A precision gene-engineered B cell medicine producing sustained levels of active factor IX for hemophilia B therapy

Abstract

Hemophilia B gene therapy treatments have not addressed the need for predictable, durable, active, and redosable factor IX (FIX). Unlike conventional gene therapy, engineered B cell medicines (BCMs) are durable, redosable, and titratable and thus have the potential to address significant unmet needs in the hemophilia B treatment paradigm. BE-101 is an autologous BCM comprising expanded and differentiated B lymphocyte lineage cells genetically engineered ex vivo to secrete factor IX (FIX)-Padua. CRISPR-Cas9-mediated gene editing at the C-C chemokine receptor type 5 (CCR5) locus was used to facilitate transgene insertion of an adeno-associated virus 6-encoded DNA template via homology-directed repair. Transgene insertion did not alter B cell biology, viability, or differentiation into plasma cells. Appreciable levels of BE-101-derived FIX-Padua were detected within 1 day after IV administration in mice, and steady state was reached within 2 weeks and persisted for over 184 days. Redosing produced an increase in FIX-Padua production close to linear dose proportionality. Comprehensive genotoxicity analysis found no off-target issues of concern. No safety signals were observed in animal tolerability and Good Laboratory Practice toxicology studies. In conclusion, BE-101 produces sustained levels of active FIX-Padua with the ability to engraft without host preconditioning and with the potential for redosing and titratability.

Graphical abstract

B cell medicines (BCMs) engineered to produce factor IX (FIX) are a novel, potentially long-lasting FIX replacement therapy for people with hemophilia B. Pre-clinical studies show rapid engraftment and production of active FIX for >180 days after a single IV injection of FIX-engineered BCM.

Introduction

Mutations in the F9 gene coding for factor IX (FIX) lead to FIX deficiency and severe bleeding events in people with hemophilia B (PwHB). Recommended therapy is prophylactic infusion of exogenous FIX^1,2; however, the short biological half-life requires frequent administration once weekly to biweekly to maintain therapeutic levels. Even extended half-life FIX prophylaxis may require individual pharmacokinetic monitoring for optimal outcomes.³ Two recently approved hemostatic rebalancing agents, marstacimab and fitusiran, do not replace the missing FIX; rather, they reduce tissue factor pathway inhibitor and antithrombin, respectively, and still require weekly or bimonthly subcutaneous administration.^4,5 A considerable barrier to treatment adherence for PwHB that leads to breakthrough bleeding events is the lifelong need for frequent infusions at intervals in which trough levels are low.^6,7 Real-world outcome studies demonstrate that both standard and extended half-life FIX prophylaxes are associated with breakthrough bleeding in PwHB (87/108 [81%] and 23/34 [68%] individuals, respectively, reported bleeding events in a 12-month period).⁸

Adeno-associated virus (AAV) vector-based gene therapy products (etranacogene dezaparvovec and fidanacogene elaparvovec-dzkt) are approved for adults with HB and show superiority versus FIX prophylaxis with respect to annualized bleed rates.^9,10 An alternative approach to enable liver-mediated FIX production is the experimental gene insertion therapy REGV131-LNP1265, in which a FIX transgene is targeted for stable insertion at the albumin (ALB) locus in the patient’s liver cells via stepwise in vivo delivery of, first, a lipid nanoparticle (LNP) carrying CRISPR-editing reagents, followed by an AAV containing the F9 transgene. While promising, these therapies carry potential risks of systematic and liver toxicity, as well as genotoxicity due to random AAV chromosomal insertions. Furthermore, AAV immunogenicity precludes redosing and may decrease therapeutic efficacy.^11,12 Significantly, AAV-based in vivo gene therapy is not approved for children due to the potential waning of sufficient FIX activity with hepatic turnover and liver growth, as well as unknown safety risks. Thus, significant unmet medical needs remain for new therapies providing predictable, durable, and redosable FIX, especially in children where early intervention is critical to prevent joint damage.¹

An autologous source of stable and durable FIX is a promising therapeutic goal to address unmet needs for PwHB. Antibody-producing long-lived plasma cells (PCs) are an attractive cell type for the sustained production of therapeutic proteins. Derived from antigen-driven differentiation of B cells, PCs are inherently long-lived (half-lives ranging from 6 months to decades)^13,14 and actively home to and reside in the bone marrow compartment, where they secrete high levels of antibodies (>1,000 antibodies/cell/s).¹⁵ Terminally differentiated human PCs derived from genetically engineered B cells (B cell medicines [BCMs]) produce sustained and high levels of transgenic proteins in vivo in immunodeficient mice without pre-conditioning.¹⁶ In non-human primates (NHPs), positron emission tomography/computed tomography tracking of zirconium-89-oxine-labeled, ex vivo-differentiated PCs show that the cells rapidly home to and engraft in the bone marrow, spleen, and liver after infusion into unconditioned autologous subjects.¹⁷ Moreover, in a redosing experiment using this approach, the biodistribution was identical to what was observed following the initial dose, demonstrating the feasibility of redosing ex vivo differentiated PCs.¹⁷ Ex vivo CRISPR-Cas9 genome editing mitigates concerns of immune responses or insertional mutagenesis in other cell types while enabling BCMs to secrete therapeutic proteins at high levels with long-term in vivo engraftment in mouse models. In contrast to gene therapy, BCMs have the potential for redosing, making them an attractive platform for the sustained supply of biologics where continuous and adaptive dosing is required to achieve therapeutic benefit, for example, in children with HB.

In this paper, we describe pre-clinical studies characterizing an autologous ex vivo precision gene-edited cell therapy (termed BE-101) comprising expanded and differentiated B lymphocyte lineage cells that have been genetically engineered to express and secrete FIX-Padua.

Results

F9 transgene integration does not alter BCM biology, viability, or differentiation

BE-101 cells were engineered using a CRISPR-Cas9 single-guide RNA (sgRNA) ribonucleoprotein (RNP) complex targeting the CCR5 locus and an AAV6 vector carrying a homology-directed repair (HDR) template containing a transgene expression cassette for FIX-Padua. BCM phenotype at the end of the culture process (Figure 1A) was compared for BE-101 cells and BCMs electroporated with RNP only without subsequent AAV transduction. Insertion or deletion (indel) frequencies at the CCR5 locus were measured relative to unedited control cells and were comparable in BE-101 cells and RNP-only controls as expected (∼80%; Figure 1B), while F9 transgene integration was observed only in BE-101 cells (∼15%–20%; Figure 1C). FIX-Padua production was evident on day 6 and remained through day 13 (Figure 1D). Neither viability nor phenotype differed between BE-101 cells and RNP-only control cells (Figures 1E–1H).

Figure 1.

Impact of engineering on BCM viability, phenotype, and FIX-Padua production

(A) Diagrammatic representation of B cell culture and differentiation process in vitro. (B) Percentage of cutting frequency at the CCR5 locus, measured by inference of CRISPR edits, of RNP-only or BE-101 engineered BCM. (C) Percentage of transgene-containing alleles at days 6 and 13, determined by ddPCR, comparing BE-101 to the RNP-only control. (D) FIX-Padua production (ng/10⁶ cells/24 h), determined by ELISA, at days 6 and 13. (E) Percentage of cell viability during the B cell differentiation process relative to RNP-only control. (F) Representative flow cytometry data from day 13 BCMs, illustrating the gating strategy. (G and H) Percentage of plasmablasts (CD27⁺CD38⁺CD138^low) and plasma cells (CD27⁺CD38⁺CD138^high) among the total live cells. The data are a repetitive experiment for n = 3 donors. Error bars represent mean ± standard deviation; ∗p < 0.01; ∗∗p < 0.001; ∗∗∗p < 0.0002; ∗∗∗∗p < 0.0001 by 1-way ANOVA; Holm-Sidak correction for multiple comparisons.

Production of functional FIX-Padua protein by BE-101 in vitro

BE-101 products generated from 7 B cell donors secreted an average of 634 ng/10⁶ cells/day of FIX-Padua protein, approaching 45% of concomitant PC immunoglobulin G (IgG) secretion rate at a molar level (Table 1). The measured FIX-Padua production rate in vitro was estimated at ∼540 molecules/cell/s.

Table 1.

In vitro secretion rate of FIX-Padua protein in timed culture experiment in comparison with hIgG from BE-101

Donor no.	hIgG secretion rate (ng/106 cells/24 h)	FIX-Padua secretion rate (ng/106 cells/24 h)	Secretion rate of FIX-Padua versus hIgG (%)a
D1	2,690	515	50
D2	2,266	293	34
D3	5,220	414	21
D4	2,954	802	71
D5	3,798	696	48
D6	5,124	855	44
D7	5,185	867	44
Mean (SD)	3,891 (1,287)	634 (229)	45 (15)

hIgG, human immunoglobulin G.

^aCalculated as FIX-Padua secretion rate divided by hIgG secretion rate and normalized by molecular weight difference of human IgG (i.e., 150 kDa) and human FIX-Padua protein (i.e., 57 kDa).

BE-101-derived FIX-Padua protein function in vitro was determined in ex vivo cell cultures in the presence of vitamin K.¹⁸ Peripheral blood B cells isolated from 3 donors were engineered and cultured with menadione sodium bisulfite (synthetic vitamin K3). The mean amount of FIX-Padua protein produced was 407.3 ± 111.4 ng/mL (Figure 2A). Vitamin K-dependent active FIX-Padua protein in the range of 1.82–11.6 mIU/mL corresponded to an activity/mg of protein of 15.8–25.4 IU/mg (Figure 2B). In the activated partial thromboplastin time (aPTT) assay, purified protein from cells cultured in the presence of vitamin K altered clotting time with percentages of FIX-Padua activity at 28.9%, 15.4%, and 26.4% for donors A, B, and C, respectively, using 5 μg BE-101 purified protein (Figures 2C and 2D). Activity levels observed in FIX-Padua-deficient plasma in the aPTT assay corresponded to a specific activity of 26.7–65.2 IU/mg protein. Minimal activity in the absence of vitamin K was observed. Finally, liquid chromatography-tandem mass spectrometry (LC-MS/MS) analysis of trypsin-digested FIX protein derived from BE-101 cells cultured in the presence of vitamin K for 48 h showed that up to 11 out of 12 possible glutamic acid residues in the GLA domain underwent γ-carboxylation (Figure 2E).

Figure 2.

In vitro FIX-Padua protein secretion and activity from BE-101

(A) FIX-Padua protein levels in supernatant production from engineered B cells on day 13 of culture as measured by ELISA. The bar graph depicts mean and standard deviation of 3 donors. (B) Active FIX-Padua protein production by BE-101 culture in the presence of vitamin K, added for 48 h from days 11–13. Activity was measured from culture supernatant using a human FIX selective antibody capture step coupled with activity determination using the Rossix FIX activity assay. The bar graph represents the calculated mean specific activity and standard deviation of 3 donors. (C) FIX-Padua activity was measured using activated partial thromboplastin time (aPTT) using PTT-automate activation and measurement of clotting times on the Stago Satellite. FIX-Padua activity curve was made from dilutions of normal pooled plasma into hemophilia B plasma. FIX-Padua protein purified from engineered B cell cultures at 5 μg/mL was measured for donors A–C (+Vit K, in the presence of vitamin K) and 1 control donor (−Vit K, in the absence of vitamin K). (D) aPTT times from (C), donors A–C. (E) LC-MS/MS peptide coverage of FIX-Padua GLA domain after trypsin digestion with corresponding glutamate carboxylation status.

Engineering efficiency, FIX-Padua protein production, and cell phenotype of BE-101 produced from the scaled-up process

After demonstrating F9 transgene integration does not alter BCM biology, viability, or differentiation and production of functional FIX-Padua protein by BE-101 in vitro, we optimized gene editing and ex vivo culture methods toward a large-scale process for potential clinical application. Parameters varied included the electroporation and AAV transduction conditions.

Optimization resulted in an average of 88% cutting frequency at the CCR5 locus in activated primary human B cells from 5 donors (Figure 3A). An average of 51% F9 gene integration was measured by droplet digital PCR (ddPCR) after optimizations were made (Figure 3B), which represented a ∼2.5- to 3-fold boost in engineering efficiency (Figure 1C), resulting in FIX-Padua production of 1,343.4 ng/mL (Figure 3C). These optimized procedures reproducibly yielded 95% CD38⁺ BCMs from 5 independent donors comprising approximately 20% CD27⁺/CD38⁺/CD138^high PCs and approximately 60% CD27⁺/CD38⁺/CD138^low plasmablasts measured from 4 out of the 5 independent donors (Figures 3D and 3E; Tables S1 and S2).

Figure 3.

Engineering efficiency, FIX-Padua production, and cell phenotype at the scale-up process

(A) Percentage cutting frequency at the CCR5 locus. (B) Percentage targeted integration (HDR) of the F9 transgene at the CCR5 locus. (C) Engineered B cell FIX-Padua production during the final stage of cell culture, as measured in the harvest supernatant using an FIX ELISA. (D) Representative flow cytometry plot illustrating the differentiation of B cells at various time points during the B cell differentiation process. (E) Percentage of antibody secreting cells (CD38⁺) among live cells during and at the end of the culture process (day 13). Error bars represent the standard deviation (±) for 5 donors and 5 individual cell expansion experiments.

Analysis of on- and off-target editing and overall genome integrity in BE-101 cell products

Following optimization of the gene editing reaction for transgene insertion efficiency at the CCR5 on-target site, we characterized the genome integrity in BE-101 cells. We used both computational and experimental methods to comprehensively nominate candidate off-target sites for the gCCR5_232 gRNA. The CRISPRitz algorithm,¹⁹ in association with a cutting frequency determination (CFD) score threshold of ≥0.2,²⁰ identified 422 NGG-protospacer adjacent motif (NGG-PAM) containing human genomic sites of homology to the gCCR5_232 on-target sequence (with no more than 4 nucleotide mismatches and 2 bulges). Additionally, we performed cellular off-target nomination using G-GUIDE (a modified version of GUIDE-seq [genome-wide unbiased identification of double-stranded breaks enabled by sequencing]²¹) and biochemical off-target discovery by SITE-seq (selective enrichment and identification of adapter-tagged DNA ends by sequencing,²² to identify potential off-target cleavage sites of gCCR5_232 in primary human B cells. G-GUIDE nominated 9 candidate off-target sites that were found reproducibly across technical duplicates in at least 1 out of 5 tested B cell donors, and as expected in primary human B cells, double-strand breaks were detected in the switch regions of the IgH locus as well. SITE-seq nominated 120 candidate off-target sites that were found in at least 2 out of 6 B cell donors tested (with no more than 7 nucleotide mismatches and 5 bulges relative to the gCCR5_232 on-target site). Interestingly, the list of 422 sites nominated in silico and the sites nominated experimentally had only 22 sites in common, with 3 sites identified by all 3 methods, one of which was the intended on-target site of gCCR5_232 (Figure 4A). Off-target sites identified by the 3 discovery efforts are listed in Table S3.

Figure 4.

Analysis of on- and off-target editing and overall genome integrity in BE-101 cell products

(A) Venn diagram showing the number of potential gCCR5_232 off-target sites discovered by CRISPRitz, G-GUIDE, and SITE-seq. GMP-grade gCCR5_232 was used for the nomination of sites via G-GUIDE and SITE-seq. (B) Mean DNA editing rates ± standard error of the mean (SEM) at 457 nominated sites across the genome as measured via targeted amplicon sequencing in 5 lots of BE-101 cells (treated) and unengineered, donor-matched controls (control) plotted on a scale between 0 and 100% (top graph) or 0%–1.5% (bottom graph). The x axis represents chromosomal location. Sites with editing rates >1% in control samples are marked in gray (background). (C) Circus plots depicting structural variants (SVs) as identified by Bionano rare variant analysis (RVA) in 3 lots of BE-101 (top) with magnification of the on-target insertion site on chromosome 3 (below). The outermost numerical track corresponds to the chromosome number. The cytoband information is shown in the black-and-white banding pattern, with red lines indicating the centromere. The next circle indicates SVs that were uniquely identified in engineered cells but not donor-matched unengineered controls. Insertion events are represented by green dots. The next circle shows copy-number variations. The inner space indicates translocation events found uniquely in engineered cells.

To determine bona fide off-target sites in BE-101 cells, DNA editing rates at the nominated loci were compared in 5 batches of BE-101 cells and donor-matched unengineered controls. Of the 526 candidate off-target sites, 457 were amenable to targeted DNA sequencing and reached the required minimal sequencing depth of 5,000 reads per site to allow for detection of differential editing frequencies as low as 0.2%. We observed high on-target editing at CCR5 (80.58% indels, p = 0), and off-target activity was validated at only 1 candidate off-target site (0.97% indels, p = 2.18E−46) (Figure 4B), located in an intergenic region on chr5.

To assess genome integrity in BE-101 cells, we performed genome-wide unbiased optical genome mapping (OGM) in 3 lots of BE-101 cells and donor-matched unengineered controls. De novo assembly of data obtained by OGM identified insertion events of the size expected for single-copy F9 transgene integrations at the CCR5 on-target locus, with frequencies of ∼45%, in line with ddPCR results in these cells (Figure S1; Table S4) confirming HDR-mediated insertion of the full length F9 expression cassette. Rare variant analysis (RVA) of the same data identified only 1 type of structural variant (SV) that was uniquely found in BE-101 cells but not donor-matched unengineered controls: large insertions at the on-target locus (detected with low molecule read counts, 5–25) in all 3 lots of BE-101 (Figure 4C; Table S5). The sizes of these large insertions increased in a stepwise fashion consistent with potential concatemeric AAV insertions (Table S5). No other reproducible SVs were identified that reached the required minimum of 5 molecules to support a SV identification.

Rapid homing of BE-101 to bone marrow with stable engraftment and FIX-Padua activity in the plasma of NOG-hIL6 mice

To study the biology of BE-101 in a small animal model we used NOG-hIL6 mice. Interleukin-6 (IL-6) is a pleiotropic cytokine primarily produced by antigen-presenting cells and is also known to drive antibody-secreting cell (ASC) differentiation and survival in multiple models.²³ Murine IL-6 does not efficiently activate the human IL-6 receptor (hIL-6R),²⁴ thus, hIL-6 in vivo is required for the efficient engraftment of human BCMs.²³ Biodistribution of BE-101 was assessed using a quantitative polymerase chain reaction (qPCR) assay designed to detect and quantify the human-specific Alu sequence as a surrogate for human genomic DNA in mouse organs, bone marrow cells, and whole-blood samples following BE-101 dosing in NOG-hIL6 mice. After a single intravenous (IV) administration of BE-101 at 20 × 10⁶ viable cells/mouse, human DNA was detected in femurs and spinal columns of NOG-hIL6 mice as early as day 1 and remained at comparable levels on day 28 (Figures 5A and 5B), consistent with the characteristics of PCs homing to the bone marrow niche. An equivalent number of human cells persisted in femurs and spinal columns on day 28 and was estimated to be less than 1% of cells dosed (calculated using a value of 6.41 pg DNA per human cell from a male donor²⁵). Human DNA levels decreased significantly or were found to be below the limit of quantitation by day 28 in all non-bone marrow compartment tissues examined, including whole blood, spleen, kidney, lung, liver, heart, brain, and testis (Figure 5C). Similar homing and engraftment were observed by in vivo bioluminescent imaging of surrogate luciferase-engineered BCMs (Figures 5D–5F). Among all non-homing tissues evaluated, the apparent clearance of human DNA in kidney was not as rapid as other tissues. However, the Alu qPCR assay cannot discriminate fragments of human DNA from DNA isolated from intact human cells. It is known that the main organ for the clearance of oligonucleotides²⁶ is the kidney, and thus it is possible that the Alu-specific signal detected in kidney at later time points originated from fragments of human DNA.

Figure 5.

Bone marrow trafficking and stable engraftment of BE-101 in NOG-hIL6 mice

Biodistribution of BE-101 in NOG-hIL6 mice was measured by a quantitative polymerase chain reaction (qPCR) method assay designed to detect and quantify the human-specific Alu sequence as a surrogate for human genomic DNA in mouse organs, bone marrow cells, and whole-blood samples. (A) Mean human DNA mass detected in femurs and spinal columns of NOG-hIL6 mouse mice on days 1 and 28 post-IV administration of BE-101 at 20 × 10⁶ viable cells/mice. Error bars represent ±SEM for n = 3 mice on days 1 and 28. Mann-Whitney U test was used to assess statistical significance; ns, not significant. (B and C) Shown are the mean percentages of human genomic DNA detected in (B) bone marrow-containing tissues and (C) whole-blood and various NOG-hIL6 mouse organs on days 7 and 28 relative to the day 1 post-IV administration of BE-101 at 20 × 10⁶ viable cells/mice. Error bars represent ±coefficient of variation for n = 3 mice at each time point. (D) Schematic of the firefly luciferase transgene for targeted integration at the CCR5 locus. (E) Quantification of bioluminescence signal over time. Each data point represents the mean and the error bars represent ±SEM for n = 3 mice in each group. (F) Bioluminescence imaging of NOG-hIL6 on days 8 and 44 after a single IV administration of 10 × 10⁶ viable unengineered or luciferase-engineered BCMs on day 1.

The durability of FIX-Padua production in vivo was assessed following a single dose of BE-101 (20 × 10⁶ viable cells/mouse) by monitoring FIX-Padua protein levels in mouse plasma for up to 184 days (Figure 6A). Following an initial decrease in levels immediately post-infusion, plasma FIX-Padua levels stabilized within 2 weeks and were consistently detected throughout the study (Figure 6B). The FIX-Padua protein was below the limit of detection in the mouse plasma collected from the vehicle control treatment group. Similarly, sustained plasma levels of human IgG (Figure 6B) and IgM (Figure S2) were detected. Human-specific FIX-Padua activity was measured in 3 mouse plasma samples using an immunocapture chromogenic activity assay from study no. 2 (Figure 6B), with specific activity ranging from 136 to 274 IU/mg at selected time points from days 28 to 126.

Figure 6.

In vivo secretion of engineered B cell-derived FIX-Padua and human IgG after a single administration and repeated administrations of BE-101 in NOG-IL6 mice

(A) NOG-hIL-6 mice were dosed with 20 × 10⁶ total viable cells of BE-101 and endogenous B cell-derived human IgG and engineered B cell-derived FIX-Padua levels were measured in blood plasma up to 196 days of study. (B) Data are representative of 3 individual studies. The lower limit of quantitation (LLOQ) of FIX-Padua protein and human IgG assay is 3.6 and 3.4 ng/mL, respectively. (C) NOG-hIL6 mice (n = 4) were dosed with 20 × 10⁶ total viable cells of BE-101 on day 0 and re-dosed on day 49 of the study. Human IgG and FIX-Padua protein levels were measured in blood plasma after 1 day post-BE-101 administration and then weekly up to 77 days of study. (D) The LLOQ of human FIX-Padua protein and human IgG is 2.5 and 0.781 ng/mL, respectively. (B and D) Each data point represents the mean across all groups at that time point. All vehicle control group measurements were below the LLOQ. The error bars indicate the SEM.

Although direct assessment of the AAV vector copies remaining in cells at the end of the cell culture or in vivo was not conducted, FIX-Padua production from potential episomal AAV genomes was assessed. In research-scale studies, cells transduced in the absence of RNP (BE-101 no RNP) produced less than 4% of FIX-Padua protein compared to BE-101 control cells (Table S6). This was confirmed in a clinical-scale process development run where cells transduced in the absence of RNP generated less than 1% of the FIX-Padua protein as compared to the average FIX-Padua protein production for clinical manufacturing runs of BE-101 where the RNP was included. These data demonstrated that the contribution of non-integrated F9 transgene to total FIX-Padua protein production ex vivo was very low and suggested that non-integrated F9 transgene would not contribute significantly to FIX-Padua expression, nor would it affect the potency of BE-101 in vivo after clinical administration.

BE-101 redosing potential in the NOG-hIL6 model was tested after a second 20 × 10⁶ dose of identical cryopreserved BE-101 cells from the same donor was administered on day 49 (Figure 6C). The mean level of FIX-Padua protein on day 56 increased 2.4-fold compared to day 42 (59 versus 24 ng/mL), and levels were sustained for at least 4 weeks (Figure 6D). Similarly, mean levels of human IgG (Figure 6D) and IgM (Figure S3) on day 56 increased 2.2–3.0-fold compared to day 42 (312 versus 143 μg/mL for human IgG and 3.74 versus 1.26 μg/mL for human IgM) and were sustained for at least 4 weeks.

BE-101 safety

In a 28-day Good Laboratory Practice (GLP) toxicology study, 18 mice were treated with BE-101 manufactured from 2 donors (2.0 × 10⁷ total viable cells/mouse). In 3 long-term in vivo pharmacology and tolerability studies, 36 mice were treated with BE-101 manufactured from 8 donors (2.0 × 10⁷ total viable cells/mouse). No unscheduled deaths occurred, no abnormal clinical observations were observed, and at termination there were no adverse effects noted in clinical pathology samples or histopathology samples (Figures S4 and S5; Tables S7–S13).

Discussion

Autologous, cell-based therapeutics are a promising treatment approach for many diseases.^{27,28,29,30,31,32,33,34} Ex vivo gene editing strategies allow precise control when modifying cellular DNA and mitigate concerns of immune responses or insertional mutagenesis in other cell types arising from in vivo viral vector-delivered gene therapy.^35,36,37 BE-101 is a precision ex vivo engineered BCM that produces sustained levels of active FIX-Padua. Terminally differentiated human PCs derived from genetically engineered B cells offer natural longevity, capacity for high levels of protein secretion, the ability to engraft without host preconditioning, and the possibility to redose, making them an attractive platform for the sustained supply of biologics in vivo where continuous dosing is required to achieve therapeutic benefit.

BE-101 exhibited robust FIX-Padua protein production in vitro (∼540 molecules/cell/s), greater than previous reports.^16,38 BE-101 rapidly engrafted into mice with sustained circulation of FIX-Padua protein for up to 184 days. To our knowledge, this is the first report of ex vivo CRISPR-Cas9-edited, FIX-Padua-expressing PCs demonstrating stable long-term engraftment and secretion of active FIX-Padua in an animal model.

The expected BE-101 biodistribution to bone marrow tissue was confirmed in the NOG-hIL6 mouse model, and BCMs were stable in marrow-containing tissues but not other organs and tissues. Furthermore, IV administration of luciferase-expressing BCMs into NOG-hIL6 confirmed rapid homing and stability in bone marrow-harboring skeletal regions.

Although the NOG-hIL6 model is sufficient to demonstrate the durability and safety of BE-101, it does have limitations. Engraftment is estimated by the human Alu element biodistribution data to be less than 1% of input cells. Thus, quantifying the percentage of B cells and ASCs that contained the integrated transgene was not feasible. However, previous work on human PCs engineered to secrete bispecifics showed effective in vivo leukemia killing, suggesting that edited cells stably engrafted at reproducible percentages.³⁹ In humans, additional PC cytokines (e.g., BAFF, APRIL) and chemokines supplied by bone marrow-resident cells such as dendritic cells, megakaryocytes, and eosinophils, as well as the stromal cell extracellular matrix, may support higher levels of BCM engraftment. Support for this hypothesis comes from recent studies in NHPs, where we determined engraftment rates of rhesus macaque ASCs.¹⁷ Upon direct IV administration without preconditioning, ∼10% of the initial dose engrafted into bone marrow and ∼10% engrafted into spleen. Based on these NHP data, we projected that similar levels of BE-101 engraftment would occur in humans, which would be at least 20-fold higher than what was observed in the NOG-hIL6 mouse model.

Construct design and therapeutic transgene delivery are critical attributes in human gene therapy, and all aspects of BE-101 design were taken into consideration for clinical application. The FIX-Padua variant used in BE-101 was reported to be 4- to 8-fold more active than native FIX, a potential advantage in BCM dosing.^40,41,42 Additionally, the Malmö variant, T148A, used in the BE-101 construct, is the same variant used in the approved FIX therapies Benefix⁴³ and Hemogenix.⁴⁴

The CCR5 locus is minimally expressed in normal and mature B cells,^45,46,47 and disruption of this genomic region is not expected to affect the normal biological phenotype of B lymphocyte lineage cells. Furthermore, the CCR5 gene has been described as a safe harbor locus for stable transgene integration,⁴⁸ and clinical evaluation of CCR5-edited CD4⁺ T cells has supported the safety of this gene editing approach in humans.^49,50 The MND (myeloproliferative sarcoma virus enhancer, negative control region deleted, dl587rev primer-binding site substituted) promoter was selected to drive F9 transgene expression in BE-101 due to successful use in previous ex vivo gene therapy clinical trials^51,52 and its resistance to transcriptional silencing. While prior clinical studies utilizing the MND promoter have generally exhibited a favorable safety profile, the relative strength of this promoter raises the theoretical concern of endogenous gene activation if the site(s) of genome integration are not properly controlled, as is the case for cell and gene therapies that use lentiviral and gammaretroviral vectors, which promote almost random transgene insertion into the host cell genome.⁵³ This potential risk of insertional mutagenesis is likely influenced by multiple factors, including transduced cell type. A recent publication⁵⁴ described the risks of insertional mutagenesis following lentiviral gene therapy in CD34⁺ hematopoietic stem cells for cerebral adrenoleukodystrophy, which appears to be associated with the inclusion of the MND promoter. However, a systematic review⁵⁵ of T cell lymphomas in recipients of chimeric antigen receptor T cell (CAR-T) cells generated using lentiviral and gammaretroviral vectors concluded there is a low susceptibility of mature T cells to insertional mutagenesis and documented the almost complete lack of T cell transformation after natural HIV infection. Moreover, the US Food and Drug Administration estimated that a minimum of 27,000 doses of CAR-Ts, which used randomly integrating vectors containing the MND/MSCV promoters, have been infused in the United States and concluded that the risk for insertional mutagenesis was likely very low. While lentiviral and gammaretroviral transduction achieves stable genomic integration in a non-targeted fashion, BE-101 utilizes targeted CRISPR-Cas9-mediated transgene insertion at the CCR5 safe harbor locus via HDR. Moreover, using OGM we did not detect transgene insertion at the single bona fide off-target site identified by our genotoxicity assessments. While AAV genomes are generally considered to remain episomal, random AAV integration into the genome of transduced cells has previously been described. Unlike integrating viral vectors, random AAV vector integration frequencies are rare, with wild-type AAV integration into the known AAVS1 insertion site estimated to be about 0.1%.⁵⁶ In addition, recombinant AAV (rAAV) integration into the host genome is predicted to be greatly reduced because the sequences coding for the native AAV proteins needed for integration are not present in the rAAV.⁵⁷ Importantly, no abnormal effects attributed to the administration of BE-101 were observed in long-term in vivo studies up to 7 months in duration. These studies included 36 total immunodeficient mice treated with BE-101 manufactured from 8 different human donors. Based on these results, we view the potential for insertional mutagenesis related to random integrations to be low.

The approach of manufacturing BE-101 using targeted CRISPR-Cas9-mediated transgene insertion at the CCR5 locus is expected to decrease the potential for insertional mutagenesis related to random integrations observed with lentiviral or transposon-mediated transgene insertion.^55,58 A comprehensive genotoxicity assessment was performed and submitted as part of the BE-101 regulatory filings, with no issues of significance identified. Our rigorous characterization studies demonstrated that off-target effects of gCCR5_232 in BE-101 cells were limited to DNA editing rates of ∼1% at a single intergenic region over 40 kb from the nearest protein coding sequence, thus posing minimal genotoxic risk. OGM confirmed the presence of insertions conforming to HDR-mediated transgene integration at the CCR5 on-target locus in agreement with independent ddPCR measurements. Larger insertions at this site were also detected. Although sequence-level information is missing from OGM analysis, the stepwise increase in insertion size indicates potential concatemerization of viral DNA, a phenomenon observed previously in T cell therapy⁵⁹ and in nonclinical research.^60,61 No other SVs above the OGM detection level were identified, indicating that genome integrity was largely preserved. In murine safety/tolerability studies (both short- and long-term) there were no safety signals identified in laboratory, histopathology, or mortality assessments following the administration of transduced BCM cells.

The BE-101 pharmacokinetic profile is unique and distinct from what has been reported with in vivo FIX AAV gene therapy. Appreciable levels of FIX-Padua protein were detected within 1 day after BE-101 IV administration in mice, and steady state was reached within 2 weeks. In contrast, the reported gene therapy FIX concentration-time profiles show initial FIX activity observed at ∼3 weeks, with steady-state levels observed at 14 weeks after vector infusion,⁹ requiring PwHB to maintain FIX prophylactics until steady-state levels are reached. Future studies will assess whether the rapid onset of FIX-Padua production with BE-101 pharmacokinetics translates to the clinic with resulting benefits to PwHB.

Equally important to the rapid onset of FIX-Padua production with BE-101 is the ability to redose/re-engraft without host pre-conditioning in the NOG-hIL6 mouse model, achieving increases in FIX-Padua production in a manner close to linear dose proportionality and similar human IgG and human IgM kinetics between the first dose and repeated dose. To our knowledge, this is the first demonstration in an animal model that engineered B cells producing a human biological are redosable and titratable. These BE-101 attributes may have significant clinical advantages over AAV-based in vivo gene therapy approaches in HB, where variable efficacy and toxicity may result in inadequate therapeutic windows.¹²

In conclusion, BE-101, an ex vivo precision gene engineered BCM, produces active and sustained levels of FIX-Padua with the ability to engraft without host preconditioning. This unique biologic delivery system has the potential for redosing and titratability and may provide a novel treatment for PwHB irrespective of age and disease severity.

Materials and methods

BE-101 production

BE-101 was produced from peripheral blood B cells collected via leukapheresis from healthy human donors (STEMCELL Technologies, Cambridge, MA; iSpecimen, Lexington, MA; OrganaBio, South Miami, FL; and HemaCare, Lowell, MA) using commercially available selection kits (STEMCELL Technologies or Miltenyi Biotec, Charlestown, MA) in accordance with Institutional Review Board guidelines. Purified CD19⁺ B cells were cryopreserved prior to BE-101 production.

The 3-phase process used for activation, expansion, and differentiation of human primary B cells and production of engineered B cells secreting therapeutic proteins has been previously described,^16,62 and BE-101 was produced with few modifications. Isolated human B cells were thawed and cultured in Iscove’s modified Dulbecco’s medium with 10% fetal bovine serum and human cytokines (phase I media: IL-21, CpGs, and CD40L; phase II media: IL-2, IL-6, IL-10, and IL-15; phase III media: interferon-α2β, IL-6, and IL-15) to promote B cell differentiation into ASCs. After activation in phase I media for 2 days, human B cells were washed into electroporation buffer (Maxcyte, Gaithersburg, MD). Pre-complexed CRISPR-Cas9 and sgRNA RNP complexes (at a 2:1 guide:Cas9 ratio) were added (final concentration 4.58 μM), and the cell-RNP mixture was electroporated (B cell program no. 3). The sgRNA targeted the C-C chemokine receptor type 5 (CCR5) locus located at chromosome position 3p21.31. (Table S14 has the sgRNA sequence).

AAV6 vectors were packaged with a transgene encoding a human F9 gene (hyperactive FIX-R338L, Padua variant,^40,41,42 and Malmö T148A variant⁶³) downstream of the MND promoter,⁶⁴ followed by a bovine growth hormone gene polyadenylation signal, and flanked by 600-nt sequences homologous to genomic DNA on either side of the CCR5 sgRNA cut site located at exon 2 of CCR5 to mediate transgene insertion via HDR (Table S15 shows the annotated DNA sequence of the transgene expression cassette). Following electroporation with RNP complexes, AAV6 vector was added to cells at variable multiplicities of infection (MOIs) and incubated for 12–15 min at 37°C followed by dilution to 1 × 10⁶ cells/mL in serum-free media (ExCellerate or X-VIVO 10) supplemented with phase I cytokines, then expanded and differentiated toward the PC lineage during the remainder of the 3-phase process. High HDR rates were dependent on the choice of insertion site and specific guide, as well as the use of high-quality AAV, appropriate MOI, and transduction in serum-free media. Control cells were cultured similarly but without the addition of AAV6 vector.

The final BE-101 drug product was a sterile cell suspension. The cryopreserved preparation, in equal quantities of Plasma-Lyte A with 5% human serum albumin and CryoStor CS10, is identical to the product to be used in clinical trials.

BE-101 product characterization

Characterization of BE-101 included assessment of viability, cellular immunophenotype, on-target indels, and transgene integration. Cell counts and viability were determined using Cellaca MX (Nexcelom Bioscience, Lawrence, MA) or NC200 (ChemoMetec A/S, Lillerød, Denmark).

Cellular immunophenotyping was measured by flow cytometry. Cells suspended in 100 μL PBS in 96-well plates or 1.5-mL Eppendorf tubes received viability dye (100 μL) and were incubated at 4°C for 10–15 min followed by washing 2× with flow cytometry staining buffer (Invitrogen, Waltham, MA) and pelleting at 450 × g for 3 min. Following live/dead staining, surface-staining antibodies were added according to the manufacturer’s recommendation with specific fluorochrome-conjugated monoclonal antibodies APC-CD19, FITC-CD20, PECY7-CD27, BV421-CD38, and PE-CD138 (Becton Dickinson, Franklin Lakes, NJ, and BioLegend, San Diego, CA) and incubated for 25–30 min at 4°C. Cells were washed and pelleted, and the pellets were resuspended in 150 μL flow cytometry staining buffer to assess live singlet cells and B cell differentiation using the NovoCyte Penteon Flow Cytometer (Agilent Technologies, Santa Clara, CA). Over 10,000 events per sample were collected and analyzed using FlowJo software (BD Biosciences, Franklin Lakes, NJ).

Indel analysis at the on-target sgRNA cut site was performed as described previously⁶⁵ (Table S16 has sequences of primers designed via the IDT PrimerQuest tool). Transgene integration frequency was measured by ddPCR (Bio-Rad, Hercules, CA) using isolated gDNA from the final BE-101 product. ddPCR probes were designed in-house (Table S17 has sequences of designed ddPCR probes). Transgene integration quantification was calculated as copies/μL FAM⁺ droplets (target amplicon or engineered alleles)/copies/μL HEX⁺ droplets (reference amplicon or total alleles) × 100 = %Targeted integration (Table S17 has primers and probes).

Measurement of FIX-Padua secreted by engineered B cells in vitro and in vivo

FIX-Padua secretion in vitro was quantified by ELISA (Human Factor IX ELISA Kit, Abcam, Waltham, MA). In vivo analysis of plasma FIX-Padua was performed by ELISA following a selective human FIX protein capture step using human FIX monoclonal antibodies GMA102 and GMA184 (Green Mountain Antibodies, Burlington, VT) prior to determination of human FIX-Padua only in mouse plasma. FIX-Padua and Malmö variant did not interfere with the human FIX selective capture antibodies used to detect and measure the circulating human FIX-Padua protein derived from BE-101 in NOG-hIL6 mouse plasma. Activity of FIX-Padua was assessed using a chromogenic assay (Rox Factor IX chromogenic activity kit, Diapharma, West Chester, OH) following the selective human FIX protein capture step. FIX-Padua activity was also assessed in an aPTT assay following purification from cell culture supernatants using size-exclusion chromatography-ultraviolet (performed at NovaBioAssays Woburn, MA). Specific activity was calculated based on measured protein activity divided by calculated amount of protein present in the sample. Corresponding clotting times were converted to percent activity using a log-log regression; percent activity, IU/mL, and IU/mg are reported.

Process for potential off-target gene editing site nomination

In silico off-target prediction

Computational prediction of potential off-target sites in GRCh38 with homology to the intended target site of gCCR5_232 (with no more than 4 nucleotide mismatches and 2 bulges, CFD score ≥0.2,²⁰ and NGG-PAM sequence) was performed using CRISPRitz version 2.5.8.¹⁹

G-GUIDE

Transfections of primary human B cells from 5 unique donors were performed with Cas9-sgRNA RNP complexes containing 4.58 μM Cas9 (Aldevron, Madison, WI) and 2-fold molar excess Good Manufacturing Practice (GMP)-grade gCCR5_232 sgRNA. RNPs and 100 pmol of end-protected double-stranded oligodeoxynucleotides (dsODNs) were added directly to 2.75 × 10⁶ cells resuspended in MaxCyte buffer to a final volume of 55 μL before electroporation on the MaxCyte ATx using B cell program no. 3. Transfected cells were cultured for 48 h and collected for gDNA extraction. The G-GUIDE assay was provided/conducted by GeneGoCell. DNA was sheared by sonication to an average length of 500 bp, end-repaired, A-tailed, and ligated to Y-shape adapter. The adapter-ligated library was subjected to 3 rounds of nested PCR using dsODN sense- and antisense-specific primers in separate reactions. Libraries were quantified on a Qubit Fluorometer 4. Library size was measured on a TapeStation 4200. Successful libraries were then sequenced on an Illumina NextSeq 2000. G-GUIDE analysis was performed as previously described.²¹ Nomination of genomic loci as potential off-targets required the left and right side of the dsODN tag integration site to be detected (using GeneGoCell’s default settings) in both technical replicates in at least 1 B cell donor.

SITE-seq

SITE-seq and data analyses were performed as previously described.²² Libraries were quantified by qPCR and sequenced with 150-bp paired-end reads on an Illumina MiSeq. Calling of Cas9 cleavage sites required ≥5 reads terminating at the same nucleotide and a sequence with no more than 7 mismatches and 5 bulges relative to the intended target site of gCCR5_232 to be found in a ±35-bp window around the cut site. Sites were nominated as potential off-targets if found in all 3 technical replicates of at least 2 out of 6 profiled donors.

Assessment of DNA editing at potential gCCR5_232 off-target sites in BE-101

On- and off-target sites identified by CRISPRitz, G-GUIDE, and SITE-seq were amplified from genomic DNA of 5 batches of BE-101 cells and donor-matched unengineered control cells using multiplexed amplicon sequencing (rhAmp-seq, IDT, Coralville, IA) or standard amplicon sequencing. Sequencing libraries were generated according to the manufacturer’s instructions and sequenced in an Illumina MiSeq targeting a sequencing depth of ≥5,000 reads per amplicon. Indel analyses were conducted using CRISPAltRations.⁶⁶ For all sites reaching the targeted sequencing depth, an effect-size and p value were calculated by fitting a negative binomial generalized linear model. Statistically validated off-target sites were defined as those with estimated differential editing above 0.2% and a nominal p value below 0.05 in a 1-sided Z test.

Evaluation of genome integrity by OGM

Sample preparation and data acquisition for OGM (Bionano Genomics) were performed by Boston Children’s Hospital IDDRC Molecular Genetics Core Facility (Boston, MA). Ultra-high molecular weight gDNA from 3 lots of BE-101 cells and donor-matched unengineered controls was extracted and labeled, loaded on the flow cell Saphyr G3.3 chip, and imaged on the Saphyr instrument. RVA, de novo assembly, and variant calling were executed on Bionano Solve software (version 3.8.1).⁶⁷ Reporting and visualization of SVs was done on Bionano Access (version 1.8.1) using hg38 as the reference genome. Each engineered sample was compared against its donor-matched unengineered control. With 300× coverage, Bionano can detect SVs with 5% frequency and with over 90% sensitivity.

Animal studies for BE-101 engraftment, biodistribution, and persistence

Immunodeficient NOG-hIL6 mice (NOD.Cg-Prkdcscid Il2rgtm1SugTg(CMV-IL6)1-1Jic/JicTac, Taconic Biosciences, Rensselaer, NY) engineered to express anhIL-6 transgene,⁶⁸ 8–14 weeks old, were housed (up to 3 animals per cage) in a HEPA filtered, individually isolated cage system (Innovive, San Diego, CA) and acclimated for at least 3 days. Housing was an American Association for Accreditation of Laboratory Animal Care-accredited vivarium operated by SmartLabs (Cambridge, MA); all animal care and use procedures were approved by the SmartLabs Institutional Animal Care and Use Committee prior to study start. Mice were randomly assigned to study groups.

A BE-101 surrogate construct expressing firefly luciferase was used to assess biodistribution and persistence. Surrogate cells formulated in PBS buffer at a concentration of 5 × 10⁷ cells/mL were injected IV at 1 × 10⁷ viable cells/mouse. Cryo-preserved and thawed BE-101 cells were formulated in PlasmaLyte or PBS buffer at a concentration of 1 × 10⁸ cells/mL and injected IV at 2 × 10⁷ viable cells/mouse. For redosing, cryopreserved BE-101 cells were dosed at 2 × 10⁷ total viable cells/mouse on study days 0 and 49. Animals were observed twice per week for changes in body condition and weights. Blood samples were collected by tail bleeding into Microvette CB300 EDTA K2E containers (Sarstedt, Nümbrecht, Germany). Plasma FIX-Padua protein level and immunocapture chromogenic activity were measured as described above. At study end, mice were euthanized, and blood and tissue samples were collected for biodistribution, clinical chemistry, or histopathology assessments.

The supplemental information includes methodology on LC-MS/MS analysis of post-translational modification of the GLA domain in BE-101-derived FIX-Padua protein, nomination of potential off-target sites (i.e., CRISPRitz, G-GUIDE, and SITE-seq), human selective IgG and IgM measurements in mouse plasma, Alu PCR assay assessing BE-101 biodistribution, whole-body bioluminescence of a BE-101 surrogate construct expressing firefly luciferase dosed to NOG-hIL6 mice, and safety based on 28-day GLP toxicology and long-term in vivo pharmacology and tolerability studies in NOG-hIL6 mice.

Data availability

All relevant data are included in the paper. Any additional data are available on request from the corresponding author, Richard A. Morgan, at rmorgan@be.bio.

Acknowledgments

The authors thank Dr. David Rawlings and Dr. Richard James at Seattle Children’s Research Institute, Dr. Glenn Pierce at the World Federation of Hemophilia, and Dr. Wing Yen Wong and Dr. Krishnan Viswanadhan at Be Biopharma for reviewing and providing valuable input to the manuscript. The authors thank the Technical Development Team at Be Biopharma for generating BE-101 materials for in vivo studies and genotoxicity assessments. The authors also thank Dr. Daniel Ferguson for help with the BE-101 genotoxicity characterization. Patrice C. Ferriola, Ph.D., of KZE PharmAssociates aided in preparation of the manuscript and was funded by Be Biopharma.

Author contributions

H.L., S. Singh, T.J.M., C.B., S.K., T.P., S.T., A. Lundberg, R.C., C.Y., S.D., L.L., W.B., E.L., C.R.M., S.C., R.K., A.F.H., S.P.A., A. Lazorchak, and C.S. designed and/or performed the experiments; S. Shenker, T.K.O., E.L., and H.-M.C. performed computational data analysis; and H.L., S. Singh, T.J.M., S.K., C.R.L., H.-M.C., A.F.H., S.P.A., A. Lazorchak, and R.A.M. wrote the manuscript.

Declaration of interests

H.L., S. Singh, T.J.M., C.B., S.K., S.T., A. Lundberg, R.C., S.D., L.L. W.B., E.L., C.R.L., H.-M.C., R.K., A.F.H., S.P.A., A. Lazorchak, and R.A.M. reported being current employees at Be Biopharma. H.L., S. Singh, T.J.M., C.B., S.K., T.P., S.T., A. Lundberg, R.C., C.Y., S.D., L.L. W.B., E.L., C.R.M., C.R.L., T.K.O., S.C., H.-M.C., R.K., A.F.H., S.P.A., A. Lazorchak, and R.A.M. reported holding stock options in Be Biopharma, a privately held company. T.J.M., H.L., A.F.H., S.P.A., S. Singh, S.K., and R.A.M. are co-inventors of the patent “Engineered cell preparations for treatment of hemophilia” (WO2024220446A2).