Week 1 HW: Final Project Ideas

Killer T cells surrounding a cancer cell.
Image source: Harvard Medical School

A Programmable SynNotch CAR-T Platform for Neuroblastoma

1.🪄 What I would like to biomanufacture, and Why ?

Neuroblastoma is a cancer of the developing nervous system that primarily affects children. Although treatment outcomes have improved, patients with high-risk, relapsed, or refractory disease continue to have limited effective therapeutic options.

CAR-T-cell therapy has shown encouraging activity in neuroblastoma, but its effectiveness is limited immune escape, limited or variable T-cell persistence, functional exhaustion within the tumor microenvironment, and the potential for on-target, off-tumor toxicity.

To address some of these limitations and to improve the specificity and functionality of CAR-T-cell therapy, I propose to biomanufacture a programmable, dual-antigen, logic-gated theranostic CAR-T-cell system that restricts full therapeutic activation to the tumor environment.

The engineered T cells would use a synthetic IF–THEN recognition circuit to increase tumor specificity while reducing constitutive CAR signaling providing autonomous control over T-cell activity.

The core of my proposed system would therefore use dual-antigen recognition rather than constitutive recognition of a single tumor antigen.

The engineered T cell would first recognize the neuroblastoma-associated antigen GD2 through a synthetic Notch (SynNotch) receptor.

GD2 recognition would act as a priming signal, releasing a synthetic transcription factor that induces expression of a CAR directed against a second neuroblastoma antigen, GPC2.

The resulting logic is:

Additionally, an inducible fluorescent reporter (such as GFP) generate a detectable signal upon GD2-SynNotch binding and an activation-dependent, membrane-tethered IL-15 module is added to enhance T-cell survival and persistence after GPC2 recognition allowing the engineered T cells to function as both therapeutic effectors and living biosensors.

The concept is therefore not simply to manufacture a CAR-T cell, but to manufacture a living therapeutic system capable of sensing its environment, making a decision, and executing an antitumor response.

2.🦠 What living system would I use?

The manufacturing chassis would be primary human T lymphocytes obtained from the patient, producing an autologous engineered T-cell product.

The inputs to the living system are therefore tumor-associated molecular signals, primarily GD2 and GPC2. The biological processes performed by the engineered T cell include antigen sensing, synthetic transcriptional regulation, CAR expression, signal transduction, cytotoxicity, cytokine-mediated support.

The final output is a population of engineered autologous T cells capable of conditional tumor recognition and regulated antitumor activity.

3.⚙️ How would I use automation?

I would integrate the engineered T-cell system with a closed, automated cell-processing platform such as the CliniMACS Prodigy. CAR-T-cell manufacturing involves multiple interconnected operations including T-cell enrichment, activation, genetic modification, expansion, sampling, washing, harvesting and final formulation. Performing these operations manually increases operator workload and introduces opportunities for variability and contamination.

Automation would also improve reproducibility and enable multiple patient-specific manufacturing batches to be processed in parallel.

During early-stage development, an Opentrons Flex platform could be used to automate high-throughput liquid handling, construct screening, and characterization of the engineered T-cell circuit.

Manual vs automated manufacturing of CAR-T Cells Image sources: Source 1 | Source 2

4.a ⚖️ How much would I need to manufacture ?

The scale I define is manufacturing enough therapy for the approximate annual worldwide population of patients with high-risk neuroblastoma.

A global analysis estimated approximately 5,560 new neuroblastoma cases in children aged 0–14 years in 2021. Approximately half of neuroblastoma cases are classified as high risk.

Therefore:

A recent phase 1/2 study of GD2-CART01 used a target dose of:

I will use 20 kg as an assumption.

For one 20-kg patient:

Therefore, the approximate annual number of infused CAR-positive cells for 2,780 patients would be:

However, because the proposed treatment is autologous, approximately 2,780 separate patient-specific manufacturing batches per year would be required.

4.b ⏳ How long would it take?

Published automated CAR-T manufacturing processes using the CliniMACS Prodigy typically require approximately 7–13 days, depending on the product and manufacturing process.

I will assume a 12-day manufacturing cycle. One instrument continuously occupied by 12-day manufacturing runs could theoretically complete:

Since real facilities cannot operate at 100% utilization (time for scheduling, setup, failures, maintenance, etc.), assuming 80% effective utilization:

For 2,780 patients:

So, under these assumptions, treating every newly diagnosed high-risk neuroblastoma patient would require approximately 100–120 parallel automated manufacturing units. The actual number could be higher because the proposed synthetic circuit is considerably more complex than a conventional single CAR.

SynNotch receptors, inducible CAR expression, cytokine modules and negative-feedback circuitry increase the genetic payload and may reduce genetic-modification efficiency or cell yield.

Thus, at a larger commercial scale, I would consider transferring the manufacturing process to a high-throughput closed platform such as the Cellares Cell Shuttle, which can handle up to 16 independent cartridges asynchronously, meaning different patient batches can run simultaneously.

CliniMACS Prodigy

🔗 Source: Source 1

Cellares Cell Shuttle

🔗 Source: Source 2

5.🧠❓ What do I not know yet?

1. ❓ whether the dual-antigen logic operates with appropriate kinetics and specificity in neuroblastoma

   I would need to determine how much GD2 is required to activate the SynNotch receptor, how rapidly GPC2 CAR appears after priming, how long CAR expression persists after the priming signal disappears, and whether heterogeneous or antigen-low tumor populations can escape the circuit.

2. 🧬 whether such a complex genetic program can be manufactured reproducibly without compromising T-cell fitness

   SynNotch-CAR constructs are already substantially larger than conventional CAR constructs. Adding cytokine, reporter and feedback modules may decrease vector production and genetic-modification efficiency or generate heterogeneous cell populations. The architecture may therefore eventually need to be simplified, divided between multiple genetic constructs, or implemented using an alternative genome-engineering strategy.

3. ⚖️ The balance between T-cell persistence, antitumor activity, and safety.

   Activation-dependent membrane-tethered IL-15 may enhance T-cell survival and persistence, but its expression level and duration would need to be optimized to avoid excessive stimulation, uncontrolled expansion, or toxicity.

Thus, the success of the project would ultimately depend not only on whether each individual module works, but also on whether the complete engineered cell behaves as a stable and predictable biological system.

Week 2 HW: DNA Design

Building the GD2-SynNotch–GPC2 CAR-T Genetic Circuit

This week, I translated the therapeutic concept proposed in Homework 1 into defined genetic parts and three buildable lentiviral constructs.

1. 🧬 Define the expression target

1.a What molecules am I expressing?

For my project, I will express two synthetic receptors (GD2-SynNotch and GPC2-CAR), one membrane-tethered cytokine (IL-15), and a fluorescent reporter (EGFP).

The current CAR response vector also retains constitutive tBFP expression as a marker of vector expression. 1.b Sequence sources GD2-SynNotch. The design is based on Moghimi et al. (2021), using the E101K-mutated anti-GD2 scFv. The remaining domains use the established Notch1 regulatory core and Gal4-VP64 architecture. Moghimi et al. (2021) | Addgene #79125 GPC2-CAR. I selected the humanized hCT3-1 anti-GPC2 scFv (VL–linker–VH), reported in patent US20250281538A1, SEQ ID NO: 12. It is incorporated into a second-generation CAR containing a CD8α hinge/transmembrane region, 4-1BB and CD3ζ. GPC2 antibody patent | Addgene #79123 Membrane-tethered IL-15. The cytokine module is based on Hurton et al. (2016), using human IL-15 connected to IL-15Rα through a flexible linker. Reference proteins: IL-15 NP_000576.1 and IL-15Rα NP_002180.1. Hurton et al. (2016) EGFP. Enhanced green fluorescent protein derived from Aequorea victoria, NCBI GenBank U55763.1. 1.c Host / chassis The chassis is primary human T lymphocytes, including CD4+ and CD8+ T cells. For the proposed autologous therapy, these cells would be obtained from the patient and genetically modified ex vivo. 1.d Where do the products go?

2. 🧩 Design the genetic parts I organized the system into three functional constructs, each with a different regulatory role.
   Construct 1 — Constitutive GD2 sensor

   Expression: Constitutive Architecture: PGK promoter → 5′ UTR → Kozak → CD8α signal peptide → GD2 scFv (14G2a-E101K) → Notch1 regulatory core → Gal4-VP64 → Stop The constitutive PGK promoter maintains expression of the GD2 sensor. After GD2 recognition, the SynNotch regulatory core releases Gal4-VP64, which activates the downstream Gal4-responsive cassette. Reference backbone: Addgene #79125, modified by replacing the original anti-CD19 scFv with the selected anti-GD2 scFv.

   Construct 2 — SynNotch-inducible effector

   Expression: Inducible by GD2-SynNotch activation Architecture: Gal4 UAS → minimal CMV → GPC2-CAR → IRES → EGFP → poly(A) CAR: CD8α signal peptide → GPC2 scFv → CD8α hinge → CD8 TM → 4-1BB → CD3ζ Gal4-VP64 released by activated SynNotch binds the Gal4 UAS and induces GPC2-CAR expression. EGFP is translated from the same transcript through an IRES and reports activation of the response cassette. The current vector also contains a separate constitutive PGK → tBFP unit. Reference backbone: Addgene #79123.

   Construct 3 — Activation-dependent cytokine support

   Expression: Activation-inducible through NFAT signaling Architecture: 3× NFAT-responsive elements → minimal IL-2 promoter + Kozak → IgE signal peptide → human IL-15 → flexible linker → IL-15Rα membrane-anchoring component → Stop Following T-cell activation, NFAT can activate transcription of the mbIL-15 cassette. Membrane tethering is intended to provide localized cytokine support rather than constitutive soluble cytokine secretion. Reference backbone: pRRLSIN.cPPT.PGK-GFP.WPRE — Addgene #12252, with the original expression cassette replaced by the NFAT-inducible mbIL-15 cassette.

   2.d Genetic-part summary Part type Selected part Source / role Constitutive promoter PGK Addgene #79125; constitutive SynNotch expression Inducible regulatory element Gal4 UAS + minimal CMV Addgene #79123; activated by Gal4-VP64 Activation-responsive promoter 3× NFAT + minimal IL-2 promoter Activation-dependent mbIL-15 expression Translation initiation 5′ UTR + Kozak Translation initiation in human cells Antigen-binding domains GD2 scFv; GPC2 hCT3-1 scFv Tumor-antigen recognition Signaling domains Notch core; 4-1BB; CD3ζ Synthetic transcriptional and CAR signaling Targeting / membrane domains CD8α signal peptide, hinge/TM; IL-15Rα anchor Membrane localization Translational control IRES EGFP expression from the inducible CAR transcript Reporter EGFP; tBFP Inducible response reporter; constitutive vector marker Lentiviral context WPRE / LTRs / backbone elements Retained from the selected vectors

Notch1 regulatory core

NFAT responsive elements  (3X , 2 Forward 1 reverse) 

Stop codon → 3′ UTR → Poly(A) signal

Construct 1 — GD2-SynNotch

Kernel schematic of the GD2-SynNotch construct.

Construct 2 — GPC2-CAR–IRES–EGFP

Kernel schematic of the Gal4-inducible GPC2-CAR–IRES–EGFP construct.

Construct 3 — NFAT-inducible mbIL-15

Kernel schematic of the NFAT-inducible membrane-bound IL-15 construct.

4. 🔎 Optimize and quality-check 4.a Codon optimization I codon-optimized the protein-coding regions of the three constructs for expression in Homo sapiens. Regulatory regions such as the PGK promoter, Gal4 UAS, NFAT-responsive elements, minimal promoters, IRES, WPRE and LTRs were left unchanged because their nucleotide sequences are functionally important. The optimization changed synonymous codon usage while preserving the intended protein products. The optimized CDS versions were retained as the proposed sequences for synthesis.

Codon optimization of the coding regions for human-cell expression.

4.b Sequence verification by BLAST I used BLASTp to verify the identity and architecture of the encoded proteins.

The complete CAR and mbIL-15 sequences are engineered fusion proteins, so they are not expected to be identical to one natural protein.

Representative BLASTp verification results.

4.c Biosecurity screening I ran Kernel’s DNA biosecurity screen on each complete final construct. All three constructs passed without a reported biosecurity flag. The screen evaluates submitted DNA sequences for matches associated with sequences of biosecurity concern. A clear result supports the required sequence-level security check, but it does not assess the biological or clinical safety of the engineered T cells themselves.

Kernel DNA biosecurity screening of the final constructs.

5. 🧪 From design to build 5.a How would I build it? I would reuse the three lentiviral backbones and synthesize only the new or modified DNA modules rather than complete plasmids. Each backbone would be PCR-linearized while excluding the sequence being replaced. The new fragment would be synthesized with homologous terminal overlaps and joined to the backbone by Gibson/HiFi assembly.

: GD2-SynNotch: replace the original anti-CD19 scFv with the GD2 scFv.

GPC2-CAR: replace the inducible payload with the GPC2-CAR–IRES–EGFP cassette.

mbIL-15: replace the original pRRLSIN expression cassette with the NFAT/minimal IL-2–mbIL-15 cassette. Gibson assembly is suitable because it enables seamless joining of overlapping DNA fragments without depending on restriction sites (Gibson et al., 2009).

Proposed assembly strategy for the three lentiviral constructs.

5.b What does synthesis constrain? The main constraints are sequence length, repeated regions and synthesis errors. My design contains repeated regulatory elements, particularly the NFAT sites and Gal4 UAS, as well as repetitive linker sequences. These regions may be harder to synthesize or assemble accurately. If a complete cassette is difficult to synthesize, I would divide it into smaller fragments and assemble them together with the backbone. DNA synthesis is not inherently error-free, so the assembled plasmids would be sequence-verified before use. Error-correction approaches can reduce mutations in de novo synthesized DNA (Carr et al., 2004).

1. ❓ Will the GD2-SynNotch gate be sufficiently specific and tightly controlled? The main uncertainty is whether GD2 recognition will induce enough Gal4-VP64 activity to express the GPC2-CAR while maintaining low background expression in the absence of GD2. I would compare EGFP/CAR expression after exposure to GD2-negative cells and cells expressing different GD2 levels. Synthetic transcriptional circuits can show behaviors that are difficult to predict from individual components alone, so this will need empirical tuning rather than sequence design alone (Elowitz & Leibler, 2000).
2. 🧬 Will the three constructs function efficiently together in the same T cell? Because the system uses separate lentiviral constructs, not every T cell may receive and express all three modules at appropriate levels. I would measure expression of each module and compare different vector-delivery conditions, then test whether cells containing the complete circuit respond differently from cells missing one component.
3. ⚖️ Will NFAT-inducible mbIL-15 remain sufficiently activation-dependent? NFAT can also be activated through endogenous T-cell signaling, so mbIL-15 may not be exclusively triggered by the GPC2-CAR. I would compare mbIL-15 expression in resting cells, after CAR stimulation and after independent TCR stimulation, together with proliferation and activation/exhaustion markers. More generally, maintaining functional separation between synthetic regulatory modules is an important consideration in engineered logic circuits (Stanton et al., 2014). The main challenge is therefore not simply whether each individual genetic module works, but whether all three modules behave together as a predictable and controllable T-cell circuit.

Final Project