Main

Designer nucleases such as zinc-finger nucleases (ZFNs), TAL effector nucleases (TALENs), and clustered regularly interspaced short palindromic repeats (CRISPR)–Cas systems can greatly simplify the creation of tailored genetic modifications1,2,3. Despite its apparent simplicity, reprograming the CRISPR apparatus to consistently achieve high levels of gene editing remains a complex task. In part, this difficulty is due to the current limited prediction accuracy of computational guide-RNA selection algorithms4,5. In addition, cell type and cell-cycle stage play a major role in determining the fate of genome editing, because human cells dictate repair outcome and have a preference for NHEJ over HDR6. Thus, identifying active guide RNAs, screening, and isolating clones with desired genetic modifications can be time consuming and costly. In addition, achieving high levels of gene editing for accurate phenotypic analysis within bulk cell populations remains challenging.

At its most basic level, higher genome-editing frequencies are associated with higher nuclease levels and activity as well as efficient delivery. Consequently, several approaches have been implemented to capture and isolate these subpopulations of cells (Supplementary Note). Other strategies have focused on altering cell-cycle parameters, the timing of nuclease expression, chemical inhibition of NHEJ, and the use of HDR agonists7,8,9,10,11,12. These strategies have produced promising results, but their general applicability and the absence of negative effects on genome integrity must be thoroughly evaluated.

Conceptually distinct genetic approaches based on the creation of gain-of-function alleles have been developed in the worm Caenorhabditis elegans. These methods, termed 'coconversion' or 'co-CRISPR', markedly increase the odds of detecting a phenotypically silent targeted mutation through the simultaneous coconversion of a mutation in an unrelated target that causes a visible phenotype13,14,15,16. It has been found that simultaneous introduction of single-guide RNAs (sgRNAs) to two different endogenous loci results in double-editing events that are not statistically independent, and this concept has also been adapted for use in Drosophila17,18.

A related approach has been described to isolate human cells with NHEJ-driven mutations by cotargeting the X-linked hypoxanthine phosphoribosyl-transferase (HPRT1) gene19,20. The technique uses a mutagenic chemotherapy drug, does not select for HDR-based events, and may affect the salvage pathway of purines from degraded DNA20. More recently, it has been found that CRISPR–Cas9-mediated insertion of a drug-selectable marker at one control site frequently coincides with the insertion at an unlinked and independently targeted site in mammalian cells21,22. Nonetheless, the insertion of a heterologous selection cassette into the genome of the edited cells may limit application of this technique. Hence, further refinement to these approaches is needed.

Here, we devised a robust coselection strategy by generating dominant cellular resistance to ouabain, a highly potent plant-derived inhibitor of the ubiquitous and essential Na+/K+ ATPase23,24. Using CRISPR–Cas9 and CRISPR-Cpf1, we generated gain-of-function alleles to coselect for mechanistically related editing events at a second locus of interest. This strategy is portable to many guide RNAs, in a manner independent of cell type, and provides a general solution for facilitating the isolation of genome-edited human cells.

Results

Editing ATP1A1 locus confers resistance to ouabain

The Na+/K+ ATPase (also known as the sodium/potassium pump), encoded by the ATP1A1 gene, is responsible for maintenance of the electrochemical gradients of Na+ and K+ across the plasma membranes of animal cells. Cardiotonic steroids (CTSs), such as ouabain (PubChem CID 439501), constitute a broad class of specific inhibitors of this enzyme and have been prescribed for congestive heart failure for more than 200 years (ref. 24). Their mechanism of action has been defined through crystallography and mutagenesis studies, which have led to the identification of inhibitor-resistant enzymes23,24,25,26. Most prominently, replacement of the border residues (Q118 and N129) of the first extracellular loop with charged amino acids generates highly resistant enzymes when ATP1A1 is overexpressed in cells26 (Fig. 1a). However, it is unknown whether deletions in this region can disrupt ouabain binding while preserving the functionality of ATP1A1 and whether modification of the endogenous locus, as opposed to overexpression of the mutant enzyme, results in cellular resistance to ouabain. Therefore, we tested whether targeting this region with the CRISPR–Cas9 system might induce such a phenotype.

Figure 1: NHEJ-driven editing at ATP1A1 induces cellular resistance to ouabain.
figure 1

(a) Schematic representation of SpCas9-target sites surrounding DNA encoding the first extracellular loop of ATP1A1. The positions of residues Q118 and N129, exon–intron boundary, PAM and four potential target sequences are shown. (b) K562 cells stably expressing SpCas9 were transfected with the indicated sgRNA-expression vectors (500 ng), and the Surveyor assay was performed 10 d later to determine the frequency of indels, as indicated at the base of each lane. Where indicated, cells were treated with 0.5 μM ouabain for 7 d starting 3 d after transfection. An expression vector encoding EGFP (−) was used as a negative control. (c) Amplicons from b were TOPO cloned and sequenced.

Full size image

First, we identified two active sgRNAs targeting Streptococcus pyogenes Cas9 (SpCas9) to the exon encoding the first extracellular loop (hereafter named G2 and G4) and one sgRNA targeting the adjacent intron (hereafter named G3) (Fig. 1a,b). Then we observed that active nucleases targeting the coding sequence (G2 and G4) induced cellular resistance to ouabain, whereas cells cleaved in the intron (G3) died within 48 h (Fig. 1a,b). We did not observe any spontaneous resistance to ouabain treatment, in agreement with findings from previous reports25,26. Because ATP1A1 is essential for cell survival, these observations suggest that in-frame insertions and deletions (indels) were created in the first extracellular loop, thus preventing ouabain binding without blocking enzymatic activity. To assess the spectrum and frequency of targeted mutations generated in these pools of cells, we used the Tracking of Indels by Decomposition (TIDE) method27. This analysis revealed that in-frame deletions were selected for over time and after ouabain treatment and that G2 generated a much more diverse set of mutations that correlated with improved growth (Supplementary Fig. 1). The analysis also showed that the Surveyor nuclease assay used to determine the frequency of indels characteristic of imprecise double-strand-break (DSB) repair by NHEJ (hereafter, the term NHEJ will be used to describe mutagenic repair, because the precise religation of DSBs cannot be detected with this assay) saturated when samples with high levels of modification were tested (Fig. 1b and Supplementary Fig. 1). Cloning and sequencing of ATP1A1 alleles from ouabain-resistant cells identified in-frame deletion products that disrupted the first extracellular loop of the pump (Fig. 1c).

Next, we sought to test whether gain-of-function alleles could be created by HDR. Reaching a high threshold of HDR in human cells is a major challenge in the genome-editing field because, at the population level, cells favor DSB repair via NHEJ over HDR6. Therefore, cleaving within the coding sequence of ATP1A1 would disfavor the recovery of cells edited through HDR at the expense of cells mutated via NHEJ, because ouabain would select for both types of repair events. To achieve selection exclusively via HDR-driven events, we took advantage of sgRNA G3, which targets SpCas9 to the intron (Fig. 1a,b). Two single-stranded oligonucleotides (ssODNs) were designed to create ATP1A1 alleles conferring ouabain resistance26. The ssODN donors created the double replacements Q118R N129D (RD) and Q118D N129R (DR), destroyed the protospacer-adjacent motif (PAM), and included additional silent mutations to create restriction sites to facilitate genotyping (Supplementary Fig. 2). Cas9-expressing K562 cells were cotransfected with sgRNA G3 along with ssODNs, and growth was monitored after the addition of ouabain. Cells survived and grew robustly only in the presence of the nuclease and either donor. Restriction fragment length polymorphisms (RFLP) assays confirmed the introduction of the desired sequence changes and their enrichment after ouabain treatment (Supplementary Fig. 2). In addition, increasing the dose of ouabain selected for the double mutants within the population (Supplementary Fig. 2). This result was not entirely unexpected, because single mutations at either position confer an intermediate level of resistance26. We speculated that it might be possible to select for cells with longer gene-conversion tracts by increasing the dose of ouabain during selection28. Titration of ouabain in the culture medium indicated that cells modified through HDR were resistant to a concentration of the drug of at least 1 mM, which is more than 100-fold higher than that for NHEJ-induced mutations and more than 2,000-fold higher than that needed to kill the cells (0.5 μM), thus highlighting the wide range of doses that can be used for selection. This positive selection was also observed in U2OS, HEK293, and diploid hTERT-RPE1 cells (Supplementary Fig. 3 and below). Notably, in selected cells with more than two copies of ATP1A1, the fraction of edited alleles could be lower than 50%. For example, in a triploid cell line, the minimal expected signal for these dominant gain-of-function mutations was 33%. The above results were reproduced by using the type V CRISPR system from Acidaminococcus sp. Cpf1 (AsCpf1), a single-RNA-guided (crRNA) enzyme that recognizes a TTTV-sequence (where V is any base but T) PAM and produces cohesive DSBs29,30 (Supplementary Fig. 3). Thus, we identified highly active CRISPR–Cas9 and CRISPR–Cpf1 RNA-guided nucleases capable of producing gain-of-function alleles at the ATP1A1 locus via either NHEJ or HDR.

Coselection enriches for CRISPR-induced indels

To test whether selection for gain-of-function alleles in ATP1A1 could result in coenrichment of NHEJ-driven mutations at a second locus, we constructed an all-in-one vector containing tandem U6-driven sgRNA-expression cassettes along with CBh-driven high-specificity eSpCas9(1.1)31 (Supplementary Fig. 4). Cells were transfected with a vector for targeting both EMX1 and ATP1A1 and either selected with ouabain or left untreated. The frequencies and spectrum of indels, as determined by TIDE and Surveyor assays, revealed a marked increase in gene disruption after selection (Table 1, Supplementary Fig. 4 and Supplementary Data 1). Similar results were obtained when the AAVS1 and ATP1A1 loci were cotargeted (Table 1, Supplementary Fig. 4 and Supplementary Data 1). These data illustrate the effects of the initial modification rate at ATP1A1 on the overall coselection process. In addition, despite observing nearly saturating on-target disruption rates, we were unable to detect activity at known off-target sites for both EMX1 and AAVS1 in these stably modified cell populations31,32 (Supplementary Fig. 4). Thus, the coselection process does not negatively affect the enhanced specificity of the eSpCas9(1.1) variant.

Table 1 Coselection for CRISPR-induced indels in cell populations
Full size table

Next, we took advantage of the multiplexing capacity of the CRISPR–Cpf1 system to perform coselections by using all-in-one AsCpf1-expression vectors containing crRNA arrays33. We observed improvements in gene-disruption efficiency at four target sites expressed either in pairs with the ATP1A1-targeting guide or as a full array simultaneously coexpressing five guides (Table 1, Supplementary Fig. 5, Supplementary Table 1 and Supplementary Data 1). As previously observed for SpCas9, the pattern of DNA repair after eSpCas9(1.1) and AsCpf1 cutting at each site was nonrandom, consistent across cell lines and independent of absolute efficacy34 (Supplementary Data 1). Although we observed off-target cleavage for the DNMT1-targeting guide in transiently transfected K562 cells, we were unable to detect mutagenesis in ouabain-selected cells, even though the on-target activity was superior35 (Supplementary Fig. 6). In contrast, in HEK293, off-target activity was low but apparent (Supplementary Fig. 6). Thus, it appears that the coselection process did not result in overt off-target activity. Collectively, these data indicated that CRISPR-driven gain-of-function mutations at the endogenous ATP1A1 gene can be used efficiently for coselection via NHEJ.

Robust coselection of cells with CRISPR-driven HDR events

We then tested whether selection of cells with a CRISPR-driven HDR event at ATP1A1 could substantially enrich for correctly targeted cells at a second locus. We targeted two endogenous genes to generate N- and C-terminal fusions with fluorescent proteins to facilitate the quantification of HDR events through FACS-based analysis. First, we inserted the coding sequence for the green fluorescent protein Clover or the red fluorescent protein mRuby2 after the second codon of the LMNA gene, which encodes the lamin A and lamin C isoforms12 (Supplementary Fig. 7). For both Clover and mRuby2 donors, we detected a marked increase in signal ranging from ∼5–6% to ∼40–50% after ouabain selection, and the cells displayed the distinct localization pattern of fluorescence enriched at the nuclear periphery (Fig. 2 and Supplementary Fig. 7). Cotransfection of the Clover and mRuby2 donors along with the LMNA and ATP1A1-targeting nucleases allowed us to visualize the enrichment of double-positive cells, thus demonstrating that biallelic targeting can be achieved after ouabain selection (Fig. 2 and Supplementary Fig. 7). The level of improvement in gene targeting at the coselected LMNA locus paralleled HDR rates at ATP1A1, as determined by RFLP assays (Supplementary Fig. 7). Coselection was efficient for wild-type (WT) SpCas9, eSpCas9(1.1), and AsCpf1 in this system (Supplementary Fig. 7).

Figure 2: Selection for targeted integration at LMNA by coediting ATP1A1 via HDR.
figure 2

FACS-based quantification of Clover and mRuby2 targeting to the N terminus of lamin A/C in K562 cells cotransfected with nucleases and donors targeting LMNA and ATP1A1. Where indicated, cells were treated with 0.5 μM ouabain for 10 d starting 3 d after transfection.

Full size image

To determine whether the enrichment occurred solely for alleles that repaired the DSB via HDR, or whether NHEJ-produced alleles were also enriched, we performed out-out PCR analysis on ouabain-selected samples and subsequent TOPO cloning and sequencing. Sequencing of 44 nontargeted LMNA alleles revealed seven WT sequences and 37 alleles with indels at the predicted cleavage site (Supplementary Fig. 8). Thus, NHEJ-produced alleles were also enriched, but a fraction of the cells were likely to have both a targeted and a WT allele. Similarly, sequencing of ATP1A1 alleles revealed 10 WT, 35 NHEJ, and 39 HDR-related events out of 84 reads (Supplementary Fig. 8). Among HDR events, directional coconversion of SNPs from the DSB was evident28. All 39 clones had incorporated the ClaI site (mutated PAM), 30 had incorporated both the ClaI and the BmgBI sites, and 17 had integrated the three RFLP sites (Supplementary Fig. 2).

To label chromatin, the HIST1H2BK locus was targeted to create a C-terminal fusion of H2B with monomeric Azami-Green (mAG1). For both WT and eSpCas9(1.1), the fraction of cells expressing the fusion protein increased from below 1% to ∼13–15% after ouabain treatment (Supplementary Fig. 9). The absence of promoter elements in the homology arms of the donor vector along with the clear chromatin-linked fluorescent signal suggested that the process enriched for correctly targeted cells (Supplementary Fig. 9). Finally, stimulation of targeted integration of transgene cassettes was also successful at AAVS1 and HPRT1 (Supplementary Fig. 9). Together, these data demonstrated that coselection for ouabain-resistant cells markedly improved the outcome of HDR-driven gene-editing experiments, irrespective of the target loci.

Enabling TAP tagging and affinity purification of endogenous protein complexes from coselected polyclonal cell populations

We next tested whether functional assays could be directly performed in modified cell pools, thus bypassing single-cell cloning steps. Using coselection, we tagged the enhancer of polycomb homolog 1 (EPC1) and the E1A-binding protein p400 (EP400), two essential subunits of the NuA4–TIP60 acetyltransferase complex that promote homologous recombination by regulating 53BP1-dependent repair36,37 (Supplementary Fig. 10). Out-out PCR-based assays and western blotting confirmed the correct integration of the affinity tag at both loci and the enrichment of tagged cells after ouabain treatment (Fig. 3a,b). Tandem affinity purification (TAP) from the cell pools yielded protein complexes virtually identical to those obtained from clonal cell lines37 (Fig. 3c). These results represent an additional step toward high-throughput genome-scale purification of native endogenous protein complexes in human cells.

Figure 3: Endogenous TAP tagging and purification of the NuA4–TIP60 complex from coselected cell pools.
figure 3

(a) K562 cells cotransfected with nucleases and donors targeting EPC1 or EP400 and ATP1A1 were treated with 0.5 μM ouabain or left untreated for 10 d starting 3 d after transfection. Targeted integration of the tag sequence was assayed by PCR, by using primers binding outside of the homology arms and designed to yield a longer PCR product if the tag is inserted. (b) Western blot analysis of whole cell extracts. Anti-FLAG M2 was used to detect tagged proteins, and anti-α-tubulin was used as a loading control. (c) Silver-stained SDS–PAGE gels showing the purified EPC1 and EP400 complexes. Mock purifications were performed on WT K562 nuclear extracts. MW, molecular weight.

Full size image

Efficient enrichment of gene-edited human hematopoietic stem and progenitor cells

To begin to explore the potential for the clinical translation of our method, we tested the coselection strategy during ex vivo expansion of cord-blood-derived human hematopoietic stem and progenitor cells (HSPCs). We used previously developed tools to introduce a mutation in the beta-globin (HBB) gene, which causes sickle-cell disease38,39,40 (Fig. 4a). Purified human CD34+ cells were electroporated with preformed SpCas9 ribonucleoprotein complexes (RNPs) containing synthetic crRNAs and transactivating crRNAs (tracrRNAs) along with ssODNs, and were expanded ex vivo with or without ouabain (Supplementary Fig. 11). Cells were cultivated in the presence of UM171 to promote expansion and maintenance of primitive progenitors during selection41. RFLP-based assays clearly indicated that cells edited at HBB were efficiently enriched by ouabain treatment, and this phenomenon was observed in cells isolated from various donors and with the use of different ssODNs (Fig. 4b–d and Supplementary Fig. 11). Although the delivery and overall efficacy require improvement, these results suggest that the process can be adapted and tested in preclinical settings. Critically, these data demonstrate that the procedure is applicable to diploid primary cells.

Figure 4: Introduction of the sickle mutation in primary human cord blood (CB) CD34+ cells by coediting ATP1A1 via HDR.
figure 4

(a) Schematic representation of the SpCas9 target site in HBB and predicted HDR outcome. The positions of the E6 residue, 5′ untranslated region (UTR), PAM, and novel restriction sites to monitor the insertion of ssODN-specified mutations are shown. (b) Cultured CD34+ cells were electroporated with ATP1A1 and HBB RNPs along with HBB #1 and ATP1A1 G4 RD ssODNs, and treated as shown in Supplementary Figure 11. Genomic DNA was harvested at the indicated time points, and a PstI RFLP assay was used to determine the frequency of HDR at HBB. (c) As in b, but with PshAI. (d) As in b, but with the Surveyor assay to determine the total frequency of edited alleles (NHEJ + HDR).

Full size image

Discussion

Our results corroborate earlier observations that cells proficient at completing one genomic manipulation have an increased probability of completing a second independent genomic manipulation, provided that the manipulations are mediated by sufficiently similar mechanisms of DNA repair14,15,21. A defining aspect of our system is that manipulations can be initiated independently through NHEJ- or HDR-driven events. The extent of enrichment for edited cells varies at different sites and according to the type of modification created at the locus of interest. Crucially, robust selection is achieved without the use of exogenous DNA markers, thus making this technique potentially compatible with therapeutic applications. The well-defined mechanism of action of ouabain acting on a nonsignaling ion pump independently of proliferation is another distinct advantage of the method. Ouabain treatment kills cells within 48 h of exposure, and targeted cells display no apparent growth delay resulting from the selection process. Accordingly, the point mutations engineered to confer ouabain resistance are naturally occurring in metazoans42. In addition, these mutant enzymes function normally, as shown by 86Rb+ uptake and ATP hydrolysis assays26. The turnover of ATP1A1 at the plasma membrane appears to be rapid, because ouabain can be added to the culture medium as early as 15 h after transfection of the CRISPR components, even if they are encoded by plasmids (data not shown).

Remarkably, coselection does not appear to yield overt off-target activity. Because the amount of CRISPR components can be titrated to improve the ratio of on-target to off-target mutation rates, an unexpected benefit may be that this balance can be achieved without decreasing efficacy43; this will hold true if transfection of lower amounts of plasmids results in lower levels of the nuclease on a per-cell basis. However, translocations between ATP1A1 and the locus of interest will occur at some frequency, and there is a slight probability that this event might be found in the selected populations44.

Coselection schemes may allow for efficient correction of a mutation 'at distance' despite the preponderance of relatively short gene-conversion tracts occurring in mammalian cells28. Because 80% of human exons are <200 bp in length, it may be feasible to cleave within an intron to seamlessly induce recombination in the juxtaposed exon. This approach might prevent complications caused by on-target mutagenesis of uncorrected alleles.

Methods

Additional methodology.

A step-by-step protocol is available as a Supplementary Protocol and an open resource in Protocol Exchange45.

Cell culture and transfection.

K562 cells were obtained from the ATCC (CCL-243) and maintained at 37 °C under 5% CO2 in RPMI medium supplemented with 10% FBS, penicillin–streptomycin and GlutaMAX. U2OS cells were obtained from the ATCC (HTB-96) and maintained at 37 °C under 5% CO2 in McCoy's 5A medium supplemented with 10% FBS, penicillin–streptomycin and GlutaMAX. HEK-293LTV (LTV-100) cells were purchased from Cell Biolabs, and hTERT RPE-1 cells (authenticated by STR analysis) were a kind gift from A. Fradet-Turcotte. Both cell lines were maintained at 37 °C under 5% CO2 in DMEM supplemented with 10% FBS, penicillin–streptomycin and GlutaMAX. All cell lines were tested for the absence of mycoplasma contamination. Cells (2 × 105 per transfection) were transfected with an Amaxa 4D-Nucleofector (Lonza), per the manufacturer's recommendations. Transfection conditions used in coselection experiments are detailed in Supplementary Table 2. Ouabain octahydrate (Sigma) was dissolved at 50 mg/ml in hot water and stored at –20 °C. Working dilutions were prepared in water and added directly to the culture medium.

Genome editing vectors and reagents.

All eSpCas9(1.1) and AsCpf1 vectors generated in this study for targeting ATP1A1 are available from Addgene (Supplementary Fig. 12). The expression cassette for hCas9 (ref. 46) (Addgene plasmid no. 41815, a gift from G. Church) was transferred to AAVS1_Puro_PGK1_3×FLAG_Twin_Strep37 (Addgene plasmid 68375) to establish the K562 cell line constitutively expressing SpCas9 under the control of a CAG promoter. All sgRNA-expression vectors were built on the SP_gRNA_pUC19 (ref. 37) backbone (Addgene plasmid 79892), with the exception of gRNA_AAVS1-T2 (ref. 46), which was a gift from G. Church (Addgene plasmid 41818). sgRNAs were designed with the GPP Web Portal (http://portals.broadinstitute.org/gpp/public/) and the MIT web-based CRISPR design tool (http://crispr.mit.edu/). Guide sequences are provided in Supplementary Table 3. When required, DNA sequences for the guides were modified at position 1 to encode a G, owing to the transcription-initiation requirement of the human U6 promoter. Reagents for LMNA targeting were as previously described12. To test the high-specificity eSpCas9(1.1)31 variant, guide sequences were cloned into the eSpCas9(1.1)_No_FLAG vector (Addgene plasmid 79877). The human-codon-optimized AsCpf1 ORF from pY010 (ref. 29) (Addgene plasmid 69982, a gift from F. Zhang) was transferred to AAVS1_Puro_PGK1_3×FLAG_Twin_Strep37 (Addgene plasmid 68375) to establish the K562 cell line constitutively expressing AsCpf1 from a CAG promoter. A crRNA-expression vector was built on the pUC19 backbone by linking the AsCpf1 5′ DR to a human U6 promoter29. Guides for AsCpf1 were as previously described33 or were designed with Benchling (https://benchling.com/academic), and sequences are provided in Supplementary Table 4. The dual AsCpf1 and U6-driven crRNA-array-expression vectors were built on pY036 (a gift from F. Zhang) as previously described33. Desalted ssODNs (Supplementary Table 5) were synthesized as ultramers (IDT) at a 4 nmol scale. All plasmid donor sequences contained short homology arms (<1 kb) and were modified to prevent their cleavage by Cas9, as previously described37. The only exception to this rule was for the targeting of Clover to the LMNA locus with AsCpf1.

Surveyor nuclease, RFLP knock in, out-out PCR, and TIDE analysis.

Genomic DNA from 2.5 × 105 cells was extracted with 250 μl of QuickExtract DNA extraction solution (Epicentre), per the manufacturer's recommendations. The various loci were amplified with 30 cycles of PCR with the primers described in Supplementary Table 6. Assays were performed with a Surveyor mutation detection kit (Transgenomics) as previously described47. Samples were separated on 10% PAGE gels in TBE buffer. For RFLP assays, the PCR products were purified and digested with the corresponding enzyme and resolved by 10% PAGE. Gels were imaged with a ChemiDoc MP (Bio-Rad) system, and quantifications were performed with Image Lab software (Bio-Rad). TIDE analysis was performed with a significance cutoff value for decomposition of P <0.005 (ref. 27). To analyze targeted integration via out-out PCR, genomic DNA extracted with QuickExtract DNA extraction solution was subjected to 30 cycles of PCR for LMNA and 35 cycles for EPC1 and EP400 with the primers described in Supplementary Table 7. Amplicons were loaded on 1% agarose gels in TAE buffer.

Flow cytometry.

The frequency of cells expressing EGFP, mAG1, Clover, and mRuby2 was assessed with a BD LSR II flow cytometer and gated for viable cells with 7-aminoactinomycin D (7-AAD).

TAP.

Nuclear extracts and purifications were performed from ∼5 × 108 cells, as previously described37,48. Approximately 1/30 of the final eluates was loaded, without prior precipitation, on Bolt 4–12% Bis-Tris gels (Life Technologies) and was run for 45 min at 200 V in MOPS buffer. Silver staining was performed with a SilverQuest kit (Life Technologies). Anti-FLAG M2 (Sigma, A8592) and anti-α-tubulin (DM1A, sc-32293) were used for western blot analysis.

Human cord blood (CB) CD34+ cell collection and processing.

Human umbilical CB samples were collected from donors who had provided informed consent, and procedures were approved by the Research Ethics Board of the CHU de Québec–Université Laval. Mononuclear cells were first isolated through Ficoll-Paque Plus density centrifugation (GE Healthcare). Human CD34+ hematopoietic stem and progenitor cells (HSPCs) were isolated through CD34+ selection according to the manufacturer's instructions (EasySep, StemCell Technologies). Purified CD34+ cells were cryopreserved in Cryostor CS10 (StemCell Technologies).

HSPC culture, editing and selection.

CD34+ HSPCs were thawed and cultured in StemSpan ACF (StemCell Technologies) supplemented with 100 ng/ml SCF (Feldan), 100 ng/ml FLT3-l (Peprotech), 50 ng/ml TPO (StemCell Technologies), 10 μg/ml LDL (StemCell Technologies), and 35 nM UM171 (StemCell Technologies) for 16–24 h after thawing. Cells were then nucleofected with Cas9 RNP and ssODNs with an Amaxa 4D Nucleofector X unit (Lonza) and the E0-100 program, according to the manufacturer's recommendations. After nucleofection, cells were incubated at 30 °C for 16 h and then transferred at 37 °C for 2 d. From day 3 after nucleofection, cells were transferred to StemSpan ACF (StemCell Technologies) containing 1× StemSpan CD34+ Expansion supplement (StemCell Technologies) and UM171. Ouabain (0.5 μM) was added to the cells during the culture-medium change at day 5 after nucleofection, replenished every 3 or 4 d, and maintained for 8 d before RFLP analysis at day 14 after thawing.

Synthetic crRNA, tracrRNA, and rRNP complexes for editing in HSPCs.

The Alt-R CRISPR system (Integrated DNA Technologies) was used to produce the HBB38,39,40 (CTTGCCCCACAGGGCAGTAA) and ATP1A1 G4 (GTTCCTCTTCTGTAGCAGCT) guides. crRNA and tracrRNA were first resuspended to 200 μM stock solutions in Nuclease-Free IDTE Buffer (Integrated DNA Technologies). For formation of crRNA–tracrRNA complexes, the two RNA oligonucleotides were mixed in equimolar concentrations, heated at 95 °C for 5 min and immediately transferred to ice. Immediately before nucleofection, 50 pmol of Cas9 protein (Integrated DNA Technologies) was incubated with 125 pmol of crRNA–tracrRNA complexes at RT for 10 min to form the RNP complexes along with 100 pmol of each ssODNs. Notably, the ATP1A1 G3 guide was not active in the Alt-R CRISPR system (Integrated DNA Technologies); therefore, we used the ATP1A1 G4 guide to generate RNPs. This guide generated low levels of ouabain resistance via NHEJ in HSPCs, and robust cell growth was observed only in the presence of ssODN donors.

Data availability.

All data supporting the findings of this study are available in the Supplementary Information files.

Additional information

Publisher's note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.