Learning language cells use to communicate with each other

Cells are the building blocks of life. The way cells recognize other cells and external cues can lead to different biological fates, including cell growth, death, and locomotion. Researchers seek to understand cellular communication, reverse engineer it, and ultimately sculpt cellular interactions that transcend natural capabilities. Although cell therapies already exist, it is likely that the future of this cell therapy will involve deeper modification of a patient’s own cells to treat a range of diseases and repair tissues.

in previous article, we reviewed a study that standardly substituted the extracellular part of a protein to identify different ligands; This “reassembled” protein switches the same signaling pathway as long as the transmembrane and intracellular compartments remain intact. Here, we discuss paper which emphasizes the intracellular part of the cell instead. Researchers from the University of California, San Francisco have theoretically reconfigured the signaling domains of CAR T cells and explored possible effects on cell-cell communication.

Making a chimeric antigen receptor

cHemerick antigen RThe CAR receptors require genetic modification to express novel prosthetic components. Figure 1 shows the three major regions of a CAR T cell: antigen-binding domain, transmembrane domain, and signaling domain. Scientists often focus on and tailor a binding domain to a specific therapeutic target (eg: proteins found in cancer cells). However, here the researchers focus on signaling domain formation and its impact on CAR T cell functioning.

cost particles

The signaling domain of a CAR T cell normally contains the CD3ζ T cell receptor (TCR) molecule and any combination of costimulators. Costemic molecules contain multiple signaling isoforms, or short peptides that bind to specific downstream signaling molecules. These molecules affect T-cell signal transduction with different effects. Two examples include 1-4BB, which can increase T-cell memory and persistence, and CD28, which is associated with effective T-cell killing but reduces T-cell persistence.

Expanding possibilities through machine learning

Researchers in Wendell Lim’s lab sought to find unspoken rules that govern cost signaling and thus improve CAR T-cell properties. They used a library of synthetic signaling models, machine learning, and a unique conceptual approach to discover combinations beyond what naturally occurs.

From words to sentences to language

The researchers looked at natural signaling proteins, pulled signaling motifs from them, and assembled synthetic combinations of signaling motifs to form unique signaling programs. This approach can be conceptualized as an exploration of syntax.

Figure 2 shows the rearrangement of different ‘words’ – the signaling models – into distinct ‘sentences’ or signaling programmes. To understand and predict the “language” of these clusters, the team then used machine learning algorithms called neural networks to discover the underlying “rules” of the datasets. This revealed the importance of word order, word meaning, and word combinations in the final product—reframed as the effect of signal motif identity, function, and order on T-cell phenotype.

The team curated a library of anti-CD19 CAR T cells with diverse cost domains. Each cell contained either one, two, or three signaling isoforms taken from normal signaling proteins (see Figure 2). The team randomly entered 12 shapes of the original signal along with one spacer shape in the positions IJ And K to produce a total of 2,379 distinct configurations, as shown in Figure 3.

Next, the researchers screened randomized subsets of the library to rate the cytotoxicity of T cells and their ability to proliferate (stem). This process formed unique and unusual combinations, including those similar to the cost molecule 4-1BB (ex: M10-M1-M1-).

Decode ‘language’ using predictive neural networks

The signal isoform sequences possess varying levels of cytotoxicity and stemness, according to experimental analysis. Next, the team leveraged this data to understand the unseen rules surrounding cost molecule design.

An artificial neural network has proven to be crucial in this investigation. As illustrated in Figure 4, the data were segmented to either train or test the algorithm to predict the cytotoxicity or stemness of a chimeric antigen receptor. This process explained several correlations, such as the ability of 4-1BB-like costimulatory domains to promote cytotoxicity and stemness.

Successful prediction using the M1 cost molecule

Can a neural network accurately predict T-cell fate with a given cost combination? The team tested the waters by adding the cost molecule M1 to the 4-1BB-like signaling domains against the CD28 signaling domains. The neural network predicted that addition of M1 isoforms would show enhanced cytotoxicity and stemness in 4-1BB-like domains while having no effect in the CD28 isotype.

In an in vitro model, CAR T cells with 4-1BB-like domains and M1 motifs effectively killed tumor cells and preserved T cell stemness; On the other hand, addition of the M1 isoforms did not cause any change to the CD28-like derivatives. This correct prediction translates into the results of the mouse model as well. 4-1BB/M1 CAR T cells delayed tumor cell growth by 2 weeks longer than 4-1BB-only CAR T cells. These observations demonstrate how a neural network can be used to accurately predict T-cell characteristics depending on the forms of synthetic signaling involved.

CAR T treatment possibilities

It can be difficult to predict how a synthetic receptor component will affect the resulting cell properties. This study unpacks part of this puzzle with signaling and machine learning model libraries. By analyzing CAR T cell cost domain populations, the team created a neural network that predicts T cell phenotype success based on current cost molecules. This, in turn, revealed cost-effective CAR T signaling rules that can be used to design better synthetic signaling domains. Similar libraries and subsequent analyzes can be applied to improve other CAR T cell modulatory regions.