When Population Models Travel: Calibration Drift and Decision Risk in Phage PK/PD, antimicrobial therapy and mechanistic dosing

Population Models Travel is a problem that has to do with the center of phage pharmacokinetics and pharmacodynamics, in which a modeled group can swing when host immunity, the density of the bacteria, the susceptibility of the bacteria to the phage, or the route of introduction changes. The more successful development strategy is based upon a measurable mechanism and not just an impressive assay result. That change creates a single translational argument with parts being chemistry, biology, exposure, resistance, safety, and production.

The scientific basis is the process of which mechanistic phage PK/PD pertains to distribution, immune clearance, bacterial killing, phage replication, adsorption, latent period, burst size and emergence of resistance. The mechanistic claim must be expressed in terms of each individual component, otherwise no isolated measurement could carry the entire mechanistic claim. By linking evidence, the probability of mistaking potency, binding, formulation performance or model fit for therapeutic understanding is minimized.

Unlike conventional antibiotics, phages can replicate in areas of susceptible bacteria and not survive in areas devoid of bacterial hosts or inaccessible to phages. The adsorption, burst size, latent period, bacterial density, immune clearance, and tissue distribution are thus nonlinearly related. In pneumonia, biofilm disease or abscess, there is a change of balance between transport and local amplification of the model, which can cause its drift.

There are several variables, such as phage strain, bacterial species, inoculum, tissue site, route, dose, immune status, antibiotic exposure and sampling schedule, that can change in concert to affect performance and must be treated together. When optimizing one factor, there is a hidden penalty in another, in particular when potency is related to stability, selectivity, delivery or manufacturability. Multiparameter decision rules are thus a more realistic criterion for ranking candidates than a single number.

Early measurement of failure routes is also a key factor in establishing mechanistic confidence. Receptor loss, restriction systems, CRISPR defense, biofilm protection, spatial refuges, immune neutralization and bacterial physiology changes are important escape mechanisms. Deliberate stress testing transforms these routes into design information which can be used for the next chemistry cycle, formulation, dosing or more model cycles.

Exposure is the link between the nominal dose and the biological effect. These are such parameters as concentration of circulating phage, penetration into tissues, local amplification, reticuloendothelial clearance, adsorption rate and number of active bacteria. Strong intrinsic activity and weak exposure at the causal site may present a candidate with a great performance in simplified assays, but be unsuitable for translation.

Interpretation of direct engagement or functional perturbation must be done at realistic concentrations. Very high doses may induce secondary stress responses, have broad off target effects, or cause assay artifacts that appear to be a desired mechanism. Concentration-response analysis, coupled with exposure and occupancy, can be used to pinpoint a range where the proposed causal process is thought to have an action.

There are also multidrug-resistant Gram-negative and other challenging bacterial infections that are influenced by biological context and the interpretation of the context. Target dependence and effective exposure can vary between individuals due to differences in genotype, cell state, tissue architecture, immune tone, metabolic environment and previous treatment. Repeated testing in a single system is not as compelling as representative model panels.

Retrospective explanations should be deferred until the time for prospective measurements of resistance and adaptation. Which mechanisms survive continued pressure can be determined in a serial passage, long-term selection, genetic perturbation, repeated dosing or environmental stress. This isn't to say that a weak candidate with a high escape threshold is better than a strong, fragile screening candidate, just that it has a higher potential for translating.

Mechanism provides information to help assess safety. Endotoxin control, inflammatory response, the speed of bacterial lysis, formation of immune complexes, effects on the microbiome, and rout-specific tolerability are all important concerns. A comparison of desired from unwanted biological responses informs on the distinction between mechanism-related pharmacology and broad-based cellular injury and formulation-related toxicity.

Manufacturing cannot be a downstream logistics exercise since the identity, purity, potency, host range, genetic characterization of the phages, endotoxin removal, stability and composition of the phage cocktail all affect the product presented to the biological systems. Even if the product name is the same, a small change in the process could result in a change in the product's composition, physical state, impurity load, stability, or dose reproducible. Enhancing the robustness of development is facilitated by having critical process attributes be directly tied to mechanistic performance.

Population PK/PD, predator-prey models, hierarchical Bayesian models, stochastic simulation, posterior predictive checks, and decision analysis are all tools available to computational tools that add leverage. There is most usefulness in prediction when there is ambiguity and applicability limits still present, particularly where there are new chemical structures, unusual biological states or where there is little data. Models that reveal uncertainty allow for improved experiment selection than models that report confident rankings without revealing areas of uncertainty.

Prospective validation involves the use of patient holds, route changes, independent preparations of phages, use of alternative bacterial loads, and the calculation of prospective prediction intervals. Orthogonal tests are important because if repeated, tests from the same assay family can give the same hidden bias which doesn't strengthen the causality. When two or more independent methods agree there is greater confidence in the link from design to biological effect; when they disagree, then the assumptions must be changed.

It is important that data integration maintains traceability from raw experiment down to a development decision. Chemical structures, assay conditions, assay metadata, exposure data, omics data and resistance data and safety signals must be identified and versioned. This discipline makes it harder for later discussion to get off course from the conditions which resulted in a product.

Decision metrics also must incorporate translation and not laboratory show. The components of useful scorecards include reproducibility, exposure normalized activity, mechanistic confidence, resistance frequency, safety margin, stability and manufacturing robustness. This way, they will be paid for maintaining a balance and it will show them the trade-offs in advance, before they end up taking on a lot of expensive commitments.

In this context, negative result is a special value. The edge of a hypothesis is determined by: Failed synthesis, missing target engagement, unstable exposure, absence of rescue, emergence of resistance, poor batch reproducibility, loss of activity in realistic models. Documenting failures helps to enhance the design of future experiments, and avoids reinvestments in appealing, but unsupported explanations.

A third filter is the reproducibility across laboratories or batches. When it appears the results are significant, a mechanism relying on a single operator, single preparation, single cell line, or single modeling option is inherently fragile. When conditions are repeated independently, and with exposure matched, it is then a matter of durable biology or local experimental success.

There is also a practical requirement for a translation in clinical and/or deployment conditions. Technically complex products may have little value if their route of administration or diagnostic needs are complex, if storage conditions are critical, if there is a large administrative burden, if the cost is prohibitive, or if the timing of the treatment is too critical. Mechanistic excellence has impact when the product design is appropriate in the context in which biological benefit needs to happen.

However, there is another level of complexity when considering biological systems as they respond to a short pulse, a prolonged exposure, washout, and repeated dosing; this is temporal design. Time resolved measurements can help to distinguish primary engagement from downstream adaptation, and can show if there is any early activity that lasts after the initiating signal has subsided. Therefore, sampling schedules must be compatible with the kinetics of phages rather than being based on one convenient endpoint of phages pharmacokinetics and pharmacodynamics.

An integrated development record (IDR) helps to facilitate better communication among medicinal chemistry, formulation science, pharmacology, computation, manufacturing and clinical planning. Rather than a single "definitive" story, the record can record links that have been established, hypotheses of reasonable probability, unresolved uncertainties, and the next experiment with the highest information value. With that transparency, iteration becomes quicker and there's less pressure to stick to initial assumptions in the face of conflicting data.

When considering novelty, it is important to compare it to acceptable reference conditions. Whether a new approach is better than established compounds (inactive controls), standard formulations, baseline models, or accepted analytical methods is defined by these. A head-to-head comparison also reveals gains which cannot be observed after normalization to exposure, assay sensitivity, batch quality or experimental complexity.

The only exception to this is that, in a phage model, there must be uncertainty, because if one could be confident in a prediction in an inappropriate biological regime, this would be more dangerous than a prediction that was open to question. It is not the value of one assay that is the deciding factor, but the consistency of the proposed mechanism with regards to exposure, biological variation, resistance, safety and yield. That standard puts in place a more "disciplined" way to go from discovery signal to a therapeutic concept that will be able to stand the complexity of the real world.