Efficient Immune Cell Genome Engineering with Improved CRISPR Editing Tools

CRISPR (clustered regularly interspaced short palindromic repeats)-based methods have revolutionized genome engineering and the study of gene-phenotype relationships. However, modifying cells of the innate immune system, especially macrophages, has been challenging because of cell pathology and low targeting efficiency resulting from nucleic acid activation of sensitive intracellular sensors. Likewise, lymphocytes of the adaptive immune system are largely refractory to CRISPR-enhanced homology-directed repair (HDR) due to inefficient or toxic delivery of donor templates via transient transfection methods. To overcome these challenges and limitations, we developed three improved methods for CRISPR-based genome editing using a hit-and-run transient expression strategy to minimize off-target effects and generate more precise genome editing. Overall, our enhanced CRISPR tools and strategies designed to tackle both murine and human immune cell genome engineering are expected to be widely applicable not only in hematopoietic cells but also other mammalian cell types of interest. All animal experiments were done in accordance with the guidelines of the NIAID/NIH Institutional Animal Care and Use Committee.


Introduction 37
The bacterial innate defense system CRISPR was discovered more than 3 decades ago (1), 38 yet this game-changing genome engineering technology was not harnessed for human gene 39 editing applications until a series of landmark studies published in 2013 (2-4). Since then, there 40 has been an exponential increase of CRISPR-related publications, mainly focusing on the use of the 41 Cas9 nuclease that comes with unpredictable off-target effects (5-7), hindering its direct 42 application to the clinic. Many research groups have developed novel methods to minimize these 43 potential side-effects (8)(9)(10)(11)(12)(13)(14). The Cas9 nickase method is a promising approach due to its strict 44 pairing distance requirement, which makes off-target pairing very unlikely to occur and therefore 45 from the human GeCKOv2 library A (33). The PX461 and PX462 plasmids were gifts from Feng 161 Zhang (Addgene plasmid #48140 & #48141) (35). The pKLV-BFP plasmid was a gift from Kosuke 162 Yusa (Addgene plasmid #50946) (36). All lentiviral and other gRNA expression plasmids were 163 prepared by sticky-end ligation before any further modification described. The PX462-hBCL10-164 gRNA-A1 plasmid was constructed by Gibson Assembly in the PciI-linearized PX461 backbone, 165 followed by sub-cloning of the double gRNA expression fragment into the PX462 backbone with 166 PvuI-XbaI sticky-end ligation. CRISPRa gRNAs were adopted from the Human CRISPR Activation 167 Library (SAM -2 plasmid system) from Feng Zhang (32). 168 Production, purification and concentration of lentiviral particles. Lentiviral particles were 170 produced in HEK293T cells following transient transfection of the packaging plasmids pVSVg 171 (Addgene plasmid #8454) and psPAX2 (Addgene plasmid #12260), and the lentiviral expression 172 plasmid in a 1:10:10 ratio using Lipofectamine 3000. Lentiviral supernatant was harvested 44 to 173 52 h post-transfection and spun at 500 X g for 5 minutes to clear cell debris. NIL particles were 174 produced the same way except with the use of the psPAX2-D64V plasmid (Addgene plasmid 175 #63586) (29) instead of the psPAX2 plasmid. All lentiviral particles other than those used in the 176 RAW cell transductions and Figure 2B (top panel) were further purified and concentrated with the 177 Lenti-X Concentrator reagent (Takara) according to the manufacturer's instructions. 178 Cell culture, transfection and antibiotic selection conditions. HEK293T, RAW264.7 and Jurkat 180 cell lines were purchased from ATCC. Cell culture was performed in a humidified environment 181 with 5% CO 2 at 37°C. Both HEK293T and RAW264.7 cell lines were cultured in DMEM 182 in RPMI supplemented with 10% FBS, penicillin/streptomycin, glutamine and 50 M -184

mercaptoethanol (complete RPMI). For plasmid transfections of HEK293T and Jurkat cells, 185
Lipofectamine 3000 (ThermoFisher) was used according to manufacturer's protocol. Puromycin 186 (0.5 to 3 g/mL) or blasticidin (1 to 3 g/mL) was added to cell culture medium 1 to 2 days after 187 transduction of both RAW and Jurkat cell lines. 188 Genomic DNA extraction, T7EN assay and sequencing. Genomic DNA was extracted from RAW,190 HEK293T or Jurkat cells using the QIAmp DNA Mini Kit (Qiagen) following the manufacturer's 191 protocol. Genomic regions of interest were then amplified by PCR for follow-up analysis. For the 192 T7EN assay, equal amounts of genomic PCR products were used within each comparison group for 193 the hybridization reaction in a thermocycler programmed to 95°C for 10 minutes, -2°C per second 194 ramp to 85°C, -0.1°C per second ramp to 25°C, and then held at 4°C until a subsequent T7 195 Endonuclease I (NEB) treatment at 37°C for 30 to 90 minutes. 25 mM final concentration of EDTA 196 was added to each reaction before agarose gel electrophoresis analysis. Indel percentage was 197 determined as described previously (8). For the genomic DNA sequencing, each DUSP gene locus 198 surrounding the CRISPR gRNA target site (approximately -500 bp to +500 bp) was amplified and 199 cloned into a XhoI-NotI linearized pcDNA3.1 backbone by Gibson Assembly. Sequencing was 200 performed with the BGH-R (Table 1)  were stimulated with TLR4 ligand Kdo2-Lipid A (KLA, Avanti Polar Lipids). Ligand addition was 215 staggered so that for all time points, cells were fixed at the same time by addition of 216 paraformaldehyde to cell cultures at a final concentration of 1.6% (10 minutes). After one wash 217 with PBS 1% FBS, cells were gently harvested from plates by scraping, were permeabilized 218 overnight using ice cold MeOH at -20°C, blocked using 5% goat serum and Fc receptor specific 219 antibody, and stained for 1 h at room temperature with anti-p38 (phospho-Thr180/Try182, BD 220 36/p38) antibody. 221 222 Cell sorting, FACS analysis and live cell confocal imaging. Cell sorting was conducted by the 223 Flow Cytometry Section, Research Technologies Branch of NIAID. FACS analysis data were 224 collected using a BD LSRII or BD Fortessa Cell Analyzer and further processed with FlowJo. 225 Human PDCD1 or PDL1 cell surface protein staining was analyzed using anti-PD-1 (ThermoFisher 226 clone MIH4) or anti-PD-L1 (ThermoFisher clone MIH1) antibody respectively. Live cell confocal 227 time-lapse imaging data were collected using a Leica SP8 microscope with a 63x NA 1.4 oil 228 chamber (ThermoFisher 155411) was coated with poly-D-lysine (Sigma P7280) for 1 h at 37°C 230 and washed twice with PBS. Cells were imaged in a heated 37°C environment with 5% CO 2 . 231 Imaging data were processed by Imaris (Bitplane). 232 233 Chemical reagents, lysis buffer and western blots. Doxycycline hydrochloride (Sigma D3447) 234 was reconstituted in DMSO at 50 mg/mL and stored at -20°C. Jurkat T cells were stimulated with 235 50 ng/mL PMA (Santa Cruz Biotechnology) and 1 M ionomycin (Sigma) (37). RAW cells were 236 lysed in a buffer containing 50 mM Tris pH7.5, 150 mM NaCl, 1 mM EDTA, 10% glycerol and 1% 237 Igepal. Western blots were analyzed using anti-DUSP3 antibody (Abcam clone EPR5492), anti-238 DUSP1 antibody (Millipore 07-535), and anti-FLAG antibody (Santa Cruz Biotechnology sc-239 166355). 240 241 Human PBMC isolation and transduction. Human whole blood from healthy anonymous 242 volunteer donors was purchased from a NIH blood bank. This was exempted from the need for 243 informed consent and Institutional Review Board review, as determined by the NIH Office of 244 Human Subjects Research Protection. Human peripheral blood mononuclear cells (hPBMC) were 245 isolated from whole blood by Ficoll density gradient separation, cultured in complete RPMI 246 medium mentioned above, and stimulated with 1 g/mL anti-CD3 (BioLegend 300334) and 1 g/mL 247 anti-CD28 (BioLegend 302944) soluble antibodies for 18 h. Purified and concentrated lentiviral 248 particles were then added to the hPBMC culture, with 8 g/mL Polybrene (Millipore Sigma TR-1003) 249 and 100 U/mL recombinant human IL-2 (TECIN (teceleukin)). These cells were then spun at 400 X g 250 for 90 minutes in a pre-warmed centrifuge at 34 °C. 251

Results 253
Dual-color Inducible CRISPR-Cas9 Editing System in RAW264.7 Macrophages. 254 To achieve a controllable balance between on-target and off-target mutations induced by 255 CRISPR, we adopted the doxycycline-inducible pCW-Cas9 system to enable precise manipulation 256 of the genome editing time period. In pCW-Cas9-transduced RAW macrophages, a range of Cas9 257 protein expression was induced by doxycycline dose titration (Figure 1 -figure supplement 1A). 258 We modified this existing system to enable single-cell flow cytometric analysis and developed the To test this method, we targeted a family of dual-specificity phosphatases (DUSPs) in RAW 271 cells. The RAWiCE cells transduced with LGB targeting each DUSP genomic locus were first 272 selected by blasticidin to maximize the gRNA-expressing (tBFP+) population. Following 273 doxycycline induction at 1.0 µg/mL for 6 days, we analyzed DUSP3 expression by intracellular 274 DUSP3 protein expression ( Figure 1F) within the CRISPR-targeted EGFP+ tBFP+ population, 276 indicating that the majority of targeted cell clones ( Figure 1E) were indeed complete knockouts 277 (KO) of DUSP3 expression. Due to limited antibody availability, we performed T7 Endonuclease I 278 (T7EN) assays to confirm the presence of insertion/deletion (indel) mutations at the targeted 279 genomic loci ( Figure 1G). Since the T7EN assay does not yield the actual DUSP3 KO percentage, from HEK293T packaging cells seen as a dramatic post-transduction decline in the live BMDM 296 percentage ( Figure 2B). To optimize BMDM survival, we further purified and concentrated 297 approach improved the live BMDM percentage as the LGB titration reached a peak transduction 299 efficiency ( Figure 2B). Using this improved transduction method for LGB expression in Cas9+ 300 EGFP+ BMDM, we performed intracellular protein staining and validated the editing efficiency of 301 DUSP3 gRNA ( Figure 2C). Consistent with previous report using BMDM from DUSP1-KO mice (38), 302 we observed sustained TLR4-induced p38 phosphorylation (p-p38) in BMDM transduced with the 303 LGB-DUSP1 gRNA expression, as compared to the EV (empty vector) and the DUSP4 controls 304 ( Figure 2D). Increased p-p38 phenotype generally correlated with the LGB-DUSP1 gRNA 305 expression level, as indicated by tBFP fluorescence in Cas9+ EGFP+ BMDM ( Figure 2E). Our 306 results suggest that TLR4-induced p38 is differentially regulated between primary BMDM and the 307 transformed RAW macrophages and illustrate the utility of our method for study of these 308 otherwise hard-to-modify cells. 309 310

An Efficient Selection Strategy for CRISPR-induced HDR Clones in Jurkat T Cells. 311
We next turned to new strategies for editing the genomes of lymphocytes. To minimize off-312 target effects, we designed a pair of gRNAs for targeting with the Cas9-D10A nickase (Cas9n) 313 strategy, choosing the human STAT3 locus as an exemplar ( Figure 3A). This pair of gRNAs 314 together with Cas9n expression in HEK293T cells was confirmed to create indel mutations at the 315 targeted STAT3 loci, as measured by the T7EN assay ( Figure 3B) following co-transfection of the 316 PX461 and the pKLV-BFP plasmids. However, our goal was more ambitious than just achieving 317 gene expression loss in these cells. In addition to expression of Cas9n and a pair of gRNAs lying in 318 close proximity, a DNA donor template is required to induce precise knock-in modification via the 319 HDR repair mechanism. The STAT3 HDR donor plasmid ( Figure 3C) was designed with 2000-bp 320 followed by the T2A sequence and the mNeonGreen (mNG) fluorescent protein-coding gene. 322 Following co-transfection of PX461-STAT3-gRNA-A, pKLV-BFP-STAT3-gRNA-1 and STAT3 HDR 323 donor plasmids, approximately 6e5 total (<0.5% co-transfection efficiency) Jurkat T cells were 324 selected with puromycin and separated by dilution cloning before further validation ( Figure 3D). 325 Based on the low HDR efficiency for large fragment insertion (<10%), it is technically challenging 326 to isolate this tiny (<0.05% or <300 cells) population of successful HDR cell clones without using 327 any selection method. After puromycin selection, flow cytometric analysis of 40 puromycin-328 resistant Jurkat cell clones revealed that most of the selected population had acquired mNG 329 expression to various extents ( Figure 3E). Further genomic PCR analysis of 7 clones that express 330 medium to high level of mNG showed that 6 out these 7 clones were indeed complete knock-in for 331 all STAT3 alleles ( Figure 3F). Based on these results, we estimated the clonal distribution of the 332 parental mNG-STAT3 Jurkat cell line ( Figure 3G). The high percentage (82.5%) of partial or 333 complete mNG-STAT3 knock-in clones suggests that this kind of antibiotic selection strategy 334 efficiently enriches the desired HDR cell clones. 335 We extended studies of this method by targeting the human BCL10 locus to express a 336 fluorescent protein-tagged fusion protein, using 6 nickase gRNA pairs ( Figure 4A) and measuring 337 the targeting efficiency by T7EN assay ( Figure 4B) as in Figure 3B. Despite the higher indel 338 frequency induced by the D3 gRNA combination ( Figure 4A & 4B), we selected the highest-scoring 339 (crispr.mit.edu) A1 gRNA combination to minimize potential off-target effects. To enhance co-340 transfection efficiency, we created a double gRNA expression plasmid ( Figure 4C) based on the 341 PX462 backbone. Furthermore, we attempted to minimize the genome editing footprint by 342 incorporating loxP sites flanking the PuroR-T2A selection cassette in the BCL10 HDR donor 343 genome editing with the same strategy can be applied because puromycin sensitivity is restored 345 ( Figure 4E). After plasmid co-transfection and puromycin selection, most Jurkat cells had 346 acquired mNG-BCL10 expression ( Figure 4F) and showed dispersed protein localization, as 347 compared to the nuclear H2B-BFP marker (data not shown). After NIL-CRE (non-integrating 348 lentiviral particles; see Figure 5A for details) treatment, we observed enhanced mNG-BCL10 349 expression level ( Figure 4G) and a substantially higher frequency of mNG-BCL10 aggregate 350 formation (39) upon antigen receptor stimulation (data not shown). Puromycin re-treatment 351 killed approximately 99.8% of the NIL-CRE-treated mNG-BCL10 Jurkat cells but did not affect the 352 survival of the original mNG-BCL10 line (data not shown). Finally, based on this NIL-CRE-treated 353 or Puro-T2A-excised mNG-BCL10 Jurkat parental cell line, we repeated STAT3 genome editing as 354 in Figure 3 but using a modified STAT3 HDR donor plasmid ( Figure 4H) in which the mScarlet 355 element replaces the mNeonGreen element ( Figure 3C). Following plasmid co-transfection and 356 puromycin selection, more than 75% of mNG-BCL10+ Jurkat cells had gained the targeted 357 mScarlet-STAT3 expression ( Figure 4I). With minimal off-target editing concern, these results 358 demonstrated a highly efficient yet sustainable selection method to precisely enrich targeted 359 knock-in cell population from very few recombinant clones created by CRISPR-mediated HDR. 360 361

All-in-one Non-integrating Lentiviral Delivery of the CRISPR-HDR Blueprint. 362
As an alternative approach, we sought to deliver Cas9, gRNA, and the HDR donor template 363 to the entire T cell culture in a transient transfection manner. We therefore turned to the use of 364 integrase-deficient lentiviral particles. We packaged the Lenti-CMV-mCherry-P2A-CRE viral 365 vector either with the wild-type integrase in the psPAX2 plasmid to produce integrating lentiviral 366 integrating lentiviral (NIL) particles. While the IL transduction created stable mCherry expression 368 in Jurkat T cells, the NIL delivery failed to integrate into the genomes and led to transient mCherry 369 expression ( Figure 5A). To test whether NIL delivery of the entire CRISPR-HDR instruction can 370 generate targeted knock-in cells, we assembled the LentiCRISPRv2B-HDR-mCherry-2A-hCD3E 371 viral vector ( Figure 5B) and examined two gRNAs targeting the human CD3E locus ( Figure 5C). 7 372 days after the NIL-CRISPR-Cas9-HDR delivery, we indeed observed a distinct population of 373 mCherry+ Jurkat cells by flow cytometry ( Figure 5D). Using primers that distinguish the 374 endogenous locus from randomly-integrated HDR template sequences, genomic PCR analysis 375 confirmed the presence of targeted knock-in alleles within the mCherry-sorted Jurkat population 376 ( Figure 5E). 377 Using NIL delivery of the LentiCRISPRv2S-HDR-tBFP-hRelA viral vector ( Figure 5F), we 378 attempted to generate a TagBFP-RelA fusion knock-in Jurkat cells, employing a gRNA targeting the 379 human RelA locus ( Figure 5G). This HDR attempt yielded a distinct and stable tBFP+ Jurkat 380 population that was further purified by FACS ( Figure 5H). We used live cell confocal time-lapse 381 imaging to visualize the RelA nuclear translocation response to PMA-ionomycin stimulation 382 ( Figure 5I). As expected, in the steady-state tBFP-RelA was mostly excluded from the H2B-383 sfCherry-defined nuclear region. Upon NF-B pathway activation, a rapid but transient increase in 384 the nuclear to cytoplasmic ratio of tBFP-RelA was observed in approximately 85% of the Jurkat 385 population ( Figure 5I). To distinguish targeted genomic knock-in from random lentiviral Cas9 expression allows more precise control of the genome editing time window and therefore 416 minimizes long-term off-target effects that cannot be avoided by constitutive Cas9 expression in 417 many lentiviral-based CRISPR expression systems. For gRNAs that efficiently knockout essential 418 cell survival genes or oncogenes, percentage changes in EGFP+ tBFP+ cells over doxycycline 419 treatment time help quantify the impacts of these gRNAs. 420 The high genome editing efficiency and the rapid single-cell phenotyping option of our 421 dual-color inducible CRISPR system in RAW cells makes this platform suitable for high-throughput 422 genome-wide gRNA library screening. Our inducible CRISPR system allows for stable uninduced 423 cell lines to be frozen for long-term storage and future analysis. Furthermore, we have optimized 424 these methods for use in primary immune cells. Lentiviral or retroviral supernatant produced 425 from HEK293T packaging cells often influences target cell growth rate and morphology, especially 426 when a high volume-ratio of viral supernatant to cell culture medium is applied to the target cell 427 line. Our data illustrate that further purification of lentiviral particles helps minimize BMDM cell 428 death. Thus, our methods achieve high transduction efficiency without sacrificing cell health in 429 culture. With these approaches, we were able to identify Cas9 and gRNA double positive 430 macrophages, using both the RAW murine cell line and primary BMDM. Interestingly, these 431 studies revealed distinct MAPK regulation between these cells. CRISPR deletion of the 432 phosphatase DUSP1 in BMDM resulted in sustained p38 phosphorylation, consistent with 433 published studies in BMDM generated from DUSP1 deficient mice (38). In contrast, DUSP1 434 deletion in RAW cells, using the same gRNA, did not yield detectable changes in p38 activation. 435 compared to BMDM (40). 437 The Cas9 nickase strategy requires two gRNA targets to be in close proximity to efficiently 438 induce DNA double-strand break repair mechanisms (8,9). Therefore, it is extremely unlikely to 439 find another off-target pairing site in the genome. Our data support previous reports that the pair 440 of Cas9 nickases are optimized with minimized offset of the gRNA pairs (8,9). Without including 441 an upstream promoter in the donor template design, our PuroR selection strategy reduces the 442 probability of selecting randomly integrated clones. Obviously, this strategy relies on genomic loci 443 that are well expressed to generate sufficient expression of the drug resistance gene. Among the 444 target genes that we have tested, we observed that this strategy worked best with the N-terminal perfect delivery vehicles for human T cell genome engineering. In Jurkat cells, lentiviral delivery is 459 much more efficient than any existing transfection method including electroporation, which often 460 leads to substantial cell death upon electric shock and reduces cell growth and overall protein 461 production for the surviving population. In addition, lentiviral particles are easy to prepare by 462 most laboratories, carry double to triple the payload as compared to the AAV system, and 463 therefore enable all-in-one design and delivery of a Cas9, gRNA expression and HDR donor 464 template. Although there is now an existing report (41) Video 1. Live cell confocal time-lapse imaging data of tBFP-RelA translocations following 714