first inducible shRNA and CRISPR/Cas9 technologies that enable streamlined functional studies of multiple genetic variants in hPSCs and in a diversity of differentiated cell types (Fig. 8G).

sOPTiKD and sOPTiKO each presents distinct advantages. On the one hand, the ability to control the level of knockdown using sOPTiKD allows the study of genes for which complete loss-of-function induces cell death, and facilitates the examination of gene dosage mechanisms. On the other hand, phenotypic studies following full gene knockout using sOPTiKO are more relevant in the case of genes that are still functional even when expressed at low levels. Moreover, sOPTiKO is applicable not only to genes, but also to non-coding genomic regulatory regions, which could represent a majority of disease-associated genetic traits (Cooper and Shendure, 2011).

Aside from the examples reported in this manuscript, we envision several other potential applications of the sOPTiKD/KO technologies. With regard to cellular and developmental biology, we anticipate that sOPTiKO could efficiently accommodate variants of the Cas9 gene with catalytically inactive domains (Dominguez et al., 2015). For instance, Cas9 fusion proteins with epigenetic modifiers could allow functional validations of putative genomic regulatory regions. Similarly, sOPTiKD could be repurposed to drive other types of inducible non-coding RNAs, such as antagomir or miRNA sponges to probe microRNA function (Ebert and Sharp, 2010). Remarkably, the high targeting efficiency and scalability of sOPTiKD/KO could allow high-throughput screenings by targeting inducible shRNA or gRNA pools. Compared with viral-based approaches (Chen et al., 2012), the isogenic integration of inducible shRNAs/gRNAs would reduce heterogeneity in the targeted population, hence increasing the screening sensitivity and specificity. With regard to disease modeling applications, sOPTiKD/KO could allow the simultaneous targeting of several hiPSC lines to probe gene function in different genetic backgrounds. Such an approach could facilitate the identification of genetic disease modifiers and the discovery of novel potential drug targets in the context of personalized medicine. Multiplex inducible gene knockdown or knockout could also be used to model complex genetic disorders. Finally, sOPTiKD/KO could be easily transferred to other cell types amenable to genetic manipulation, including established cell lines and adult stem cells (Drost et al., 2015; Mandal et al., 2014), thus allowing functional studies in a multitude of systems. In conclusion, we expect that sOPTiKD/KO technologies will have a broad impact on our ability to study human development, physiology and disease.

Feeder- and serum-free hESC (H9 line; WiCell) and hiPSC (A1AT^R/R line; Rashid et al., 2010) culture and differentiation were as previously described (Vallier, 2011). Details of media compositions and protocols are provided in the supplementary Materials and Methods.

Sequences of all plasmids used in this study are provided in Appendix S1, and all cloning procedures and targeting experiments are described in detail in the supplementary Materials and Methods. Briefly, AAVS1 targeting for OPTiKD and sOPTiKD was performed by lipofection, while AAVS1 targeting for sOPTiKO was performed by nucleofection (Bertero et al., 2015; Vallier et al., 2004). Clonal lines were selected using 1 μg/ml puromycin (Sigma; for OPTiKD and sOPTiKD) or 25 μg/ml geneticin (G418 sulfate, Gibco) and 0.5 μg/ml puromycin (for sOPTiKO).

Unless otherwise described in the results or figure legends, tetracycline hydrochloride (Sigma-Aldrich) was used at 1 μg/ml to induce gene knockdown or knockout. Refer to the supplementary Materials and Methods for details on the timing of DPY30 inducible knockdown during hESC differentiation.

Quantitative real-time PCR (qPCR), western blot, flow cytometry and immunofluorescence were performed according to standard protocols as previously described (Bertero et al., 2015). Details, including the primer sequences and antibodies used, are provided in the supplementary Materials and Methods.

Statistical analyses were performed using GraphPad Prism 6. The type and number of replicates, the statistical test used, and the test results are described in the figure legends. All statistical tests employed were two-tailed. Unless stated otherwise in the figure legends, all graphical data are presented as mean±s.e.m. No experimental samples were excluded from the statistical analyses. Sample size was not pre-determined through power calculations, and no randomization or investigator blinding approaches were implemented during the experiments and data analyses. When representative results are presented, the experiments were reproduced in at least two independent cultures.

We thank Dr Kosuke Yusa for providing the AAVS1 ZFN plasmids and Dr Andrew Knights for technical support.