Genetic Retargeting of E3 Ligases to Enhance CAR T Cell Therapy

Summary

Chimeric antigen receptor (CAR) T cell therapies are medical breakthroughs in cancer treatment. However, treatment failure is often caused by CAR T cell dysfunction. Additional approaches are needed to overcome inhibitory signals that limit anti-tumor potency. Here, we developed bifunctional fusion “degrader” proteins that bridge one or more target proteins and an E3 ligase complex to enforce target ubiquitination and degradation. Conditional degradation strategies were developed using inducible degrader transgene expression or small molecule-dependent E3 recruitment. We further engineered degraders to block SMAD-dependent TGFβ signaling using a domain from the SARA protein to target both SMAD2 and SMAD3. SMAD degrader CAR T cells were less susceptible to suppression by TGFβ and demonstrated enhanced anti-tumor potency in vivo. These results demonstrate a clinically suitable synthetic biology platform to reprogram E3 ligase target specificity for conditional, multi-specific endogenous protein degradation, with promising applications including enhancing the potency of CAR T cell therapy.

eTOC Blurb

Many genes that limit CAR T cell therapy are difficult to target with existing modalities. Lane et al. developed bifunctional degrader proteins to bridge an E3 ligase and one or more endogenous target proteins. A multi-specific SMAD degrader reduced CAR T cell responsiveness to TGFβ and enhanced anti-tumor potency.

Graphical Abstract:

Introduction

Cellular immunotherapies have emerged as medical breakthroughs in cancer therapy. Next-generation approaches will encode genetic modifications that further enhance expansion, dysfunction-resistance, and effector function. A dominant strategy for the identification of genes that limit anti-tumor T cell potency is CRISPR/Cas9 knockout screening^1,2; CRISPR knockout T cell therapies have advanced to clinical testing³. However, many genes that limit anti-tumor T cell potency remain challenging to target. First, irreversible knockout of genes related to hematologic malignancies such as TET2 and DNMT3A may increase the risk of engineered cell transformation^4–6. Second, core pathways that limit immune cell functions are most often genetically redundant and thus difficult to target without inducing multiple genome edits and DNA double strand breaks. Examples include multiple inhibitory receptors and suppressive transcription factor ensembles^7,8. Third, negative regulatory genes often also serve beneficial roles of preventing overstimulation, thus favoring conditional perturbation to irreversible genetic ablation^9,10. In sum, innovative technologies are needed for conditional, multi-specific control of proteins that limit cellular immunotherapy function without targeting genetic loci associated with cancer.

Inducible protein-protein interactions to control endogenous protein abundance is an alternative approach to rewire cellular circuitry¹¹ and target suppressive pathways in cellular immunotherapies. Indeed, the first proteolysis targeting chimeras (PROTACs) were partially peptide-based bifunctional molecules that bridged E3 ubiquitin ligases and target proteins, thereby inducing target protein ubiquitination and degradation¹². This E3 recruitment strategy to induce target protein depletion has been used to induce the degradation of GFP-tagged fusion proteins in model systems¹³, as well as targeting endogenous proteins such as beta-catenin and PCNA in human cells^14–16. While such bifunctional fusion proteins are emerging as well-validated experimental tools, they have not to the best of our knowledge been envisioned for use in gene- and cell-based therapies.

Here, we report the development of all-human sequence E3-recruiting fusion proteins, termed SYnthetic Substrate Receptors (SySRs) to induce conditional, multi-specific degradation of sets of proteins that limit CAR T cell potency. We optimized binding domains for multiple broadly expressed E3 ligases, enabling efficient >90% target degradation in multiple cell types. For synthetic circuit applications, two strategies were explored for conditional target degradation gated by inducible promoters and lenalidomide-inducible target recruitment to CRL4^CRBN. To demonstrate the concept of cell therapy application, we developed SySRs targeting the multiple SMAD transcription factors that enforce T cell-suppressive TGFβ signaling. SMAD degradation enhanced CAR T cell proliferation and anti-tumor function. SySRs are a generalizable and clinically suitable platform to degrade suppressive protein ensembles to augment next-generation cellular immunotherapies.

Design

Many of the proteins that suppress cell therapies remain undruggable, and the genes that encode them are often not amenable to genetic knockdown or knockout. To overcome these challenges, we envisioned a technology with which to deplete endogenous suppressive proteins using transgene-based targeted protein degradation. The transgene-encoded engineered proteins were directly inspired by PROTACs and other modular, bifunctional molecules that induce degradation by recruiting a neosubstrate to an E3 ubiquitin ligase. This design strategy enabled modular programming with native or synthetic target binding domains, and incorporation into synthetic circuits or user-controlled chemical genetic switches. Finally, these clinically suitable, human sequence-based degraders have the potential for direct translation to enhance gene- and cell-based therapies, as exemplified by co-delivery with a CAR and proof-of-concept to overcome suppressive signals that limit anti-tumor T cell potency.

Results

SySR expression redirected E3 ligase activity for target protein degradation

We first sought to develop GFP-targeted degrader fusion proteins for platform optimization. These SySRs were composed of a previously described GFP-binding nanobody vhhGFP4¹⁵, a glycine-serine linker, and an E3 ubiquitin ligase-binding domain (Table S1). The SySR proteins are designed to simultaneously bind to an E3 ubiquitin ligase complex and the target protein, thereby mediating ubiquitination and degradation of the target protein (Figure 1A). SySRs with ligase-binding domains from multiple genes, especially SOCS2, Vpx, and Vpr efficiently reduced GFP abundance (Figure 1B–C and Figure S1A). GFP depletion was partially inhibited by treatment with the neddylation inhibitor MLN7243, consistent with ubiquitin-dependent target degradation (Figure 1D). CRL4 and CRL5 E3 ubiquitin ligase complexes are broadly expressed¹⁷, therefore we hypothesized that SySRs targeted to these E3 ligase complexes using domains from SOCS2 and Vpx would be effective in multiple cell types. GFP degradation with SOCS2- and Vpx-based degraders was efficient in HEK293 cells, multiple T cell lines, and multiple myeloid cell lines (Figure 1E). In summary, GFP-targeted SySRs engaging multiple E3 ubiquitin ligase complexes efficiently degraded the target protein in multiple cell types.

Figure 1. Reprogramming E3 ligase specificity using synthetic substrate receptors (SySRs).

(A) Schematic of the degradation system using a GFP-binding SySR. (B) Representative GFP intensity in Jurkat cells that were untransduced, transduced with GFP only, or co-transduced to express GFP and vhhGFP4-SOCS2. (C) Normalized GFP fluorescence intensity in Jurkat cells dually transduced with GFP and a GFP degrader composed of the vhhGFP4 nanobody linked to the indicated ligase-binding domain. Control cells were singly transduced with GFP. (D) Normalized GFP fluorescence intensity in Jurkat cells treated with the neddylation inhibitor MLN7243. Cells were treated with 0.5uM MLN7243 for 4 hours and analyzed by flow cytometry. Two-sided student’s t-test was performed. (E) Normalized GFP fluorescence intensity in multiple cell lines transduced with GFP and either the SOCS2- or Vpx-based GFP degrader. Two-sided student’s t-test was performed. (F) Cells were exposed to 2 ng/mL recombinant IFNɑ for the indicated time period and then assessed by intracellular flow cytometry for pSTAT1 abundance. Two-sided ANOVA was performed. Experiments performed in technical triplicate (D, F) or quadruplicate (C,E). Error bars indicate mean ± SD. ns, P > 0.05.

The overexpression of fusion proteins with E3 ligase substrate receptor domains could alter protein degradation by the E3 ligase complex. In theory, endogenous substrate receptors could be displaced by the synthetic proteins, leading to stabilization of endogenous substrate proteins. SOCS2 is one of multiple SOCS box-containing proteins that bind to the CRL5 E3 ligase and act as substrate receptors to direct CRL5 to target proteins. Because cells can simultaneously co-express many endogenous SOCS box-containing proteins, we hypothesized that the expression of one additional SOCS box-containing SySR would not disrupt endogenous CRL5-dependent functions. Type 1 interferons induce JAK/STAT signaling and downstream upregulation of multiple SOCS genes, which then enforces a negative feedback loop to inhibit and deplete phosphorylated STAT1 (pSTAT1)^18,19. Indeed, SOCS1 and SOCS3 are sufficient to suppress interferon-dependent pSTAT1, and SOCS1 or SOCS3 knockdown or knockout potentiates pSTAT1 after interferon stimulation^20–23. In Jurkat cells stimulated with IFNɑ, the dynamics of pSTAT1 accumulation and depletion were comparable with or without SOCS2-based SySR expression (Figure 1F). To assess for systems-level effects, we performed proteome-scale mass spectrometry on Jurkat cells with or without SOCS2-based SySR expression (Figure S1B). Gene set enrichment analysis did not identify differentially regulated biologic processes with or without SySR expression (Table S2). In summary, expression of a SOCS2-based SySR had no observable effect on a canonical CRL5 substrate or biologic processes described by gene ontology terms.

Targeting endogenous proteins for degradation

We next sought to develop degraders of proteins that limit CAR T cell therapy. TGFβ is a central mechanism of immunosuppression in the tumor microenvironment²⁴. Many strategies to inhibit TGFβ signaling have been developed, including dominant negative TGFβR2 proteins that have been deployed in solid tumor-targeted CAR T cell clinical trials^25,26. As a first proof of concept, we hypothesized that degrading the SMAD transcription factors downstream of TGFβ receptors could more completely block suppressive SMAD signaling than receptor modifications, potentially also blocking alternative pathways to SMAD activation that have been implicated in cancer immune escape^27,28 (Figure 2A). SMAD2 and SMAD3 transcription factors are recruited to the cell surface and TGFβ-family receptors by the protein SARA²⁹. We created a SARA-SOCS2 fusion protein (Figure 2B), and this putative SMAD degrader induced the depletion of SMAD2-GFP and SMAD3-GFP fusion proteins (Figure 2C–D). The SARA-SOCS2 SMAD degrader reduced endogenous SMAD2/3 abundance (Figure 2E–F) and also reduced SMAD transcriptional activity (Figure 2G). In summary, using natural target binding domains, we developed a multi-specific endogenous protein degrader.

Figure 2: Targeted degradation of cell therapy-relevant endogenous proteins.

(A) Schematic of SMAD2/3-dependent signaling in the TGFβ pathway. (B) Schematic of the degradation system using a SARA-based SySR to bind SMAD2/3. (C) Normalized GFP fluorescence intensity in Jurkat cells dually transduced with a SMAD2- or SMAD3-GFP reporter, with or without SARA-SOCS2 co-transduction. (D) Flow plot showing GFP and mTagBFP2 intensity in Jurkat cells dually transduced with either a SMAD2- or SMAD3-GFP reporter and SARA-SOCS2 with a mTagBFP2 transduction marker. (E) Endogenous SMAD2/3 intensity in Jurkat cells transduced with SARA-SOCS2. Cells were analyzed via intracellular flow (ICF). (F) Mean fluorescent intensity (MFI) of SMAD2/3 in Jurkat cells transduced with SARA-SOCS2. (G) Normalized levels of SEAP transcriptional reporter activity in HEK Blue TGFβ cells with or without SARA-SOCS2 expression. (H) Schematic of synNotch circuit with a CD19-directed priming receptor coupled to transcriptional control of BFP only, BFP and vhhGFP4-SOCS2, or BFP and SARA-SOCS2 expression. (I) GFP fluorescence intensity of synNotch GFP degrader circuit-activated (BFP+) Jurkat cells normalized to BFP− Jurkat cells after co-culture with CD19+ NALM6 cells. (J) SMAD 2/3 MFI of SMAD degrader circuit-activated (BFP+) Jurkat cells normalized to BFP− Jurkat cells after co-culture with CD19+ NALM6 cells. Two-sided student’s t-tests were performed as indicated. Experiments performed in technical triplicate. Error bars indicate mean ± SD.

Engineering conditional protein degradation systems

While regulated perturbation may not be specifically required to engineer optimal TGFβ-insensitive CAR T cells, we sought to develop generalizable methods for conditional endogenous protein degradation that could be applied to future targets. Drug-inducible E3 ligase recruitment could enable target protein degradation with fast dynamic control. Therefore, we sought to develop an inducible SySR using a previously engineered zinc-finger domain whose association with the CRL4^CRBN E3 ubiquitin ligase is conditional upon the addition of the molecular glue anti-cancer drug lenalidomide or its analogs³⁰ (Figure S2A). This inducible SySR rapidly degraded GFP, as the majority of GFP depletion occurred within one hour after lenalidomide treatment (Figure S2B). The half-maximal degradation concentration was approximately one nanomolar (Figure S2C). A PD-1-GFP fusion protein was effectively degraded with lenalidomide (Figure S2D–E). Given these promising performance characteristics with inducible GFP degradation, we next generated a lenalidomide-inducible SMAD-targeted degrader composed of a SARA-zinc-finger fusion protein. When co-expressed with SMAD2-GFP and SMAD3-GFP reporters, the lenalidomide-inducible SMAD degrader demonstrated drug-independent destabilization of the target protein and only partial lenalidomide-induced target degradation (Figure S2F–G). In summary, lenalidomide-inducible SySRs enabled control of target protein abundance with small molecule kinetics. While the lenalidomide-inducible SMAD degrader had inferior performance characteristics compared to the constitutive SARA-SOCS2 SySR, the drug-inducible system may have future utility for targeting other endogenous proteins.

Having explored post-translational control of SySR function, we developed a second approach to conditional regulation of target protein degradation using synthetic circuits to regulate SySR transgene transcription. We generated a synNotch circuit³¹ in which a CD19 sensor induced the expression of a GFP degrader transcriptional response element (Figure 2H). When co-cultured with CD19+ tumor cells, Jurkat-synNotch cells induced to express the degrader transgene depleted GFP (Figure 2I). When a SARA-SOCS2 SySR was substituted for the GFP-targeted SySR in the synNotch circuit, the cells responding to the CD19 stimulus depleted endogenous SMAD2/3 (Figure 2J). In summary, a synthetic circuit enabled conditional SMAD degradation in response to tumor antigen recognition. Taken together, we performed platform development targeting GFP and credentialed both constitutive and conditional SMAD degraders in cell lines. Given that constitutive TGFβ blockade is a promising strategy to enhance CAR T cell potency²⁶, we next advanced a constitutive and multi-specific SMAD degrader for testing in primary human CAR T cells.

SMAD Degrader CAR T cells

To test the function of CAR T cells additionally engineered to express SySRs targeting proteins that limit T cell functions, we generated multi-cistronic lentivectors encoding the SMAD degrader and an anti-CD19 CAR (Figure 3A). The SMAD degrader CAR T cells had reduced CAR surface abundance, as expected with a longer multi-cistronic transgene (Figure 3B). In overnight and multi-day NALM6 co-culture assays, without exogenous supplementation of TGFβ the SMAD degrader CAR T cells had reduced (Figure 3C) and comparable (Figure 3D) target cell lysis versus conventional CAR T cells, respectively. With TGFβ in the same assays, the SMAD degrader CAR T cells had comparable (Figure 3C) and increased (Figure 3D) NALM6 cytolysis versus conventional CAR T cells, respectively. Overall, cytolytic function was robust in all conditions and the effect sizes of TGFβ- and SMAD degrader-dependent differences in short-term target cytolysis were modest. In the same multi-day live cell imaging co-culture experiment, the SMAD degrader CAR T cells proliferated more than conventional CAR T cells (Figure 3D).

Figure 3: Proliferation and tumor cytolysis of SMAD degrader CAR T cells.

(A) Design of a CD19-targeted CAR lentivectors, with or without the SMAD degrader SARA-SOCS2 (B) Intensity of mCherry and CAR, as detected with an extracellular Myc epitope tag, in primary T cells transduced with either a conventional CD19 CAR or a SMAD degrader CD19 CAR. (C) Percent cytolysis of NALM6 tumor cells after overnight co-culture with CAR T cells at multiple effector:target ratios, with or without 10 ng/mL recombinant TGFβ1. Two independent experiments with separate normal donors are presented, each experiment performed in technical triplicate, two-sided ANOVA tests were performed comparing CAR constructs in each TGFβ condition. UTD, untransduced primary T cells. (D) Normalized tumor and CAR T cell area in Incucyte live cell imaging co-culture assays. CAR T cells were cultured at a 1:1 ratio with CD19-expressing NALM6 tumor cells for 93h. Two independent experiments with separate normal donors are presented, each experiment performed in technical triplicate, two-sided student’s t-tests were performed with the final 93h imaging timepoint comparing CAR constructs or UTD in each TGFβ condition. Error bars indicate mean ± SD. ns, P > 0.05.

To further evaluate the long-term proliferation of conventional and SMAD degrader CAR T cells, these cells were repetitively stimulated with irradiated K562 cells engineered to express CD19 with and without exogenous TGFβ. Whereas conventional CAR T cell proliferation was inhibited by TGFβ, SMAD degrader CAR T cell proliferation was not inhibited by TGFβ (Figure 4A), and without exogenous TGFβ the SMAD degrader CAR T cell proliferation was greater than control in one of two donors (Figure 4A and Figure S3). Taken together, across the multi-day co-culture model and the multi-week repetitive stimulation assay in two T cell donors with and without exogenous TGFβ, the SMAD degrader CAR T cells demonstrated increased proliferation versus the conventional CAR T cells in seven out of eight comparisons. The CAR T cells differentiated to effector subsets over the course of the assay with both constructs (Figure 4B and Figure S4). Dynamics of the surface abundance of the inhibitory receptors PD-1, LAG-3, and TIM-3 were complex (Figure S5); overall, the co-expression of multiple inhibitory receptors was less frequent for the SMAD degrader CAR T cells without exogenous TGFβ in most cases, and the enhanced proliferation of the SMAD degrader CAR T cells with TGFβ was associated with more frequent co-expression of multiple inhibitory receptors (Figure 4C and Figure S3). Dynamics of cytokine expression after stimulation were also complex, and enhanced proliferation of the SMAD degrader CAR T cells with TGFβ was not associated with a reduction in cytokine responsiveness after stimulation (Figure 4D). In summary, SMAD degrader CAR T cell proliferation was not suppressed by TGFβ, and this resistance to TGFβ-mediated suppression was associated with increased inhibitory receptor expression but no loss in cytokine responsiveness after multiple rounds of antigen stimulation.

Figure 4: Enhanced long-term proliferation in vitro with SMAD degrader CAR T cells.

(A) Normalized CAR T cell expansion after repetitive stimulation with irradiated CD19+ K562 tumor cells. K562 cells were added on Days 0, 7, 14, and 21. A flow cytometry panel to assess T cell phenotype and inhibitory receptor surface abundance was performed on Days 0, 7, 14, 21, and 28. An intracellular flow cytometry panel to assess IFNɣ and TNFɑ abundance was performed on Days 0, 1, 8, 15, and 22. Two-sided ANOVA tests were performed. (B) Percentage of CCR7-CD45RA− T_EM and CCR7-CD45RA+ T_EMRA cells of CD4+ and CD8+ cells. (C) Percentage of CD4+ and CD8+ cells expressing either two or three inhibitory receptors of PD-1, TIM-3, and/or LAG-3. (D) Percentage of CD4+ and CD8+ cells expressing IFNɣ or co-expressing IFNɣ and TNFɑ. For B, C, and D, two-sided student’s t-tests of the final timepoint are shown. Experiment performed with or without continuous exposure to 10 ng/mL recombinant TGFβ1, with two independent T cell donors in technical triplicate. Donor 1 is shown in this figure, Donor 2 is shown in Figure S3. Error bars indicate mean ± SD. ns, P > 0.05.

To evaluate the transcriptional response of conventional and SMAD degrader CAR T cells to TGFβ, we performed gene expression profiling with a 780 gene expression panel on FACS-purified conventional or SMAD degrader CAR T cells after 24 hours of exposure to recombinant TGFβ (Figure 5A, Figure S6, Table S3, Table S4). Whereas we identified 24 genes in conventional CAR T cells that were upregulated with TGFβ, four of these were also upregulated in SMAD degrader CAR T cells with TGFβ (Figure 5B). Similarly, we identified 24 genes in conventional CAR T cells that were downregulated with TGFβ, and one of these was also downregulated in SMAD degrader CAR T cells with TGFβ (Figure 5B). To examine a subset of previously reported T cell TGFβ-responsive genes^32,33 by an orthogonal mechanism, we measured transcript abundance for STAT3, GZMA, and ITGAM with qRT-PCR, and observed that the fold expression change in response to TGFβ was reduced for each gene with the SMAD degrader CAR T cells (Figure 5C). In summary, the SMAD degrader inhibits the transcriptional response of CAR T cells to TGFβ.

Figure 5: Reduced responsiveness of SMAD degrader CAR T cells to TGFβ.

(A) Log₂(fold change) in gene expression upon TGFβ treatment in conventional CAR19 and CAR19 SMAD degrader T cells. (B) Venn diagrams illustrating significantly upregulated and significantly downregulated genes in CAR19 and CAR19 + SMAD degrader. A p-value of 0.01 was used as the significance cutoff. (C) Three TGFβ-related gene probes (SMAD3, GZMA, and ITGAM) were chosen for further gene expression analysis with qRT-PCR. Fold change in the expression of each gene upon TGFβ treatment were shown, in conventional CAR19 and in CAR19 + SMAD degrader. Two-sided student’s t-tests were performed. Error bars indicate mean ± SD.

To test in vivo anti-tumor potency, we next compared conventional and SMAD degrader CAR T cells in a murine xenograft JEKO-1 tumor model (Figure 6A). The SMAD degrader CAR T cells depleted tumor cells in all mice across two normal donors (Figure 6B). The SMAD degrader CAR T cells exhibited increased expansion in the peripheral blood 14 days after injection (Figure 6C). Overall, CAR T cell counts in the blood across multiple timepoints were comparable (Figure 6D). On day 46, CAR T cell chimerism was comparable in the bone marrow and spleen (Figure 6 E–F). In summary, addition of the SMAD degrader enhanced in vivo CAR T cell early expansion and anti-tumor potency.

Figure 6: Enhanced in vivo anti-tumor potency with SMAD degrader CAR T cells.

(A) Schematic of in vivo experimental design. (B) Whole mouse bioluminescence, reflective of tumor burden. Thin lines denote individual mice, and thick lines denote group means. Mann-Whitney test of endpoint bioluminescence was performed. (C) Absolute quantification of CAR T cells in peripheral blood. Two-sided student’s t-test was performed. One SMAD degrader sample was aspirated during sample preparation and thereby lost. (D) Absolute quantification of CAR T cells in peripheral blood from multiple timepoints. Two-sided mixed effects model analysis was performed. (E) Percentage CAR T cells of total bone marrow cells. Two-sided student’s t-test was performed. (F) Percentage CAR T cells of total spleen cells. Two-sided student’s t-test was performed. CAR T cells were manufactured from two normal donors, with five mice per construct and donor. Error bars indicate mean ± SD. ns, P > 0.05.

Discussion

Here, we developed bifunctional fusion proteins to bridge E3 ligases and target proteins, with enforced proximity resulting in target ubiquitination and degradation. For conditional target degradation, inducible promoters and a lenalidomide-dependent E3 recruitment system were credentialed for targeting GFP. Using a subset of these E3 recruitment strategies and a binding partner of cell therapy-relevant proteins, we developed multi-specific degraders of SMAD2 and SMAD3. Expression of these synthetic substrate receptors was well-tolerated in T cells. A constitutively expressed multi-SMAD degrader armored CAR T cell against suppression by TGFβ and enhanced in vivo anti-tumor potency. Compact (as small as 121 amino acids) and human-sequence derived, here we advance a clinically suitable gene therapy technology well-positioned to augment next-generation adoptive immune cell therapies.

TGFβ is a central immunosuppressive signal in tumor microenvironments^34,35. Many therapeutic modalities have been developed to target TGFβ signaling in cancer²⁴. In the context of genetically modified cellular immunotherapy, CAR T cells have been engineered with dominant negative TGFβR2 receptors, switch receptors that convert TGFβ binding into stimulatory signaling, CARs activated by soluble TGFβ ligation, and CRISPR/Cas9-based TGFβR2 knockout^25,26. These are all attractive and, in some cases, clinically validated approaches to overcome TGFβ-mediated T cell dysfunction. We explored an alternative approach to this central pathway to highlight engineering possibilities enabled by the direct targeting of ensembles of transcription factors that often act as central nodes integrating diverse upstream signaling inputs. In comparison to established approaches, the SMAD degrader is designed to 1) prevent ligand-independent SMAD transcriptional activity, 2) prevent SMAD-dependent but not SMAD-independent TGFβ signaling, and 3) overcome escape mechanisms from TGFβ receptor blockade such as receptor upregulation or alternative TGFβ-family ligand/receptor signaling^27,28,36. A comparable genetic knockout strategy, for example CRISPR/Cas9 targeting of SMAD2 and SMAD3, with present technologies would require Cas9 delivery and the simultaneous generation of four double-strand breaks in clinical manufacturing. By comparison, inclusion of a 121 amino acid SMAD degrader transgene in the CAR lentivector is a promising approach to enhance CAR T cell potency and overcome TGFβ-associated dysfunction that adds minimal additional complexity to cell therapy product manufacturing. Beyond TGFβ, the multi-specificity of protein-protein interaction-mediated degradation could in the future allow for targeting of larger gene families such as inhibitory KIRs.

Chemically inducible dimerization and degradation underlie cornerstone chemical genetic systems that have enabled the remote control of CAR signaling, in some cases to augment potency by preventing excessive tonic signaling^37–39. While molecular glues and PROTACs have been used extensively to regulate the abundance and function of CARs or other fusion proteins tagged with drug-interaction domains with fast ON/OFF dynamics, examples of small molecule degraders specific for cell therapy-relevant endogenous proteins are exciting⁴⁰ but few and limiting. SySRs, bioPROTACs, and other genetic degraders leverage an orthogonal and exceptionally broad potential design space of protein-protein interactions to enforce endogenous target protein degradation in engineered cells. As building blocks for synthetic circuits, inducible expression of SySRs can be used to link cell state to endogenous protein abundance. In future iterations, recognition of tumor antigen or microenvironment cues can be used to gate SySR expression, allowing for the depletion of potent negative regulators of cellular immunotherapies that require spatiotemporal control. SySRs are generalizable elements amenable to diverse next-generation gene- and cell-based therapeutic applications.

Limitations of the study

With respect to limitations, each target-specific degrader is bespoke and requires individual optimization. Target-binding domains are a prerequisite, although scalable affinity-reagent panning can be performed for targets without suitable binding partners⁴¹. As with small molecule PROTACs, for a given liganded target protein, only a subset of E3 recruitment strategies are likely to enforce robust degradation for a given target. These complexities in many ways mirror the development path of PROTAC chemistry, as platform-defining measurements at scale⁴² lead to more precise, programmable chemistry and now therapeutic campaigns with extraordinary potential in advanced clinical development. A similar path of optimization and rational development can be envisioned for protein-based degraders. In vivo conditional target degradation to enhance CAR T cell potency was not attempted in this study, largely because the simpler strategy of constitutive SMAD degrader expression was effective in protecting CAR T cells from TGFβ-mediated suppression and enhancing in vivo anti-tumor potency. Future SySR development to safely enhance engineered cell function will likely require conditional degradation for some targets, including proteins with dual roles in beneficial negative feedback and T cell dysfunction, such as PD-1 and TOX, and proteins with cancer risk from irreversible genetic knockout, such as TET2 and DNMT3A.

Significance

Many genes limiting cellular immunotherapies are difficult to target using existing methods. Instead, these genes may be optimally targeted with precision perturbations at the protein level to minimize transformation risk, overcome genetic redundancy, and optimize dynamics and tuning. In contrast to dominant genetic strategies such as CRISPR/Cas9-based knockout and shRNA, here we present a transgene-based targeted protein degradation system to enforce conditional and/or multi-specific target protein knockdown in engineered cells. Clinically suitable, all-human sequences were used to develop a degrader of multiple suppressive transcription factors that enhanced CAR T cell potency. This approach to targeted endogenous protein degradation has implications beyond CAR T cells, with widespread applications in gene- and cell-based therapy.

STAR★METHODS

Resource Availability

Lead contact

All requests for reagents and resources should be directed to the lead contact, Max Jan (mjan@mgh.harvard.edu).

Materials availability

All sequences are provided in Table S1 and can be regenerated by DNA synthesis.

Data and code availability

• Mass spectrometry raw data have been deposited at the MassIVE data repository (massive.ucsd.edu) and are publicly available as of the date of publication. Accession numbers are listed in the key resources table.

• This paper does not report original code.

• Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Key resources table.

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
CCR7	BD Biosciences	Cat#561271
PD1	BD Biosciences	Cat#563789
TIM3	BD Biosciences	Cat#565566
LAG3	BD Biosciences	Cat#565716
CD45	BioLegend	Cat#304126
CD8	BD Biosciences	Cat#647458
CD45	BD Biosciences	Cat#651850
CD3 (APC-H7)	BD Biosciences	Cat#641406
GZMB	BioLegend	Cat#372204
phospho-STAT1 (Tyr701)	Cell Signaling Tech	Cat#9174S
SMAD2/3 (PE)	Cell Signaling Tech	Cat#72255
SMAD2/3 (Biotinylated)	Cell Signaling Tech	Cat#12470
Streptavidin-APC	BioLegend	Cat#405207
IFNgamma	BioLegend	Cat#506504
TNFalpha	BioLegend	Cat#502930
Ly6G/C	BioLegend	Cat#108412
Ter-119	BioLegend	Cat#116212
CD11b	BioLegend	Cat#101212
NK1.1	BioLegend	Cat#108710
CD3 (BUV395)	BD Biosciences	Cat#563548
CD71	BioLegend	Cat#334102
Myc-tag	Cell Signaling Tech	Cat#2233S
Chemicals, peptides, and recombinant proteins
DAPI	Biolegend	Cat#422801
BV Stain Buffer	BD Biosciences	Cat#566349
Opti-mem	Thermo Scientific	Cat#31985088
Rpmi 1640 Medium, GlutaMAX™, HEPES	Thermo Scientific	Cat#72400120
Phosphate Buffered Saline	Thermo Scientific	Cat#20012050
Fetal Bovine Serum	Thermo Scientific	Cat#A3160502
Penicillin/Streptomycin	Thermo Scientific	Cat#15140122
CD3/CD28 Dynabeads	Thermo Scientific	Cat#11132D
Recombinant human IL-2	Peprotech	Cat#200–02
DMSO	Thermo Scientific	Cat#85190
Luciferase Cell Culture Lysis 5x Reagent	Promega	Cat#E1531
Interferon alpha	Invivogen	Cat#rcyc-hifna2b
Cytofix/Cytoperm Kit	BD Biosciences	Cat#554714
TGFbeta1	Peprotech	Cat#100–21
T cell Rosette Sep Isolation Kit	StemCell Tech	Cat#15061
Nano-glo Luciferase Assay System	Promega	Cat#N1110
Gibson Assembly Master Mix	New England Biolabs	Cat#E2611L
FACS Lysing Solution	BD Biosciences	Cat#349202
Critical commercial assays
TaqMan Gene Expression Master Mix	Thermo Scientific	Cat#4369016
GAPDH VIC	Thermo Scientific	Cat#4448489; Assay ID Hs02786624_g1
SMAD3 FAM	Thermo Scientific	Cat#4453320; Assay ID Hs00969210_m1
GZMA FAM	Thermo Scientific	Cat#4453320; Assay ID Hs00989184_m1
ITGAM FAM	Thermo Scientific	Cat#4453320; Assay ID Hs00167304_m1
SuperScript IV VILO Master Mix with ezDNase	Invitrogen	Cat#2671536
nCounter Master Kit Prep Pack, Cartridge & Prep Plate	NanoString Tech	Cat#050523
nCounter 12-well Notched Strip tubes	NanoString Tech	Cat#207108
nCounter 12-well Notched Strip tube lids	NanoString Tech	Cat#104727
nCounter Prep Station Tips	NanoString Tech	Cat#JN221001
nCounter CAR-T Panel	NanoString Tech	Cat#050523
Trucount Absolute Concentration Tubes	BD Biosciences	Cat#340334
Deposited data
Proteomics data generated in this study	MassIVE data repository	MSV000092850
Experimental models: Cell lines
HEK293	ATCC	Cat#CRL-3216
HEK Blue TGF-beta	InvivoGen	Cat#hkb-tgfbv2
Jurkat	ATCC	Cat#TIB-152
SupT1	ATCC	Cat#CRL-1942
NALM6	ATCC	Cat#CRL-3273
JEKO-1	ATCC	Cat#CRL-3006
Kasumi	ATCC	Cat#CRL-2724
K562	ATCC	Cat#CCL-243
OCI-AML2	DSMZ	Cat#ACC 99
Experimental models: Organism/strains
NSG mice (NOD.Cg-PrkdcscidIL2rgtm1Wjl/SzJ)	Jackson Laboratory	Cat#005557
Recombinant DNA
SySR sequences	This paper	See Table S1
Software and algorithms
Prism	Graphpad	www.graphpad.com
LIMMA	Ritchie et al.49	https://bioconductor.org/packages/release/bioc/html/limma.html
gsea2–2.2.2	Subramanian et al.50	www.gsea-msigdb.org
MSigDBv2022	Liberzon et al.51	www.gsea-msigdb.org
NSolver	NanoString Tech	https://nanostring.com/
FlowJo	FlowJo, LLC	www.flowjo.com
Aura	Spectral Instruments Imaging	spectralinvivo.com
R	R Core Team52	www.R-project.org

Experimental Model and Study Participant Details

Cell lines

HEK293 (female, fetal), Jurkat (male, 14 year-old), SupT1 (male, 8 year-old), NALM6 (male, 19 year-old), JEKO-1 (female, 78 year-old), Kasumi (male, 7 weeks-old), and K562 (female, 53 year-old) cell lines were obtained from ATCC. OCI-AML2 (male, 65 year-old) cells were obtained from DSMZ. HEK Blue TGF-beta (female, fetal) cells were obtained from InvivoGen. Cell cultures were maintained at 37 C and 5% carbon dioxide, and RPMI 1640 medium with GlutaMAX and HEPES supplemented with 10% fetal bovine serum (FBS), penicillin, and streptomycin unless otherwise indicated. Mycoplasma-free cultures were confirmed routinely by PCR.

Human T cells

Human T cells were purified by negative selection according to manufacturer’s instructions (Stem Cell Technologies, 15061) from anonymous male and female human healthy donor leukopaks obtained from the Massachusetts General Hospital blood bank under an Institutional Review Board–approved protocol (2016P001219). After selection, the enriched T cells were cryopreserved in multiple aliquots in 90% FBS and 10% DMSO to allow for multiple donor-matched experiments. All donor attributes, including sex, were de-identified prior to specimen receipt. Cells from at least two donors were used in primary cell experiments, and cells from three donors were used in gene expression experiments performed in biologic triplicate. For the in vivo studies, sample sizes as indicated in the Figure 6 legend were consistent with prior donor and sample allocation^30,43. Prior to T cell transfer, the mice were randomly assigned to groups on the basis of tumor burden.

Animals

NOD.Cg-Prkdc^scid Il2rg^tm1Wjl/SzJ (NSG) mice were obtained from Jackson Laboratory and maintained in a breeding colony at MGH. Mice were housed in cages of up to 5 mice under pathogen-free conditions, maintained at 70–76 F, 30–70% humidity, and 12-hour light/12-hour dark cycles. The mice used in experiments were female mice between 6–10 weeks of age that were genetically immunosuppressed but otherwise healthy, and were naïve to previous procedures or treatments. Animal care and experiments were performed in accordance with the Massachusetts General Hospital Institutional Animal Care and Use Committee (IACUC protocols 2020N000114, 2022N000102).

Primary human CAR T cell transduction and culture

Primary human T cells were cultured in RPMI 1640 medium with GlutaMAX and HEPES supplemented with 10% FBS, penicillin, streptomycin, and 20 IU/mL recombinant human IL-2. The T cells were activated on day 0 post-thaw using a 1:3 cell:bead ratio of CD3/CD28 Dynabeads and a starting cell density of 1e6/mL. On day 1, cells were transduced with lentivirus at a multiplicity of infection of 5. Starting on day 3, the media including recombinant IL-2 was doubled every 2 days. On day 7, a magnet was used to remove the CD3/CD28 beads. On Day 12–14, lentiviral transduction efficiency was determined by flow cytometry and the cells were either directly applied to functional assays or cryopreserved in 90% FBS and 10% DMSO.

Method Details

Lentiviral Production

HEK 293 cells were cultured in RPMI 1640 medium with GlutaMAX and HEPES supplemented with 10% fetal bovine serum (FBS), penicillin, and streptomycin. Lentiviral particles were produced in HEK 293 cells transduced with a mixture of, for each mL of HEK 293 cell culture media, 34 μL Optimem, 0.27 μg transfer plasmid, 0.45 μg VSV-G (Addgene plasmid # 14888, generated by Dr. Tannishtha Reya), 0.18 μg pSPAX2 (Addgene plasmid # 12260, generated by Dr. Didier Trono) and 3 μL of Fugene HD. For cell culture experiments, lentiviral particle-containing supernatant was collected 48 hours after transfection, passed through a 0.45 μM PVDF filter, and applied to target cells or stored at −80 C. For primary human T cell experiments, lentiviral particle-containing supernatant was concentrated by ultracentrifugation at 25,000 RPM in a SW28 rotor, resuspended in approximately 1/100 of the starting volume in RPMI or PBS, and cryopreserved at −80 C.

Vector Cloning

Entry vectors were linearized by restriction digest. Double stranded DNA fragments were generated by gene synthesis. DNA fragments were introduced by Gibson Assembly (New England Biolabs E2611L) or T4 ligation (Thermo Fisher Scientific EL0011) according to manufacturers’ specifications. Cloning products were transformed into STBL3 bacteria (Thermo Scientific, C737303). Vector sequences were confirmed by Sanger sequencing. Sequence elements are presented in Table S1.

Mass Spectrometry

Jurkat cells were lentivirally transduced to express either a GFP-targeted degrader or a control vector. Transduced cells were enriched by FACS using the mTagBFP2 transduction marker (Sony SH800) and expanded in cell culture. 17 million cells per sample were collected, washed twice in PBS, pelleted, and snap frozen. Subsequently, the cell pellets were lysed, reduced, and alkylated following the previously published method⁴⁴. Subsequently, 25 μg of the resulting peptides were labeled using TMTpro reagents from Thermo Scientific, as per the manufacturer’s instructions. The labeled samples were then combined and subjected to fractionation using high pH reversed-phase HPLC⁴⁴. The resulting fractions were analyzed using a 3-hour reversed-phase LC-MS2/MS3 run on an Orbitrap Fusion Lumos instrument. For quantification, MS3 isolation employed Simultaneous Precursor Selection (SPS), as described in earlier studies^45–47. Protein identification was based on MS2 spectra using the sequest algorithm, which searched against a human database (uniprot 2020) on an in-house-built platform⁴⁸. To ensure high confidence in protein identifications, a target-decoy database-based search strategy was utilized to filter against a false-discovery rate (FDR) of protein identifications of less than 1%⁴⁹. The proteome profiles were normalized across all samples in the run using an average normalization method. Briefly, all protein row intensities were scaled so that the average of each row was equal to the median of all row averages. Then the row scaled intensities were adjusted so that each sample column’s median scores was set to the average of all sample column median values. After normalization the data was converted into relative protein intensities by dividing each row by its median value and then applying a log2 transformation.

Differential Proteome Analyses

A total of 6900 proteins were quantified across all samples in the run. The triplicate relative intensities were compared by Pearson correlation in order to assess sample quality. Pairwise replicate correlations indicated that two samples were of poor quality (low correlation with other replicates) and removed from further analysis (Jurkat_Vector_Ctl_2, Jurkat_vhhGFP4SOCS2_1). Differential analysis between condition pairs was performed using the R package LIMMA⁵⁰ that generates a moderated T-statistic to determine statistical significance. Proteins with a Bejamini-Hochberg corrected p-value (FDR) less than 0.1 were identified as differentially expressed between sample conditions.

Gene Set Enrichment Analysis

Functional enrichment of differentially expressed proteins was performed for each comparison. The log2 fold change for all proteins were sorted and the protein names were converted into ENTREZ gene identifiers. The data was reformatted into an RNK file (file format described in the GSEA wiki, http://software.broadinstitute.org/cancer/software/gsea/wiki/index.php/Data_formats#RNK:_Ranked_list_file_format_.28.2A.rnk.29) and we used the command line GseaPreranked tool from gsea2–2.2.2⁵¹ to calculate enrichments of the GO Biological Process gene sets retrieved from MSigDB v2022 (we used default parameters except for the following: -nperm=10000, -set_max=4000, -set_min=1)⁵². Gene sets with an FDR<25% were identified as significantly differentially enriched. All statistical analyses were performed in R⁵³.

HEK Blue Reporter Assay

HEK Blue TGFβ cells (InvivoGen hkb-tgfbv2) were maintained in selection media according to manufacturer’s specifications. HEK Blue TGFβ cells were transduced with the SARA-SOCS2m-TagBFP2 lentivirus, and then purified by FACS (Sony SH800). In each well of a flat-bottom 96-well plate, 25,000 control of SMAD degrader HEK Blue TGFβ cells were plated in 100 μL of media with 10 ng/mL recombinant TGFβ1. Cell culture supernatant was collected 24 hours later, and secreted alkaline phosphatase (SEAP) activity was detected according to manufacturer’s instructions (InvivoGen rep-qbs) and quantified using a 96-well spectrophotometer (Biotek Synergy Neo2).

CAR T cell cytotoxicity assay

RPMI 1640 medium with GlutaMAX and HEPES supplemented with 10% FBS, penicillin, and streptomycin was used for all CAR T cell in vitro functional assays unless otherwise noted. Primary human CAR T cells were cocultured at defined cell ratios with 1e5/mL NALM6 target cells engineered to express GFP and luciferase for 16 hours in a 96-well round-bottom plate. To quantify the cellular luciferase, the plate was centrifuged, cell culture supernatant removed, and and 100 μL lysis buffer was added (Promega E1531). A 2:1 ratio of cellular lysate and luciferase assay (Promega N1110) was mixed and quantified for luciferase activity by microplate reader (Biotek Synergy Neo2). Percentage specific lysis was calculated from relative luminescence units (RLU) as follows: 100% × (target cells-only RLU − test RLU) / (target cells-only RLU).

Live cell analysis of CAR T cell and tumor cell line co-culture

On day 0, tissue culture plates were coated with anti-CD71 antibody diluted in PBS. On day 1, the plates were washed with PBS and then Nalm6-luciferase-GFP cells were added to the coated plates. Four hours later, defined cell ratios of primary human CAR T cells were added. Plates were imaged every hour for at least 90 hours (Incucyte ZOOM). Total green and red object areas were collected, indicative of GFP+ tumor and mCherry+ CAR T cell area, respectively.

Long term in vitro culture

Each week for four weeks, primary human CAR T cells were repetitively stimulated with irradiated K562 cells that had been engineered to express CD19 at a 1:1 ratio and total live cell density of 1 million cells per mL. The cultures were maintained by doubling media mid-week after stimulation and counted prior to weekly re-stimulation. Cells were serially collected from these cultures to perform intracellular flow cytometry on cells pre-stimulation (day 0) and one day post-stimulation (days 1, 8, 15, 22). Intracellular flow analysis was also attempted on day 29, but not reported due to insufficient cell number. Cells were stained with DAPI and antibodies to CD4 and CD8, washed, and then fixed and permeabilized according to manufacturer directions (BD Cytofix/Cytoperm 554714). Cells were then stained with TNFα and IFNɣ antibodies, washed, and analyzed by flow cytometry. Cells were also serially collected from these cultures to perform analysis of T cell phenotype and inhibitory receptor surface abundance on days 0, 7, 14, 21, and 28. Cells were stained with DAPI and antibodies to CD3, CD4, CD8, CCR7, CD45RA, PD1, TIM3, and LAG3.

CAR T cell production for transcriptional analysis

On day 8 of primary CAR T cell transduction and culture, mCherry+ cells were purified by fluorescence-activated cell sorting (FACS) (Sony SH800) and then returned to culture with media supplementation every 2 days to maintain a minimum density of 1e6 cells/mL. On day 12, cells were either untreated or treated with 10 ng/mL recombinant human TGFβ1 (Peprotech 100–21). 24 hours later, cells were washed once in PBS and collected for RNA purification according to manufacturer’s instructions (Qiagen RNeasy Mini Kit).

Nanostring nCounter gene expression analysis

25–100 ng of sample RNA (up to 5 μL) was mixed with the hybridization buffer (5 μL), Reporter CodeSet (3 μL)(CAR-T Panel Gene Expression Kit; NanoString Technologies, Seattle, WA), and RNAse-free water (up to 13 uL total volume). Capture CodeSet (2 μL) (CAR-T Panel Gene Expression Kit; NanoString Technologies, Seattle, WA) was then added to each tube immediately before placing the mixture at 65°c for 16 hours for the hybridization reaction. Reaction was ramped down to 4°c until analysis with the nCounter XT Assay (NanoString Technologies, Seattle, WA). Differential gene expression and pathway score analysis was performed using nSolver analysis software (NanoString Technologies, Seattle, WA); the expression of each gene was predicted with single linear regression and multiple hypothesis correction was performed with the Benjamini-Yekutieli procedure.

cDNA synthesis

cDNA were reverse transcribed from purified RNA with the SuperScript^™ IV VILO^™ Master Mix with ezDNAse enzyme (Invitrogen, Waltham, MA). Up to 2.5 ug total RNA were mixed with 10X ezDNAse buffer (1 μL), ezDNAse enzyme (1 μL), and nuclease-free water (to 10 μL) on ice, followed by digestion on 37°c for 2 minutes. After brief centrifugation, nuclease-free water (6 μL) and either SuperScript^™ IV VILO^™ Master Mix (4 μL) or SuperScript^™ IV VILO^™ No RT Control (4 μL) were added, depending on whether the reaction was RT or no RT control. The mixture was then incubated for reverse transcription at 25°c for 10 minutes, 50°c for 10 minutes, and 85°c for 5 minutes.

TaqMan gene expression analysis

The resulting cDNA were analyzed using Applied Biosystems TaqMan^™ gene expression analysis. 1–100 ng/μL of cDNA per reaction were mixed with the TaqMan^™ gene expression assay Master Mix (2.5 μL) (Applied Biosystems, Waltham, MA), target gene assays (SMAD3, GZMA, and ITGAM) conjugated to FAM (0.25 μL) (ThermoScientific, Waltham, MA), and the reference gene assay (GAPDH) conjugated to VIC (0.25 μL) (ThermoScientific, Waltham, MA). Four technical replicates were created for every treatment condition for each cDNA. Real-time PCR was performed on a Roche LightCycler 480 Instrument II (Roche Diagnostics GmBH, Basel, Switzerland). The data from each experiment were then normalized to the control to determine the relative fold change in gene expression. All quantitative calculations were performed using the 2−ΔΔCt method using GAPDH as the reference gene.

Murine xenograft model

NSG mice were injected with 1e6 Jeko-1 target cells engineered to express GFP and luciferase. Mice were treated at Day 7 with 5e5 CD19 CAR T cells or untransduced control T cells. Mice were imaged weekly for tumor bioluminescence emission using an AMI HT imaging system. For bioluminescent imaging, outliers attributed to missed luciferin injection were censored and were defined as the following: values 50x lower than the previous value and below 5e5. Blood was drawn on Days 14, 21, 35, and 46, subjected to red blood cell lysis (BD Biosciences 349202), and analysed for CAR T cell concentration using absolute concentration tubes (BD Biosciences 340334) and antibodies for Ly6G/D, Ter-119, NK1.1, murine CD11b, and human CD3. Mice were sacrificed after 46 days and spleen and bone marrow were harvested, homogenized and filtered to collect a single cell suspension, and analyzed by flow cytometry to enumerate CAR T cells.

Quantification and Statistical Analysis

Statistical tests were performed in GraphPad Prism 10 (GraphPad). Data are presented as means ± standard deviation unless otherwise indicated. The statistical tests used are described in each figure legend, and additionally in the Method Details section for proteomics and differential gene expression analysis. All tests were two-sided. Statistical tests include student’s t-test, Mann-Whitney, ANOVA, and mixed-effects model. The D’Agostino-Pearson omnibus test was used to assess normality. The ɑ level was 0.05 unless otherwise indicated.

Highlights.

• Bifunctional degrader proteins bridge an E3 ligase and endogenous target proteins

• Clinically suitable, human sequence-based transgenes amenable to cell therapy

• Optionally conditional degradation gated by small molecules or synthetic circuits

• A SMAD2/3 degrader enhanced CAR T cell anti-tumor potency