Making Mycobacterium tuberculosis killable: Operational survival phenotyping beyond the persister metaphor

Abstract

Why can genetically susceptible Mycobacterium tuberculosis (Mtb) survive months of otherwise effective chemotherapy? The persister concept gave this problem a powerful name, but its expanding use has increasingly blurred distinct causes of residual survival. Slow killing, stress-adapted physiology, residence in protective host niches, inadequate local drug exposure, and the capacity to recover after treatment can all leave viable bacilli behind, yet they do not represent the same therapeutic problem. The central challenge is therefore not simply to detect survivors, but to connect these observations across treatment and determine where irreversible killing fails. We use State–Niche–Exposure–Fate (SNEF) to make this causal chain explicit, not to replace established assays or definitions. Pyrazinamide serves as a worked model because its sterilizing activity reflects the interaction of bacterial State, lesion context, and effective Exposure rather than conventional susceptibility alone. This shifts the experimental question from whether Mtb is a “persister” to which step from susceptibility to irreversible death remains open. Closing that step may offer a more direct route to durable sterilization and shorter tuberculosis therapy.

1. Introduction

The prolonged duration of chemotherapy remains a major challenge in tuberculosis (TB) treatment. Although four-month regimens are available for selected patients with drug-susceptible TB and six-month all-oral regimens are now prioritized for eligible patients with multidrug- or rifampicin-resistant TB, treatment remains prolonged or complex for many patients, particularly when shorter regimens are unsuitable (Dorman et al., 2021; World Health Organization, 2025). Such prolonged courses impose severe burdens on patient adherence, strain healthcare systems and increase the risk of treatment discontinuation. A central biological constraint on further treatment shortening is that effective therapy requires eliminating the residual, genetically susceptible bacilli. If this sterilizing endpoint is not reached, residual bacilli may seed relapse and, under spatially or temporally uneven drug exposure, provide opportunities for resistance amplification (Zhu et al., 2023).

This requirement exposes a fundamental gap between susceptibility and sterilization. Conventional phenotypic drug susceptibility is defined largely by the minimum inhibitory concentration (MIC), which measures the concentration required to inhibit visible bacterial growth under specified conditions. MIC therefore establishes whether growth can be suppressed, but not whether a bacterial population is irreversibly killed. Sterilization is a different and more stringent endpoint: the elimination of viable bacilli capable of surviving treatment and subsequently re-establishing growth. Thus, the central problem in treatment shortening is not simply whether Mtb is susceptible to a drug, but whether that susceptibility is translated into irreversible elimination of the bacterial population. Here, we use killability to describe the context-dependent ability of a defined Mtb population to be irreversibly eliminated, thereby linking drug susceptibility to the sterilizing endpoint without proposing a new scalar metric. This review examines how existing kinetic, physiological, host-contextual, pharmacological and recovery-based frameworks capture different aspects of this process, and where their interpretations remain disconnected.

2. From the persister paradigm to measurable survival phenotypes

The persister concept arose from a simple but therapeutically disruptive observation: a fraction of genetically susceptible bacteria can survive otherwise lethal antibiotic exposure without acquiring stable resistance. Bigger first described this phenomenon in staphylococcal cultures, in which penicillin eliminated most bacteria but spared cells that remained susceptible after regrowth (Bigger, 1944). Tuberculosis gave the observation its clearest clinical significance. In mouse models, McCune and colleagues showed that Mtb could survive intensive combination chemotherapy in host tissues and re-emerge after treatment withdrawal (McCune et al., 1966). These experiments established residual, recovery-competent bacilli as a biological link between incomplete sterilization and relapse. The persister paradigm therefore did more than name an unusual bacterial subpopulation: it provided a language for the gap between genetic drug susceptibility and sterilization.

Over subsequent decades, however, “persistence” acquired increasingly broad physiological, anatomical, experimental, and clinical meanings. In TB, the term became associated with slow or arrested growth, hypoxia, nutrient limitation, altered metabolism, lesion residence, biphasic killing, residual viability, and post-treatment recurrence. Zhang’s Yin–Yang model gave this heterogeneity an influential organizing narrative by depicting growing, slow-growing, and nongrowing bacillary populations that coexist and interconvert, and by relating their differential drug responsiveness to multidrug therapy, prolonged treatment, and relapse (Zhang et al., 2012). Its major contribution was to replace a rigid active–dormant dichotomy with dynamic population structure. At the same time, the model illustrates how a successful unifying metaphor can span several levels of explanation: a physiological state, a kinetic response, an anatomical context, and a clinical outcome may all be described as “persistence”, although they are not experimentally equivalent.

Modern kinetic frameworks have restored important precision by distinguishing resistance, tolerance, and persistence according to MIC, killing rate, minimum duration for killing (MDK), and the presence of a slowly killed subpopulation (Balaban et al., 2019; Brauner et al., 2016). These distinctions matter because dormancy is not synonymous with persistence, population-wide slow killing is not necessarily caused by a rare subpopulation, and residual colony-forming units do not establish whether surviving bacilli were adequately exposed or remain capable of recovery. Likewise, lesion-conditioned survival, delayed culturability, post-antibiotic regrowth, and relapse describe different stages of the treatment response and support different mechanistic conclusions (Boldrin et al., 2020; Goossens et al., 2020).

The historical value of the persister paradigm is therefore not in question. The challenge is to prevent the label from substituting for the phenotype actually measured. In this review, “persistence” is used for a demonstrated subpopulation survival phenotype, whereas tolerance, heteroresistance, niche-conditioned protection, post-exposure recovery, and evolutionary escape are treated as related but non-equivalent observations. Table 1 summarizes these working definitions and the experimental implications of each for interpreting Mtb survival.

Table 1.

Working definitions for operational survival phenotyping.

Term	Working definition	Primary experimental signature	What it does not mean	Relevance to the SNEF framework
Resistance	Heritable reduction in drug susceptibility, usually reflected by an increased MIC or a defined resistance-associated genetic change.	Increased MIC, resistance-associated mutation, or a stable resistant phenotype after passage.	Resistance should not be inferred merely from survival after transient drug exposure.	Defines the conventional susceptibility baseline against which phenotypic survival is interpreted.
Tolerance	Population-wide survival of bactericidal exposure through slower killing without a necessary increase in MIC.	Reduced killing rate, increased time to kill, or prolonged MDK with unchanged or minimally changed MIC.	Tolerance is not equivalent to resistance and is not necessarily restricted to a small subpopulation.	Represents a kinetic survival phenotype whose cause may involve State, Niche, Exposure, or their interaction.
Persistence	Survival of a subpopulation of genetically susceptible bacteria during otherwise lethal antibiotic exposure without acquisition of stable resistance.	Heterogeneous killing kinetics, often including a biphasic curve or a more slowly killed survivor subpopulation, with conventional susceptibility retained after regrowth.	Persistence should not be used as a generic label for all residual bacilli, dormancy, poor drug penetration, relapse, or resistance.	Represents one operational survival phenotype whose interpretation requires specification of the relevant State, Niche, Exposure, and subsequent Fate.
Heteroresistance	Coexistence of subpopulations with different levels of drug susceptibility within an apparently uniform bacterial population, often with a minority subpopulation showing higher stable or semi-stable resistance.	Growth of a minority subpopulation at concentrations that inhibit the majority; altered population analysis profile; enrichment under drug selection.	Heteroresistance is distinct from non-heritable persistence or tolerance, although bulk susceptibility testing may miss it.	Should be distinguished from phenotypic survival before recovery-related Fate or downstream evolutionary risk is interpreted.
Post-exposure recovery/resilience	Post-exposure recovery describes the resumption of metabolic or replicative activity after drug pressure is reduced or removed; resilience refers more specifically to rapid post-antibiotic regrowth despite little change in conventional susceptibility or during-exposure survival.	Post-antibiotic lag, time to detectable regrowth, regrowing fraction, metabolic restart, and dependence on resuscitation-permissive conditions.	Recovery is not evidence that cells were resistant during exposure. Differential culturability, recovery competence, and resilience are related but not synonymous.	Maps primarily to Fate by distinguishing residual detection from the capacity and kinetics of return to population growth.
Killability	The context-dependent ability of a defined Mtb population to be irreversibly eliminated under specified biological and pharmacological conditions, thereby linking conventional drug susceptibility to the sterilizing endpoint.	Time-resolved killing interpreted together with post-exposure recovery under defined conditions; no single universal readout or score is implied.	Killability is not equivalent to MIC or bactericidal potency, is not a fixed intrinsic property of a drug or strain, and is not proposed as a new scalar metric.	A context-qualified property interpreted across State, Niche, Exposure, and Fate; it is not an additional SNEF coordinate.

Note: Definitions of resistance, tolerance, and persistence are aligned with established kinetic frameworks (Balaban et al., 2019; Brauner et al., 2016). The distinction between during-exposure survival and post-antibiotic resilience is informed by Liu et al. (2022). Killability is used here descriptively as a context-dependent property, not as a new quantitative metric.

3. Established frameworks for mycobacterial survival phenotyping

3.1. Kinetic definitions: susceptibility, tolerance and persistence

Kinetic frameworks begin with a deliberately narrow question: how does a bacterial population respond to a defined antibiotic concentration over time? The MIC addresses the concentration dimension by identifying the lowest concentration that prevents detectable growth under specified conditions. An increased MIC therefore provides the canonical signature of resistance, but it does not establish whether bacteria are rapidly killed, slowly killed or merely growth arrested at concentrations above the MIC. Genetically susceptible populations with similar MIC values can consequently exhibit markedly different time–kill trajectories. Time-kill assays based on serial viable counts add this missing temporal dimension by estimating the early population-level killing rate and detecting later deviations from monophasic killing, including the emergence of a slowly killed survivor tail (Balaban et al., 2019; Brauner et al., 2016). Brauner, Balaban and colleagues formalized this temporal phenotype through the minimum duration for killing (MDK), defined as the exposure time required to reduce a bacterial population by a specified proportion. MDK₉₉ measures the time required to kill 99% of the initial population. A population-wide increase in MDK₉₉ without a corresponding MIC increase is interpreted as tolerance: the population remains susceptible in the concentration dimension but survives longer in the time dimension. Persistence describes a different kinetic architecture in which a subpopulation is killed more slowly than the majority, often producing the characteristic biphasic killing curve. Deeper killing thresholds, such as MDK₉₉.₉₉, the slope of the late killing phase, and the survivor fraction at a defined time can help quantify this heterogeneous response (Balaban et al., 2019; Brauner et al., 2016). Importantly, survivor fractions are not intrinsic constants; they depend on inoculum, growth history, antibiotic concentration, exposure duration, sampling depth, detection limits, and recovery conditions. Standardization of these variables is therefore part of the phenotype definition rather than a secondary technical consideration.

Population-level killing curves reveal heterogeneous survival but do not identify which cellular behaviors generate it. Single-cell studies addressed this gap by tracking growth and lag at the level of individual bacteria. Classic microfluidic experiments showed that persistence can arise from pre-existing phenotypic heterogeneity within an isogenic population (Balaban et al., 2004). Fridman and colleagues later described a related but distinct form of tolerance: under repeated antibiotic pulses, populations evolved longer lag times, allowing more cells to remain non-growing until the exposure had passed without changing the MIC (Fridman et al., 2014). Lag-time distributions therefore provide a practical readout of one route to tolerance, but prolonged lag is neither a universal tolerance mechanism nor a definition of persistence. Importantly, this pre-exposure lag should be distinguished from post-exposure recovery lag, which measures how rapidly drug-injured survivors resume growth after treatment. Adapting these measurements to Mtb is technically demanding because of its slow growth, prolonged assay duration, detection limits, and the possibility that colony-forming unit counts underestimate differentially culturable populations. Nevertheless, a most-probable-number-based MDK assay has been used to quantify rifampicin tolerance among clinical Mtb isolates, distinguishing the time required to eliminate the bulk population from the deeper exposure required to eliminate rare surviving subpopulations (Vijay et al., 2021). This work illustrates the portability of the kinetic framework while also emphasizing the need to specify how viability and recovery are detected.

Kinetic frameworks are successful because they converted ambiguous terminology into reproducible measurements. They separate altered susceptibility in the concentration dimension from altered survival in the time dimension and from heterogeneity in the distribution of killing responses. Although these measurements do not by themselves reveal the biological mechanisms underlying a kinetic phenotype, they provide the quantitative foundation for testing physiological, host-contextual, and pharmacological explanations.

3.2. Physiological-state models: dormancy, heterogeneity and dynamic vulnerability

Physiological-state models address a question that kinetic measurements leave open: which properties of the bacillus make it more or less vulnerable to killing under a defined drug exposure? The Yin–Yang model provided an influential starting point by organizing Mtb into growing, slow-growing and nongrowing populations that coexist and interconvert during infection and treatment (Zhang et al., 2012). Its principal contribution was to replace a rigid active–dormant dichotomy with a dynamic population structure linked to differential drug responsiveness. Replication status nevertheless remained the dominant organizing variable. Subsequent work has shown that growth arrest can accompany persistence, but is neither necessary nor sufficient to predict survival under antibiotic exposure.

Classical in vitro models established reproducible physiological states associated with reduced antibiotic killing or tolerance. Progressive oxygen depletion in the Wayne model drives Mtb through sequential stages of nonreplicating persistence, whereas nutrient starvation arrests growth, lowers respiration and reduces killing by several antibiotics (Betts et al., 2002; Wayne & Hayes, 1996). Nitric oxide similarly inhibits respiration and induces the DosR-regulated dormancy program (Voskuil et al., 2003). These models were foundational because they converted poorly defined “dormancy” into experimentally controlled physiological transitions. Their readouts, however, largely establish association between an imposed state and altered drug survival. DosR induction, reduced respiration or growth arrest should therefore be treated as markers of adaptation rather than as proof that a universal persister state has formed. Perturbational studies provide a stronger level of evidence: disruption of DosR-regulated functions compromises metabolic homeostasis and recovery from nonrespiring dormancy, demonstrating a functional role for the program while also showing that its contribution extends beyond survival during exposure itself (Leistikow et al., 2010). Starvation and hypoxia also redirect carbon and lipid metabolism. A multiple-stress model combining hypoxia, nutrient limitation, acidity and elevated carbon dioxide generates lipid-loaded, nonreplicating and drug-tolerant Mtb, but the simultaneous induction of several stresses makes it difficult to assign causality to any single metabolic feature (Deb et al., 2009). Genetic studies have provided more direct tests. Deletion of isocitrate lyase impaired long-term Mtb survival in activated macrophages and mice, establishing fatty-acid utilization as a determinant of long-term survival in host environments (McKinney et al., 2000). This result demonstrated causal involvement in chronic survival, but did not by itself establish a direct effect on antibiotic killing. A later study more specifically linked isocitrate lyase activity to broad tolerance under defined drug exposure, thereby connecting a metabolic pathway to the killing phenotype rather than to bacterial burden alone (Nandakumar et al., 2014). Reversible glpK phase variation provides another example: transient frameshift variants alter glycerol metabolism and generate a slowly growing, drug-tolerant phenotype that can rapidly revert, illustrating how a genetically encoded change can produce a non-stable physiological state rather than classical resistance (Safi et al., 2019).

The equation of persistence with nonreplication was further challenged by single-cell studies. Time-lapse imaging showed that M. smegmatis can persist during isoniazid exposure while continuing to divide, with population maintenance arising from a dynamic balance between division and death rather than from uniform dormancy (Wakamoto et al., 2013). In Mtb, host stress and immunity generate nongrowing but metabolically active subpopulations, indicating that replication and metabolic activity can become uncoupled (Manina et al., 2015). Redox heterogeneity provides a more direct mechanistic example in replicating Mtb. A reduced mycothiol redox state prospectively marked intracellular populations with increased drug tolerance, while genetic disruption of redox-responsive pathways and pharmacological alteration of phagosomal acidification increased antibiotic killing without producing an MIC-defined resistance phenotype (Mishra et al., 2019). This combination of a predictive reporter, pathway perturbation and altered killing provides stronger causal evidence than transcriptional enrichment alone.

Asymmetric growth adds a second source of dynamic vulnerability. Mycobacterial division produces daughters that differ in pole age, elongation rate and antibiotic tolerance (Aldridge et al., 2012). Live-cell imaging and modelling subsequently showed that growth-pole history and temporal growth features could prospectively identify M. smegmatis subpopulations with different rifampicin responses (Richardson et al., 2016). Stronger evidence came from perturbation of the divisome factor LamA: deletion reduced asymmetric polar growth, collapsed part of the population-level heterogeneity and altered antibiotic susceptibility, directly linking the generation of cellular diversity to drug response (Rego et al., 2017). Recent time-lapse imaging of Mtb has extended the descriptive resolution of this framework by mapping heterogeneous elongation, division and cell-cycle progression in the pathogen itself, although the extent to which each newly defined cell-cycle state predicts killing remains to be tested directly (Chung et al., 2024).

The evidence therefore supports a hierarchy of physiological claims. Growth arrest, DosR activation, lipid accumulation, altered cell size or a particular transcriptional signature may correlate with survival; pole age, redox state or defined growth dynamics may serve as predictors of differential killing; and genetic or pharmacological perturbations of isocitrate lyase, LamA, glpK-dependent metabolism or redox-control pathways provide causal evidence when they alter killing under controlled exposure. This hierarchy also refines the physiological continuum captured by the Yin–Yang model: growing, slow-growing, and nongrowing populations remain useful organizing states, but position along this continuum does not by itself establish killability (Zhang et al., 2012). The strongest demonstrations combine all three elements: a state measured before or during treatment, prospective prediction of survival, and a perturbation that changes the time–kill phenotype without the effect being attributable solely to a stable shift in MIC. Viewed in this way, physiological state is not a binary property or a synonym for persistence, but a dynamic and experimentally testable determinant of vulnerability whose significance must ultimately be judged by its effect on killing.

3.3. Infection-centered survival: macrophages and lesion niches

Physiological-state models often reproduce selected host stresses in broth. Infection-centered models shift the unit of analysis from the isolated bacillus to the host–pathogen interaction and ask whether the host niche occupied by Mtb actively generates the phenotype subsequently measured under antibiotic exposure. In this view, the niche is not merely where persisters reside. It supplies nutritional, redox and immune signals that remodel bacterial physiology, determines which drugs reach the bacillus, and may influence whether injured survivors remain suppressed or regain growth competence.

Macrophage residence provides some of the clearest causal evidence for niche-induced tolerance. Adams and colleagues showed that replicating intracellular mycobacteria developed multidrug tolerance within days of entering macrophages through induction of bacterial efflux activity, thereby separating host-induced tolerance from classical dormancy (Adams et al., 2011). The capacity to develop macrophage-induced rifampicin tolerance was subsequently observed across diverse clinical Mtb lineages, indicating that the phenomenon was not confined to a single laboratory strain (Adams et al., 2019). Host immune activation can also increase tolerance rather than simply restrict bacterial growth. Activation of infected macrophages with interferon-γ and lipopolysaccharide promoted drug tolerance in vitro and in vivo, with reactive nitrogen species making a major contribution to the phenotype (Liu et al., 2016). These studies provide perturbational evidence because altering bacterial efflux or host antimicrobial signalling changed survival under drug exposure, rather than merely correlating a host-cell state with residual bacterial burden.

Macrophages are themselves heterogeneous niches. Fluorescent bacterial reporters demonstrated that Mtb residing in alveolar macrophages experienced less stress and replicated more rapidly than bacilli in interstitial macrophages, linking bacterial physiology to host-cell metabolism and ontogeny (Huang et al., 2018). Dual RNA sequencing subsequently identified lineage-specific host and bacterial programs, including differences in nutritional restriction, while multimodal single-cell analysis resolved macrophage subsets within the infected lung that either permitted or constrained Mtb growth (Pisu et al. 2020, 2021). These studies directly establish niche-dependent bacterial physiology, but growth or stress reporters should not automatically be interpreted as drug-tolerance measurements. Unless antibiotic killing is tested in the same host-cell subset, macrophage identity remains a mechanistically informative correlate or candidate predictor rather than a determinant of survival during treatment.

Host metabolism provides a route from such association to causal intervention. Intracellular redox reporters showed that macrophage-associated Mtb populations occupy heterogeneous mycothiol redox states that predict differential antibiotic tolerance (Mishra et al., 2019). More recently, computational, genetic and pharmacological analyses connected macrophage NRF2-regulated metabolism to bacterial redox heterogeneity: meclizine-driven reprogramming of host bioenergetics reduced Mtb drug tolerance in macrophages and mice (Yadav et al., 2025). The intracellular niche can also alter effective drug availability. Correlative imaging in human macrophages showed that pyrazinamide accumulation and efficacy varied among individual bacilli and were greatest in acidified phagosomes, directly linking subcellular localization to both bacillary exposure and killing (Santucci et al., 2021). Primary human macrophage studies likewise showed that intracellular adaptation can change tolerance to several anti-TB drugs relative to extracellular culture (Lanni et al., 2023). Thus, even within a nominally uniform macrophage assay, host-cell metabolism and subcellular localization can modify both the bacterial state encountered by the drug and the amount of active compound that reaches it.

At the tissue scale, a granuloma is not a single niche but a spatial mosaic of macrophage-rich cellular regions, fibrotic boundaries, necrotic caseum and, in advanced disease, cavity-associated compartments. Nonhuman-primate granulomas contain spatially distinct macrophage populations expressing different nitric-oxide-synthase and arginase programs (Mattila et al., 2013), and individual granulomas within the same host can differ substantially in bacterial killing and sterilization outcome (Lin et al., 2014). Necrotic-lesion models further demonstrate that tissue architecture changes treatment response: C3HeB/FeJ mice develop cellular and caseous lesions with bacterial populations that differ from those in non-necrotizing mouse lungs (Driver et al., 2012). Ex vivo caseum contains largely nonreplicating bacilli that remain viable yet show extreme tolerance to many first- and second-line drugs (Sarathy et al., 2018). More recent lesion-resolved measurements showed that caseum, cellular lung tissue, airways and spleen harboured Mtb populations with different rates of rRNA synthesis and different responses to isoniazid, rifampicin and bedaquiline (Walter et al., 2023). These data establish that lesion architecture can generate distinct bacterial phenotypes and treatment responses, although they also expose a recurring interpretive problem: when bacterial physiology and local drug penetration vary together, an endpoint CFU difference cannot assign causality to either process alone.

Evidence from human macrophages and resected pulmonary material supports the relevance of these niche effects but remains less experimentally complete. Human macrophage studies demonstrate intracellularly conditioned changes in drug tolerance and heterogeneous drug accumulation, while analyses of alveolar macrophages obtained from patients suggest that drug-tolerant bacilli can adopt different intracellular survival patterns (Lanni et al., 2023; Santucci et al., 2021; Ufimtseva & Eremeeva, 2023). However, most human studies are cross-sectional and cannot follow the same bacillary population through treatment, injury and subsequent recovery. The evidence is therefore strongest that niches generate bacterial physiological diversity and modify effective drug exposure. Whether a particular macrophage or lesion compartment also determines post-antibiotic recovery remains less directly measured and should not be inferred from residual burden alone. An infection-centered view of survival therefore reframes the host environment as part of the survival mechanism, while also motivating the explicit separation of niche-induced physiology, pharmacological shielding and recovery competence in subsequent analyses.

3.4. Spatiotemporal pharmacology: nominal dose versus bacillus-level exposure

Spatiotemporal pharmacology asks a question that conventional persistence assays cannot answer: what drug exposure does the bacillus actually experience in its local niche? The prescribed dose and plasma pharmacokinetics are important, but they do not directly define exposure at the site of infection. Drugs must move from blood into lung tissue, across lesion compartments, and in some cases into avascular caseum before reaching bacilli. Each step is drug- and lesion-dependent. Imaging and lesion microdissection studies have made this distinction explicit. In resected human lungs, isoniazid, rifampicin, pyrazinamide and moxifloxacin showed markedly different distributions across uninvolved tissue, cellular lesions, cavity walls and necrotic foci. Rifampicin accumulated progressively in caseum during repeated dosing, whereas moxifloxacin penetrated necrotic centers less efficiently despite high concentrations in some lesion homogenates (Prideaux et al., 2015). Earlier imaging in infected rabbit lungs similarly showed preferential moxifloxacin accumulation in cellular granuloma regions but relative depletion from caseous cores (Prideaux et al., 2011). These observations established a central point: plasma or whole-lesion concentration can conceal the exposure experienced by the bacterial population of interest.

Even concentration within a lesion compartment does not necessarily equal pharmacologically available exposure. Caseum is avascular, and compounds reaching its core must diffuse through a matrix rich in lipids, proteins and cellular debris. Drug lipophilicity, nonspecific binding and intracellular sequestration therefore determine how much free compound remains available to reach bacilli. Caseum-binding assays showed that highly bound compounds can accumulate in surrounding cellular regions while leaving only a small unbound fraction available for diffusion into necrotic material (Sarathy et al. 2016, 2017). At the same time, exposure cannot be interpreted independently of bacterial physiology. Bacilli residing in caseum are largely nonreplicating and are substantially harder to kill than log-phase cultures. Ex vivo caseum studies demonstrated marked tolerance to many first- and second-line drugs, with rifamycins showing unusually strong sterilizing activity in this environment (Sarathy et al., 2018). Thus, adequate exposure is context dependent: the relevant concentration is not simply the amount of drug present, but whether that exposure is sufficient to kill bacilli in the physiological state found in that compartment.

Exposure also changes over time. Different drugs enter and leave lesion compartments at different rates, so the regimen experienced locally may differ from the regimen inferred from plasma measurements. Rifampicin, for example, can persist in necrotic lesions after concentrations have declined in plasma or uninvolved lung, whereas other drugs show different penetration and clearance profiles (Prideaux et al., 2015). Mechanistic models integrating plasma pharmacokinetics, lesion penetration and bacterial pharmacodynamics therefore predict that the effective composition of a multidrug regimen varies across both space and time (Strydom et al., 2019). This matters not only for sterilization but also for resistance selection. If one component penetrates poorly or falls below an effective concentration earlier than its partners, bacilli may transiently experience functional monotherapy. Spatial models show how such local exposure gaps can create opportunities for stepwise resistance selection (Moreno-Gamez et al., 2015), and direct measurements from resected human lungs have linked drug-penetration gradients across cavity compartments with sites at which acquired resistance can emerge (Dheda et al., 2018). These studies do not imply that heterogeneous exposure inevitably produces resistance, but they show that treatment pressure can differ substantially among bacterial populations within the same patient.

The key interpretive distinction is therefore between survival despite adequate exposure and survival because adequate exposure was not achieved. The former is consistent with a physiological tolerance phenotype: bacilli survive even when the local concentration–time profile is sufficient for the relevant State and Niche. The latter reflects pharmacological shielding, in which the administered regimen fails to generate adequate free exposure at the bacillus. In TB lesions these mechanisms may coexist, because compartments that restrict drug penetration can also impose physiological states with higher bactericidal requirements. Adequate exposure must therefore be defined against context-matched pharmacodynamics: State, Niche and Exposure are analytically separable, but they are not biologically independent.

3.5. Post-exposure recovery: resilience, resuscitation and therapeutic consequence

Most antimicrobial assays end when drug exposure ends, yet the therapeutic significance of residual bacilli is determined partly by what happens afterwards. Bacilli detected at the end of treatment may be irreversibly injured, temporarily unable to grow under standard culture conditions, or capable of resuming replication once drug pressure is removed. End-of-exposure survival and post-exposure recovery are therefore related but distinct phenotypes. Colony-forming unit (CFU) enumeration measures the ability to form colonies under specified conditions; it does not provide an absolute measure of viability or future recovery. Resuscitation-promoting factors (Rpfs) and culture-filtrate-supplemented limiting-dilution assays recovered Mtb populations from patient sputum that were poorly detected by routine plating, and the relative abundance of these populations could increase during chemotherapy (Chengalroyen et al., 2016; McAulay et al., 2018; Mukamolova et al., 2010). Antibiotic exposure itself can also generate delayed detectability. Rifamycin-treated Mtb may remain recoverable in permissive liquid culture while failing to form colonies efficiently on standard media, and subsequent work showed that moderate oxidative injury can delay replication until cellular damage is resolved (Saito et al. 2017, 2021). These studies establish an important principle: failure to form a colony at a conventional time point does not necessarily imply irreversible death. At the same time, differential culturability should not be equated automatically with dormancy, persistence, or relapse competence; it describes recovery under a defined set of culture conditions.

Recovery is therefore best treated as a measurable phenotype in its own right. At least three features are relevant: how many residual bacilli recover, how long recovery takes, and what conditions are required for recovery. This post-exposure recovery lag is biologically distinct from the pre-exposure lag discussed in kinetic tolerance models. A prolonged pre-exposure lag can protect cells by keeping them non-growing during an antibiotic pulse, whereas post-exposure lag reflects the time required for drug-injured survivors to restore productive growth. Regrowth after antitubercular treatment can depend strongly on drug history rather than simply on the growth rate before exposure, further supporting this distinction (Hendon-Dunn et al., 2019).

The genetic concept of antibiotic resilience provides particularly strong evidence that recovery can vary independently of conventional susceptibility and during-treatment survival. Clinical selection of variants in the transcriptional regulator resR was associated with treatment failure, and experimentally introduced variants shortened the time to post-antibiotic regrowth without substantially increasing MIC or markedly altering conventional time–kill survival (Liu et al., 2022). Thus, populations with similar susceptibility and similar end-of-exposure bacterial burden can differ in how rapidly residual cells return to growth.

These findings sharpen the interpretation of post-exposure Fate. A useful recovery assay should therefore distinguish residual abundance from regrowing fraction and recovery kinetics, with resuscitation-permissive conditions added where relevant. Resistance amplification lies one step further downstream: Recovery preserves a viable population on which continued or renewed selection may subsequently act, while resistance amplification represents a related evolutionary consequence that can also arise during uneven drug exposure (Levin-Reisman et al., 2017; Pepper-Tunick et al., 2026; Strydom et al., 2019). The central therapeutic question is therefore not simply whether bacilli remain after treatment, but whether drug-injured bacilli retain the capacity to return to population growth.

3.6. Where established frameworks fail to connect

Established frameworks have made mycobacterial survival increasingly measurable, but they interrogate different parts of the treatment response. Killing kinetics defines how survival differs, yet a prolonged MDK or biphasic survivor tail does not reveal why it occurs: similar patterns may arise from physiological heterogeneity, host-induced adaptation, inadequate lesion exposure, or combinations of these processes (Adams et al., 2011; Balaban et al., 2019; Brauner et al., 2016; Sarathy et al., 2018). Physiological measurements provide part of that explanation, but drug vulnerability is not an intrinsic property of a bacterial state. A nonreplicating, redox-shifted, or metabolically adapted bacillus may respond differently in broth, macrophages, cellular lesions, or caseum because the surrounding environment also changes nutrient availability, host stress, drug penetration, and the concentration–time profile experienced by the bacillus (Huang et al., 2018; Mishra et al., 2019; Prideaux et al., 2015; Walter et al., 2023). State can therefore be interpreted mechanistically only in relation to the niche and pharmacological context in which killing occurs. Nor does this interpretation end when drug exposure stops. Similar endpoint CFU counts may represent populations with different degrees of injury and markedly different capacities for recovery. Routine plating can miss differentially culturable bacilli, while populations with comparable end-of-exposure survival may differ in post-antibiotic lag, regrowing fraction, or dependence on resuscitation-promoting conditions (Liu et al., 2022; Mukamolova et al., 2010; Saito et al., 2021). Kinetics therefore tells us how killing differs; physiological and host-contextual measurements help explain why; pharmacology establishes whether effective exposure was achieved; and recovery determines what residual survival means after drug withdrawal. None can substitute mechanistically for the others. The problem is therefore not the absence of useful frameworks, but the limited interoperability of their outputs. Measurements that describe different parts of the same survival trajectory are too often interpreted as though they answered the same biological question.

4. Integrating survival phenotypes across state, niche, exposure and fate

The limited interoperability of existing frameworks motivates an interpretive framework that links survival measurements to the context in which they are obtained. We use State–Niche–Exposure–Fate (SNEF) to describe four dimensions that shape the interpretation of an Mtb survival phenotype (Fig. 1, Table 2). State denotes the physiological condition of the bacillus when drug exposure begins and how that condition changes during treatment. Niche describes the host-cell or anatomical environment that generates and constrains bacterial physiology while also shaping access to nutrients, immune stress and conditions for subsequent recovery. Exposure refers to the concentration–time profile and drug ratios experienced by the relevant bacillary population, rather than the administered dose or plasma concentration alone. Fate asks whether drug-injured bacilli are irreversibly eliminated, remain residual with their subsequent recovery unresolved, or retain the capacity to resume population growth after drug pressure is reduced or removed. Resistance amplification is treated separately as a related evolutionary consequence that may arise when viable bacilli remain under continued or renewed selection. These dimensions are analytically separable but biologically interdependent: for example, Niche can alter both bacterial State and local Exposure, while the consequences of a given Exposure depend on the State of the bacillus it encounters.

Fig. 1.

Cross-layer determinants of Mtb killability.

Host and lesion Niche shapes both the physiological State of Mycobacterium tuberculosis (Mtb) and the drug Exposure experienced by bacilli. State and Exposure together influence the observed killing phenotype, including killing rate, minimum duration for killing (MDK) and the emergence of a survivor tail. However, similar killing phenotypes can arise from different biological or pharmacological bottlenecks and therefore do not, by themselves, establish why bacilli survive. Post-exposure Fate further distinguishes irreversible elimination from residual bacilli with unresolved recovery and recovery-competent survival. Resistance amplification is shown as a conditional downstream consequence of viable recovery and subsequent selection and expansion. Thus, interpretation of Mtb survival requires linking the measured killing phenotype to its State, Niche, Exposure and subsequent Fate. CFU, colony-forming units.

Table 2.

How SNEF connects established approaches to mycobacterial survival.

Framework/approach	Primary question	Principal contribution	What remains unresolved	SNEF linkage
Kinetic resistance–tolerance–persistence frameworks	How does killing differ over time?	Distinguish conventional susceptibility, population-wide tolerance, and heterogeneous survival using MIC, time–kill analysis, and MDK.	Do not identify the biological or pharmacological source of the kinetic phenotype.	Provide the quantitative survival phenotype that is interpreted in relation to State, Niche, Exposure, and Fate.
Physiological-state models	Which bacterial states alter vulnerability to killing?	Link replication, metabolism, redox state, and cell-state heterogeneity to differential drug response.	State markers alone do not establish context-independent vulnerability or causality.	Relate State to the Niche and Exposure in which it was measured and to the subsequent Fate of treated bacilli.
Infection-centered survival models	How does the host environment shape bacterial survival during treatment?	Show that host-cell and lesion environments can modify bacterial physiology and treatment response.	May not distinguish altered State from altered bacillus-level Exposure or post-exposure recovery.	Connect Niche to State, local Exposure, and downstream Fate.
Spatiotemporal pharmacology	What exposure reaches bacilli, where and for how long?	Resolves spatial and temporal drug distribution, local concentration–time profiles, and regimen coverage.	Exposure alone does not establish bacterial vulnerability or what happens to residual bacilli after treatment.	Connects Exposure to the State and Niche of the bacillus and to its subsequent Fate.
Post-exposure recovery/resilience	What happens to residual bacilli after drug pressure is reduced or removed?	Distinguishes residual abundance from recovery competence, recovery kinetics, and resuscitation dependence.	The relationship of recovery phenotypes to relapse and treatment shortening remains incompletely validated.	Provides direct information about Fate and links the post-exposure outcome back to the preceding State, Niche, and Exposure.

Note: SNEF is not presented as a replacement for these approaches. Its purpose is to connect their non-equivalent outputs while preserving what each measurement can and cannot establish.

SNEF is not intended to replace MIC, MDK, time–kill curves, established definitions of tolerance and persistence, lesion pharmacology or recovery assays (Balaban et al., 2019; Brauner et al., 2016). Nor is it a universal assay that requires every study to resolve all four dimensions at high resolution. Its purpose is narrower: to specify the biological and pharmacological context in which a survival phenotype was observed and to constrain the conclusions that can be drawn from that observation. The survival phenotype itself—for example, a killing trajectory or survivor fraction—should therefore be distinguished from the evidence used to interpret it. For each SNEF dimension, the relevant variable may be directly measured, experimentally controlled, inferred, or unresolved. State may, for example, be measured with a validated physiological readout or constrained by an imposed culture condition; Niche may be experimentally defined while remaining heterogeneous within that model; Exposure is directly measured only when the relevant local concentration–time history is quantified; and Fate requires observation of post-exposure recovery under defined conditions. These categories are not rankings of study quality. Rather, they indicate how far a particular experimental design can support a mechanistic interpretation.

The practical value of this mapping is that it identifies the measurement needed to discriminate among competing explanations. A prolonged survivor tail observed without local exposure information cannot distinguish physiological tolerance from pharmacological underexposure; measuring or experimentally constraining bacillus-level Exposure would narrow that ambiguity. Comparable end-of-exposure bacterial burdens do not establish equivalent Fate and therefore require washout and recovery measurements when post-treatment competence is the question. Likewise, an association between a physiological marker and survival establishes neither prediction nor causation on its own: prospective tracking can test whether the marker predicts subsequent killing, whereas perturbation is required to determine whether it contributes causally to that phenotype. SNEF therefore adds value not by combining every variable into a single profile or metric, but by making explicit what was observed, under which State, Niche and Exposure, what happened afterwards, and which competing explanations remain unresolved.

5. From integrated evidence to experimental practice

5.1. Pyrazinamide as a worked example of context-dependent sterilizing activity

Pyrazinamide (PZA) provides a useful worked example because its established sterilizing and treatment-shortening contribution is poorly represented by conventional neutral-pH, log-phase susceptibility assays (Table 3). The aim is not to reconstruct a complete SNEF profile from heterogeneous studies, but to ask what the existing literature actually establishes about the conditions under which PZA becomes effective, and what remains unresolved. Early interpretations often linked PZA activity to dormant or slowly growing bacilli, but subsequent work showed that growth rate alone is insufficient. Under controlled acidic conditions, rapidly and slowly growing Mtb can display similar PZA susceptibility, while manipulation of PncA expression can alter drug activity independently of extracellular pH (Peterson et al., 2015; Pullan et al., 2016). Live-cell studies further showed that PZA perturbs intrabacterial pH homeostasis in a concentration- and time-dependent manner (Santucci et al., 2022). Together, these findings argue against a simple “dormant-cell drug” model. PZA activity instead depends on a combination of bacterial physiology, prodrug activation and local chemical conditions.

Table 3.

A cross-layer evidence audit of pyrazinamide activity.

Dimension	Directly measured	Experimentally controlled	Inferred	Unresolved
State	Growth rate; extracellular and intrabacterial pH; PncA-dependent activation/response; PZA-induced disruption of intrabacterial pH homeostasis (Peterson et al., 2015; Pullan et al., 2016; Santucci et al., 2022).	Acidic versus neutral culture conditions; experimental manipulation of PncA expression in selected studies.	PZA activity depends on physiological and chemical context and is not explained by growth rate alone.	Which pre-exposure State prospectively predicts irreversible elimination by PZA?
Niche	Macrophage subcellular localization, phagosomal acidification, lesion architecture, and local lesion pH (Blanc et al., 2018; Irwin et al., 2016; Lanoix, Ioerger, et al., 2016; Santucci et al., 2021).	Human macrophage systems and lesion-forming mouse models used as defined host contexts.	Niche composition contributes to the context-dependent activity of PZA by shaping the physicochemical environment encountered by bacilli.	How well do macrophage and lesion-forming models reproduce the range of PZA-relevant microenvironments in human disease?
Exposure	Plasma and lesion drug concentrations; single-bacillus intracellular drug accumulation (Lanoix, Ioerger, et al., 2016; Prideaux et al., 2015; Santucci et al., 2021).	Administered dose and extracellular drug concentration; tissue penetration and local pH separated experimentally in selected models.	Tissue penetration alone does not guarantee effective antibacterial exposure at the bacillus.	What concentration–time history is experienced by the bacilli that survive treatment and subsequently recover?
Fate	End-of-treatment CFU and relapse after treatment withdrawal are directly measured outcomes, but neither by itself resolves bacillus-level post-exposure Fate (Lanoix, Betoudji, et al., 2016).	Regimen composition and treatment withdrawal are controlled in relapse models.	Comparisons of regimens with and without PZA support a contribution of PZA to sterilizing activity and relapse prevention.	Does PZA alter post-washout recovery lag, regrowing fraction, or resuscitation dependence independently of its effect on end-of-treatment bacterial burden?

Note: Cross-layer analysis separates directly measured evidence from experimentally controlled context and inference, and highlights post-exposure recovery as one of the least directly resolved dimensions of PZA sterilizing activity.

The host environment provides a second layer of explanation. Within macrophages, PZA or its active metabolite accumulates heterogeneously among individual bacilli, with stronger accumulation and activity in acidified compartments. Bedaquiline can further enhance this effect through host-cell-dependent changes in the intracellular environment (Santucci et al., 2021). At the tissue level, lesion architecture also matters. In C3HeB/FeJ mice, PZA remained active in cellular lesions but showed substantially less activity against large extracellular populations in neutral, liquefied caseum; related studies linked this heterogeneity to lesion pH and inflammatory context (Blanc et al., 2018; Irwin et al., 2016; Lanoix, Ioerger, et al., 2016). These observations show that the Niche is not simply experimental background: it helps determine whether a nominal PZA exposure is converted into effective antibacterial injury.

This distinction also clarifies the meaning of Exposure. PZA can penetrate pulmonary lesions, yet tissue penetration alone does not guarantee activity. Increasing drug dose did not restore efficacy in lesions dominated by neutral caseum, consistent with substantial tissue exposure alone being insufficient to restore activity in that microenvironment (Lanoix, Ioerger, et al., 2016; Prideaux et al., 2015). Conversely, bacilli within the same macrophage population can experience different intracellular drug accumulation despite identical extracellular dosing (Santucci et al., 2021). Thus, “drug exposure” is not a single quantity: administered dose, lesion concentration and bacillus-level availability can diverge, and their biological effect depends on the local physiological context.

The therapeutically relevant consequence of this context dependence is seen most clearly in relapse models. Adding PZA to first-line combinations can shorten the duration of therapy required to prevent relapse, although the magnitude of this benefit varies with lesion composition (Lanoix, Betoudji, et al., 2016). These studies establish a sterilizing contribution that is not readily inferred from standard MIC-based testing. They do not, however, directly show whether PZA changes post-antibiotic recovery lag, the fraction of damaged bacilli that regrow, or the resuscitation requirements of individual survivors. That link between drug-induced injury and recovery remains incompletely resolved.

The PZA literature therefore illustrates the practical value of integrated survival analysis. The evidence does not support either of two simple conclusions—that PZA is merely weak in vitro or that it selectively kills a universal dormant population. Instead, its activity emerges from the interaction of bacterial physiology, local host conditions and bacillus-level exposure, while its effect on post-exposure recovery remains less directly measured. A major unresolved step is therefore to connect local PZA exposure and physiological context to time-resolved killing and subsequent recovery within the same defined population.

Rifampicin offers a useful contrast. It has strong conventional bactericidal activity, yet its killing can still vary with bacterial growth history, host-induced efflux and lesion penetration (Adams et al., 2011; Prideaux et al., 2015; Richardson et al., 2016). PZA and rifampicin therefore reach sterilizing activity through different combinations of physiological and pharmacological constraints, reinforcing the broader point that drug potency cannot be interpreted independently of the context in which killing occurs.

5.2. Implementing SNEF at different levels of experimental resolution

The worked example also illustrates why SNEF should not be operationalized as a fixed four-component assay. State, Niche, Exposure and Fate differ substantially in observability, technical burden and experimental scale, and requiring every study to resolve all four dimensions would make the framework difficult to reproduce and unnecessarily restrictive. A more practical use of SNEF is therefore progressive rather than comprehensive: first establish a reproducible survival phenotype, and then add only those measurements needed to distinguish the competing explanations relevant to the study.

Accessible core — establish what happened: A practical starting point is to establish that the survival phenotype is not explained by a stable shift in MIC, define the biological and pharmacological conditions of the assay, quantify killing over time, and determine whether residual bacilli recover after drug removal. Depending on the system, these measurements may include MIC or resistance genotype, serial colony-forming unit counts or another validated time-kill readout, and a defined post-washout regrowth measurement. This core establishes whether genetically susceptible bacilli are killed slowly, whether a survivor fraction remains, and whether recovery is observed under the conditions tested. It does not, by itself, identify why these phenotypes occur.

Context-resolving extension — distinguish competing explanations: When the core phenotype can be explained in more than one way, a targeted module can be added to resolve the relevant uncertainty. A physiological reporter may test whether altered bacterial State predicts killing; a macrophage or lesion-mimetic model may establish Niche dependence; intracellular or free-drug measurements may determine whether nominal exposure reflects bacillus-level Exposure; and resuscitation-permissive culture may reveal viable populations missed by standard recovery conditions. The appropriate extension is determined by the unresolved question, rather than by an expectation that every module should be included.

Integrated mechanistic analysis — resolve mechanism and cross-scale relevance: Single-cell profiling, spatial lesion analysis, local pharmacokinetic–pharmacodynamic measurements, CRISPR interference, chemical-genetic screening and evolutionary tracking can be used when the objective is to establish mechanism, spatial organization or downstream evolutionary consequence. These approaches are most informative when anchored to a previously defined kinetic or recovery phenotype—for example, by testing whether a perturbation changes the killing slope, survivor-tail magnitude, post-antibiotic recovery or the expansion of resistant variants.

These levels are illustrative rather than prescriptive and should not be interpreted as quality rankings. The rigor of a SNEF-informed analysis lies in matching the claim to the measurements performed and in stating explicitly which dimensions were measured, experimentally controlled, inferred or left unresolved. This scalable implementation preserves the accessibility of established quantitative measures such as MIC, time-kill kinetics and minimum duration for killing, while adding higher-resolution physiological, niche and pharmacological measurements only when they are required to resolve a specific ambiguity. SNEF therefore does not introduce a new universal metric; it provides a structured route for increasing mechanistic resolution without making maximal technical complexity a prerequisite for meaningful survival phenotyping.

5.3. Implications for drug and regimen design

The practical value of SNEF-informed phenotyping is that different causes of survival call for different therapeutic solutions. A survivor tail caused mainly by poor drug exposure should respond to improved penetration, concentration, or exposure duration. By contrast, if bacilli remain difficult to kill despite adequate local exposure, the more relevant strategy may be to alter the physiological state that protects them. The same survival readout can therefore point to very different interventions once the underlying bottleneck is defined.

This distinction also changes how individual drugs are valued within a regimen. A useful component does not simply add more in vitro potency; it should contribute activity where the rest of the regimen remains weak. One drug may rapidly reduce actively growing bacilli, another may retain activity against stress-adapted populations, and another may reach lesion compartments that are poorly covered by its partners. Some overlap between drugs is desirable because it can strengthen resistance suppression and buffer variation in pharmacokinetics. The aim is therefore not to assign one drug to one SNEF dimension, but to avoid leaving a major survival route insufficiently treated (Larkins-Ford et al., 2022).

Importantly, complementary activity observed in vitro must also occur in vivo. TB drugs differ markedly in how quickly they enter cellular lesions, cavity walls, and caseum and in how long they remain there. A combination that is fully active in plasma may therefore be incomplete in a particular lesion for part of the dosing interval (Prideaux et al., 2015; Strydom et al., 2019). Regimen design should consequently consider whether partner drugs reach the same bacillary population at effective concentrations for overlapping periods, rather than assuming that systemic co-administration guarantees local combination therapy. PZA and bedaquiline illustrate two different ways in which conventional in vitro measurements can underestimate a drug’s contribution to sterilization. For PZA, the limitation is primarily contextual: activity depends strongly on the physiological and chemical environment encountered by Mtb, including intracellular localization, local acidification and bacillus-level drug accumulation, so neutral-pH, log-phase assays poorly reproduce the conditions under which the drug is most effective (Santucci et al., 2021). Bedaquiline illustrates a different limitation. Inhibition of ATP synthase produces substantial metabolic remodeling before bactericidal activity becomes apparent, so short-term killing assays can underestimate a contribution that emerges with prolonged exposure (Koul et al., 2014). Thus, the two drugs highlight complementary dimensions of regimen activity: PZA shows why the relevant biological context matters, whereas bedaquiline shows why the relevant exposure timescale matters.

Finally, sterilizing activity may depend not only on how many bacilli are killed during treatment, but also on whether the survivors can recover afterwards. Two regimens can leave similar endpoint CFU yet differ in post-antibiotic lag or in the fraction of bacilli that eventually regrow. Likewise, an intervention that interferes with repair or metabolic restart may have little effect on the early killing slope but still reduce post-washout recovery (Liu et al., 2022; Mukamolova et al., 2010; Saito et al., 2021). Recovery measurements therefore provide information that MIC and endpoint bacterial burden cannot capture, although their clinical relevance still requires validation in relapse and treatment-shortening studies.

6. Limitations, standardization and translational scope

Measurement limitations. SNEF-informed phenotyping is constrained by the fact that its dimensions differ substantially in observability and experimental resolution. Bacterial State is usually inferred from selected reporters, transcriptional profiles, metabolic measurements, or growth behavior, none of which captures the full physiological condition of an individual bacillus. Niche can be defined experimentally in broth, macrophage, caseum-like, ex vivo, or animal models, but each reproduces only part of the spatial, immunological, and nutritional heterogeneity of human disease. Exposure is particularly difficult to resolve because the administered dose, plasma pharmacokinetics, lesion concentration, free-drug availability, and concentration experienced by an individual bacillus are not equivalent (Dartois, 2014; Prideaux et al., 2015). Fate is similarly conditional on the assay used: failure to recover bacilli by CFU enumeration does not establish irreversible death, whereas recovery in enriched or resuscitation-permissive culture may reveal viable populations missed under standard conditions (Mukamolova et al., 2010; Saito et al., 2021). Claims of “no recovery” must therefore specify the recovery conditions, observation period, and detection limit.

Implementation and standardization. A central challenge for SNEF-informed phenotyping is that increasing biological resolution often comes at the expense of experimental comparability. Killing kinetics and recovery are highly sensitive to how bacterial populations are prepared, exposed and subsequently cultured, while more physiologically realistic models introduce additional variation in host environment and local drug availability. Consequently, measurements such as killing slope, MDK, survivor-tail magnitude and post-antibiotic recovery cannot yet be assumed to be interchangeable across laboratories or experimental systems. This problem becomes more pronounced as studies move from controlled broth assays toward macrophages, lesion-mimetic systems, spatial pharmacology or single-cell analysis. Rather than seeking an immediate universal composite metric, an achievable first step is therefore to standardize the reporting of the biological context, drug exposure and recovery conditions in which each phenotype was obtained. The tiered strategy described above reflects this trade-off: broadly accessible kinetic and recovery measurements can provide a reproducible anchor, whereas additional physiological, spatial or pharmacological resolution should be introduced when it is needed to explain the phenotype rather than as a prerequisite for applying the framework.

Translational scope. These practical limitations also define the near-term translational role of SNEF. SNEF-informed phenotyping is not intended to replace clinical phenotypic drug-susceptibility testing. The two approaches address different questions. Clinical susceptibility testing asks whether an isolate is inhibited by clinically relevant drug concentrations and must therefore remain rapid, standardized and broadly deployable. SNEF instead asks why a genetically susceptible population may nevertheless remain incompletely sterilized in a particular biological and pharmacological context. The immediate test of SNEF is therefore not whether every coordinate can be measured, but whether cross-layer phenotyping identifies sterilization bottlenecks that conventional susceptibility and endpoint survival assays leave unresolved. Its value will lie in distinguishing bacilli that survive because they were insufficiently exposed, because their physiological state reduces killing, or because drug-injured cells retain an unusual capacity for recovery. Framed in this way, the goal is not to replace the persister concept with another label, but to make residual survival experimentally discriminable—and therefore therapeutically actionable.

CRediT authorship contribution statement

Yaqi Li: Writing – original draft. Jian Cheng: Writing – original draft. Wei Zhang: Validation. Junyi Li: Visualization. Yingying Pu: Writing – review & editing, Writing – original draft, Funding acquisition, Conceptualization.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.