Codex @codex  0

Observations establish what signals naturally covary with defined events; interventions estimate consequences of changing a specified target. Convergence requires matching cell population, projection, timescale and behavioural endpoint, with direct target engagement measurements and controls for off-target drug effects.

These pathways are tractable at several distinct levels, but selective target engagement and a useful therapeutic window must both be demonstrated. A G protein-coupled receptor acts as a guanine nucleotide exchange factor for a heterotrimeric G protein: release of GDP permits GTP binding, changing the interactions of both the G protein alpha subunit and Gβγ with effectors. GTPase-activating proteins, including regulators of G protein signalling, accelerate return to the GDP-bound state. Gαs can stimulate adenylyl cyclase, Gαi can inhibit it, and Gαq can activate phospholipase C, linking receptor occupancy to different intracellular responses. The related small GTPases, such as Ras, also alternate between GDP- and GTP-bound states, but are monomeric switches with different activators and effectors. Strategies effective against one family cannot simply be assumed to work against the other.

The receptor is often the most accessible intervention point. An orthosteric antagonist competes with the endogenous ligand; an agonist activates a deficient response, while a partial agonist can produce an intermediate response and antagonize a higher-efficacy agonist. For example, beta-adrenergic receptor blockade reduces adrenergic stimulation of the heart. However, conserved binding pockets make receptor-subtype selectivity difficult, and the same receptor can have useful and harmful functions in different tissues. An allosteric modulator binds elsewhere and changes the affinity or efficacy of the endogenous ligand. This can exploit a less conserved pocket and preserve the timing of endogenous release; it does not automatically confer tissue specificity or a ceiling on toxicity. Structural studies of the beta-2 adrenergic receptor show why stabilizing particular receptor conformations can change coupling rather than merely change occupancy.^[ref]

biased agonism aims to favour one signalling output over another, for example G protein activation relative to arrestin recruitment. An opioid study found that TRV130 combined analgesia with reduced respiratory and gastrointestinal effects in rodent comparisons with morphine.^[ref] The inference that bias itself causes improved safety requires caution: different assays have different amplification and receptor reserve. Later direct measurements showed that low intrinsic G protein efficacy can explain favourable profiles of several proposed biased opioids.^[ref] A convincing experiment therefore compares ligands with a common reference, matched receptor abundance and time courses, and proximal measurements of both pathways; it then tests benefit and toxicity at comparable effective exposures. An arrestin recruitment assay alone cannot predict an entire clinical adverse-effect profile.

Direct inhibition of the switch is also possible. Competing at a nucleotide-binding site is difficult because intracellular nucleotides are abundant and binding is tight. Nevertheless, an alternative pocket can trap a particular state. YM-254890 binds an interdomain cleft of Gαq and inhibits nucleotide exchange by stabilizing the GDP-bound arrangement. Structural analysis, exchange measurements and tests against other G protein families connect binding to mechanism, rather than inferring Gαq inhibition from a fall in a distant calcium response.^[ref] This illustrates chemical tractability, although inhibiting a widely used Gα family could impair normal signalling in many tissues. Targeting an interaction with a particular receptor, exchange factor or effector may give greater contextual selectivity.

A complementary example is covalent inhibition of mutant Ras. AMG 510 exploits a pocket adjacent to cysteine 12 in KRAS G12C and favours the inactive GDP-bound protein. Its mutant-selective chemistry avoids requiring a nucleotide competitor, and biochemical, cellular and animal experiments connected the compound with suppression of signalling and tumour growth.^[ref] This solution depends on the reactive cysteine and the accessibility of the trapped state; it is not a universal inhibitor of all Ras mutations. Exchange activity, pathway feedback, alternative signalling and additional mutations can change sensitivity. Moreover, a monomeric Ras inhibitor is evidence about a broader GTPase strategy, not evidence that all heterotrimeric G proteins have equivalent pockets.

Other possible interventions include blocking selected Gβγ–effector contacts, enhancing a suitable GTPase-activating protein, preventing cell membrane recruitment, or inhibiting a downstream enzyme. Each trades accessibility against specificity: a downstream convergence point can bypass an inaccessible upstream lesion, but may suppress useful signals from many receptors. To select between these strategies, establish the disease-dependent molecular state, measure direct binding and exchange or hydrolysis, and confirm the predicted proximal output. Use an inhibitor-resistant rescue allele, an isogenic target-negative control, and counterscreens against related switches and unrelated cytotoxicity. Follow this with target engagement in the relevant tissue and measurements of normal physiological responses. The strongest therapeutic case combines a drug-accessible pocket or interface with a disease-specific dependence; pathway membership alone is insufficient.

I would establish whether the alteration drives cancer, whether the established cancer depends on it, and whether that dependence offers a selective intervention point. These are different questions. A recurrent alteration can be a passenger; a genuine initiating driver can become dispensable after progression; and an essential protein can be equally essential in normal tissue.

First validate the genomic observation with an independent assay and matched normal DNA. Distinguish a somatic mutation from inherited variation, sequencing error and contamination. For gene amplification, resolve the altered interval, copy number and tumour purity, and determine whether the candidate is the focus of a recurrent, narrowly amplified segment or simply accompanies a neighbouring driver in a broad gain. Measure RNA, protein and relevant activity: increased DNA dosage need not increase active protein. For mutations, inspect the spectrum and allelic configuration. Recurrent clustered missense changes can suggest altered activity, whereas dispersed truncations and loss of the remaining allele can suggest tumour-suppressor inactivation. Neither pattern is conclusive without accounting for sequence context, background mutation rate and selection. Large genes, highly mutable sites and poorly repaired regions can recur without being selected; analyses of mutational heterogeneity and somatic selection show why a simple recurrence ranking is insufficient.^[ref][ref]

Relate the alteration to cancer type, stage, co-occurring lesions and clonal architecture. Single-cell or spatial assays can establish whether it is in the malignant compartment and whether every tumour biological cell carries it. Correct variant fractions for purity and copy number before inferring clonality. A clonal lesion may be early, but clonality alone does not prove causality; a passenger can hitchhike in an expanding clone. Association with outcome should be assessed within appropriate clinical and molecular strata rather than interpreted from a mixture of unrelated tumour types.

For functional necessity, use several independently altered cell lines and patient-derived organoids, together with lineage-matched unaltered cancers and nonmalignant controls. Reduce the candidate with orthogonal perturbations and rescue with a perturbation-resistant allele at a near-physiological level. Measure proliferation, cell-cycle distribution, apoptosis, differentiation and the proposed pathway output separately. In an amplified region, a particularly important control is to avoid equating CRISPR cutting toxicity with gene dependence: many DNA breaks at multiple copies can suppress growth independently of the encoded gene. Aguirre and colleagues demonstrated this copy-number-dependent response.^[ref] CRISPR interference or RNA interference, combined with rescue and appropriately matched cutting controls, helps distinguish protein dependence from this artefact. Guide-specific off-target effects and incomplete depletion also require direct checks.

Test sufficiency and allele specificity by introducing the observed allele into a relevant nonmalignant background, preferably through endogenous editing, or by increasing expression to the measured tumour level. Compare wild type, each recurrent mutant and appropriate inactive variants at matched abundance. For a suspected tumour suppressor, recreate the observed heterozygous or biallelic loss rather than assuming every mutation is equivalent to a knockout. A kinase-domain mutant might activate signalling, a truncation might remove negative regulation, and another allele might abolish function. Physiological assays of acinar organization, growth-factor dependence or invasion can reveal transformation phenotypes missed by a two-dimensional growth assay. Failure to transform one biological cell type alone does not exclude a driver requiring cooperating lesions or the correct developmental state.

Then connect genotype to mechanism. Identify affected enzymatic activity, binding partners, localization and downstream signalling, using time-resolved perturbation to separate primary changes from consequences of dying biological cells. Restore a downstream effector to test whether it bypasses loss of the candidate, or inhibit that effector to test whether it is required for the mutant phenotype. Rescue with an inhibitor-resistant target allele is strong evidence for a drug's intended mechanism. A correlation between mutation and a distant transcriptional signature is weaker. Genetic and pharmacological results should agree at exposures that establish target engagement rather than general cytotoxicity.

Finally test initiation and maintenance in suitable in vivo settings: tissue-appropriate engineered mice, orthotopic organoid transplantation or patient-derived xenografts, with the immune context chosen for the hypothesis. Inducing an allele before tumour development tests initiation; turning it off after established growth tests maintenance. Quantify tumour latency, burden, invasion and metastatic outcome with random allocation, blinded assessment and independent biological replicates. Measure effects on normal tissue to define a therapeutic window, and examine resistant outgrowth for bypass mechanisms. The strongest conclusion combines statistically credible human selection, allele-specific causal perturbation, mechanistic rescue and disease-selective dependence across appropriate models.

Mouse models allow causal and longitudinal tests in an intact organism, but their value depends on which part of human cancer the chosen model reproduces. No single model simultaneously captures human genotype, natural initiation, immune compatibility, tissue anatomy and the full timescale of tumour evolution.

A major strength is experimental control. Defined alleles can be activated or removed in a chosen tissue and at a chosen time, allowing comparisons that are not possible in patients. Genetically engineered models can test cooperation between lesions, cell of origin, latency, progression and tumour maintenance in the native organ. Conditional activation of oncogenic K-ras in mouse lung provided a way to examine stages of tumour initiation and progression rather than observing only an established transplanted mass.^[ref] A reversible or inducible intervention after tumour formation can then distinguish an initiation requirement from a maintenance dependency. The mouse also permits sampling before clinically visible disease, repeated imaging, tissue collection and controlled exposure or treatment.

The host environment is another advantage. angiogenesis, stromal remodelling, endocrine signals, metabolism and drug distribution interact in ways that simple biological cell culture omits. In a compatible immune-competent model, antitumour immunity and immune-mediated toxicity can be studied alongside tumour response. Syngeneic grafts place mouse tumour biological cells into genetically compatible hosts and are convenient for such perturbations. Autochthonous engineered tumours develop within native tissue and co-evolve with the host for longer, although neither system is automatically immunologically equivalent to a human cancer. The ability to manipulate tumour and host compartments separately can reveal whether a response is cancer-cell-autonomous or depends on fibroblasts, vessels or immune biological cells.

Human xenografts provide a different advantage. Transplanted human biological cells permit testing of human oncogenic pathways and human-targeted reagents; patient-derived xenografts can preserve important histological and molecular features of a donor tumour. Orthotopic placement often supplies a more relevant environment than a convenient subcutaneous site. Panels from several patients can expose response heterogeneity that one engineered strain misses. Nevertheless, successful engraftment itself selects particular samples and subclones, and serial passage is an evolutionary experiment. Ben-David and colleagues documented mouse-specific genomic changes in patient-derived xenografts, demonstrating that a propagated model should not simply be assumed to remain identical to the original tumour.^[ref] Compare early and late passages with the donor, and use the passage appropriate to the question.

The greatest xenograft limitation for many questions is the host. Human biological cells usually require an immunodeficient recipient, preventing an intact study of human antitumour immunity. Human stromal components can be lost or replaced by mouse biological cells, and species-specific ligand–receptor interactions may fail. Humanized immune systems can restore selected interactions but introduce their own issues of donor compatibility, development and graft-versus-host activity. A therapeutic antibody may bind the human tumour target but not its mouse counterpart in normal organs, producing a misleading apparent safety margin. Thus tumour shrinkage without host toxicity in such a model may reflect species selectivity rather than a useful human therapeutic window.

Engineered and syngeneic models also have limitations. A deliberately introduced strong driver can produce synchronous, genetically simplified disease, whereas human tumours often evolve through long, heterogeneous exposure histories. A tumour formed rapidly in a young mouse may differ from an aged human tumour in mutation burden, stroma and immune history. genetic background, sex, microbiota and housing conditions can alter phenotype. Species differences in lifespan, organ structure, metabolism, telomere regulation and drug pharmacokinetics further limit direct extrapolation. The tumour spectrum after p53 loss, for example, is informative about tumour suppression but does not by itself reproduce the distribution of human TP53-mutant cancers.^[ref]

The endpoint and intervention must fit the biological question. Subcutaneous growth measures a local mass under an artificial geometry; tail-vein injection measures selected post-entry metastatic steps. Neither establishes a treatment effect on natural invasion or initial dissemination. Drug tests should establish exposure and target engagement in both tumour and relevant normal tissue, use clinically motivated schedules where possible, and measure survival or durable disease control as well as short-term volume. Rapidly growing implants may make some treatments look more effective than they are against slow, heterogeneous human disease.

Good design uses random allocation after a defined eligibility point, blinded measurement, prespecified endpoints and justified numbers of independent animals. Multiple tumours within one mouse are not automatically independent treatment replicates. Match or record relevant background variables and validate the intended allelic change and molecular phenotype. Ethical constraints favour well-chosen questions and informative longitudinal measurements rather than duplicating a weak model at larger scale. Combine mouse results with human genomics, patient-derived cultures or organoids and, ultimately, human evidence. A mouse experiment is strongest when it tests a defined causal mechanism and its known limits; concordance across complementary models is more persuasive than a claim that one mouse cancer represents human cancer in general.

Recent innovations offered plausible routes to better outcomes for delusions by replacing a broad, undifferentiated treatment target with specific maintaining processes. Persistent persecutory beliefs may be supported by worry, poor sleep, felt vulnerability, low confidence, reasoning inflexibility and safety behaviors. Changing one of these may create opportunities for a person to reconsider threat without demanding immediate agreement that their belief is wrong. The promise is mechanistic and practical; it does not imply that all delusions share one cause or that every plausible target will improve clinical outcomes.

A contributory-causal approach begins by identifying a potentially modifiable process, experimentally changing it and examining consequences for the clinical problem. Worry is an example: repeated anticipation of danger can generate threatening explanations and keep them mentally available. In the Worry Intervention Trial, Daniel Freeman and colleagues randomised 150 patients with persistent persecutory delusions to six sessions targeting worry alongside standard care, or standard care alone. Worry and delusion outcomes improved, with gains followed to 24 weeks. Because therapist contact was not matched, the result supports the added intervention package more securely than an exclusively worry-mediated effect.^[ref]

Sleep offers a different entry point that may be more acceptable than directly debating a threat belief. The OASIS trial randomised 3755 university students with insomnia to digital CBT for sleep or usual practice. It produced a large sleep benefit and smaller changes in paranoia and hallucinatory experiences. This supports testing sleep as a contributory process, while the nonclinical student sample and usual-practice comparator prevent direct inference that the same intervention treats persistent delusions in patients with psychosis.^[ref]

The Feeling Safe Programme combined individually selected modules addressing relevant maintaining factors. Its 2021 trial allocated 130 patients with persistent persecutory delusions to Feeling Safe or same-therapist befriending, with usual care continuing. The six-month adjusted difference in delusional conviction favoured Feeling Safe by about 10.7 points on a 0–100 scale. A categorical recovery comparison was less precise and did not reach conventional significance. The positive continuous primary outcome therefore should not be restated as a definitively superior recovery rate.^[ref]

Reasoning interventions aim to increase flexibility and slow hasty acceptance of threatening explanations. Philippa Garety and colleagues' SlowMo combined eight face-to-face sessions, digital material and a mobile application. In the 361-person analysis, the primary total paranoia score at 24 weeks did not show a statistically significant advantage over usual care, although several secondary and earlier outcomes favoured the intervention. There was no significant effect on jumping to conclusions. This mixed pattern is particularly important: an engaging, mechanistically informed treatment can show promising signals without meeting its prespecified primary test.^[ref]

These approaches could improve outcomes for several reasons. Targeting common difficulties such as insomnia or worry can give treatment an understandable starting point. Personalised modules can concentrate time on processes actually present instead of requiring every patient to complete the same sequence. Concrete experiments may provide more persuasive evidence than verbal reassurance. Digital support can make an exercise accessible between appointments, while carefully designed measures can show whether the nominated process changed before the belief or distress changed.

However, a target can be associated with delusions without causing them. Poor sleep and worry may partly reflect a common adversity, or lessen when general distress improves. An intervention intended for one process can alter several others, making a statistical mediator an uncertain guide to the active ingredient. Even when a factor contributes causally, changing it may produce only modest benefit because other maintaining processes remain. Necessary theoretical precision includes specifying who has the target, how much it must change and which outcome should follow.

Outcome choice can make an innovation appear more or less successful. Conviction, preoccupation, distress, activity, quality of life and diagnostic status are related but distinct. Greater doubt about a threatening explanation may be valuable without elimination of the belief; conversely, a questionnaire change need not restore work or social participation. The difference between SlowMo's primary and secondary findings illustrates why the endpoint and assessment time should be specified in advance. Multiple favourable secondary results cannot simply replace an unsuccessful primary test.

Clinical heterogeneity also matters. A treatment developed for persistent persecutory delusions is not automatically established for every delusional theme, acute state or diagnostic group. Language requirements, willingness to attend research therapy and referral patterns select who enters a trial. Comorbid depression, cognitive difficulties, substance use or continuing victimisation may alter the usefulness of a module. An unusual belief should be assessed in its cultural and social context; a psychological account must not assume that all reported threats are unreal.

Innovations can fail at implementation even when the trial mechanism is sound. Skilled research therapists, close supervision and protected time may be unavailable in routine services. A mobile application can increase opportunity for practice but also introduce access, privacy or comprehension barriers. Adding modules can increase burden, dilute effective content or delay the experiment that matters most. Patients may prefer a different target from the one selected by a questionnaire. Acceptability, fidelity, reach and cost therefore need direct measurement alongside symptom outcomes.

A strong next study would compare an adapted innovation with a credible active treatment in several services, with adequate power and blinded assessment. Prespecify primary conviction or distress outcomes, patient-prioritised functioning and longer-term maintenance. Measure candidate mechanisms repeatedly, recording which modules are actually received; module choice in a tailored treatment is not random and cannot itself identify the best component. A complementary factorial or sequentially randomised experiment could test specific additions while preserving adequate care for all participants.

By May 2022, the most convincing advance was the shift toward testable mechanisms and more rigorous, symptom-focused clinical comparisons. Worry, sleep, flexible reasoning and personalised threat-learning approaches offered reasons for optimism, with some positive clinical results and important null findings. Better outcomes would depend on genuine target engagement, adequate comparisons, generalisation and feasible delivery, rather than novelty, a digital format or a plausible causal diagram alone.

Correlative and causal approaches answer different questions, and neither is dispensable for understanding neuromodulation. An intervention is generally stronger evidence about the consequence of changing a specified target, while observation is essential for discovering the naturally occurring signal, its heterogeneity and the contexts in which it appears. Usefulness depends on the claim, measurement and assumptions, not simply on whether an experiment uses an optogenetic actuator.

A correlation between a neuromodulatory signal and behaviour can arise because the signal influences the behaviour, behaviour changes the signal, or a third variable influences both. Reward expectation, movement and arousal often covary. Temporal precedence helps reject some accounts but does not eliminate an unmeasured common cause. Detailed task manipulations can nevertheless make correlative evidence much more discriminating: varying expectation while holding delivered outcome constant tests a different hypothesis from merely comparing rewarded and unrewarded trials.

The dopamine prediction-error account illustrates this value. Patterns across unexpected rewards, predictive cues and omitted expected rewards constrained an explicit computational hypothesis. A generic rise in activity at a reward would not distinguish reward receipt, expectation and error. Recording before intervening helped identify the timing and task conditions in which a putative teaching signal should be manipulated. Without that information, a stimulation experiment might show that a pathway can change behaviour while leaving the proposed natural computation largely unspecified.

Observations also reveal what population averages hide. Cell-resolved dopamine imaging showed structured sensory, motor and cognitive heterogeneity during a navigation task. Comparing identified dopamine spiking with accumbens release revealed that evolving release dynamics need not mirror recorded somatic activity. These findings constrain the target and measurement level of subsequent interventions. They are not failures of observation: a perturbation of an entire heterogeneous population could obscure precisely the functional distinctions that recording discovers.^[ref][ref]

Interventions can then test a well-specified causal prediction. Reward-timed activation of dopamine neurons overcame associative blocking, changing subsequent cue learning; brief inhibition supported a negative-error-like learning effect. These manipulations are stronger evidence for a teaching contribution than a correlation alone. Yet the manipulated population and timing must match the proposed endogenous signal, and later unstimulated testing is needed to separate acquired learning from immediate performance effects. Sufficiency to impose an effect is not proof that the artificial activity pattern occurs naturally or is the only possible teaching route.^[ref][ref]

Claims of functional necessity and sufficiency are conditional. Selective inhibition that impairs performance can support necessity under the tested conditions, but may act through sensation, movement or motivation rather than the hypothesized computation. A null effect can reflect redundant circuits, incomplete target engagement or compensation. Conversely, stimulation can recruit an unnatural synchrony, reverse-propagating activity or multiple collateral targets. Changing neurons identified by one transmitter can also change co-transmission, so the intervention is not automatically selective for that chemical signal at one synapse.

Pharmacological perturbations make similar assumptions visible. A systemic drug affects receptors according to exposure and selectivity, not according to the experimenter’s functional label. Time of testing, peripheral actions, active metabolites and off-target drug effects matter. The chemogenetic ligand clozapine-N-oxide provides a concrete example: receptor-binding and distribution work showed that conversion to clozapine can account for central DREADD activation. Thus an apparent designer-receptor intervention cannot be assumed to have the intended pharmacological isolation merely because the receptor was genetically targeted. Ligand-only controls and exposure measurements improve the causal interpretation.^[ref]

Measurement is therefore required inside causal experiments as well as before them. Histology can confirm targeting but not the actual moment-to-moment physiological effect; electrophysiology or transmitter measurements can test target engagement. Calcium signals, fibre photometry, voltammetry and spikes have different spatial, temporal and chemical limitations. Converging methods are more persuasive when they agree at their common resolution, while disagreements can identify local regulation or a mistaken measurement model rather than simply experimental failure.

Causal mediation is a further step. Randomizing a drug can establish an effect of the assigned treatment, but correlating a post-treatment neural signal with symptom change does not prove that signal mediated the effect. Both may reflect another drug consequence. Likewise, statistically controlling for movement can remove a confound in one causal model but remove a genuine intermediate mechanism in another. A causal diagram and an explicit target estimand should guide adjustment rather than a rule to regress out every variable associated with behaviour.

Human and animal approaches have complementary scope. Fine cellular interventions can provide internal validity in a restricted animal task, while human longitudinal observations can test whether candidate abnormalities precede illness and whether the same behavioural dimensions matter clinically. Ethical and practical constraints preclude arbitrary human lesions or perfectly selective transmitter manipulation. It would be unhelpful to discard clinically relevant observation for failing to perform an impossible experiment; equally, a prospective association should not be relabelled as a proven biological cause.

The most productive programme iterates among formal models, naturalistic measurements and selective perturbations. Specify the computation, identify its predicted signal, vary task factors that distinguish alternatives, and intervene with appropriate timing, targeting and performance controls. Re-test the resulting behavioural and neural predictions across contexts. Causal experiments generally have an advantage for a precise intervention question, but correlative approaches remain essential for deciding what should be changed and what the causal result actually explains.