When Disease Becomes Part of the Design

In 2012, researchers studying BRAF V600E encountered a result that exposed a limit of molecular precision. In melanoma, inhibiting the altered BRAF protein produced substantial responses. In colorectal cancers carrying the same mutation, the response was far weaker. The molecular lesion and intervention were the same. What differed was the regulatory system surrounding them.

In BRAF V600E colorectal-cancer cells, BRAF inhibition rapidly activated signalling through the epidermal growth factor receptor, or EGFR. That feedback supported continued downstream signalling and proliferation. Melanoma cells expressed much less EGFR and were therefore less able to produce the same response. The same molecular lesion, acted on by the same intervention, produced a different outcome because the biological context was different.^[1]

Studies of mTOR complex 1, or mTORC1, revealed a related problem. In the cancer contexts examined, mTORC1 signalling participated in feedback that restrained upstream receptor and PI3K activity. Rapalog or mTORC1 inhibition relieved parts of that restraint and increased signalling through AKT. Other work found activation of the MAPK pathway through an S6K–PI3K–RAS feedback route after mTORC1 inhibition.^[2,3]

These findings did not show that direct inhibition was misguided. They showed that inhibiting an intended target and controlling the resulting biological state are not necessarily the same achievement. An intervention acts on its target within a network that can absorb, redirect or oppose it.

Drug discovery has become increasingly capable of identifying molecular targets, resolving their structures and measuring their engagement. Those capabilities remain essential. They do not, by themselves, establish what physiological state will follow. That requires a broader question: what information is contained in the biological context where the drug acts?

Disease-associated activity is usually described by the harm it produces. Inflammation persists. Cells proliferate without appropriate restraint. Glucose rises beyond a healthy range. A signalling pathway remains active when it should recede.

That description is necessary, but biological signals can also carry information through amplitude, duration, frequency and location.^[4] A concentration may reflect intensity. Timing may indicate when a system has entered a particular state. Relationships among signals may help distinguish an isolated molecular event from a coordinated physiological response.

This does not make every disease-associated signal a useful therapeutic input. A signal may merely accompany pathology rather than cause it. It may occur in healthy tissue as well as diseased tissue. Its relationship to severity may differ between people or change as disease progresses. In some settings, suppressing it as completely as possible is the appropriate objective.

The design question is narrower. Where a biological condition reliably reflects therapeutic need, can that condition help govern where, when or how strongly an intervention acts?

Pharmacological effects are already shaped by context. Target abundance, tissue exposure, metabolism and pathway state all influence them. The distinction is whether those dependencies are incidental or deliberately incorporated into a design. An exposure-driven intervention relies primarily on the amount of drug reaching its target. A context-dependent intervention would still require sufficient exposure, but part of its activity would also depend on a defined biological input.

The difference matters because a drug can be molecularly selective while remaining physiologically broad. It may bind the intended target and still act across tissues, signalling states or periods in which the same degree of activity is not required. Target selectivity is therefore necessary for many programs, but it does not answer every question about therapeutic precision.

Several experimental approaches have begun to make this principle concrete.

Glucose-responsive insulin offers a direct example. Insulin is indispensable, but matching its activity to changing glucose levels remains difficult. Too little activity permits hyperglycaemia; too much can cause hypoglycaemia. Researchers designed the investigational insulin NNC2215 with a glucose-responsive molecular switch that altered insulin-receptor affinity across changing glucose concentrations. The molecule showed reversible glucose-sensitive bioactivity in vitro. In rat and pig studies, it attenuated hypoglycaemia while partly covering glucose excursions.^[5]

These findings remain preclinical. They do not establish benefit in humans, eliminate hypoglycaemia or remove the need to manage exposure. They demonstrate a narrower design principle: a changing biological condition can contribute to the activity of an intervention.

A different form of conditionality has been explored in cancer. EGFR is a useful therapeutic target, but it is also present in healthy tissue. Researchers developed a masked EGFR-directed antibody that became available for target binding after cleavage by proteases commonly active in the tumour microenvironment. In preclinical models, this design produced conditional activity and improved the therapeutic index relative to the unmasked antibody.^[6]

Here, the relevant input was not EGFR abundance alone. It was protease activity associated with the local tissue environment. The design attempted to separate where the antibody travelled from where it could productively engage its target.

Synthetic biology has extended conditional control further. Researchers engineered T cells with a synthetic Notch, or synNotch, receptor that recognized one antigen and then induced expression of a chimeric antigen receptor directed against a second. In preclinical systems, the circuit enabled sequential combinatorial tumour recognition and improved discrimination between cells carrying both antigens and cells carrying only one.^[7]

A glucose-sensitive protein, a protease-activated antibody and an engineered immune-cell circuit are not members of a single therapeutic class. They use different modalities, inputs and mechanisms, and none establishes that conditional control will be feasible in every disease. Collectively, they show that biological context can add a source of therapeutic control beyond exposure alone.

Designing around biological state creates more ways for a hypothesis to fail.

The proposed input must distinguish a therapeutically relevant state with sufficient reliability. Its concentration or activity must vary across a range that the intervention can detect. The relationship must persist across dose, time and tissue rather than appearing only under one experimental condition. If the same input exists elsewhere in the body, conditional activation may simply move unwanted activity rather than prevent it.

The input may also change because of the treatment itself. A state-dependent mechanism that responds appropriately at the beginning of an intervention may behave differently after feedback, adaptation or disease progression alters the surrounding biology. Population averages can conceal subgroups in which the proposed relationship is absent. Measurements from isolated cells can disappear in tissue, and effects observed in tissue can fail to survive at the level of an organism.

These are not secondary development questions. They determine whether biological context provides meaningful control or only a more complicated description of conventional pharmacology.

Testing the idea therefore requires more than demonstrating that a molecule binds or that an effect changes in the presence of one signal. The intervention must be evaluated across relevant combinations of drug exposure and biological state. Its activity should be compared in conditions representing both therapeutic need and physiological function. Alternative explanations, including changes in target expression, distribution or metabolism, must be separated from the proposed context dependence.

A convincing result would show not only that the desired effect can occur, but that its relationship to biological state is reproducible, mechanistically coherent and useful within an acceptable therapeutic window. A negative result must be allowed to end the hypothesis.

Computational and structural methods can help identify candidate relationships, explore plausible mechanisms and prioritize experiments. They cannot establish that a biological input provides therapeutic control. That conclusion requires functional evidence across increasingly complete experimental systems.

Aeviant was founded to investigate whether biological state can become a deliberate variable in therapeutic design.

Aeviant’s research position is that molecular selectivity, although essential, is not a complete description of pharmacological precision. A discovery program should also define the conditions under which activity is intended to occur, the physiological functions that must remain intact and the evidence that would show the proposed dependence is not real.

This is not a validated platform, and Aeviant has not demonstrated state-dependent therapeutic control experimentally. It is a research position that must remain answerable to evidence. Each proposed mechanism creates its own advancement requirements and its own reasons to stop.

Some biological systems will not contain a useful conditional input. Some apparent relationships will collapse when tested outside a simplified model. In other cases, direct suppression, replacement or removal will remain the most appropriate therapeutic strategy. Context-dependent design should be used only where it produces a measurable advantage, not because it appears more sophisticated.

The proposition is deliberately limited: biological state can be treated as an experimental variable in drug design. In some systems, that variable may improve therapeutic control. In others, it will add complexity without benefit. Aeviant will judge the idea by the evidence, including when the evidence shows that biological state does not provide useful control.