facilitate assessment of the effectiveness of double transgenic selection, we cloned markertrons onto lentiviral vectors expressing TagBFP or mCherry fluorescent proteins as test transgenes (Fig. 2a). Viral preparations were transduced into U2OS cells, which were then split into replicate plates with non-selective or selective media. Following appropriate passages for antibiotics selection, the two cell cultures were analyzed by flow cytometry. For Hygromycin resistance (HygroR) gene, one “native” SspDnaB split point (SspDnaB-200 = G200:S201; Plasmid pair 5,6) with flanking residues “GS” and one “native” NpuDnaE split point (NpuDnaE-89 = Y89:C90; Plasmid pair 3, 4) with “YC” residues were tested (Supplementary Fig. 1a). Both enabled successful selection when both N- and C-markertrons were transduced yielding >95% BFP + mCherry + double transgenic cells in selected cultures compared to <40% double-positive cells in nonselected culture (Fig. 2b; Plasmid pairs 3, 4 and 5, 6). Cells transduced with either of the single markertrons did not survive Hygromycin selection. In contrast, double transgenesis with conventional full-length non-split HygroR vectors only allowed for ~20% enrichment of BFP + mCherry + cells (Plasmid pairs 97,98) at lower titers and for up to ~50% at higher titers. We screened three additional potential split points (NpuDnaE-52 = 52 S:53 C; Plasmid pair 7,8), (NpuDnaE-240 = 240 A:241 C; Plasmid pair 9,10), and (NpuDnaE-292 = 292 R:293 C; Plasmid pair 11,12) for NpuDnaE with the obligatory cysteine residue on the C-extein junction and a residue on the N-extein junction reported to support substantial trans-splicing activities7. We also tested six additional NpuDnaE split points (NpuDnaE-69, 131, 171, 218, 259, and 277) by inserting an “artificial” cysteine on the C-extein junction to support splicing at ectopic sites yielding additional split points. In total, eight out of eleven split points tested supported Hygromycin selection (Fig. 2b). Two of the Hygro Intres designs (NpuDnaE-131 and 292) failed to provide resistance in two of the four replicate experiments at lower titers, while three designs (NpuDnaE-218, 259, and 277) failed to provide resistance in any experiments. These positions may reside within less efficient splicing sequence and structural contexts or may disrupt folding of the HygroR protein upon reconstitution. Indeed, western blot analysis using terminally tagged markertron fragments revealed that among split points 52, 68, 89, 131, and 171, trans-splicing is least active at split point 131 (Supplementary Fig. 1b, lane 5). This is consistent with its failure to consistently provide resistance at a lower titer (Fig. 2b). In addition, the insertion of the artificial cysteine at the NpuDnaE-69, 131, and 171 C-markertrons is required for protein splicing mediated by NpuDnaE intein at these positions (Compare lanes 2/3, 5/6, and 7/8), consistent with a well-established requirement7. Nonetheless, the six successful designs validate our screening strategy and demonstrate that HygroR is amenable to splitting at different positions spanning a large portion of the protein. Similarly, for Puromycin resistance (PuroR) (Fig. 3a), Neomycin/ G418 resistance (NeoR) (Fig. 3b) or Blasticidin resistance (BlastR) (Fig. 3c) genes, we identified four, two, and one functional 2-split Intres pair(s), respectively. In all of these cases, cells transduced with single markertrons did not survive selection, while cells transduced with both yielded 88–100% double transgenic cells in selective cultures compared to <50% in non-selective cultures. Details of the split points of Intres genes and plasmids are presented in Supplementary Figs. 1–4 and Supplementary Table 1. To facilitate adoption of Intres markers, we created Gatewaycompatible lentiviral vectors for convenient restriction-ligationindependent LR clonase recombination of transgenes8 (Supplementary Fig. 5). We tested the functionality of these vectors by recombining TagBFP and mCherry, respectively to the N- and CIntres vectors and found robust selection of double transgenic cells (Supplementary Fig. 5b). One potential utility of Intres a 5LTR 5LTR

100 + rry ) e su hCm irvL 50 +FBP (25μ % + rry ) he irsu +FBPCm (v100Lμ 50 % s e t a c il p e

R 0

D N Key n .. s +Hygro

ND = Experiment not performed X = No survival vectors is to install different fluorescent markers in cells to label different cellular compartments. To explore this application, we cloned in NLS-GFP and LifeAct-mScarlet9, which label nucleus and F-actin, respectively, by Gateway recombination to conventional full-length (FL) non-split Hygromycin selectable vectors or 2-split Hygromycin Intres vectors. We transduced cells with either sets of plasmids and subjected them to antibiotic selection (Supplementary Fig. 5c). The sample transduced with non-split selectable plasmids contained both singly and doubly labeled cells, while cell transduced with Intres plasmids were all doubly labeled.

Split mScarlet fluorescent genes for double transgenesis. To test whether split fluorescent markers can be used for transgene selection, we screened for NpuDnaE split points in the mScarlet fluorescent protein (Supplementary Figs. 6 and 7a) and identified four split points allowing for >96% enrichment of double transgenic cells and three other split points enabling >60% enrichment of double transgenic cells in an mScarlet-gated population, compared to <20% double transgenic cells in non-gated population (Supplementary Fig. 7b, c).

Three-split Hygromycin Intres for triple transgenesis. With the split points identified for 2-markertron Intres genes, we set out to engineer higher degree split markers. We tested combinations of splits points to partition a marker gene into three or more markertrons to allow for co-selection of more than two “unlinked” transgenes with one antibiotic (Fig. 4a, b). To identify pairs of split points that would allow such an “Intres chain”, we cloned 3-split markertrons into three lentiviral vectors each carrying one of three fluorescent transgenes TagBFP, EGFP, or mCherry, that will allow us to assess effectiveness of selection by flow cytometry (Fig. 4c). Since the Hygromycin resistance gene is the longest and provides the most split points for testing, we focused on engineering 3-split Hygromycin Intres. We tested two 3-split Hygromycin Intres using two intervening NpuDnaE

BlastR 57 b

NeoR inteins (i.e., homogeneous intein), two using NpuDnaE for the first intein and SspDnaB for the second intein, as well as two using SspDnaB for the first intein and NpuDnaE for the second intein (i.e., heterogeneous “orthogonal” inteins) (Fig. 4d). The four heterogeneous-intein 3-split Hygromycin Intres enabled 95–100% triple transgenic selection and the two homogeneousintein Hygro Intres enabled 74–99% triple transgenic selection in Hygromycin-selected cultures compared to <20% in non-selected cultures. Samples with “leave-one-out” transduction did not yield any viable cells after Hygromycin selection while cells transduced with non-split Hygromycin vectors yielded only 7–17% triple transgenic cells after selection. The observation that 3-split Intres designs using two orthogonal inteins yielded more consistent results than those using the same inteins for the two split points suggest that the use of the same inteins for joining multiple split points may result in artifacts caused by combinatorial splicing that generates “misjoined” fragments. To facilitate the use of 3split Intres, we created Gateway compatible lentiviral vectors with three of the 3-split Hygromycin Intres (Supplementary Fig. 8a). Three sets of these vectors were each tested by recombining TagBFP (as transgene 1), EGFP (as transgene 2) and mCherry (as transgene 3) into the N-, M-, and C-Intres Gateway destination vectors. Lentiviruses derived from the resultant vectors were used to transduce U2OS cells, which were then split into Hygromycin selective or non-selective media (Supplementary Fig. 8b). Two weeks after selection, cells were analyzed by flow cytometry. All three sets of 3-split Hygromycin Intres plasmids support triple transgenic cell selection of >97% compared to <40% in the nonselected cultures (Supplementary Fig. 8c).

Application of Intres in CRISPR-Cas-mediated knock-in. Another potential application of split selectable markers is to facilitate genome engineering and editing via the CRISPR-Cas system10. Although gene knockout based on NHEJ-mediated insertions/deletions (indels) occurs at high frequency, precise editing and knock-in based on homology directed repair (HDR) using exogenous repair templates are inefficient11. We tested whether split selectable markers can be used to select for cells with CRISPR-mediated biallelic knock-in at the AAVS1 locus12. We constructed targeting constructs with homology arms flanking the target site, and splice acceptor-2A peptide to trap the markertrons within intron one of the host gene PPP1R12C. However, we did not obtain any live cells after CRISPR-Cas knock-in experiments in HEK293T cells using these targeting constructs and two weeks of antibiotic selection. We suspected that the endogenous promoter of the host gene PPP1R12C might not drive sufficient expression of markertrons to reconstitute enough antibiotic resistance protein to counter the antibiotic. We thus tested an alternative strategy to express Intres markertrons using the TetO promoter which allows activity to be tuned by doxycycline (dox). To allow comparison of Intres-mediated biallelic selection versus full-length (FL) non-split selectable markers, we implemented several different targeting construct designs. First, we drove expression of a full-length (FL) resistance gene (e.g., Hygro) together with rtTA under a constitutive EF1a promoter and a separate test Intres (e.g., Blast Intres) under a dox-inducible TetO promoter (Supplementary Fig. 9b, Plasmids 109 and 110). This allows comparison of full-length and split selectable markers within the same constructs. To allow valid comparison of full-length versus split markers driven by the same TetO promoter, we constructed two similar plasmids 107 and 108 (cf. Plasmids 109 and 110), wherein the full-length antibiotic resistance gene (Blast) is placed downstream of the TetO promoter. To enable single-cell quantification of biallelic targeting and to demonstrate the feasibility of incorporating two transgenes into two AAVS1 alleles, we appended EGFP and mScarlet fluorescent genes downstream of the test split or non-split markers via the self-cleaving 2A peptide. Similarly, to test Hygro Intres, we swapped the EF1a and TetO-driven markers so that FL Hygro or Hygro Intres were placed downstream of TetO and FL Blast downstream of EF1a (Supplementary Fig. 9c, d; Plasmids 111–114). We co-transfected pX330-AAVS1 (Plasmid 106) M1 Ick-1

Mk M1

M2

Mk

Reconstituted marker d containing Cas9 and sgRNA targeting AAVS1, and the different pairs of targeting constructs (TC) into HEK293T cells, split into triplicate doxycycline-containing media without antibiotics, with Blasticidin, or with Hygromycin at the subsequent passages. Two weeks after selection, we analyzed the cultures for biallelic targeting by flow cytometric measurement of GFP and RFP fluorescence (Supplementary Fig. 9e). As expected, non-selected cultures harbored a small fraction (<1%) of biallelic knock-in GFP+/RFP+cells (Supplementary Fig. 9e; Selection = None). Selection of antibiotics where corresponding FL antibiotic resistance genes were present on targeting constructs yielded <30% biallelic knock-in cells (Supplementary Fig. 9e; Blast: TC a, c, d; Hygro: TC a, b, c). In contrast, selection by antibiotics where corresponding Intres are present on the targeting constructs yielded 75% (Supplementary Fig. 9e; Blast Intres: TC b) and 88% (Supplementary Fig. 9e; Hygro Intres: TC d) biallelic knock-in cells. Selection for an additional two weeks allowed split Blast and Hygro TCs to achieve 96.5% and 97.0% biallelic knock-in, respectively (Supplementary Fig. 9f, g). We next tested biallelic engineering in KOLF2-C1 human induced pluripotent stem cells (hiPSCs), which are karyotypically normal with a stable diploid genome13 (Fig. 5). The full-length non-split Blast targeting constructs (Fig. 5a) and 2-split Blast Intres targeting constructs (Fig. 5b) were tested for selection of biallelically modified clones.

A R T I C L E

HA-L

SA-2A

Neo PPP1R12C e1 HA-L

SA-2A

Neo

PPP1R12C e1 Non-split TCs 107, 108

Intres TCs 109, 110 P F G E s t e l r a c S m

Purified Cas9 proteins were complexed with synthetic sgRNA to form Cas9 ribonucleoprotein (RNP) and co-nucleofected with the targeting constructs into KOLF2-C1, followed by dox-induction and antibiotic selection. Surviving colonies were picked into separate wells for establishing single-cell clones. Genotyping PCR revealed that targeting using non-split Blast resistance gene generated only 8% biallelic clones, while targeting using Blasticidin Intres yielded exclusively (100%) biallelically modified clones (Fig. 5c, d), showing both fluorescent signals (Fig. 5e) indicative of the targeting by each targeting construct at the two alleles of AAVS in these hiPSCs.

Selection of four or more transgenes with Intres. The utility of Intres may become more apparent in cases where more than three transgenes are to be selected. As we have observed in our 3-split Hygromycin Intres engineering exercise that the use of a set of orthogonal inteins represent a better design for a more robust split marker, we tested four other inteins (gp411, gp418, NrdJ1, IMPDH1)14 in splitting HygroR or PuroR. We identified additional functional splits of HygroR and PuroR at different positions (Supplementary Figs. 10 and 11). Some of these additional Intres were further adapted to the Gateway cloning system (Supplementary Figs. 12 and 13). To directly observe protein splicing as well as to confirm these inteins are indeed orthogonal, we conducted western blot analysis of protein trans-splicing between Nmarkertrons N-terminally tagged with 3xFLAG-epitope and Cmarkertrons C-terminally tagged with HA-epitope (Supplementary Fig. 14). As expected, while cognate markertrons with matching N- and C-inteins supported reconstitution of the fulllength HygroR (lanes 3,5,6), markertrons with unmatched N- and C-inteins did not yield full-length HygroR (lanes 7,8). To introduce and select cells with four or more transgenes, one approach is through sequential transduction/selection of two or more sets of 2-split Intres vectors. By subjecting cells to two rounds of 2split Intres transduction/selection (Hygro → Puro or Puro → Hygro) with each round carrying two transgenes, we obtained quadruple transgenic cells (Supplementary Fig. 15). These results demonstrated that four transgenes can be sequentially introduced, and that the Intres system is compatible with sequential cell engineering. Another way to introduce four or more transgenes is with higher-degree split Intres markers. By combining the multiple inteins and positions tested for HygroR, we designed and tested 6-split Hygro Intres marker (Supplementary Fig. 16). While cultures transduced with all markertrons yielded viable cells, leave-one-out cultures missing any one of the markertrons did not produce any viable cells after selection. This result demonstrates that up to at least 6 transgenic vectors can be selected simultaneously by one selection scheme using a split selectable marker.

Proviral copy number analysis. We validated Intres lentiviral vectors in additional cell lines (HEK293T and HeLa) (Supplementary Fig. 17). To ask whether split markers require a substantially higher copy number than non-split markers to support selection, we conducted proviral copy number analysis in nonselective and selective cultures of cells transduced with non-split HygroR or split Hygro Intres markers (Supplementary Fig. 18) in U2OS, HEK293T and HeLa cells. In general, we observed 1.3–3.1fold proviral copy numbers in the split marker cultures compared to the non-split cultures. Since the two-split markers require the presence of the two different viral genomes hosting the two markertrons to reconstitute a full resistance protein, it is expected to have ~2-fold equivalence of viral integration to support selection.