VE-cadherin NOT-gated CD93 CAR T cells discriminate between AML and healthy endothelial cells

This study demonstrates that engineering CD93-targeting CAR T cells with a VE-cadherin-based inhibitory NOT-gate effectively preserves healthy endothelial integrity while maintaining potent anti-leukemic activity against acute myeloid leukemia, thereby overcoming a major barrier to clinical translation.

The Gist

The Big Problem: The "Friendly Fire" Dilemma

Imagine you are a general trying to defeat an enemy army (Acute Myeloid Leukemia, or AML) that is hiding inside a city. You have a special weapon: CAR T-cells. These are super-soldiers you engineer to recognize a specific flag (a protein called CD93) that only the enemy soldiers are wearing.

The Catch: The enemy soldiers are wearing that flag, but unfortunately, so are the innocent civilians (healthy blood vessel cells or endothelial cells) living in the same city.

If you send your super-soldiers in, they will see the flag and attack. They will kill the enemy, but they will also accidentally kill the innocent civilians. In medical terms, this is called "On-target, off-tumor" toxicity. It's like a smart missile that hits the bad guy but also destroys the house he's standing in. This has been a major roadblock in treating AML with CAR T-cells.

The Solution: The "Smart Gate" (NOT-Gating)

The researchers in this paper came up with a clever solution. Instead of just giving the soldiers a "Go" signal when they see the enemy flag, they gave them a "Stop" signal if they see a different flag that only the innocent civilians wear.

They chose VE-cadherin as the "Stop" flag. This protein is found on healthy blood vessel cells but not on the leukemia cells.

They built a "Smart Gate" system (called a NOT-gated CAR) that works like this logic puzzle:

• IF you see the Enemy Flag (CD93) AND
• IF you do NOT see the Innocent Flag (VE-cadherin)...
• THEN Attack!

But if the soldier sees both flags (meaning they are near a healthy blood vessel), the "Stop" signal overrides the "Attack" signal, and the soldier stands down.

How They Tested It

The researchers didn't just guess; they built these smart soldiers and put them through rigorous training:

1. Finding the Right "Stop" Flag: They looked at thousands of genetic maps to find a protein that stays on healthy blood vessels even when the body is inflamed (which happens during cancer treatment). They found VE-cadherin was the perfect candidate because it's stable and specific to healthy vessels.
2. The Training Ground (2D): First, they tested the soldiers in a simple petri dish. When they faced leukemia cells, the soldiers attacked fiercely. When they faced healthy blood vessel cells, the soldiers ignored them completely.
3. The Realistic Simulation (3D): Petri dishes are flat, but blood vessels are tubes. The researchers used a high-tech, 3D micro-chip system (called LumeNEXT) that mimics real blood vessels. Even in this complex, tube-like environment, the smart soldiers killed the leukemia but left the blood vessel walls intact.
4. The "Switch" Test: They wanted to make sure the soldiers didn't get confused if they fought a battle and then had to protect a civilian. They found that the soldiers could switch from "Attack Mode" to "Protect Mode" instantly. If they killed some leukemia cells first, they could still immediately stop and protect a blood vessel they encountered next. The system is reversible and flexible.

Why This Matters

This research is a game-changer for two reasons:

1. Safety: It proves we can target dangerous cancers like AML without destroying the patient's blood vessels, which would be fatal.
2. The Blueprint: The method they used to find the "Stop" flag (using big data to compare healthy cells vs. cancer cells) can be used for other cancers. If a cancer target is also found on healthy tissue, this "Smart Gate" strategy can be applied to make the treatment safe.

The Bottom Line

Think of this like giving your security guards a smart badge.

• Old guards: "If you see a red hat, shoot." (Result: They shoot the bad guys and the mailman who also wears a red hat).
• New guards: "If you see a red hat, check for a blue badge. If the blue badge is missing, shoot. If the blue badge is present, stand down."

This paper shows that this new strategy works perfectly. It allows the immune system to hunt down leukemia with precision, sparing the healthy blood vessels and giving patients a real chance at a cure without the deadly side effects.

Technical Summary

1. Problem Statement

The translation of Chimeric Antigen Receptor (CAR) T-cell therapy from B-cell malignancies to Acute Myeloid Leukemia (AML) is hindered by on-target, off-tumor (OTOT) toxicity.

• The Target: CD93 is a promising antigen for AML therapy as it is highly expressed on leukemic stem cells (LSCs) and blasts but absent on healthy hematopoietic stem and progenitor cells (HSPCs).
• The Obstacle: CD93 is also expressed on healthy endothelial cells (ECs). Targeting CD93 directly risks severe vascular damage, a potentially fatal side effect.
• The Challenge: Existing strategies to mitigate OTOT toxicity often narrow the therapeutic window or fail to account for the dynamic, pro-inflammatory microenvironment created during CAR T-cell therapy, which can alter antigen expression on healthy tissues.

2. Methodology

The authors employed a multi-step approach combining bioinformatics, protein engineering, and advanced physiological modeling:

• Bioinformatics-Driven Target Selection:

  • Analyzed bulk RNA-seq data from EC lines (iHUVEC, TIME) and AML cell lines (Kasumi-1, NOMO-1, THP-1) under resting and pro-inflammatory conditions (IFN$\gamma$, TNF$\alpha$).
  • Screened the Tabula Sapiens single-cell atlas and the TARGET-AML patient database (3,225 samples) to identify EC-specific surface markers that remain stable during inflammation and are absent in AML.
  • Selection: VE-cadherin (VC/CDH5) and VEGFR2 were identified as top candidates. VC was prioritized because its protein expression remained stable on ECs under inflammatory stress, whereas VEGFR2 expression decreased.

• CAR Engineering (NOT-Gating Strategy):

  • Activating CAR (aCAR): Engineered a CD93-specific CAR (CD93.28$\zeta$) to target AML.
  • Inhibitory CAR (iCAR): Engineered a novel VE-cadherin-specific single-chain variable fragment (scFv) derived from the 2E11d antibody. This scFv was fused to the intracellular inhibitory domain of PD-1 (VC.PD1) to create an iCAR.
  • Logic Gate: Created a bicistronic retroviral vector co-expressing the VC.PD1 iCAR and the CD93.28$\zeta$ aCAR. The logic is: Kill if CD93+ AND VC-. If VC is present (healthy EC), the PD-1 signal inhibits activation.
  • Controls: Generated a "Pdel" control (VC.Pdel/CD93.28$\zeta$) lacking the PD-1 inhibitory domain to confirm that inhibition was specific to the PD-1 domain.

• Experimental Models:

  • In Vitro Cytotoxicity: 2D co-cultures of CAR T cells with AML (THP-1, OCI-AML3) and EC (TIME, iHUVEC) lines. Measured cytokine release (IFN$\gamma$, GzmB) and target cell lysis.
  • 3D Microphysiological System: Utilized LumeNEXT devices (PDMS microfluidic chips) to create 3D vascularized lumens lined with HUVECs. This model mimics native vessel geometry and polarity, allowing for the assessment of EC integrity in a more physiologically relevant context.
  • Sequential Exposure Assays: Tested the reversibility of the NOT-gate by exposing CAR T cells to AML then ECs, or ECs then AML, to simulate dynamic tumor microenvironments.

3. Key Contributions

• Identification of VC as an Optimal iCAR Ligand: Demonstrated that VE-cadherin is a superior iCAR target compared to previous surrogates (like CD19) due to its stable expression on ECs even under the pro-inflammatory conditions induced by CAR T-cell therapy.
• Novel scFv Development: Successfully engineered and validated a new VE-cadherin-specific scFv (in the Heavy-Light orientation) that functions effectively within a CAR structure.
• Reversible Logic Gating: Proved that the NOT-gated system is dynamic; the inhibitory signal does not permanently exhaust the T cells, allowing them to switch between "kill" (AML) and "spare" (EC) modes based on immediate target engagement.
• Validation in 3D Models: Moved beyond standard 2D assays to validate EC sparing in a 3D vascularized microphysiological system, a critical step for predicting clinical safety.

4. Key Results

• Specificity Validation:
  • VC expression was confirmed to be exclusive to ECs and stable under inflammatory cytokine exposure, while CD93 was confirmed on AML.
  • VC-specific aCARs (VC.28$\zeta$) successfully killed ECs but not AML, confirming the specificity of the scFv.
• NOT-Gated Functionality:
  • Cytokine Production: VC.PD1/CD93.28$\zeta$ CAR T cells secreted high levels of IFN$\gamma$ and GzmB against AML targets but showed near-complete abrogation of cytokine release when exposed to ECs.
  • Cytotoxicity: In 2D assays, the NOT-gated cells killed AML cells effectively while sparing ECs. The control (Pdel) cells killed both.
  • 3D Vascular Protection: In LumeNEXT devices, VC.PD1/CD93.28$\zeta$ cells preserved endothelial lumen coverage and viability, whereas control cells caused significant EC death and lumen collapse.
• Reversibility and Durability:
  • Prior Activation: CAR T cells that had previously killed AML cells retained their ability to be inhibited by VC on ECs (no "exhaustion" of the inhibitory gate).
  • Prior Inhibition: CAR T cells that had been inhibited by ECs retained the ability to activate and kill AML cells upon subsequent exposure.
  • Phenotype: The presence of the iCAR did not induce baseline T-cell exhaustion (no significant increase in PD-1, TIM-3, or LAG-3 expression) compared to controls.

5. Significance

• Clinical Translation for AML: This study provides a viable pathway to utilize CD93, a potent AML target, by solving the critical safety hurdle of vascular toxicity.
• Generalizable Strategy: The bioinformatics-guided approach to selecting iCAR ligands based on inflammatory stability and spatial proximity (noting that VC and CD93 are physically close in EC junctions) offers a blueprint for developing safe CAR therapies for other malignancies where OTOT toxicity is a barrier.
• Mechanistic Insight: The findings challenge the notion that strong activation leads to irreversible inhibition; instead, they demonstrate that CAR T cells can dynamically toggle between activation and inhibition states, a crucial feature for safety in complex tumor microenvironments.

In conclusion, the authors successfully engineered a "smart" CAR T cell that discriminates between malignant and healthy endothelial cells using a VE-cadherin NOT-gate, preserving vascular integrity while maintaining potent anti-leukemic activity.