The research focuses on nuclear factor kappa B, or NFκB, the master transcription factor that activates immune genes in macrophages. Decades of work have established that NFκB does not simply switch on and off. Instead, it shuttles in and out of the nucleus with distinctive temporal patterns — oscillatory for some stimuli, transient for others — and these dynamic signatures, sometimes called signaling codons, carry information about which ligand triggered the response. Six such codons have been characterized: activation speed, peak amplitude, duration of signaling, total integral activity, the balance of early versus late activity, and oscillatory content. Together, they form something like a Morse code of immunity, allowing the cell to distinguish, for example, a cytokine signal from a bacterial lipopolysaccharide encounter.

The problem with most previous studies, the authors note, is that they examined one ligand at a time. Nature rarely cooperates so neatly. During an E. coli infection, a macrophage may first encounter LPS from the bacterial envelope, then CpG DNA from the ruptured genome, and simultaneously inflammatory TNF produced by neighboring cells. In the gut, immune cells confront a rolling storm of microbial and host-derived signals. The combinatorial space of possible ligand mixtures is astronomically large, making comprehensive experimental exploration essentially impossible. The UCLA team’s solution was to build a virtual laboratory: a mechanistic mathematical model of NFκB signaling, comprising 52 ordinary differential equations, 101 reactions, and 133 parameters, organized into five receptor modules corresponding to the TNF, TLR2, TLR4, TLR9, and TLR3 pathways, all converging on a shared IKK–IκBα–NFκB core.

Crucially, the model was not trained on mixture data. It was fitted to live-cell imaging trajectories of hundreds of individual macrophages responding to each of five ligands — TNF, LPS, Pam3CSK4, CpG, and poly(I:C) — at multiple doses. To extend the model into combinatorial territory, the researchers developed an elegant statistical workaround. Because each simulated cell’s parameters had only been estimated for one receptor module plus the shared core, they imputed the missing receptor parameters using nearest-neighbor hot-deck imputation, anchoring each match on the core module parameters that all pathways share. In this way, the team generated virtual cell populations capable of responding to any ligand combination — thirty-one conditions in total, spanning single ligands through the full five-ligand cocktail.

The first validation was encouraging. Experimental responder fractions, measured by live-cell microscopy of fluorescently tagged RelA, rose with the number of ligands present and plateaued much as the model predicted, suggesting that signaling from multiple ligands generally combines in an integrative fashion and that non-responding cells reflect receptor-level heterogeneity rather than cellular ill-health. But one prediction failed in a telling way. The model expected the CpG plus poly(I:C) combination to produce a stronger response than either ligand alone; the experiments showed the opposite, with the pair dampening signaling below the CpG-only level. That discrepancy became the thread the researchers pulled to unravel a previously unknown mechanism.

Both CpG and poly(I:C) are sensed by Toll-like receptors located in endosomes, and both depend on the same endosomal transport machinery to reach those receptors. The original model had treated these transport processes as independent, but reality is stingier: the cell has a finite capacity for hauling cargo into endosomes. By reformulating the transport reactions with competitive Michaelis–Menten kinetics — adding no new parameters — the researchers captured the bottleneck. The result was a negative correlation between CpG–TLR9 and pIC–TLR3 signaling complexes: because poly(I:C) signals relatively weakly, it effectively poisons the stronger CpG response by saturating their shared transport pathway. The revised model accurately reproduced the non-integrative experimental result.

With the model refined, the team asked whether ligand identity remains readable within mixtures. Using Wasserstein distance analysis and machine learning classifiers trained on the six signaling codons, they found that specificity does degrade as ligands pile up — the five-ligand condition was hardest to classify — but it never vanished. Even the five-ligand mixture was recognized well above chance. More strikingly, binary classifiers asked a biologically natural question: is this ligand present in the mixture at all? For TNF, the answer could be read from NFκB dynamics with an ROC area under the curve of 0.94; LPS and Pam followed closely, and experimental data confirmed the pattern. Macrophages, it appears, can partially detect specific threats even in a molecular crowd.

To map synergy and antagonism systematically, the team simulated roughly 360,000 single-cell trajectories across all ten ligand pairs at six doses each. Synergy emerged mainly at low doses and traced back to ultrasensitive IKK activation: in cells with low receptor abundance and constrained TAK1 signaling, two weak inputs could combine to push the ultrasensitive IKK switch over threshold, amplifying NFκB peaks. Antagonism, by contrast, arose at high doses through resource competition. For the LPS–Pam pair, the shared co-receptor CD14 proved to be the limiting resource. Pam binds and releases CD14 more rapidly, peaking early and sequestering the co-receptor before LPS signaling can fully mature, thereby diverting CD14 away from productive TLR4 signaling and weakening the downstream NFκB response.

The most consequential discovery concerned what these interactions do for stimulus-response specificity. When the researchers compared a model lacking CpG–pIC antagonism with the updated model that includes it, the antagonistic version showed significantly greater separability between conditions in Wasserstein distance analyses. Multidimensional scaling visualizations made the point vividly: without competition, the CpG–pIC condition drifted into the cluster of other mixtures, becoming just another blended signal. With competition included, it stood apart, carrying a distinctive dynamic fingerprint. Antagonism, born of a mundane biochemical bottleneck, was actively manufacturing uniqueness. Synergy, though dramatic at the single-cell level, did not confer any similar benefit — it boosted peak responses without improving the distinguishability of the informative signaling codons.

The implications reach well beyond this particular pair of ligands. Resource-limited antagonism — competition for endosomal transport, for co-receptors like CD14, or for shared kinase modules — may represent a general design principle by which immune signaling networks preserve information under combinatorial overload. The computational workflow itself is a contribution: by combining mechanistic ordinary differential equation models with nonlinear mixed-effects parameter estimation and anchor-based statistical matching, the team demonstrated a path toward virtual cell models that retain single-cell heterogeneity while extending predictions far beyond their training data. With all code and data released openly, the framework is ready to absorb additional pathways such as AP1, p38, and the interferon axis, and to explore sequential stimulation over longer timescales. For now, the message is clear: in the immune system’s crowded molecular marketplace, a little rivalry among signals may be exactly what keeps the message intelligible.

Subject of Research: How signaling pathway antagonism shapes macrophage stimulus-response specificity to mixtures of immune ligands

Article Title: Macrophage response specificity to ligand mixtures is improved by signaling pathway antagonism

Article References: Guo, X., Sen, S., Gonzalez, J., & Hoffmann, A. (2026). Macrophage response specificity to ligand mixtures is improved by signaling pathway antagonism. Molecular Systems Biology. https://doi.org/10.1038/s44320-026-00230-9

Image Credits: AI Generated

DOI: 10.1038/s44320-026-00230-9

Keywords: macrophages, NF-kB, signaling dynamics, Toll-like receptors, stimulus-response specificity, ligand mixtures, mathematical modeling, systems biology, innate immunity, antagonism, synergy, single-cell imaging

The investigation, led by Yu Hasegawa of the Women’s Biomedical Research Institute at the Icahn School of Medicine at Mount Sinai, together with colleagues at the University of Maryland School of Medicine and other institutions, focused on bacterial extracellular vesicles, or bEVs. These are nanoscale, membrane-bound packages that bacteria shed continuously, carrying proteins, toxins, DNA, RNA, and fragments of cell-wall components. Far from being inert debris, bEVs act as messengers: they travel through the mucus and fluid lining the vaginal tract, dock onto epithelial and immune cells, and deliver molecular signals that can calm or inflame the host. In effect, they form a communication network layered on top of the microbiome itself, and it is this network that the new work shows is scrambled by antibiotic exposure.

The team worked with three species that define very different vaginal states. Lactobacillus crispatus dominates a healthy vaginal microbiome, classified as Community State Type I, and its presence is associated with low inflammation and protection against infection. By contrast, microbiomes dominated by anaerobes such as Gardnerella vaginalis and Mobiluncus mulieris, known as Community State Type IV, are linked to bacterial vaginosis and to serious downstream consequences, including increased susceptibility to sexually transmitted infections, infertility, and preterm birth. Understanding how these species signal to host tissue, and how treatment alters that signaling, is therefore central to explaining why bacterial vaginosis so often recurs after antibiotic therapy.

To probe this, the researchers grew each species in the laboratory and exposed cultures to three clinically relevant antibiotics, ampicillin, clindamycin, and metronidazole, at doses below the minimum inhibitory concentration, the threshold normally required to stop bacterial growth. This sub-minimal dosing was deliberate: in a real vaginal environment, antibiotic concentrations vary widely across time and space, and bacteria in protected niches frequently experience levels that stress rather than kill them. The team then isolated the vesicles released by treated and untreated cultures and quantified them using nanoparticle tracking analysis, a laser-based technique that counts and sizes individual nanoparticles in suspension.

The isolated vesicles were next applied to laboratory cultures of human epithelial and immune cells, and the inflammatory response was measured with cytokine profiling using multiplex immunoassays, which can quantify dozens of immune signaling molecules simultaneously. The researchers also performed Toll-like receptor pathway analyses. Toll-like receptors are pattern-recognition molecules on the surface of host cells that detect conserved microbial structures such as lipoproteins and flagellin, and they sit at the front line of innate immunity, deciding whether a bacterial signal is interpreted as benign or threatening.

The results were striking and highly specific. Antibiotic exposure significantly enhanced the inflammatory properties of the vesicles, but the effect depended on both the bacterial species and the antibiotic involved. Vesicles from antibiotic-exposed Gardnerella vaginalis cultures provoked elevated cytokine responses in host cells primarily through activation of TLR2, the receptor that recognizes bacterial lipoproteins. Vesicles from Mobiluncus mulieris, meanwhile, stimulated inflammation through both TLR2 and TLR5, the latter being the receptor dedicated to detecting flagellin, the protein that builds bacterial tails. These findings indicate that the drugs do not merely reduce bacterial numbers; they change the molecular cargo and immune visibility of the vesicles the surviving bacteria release.

Perhaps the most unexpected result concerned Lactobacillus crispatus. Vesicles from this beneficial species are typically non-inflammatory, consistent with its role in maintaining a calm vaginal environment. Yet when the cultures were exposed to metronidazole, the workhorse drug for bacterial vaginosis, even at a dose lower than what is used clinically, the resulting vesicles became immunostimulatory. In other words, a treatment intended to restore vaginal health may inadvertently push its most protective resident into producing inflammatory signals, an unintended proinflammatory consequence that could help explain why some patients feel worse before they feel better, and why the microbial ecosystem sometimes fails to settle back into a healthy state after therapy.

The study did not stop at laboratory cultures. Examining human vaginal swabs, the researchers detected bacterial extracellular vesicles directly in patient samples, including vesicles carrying vaginolysin, a toxin characteristic of Gardnerella. Notably, such vesicles were found even in swabs from women whose microbiomes were classified as Community State Type I, dominated by Lactobacillus. This indicates that low-abundance microbes, including potential pathogens, remain transcriptionally active even when they are too scarce to register in standard microbiome profiling, quietly releasing vesicles that continue to shape the local immune environment. The vesicle layer, in other words, preserves a memory of the microbial community that simple bacterial census methods miss.

Taken together, the findings suggest that antibiotics reshape bacterial communication in the vagina through their vesicles, potentially contributing to inflammation, disruption of the epithelial barrier that protects underlying tissue, persistent dysbiosis, and the recurrent infections that plague bacterial vaginosis patients. The authors argue that the work underscores the need for precision antimicrobial strategies that eliminate pathogens while preserving beneficial bacteria and their functional vesicles. Future therapies, they suggest, should account for the ecosystem-wide effects of antibiotics on the vaginal microbiome and on the vesicle-mediated signaling network that connects microbes to host immunity. Given the high recurrence rates of bacterial vaginosis and its links to preterm birth and infection risk, treating the vaginal microbiome as an ecosystem to be managed, rather than a pathogen to be eradicated, may prove a decisive shift in women’s health.

Subject of Research: Antibiotic-driven changes in the immune profiles of bacterial extracellular vesicles from vaginal anaerobes

Article Title: Exposure to antibiotics modifies the immune profiles of bacterial extracellular vesicles from common vaginal anaerobes

Article References: Hasegawa, Y., Swain, O., Rajpal, U., France, M., Ncube, L., Mogno, I., Zierden, H., Ravel, J., & Elovitz, M. A. (2026). Exposure to antibiotics modifies the immune profiles of bacterial extracellular vesicles from common vaginal anaerobes. Microbiome. https://doi.org/10.1186/s40168-026-02510-w

Image Credits: AI Generated

DOI: 10.1186/s40168-026-02510-w

Keywords: vaginal microbiome, bacterial extracellular vesicles, bacterial vaginosis, Gardnerella vaginalis, Lactobacillus crispatus, Mobiluncus mulieris, metronidazole, clindamycin, Toll-like receptors, inflammation, dysbiosis, preterm birth