C1: Multi-target coordinated intervention

Logic-gated cell therapies are engineered cells — most maturely T cells, increasingly other immune cells and stem cells — that activate only when specific combinations of input conditions are met, rather than firing against any cell displaying a single target.

The molecular implementations:

AND gates require two antigens to be present simultaneously before the cell activates. The leading implementation is the synthetic Notch (synNotch) receptor: binding of the first antigen cleaves the synNotch receptor, releasing a transcription factor that then drives expression of a conventional CAR targeting the second antigen. The cell only becomes cytotoxic where both antigens co-occur — useful for distinguishing a tumour (which may express an unusual combination of common antigens) from healthy tissue (which expresses each antigen individually elsewhere).

NOT gates use inhibitory CARs (iCARs): an activating receptor against a tumour-associated antigen is paired with an inhibitory receptor against a "safe self" antigen present on healthy tissue but absent on tumour. The cell attacks targets bearing the tumour antigen unless the safe-self antigen is also present, sparing healthy tissue.

Inducible (ON/OFF) systems place CAR expression under the control of a small-molecule switch, so a clinician can turn therapy activity up or down by administering or withdrawing a drug — a control modality directly relevant to AIHS platform orchestration.

Current state of the art: two-input logic gates are in clinical trials; three-input gates are in preclinical research. The complexity ceiling is set by how many inputs can be reliably integrated in a single cell without disrupting normal cellular function or creating unintended activation conditions. AIHS Phase II should see four-input gates demonstrated with confidence; higher complexity remains research-grade. Companies including Senti Bio and Asher Bio, plus numerous academic groups, are advancing this rapidly.

Single-payload delivery — getting one drug or one gene therapy to one cell type — is now mature for many use cases. Multi-payload delivery requires several things to align simultaneously, and each adds a failure mode.

Vehicle coexistence. The delivery vehicles must circulate together without aggregating or interfering. Lipid nanoparticles with different surface chemistries can interact; viral vectors can compete for the same cellular entry receptors. Co-formulation requires that each vehicle remain stable and functional in the presence of the others.

Orthogonal targeting. The targeting moieties must address different cell types without cross-reaction. If payload A is meant for hepatocytes and payload B for T cells, the targeting ligands must be specific enough that A does not end up in T cells or vice versa — and the available repertoire of truly orthogonal targeting ligands is limited.

Payload non-interference. The payloads must function in their target cells without cross-talk. Two gene therapies sharing transcriptional machinery, two small molecules competing for the same metabolic enzyme, or two engineered circuits using the same synthetic transcription factor will interfere.

Pharmacokinetic coordination. Different payloads have different absorption, distribution, metabolism, and clearance profiles. Getting them all to effective concentration at their respective targets simultaneously — when one clears in hours and another persists for weeks — requires careful dosing design.

Current approaches include sequential dosing (one payload at a time, simplest but slowest and not truly coordinated), combination nanoparticles (multiple cargoes in one vehicle, limited by cargo compatibility), and engineered cells that themselves deliver multiple coordinated effects (most flexible, least mature). Each has tradeoffs; none yet achieves the seamless multi-target coordination AIHS requires for routine clinical use.

A single therapy is a perturbation to a biological system. Two therapies are two perturbations whose joint effect is not, in general, the sum of their individual effects — because the body's regulatory loops respond to the combined disturbance differently than to either alone.

The classic illustration is drug-drug interaction. Combining two drugs that work through different mechanisms can produce unexpectedly large effects (synergy), unexpectedly small effects (antagonism), or entirely new effects (a toxicity neither drug produces alone). Some of these interactions are pharmacokinetic — one drug alters the metabolism or clearance of another, as when a drug inhibits the CYP450 enzymes that process a second drug, raising its concentration to toxic levels. Others are pharmacodynamic — the drugs act on the same downstream pathway in ways that compound or cancel.

For AIHS coordinating dozens of simultaneous interventions, the combinatorial interaction space is enormous. With n possible interventions, there are on the order of n² pairwise interactions and vastly more higher-order combinations — far too many to characterise exhaustively through experiment. This means the system cannot rely on having tested every combination it might deploy.

The fundamental tool for predicting these interactions in advance is causal modelling (B3). Without robust causal models of the biological systems involved, multi-target coordination becomes either conservative (only combinations with direct empirical validation are used, sharply limiting capability) or risky (untested combinations are deployed on the assumption that individually-safe components remain safe together, increasing adverse events).

This is why C1 progress is strongly gated by B3 progress, despite C1 being technically simpler than B3 in isolation. The delivery and control engineering for multi-target intervention can mature on a 10-20 year timeline, but the ability to deploy that capability safely at scale depends on the causal-modelling capability that lets the system predict interactions it has never directly observed. Safe deployment also depends on the bounded-reversibility (C5) and safety-architecture (C6) capabilities that let the system halt or reverse a coordinated intervention that turns out to interact badly.