A bacteriophage that infects one bacterial strain may have little activity against another. For personalized phage therapy, that specificity creates a practical problem. Clinicians need a phage active against the bacterium infecting a particular patient, but finding activity in a laboratory test is only the beginning. The phage must be prepared to an appropriate quality standard, delivered as part of a treatment plan, and assessed for its effects. Researchers must also determine how much a laboratory susceptibility result can tell them about the patient’s likely response.
Recent publications address different parts of this challenge. One group used machine learning to design patterns of phage activity against bacterial hosts.1 Two others developed methods for disrupting genes across phage genomes to learn which genes a phage needs, including under particular conditions.2,3 A new consensus guideline sets out recommendations for organizing personalized phage treatment.4 These are discrete advances in design, biological understanding, and clinical practice, respectively. They were not tested as a combined approach.
Their proximity raises a question for the field: can phage therapy move toward a process in which researchers specify desired properties, understand what allows a phage to infect its target, and test and prepare that phage in ways that support a clinical decision? The new work makes the question more precise. It also exposes how much remains between an engineered laboratory result and a dependable treatment.
Designing a Phage’s Targets
Personalized phage therapy begins by testing available phages against a bacterial isolate. If a candidate shows activity, clinicians may consider it for treatment. Naia Novy and colleagues investigated a different possibility: modifying a phage to achieve a specified pattern of activity across bacterial hosts.1
The researchers focused on the receptor-binding protein of bacteriophage T7. They used measurements of how protein variants performed across five Escherichia coli strains to train models that predicted the effects of changes to the protein. They then designed and experimentally tested phages against 26 objectives. Some objectives favored activity against an individual strain. Others sought particular combinations of target and nontarget strains or activity across all strains used for training. The researchers identified phages satisfying each of the 26 design objectives.
Those objectives capture a distinction that could matter in phage development. Maximizing activity against one bacterium is a different task from increasing activity against several while limiting it against another. A broadly active design and a highly selective design may both be technically successful, yet answer different needs. Novy and colleagues showed that host-targeting patterns could be specified in advance and pursued through model-guided changes, rather than characterized only after a phage was found or modified.
The study also suggests that seemingly different targeting profiles can be separated by relatively few changes to the receptor-binding protein. That gives researchers a way to investigate how changes in a phage protein alter activity across hosts. It does not establish that every phage offers the same flexibility, or that the best-designed profile in this experiment would prove useful against a patient’s infection. The models and experimental results concern T7 and the bacterial strains examined in the study.
Designing a phage that meets a stated objective in a defined laboratory system demonstrates control over the property being measured. Clinical development would require researchers to decide which targeting profile they need, show that they can produce it in a relevant phage, and test its performance against the bacteria and under the conditions that matter for treatment. The Novy study advances the first of those questions.
Changing a receptor-binding protein can alter phage activity across hosts, but it does not reveal which other phage genes or bacterial defenses contribute to that activity. To investigate those other requirements, researchers need ways to test what different changes impact activity under different conditions.
Which Phage Genes Matter?
Phage genomes contain many genes with unknown or incompletely understood functions. This makes engineering less predictable: a researcher may know what change they want to make without knowing whether a genomic region can tolerate disruption or whether an apparently dispensable gene will prove important against another bacterial host. Two recent studies published in Nature Microbiology used transposon insertion to investigate such questions across phage genomes.2,3 Their methods differ, as do the capabilities they emphasize.
Natalie Kyte and colleagues developed phage Tn-seq, which combines transposon mutagenesis, anti-CRISPR-based selection, and deep sequencing.2 By examining where insertions persist among viable phages, the researchers identified genes and genomic regions required under the conditions tested and those that tolerated disruption. They applied the approach across diverse phages, including a nucleus-forming jumbo phage, and compared essentiality assignments with other evidence about the phages’ genes and proteins.
An essentiality map is useful for more than labeling genes. If an insertion disrupts a function a phage needs, that location is a poor candidate for adding new material. A location that tolerates an insertion may offer room for further investigation or modification. Kyte and colleagues put that possibility to the test by using transposons to introduce genetic cargo, including a fluorescent reporter and anti-CRISPR genes. In one experiment, they identified recombinant phages within two days and confirmed reporter expression on the third. They also achieved transposon insertion in phages with hypermodified DNA.
These results bring experimental speed to a specific engineering problem: researchers could introduce cargo and identify viable modified phages without first choosing a genomic insertion site solely on the basis of an existing annotation. However, viability and successful cargo insertion are early measures. They do not show that a modified phage has the complete set of properties needed for therapy. The results with hypermodified-DNA phages likewise establish that insertion was possible in those phages; they do not constitute a complete essentiality map for each one.
Dorentina Humolli and colleagues approached gene essentiality with HIDEN-SEQ, another transposon-insertion sequencing method.3 They used it to investigate not only which genes a phage requires generally but also which become important under specific conditions. Testing mutant libraries across bacterial hosts and growth conditions revealed previously uncharacterized factors associated with overcoming specific bacterial antiviral defenses.
The distinction between generally essential and conditionally essential genes has direct bearing on how researchers interpret phage activity. A gene that appears unnecessary when a phage grows on one bacterial host may matter when another host presents a defense that the phage must overcome. Studying the phage on a single laboratory host could miss that dependence. HIDEN-SEQ provides an experimental way to identify and investigate it.
The two new methods build on earlier genome-wide phage research. Their value here lies in the questions they make tractable. Kyte and colleagues show how insertional mutagenesis can map essential regions and rapidly introduce cargo into viable phages. Humolli and colleagues show how comparing insertions across conditions can reveal genes needed against particular bacterial defenses. One study emphasizes both genome mapping and modification; the other makes the changing requirements of infection especially visible.
The Novy models predict an activity phenotype from changes to one phage protein; the functional-genomics studies examine contributions from other parts of phage genomes under different conditions. These researchers have not combined model-guided targeting design with genome-wide functional testing to produce a therapeutic candidate, but they show why that combination might be worth investigating.
Testing Against the Patient’s Isolate
However well researchers understand or modify a phage, a personalized treatment decision must return to the patient’s bacterium. Phage susceptibility testing asks whether a candidate shows activity against the bacterial isolate. Its results, often presented as a phagogram, help inform matching. The new consensus guideline from a personalized phage therapy guideline group recommends confirming susceptibility as close to treatment initiation as possible because it can change over time.4
The difficulty is deciding how much weight to put on the result. A test can establish activity under the conditions in which the laboratory performed it. Whether that activity predicts a clinical response is a separate question. The guideline calls for standardized and harmonized susceptibility and potency testing, interlaboratory proficiency testing, and research correlating laboratory susceptibility results with patient outcomes. It also describes the clinical evidence base for phage therapy as limited.
The functional-genomics work offers one possible route toward a more informative interpretation of laboratory results. If a phage grows on one bacterial host but struggles on another, experiments that identify conditionally essential genes could help researchers investigate why. A result associated with a particular bacterial defense might explain a difference that a simple record of susceptibility leaves unexplained. That is a research possibility, not an established clinical use of HIDEN-SEQ. The study does not show that sequencing a phage or identifying one of its genes can predict a patient’s response.
Additionally, greater mechanistic detail would likely not replace direct testing. A designed targeting profile describes what researchers sought and observed in one experimental system. A functional-genomics experiment describes which genes mattered under its test conditions. A phagogram addresses the activity of a candidate against a particular isolate. Each answers a different question. The unresolved task is to learn which combination of measurements helps clinicians choose a phage and anticipate what will happen during treatment.
The connection matters for future design work. A phage could meet a laboratory targeting objective without performing as hoped in a test against a patient’s isolate. Researchers need consistent tests and clinical outcome records to determine which designed properties deserve priority, rather than assuming that a more precisely optimized laboratory measure will yield a better treatment choice.
Building a Clinical Process
Selecting a candidate still leaves the practical work of treatment. Developed within a German medical-society framework with international participation, the consensus guideline drew on 20 professional societies, patient advocacy groups, regulatory authorities, and 18 international experts. Its more than 60 recommendations propose a framework for personalized practice spanning core principles, infrastructure, preparation and quality control, administration, and future research. Nevertheless, they should not be mistaken for an internationally adopted standard.
The process begins with identifying the bacterium and assessing candidate phages against it. A treatment team then needs access to a preparation facility with defined quality controls and clinical expertise to decide whether, how, and when to administer the phage. The guideline describes coordinated roles for phage laboratories, clinicians, microbiologists, pharmacists, and interdisciplinary phage therapy boards. These capabilities need not all be located at the same institution, but their results and decisions must connect in time to support a patient’s care.
The work continues after treatment starts. Monitoring, microbiological sampling, and documentation can reveal changes relevant to an individual patient while building a record that others can study. The guideline calls for comprehensive case documentation in registries and for continued research into how susceptibility results relate to outcomes. Those recommendations make clinical learning part of the proposed system, rather than something left to isolated reports of treatment attempts
Seen in that sequence, the laboratory studies address important but bounded tasks. Model-guided design might help researchers propose a phage with a useful targeting profile. Functional genomics might help them understand requirements for infection or introduce a modification. Neither activity determines by itself whether the resulting phage can be prepared to the required standard, selected on the basis of an informative test, or administered in a way that benefits a patient. The guideline describes a framework for those later decisions; it cannot establish efficacy.
The next test for an engineering approach is whether better-defined phage properties lead to better decisions and outcomes. Researchers would need to connect what they design or discover with the measurements used to select a candidate, the controls used to prepare it, and the results observed in patients. These publications give them more precise ways to work on each part of that problem. Whether joining those parts improves treatment remains open.
