Date: 1st December 2020
Chimeric antigen receptor (CAR) T cells are T cells that have been genetically engineered to produce an artificial T-cell receptor for use in immunotherapy, and are a powerful way to treat cancer. Whilst, their use for blood cancer therapies are now well established – as a single specific antigen can be targeted, their extended use in solid cancers types has been hindered – as these types of tumours lack this specificity. Now, scientists have combined artificial intelligence (AI) with cell engineering, to design ‘living’ medicines that exploit discriminatory combinatorial antigen targets that can precisely target tumours.
Precise cellular recognition is a key challenge in treating diseases such as cancer, and current therapeutic approaches generally rely on identifying malignant cells by interacting with a single cancer-associated antigen. Furthermore, in many cases these therapies do not just target the cells of interest, they also recognise healthy cells that also express the single antigen.
Now, scientists from the University California San Francisco and Princeton University, US, led by Wendell Lim have used a machine learning approach to assemble a catalogue of antigen combinations that can precisely target cancerous cells whilst leaving normal cells alone. The team then used this information to programme T cells via a customisable molecular sensor – allowing T cells to recognise up to three antigens expressed on or inside cancer cells and integrated these inputs to achieve NOT, AND and OR logic.
The team had previously performed a comprehensive in silico screen and used machine learning techniques to identify multi-antigen signatures that would improve tumour discrimination by CAR T cells engineered to integrate multiple antigen inputs via Boolean logic, e.g., AND and NOT.
Now, using this AI-data, the team now turned to engineering T cells using synthetic Notch (SynNotch) receptors to transcriptionally interconnect multiple molecular recognition events. The SynNotch is a receptor that can be engineered to recognise a range of target antigens and can act as a regulatory layer by inducing antigen-triggered expression of synthetic transcription factors (here GAL4 or LEXA). These output responses were also programmed, so that they controlled the cell function once an antigen was recognised. This allowed the system to be programmed with both positive and negative logic, allowing precise recognition of a wide variety of cells.
For example, a synNotch receptor could be engineered so that upon the recognition of antigen A, the cell made a second synNotch receptor that recognised antigen B, this then induced the expression of a CAR (chimeric antigen receptor) that recognised antigen C. The output of this circuit was a T cell that would only kill cells which expressed all 3 antigens. In contrast, a T cell could be programmed with a NOT function, here if it was to encounter an antigen present in normal tissues but not in the cancerous ones, it could be programmed to cause the T cell carrying it to die, sparing the normal cells from attack and possible toxic effects.
Proof-of-concept
The team built several logic gates circuits for AND, NOT and OR and successfully tested them in T cells in vitro against melanoma and breast cancer cells. However, as a proof-of-concept they wanted to test a three-input AND gate in-series circuit in vivo to determine whether T cells bearing this circuit could selectively kill tumours in mouse tumour models.
To address this, they constructed a myelogenous leukemia cell line expressing either three antigens, or one containing just two of those antigens and implanted them into in a bilateral tumour model in immune compromised mice. The three-input AND gate T cells were then administered by tail vein injection and allowed to autonomously explore and act on both tumours. The T cells showed precise recognition and rapidly cleared the three-antigen tumours however, the two-antigen tumours implanted into the other flank of the mouse remained untouched by the T cells.
Conclusions and future applications
The team here have developed smart cells that harness the computational sophistication of biology and have real impact on fighting cancer. They show that sensor cells – such as a therapeutic T cell – can be flexibly engineered to recognise combinations of intra- and extra-cellular antigens and can integrate this information using a range of logic gates. This allowed precise targeting of these cells to specific cell types such as cancer.
The components of the system, such as SynNotch, work in a modular plug-and-play manner, allowing the system to be highly flexible and is therefore, compatible with new target antigens. Furthermore, the output function on antigen recognition is also one that can be manipulated and can be designed to achieve a variety of T cell outcomes.
In fact, Lim’s group is now exploring how these circuits could be used in CAR T cells to treat glioblastoma (GBM), an aggressive form of brain cancer that is nearly always fatal with conventional therapies. We have recently seen several advancements for GBM using novel therapeutics such as synthetic protein nanoparticle (SPNP) equipped with a cell-penetrating, tumour-targeting peptides, and antibody-cytokine protein fusions called immunocytokines, both of which resulted in GBM tumour regression. It will be interesting to see how these SynNotch T cells compare.
From a wider perspective it is likely that this technology could translate into treatments for other disease types, for example autoimmune disease or graft-versus-host disease, in fact any disease that requires targeting specific or dysfunctional cells. Overall, this technology will accelerate specific targeting of therapeutics, reducing toxicity and off-target effects and improving patient outcomes.
For more information please see the press release from UCSF
Atasheva, S., C. C. Emerson, J. Yao, C. Young, P. L. Stewart and D. M. Shayakhmetov (2020). “Systemic cancer therapy with engineered adenovirus that evades innate immunity.” Science Translational Medicine 12(571): eabc6659.
https://doi.org/10.1126/science.abc6270
