Reprogramming human B cells with custom heavy chain antibodies

We describe a genome editing strategy to reprogram the immunoglobulin heavy chain (IgH) locus of human B cells to express custom molecules that respond to immunization. These heavy chain antibodies (HCAbs) comprise a custom antigen-recognition domain linked to an Fc domain derived from the IgH locus and can be differentially spliced to express either B cell receptor (BCR) or secreted antibody isoforms. The HCAb editing platform is highly flexible, supporting antigen-binding domains based on both antibody and non-antibody components, and also allowing alterations in the Fc domain. Using HIV Env protein as a model antigen, we show that B cells edited to express anti-Env HCAbs support the regulated expression of both BCRs and antibodies, and respond to Env antigen in a tonsil organoid model of immunization. In this way, human B cells can be reprogrammed to produce customized therapeutic molecules with the potential for in vivo amplification.


Introduction 29
Monoclonal antibodies are important therapeutics that allow specialized antibody designs 30 not achieved by immunization, such as targeting self-antigens or highly variable pathogens. 1,2 31 However, since their administration for chronic conditions can be burdensome and expensive, 32 gene and cell therapies are also being considered for delivery. 3,4 AAV vectors have been shown 33 to support expression of custom antibodies from skeletal muscle, 4,5 although such approaches 34 have been hindered by low expression levels or immunogenicity in large animal 6,7 and human 35 studies. 8,9 An alternative approach is to use genome editing to express custom antibodies from 36 the natural immunoglobulin (Ig) locus in B cells. 10 It is hypothesized that this would have the 37 advantage of maintaining natural aspects of antibody production in the engineered cells, in 38 particular the response to the matched antigen. The VH plus L chain insertion approach presents certain challenges. It requires expression 50 of both H and L chain sequences from the same cassette, which can be achieved by including a 51 self-cleaving peptide 11,12,14 or a long flexible linker. 11,13 More significantly, edited cells could still 52 express endogenous H and L chains from unedited loci that could mispair with the engineered antibody chains and create unwanted antigen specificities, although a separate editing step to 54 disrupt at least the endogenous Igk L chain can also be included. 11,14 Finally, since the Fc  We evaluated the HCAb engineering strategy in human B cell lines using ARDs based on 109 the J3 24 and A6 25 camelid VHH domains, both of which recognize the HIV-1 gp120 Env protein.

110
Plasmid homology donors, compatible with the sg05 target site, were co-electroporated into Raji 111 B cells with sg05 Cas9 RNPs. Specific HCAb-BCR expression was evaluated by flow cytometry 112 for surface IgG, which is not normally expressed by Raji cells, 26 and by binding to recombinant 113 gp120. Cells receiving both RNPs and homology donors had higher levels of surface staining than 114 control cells or cells receiving the homology donors alone, and the correlation between IgG and 115 gp120 staining suggested both labels were binding the same BCR molecule (Fig. 2a). The 116 relatively low editing rates we obtained reflect the inefficiency of plasmid homology donors in the 117 Raji cell line, and similar editing rates were achieved with donors containing a control GFP 118 expression cassette, or in the alternate Ramos B cell line (Extended Data Fig. 2a-b).

119
To further evaluate HCAb editing, edited Raji and Ramos cells were sorted based on 120 expression of GFP or surface IgG, as appropriate (Extended Data Fig. 2c-d). Specific in-out PCR 121 and Sanger sequencing confirmed site-specific insertion for all 3 inserts in Raji cells (Fig. 2b, 122 Extended Data Fig. 2e). Since these cell lines do not secrete human IgG,26,27 HCAb secretion 123 could be measured using an anti-IgG ELISA. Secreted IgG was detected in the supernatants of 124 J3 or A6 edited cells, but not GFP-edited cells (Fig. 2c, Extended Data Fig. 2f). Together these 125 data support that the chimeric HCAb transcripts arising from the inserted cassettes could be 126 alternatively spliced to produce both membrane-bound and secreted isoforms.

127
Finally, the functionality of the secreted HCAbs was evaluated. Raji cells seeded in equal 128 numbers secreted higher levels of J3 than A6 HCAbs (Fig. 2c). When normalized, we observed 129 similar HIV neutralization activities for HCAbs originating from either edited Raji or Ramos cells, or the matched recombinant HCAbs produced from transfected 293T cells (Fig 2d,e). Moving 131 forward, we chose to focus on J3 due to its higher secretion levels and superior anti-HIV potency 132 across broad panels of HIV strains. 25,28 133 We next tested the hypothesis that the HCAb design would bypass the concern of H and 134 L chain mispairing. We co-expressed the J3 HCAb in 293T cells with both the H and L chains of 135 a conventional monoclonal human antibody, or its L chain only, and evaluated interactions using 136 ELISAs and antibodies specific for VHH domains or human L chains (Extended Data Fig. 3). We 137 did not observe any such interactions for the HCAb/L chain combinations. In contrast, we did 138 detect the expected H chain heterodimers when the HCAb was co-expressed with the H+L 139 complete antibody combination. However, such pairings are expected to create bispecific 140 antibodies that keep intact each separate antigen recognition domain, limiting the possibilities for 141 novel specificities or self-reactivity. Together, these data support the idea that HCAb expression 142 in a B cell that also expresses endogenous H and L chains will provide a better safety profile than  both H and L chains and allows antibody affinity maturation. 29 Although the mechanism whereby 150 AID targets antibody variable domains is complex and not well-understood, the link to transcription 151 start sites in the Ig locus 30 suggested that promoter-driven expression of inserted ARDs in HCAb-152 edited B cells could still support SHM, despite being inserted in a constant region (Fig 1a).

153
To evaluate this possibility, we cultured J3-edited Raji cells that constitutively express 154 AID 31 for 6 months without selection. We looked for evidence of SHM by deep sequencing at 155 different timepoints, and for changes in gp120 binding that could reflect the acquisition of mutations. Over time, there was a progressive loss of gp120 binding for both the BCR and 157 secreted J3 isoforms, along with a reduction in J3 HCAb secretion, suggesting mutations in the 158 J3 sequence that altered functionality and expression ( Fig. 3a-c, Extended Data Fig. 4a). Deep 159 sequencing confirmed that the J3 sequence in Raji cells accumulated mutations over time, and in 160 particular within the CDR3 sequence that is generally considered to be the most important region 161 for antigen binding (Fig. 3d). The observed mutations were predominantly at cytosine residues 162 within predicted AID hotspots (Fig. 3e,f), consistent with AID-mediated mutagenesis.

163
Interestingly, our data suggests that AID activity was also influenced by the specific 164 sequence of the inserted DNA cassette. Raji cells containing a control GFP cassette did not 165 develop such a mutational signature, and neither did the A6 VHH domain (Extended Data Fig.   166 4b,c). This was not due to a lack of AID hotspot motifs, as these were present in all 3 sequences 167 evaluated (Extended Data Fig. 4d,

179
We evaluated 3 different B cell stimulation conditions: a 3-step differentiation protocol (DP) 180 we and others have previously used for human B cell genome editing, 33,34 an anti-RP105 181 antibody, 11,14 and a commercial B cell activation cocktail (BAC) (Supplementary Fig. 4). Each protocol was assessed in 3 different basal media for cell expansion, viability, cell size and IgG 183 secretion (Extended Data Fig. 5a-e). As expected, the DP treatment resulted in little expansion 184 of cell numbers but robust ASC differentiation and IgG secretion. Unexpectedly, in our hands anti-185 RP105 did not support either B cell survival or expansion. In contrast, BAC stimulation in XF 186 media drove robust >100-fold B cell expansion, with or without FBS.

187
We further analyzed the phenotypes of the B cells in the different cultures. We found that 188 BAC culture drove activation and class switching from naïve to IgM single-positive cells, double-189 negative cells (likely memory or ASC precursors), 35 and memory B cells (Extended Data Fig. 5f).

190
Further, cells initially activated with BAC could still be differentiated towards ASCs upon 191 subsequent DP treatment (Extended Data Fig. 5f Fig. 6a). With 202 primary B cells, AAV6-J3 donors supported editing rates of approximately 35%, measured by flow 203 cytometry for surface J3-BCR expression, or ddPCR for edited IGHG1 alleles (Fig. 4a-c). Site-204 specific insertion of the J3 cassette was also confirmed by in-out PCR and Sanger sequencing 205 (Extended Data Fig. 6b). The edited primary B cells retained the ability to undergo robust 206 expansion after editing (Fig. 4d), and secreted J3 HCAbs (Fig. 4e). The engineered antibodies 207 also had the expected ability to neutralize HIV-1 strain NL4-3 (Extended Data Fig. 6c)

208
By varying the AAV6-J3 MOIs, we observed that although lower MOIs reduced the initial 209 editing rates achieved, they also had less of an impact on cell proliferation, resulting in similar 210 total numbers of edited cells by day 8 (Extended Data Fig 6d-h). The inhibition of proliferation at 211 higher doses could be due to the reported p53-mediated sensing of AAV ITRs. 36 Editing was also 212 possible with all of the BAC and DP activation protocols we evaluated, although none 213 outperformed BAC in serum-free XF media (Supplementary Fig. 6).

214
We next investigated the differentiation capability of edited B cells under different culture 215 conditions. The editing process did not significantly impact the phenotype of BAC cultured cells, 216 which remained largely IgM single-positive, double-negative, and memory B cells (Fig. 4f,

217
Extended Data Fig. 6i). After editing, the cells could also still be induced towards ASCs

220
When B cells differentiate towards ASCs, there is a switch from BCR to secreted antibody 221 isoforms. 23 After differentiation of unedited B cells we observed around a 50% reduction in the 222 frequency of IgG-BCR + cells and 75% reduction in the amount of IgG-BCR on the cell surface 223 ( Fig. 4g-i). Expression of J3-BCR on edited B cells paralleled these changes, with similar 224 reductions in both the frequency and magnitude of J3-BCR expression after differentiation ( Fig.   225 4g-i). The switch from J3-BCR to antibody isoforms was also confirmed at the mRNA level 226 (Extended Data Fig. 6j). We also observed a similar concordance for the secretion of antibodies, 227 since the rate of IgG secretion per cell in unedited samples correlated with the rate of J3 secretion 228 per cell from edited samples, across different matched time points and differentiation conditions 229 (Fig. 4j). As expected, both total IgG and J3 HCAb secretion rates per cell were generally highest 230 in the differentiated samples.

231
In sum, these data show that primary human B cells activated with BAC can be efficiently 232 edited at IGHG1. HCAb editing has minimal impact on B cell differentiation, with the cells adopting 233 a memory or precursor phenotype. Finally, the edited cells can undergo differentiation, which enhances antibody secretion and reduces BCR expression. These changes were similar for 235 endogenous IgG and J3 HCAb, implying proper regulation of this important process for the HCAb 236 construct.

239
We used a tonsil organoid system 37 to examine the ability of HCAb-engineered human B 240 cells to respond to stimulation by a matched antigen (Fig. 5a). We first confirmed the ability of 241 non-engineered tonsil B cells to respond to immunization in this model by treating the organoid 242 cultures with HIV gp120 or PE antigens, and further identified an effective adjuvant for gp120 243 ( Supplementary Fig. 7). Next, B cells were purified from total tonsil mononuclear cells (TMNCs) 244 and edited as before using sg05 RNPs and AAV6-J3 vectors (Fig. 5b). Four days later, the edited 245 B cell population was reconstituted with total TMNCs (including an excess of unmanipulated B 246 cells) and paired samples were cultured without immunization, or immunized with gp120 or control 247 PE protein plus adjuvant. Immunization with gp120 resulted in a significant expansion of J3-edited 248 B cells compared to the unimmunized cultures (Fig. 5c,d). No expansion was observed following 249 immunization with PE, confirming that this was an antigen-specific response.  by flow cytometry and ELISA, respectively, and editing was confirmed by ddPCR ( Fig. 6a-     pairing is still possible with endogenous unedited L and H chains in the same cell that could create 305 novel paratopes, and although endogenous Igk can also be targeted for disruption, 11,14 the 306 structure of Igl and its pseudogenes makes it much more challenging target to disrupt. In contrast, 307 the HCAb editing approach described here offers a simpler path to B cell engineering, requiring 308 only one DNA edit to insert a single polypeptide chain that can reprogram IgH for expression of a 309 custom H chain antibody.
To validate the HCAb editing approach, we selected ARDs based on the J3 and A6 anti-

311
HIV camelid VHH domains and inserted them into the intron downstream of the CH1 exon of 312 IGHG1. This location ensures that the resulting H chain polypeptides exclude CH1 and will not 313 interact with endogenous L chains. The engineered HCAbs retained the anti-HIV properties of the 314 parental ARD, mirrored the physiology of a native IgG gene by regulating expression of both the 315 membrane and secreted isoforms during ex vivo differentiation, and responded to antigen in a 316 tonsil organoid model. Of interest, we also observed evidence that insertion of the J3 VHH at this 317 locus supported SHM after extended culture in a B cell line, although the A6 VHH and a GFP 318 control sequence did not. We note that our design of the J3 DNA sequence was codon wobbled 319 from the published amino acid sequence 50 using the closest human germline VH (IGHV3-23) 320 sequence as a template. In contrast, the A6 DNA sequence was unmodified from the natural 321 antibody DNA sequence that had already undergone SHM evolution in its camelid host, which 322 could explain the differing susceptibilities to hypermutation. The presence or absence of SHM 323 could be uniquely advantageous in different settings. For instance, a variable pathogen such as

324
HIV could be better controlled by HCAbs that retain the capacity to undergo SHM and subsequent 325 affinity maturation, whereas a static target like a self-antigen may be better matched with an HCAb 326 that does not hypermutate. Understanding the mechanisms that govern SHM at these insertion 327 sites will require further research but may yield rules that allow tuning of sequences to either avoid 328 or exploit this natural process of antibody evolution.

329
We further demonstrated that the simplicity of the HCAb design allows unprecedented

354
An advantage of editing within the constant region of a defined antibody subclass is that the 355 HCAbs will not diversify through class switch recombination, but only express the selected 356 engineered isotype. This could avoid the need for ex vivo or in vivo differentiation/immunization 357 to drive isotype selection, while also generating a more consistent cell therapy product that will 358 be stable over time and avoid potentially adverse effects of ongoing class switching. 55 Conversely,

505
The VH (3GIZ_H) and VL (3GIZ_L) sequences of the fully human anti-CD20 antibody

506
Ofatumumab were obtained from NCBI. These VH and VL sequences, human IGKC, and the 507 CH1 exon of human IGHG1 were codon optimized using the GeneOptimizer tool (Thermo Fisher) 508 and synthesized as gBlocks (IDT). IGHV3-9 and IGKV3-11 leaders were placed upstream of VH

627
After overnight incubation of samples and washing, 50 µL of diluted enzyme was added to each 628 well, or the digestion buffer alone to control wells. Digestion was carried out for 16 hours at 37°C.

629
Plates were washed 8-10 times with 350 µL wash buffer per well, before continuing to detection

671
In all culture conditions, cells were edited after 3 days of pre-activation, as previously 672 described. 33 Briefly, cells were washed with PBS and resuspended at 5 x 10 7 cells/mL in serum-673 free XF media.

748
Fwd2 and IO-EEK-In-6 (Supplementary Table 5). PCR products were visualized on a 1% agarose 749 gel with GelRed Nucleic Acid Stain (Biotium). PCR products were purified using Nucleospin Gel                      Homology directed repair of the targeted DNA break inserts a cassette comprising a B cell-specific promoter, a custom ARD, and a splice donor to direct splicing to downstream endogenous exons. (b) K562 cells were edited with Cas9 RNPs programmed by gRNAs targeting the IgG1 CH1 intron and indels measured at on-target (IGHG1) and off target (IGHG2-P) genes by Sanger sequencing and ICE (n = 3). (c) K562 cells were edited by Cas9 RNPs plus AAV6 homology donors containing a GFP expression cassette, matched for each gRNA. GFP expression was measured after 3 weeks, to dilute out episomal AAV genomes (n = 3). (d) On-and off-target activity of sg05 was measured at indicated IGHG genes in primary human B cells (n = 2), 5 days after editing with sg05 RNPs, by targeted amplicon deep sequencing, with percentage mutated reads calculated as insertions, deletions, ≥ 2 bp changed. See also Extended Data Fig. 1d,e. Error bars show mean ± SEM. Statistical comparisons (c-d) were performed by 2-way ANOVA. * p < 0.05, ** p < 0.01, *** p < 0.001, **** p < 0.0001.      Week 6 Week 15 Week 24 Input plasmid Week 6 Week 15 Week   (c-e) Tonsil B cells from n = 3-5 donors were edited with RNPs plus AAV6-J3 at MOI = 1-5 x 10 5 vg/cell, cultured for 2-4 days, and reconstituted with total autologous TMNCs. Cultures were immunized with HIV gp120 plus Adju-Phos or control PE protein plus Alhydrogel and cells were harvested 12 days later for analyses. (c) Representative panel showing increase in gp120 + VHH + cells after immunization with gp120 but not PE, measured by flow cytometry, gated on CD19 + CD3 -B cells as described in Supplementary Fig. 8. (d) Response of tonsil organoid cultures containing control or edited B cells to gp120 or PE immunization. Matching colors indicate samples from the same tonsil donor. (e) Phenotypes of B cells in tonsil organoids containing control or J3-edited B cells were characterized at day 12 by flow cytometry, with or without gp120 immunization, using the gating strategy described in Supplementary Fig. 8. Error bars show mean ± SEM. Statistics in panel (d-e) were calculated by 1-way ANOVA. * p < 0.05, ** p < 0.01.  (a-f) Primary human B cells from n = 3 independent experiments were activated with BAC in XF media, starting at day -3, and edited at day 0 with sg05 Cas9 RNPs and AAV6-CD4 donors (a-c, MOI = 5 x 10 5 vg/cell) or AAV6-PGT121-scFv donors (d-f, MOI = 5 x 10 5 vg/cell). (a,d) Editing rates were measured at day 8 by flow cytometry for surface J3-BCR. (b,e) Editing was quantified by in-out ddPCR at day 8 and normalized per cell against a control reaction. (c,f) HCAb secretion was measured by gp120-IgG ELISA at day 8. (g) Design of CH2 editing approach. Cas9 gRNA CH2-g1 targets the intron downstream of CH2. Homology donor cassette contains B cell specific promoter, antigen recognition domain (ARD), codon-wobbled IgG1 Hinge (Hi) and CH2 exons, and splice donor to link to endogenous CH3 and membrane exons after insertion. (h-i) Primary human B cells from n = 3-4 independent experiments were activated with BAC in XF media, starting at day -3, and edited at day 0 with CH2-g1 Cas9 RNPs and AAV6-J3-CH2* donor (MOI = 10 4 vg/cell).
(h) Editing rates were measured at day 8 by flow cytometry for surface J3-BCR. (i) J3 HCAb secretion was measured by gp120-IgG ELISA at day 8. Error bars show mean ± SEM. Statistics were calculated by 2-tailed t-test. * p < 0.05, ** p < 0.01, *** p < 0.001.  K562 cells were electroporated with Cas9 RNPs for indicated gRNAs and a matched plasmid homology donor containing a GFP expression cassette (n = 3). HDR editing was measured by flow cytometry for GFP expression after 3 weeks. (c) Site-specific insertion of GFP expression cassettes in AAV6-edited K562 cells was confirmed by in-out PCR for each tested gRNA. Uncropped gel is available in Supplementary Fig. 3a. (d) On-and off-target activity of sg05 was measured at indicated IGHG genes in primary human B cells, 5 days after editing, by targeted amplicon deep sequencing. Aggregate mutations at each base in a 50bp window surrounding the sg05 cut site (0; orange dotted line) are shown for each gene. (e) Percentage mutated reads at each IGHG gene calculated as for all changes (≥ 1 bp changed), which gives a higher background than when a cutoff of ≥ 2 bp is selected, as shown in Fig. 1d (n = 2). Error bars show mean ± SEM. Statistics were calculated by 2-way ANOVA. * p < 0.05, *** p < 0.001, **** p < 0.0001, ns = not significant.   293T cells were transfected with expression plasmids for the J3 VHH HCAb, a full-length (FL) conventional human IgG1 derived from Ofatumumab, or its Igκ L chain only (LC), or combinations as indicated. Supernatants, including mixed supernatants as controls, were evaluated by ELISAs based on HIV gp120 binding by the J3 VHH and detection with an anti-VHH antibody (control), or anti-Igκ L chain antibodies to detect pairings between the J3 HCAb and either the FL or LC components. FabALACTICA digestion is expected to cleave FL antibodies into Fab and Fc fragments and may also cleave HCAbs. (b-c) Results of gp120-VHH (b) and gp120-Igκ (c) ELISAs, with or without FabALACTICA digestion (n = 2-4). Cross-pairing was only observed after co-transfection of J3 HCAb and FL antibody, consistent with H chain interactions that were released by FabALACTICA digestion. In contrast, the L chain alone did not pair with the J3 HCAb. Error bars show mean ± SEM. Statistics were performed using 2-way ANOVA. **** p < 0.0001.   Absorbance curve for J3 avidity over time from J3-edited Raji cells in extended culture (n = 3). IgG produced by the cells at indicated time points was measured by gp120-IgG ELISA. (b) Changes at A6 or GFP sequence in edited Raji cells over time, measured by deep sequencing. Percentage mutation at each position is the frequency of reads that did not match the wild-type sequence. CDR regions in A6 are indicated in grey. (c) Total mutations in A6 or GFP sequences at each timepoint were summed and divided by the total sequence length to determine a total % mutations. Shown in green are mutations associated with AID hotspot motif cytosines (WRCH). (d) The density of AID hotspots (number of WRCH hotspots / number of base pairs in the sequence) for each indicated sequence. (e) The distribution of AID hotspots across framework regions (FR) and complementarity-determining regions (CDR) for J3 and A6. Error bars show mean ± SEM. Week 6 Week 15 Week 24 Week 6 Week 15 Week 24   Supplementary Fig. 5. For BAC+DP, the cells were started in BAC and treated for a total of 5 days, then switched to the DP protocol for the remaining 6 days, as shown in Supplementary Fig. 4. (g) Comparison of total IgG secretion and IgG/cell at day 8 for cells treated with indicated stimulation protocols. DP was in IMDM, BAC and BAC + DP were in XF without FBS. Error bars show mean ± SEM. Statistics in panels (c,g) were calculated by one-way ANOVA. * p < 0.05, ns = not significant.  (h) J3 HCAb secretion, measured by gp120-IgG ELISA at day 8. (i) B cell phenotypes of edited cells in BAC (left) or with differentiation (BAC + DP, right) were measured by flow cytometry. Unedited matched control cells were also quantified and subtracted from the frequencies in the edited cells, to highlight any changes in differentiation after editing. (j) Edited primary B cells were differentiated in DP plus IMDM media and RNA extracted at indicated time points. RT-PCR was performed using a J3-specific forward primer and IgG reverse primers specific for the secreted (S) or membrane (M) isoforms. Uncropped gel image is provided in Supplementary Fig. 3d. Error bars show mean ± SEM. Statistics were calculated by 2-way ANVOA (a), 1-way ANOVA (d-h), or 1-sample t-test (i). * p < 0.05, ** p < 0.01, *** p < 0.001, **** p < 0.0001.