HIGHLY SENSITIVE IN VITRO ASSAYS TO DEFINE SUBSTRATE PREFERENCES AND SITES OF NUCLEIC-ACID BINDING, MODIFYING, AND CLEAVING AGENTS

Abstract

Methods and compositions for performing highly sensitive in vitro assays to define substrate preferences and off-target sites of nucleic-acid binding, modifying, and cleaving agents.

Claims

1.-15. (canceled)

16. A method of selecting for double stranded DNA sequence(s) that are bound by a DNA-binding domain, the method comprising: (i) providing a plurality of linear dsDNA oligonucleotides of known sequences; (ii) incubating the plurality of linear dsDNA oligonucleotides in the presence of the DNA binding domain affinity-tagged with an affinity tag that can be bound to a substrate molecule under conditions sufficient for binding of the affinity-tagged DNA-binding domain and one or more of the plurality of linear dsDNA oligonucleotide(s) to occur, thereby creating bound linear dsDNA oligonucleotide(s); and (iii) selecting for bound linear dsDNA oligonucleotide(s), thereby creating selected linear dsDNA oligonucleotide(s).

17. A method of identifying double stranded DNA sequence(s) that are bound by a DNA-binding domain, the method comprising determining the sequence(s) of the selected linear dsDNA oligonucleotide(s) produced by the method of claim 16, thereby identifying double stranded DNA sequence(s) that are bound by the DNA-binding domain.

18. A method of enriching for double stranded DNA sequence(s) that are bound by a DNA-binding domain, the method comprising: (i) incubating the plurality of selected linear dsDNA oligonucleotides produced by the method of claim 16 in the presence of the DNA binding domain affinity-tagged with an affinity tag that can be bound to a substrate molecule under conditions sufficient for binding of the affinity-tagged DNA-binding domain and one or more of the plurality of selected linear dsDNA oligonucleotide(s) to occur, thereby creating bound selected linear dsDNA oligonucleotide(s); and (ii) selecting for bound selected linear dsDNA oligonucleotide(s), thereby creating enriched linear dsDNA oligonucleotide(s).

19. A method of identifying double stranded DNA sequence(s) that are bound by a DNA-binding domain, the method comprising determining the sequence(s) of the enriched linear dsDNA oligonucleotide(s) produced by the method of claim 16, thereby identifying double stranded DNA sequence(s) that are bound by the DNA-binding domain.

20. The method of claim 16, wherein the affinity-tagged DNA-binding domain is a Cas9 protein complexed with a sgRNA, a variant of a Cas9 protein complexed with a sgRNA, a Cas9 fusion protein complexed with a sgRNA, or a variant of a Cas9 fusion protein complexed with a sgRNA.

21. The method of claim 20, wherein the sgRNA targets a site selected from the group consisting of EMX1, FANCF, HBB, HEK2, HEK3, HEK4, RNF2, ABE14, ABE16, ABE18, and VEGFA3.

22. The method of claim 16, wherein the affinity-tagged DNA-binding domain is inactivated Cas9 (dCas9) complexed with a sgRNA.

23. The method of claim 16, wherein the affinity-tagged DNA-binding domain is an engineered zinc finger array.

24. The method of claim 16, wherein the affinity-tagged DNA-binding domain is an engineered TALE repeat array.

25. The method of claim 16, wherein the substrate molecule is a magnetic bead carrying a molecule that binds to the affinity tag.

26. The method of claim 16, wherein the affinity tag is a molecule that can be covalently bound to benzylguanine and the substrate molecule is a benzylguanine-carrying substrate molecule.

27. The method of claim 16, wherein said selecting comprises: (i) incubating the bound linear dsDNA oligonucleotide(s) under conditions sufficient for binding of the affinity tag to the substrate molecule, thereby creating substrate bound linear dsDNA oligonucleotide(s); (ii) separating the substrate bound linear dsDNA oligonucleotide(s) from unbound linear dsDNA oligonucleotide(s); and (ii) eluting the substrate bound linear dsDNA oligonucleotide(s) in either (a) an appropriate buffer to promote dissociation of the substrate bound linear dsDNA oligonucleotide(s) or (b) a buffer containing a protease under conditions effective to degrade bead-bound protein and release substrate bound linear dsDNA oligonucleotide(s), thereby creating selected linear dsDNA oligonucleotides.

28. The method of claim 27, wherein the protease is proteinase K.

29. The method of claim 18, wherein said enriching comprises: (i) incubating the bound selected linear dsDNA oligonucleotide(s) under conditions sufficient for binding of the affinity tag to the substrate molecule, thereby creating substrate bound selected linear dsDNA oligonucleotide(s); (ii) separating the substrate bound selected linear dsDNA oligonucleotide(s) from unbound linear dsDNA oligonucleotide(s); and (ii) eluting the substrate bound selected linear dsDNA oligonucleotide(s) in either (a) an appropriate buffer to promote dissociation of the substrate bound linear dsDNA oligonucleotide(s) or (b) a buffer containing a protease under conditions effective to degrade bead-bound protein and release substrate bound linear dsDNA oligonucleotide(s), thereby creating enriched linear dsDNA oligonucleotides.

30. The method of claim 16, wherein the linear dsDNA oligonucleotides comprise 16 to 10.sup.8 different sequences.

31. The method of claim 16, wherein the linear dsDNA oligonucleotides comprise sequences that are 50 to 500 bp long.

32. The method of claim 16, wherein the linear dsDNA oligonucleotides comprise potential DNA substrate sequences comprising: (i) a set of all potential off-target sequences for the cytidine deaminase base editing enzyme in a reference genome bearing up to a certain number of substitutions, single base pair deletions, and/or single base pair insertions relative to an identified on-target site for the cytidine deaminase base editing enzyme; (ii) a comprehensive set of all potential off-target sequences for the cytidine deaminase base editing enzyme bearing up to a certain number of substitutions, single base pair deletions, and/or single base pair insertions relative to an identified on-target site for the cytidine deaminase base editing enzyme; (iii) a set of potential off-target sequences for the cytidine deaminase base editing enzyme present in a set of variant genomes from defined populations bearing up to a certain number of substitutions, single base pair deletions, and/or single base pair insertions relative to an identified on-target site for the cytidine deaminase base editing enzyme; (iv) a set of all potential off-target sequences for the cytidine deaminase base editing enzyme in the coding sequence of a reference genome bearing up to six mismatches relative to an identified on-target site for the cytidine deaminase base editing enzyme; or (v) a set of all potential off-target sequences for the cytidine deaminase base editing enzyme in the sequence of an oncogene hotspot and/or tumor suppressor gene of a reference genome bearing up to a certain number of substitutions, single base pair deletions, and/or single base pair insertions relative to an identified on-target site for the cytidine deaminase base editing enzyme.

Description

DESCRIPTION OF DRAWINGS

[0019] FIG. 1. Illustration of the differences in library complexity among base substitution library (SEQ ID NOs.:1-6, 1, and 7-11) and genomic DNA libraries (SEQ ID NOs.:1 and 12-16). For genomic libraries, the on-target site (in a red box in this example) retains very little similarity to the other roughly ˜3 billion genomic sequences. For base substitution libraries, the pre-selection libraries are enriched for sites that are not necessarily present in the genome but that are similar to the intended target site. Substitutions are indicated by lower case letters.

[0020] FIG. 2. Illustrative overview of exemplary method. A user-defined set (typically, but not limited to, 10,000 to 100,000 sequences) of potential DNA substrates is generated by synthesis on a high-density oligonucleotide array and then made double-stranded. Three potential examples of sets of sequences to use include the set of all sequences with up to six mismatches in the human genome, eight mismatches in the human exome, or all possible DNA sequences with up to three mismatches. The double-stranded DNA library can then be used in screens for sequence modification, screens for sequence depletion, or selection for sequence modification and/or cleavage. Black lines indicate constant sequence that is present in every member of the library and is used for primer binding sites during amplification steps and in bioinformatic processing.

[0021] FIG. 3. Target site libraries can be selected for cleavage (strategy 1) or screened for depletion of cleaved sequences (strategy 2). Black lines indicate constant sequence that is present in every member of the library and is used for primer binding sites during amplification steps and in bioinformatic processing. SEQ ID NOs.:17-22 are shown.

[0022] FIG. 4. Enrichment of Cas9 cleavage sites by selection from a random base substitution library. The intended target site (SEQ ID NO:23) is listed below the heat map. Each black box represents the abundance of a particular nucleotide (denoted on left) at the position corresponding to the target site nucleotide listed below, with black representing the most abundant nucleotides/per position and white representing no abundance.

[0023] FIG. 5. Identification of on-target sequences through screening of random base substitution library and of genomic DNA library. Substrates with fewer numbers of mutations (indicated by lower m numbers, where (m_d_i)->m=number of mutations, i=number of insertions, d=number of deletions) in the substrate profiling library and in the genome-inspired library where Xm0 denotes m mismatches without any insertions, RNAmd indicates a target site deletion of length d with m mismatches at the remaining base pairs in the site, DNAmi indicates a target site insertion of length I with m mismatches at the remaining base pairs in the site).

[0024] FIG. 6. Compositions of representative genomic DNA oligonucleotide libraries are shown. The number of individual genomic sites are listed according to number of mismatches and bulges. Unless otherwise specified, these libraries were utilized in the experiments outlined in the subsequent figures.

[0025] FIG. 7. Library characterization. Uniformity metrics and drop out percentages are shown for each of the libraries, after oligonucleotide synthesis, library amplification, and Illumina sequencing. 90.sup.th percentile sequencing counts refer to the number of sequencing reads obtained for the library member in the 90.sup.th percentile, when ordered in terms of increasing reads. 90/10 ratio is the ratio of the number of sequencing reads for the 90.sup.th percentile library member divided by the 10.sup.th percentile library member and is a metric of library uniformity. Drop out refers to the number of sequences that were not represented in the sequenced, amplified libraries.

[0026] FIG. 8. Enrichment of known GUIDE-seq sites. Swarm plots are shown for representative cleavage selections using a method described herein (an example of which is referred to as the ONE-seq method). Each circle represents the aggregate read counts, normalized to the on-target sequence for a given guide RNA selection (listed on top), for an individual library member. The black stars indicate the on-target library member. Filled circles denote sites that were identified by published GUIDE-seq experiments. There were no published RNF2 GUIDE-seq sites.

[0027] FIG. 9. Enrichment of highly enriched CIRCLE-seq sites. Swarm plots are shown for representative cleavage selections using the ONE-seq method. Each circle represents the aggregate read counts, normalized to the on-target sequence for a given guide RNA selection (listed on top), for an individual library member. The black stars indicate the on-target library member. Filled circles denote sites with >100 read counts that were identified by published CIRCLE-seq experiments. There were no published RNF2 GUIDE-seq sites.

[0028] FIG. 10. Enrichment of moderately enriched CIRCLE-seq sites. Swarm plots are shown for representative cleavage selections using the ONE-seq method. Each circle represents the aggregate read counts, normalized to the on-target sequence for a given guide RNA selection (listed on top), for an individual library member. The black stars indicate the on-target library member. Filled circles denote sites with 10-99 read counts that were identified by published CIRCLE-seq experiments.

[0029] FIG. 11. Enrichment of lowly enriched CIRCLE-seq sites. Swarm plots are shown for representative cleavage selections using the ONE-seq method. Each circle represents the aggregate read counts, normalized to the on-target sequence for a given guide RNA selection (listed on top), for an individual library member. The black stars indicate the on-target library member. Filled circles denote sites with 1-9 read counts that were identified by published CIRCLE-seq experiments.

[0030] FIG. 12. Venn diagrams showing that the exemplary method (darker circles) identifies all 60 out of 62 highly enriched (>100 reads) CIRCLE-seq sites (lighter circles) of six SpCas9: sgRNA, using a cutoff of 1% of the on-target ONE-seq aggregated read count. The 2 sites that were not above the 1% ONE-seq cutoff do not necessarily represent bona fide off-target sequences and may be false positives in the CIRCLE-seq method. CIRCLE-seq does not identify 478 of the ONE-seq identified sites for these guide RNAs.

[0031] FIG. 13. Validation results of three FANCF off-target sites identified by ONE-seq but not GUIDE-seq or CIRCLE-seq. Targeted amplicon sequencing was performed on three of the five most highly enriched novel off-target candidates identified by ONE-seq from HEK293T cells sorted for the top decile of expression of a SpCas9:FANCF sgRNA construct. On the left, total number of indel-containing (edited) sequence reads and total number of reference reads are shown, along with edit percentage and unedited candidate off-target sequence are shown. To the right, individual data from three separate sorting and control (untreated) experiments are shown. (SEQ ID NOs:24-39 appear in order)

[0032] FIG. 14. Reproducibility of enrichment scores in variant libraries. ONE-seq selections were performed on an EMX1 genomic off-target library and an EMX1 genomic variant off-target library. Enrichment scores (relative to the on-target sequence) are shown for the library members that are shared by both libraries. The superimposed line corresponds to equal enrichment scores from both selections.

[0033] FIG. 15. ONE-seq identifies candidate off-target sites that are present in the population but not in the reference genome. Normalized aggregate read counts (where 1.0 is the on-target site) are shown for off-target candidates identified from the reference genome and paired off-target candidates that contain SNPs found in the 1000 genomes population. Variants that are present in >40% of the population are shown with filled circles. The superimposed line corresponds to equal enrichment scores from both paired library members.

[0034] FIG. 16. Base editor screening strategy. A random base substitution library designed for the EMX1 target site was incubated in vitro with BE1, and amplified by PCR with Kapa HiFi Uracil+DNA polymerase, which converts U:G base pairs to equal mixtures of T:A and C:G base pairs during DNA synthesis (a dATP nucleotide is incorporated across from dU). Therefore, any library member that can be modified by BE1, when sequenced, will sequence as a mixture of the original barcode-linked sequence from the pre-treatment library and also a modified sequence that contains C->T substitutions (and other rarer substitutions).

[0035] FIG. 17. Base editor screen demonstrates enrichment of sites containing NGG and demonstrates high specificity at the PAM-proximal end of the target site (SEQ ID NO:23) and lower specificity at the PAM distal end. Heat map is interpreted in the same way as in FIG. 4.

[0036] FIG. 18. BE3 selection strategy. In this strategy, target site libraries are exposed to BE3 enzyme and are enriched for modified members through double-strand break creation at sites with uridine nucleotides (through USER) and nicks (through BE3).

[0037] FIG. 19. Enrichment of BE3 off-target sites by ONE-seq. Normalized aggregate read counts (where 1.0 corresponds to the on-target site) are shown for eight ONE-seq selections on genomic DNA-inspired libraries. Only sites with a score of 0.01 or greater (1% of on-target enrichment) are shown. Black stars represent on-target library members. Filled in black circles denote newly validated off-target sites compared to Digenome-seq (with the exception of ABE Site 18, which was not tested by Digenome-seq). Open black circles denote Digenome-seq candidate sites.

[0038] FIG. 20. Newly identified and validated BE3 off-target sites. Data from targeted amplicon sequencing from genomic DNA from HEK293T cells expressing the indicated BE3:sgRNA complexes are shown, in comparison to an untreated control. Experiments were performed in three replicates. Only the 28 newly identified and validated BE3 off-target sites compared to Digenome-seq are shown.

[0039] FIG. 21. ABE selection strategy. In this example, the Adenine base editor (ABE) is used. ABE creates A->I changes in DNA. The method we use here to define off-target sites of ABE is similar to that used in Example 3, except a different endonuclease, endonuclease V, is used to create a nick at a deoxyinosine site in DNA.

[0040] FIG. 22. ABE selection on a base substitution library. Heatmap is interpreted in same way as in FIG. 4. The data from the selection demonstrate enrichment of NGG PAMs, but also enrichment of sequences containing an A at position five, relative to the SpCas9 cleavage selection in FIG. 4 and the BE3 selection in FIG. 8, demonstrating the need for an A in a certain editing window for ABE to demonstrate activity. (SEQ ID NO:23) FIG. 23. Enrichment of ABE7.10 off-target sites by ONE-seq. Normalized aggregate read counts (where 1.0 corresponds to the on-target site) are shown for eight ONE-seq selections on genomic DNA-inspired libraries. Black stars represent on-target library members. Filled circles denote validated off-target sites. Open dark circles denote off-target candidates that were sequenced in the validation study.

[0041] FIG. 24. Validated ABE off-target sites. Data from targeted amplicon sequencing from genomic DNA from HEK293T cells expressing the indicated ABEmax:sgRNA complexes are shown, in comparison to an untreated control. Experiments were performed in three replicates.

[0042] FIG. 25. Capillary electrophoresis data from an experiment demonstrating the specificity of TkoEndoMS' endonuclease activity for G:U DNA mismatches in vitro. An 800 base pair PCR amplicon was incubated with purified BE protein and a variable sgRNA for two hours to induce site-specific deamination. After purification, the deaminated PCR amplicon was incubated with purified TkoEndoMS protein for 7 minutes to induce double strand breaks at G:U mismatches. The DNA was then separated by size by capillary electrophoresis and imaged.

[0043] FIG. 26. Overview of enrichment for binding sites by pulldown. In this method, dCas9 coated beads are incubated with a library of potential off-target sites. Library members that are not bound are washed into supernatant, and bound library members are eluted by digesting bead-bound protein with Proteinase K. The resulting eluted library can either be amplified and subjected to additional rounds of pulldown or subjected to analysis by high throughput sequencing.

[0044] FIG. 27. Pulldown conditions can discriminate between on and off target sites. A mixture of three double stranded DNAs with differing lengths was subjected to binding site pulldown with dCas9:EMX1 sgRNA-coated beads. The on target site was present on a 280 base pair DNA, and one of two off target sites (OT2 or OT4) were present on a 220 base pair DNA. A third 200 bp (“random”) DNA containing neither on nor off-target site was also in the mixture. Enriched DNA was run on a QIAxcel. Lane A3 denotes a size ladder. Lane A1 shows selective pulldown of the 280 base pair on target site, but no pull down of the OT2 site or the 200 bp DNA. Lane A2 shows that the OT4 site can still be bound in the method, demonstrating that the conditions used in the pulldown are able to enrich for off target sequences that are capable of being bound by a dCas9:EMX1 sgRNA. (SEQ ID NOs:23, 40-41 appear in order)

[0045] FIG. 28. Enrichment of genomic DNA-inspired library by pulldown. A pulldown conducted in the presence of 50 ug/ml heparin on the FANCF library leads to enrichment of the on-target site (black star) to the most abundant post-pulldown library member.

[0046] FIG. 29. Post-selection library composition for I-Ppol. A sequence logo is shown for sequences with at least 1% of the normalized read counts of the intended I-Ppol target site. Position in the site from 5′ to 3′ end is shown on the horizontal axis, and the height of the stack of letters denotes information content (in bits) for each position. The height of each individual nucleotide highlights the relative contribution oft hat nucleotide to the information content oft hat position. Positions 2, 13, and 14 are the most highly specified (highest information content). Position 15 is the least specified (lowest information content).

DETAILED DESCRIPTION

[0047] In vitro/biochemical strategies to understanding on- and off-target activity of DNA binding domains generally fall into two types (FIG. 1): In the first type, a set of DNA sequences in a relevant defined system (for example, the human genome) is interrogated for off-target cleavage events. Examples of methods that utilize this strategy include CIRCLE-seq (Tsai et al., Nat Meth. 14: 607 (2017)], SITE-seq (Cameron et al. Nat Meth. 14: 600 (2017)), and Digenome-seq (Kim et al. Nat Meth. 12: 237 (2015)). The scope of these genome-wide methods is restricted to off-target sites that are present in the particular genomic DNA used in the study. In contrast, in the second type of strategy, the substrate preference of the DNA binding/modifying protein is assayed in a more unbiased fashion by comprehensively interrogating a library of binding sites in which certain base positions are randomly substituted with all potential alternative bases (rather than just a limited set of base substitutions as in the case of the first strategy). These “genomic DNA” and “random base substitution library” approaches have been carried out for various nucleases (ZFNs (Pattanayak et al. Nat Meth. 8: 765 (2011)], TALENs (Guilinger et al. Nat Meth. 11: 429 (2014)), and CRISPR-Cas9 (Pattanayak et al. Nat Biotech. 31: 839 (2013))) in vitro and provide insights into the biochemical function and specificity of these nucleases.

[0048] Both types of strategies to study off-target activity have limitations that affect their abilities to identify bona fide off-target sites. In genome-wide selections, the tens to hundreds of cleaved off-target sites must be enriched from a background of billions of other sites that are not cleaved (the human genome has a length of ˜3 billion base pairs and therefore contains ˜6 billion sites to be assayed). For example, due to noise in the enrichment method and in the sequencing results, the CIRCLE-seq method is limited to detection of sites that have no more than six mismatches relative to the on-target site, which represents only ˜0.002% of the genomic material present in the assay. While some methods, such as Digenome-seq, rely on massive over sequencing of nuclease treated DNA libraries, methods like CIRCLE-seq and GUIDE-seq typically incorporate an enrichment step for edited sequences. This enrichment step can be performed in cells (GUIDE-seq) or in vitro (CIRCLE-seq). Although it is substantially more sensitive than other methods for off-target screening, the CIRCLE-seq method requires a very large input of genomic DNA (25 μg) for each experimental sample.

[0049] In vitro selections on unbiased base substitution libraries are limited by library size (the set of sequences that can practically be assayed). For example, an SpCas9 target site contains 22 potentially-specified base pairs (20 from hybridization to the guide RNA and two from the PAM sequence). To assay all potential target sites bearing all possible combinations of base substitutions at all positions, at least 4.sup.22˜10.sup.13 unique molecules of DNA, would need to be generated and interrogated, neither of which is possible using current technologies. For example, library construction methodologies are currently limited to producing 10.sup.11-10.sup.12 unique molecules of DNA. Furthermore, even if library construction methodologies were improved, it is not feasible to sequence 10.sup.12 molecules of DNA. To overcome this restriction, doped oligonucleotide synthesis is traditionally used to create a library of sites bearing base substitutions that follow a binomial distribution, such that the on-target site is present in more copies than each variant site in the library bearing a single mutation, each of which are present in more copies than each variant site in the library bearing double variant site, and so forth. Therefore, selections performed with these random base substitution libraries are limited by the fact that 1) it is not possible to create a completely unbiased library (i.e., they are heavily biased towards the intended on-target site sequence) and 2) it is not possible to create a library that uniformly represents the potential sequence space. Furthermore, using the outputs from defined libraries assays to predict or identify off-target sites in genomic sequences often requires extrapolation (Sander et al. Nucleic Acids Res. 41: e181 (2013)), because not all relevant genomic sequences are guaranteed to be covered in pre-selection (limited to 10.sup.12 sequences, which corresponds to six or seven substitutions) or post-selection libraries (limited to 10.sup.7-8 sequences by sequencing capacity).

[0050] Methods of Identifying DNA Binding, Modification, or Cleavage Sites Herein, we provide improved methods (FIG. 2) that enable identification of on- and off-target binding, modification, or cleavage sites of DNA modifying proteins/protein complexes (including but not limited to: dCas9 fused to an effector domain, Cas9-based base editors, or active Cas9 proteins) and that overcomes the disadvantages of both the “genomic DNA” and “random base substitution library” approaches. With this method, a pre-enriched library of linear DNAs consisting of particular user-specified sequences is generated by high-density oligonucleotide synthesis and then interrogated for those sequences that can be bound, modified or cleaved by sequence-specific proteins or protein complexes. Minimally, this method allows for the identification of sequences that are potential substrates for any agent whose modifying action can lead to sequence modification, binding, or cleavage of nucleic acid.

[0051] The pre-enriched linear DNA library members are initially synthesized on high-density oligonucleotide arrays as individual single-stranded DNA sequences, each bearing a unique identifier/barcode, which is present/duplicated on both sides of the oligonucleotide (FIG. 2). The synthesized oligos are released from the chip and converted into double-stranded DNA molecules by priming against a common sequence present in all DNA molecules synthesized on the chip. This pooled library is then incubated with a site-specific nuclease, modifying protein, or DNA binding domain of interest and either enriched for cleaved, modified, or bound sequences in a selection format (see examples 1, 3, and 4) or screened for modification (see example 2). The DNA sequences of cleaved sites can then be reconstructed from either of the identical barcodes that originally flanked these sites and that are now separated into two molecules. Synthesized molecules can be specified to represent 1) a set of all potential off-target sequences in a reference genome bearing up to a certain number of mismatches relative to the on-target site (analogous to genomic DNA libraries), 2) a comprehensive set of potential off-target sites bearing up to a certain number of mismatches (analogous to random base substitution libraries), 3) a library of potential off-target sequences present in a set of variant genomes from defined populations (i.e., genomic DNA libraries designed to reflect DNA sequence variants present in a population of individuals), or 4) other relevant defined sets of potential off-target sites (for example, oncogene hotspots or sequences from tumor suppressor genes). This strategy has important advantages for constructing these libraries. For random base substitution libraries, all sequences within a defined number of substitutions can be represented equally and can be easily sampled with current next-generation sequencing methodologies. For genomic or exomic DNA libraries, only the sites that are most likely to be relevant (for example, all potential off-target sites with six or fewer substitutions) are included, which eliminates the noise contributed by the ˜99.998% of sites that are not substrates. Importantly, because this method results in the generation of a double-strand DNA (or RNA) library of an enriched set of potential off-target sites of a DNA (or RNA) binding protein, it can be used to define the specificity and off-target sites of not only nucleases but also other proteins that bind or modify nucleic-acid, including but not limited to, customizable base editing enzymes (Komor et al. Nature 533: 420, 2016, Gaudelli et al. Nature. 551: 464 (2017)), transcriptional activators (Mali et al, Nat Biotech. 31: 833 (2013), Chavez et al. Nat Meth. 12: 326 (2015)), transcriptional repressors (Bikard et al, Nucleic Acids Res, 41: 7429 (2013), Thakore et al. Nat Meth. 12: 1143 (2015)), and epigenome editing enzymes (reviewed in Zentner and Henikoff. Nat Biotech. 33: 606 (2015)).

[0052] The ability to define identical barcodes flanking a defined recognition site represents a significant advance over previous in vitro profiling methods (U.S. Pat. Nos. 9,322,006, 9,163,284), because the sequence of the library member is encoded in at least three locations on each individual member of a DNA pool. This redundancy of information is particularly advantageous when seeking to define DNA modifying activity (such as base editing) where the target sequence is modified. The original sequence information can be obtained from the information content contained in a flanking barcode, even if the actual DNA sequence of the library member itself is modified. The redundancy of information in two barcodes and a recognition site also allows for an endonuclease cleavage selection (or paired base modification+cleavage selection) to be performed on potential cleavage sites that are present in a single copy per library member, as opposed to multiple copies (U.S. Pat. Nos. 9,322,006, 9,163,284). Without the present barcoding strategy, sequences of library members that get cleaved within a recognition site cannot be reassembled, since the cut separates in space the two sides of the cut site (above figure, bottom right, blue region).

EXAMPLES

[0053] The invention is further described in the following examples, which do not limit the scope of the invention described in the claims.

[0054] Target sites used in following examples:

TABLE-US-00001 target name Sequence (5′ -> 3′) SEQ ID NO: EMX1 GAGTCCGAGCAGAAGAAGAAGGG 22 RNF2 GTCATCTTAGTCATTACCTGAGG 42 FANCF GGAATCCCTTCTGCAGCACCTGG 43 HBB TTGCCCCACAGGGCAGTAACGG 44 HEK2 (HEK293_2) GAACACAAAGCATAGACTGCGGG 45 HEK3 (HEK293_3) GGCCCAGACTGAGCACGTGATGG 46 HEK4 (HEK293_4) GGCACTGCGGCTGGAGGTGGGGG 47 ABE14 (ABE_14) GGCTAAAGACCATAGACTGTGGG 48 ABE16 (ABE_16) GGGAATAAATCATAGAATCCTGG 49 ABE18 (ABE_18) ACACACACACTTAGAATCTGTGG 50 VEGFA3 (VEGFA_3) GGTGAGTGAGTGTGTGCGTGTGG 51

Example 1: DNA Cleavage Selection with SpCas9 and SpCas9-HF1

[0055] In this example, a random base substitution library designed for an SpCas9 nuclease programmed with a guide RNA (gRNA) designed against an on-target site in the human EMX1 gene (hereafter referred to as the EMX1 gRNA and EMX1 target site) and a library of potential EMX1 gRNA off-target sites from the human reference genome were selected for cleavage with SpCas9 or SpCas9-HF1.

[0056] In this example (FIG. 3), both a selection (strategy 1) and a screen (strategy 2) can be employed. In strategy 1, a pooled library of ˜50,000 barcoded library members (either a random base substitution library containing all possible sequences within three mismatches of an EMX1 SpCas9 on-target site or a genomic DNA-inspired library containing all possible sequences from the hg19 human reference genome within six mismatches of an EMX1 SpCas9 on-target site—see Methods for details of libraries.)

[0057] A selection performed using strategy 1 on the random base substitution library with a 1:1:1 ratio of SpCas9:sgRNA:DNA library (EMX1 target sites) demonstrated enrichment of sequences that could be cleaved (FIG. 4). The positions in the target site are on the horizontal axis (with the on target bases listed below). The possible bases (substitutions or on-target) in the library are indicated on the vertical axis. Data was pooled and summarized in a heatmap, where darker black rectangles indicate a larger proportion of sites in the post-selection library containing the corresponding base from the vertical axis. As a proof of principle, this heatmap agrees with previous studies that demonstrate the N of the NGG PAM sequence is not specified and that specificity at the PAM-distal end of the target site is lower than that of the PAM-proximal end.

[0058] A screen performed using strategy 2 with the random base substitution library yielded similar results (FIG. 5). Substrates with fewer numbers of mutations (indicated by lower m numbers, where (m_d_i)->m=number of mutations, i=number of insertions, d=number of deletions) in the substrate profiling library and in the genome-inspired library where Xm0 denotes m mismatches without any insertions, RNAmd indicates a target site deletion of length d with m mismatches at the remaining base pairs in the site, DNAmi indicates a target site insertion of length I with m mismatches at the remaining base pairs in the site).

[0059] Genomic libraries are generally composed of all potential off-target sites in the hg19 reference human genome that had zero to six mismatches relative to the on-target sequence, up to four mismatches in combination with a DNA bulge of one or two nucleotides, and up to four mismatches with an RNA bulge of one nucleotide, and up to three mismatches with an RNA bulge of two nucleotides (FIG. 6). Sequencing of pre-selection libraries to assess quality metrics (FIG. 7) demonstrated a low dropout rate (0.20% or less) and high uniformity (90/10 ratio >−2). These metrics have not, to our knowledge, been calculated for other specificity methods, so direct comparison is not possible.

[0060] Selections were performed using strategy 1 (referred to as ONE-seq) with genomic DNA-inspired libraries for six non-promiscuous guide RNAs (HBB, RNF2, HEK2, HEK3, FANCF, and EMX1) with relatively few expected off-target sequences. On target sequences (FIG. 8, black stars) were either the most enriched or among the top 3 most enriched library members out of tens of thousands for the six non-promiscuous guide RNAs tested. Summed over the six non-promiscuous guide RNAs, ONE-seq enriched all 163 GUIDE-seq identified off-target sites (FIG. 8, filled circles), with post-selection read counts of ranging from 11% to 120% of the on-target sequences. This method also enriches highly enriched CIRCLE-seq sites (defined here as those with >100 sequence reads, FIG. 9), and appropriately, enriches to a lesser extent CIRCLE-seq sites that are moderately enriched (10-99 reads, FIG. 10), or lowly enriched (1-9 reads, FIG. 11). If a cutoff of 1% of the on-target enrichment in the ONE-seq method described here, ONE-seq identifies 60 out of 62 highly enriched CIRCLE-seq sites (FIG. 12), while CIRCLE-seq fails to identify 478 highly enriched ONE-seq candidates. Of note, the 2 highly enriched CIRCLE-seq sites that are not highly enriched by ONE-seq may be false positives of the CIRCLE-seq method. Validation of ONE-seq sites that are novel and not identified by GUIDE-seq or CIRCLE-seq was demonstrated by sorting HEK293T cells in the top decile of SpCas9:FANCF sgRNA expression (FIG. 13). These results demonstrate that the method described here is at least as sensitive as existing methods and is likely more sensitive.

[0061] In addition, this method can be generalized to any library/defined set of nucleic acid sequences. For example, using publicly available data from the 1000 genomes project, ONE-seq selections were performed on an EMX1 genomic off-target site library that accounts for naturally occurring sequence variation on a population scale. In this example library, all sequences from the reference hg19 human genome assembly that were in the original EMX1 library (FIG. 6) and contained a SNP in the 1000 genomes database were included. In addition, the SNP containing sequences were also included as additional library members to account for the possibility that individuals may have off-target sequences that are not contained in the reference genome. A ONE-seq cleavage selection performed on this SNP-containing EMX1 library provides reproducible enrichment of off-target candidates that are present in the reference hg19 genome (FIG. 14). ONE-seq cleavage selections on the variant library demonstrate assessment of tens of thousands of variants present in the population for a candidate set of off-target sites of the EMX1 guide RNA, identifying several (FIG. 15, black circles) that are differentially enriched.

Example 2: Base Editor Screen with BE1

[0062] In this example (FIG. 16), a screening strategy is used to identify base modification created by the BE1 enzyme (Komor et al. Nature 533: 420, 2016), which canonically creates C->U changes in a defined window of DNA.

[0063] A base editor screen following the protocol above with BE1 was applied to an EMX1 target site and the substrate profiling library yielded enrichment of an expected profile of tolerated off-target sites (FIG. 17).

Example 3: Base Editor Selection with BE3

[0064] In this example (FIG. 18), a selection strategy was used to enrich for sites that are modified by the BE3 enzyme. Library members that can be recognized by the BE3 enzyme (Komor et al. Nature 533: 420, 2016) should exhibit both C->U modification and a nick on the opposite strand. The USER enzyme (NEB) was used to achieve double stranded cleavage of library members that are BE3 substrates by replacing dU nucleotides with a nick. Resulting modified library members will therefore contain two nicks with 5′ phosphates on opposite strands and are incubated with a DNA polymerase that can blunt these DNA overhangs (ex: T4 DNA polymerase or Phusion DNA polymerase). The resulting phosphorylated blunt ends are captured with double stranded DNA adapters prior to amplification/selection using one primer specific to the adapter and one primer specific to the library backbone (as in Example 1). Additional selection stringency can be obtained by performing size selection for smaller, cut fragments before or after amplification.

[0065] Using this approach, we examined BE3 targeting with genomic DNA-inspired libraries for eight target sites, including all seven BE3 targets tested previously by Digenome-seq (Kim et al. Nat. Biotech. 35:475, 2017). ONE-seq selection results revealed enrichment of the intended target sites to the top 13 of tens of thousands of library members for all eight selections (FIG. 19, black stars). For three out of the eight selections, the intended target site was the most enriched site. All 42 previously validated off-target sites by Digenome-seq were present in the enriched, post-selection libraries (FIG. 19, open black circles), and 40 out of the 42 were among the top 61 sites for each selection. To further validate our ONE-seq results, we amplified and sequenced from human HEK293T cells approximately 20-40 highly-ranking sites from each selection. Our results demonstrated 28 validated BE3 off-target sites that were not identified as candidates by Digenome-seq (FIG. 19, solid black circles). Six of the 28 new sites had edit percentages greater than 1% in cells (FIG. 20), with a high of 23.9%, suggesting a higher level of sensitivity of ONE-seq for detecting not only weak off-target sites but also high frequency off-target sites.

Example 4: Base Editor Selection with ABE

[0066] In this example (FIG. 21), we performed a selection using Adenine base editor (ABE; Gaudelli et al. Nature. 551: 464 (2017)) with the EMX1 gRNA and a base substitution profiling library and the genomic DNA library. ABEs are sgRNA-guided Cas9 nickases fused to a protein domain that can catalyze the conversion of deoxyadenosine to deoxyinosine. In this example, double strand cleavage of the pre-selection library is accomplished in two steps (FIG. 21). First, incubation with ABE enzyme and guide RNA leads to nick formation of the strand of a recognized library member that hybridizes to the guide RNA. Second, subsequent incubation with Endonuclease V, an enzyme that creates a nick to the 3′ of deoxyinosines in library members that could result from ABE activity, leads to nick formation on the non-hybridized DNA strand, leading to a double strand break with an overhang. Subsequent fill-in of the double strand break with a DNA polymerase leads to the formation of blunt ends, which can be selected for as described for Cas9 nucleases in Example 1.

[0067] Selection of the base substitution library demonstrates enrichment of substrates with an NGG. In addition, as expected, this experiment (FIG. 22) demonstrates enrichment of substrates with an A at position five of the target site (where 1 is the base pair most distal from the PAM), reflecting a preference of ABE for modification of an A more distal to the PAM than is present in the canonical EMX1 target site. Of note, of the 100 most abundant sequences in the post-selection library, 95 had an A at position 5. These results demonstrate that our strategy works to enrich and identify off-target sites of the ABE.

TABLE-US-00002 TABLE 1 Enrichment of sequences with an A in position 5 in the ABE selection. Number of times observed First five nucleotides out of the top 100 most of post-selection enriched post-selection library member library members GAGTA 83 AAGTA 12 GAGTC (canonical first 3 five nucleotides) GAAGT 1 GGAGT 1

[0068] We have also performed the above selection on the EMX1 genomic DNA library (Table 2), which demonstrates enrichment of the EMX1 on-target site (highlighted; 96.sup.th most abundant post-selection library sequence) and the EMX1 off-target site with the highest off-target recognition (bold and asterisk; 9.sup.th most abundant post-selection).

TABLE-US-00003 TABLE 2 Top 96 most-enriched sites in the post-selection library for an ABE selection on a genomic DNA library of potential EMX1 off-target sites. SEQ ID chromosome location target NO: chr4 33321459 GTACAGGAGCAGGAGAAGAATGG 52 chr17 72740376 CAAACGGAGCAGAAGAAGAAAGG 53 chr10 58848711 GAGCACGAGCAAGAGAAGAAGGG 54 chr10 128080178  GAGTACAAGCAGATGAAAAACGG 55 chr6 99699155 GAGTTAGAGCAGAGGAAGAGAGG 56 chr7 141972555  AAGTCCGGGCAAAAGAGGAAAGG 57 chr19 24250496 GAGTCCAAGCAGTAGAGGAAGGG 58 chr11 111680799  CAGTAGTGAGCAGAAGAAGATAGG 59 chr5  45359060* GAGTTAGAGCAGAAGAAGAAAGG 60 chr7 17446431 GTCCAAGAGCAGGAGAAGAAGGG 61 chr12 106646073  AAGTCCATGCAGAAGAGGAAGGG 62 chr15 22366604 GGAGTAGAGCAGAGGAAGAAGGG 63 chr10 109561613  GGAACTGAGCAAAAGAAGATAGG 64 chr11 62365266 GAATCCAAGCAGAAGAAGAGAAG 65 chr2 21489994 GCGACAGAGCAGAAGAAGAAGGG 66 chr1 234492858  GAAGTAGAGCAGAAGAAGAAGCG 67 chr2 218378101  GAGTCTAAGCAGGAGAATAAAGG 68 chr18 32722283 TGTCCAGAGCAGATGAAGAATGG 69 chr22 22762518 GAACATGAGCAGAAGAAGAGGAG 70 chr11 34538379 AGGCCAGAGCAAAAGAAGAGAGG 71 chr11 106142352  GTACAAGAGCAGGAGAAGAAGGG 72 chr15 91761953 GAGTCAGGGCAGAAGAAGAAAAT 73 chr4 87256685 GAGTAAGAGAAGAAGAAGAAGGG 74 chr4 21141327 AAGCCCGAGCAGAAGAAGTTGAG 75 chr8 128801241  GAGTCCTAGCAGGAGAAGAAGAG 76 chr7 106584579  GAGGGGAGCAAAAGAAGGAGGG 77 chr1 117139004  CAGGGAGAGCAAAAGAAGAGAGG 78 chr1 231750724  GAGTCAGAGCAAAAGAAGTAGTG 79 chr15 44109746 GAGTCTAAGCAGAAGAAGAAGAG 80 chr21 23586410 CAGGGAGAAGAAGAAGAAGGG 81 chr7  2127682 GAGTTAGAGAAGAAGAAGACTGG 82 chr10 98718174 ACAATCGAGCAGCAGAAGAATGG 83 chr1 221020698  GAGTAGGAGCAGATGAAGAGAGG 84 chr9 115729750  CAGTATGAGCAAAAGAAGAAAGA 85 chr11 102753237  GAGTCCATACAGAGGAAGAAAAG 86 chr1 48581991 GAATGAGCAAAAGAAGAAAGC 87 chr12 73504668 GAGTTAGAGCAGAAAAAAAATGG 88 chr1 184236226  AATACAGAGCAGAAGAAGAATGG 89 chr11 119322554  TAGTGAGCAGAAGAAGAGAGA 90 chr1 151027591  TTCTCCAAGCAGAAGAAGAAGAG 91 chr11 68772640 GAGTCCATACAGGAGAAGAAAGA 92 chr2  9821536 AGGTGGGAGCAGAAGAAGAAGGG 93 chr2 54284994 AAGGCAGAGCAGAGGAAGAGAGG 94 chr1 99102020 GAGGCACAAGCAAAAGAAGAAAAG 95 chr19  1438808 GAAGTAGAGCAGAAGAAGAAGCG 96 chr2 73160981 GAGTCCGAGCAGAAGAAGAAGGG 22

[0069] The two sequences highlighted are the most active cleavage off-target site (chr5: 45359060), asterisked, and the on-target site (chr2: 73160981). It is expected for the off-target site to be more enriched in the selection due to presence of an A in a more favorable position in the editing window.

[0070] We have additionally performed the above selection on genomic DNA libraries designed to identify off-target sequences of six guide RNAs (FIG. 23). Application of the modified ONE-seq selection protocol to six ABE targets revealed enrichment of the intended, on-target site to the top 3 of the post-selection libraries for the five non-promiscuous guides tested (HEK4 is a known promiscuous guide RNA). Validation by amplicon sequencing of DNA from human HEK293T cells individually transfected with the appropriate ABE7.10:sgRNA pairs of top candidate sites (approximately 20 each from each selection) identified 12 total confirmed cellular off-target sites across six target sites. This set includes the three validated off-target sites identified for the two guide RNAs that were tested by both ONE-seq and EndoV-seq (Liang et al. Nature Communications. 10: 67 (2019)) or Digenome-seq (Kim et al. Nature Biotechnology. 37: 430 (2019)) as well as nine newly validated off-target sites that were not identified as potential candidates by either of those methods. Nine of the 12 sites either had off-target modification rates below one percent or only showed evidence of a single nucleotide substitution, either of which could be caused by sequencing error despite stringent quality filtering of sequencing reads (all positions in paired reads must have quality score >Phred 30) and triplicate validation. To improve our confidence that these sites are bona fide off-target sites, we performed a second round of validation experiments with cells transfected with a plasmid expressing both ABEmax, a codon-optimized version of ABE7.10, and GFP and sorted the cells to enrich for the top decile of GFP, and therefore ABE, expression (FIG. 24). Genomic DNA extraction was performed immediately after sorting, without further expansion. In the sorted validation set, on target modification frequencies ranged from 61%-94%, compared to 31%-56% in the unsorted validation set. All 12 of the off-targets from the unsorted validation set were modified at higher frequencies in the sorted validation set, confirming that they are bona fide off-targets, and five additional off-target sites were identified at modification frequencies of less than one percent. One ABE_14 off-target site, containing a single mismatch relative to the on-target site, was modified in 85% of the DNA in the sorted validation (18% in the unsorted validation), suggesting that some ABE off-target sites can be modified at high frequencies.

Example 5: Base Editor Selections with ABE or BE3 Using an Enzyme that Creates a Double-Strand Break at Positions that have been Modified

[0071] In this example, modified library members containing a deoxyinosine could be made to have blunt, double-stranded ends through the action of the TkoEndoMS protein (Ishino et al, Nucleic Acids Res. 44: 2977 (2016)). TkoEndoMS can be used to create a double-strand breaks at the dI:dT base pairs that result from dA->dI editing by ABE. DNA with a double strand break is then subject to the same downstream steps as in Example 1, with ligation of adapters to phosphorylated, blunt ended DNA if a base editing enzyme without nicking activity is used. If a base editing enzyme with nicking activity is used, end polishing with a blunt-end creating DNA polymerase (such as T4 or Phusion), such as in Examples 4 and 5, is used to allow for enrichment of both sides of a cut library member.

[0072] We have demonstrated that TkoEndoMS can also create double-strand breaks at dG:dU mismatched base pairs that result from dC->dU editing (in this example by BE1), demonstrating its additional applicability to BE1, BE3 and other enzymes that cause dC->dU changes after DNA binding (see U.S. Ser. No. 62/571,222 and FIG. 25). This strongly suggests that we can use TkoEndoMS on our synthesized DNA site libraries to identify off-target base edits caused by the various base editors that induce dC to dU edits, regardless of whether a Cas9-induced nick is also present.

Example 6: Enrichment of DNA Binding Sites by Pulldown

[0073] SELEX (selective evolutions of ligands by exponential enrichment) has been used to define the DNA-binding specificity of DNA-binding domains (originally by Oliphant et al., Mol Cell Biol. 9: 2944, 1989). In the SELEX method, libraries of randomized DNA sequences are subjected to multiple rounds of pulldown and enrichment with an immobilized DNA binding domain of interest to identify the sequences in the initial pool that can bind to a DNA of interest. The SELEX method has been applied to the zinc finger and TALE moieties of ZFNs (Perez et al., Nat Biotech. 26: 808 (2008)) and TALENs (Miller et al. Nat Biotech. 29: 143 (2011)), however, there are no reports of SELEX studies on Cas9 proteins. We speculated that SELEX studies on Cas9 proteins are difficult due to the need to selectively enrich a 22 base pair target site from a large library, which would have to contain >10.sup.13 unique molecules, or at minimum 10.sup.12 molecules, corresponding to a 20 base pair target site, if an NGG PAM is fixed.

[0074] In this example, we took advantage of pre-enriching our pre-selection libraries for sites that are most likely to be bound by a given Cas9:sgRNA complex (or other DNA-binding domain with predictable binding motifs). We assessed Cas9 DNA binding preferences and specificity by performing sequential rounds of DNA pull down experiments on the pre-enriched libraries (FIG. 26). This was achieved by tethering inactivated Cas9 (dCas9) to magnetic beads. To chemically bind dCas9 to magnetic beads we employed Cas9 protein harboring a so called SNAP-tag. Proteins with a SNAP-tag can be covalently bound to a benzylguanine-carrying substrate molecule, such as a magnetic bead. We envision incubating either type of oligonucleotide library with bead-bound SNAP-tagged dCas9 and enriching for DNA substrates with a high binding affinity to Cas9 by magnetic bead capture of bound sequences and washing away unbound sequences. This process could be repeated in multiple cycles by amplifying eluted library members and using the resulting enriched DNA library as a starting library for bead-based selection. Using this method, with a single cycle, we have demonstrated conditions that could lead to selective pulldown with a dCas9:EMX1 sgRNA of the on target site compared to an off target site (FIG. 27). Furthermore, pulldown of a FANCF genomic DNA-inspired library, leads to maximal enrichment of the on-target site relative to other sites (FIG. 28). The on target site Detailed knowledge of Cas9 binding will be especially valuable to mechanistically study improved properties of genetically engineered Cas9 variants, such as high fidelity Cas9. Importantly, off-target patterns of Cas9 fusion proteins (or fusion proteins with other DNA binding domains) with limited interdependence of effector domain and DNA binding domain might be mainly defined by the DNA binding properties of the fusion protein. Performing DNA pulldown experiments on pre-enriched oligonucleotide libraries might therefore contribute to our understanding of fusion protein off-target distributions. Furthermore, by conducting DNA binding studies of Cas9 (or other DNA binding protein domains) on a library with limited complexity, high quality binding data with little background noise can be obtained that could be subsequently used to extrapolate and predict binding of more complex libraries, such as the genome of a cell.

Example 7: Homing Endonuclease Selections

[0075] Homing endonucleases, such as I-Ppol, represent a group of naturally occurring nucleases that have longer base recognition motifs than the majority of restriction enzymes. Though homing endonucleases (also called meganucleases) do not have specificities that can be easily reprogrammed, if they target a genomic sequence of interest, they could be of research, commercial, or clinical use. Here, we show that we could adapt our in vitro selection to analyze the specificity profile of the I-Ppol homing endonuclease. We created an unbiased library of potential I-Ppol off-targets including all sites with up to 3 mismatches and single DNA/RNA bulges. The I-Ppol library contained 15533 members. I-Ppol selections enriched 501 of the 15533 library members (Table 3) while the intended, on-target site was ranked close to the top of the selection (28 out of 15533). Sequences with one mismatch or one insertion were the most enriched library members. Analysis of mismatch positions among top scoring I-Ppol off-target candidates revealed that certain positions within the recognition motif were more important for I-Ppol cleavage than others (FIG. 29). Especially positions 2, 13 and 14 seemed to be highly conserved and most important for I-Ppol mediated DNA cleavage. The adaptation of the in vitro selections to homing endonucleases demonstrates that the selections are broadly usable to analyze the off-target profiles of a variety of nucleases, including those that get cut to reveal sticky ends (like I-Ppol and Cas12a). I-Ppol leaves a 4 bp 3′-overhang, a DNA end configuration that is known to decrease the efficiency of off-target detection by existing methods such as GUIDE-seq or CIRCLE-seq. We therefore demonstrate that the in vitro selections can be used to analyze nucleases inducing staggered DNA breaks.

TABLE-US-00004 TABLE 3 Top 30 most-enriched sites in the post-selection library for I-PpoI on a unbiased DNA library. Found Found Alignment target # seqs_cleaved seqs_cleaved_rmv 1_0_1 CTATCTTAAGGTAGTC 97. 1507 1459 1_0_1 ACTCTCTTAAGGTAGC 98. 1329 1294 1_0_1 CTATCTTAAGGTAGCC 99. 1264 1235 3_0_0 CTACCTTAAGGTAGT 100. 1100 1071 3_0_0 CTACCTTAAGGGAGC 101. 1017 989 2_0_0 CTATCTTAAGGGAGC 102. 967 951 2_0_0 CTCCCTTAAGGGAGC 103. 960 923 1_0_1 CTATCTTAAGGTAGGC 104. 947 919 1_0_1 CTCTCTTAAGGGAGCC 105. 920 896 1_0_1 CTCTCTTAAGGTAGCT 106. 913 883 2_0_0 CTCCCTTAAGGTAGT 107. 885 866 0_0_1 CTCTCTTAAGGTAGTC 108. 865 842 1_0_1 CTCTCTTAAGATAGCC 109. 858 836 2_0_0 CTACCTTAAGGTAGC 110. 829 799 1_0_1 CTCCCTTAAGGTAGTC 111. 781 765 1_0_1 CTCTCATAAGGTAGTC 112. 744 724 1_0_1 CTCTCATAAGGTAGCC 113. 744 722 1_0_1 CTCTGTTAAGGTAGTC 114. 729 710 3_0_0 CTCCCTTAAGAGAGC 115. 732 702 1_0_1 CTCCCTTAAGGTAGCC 116. 713 694 1_0_1 CTCCCTTAAGGTAGAC 117. 709 689 1_0_0 CTCTCTTAAGGTAGT 118. 679 670 2_0_0 CTATCTTAAGGTAGT 119. 687 670 1_0_0 CTCTCTTAAGGGAGC 120. 673 653 3_0_0 CTATCTTAAGGGAGT 121. 651 639 1_0_1 CTCTGTTAAGGTAGCC 122. 650 636 0_0_1 CTCTCTTAAGGTAGGC 123. 650 633 0_0_0 CTCTCTTAAGGTAGC 124. 643 628 1_0_0 CTATCTTAAGGTAGC 125. 629 620 2_0_0 CTCTCTTAAGAGAGC 126. 634 620

#, SEQ ID NO:

[0076] Mostly closely matched off-target candidates were enriched to the top of the selection. However, the selections demonstrated that I-Ppol off-target candidates are exitent in abundance

Methods: Library Generation

[0077] Oligonucleotide library synthesis on high density chip arrays were purchased from Agilent.

Substrate Profiling Library:

[0078] 1) An oligonucleotide backbone was developed that had 50% GC content and no potential canonical PAM sequences (NGG for S. pyogenes Cas9).
2) 13-14 base pair barcodes were generated that were at least two substitutions away from all other barcodes, were 40-60% GC, and did not contain any canonical PAM sequences for the minimally unbiased libraries:
3) potential off-target sites were generated for all possible combinations of substitutions, insertions, and deletions for an SpCas9 target site (this can be variable):

TABLE-US-00005 single base pair single base pair substitutions deletions insertions <=3 0 0 <=1 1 0     0 2 0 <=1 0 1
4) barcodes/potential off-target sites for all i off target sites (I˜50,000) were combined into the backbone:

TABLE-US-00006 (SEQ ID NO: 127) GACGTTCTCACAGCAATTCGTACAGTCGACGTCGATTCGTGCT (barcode.sub.i)TTTGACATTCTGCAATTGCACACAGCGT (potential_off_target_site.sub.i)TGCAGACTGTAAG TATGTATGCTTCGCGCAGTGCGACTTCGCAGCGCATCACTTCA (barcode.sub.i)AGTAGCTGCGAGTCTTACAGCATTGC

Genome-Inspired Library:

[0079] 1) Potential off-target sites were generated with CasOffFinder according to the table below (these parameters can vary) and 20-113 bp (this can be variable) of genomic flanking sequence was added

TABLE-US-00007 single base pair single base pair substitutions deletions insertions <=6 0 0 <=4 <=2     0 <=3 0 <=2         4 0 1

[0080] For an EMX1 site, here is an example of the number of sequences present given the above parameters.

TABLE-US-00008 insertion (DNA deletion (RNA # of mismatches bulge) length bulge length) sequences 0 0 0 1 2 0 0 1 3 0 0 25 4 0 0 378 5 0 0 3903 6 0 0 30213 1 0 2 1 2 1 0 6 2 2 0 7 2 0 1 17 2 0 2 161 3 1 0 130 3 2 0 126 3 0 1 566 3 0 2 7579 4 1 0 2214 4 2 0 1942 4 0 1 8279 Total 55549
2) barcodes/potential off-target sites for all i off target sites (i˜50,000) were combined into the backbone as for the minimally unbiased library with maximal genomic flanking context:

TABLE-US-00009 (SEQ ID NO: 128) GACGTTCTCACAGCAATTCGT(barcode.sub.i)(flanking genomic context.sub.i)(potential_off_target_site.sub.i) (flanking_genomic_context)(barcode.sub.i)TGCGAGTCTTACA GCATTGC

[0081] Constant backbone sequence can be increased as the flanking genomic context is varied.

[0082] For example, with 10 bp genomic flanking sequence on both sides:

TABLE-US-00010 (SEQ ID NO: 129) GACGTTCTCACAGCAATTCGTACAGTCGACGTCGATTCGTGCT (barcode.sub.i)TTTGACATTCTGCAATGT (flanking_genomic_context.sub.i) (potential_off_target_site.sub.i) (flanking_genomic_context)(AAGTATGTATGCTTCGCGCAGTGC GACTTCGCAGCGCATCACTTCA(barcode.sub.i)AGTAGCTGCGAGTCTTACA GCATTGC

Other Library Generation Strategies:

[0083] incorporate population based SNPs into genomic sequences [0084] generate libraries based on only coding DNA sequences [0085] generate libraries of sites that are oncogene hotspots or tumor suppressor genes

[0086] The following are examples of methods using off-target libraries constructed using the above principles.

Method for In Vitro Selection of Cleaved Library Members

[0087] 1. Library Amplification

[0088] We amplify the oligonucleotide libraries using primers that bind to the constant flanking regions that are found in all library members. These primers contain 5′ prime overhangs that introduce additional length and a unique molecular identifier. The libraries are amplified using the following protocol using 2 μl of 5 nM input library.

TABLE-US-00011 SV (2l of 5 nM input library) 2 Thermopol buffer 5 Taq Polym. 0.25 dNTP 10 mM 1 KP_extension_new_fw* 1 KP_extension_new_rev* 1 H2O 39.75 RV 50 PCR program cycles 12 ID 95 30 D 95 20 A 50 15 E 68 1 FE 68 30 min SV Samples Volume RV Reaction Volume *KP_extension_new_fw, Primer Sequence: GCTGACTAGACACTGCTATCACACTCTCTCANNNNNNNNAGACGTTCTCACAGCAATTCG (SEQ ID NO: 130) *KP_extension_new_rev, Primer Sequence: GCGTAATCACTGATGCTTCGTAAATGAGACANNNNNNNNTGCAATGCTGTAAGACTCGCA (SEQ ID NO: 131)

[0089] 2. DNA Purification:

[0090] DNA purification with AMPure magnetic beads at a sample:bead ratio of 0.9× according to manufacturer's protocol.

[0091] 3. Enzymatic Incubation:

[0092] Incubation of 300 ng of the chip-synthesized library with protein of interest at varying enzyme concentrations and incubation times. In most cases (Cas9, Cas9HF, BE3, ABE) it is sufficient to perform an 1-2 h incubation of the enzyme in activity buffer on 300 ng of oligonucleotide library at a molar ratio of 10:10:1 for protein, sgRNA and DNA substrate, respectively. Depending on the specific protein function, these parameters may need to be optimized.

[0093] 4. Optional DNA Nicking:

[0094] Depending on the analyzed protein, enzymatic incubation may not result in the creation of a DNA double strand break (DSB). In the case of BE3 and ABE both enzymes merely nick on strand of DNA while base editing the other. By employing USER enzyme or Endonuclease V for BE3 and ABE, repectively, it is possible to convert this DNA nick into a staggered DSB (see FIGS. 8 and 9). To achieve this, bead-purified DNA from step 4 is incubated with USER enzyme or Endonuclease V for one hour at 37° C. in their respective activity buffer.

[0095] 5. DNA Purification:

[0096] DNA purification with AMPure magnetic beads at a sample:bead ratio of 1.5× according to manufacturer's protocol.

[0097] 6. Optional DNA Blunting:

[0098] If an additional nicking step (5) was required, the staggered DSB will be blunted by incubation with Phusion Polymerase for 20 min at 72° C. and then cooled to 4° C.

[0099] 7. DNA Purification:

[0100] DNA purification with AMPure magnetic beads at a sample:bead ratio of 1.5× according to manufacturer's protocol.

[0101] 8. Adapter Ligation:

[0102] Next, half functional Y-shape adapters are ligated to the blunted DNA from step 7. To achieve this, we supply adapter in 10-fold molar excess over library fragments and ligate using the NEB quick ligation kit, incubating the reaction at 25° C. for 10 min.

[0103] 9. Gel Purification:

[0104] Next, we perform a gel purification of the ligation reaction by employing a 2.5% Agarose gel. The electrophoresis is performed at 120 Volt for 1 hour. After 1 hour the sample containing lanes are excised at around 180 bp fragment size and DNA is extracted using the Qiagen gel extraction kit according to manufacturer's protocol.

[0105] 10. PCR-Amplification:

[0106] The eluate from step 9 is subsequently used as input for two PCR reactions that amplify the Protospacer-adjacent and PAM-adjacent site of cut library members. The primers used in this PCR contain 5′ overhangs that can be subsequently used to append Illumina sequencing barcodes. Optionally, QPCR can be performed to determine the minimum number of PCR cycles required. The PCRs are performed using the following parameters:

TABLE-US-00012 Sample Volume 6 Phusion High Fidelity Buffer 5X 10 Phusion Polymerase 0.5 dNTP 10 mM 1 PrimerA 2.5 PrimerB 2.5 H2O 27.5 PCR program cycles 25-35 ID 98 30 D 98 10 A 65 20 E 72 5 FE 72 5 min

[0107] 11. DNA Purification:

[0108] DNA purification with AMPure magnetic beads at a sample:bead ratio of 1.5× according to manufacturer's protocol.

[0109] 12. Quality Control Using Capillary Electrophoresis:

[0110] Quality control is performed by examining the PCR products via capillary electrophoresis.

[0111] 13. PCR-Based NGS Library Preparation:

[0112] Sequencing adapters are appended to the PCR products from step 12 by performing a PCR with primers containing Illumina sequencing adapters. The PCRs are performed using the following parameters:

TABLE-US-00013 Sample Volume 50 ng total Phusion High Fidelity Buffer 5X 10 Phusion Polymerase 0.5 dNTP 10 mM 1 IndexPrimerA 2.5 IndexPrimerB 2.5 H2O Ad 50 PCR program cycles 10 ID 98 30 D 98 10 A 65 30 E 72 35 FE 72 10 min

[0113] 14. DNA Purification:

[0114] DNA purification with AMPure magnetic beads at a sample:bead ratio of 1.5× according to manufacturer's protocol.

[0115] 15. Next Generation Sequencing on Illumina Sequencers:

[0116] The DNA libraries from step 14 are quantified via digital droplet PCR and sequenced on Illumina sequencer's according to the manufacturer's protocol.

Method for Enrichment of DNA Binding Sites by Pulldown

[0117] 1) Resuspend Snap Capture Beads (NEB) [0118] 2) Pipette 80 uL of the beads to a new 1.5 mL Eppendorf tube [0119] 3) Place tube in a magnetic particle separator and discard the supernatant [0120] 4) Add 1 mL of Immobilization Buffer (20 mM HEPES, 150 mM NaCl, 0.5% Tween20, 1 mM DTT, pH 6.5) and vortex gently [0121] 5) Place tube in a magnetic particle separator and discard the supernatant [0122] 6) Prepare the protein: Add Engen Spy dCas9 (SNAP-tag) (NEB) (4.5 uL of 20 uM per pull down reaction) to 500 uL of Immobilization Buffer [0123] 7) Add the diluted protein to the beads and mix well via pipetting [0124] 8) Incubate for 1 hour shaking at room temperature [0125] 9) Place tube in a magnetic particle separator and discard the supernatant [0126] 10) Wash the beads. Add 1 mL of Immobilization Buffer, pipette mix well, and then place the tube in a magnetic particle separator and discard the supernatant [0127] 11) Repeat step 10 twice more for a total of 3 washes. Perform the last wash with Immobilization Buffer with 10 ug/mL Heparin [0128] 12) Resuspend the beads in 45 uL of immobilization buffer per pull down [0129] 13) Mix the following:

TABLE-US-00014 Component Amount for 1 Pull Down Reaction Water Add enough to make the final volume 60 uL after adding everything including 0.9 pmol of Library 10X Immobilization Buffer + 6 uL 100 ug/mL Heparin gRNA 3500 ng Engen Spy dCas9 45 uL (SNAP-tag) + Magnet Beads [0130] 14) Incubate for 25 deg C. for 10 min [0131] 15) Add 0.9 pmol of library [0132] 16) Incubate at 37 deg C. for 30 min [0133] 17) Place the tube on a magnetic bead separator and discard the supernatant [0134] 18) Wash the beads 5 times with 200 uL of Immobilization Buffer with 10 ug/mL Heparin [0135] 19) Add 50 uL of water and 2 uL of Proteinase K and incubate at room temperature for 10 min while shaking [0136] 20) Clean up the pulled down product with DNA purification beads (for example, Ampure) and elute in 10 uL of 0.1× Buffer EB (QIAgen)

Other Embodiments

[0137] It is to be understood that while the invention has been described in conjunction with the detailed description thereof, the foregoing description is intended to illustrate and not limit the scope of the invention, which is defined by the scope of the appended claims. Other aspects, advantages, and modifications are within the scope of the following claims.