Building a Drug Programme That Can Change Course

Drug development is unusually good at making activity look like progress. A target can generate years of meetings, assays, compounds and animal studies before anyone asks the primitive question with enough force:

Is the proposed disease mechanism actually present in the human system we intend to treat?

That is the first gate.

At Longinus Therapeutics, the operating premise is that a programme should begin with an experiment capable of making the programme unnecessary. This sounds hostile to ambition. It is the opposite. A hard early gate protects ambition from being consumed by the wrong mechanism.

A genetic association is not yet a therapeutic target

A copy-number change can identify a region of interest. It does not automatically identify the causal gene within that region, the affected cell type, the direction of the molecular disturbance or the intervention that will improve it.

Even apparently simple dosage logic can fail at several layers:

• An extra DNA copy may not produce proportionally more RNA.
• More RNA may not produce more functional protein.
• More protein may be buffered by degradation or complex assembly.
• A molecular difference may exist without producing a measurable cellular phenotype.
• A phenotype may arise from another gene in the interval—or from an interaction among several genes.
• The proposed mechanism may operate during development but disappear in the mature cell type being measured.

The gap between genotype and drug target is not a detail. It is the programme.

What the first gate should establish

A useful first gate has four layers.

1. The model is real

The cells must have the expected identity and genotype. Patient-derived induced pluripotent stem cells should retain the relevant copy-number state, remain genomically stable and differentiate reproducibly into the chosen neural model.

Without this, a negative result is uninterpretable. The biology may be absent, or the model may simply be broken.

2. The measurement is specific

Closely related genes create a recurring trap: an assay can produce a precise number for the wrong molecule. RNA primers, antibody reagents and mass-spectrometry peptides must be tested for paralog specificity rather than trusted by label.

Orthogonal measurements matter because every platform has characteristic failure modes. RNA should be measured with a validated quantitative assay. Protein should be measured with at least two independent approaches when antibody specificity is uncertain. Controls should demonstrate that the assay moves when the target is deliberately increased or reduced.

One beautiful Western blot is a photograph, not a decision system.

3. The signal is reproducible

The proposed disturbance should appear across independent cell clones and independent differentiations, not only across technical replicates from one successful plate.

Biological replication is expensive because biology is variable. It is also the only replication that answers the question. Repeating the same lysate three times measures the instrument. Repeating the differentiation measures the programme.

4. The signal connects to a reversible consequence

A target-engagement assay can show that a molecule changes RNA or protein. It does not show that the change matters.

The first useful phenotype need not capture the entire disorder. It should be measurable, reproducible and capable of moving in the predicted direction when dosage is normalized. That may be a pathway readout, neuronal morphology, synaptic organization or another cell-state feature supported by the biology.

The word reversible is critical. A developmental scar that cannot change in the model may be scientifically interesting but operationally useless for ranking therapeutic candidates.

Design the gate before seeing the data

The pass criteria must exist before the result.

Otherwise every weak result acquires a post-hoc explanation. The threshold changes. A secondary assay becomes primary. One favourable clone is elevated above two neutral clones. A suggestive pathway is promoted to a phenotype. The programme survives, but the evidence has stopped constraining it.

A credible gate specifies in advance:

• Which cell types and controls will be compared.
• How many independent clones and differentiations are required.
• Which measurements are primary and which are exploratory.
• What assay quality is acceptable.
• What constitutes reproducible direction and magnitude.
• Which result advances the programme.
• Which result permits one bounded repair.
• Which result stops the target.

This is not bureaucracy. It is a checksum against motivated reasoning.

Failure has more than one meaning

A failed gate is useful only if the failure is classified correctly.

Technical failure

The assay cannot distinguish the target from a related protein, the cells fail quality control or different platforms disagree. The right response is to repair the measurement once and repeat it.

Technical failure does not falsify the biological hypothesis. It also does not permit advancement.

Model failure

The molecular signal exists, but the selected cell model shows no usable consequence. A fast induced-neuron system may be too immature or too compositionally simple. The mechanism may require neural progenitors, astrocytes, inhibitory neurons or longer maturation.

The disciplined response is one evidence-selected model revision—not a tour of every fashionable model system. Organoids are not a substitute for a precise question.

Target failure

The proposed gene is not measurably dysregulated, while another gene in the duplicated interval is. The programme should move from narrative to perturbation: normalize each candidate gene toward the control range and observe which intervention reverses the cellular state.

This can rescue the therapeutic thesis while killing the preferred target. That is a good trade.

Interval failure

No individual candidate explains the phenotype, or normalization of the entire interval fails to move it. At that point the copy-number change may be acting through a combination of genes, a developmental window not captured by the model or a mechanism unrelated to the original hypothesis.

An isogenic comparison becomes decisive. Patient cells should be compared with genetically matched cells in which the relevant dosage has been normalized. Differences that reverse in the matched background are much more informative than differences between unrelated people.

If the interval still fails to produce a causal, reversible signal, the programme should stop treating it as the therapeutic driver.

The laboratory is a decision engine

The purpose of an early laboratory is not to possess equipment. It is to shorten the interval between a hypothesis and a reliable decision.

That changes what should be built internally. Daily cell biology, assay integration and experimental judgment may belong close to the team. Infrequent sequencing, quantitative proteomics, advanced imaging and regulated work can remain in specialist cores until outsourcing becomes slower or more expensive than ownership.

The best part is no part. The best instrument is the one that does not need to be purchased because a validated external service returns the required data tomorrow.

Infrastructure should follow the biological loop, not precede it.

The second shot must also be finite

Stopping after one ambiguous result is reckless. Refusing to stop after repeated negative results is worse.

A rational rescue sequence is:

1. Repair the measurement once.
2. Test one evidence-selected alternative model.
3. Normalize candidate genes individually and in justified combinations.
4. Use an isogenic correction to test whether the interval is causal.
5. If no reproducible phenotype reverses, reopen the causal search.

This sequence creates multiple shots on goal without converting uncertainty into an indefinite research programme.

What a pass buys

Passing the first gate does not prove that a drug will work. It earns the right to start optimizing one.

Only then does it make sense to compare therapeutic sequences, delivery conditions, potency, selectivity, durability and toxicity. Only then can a reduction in target RNA or protein be connected to a phenotype that matters. Only then does additional infrastructure compound learning rather than compound sunk cost.

The first gate is therefore not the cautious phase before the real programme.

It is the real programme’s first act: make reality expensive to misunderstand.