A growing family of antimicrobials kills bacteria by physically wrecking the lipid membrane — cationic peptides that carpet and perforate the bilayer, polymer coatings that collapse it on contact, graphene nanosheets that cooperatively extract lipid molecules until the envelope tears open. Because a bacterium cannot cheaply rebuild an entire negatively charged envelope, these agents impose a high evolutionary barrier to resistance. Yet almost everything we know about how well they work comes from 60-year-old static endpoint assays: grow bacteria in broth overnight, read turbidity, count colonies. The consequences are documented and bleak. Over 400 antimicrobial peptides have entered preclinical pipelines and only about 15% of those reaching clinical trials succeed, and the standard assay medium can misreport potency by more than 20-fold relative to physiological infection-like conditions Antimicrobial Susceptibility Testing of Antimicrobial Peptides to Better Predict Efficacy. We rank candidates by a single overnight number while the actual death process — minutes of depolarization, transient metabolic scrambling, energetic collapse — goes unwatched.
This proposal watches it. We will couple two real-time optical readouts on the same live bacterial population during a membrane-disrupting attack. The first is extracellular flux analysis (EFA), in which oxygen and pH sensors lower into a sealed micro-chamber above a cell monolayer and read respiration (oxygen consumption rate, OCR) and acidification (extracellular acidification rate, ECAR) minute by minute. The second is a dual-luciferase reporter pair, in which firefly luciferase — an enzyme whose light output strictly requires intracellular ATP — is deliberately repurposed as an ATP gauge, calibrated against an ATP-independent Renilla control. Our core hypothesis is that the ordered sequence of bioenergetic events during membrane attack, together with whether stress-response genes manage to turn on at all, predicts bacterial killing and regrowth substantially better than the minimum inhibitory concentration (MIC), and transfers across media where MIC does not.
Core Claim
Two claims, one methodological and one scientific.
The methodological claim resolves a known artifact by exploiting it. Firefly luminescence falls when ATP is depleted, so under a membrane disruptor a dropping firefly signal can mean either promoter repression or simple energy starvation — a documented confound in every dual-luciferase study of metabolic stress. We invert this: run one reporter configuration in which both luciferases are constitutively expressed, so the firefly/Renilla ratio becomes a pure intracellular ATP clock, and a second configuration in which an envelope-stress promoter drives firefly. Dividing one by the other gives an ATP-corrected stress-transcription trajectory:
$$A_{\text{stress}}(t) = \frac{F_{\text{stress}}(t)/R_{\text{stress}}(t)}{F_{\text{const}}(t)/R_{\text{const}}(t)}$$
where $F$ and $R$ are firefly and Renilla luminescence, the subscript "stress" denotes the stress-promoter configuration and "const" the constitutive ATP-clock configuration, both measured at the same disruptor dose and time $t$. If $A_{\text{stress}}$ stays flat while raw firefly signal craters, the peptide is starving the cell, not repressing the promoter.
The scientific claim is that kill kinetics — the time from dosing to respiratory collapse, the magnitude of any last-ditch acidification burst, the rate of ATP collapse, and the failure or survival of stress transcription — form a signature that predicts 24-hour log-kill and post-washout regrowth better than MIC, including in bicarbonate-containing physiological medium where MIC shifts by more than 20-fold Antimicrobial Susceptibility Testing of Antimicrobial Peptides to Better Predict Efficacy.
Research Program
The closest existing work stops short on all three sides, and we build directly on each of its edges. Strickertsson et al. placed live bacteria and host cells together in a Seahorse XF24 chamber and showed host mitochondrial respiration falling 2- to 4-fold over 8–24 hours of Enterococcus faecalis infection, with reactive oxygen species surging within 30 minutes at a multiplicity of infection of 50 Enterococcus faecalis Infection Causes Inflammation, Intracellular Oxphos-Independent ROS Production, and DNA Damage in Human Gastric Cancer Cells — but they measured host metabolism, used static microarrays for transcription, and never applied a physical disruptor. Li et al. established the full EFA stress-test dissection (oligomycin to isolate proton leak, FCCP to reveal maximal capacity, rotenone plus antimycin A to leave only non-mitochondrial respiration) in lung epithelium under cigarette smoke stress Role of mitochondrial fusion proteins MFN2 and OPA1 on lung cellular senescence in chronic obstructive pulmonary disease — but tested no antimicrobial and no dynamic reporter. And the deepAMP foundation model designed membrane-lytic peptides that matched FDA-approved antibiotics and resisted resistance development over 25 days of serial passage A Foundation Model Identifies Broad-Spectrum Antimicrobial Peptides against Drug-Resistant Bacterial Infection — but validated killing only with endpoint MIC, colony counts, and fixed electron microscopy. Dai et al. came closest to kinetics, showing with the membrane-potential dye DiSC2(5) that Pentocin ZFM94 permeabilizes cells within 3 minutes A Novel Bacteriocin From Lactobacillus Pentosus ZFM94 and Its Antibacterial Mode of Action — a single variable at a single timescale. No reported result found on Lacuna for real-time OCR or ECAR kinetics in bacteria under a membrane-disrupting agent; that measurement is the center of this program.
Concretely, we will build three modules. First, bacterial EFA: EFA platforms require firmly attached monolayers, and bacteria are non-adherent, so we will compare immobilization matrices (poly-D-lysine, capture plates, and a Cell-Tak-like coating) against dye-exclusion and CFU controls for baseline stress, then dose immobilized S. aureus, E. coli, and P. aeruginosa through the instrument's injection ports with a panel spanning modes of physical disruption: Pentocin ZFM94 (target-independent pore formation, MIC 1.75 μM against Micrococcus luteus, 2.00 μM against S. aureus D48, 2.50 μM against E. coli DH5α), the deepAMP Tier 1 and Tier 2 peptide set, graphene nanosheets as a mechanical lipid-extraction mode following the cooperative-extraction mechanism characterized by Guangxin Duan et al., and ciprofloxacin as a non-membrane control. Second, the reporter module: the two dual-luciferase configurations described above, with envelope-stress promoters driving firefly in the stress configuration. Third, a host collateral arm applying the Li et al. mitochondrial stress test and an HSF1 heat-shock luciferase reporter of the kind used in the homoharringtonine screen The small-molecule drug homoharringtonine targets HSF1 to suppress pancreatic cancer progression to mammalian cells at hemolytic-subthreshold peptide doses. Analysis will fit a small set of kill-kinetic parameters per agent and, where residual flux needs pathway interpretation, draw on open bacterial flux-analysis tooling such as WUFlux WUFlux: an open-source platform for 13C metabolic flux analysis of bacterial metabolism.
Baselines
The naive reference point is MIC-only prediction: standard broth microdilution in cation-adjusted Mueller-Hinton broth, with the MIC value alone used to rank expected outcome. This is what the field effectively does today, and it is the null we must beat — kinetic features must add predictive power over MIC on the same peptides, same strains, same labs, or the program's scientific claim fails. We will run it on the identical panel with matched replication.
The strongest existing competitor is the static endpoint pipeline exemplified by deepAMP: MIC determination, CFU time-kill plating, and fixed imaging A Foundation Model Identifies Broad-Spectrum Antimicrobial Peptides against Drug-Resistant Bacterial Infection. That paper reports that more than 90% of its 29 synthesized candidates beat penetratin in inhibition, that lead candidate T2-9 matched FDA-approved antibiotics, and that candidates cleared P. aeruginosa in a mouse wound model at levofloxacin-comparable levels — the current high-water mark for designed membrane disruptors. We will run this full static pipeline alongside our kinetic assay on every agent, with the same media, incubation, and analysis effort, so the only difference is what information each predictor uses. Beating it means achieving higher rank correlation with 24-hour log-kill and with post-washout regrowth, especially in bicarbonate-containing physiological medium where the AST literature documents MIC shifts exceeding 20-fold Antimicrobial Susceptibility Testing of Antimicrobial Peptides to Better Predict Efficacy.
The strongest real-time single-variable competitor is the DiSC2(5) depolarization assay, which detected Pentocin's membrane action within 3 minutes A Novel Bacteriocin From Lactobacillus Pentosus ZFM94 and Its Antibacterial Mode of Action. We will run it in parallel at matched doses; it reads only membrane potential, and our claim is specifically that multi-variable kinetics (respiration, acidification, ATP, transcription) carries outcome information that a single dye cannot. The decisive ablation isolates our key ingredient: a firefly-only reporter arm with no Renilla ratio and no ATP correction, scored on the same outcome predictions. If uncorrected firefly signal performs as well as the ATP-corrected trajectory, the artifact-resolution contribution collapses; a second ablation, reporters without EFA, tests whether the flux traces add anything beyond the reporter module alone.
Why not just…?
Why not just lyse cells and measure ATP with a commercial kit? The endpoint firefly-luciferase ATP assay is exactly the static snapshot the flux literature was built to replace Jarosiewicz et al.. Lysing at one or two timepoints cannot order the events — whether proton leak precedes respiratory shutdown, whether an acidification burst precedes ATP collapse — and cannot distinguish uncoupling from ATP-synthase inhibition, which the oligomycin/FCCP stress test separates Role of mitochondrial fusion proteins MFN2 and OPA1 on lung cellular senescence in chronic obstructive pulmonary disease. Strickertsson et al. showed respiration can fall 2- to 4-fold while non-mitochondrial pathways stay elevated Enterococcus faecalis Infection Causes Inflammation, Intracellular Oxphos-Independent ROS Production, and DNA Damage in Human Gastric Cancer Cells; a lysed ATP number collapses that structure into one point. This alternative is cheaper, and we will include it as a secondary readout — but our claims are about ordering and prediction, which it cannot test.
Why not just do RT-qPCR or RNA-seq on the stressed cells? Steady-state mRNA conflates synthesis with degradation: the Arid4a study shows transcript levels shifting through altered mRNA stability rather than transcription Arid4a Suppresses Breast Tumor Metastasis by Enhancing MTSS1 Expression via mRNA Stability, and Strickertsson et al.'s microarray approach captured only the net 8–24 hour outcome. Our question is whether stress promoters initiate within minutes of membrane attack, under dosing where cell counts and RNA recovery are themselves shifting — precisely the setting where the Renilla internal control earns its keep The small-molecule drug homoharringtonine targets HSF1 to suppress pancreatic cancer progression. A dense kinetic titration series by RNA-seq is also cost-prohibitive. We will, however, validate a subset of reporter trajectories with RT-qPCR as a sanity check, since the reporters are episomal and do not fully recapitulate chromatin context.
Why not just stick with MIC broth microdilution, the universal standard? Because the standard is documented to mislead for this exact agent class: cationic peptides adsorb to plastic labware, Mueller-Hinton salt suppresses electrostatic attack, and physiological bicarbonate shifts potency by more than 20-fold, contributing to an ~85% clinical failure rate among peptides reaching trials Antimicrobial Susceptibility Testing of Antimicrobial Peptides to Better Predict Efficacy. The QMAP benchmark compounds the worry: even machine-learned MIC predictors, which look strong under random splits, approach random performance on high-potency peptides (MIC below 10 μM) under homology-aware evaluation QMAP: A Benchmark for Standardized Evaluation of Antimicrobial Peptide MIC and Hemolytic Activity Regression. Our answer to the ranking-signal problem is not another predictor trained on the same weak labels but a different, physical measurement. This alternative is nonetheless the real null hypothesis, which is why MIC-only is our first baseline and the decisive check is stated against it.
Proposed Experiments and Decisive Checks
Experiment 1 establishes bacterial EFA kill traces. Immobilized S. aureus and E. coli receive Pentocin ZFM94 at 0.25× to 4× its reported MIC (2.00 and 2.50 μM respectively), with parallel DiSC2(5) depolarization reading. Expected signature, from the 3-minute leakage already reported: depolarization within minutes, a transient ECAR burst as cells try to rebuild membrane potential, then OCR collapse. Ciprofloxacin should show none of these on this timescale.
Experiment 2 runs the two reporter configurations at matched doses, with ciprofloxacin as the crucial control: it should leave the ATP clock stable while genuinely inducing stress promoters, whereas membrane disruptors should crater the ATP clock — and the corrected $A_{\text{stress}}(t)$ tells us whether envelope-stress transcription initiates before or after ATP collapse. An oxygen-clamped calibration guards the one real confound: both luciferases need oxygen, so we verify that Renilla output is insensitive to the modest oxygen swings in the transient micro-chamber.
Experiment 3 is the media transfer test. Each agent is profiled in standard medium versus bicarbonate-buffered physiological medium; the hypothesis is that MIC shifts dramatically (as the >20-fold report predicts) while the kill-kinetic signature at matched free peptide concentration is conserved — kinetics transfer where MIC does not.
Experiment 4 profiles host collateral toxicity: A549 cells under the full mitochondrial stress test Role of mitochondrial fusion proteins MFN2 and OPA1 on lung cellular senescence in chronic obstructive pulmonary disease with an HSF1 reporter The small-molecule drug homoharringtonine targets HSF1 to suppress pancreatic cancer progression, predicting that sublytic cationic peptides raise proton leak and depress spare respiratory capacity before any hemolysis is detectable — giving the field a sensitive early toxicity readout that static HC50 endpoints, which QMAP shows are poorly predicted by current models QMAP, currently lack.
Experiment 5 is the decisive check. Kinetic parameters are fit on a training subset of the 29 deepAMP peptides and used to predict 24-hour log-kill and post-washout regrowth on held-out peptides, head-to-head with MIC-only and with the full static pipeline. If kinetic features do not beat MIC-only prediction on the held-out set, the scientific claim fails even if the measurement stands — and we would report it as such, since the methodological contribution (the ATP-corrected reporter, validated against known artifacts like the transcriptional-repression-versus-ATP-starvation ambiguity documented across dual-luciferase studies) would survive independently.
References
- Antimicrobial Susceptibility Testing of Antimicrobial Peptides to Better Predict Efficacy
- Enterococcus faecalis Infection Causes Inflammation, Intracellular Oxphos-Independent ROS Production, and DNA Damage in Human Gastric Cancer Cells
- Role of mitochondrial fusion proteins MFN2 and OPA1 on lung cellular senescence in chronic obstructive pulmonary disease
- A Foundation Model Identifies Broad-Spectrum Antimicrobial Peptides against Drug-Resistant Bacterial Infection
- A Novel Bacteriocin From Lactobacillus Pentosus ZFM94 and Its Antibacterial Mode of Action
- Graphene-Induced Pore Formation on Cell Membranes
- The small-molecule drug homoharringtonine targets HSF1 to suppress pancreatic cancer progression
- WUFlux: an open-source platform for 13C metabolic flux analysis of bacterial metabolism
- The Impact of Glyphosate, Its Metabolites and Impurities on Viability, ATP Level and Morphological changes in Human Peripheral Blood Mononuclear Cells
- Arid4a Suppresses Breast Tumor Metastasis by Enhancing MTSS1 Expression via <scp>mRNA</scp> Stability
- QMAP: A Benchmark for Standardized Evaluation of Antimicrobial Peptide MIC and Hemolytic Activity Regression