Trained immunity (a durable, epigenetically encoded state of heightened responsiveness that innate immune cells acquire after an inflammatory exposure) and bystander T cell activation confer antigen-agnostic immune memory — changed immune-cell behavior that does not depend on antigen-specific receptors — but what maintains this memory over weeks to months, and whether its durability can be tuned, is unresolved. Mathematical modeling (Tripathi et al., 2025) shows that cell-intrinsic constraints — immune cell turnover, dependence on limited cytokines (secreted immune-signaling proteins), and the fidelity of epigenetic-state transmission during self-renewal and differentiation of hematopoietic (blood-forming) stem and progenitor cells — bound trained-immunity durability unless near-perfect transmission fidelity is assumed, which is biologically unlikely. The models predict that cytokine-mediated positive feedback (an interferon-γ (IFN-γ) loop in which IFN-γ secreted by memory T cells upregulates interleukin-15 (IL-15) availability in bystander activation memory, an interleukin-6 (IL-6) autocrine/paracrine loop in macrophages, or interleukin-1β (IL-1β)-driven epigenetic reprogramming of hematopoietic progenitors) can create a hysteretic circuit whose slow, feedback-strength-dependent relaxation to baseline yields durable, tunable memory. The open problem is to determine experimentally, in vivo, whether such circuit-level feedback mechanisms actually operate to maintain and tune antigen-agnostic memory durability, to quantify that tunability, and to establish whether mechanisms other than positive feedback contribute.

Background

Beyond antigen-specific memory of T and B cells, infection or vaccination can induce antigen-agnostic immune memory: changed behavior of immune cells that does not depend on recognizing a specific antigen. Two examples are bystander T cell activation memory (memory-like T cells maintained by homeostatic cytokines — secreted signaling proteins that support immune-cell survival — such as interleukin-15 (IL-15) after inflammatory challenge, with IFN-γ secreted by memory T cells upregulating IL-15 expression and presentation) and trained innate immunity, in which an inflammatory exposure such as vaccination with BCG (bacillus Calmette–Guérin, the live-attenuated tuberculosis vaccine) or Candida fungal infection leaves innate immune cells and their progenitors in an epigenetically altered, more responsive state. Trained immunity was consolidated as a concept by Netea, Quintin and van der Meer (2011), and Kaufmann et al. (2018) showed that BCG can 'educate' hematopoietic stem cells (HSCs, the self-renewing bone-marrow cells that generate all blood lineages) to produce protective innate immunity, establishing a central reservoir of trained immunity in hematopoietic stem and progenitor cells (HSPCs; the hierarchy includes HSCs, short-term HSCs (ST-HSCs), multipotent progenitors (MPPs), and branching myeloid progenitors — common myeloid progenitors (CMPs) and granulocyte–macrophage progenitors (GMPs)). The epigenetic memory state is a durable change in chromatin accessibility — how open the DNA packaging is at inflammatory-response genes — that is inherited by daughter cells during self-renewal and transmitted down differentiation to effectors such as monocytes and macrophages. What determines how long this state lasts (its durability), and whether that duration can be tuned, remained unclear. Tripathi, Liu, Lau, Sparks and Tsang (bioRxiv 2025, doi:10.1101/2025.07.18.663399) built ordinary-differential-equation models of bystander activation memory and trained immunity, instantiated with measured mouse hematopoietic self-renewal and differentiation rates, and analyzed ensembles of biologically plausible parameter sets with a machine-learning parameter-phenotype mapping (their 'MAPPA' approach, in which simulations over many parameter sets are mined by random-forest regression — an ensemble machine-learning method that ranks how strongly input parameters control an output phenotype). They found: (1) without feedback, durability is constrained by effector-cell turnover and cytokine availability, and centrally trained immunity is durable only if the probability pdivisionp_{division} that the epigenetic memory state is faithfully transmitted during progenitor self-renewal is very close to 1, if the probability pdifferentiationp_{differentiation} of preserving the state per differentiation step exceeds about 0.5, and if the state resides high in the hierarchy (HSCs, ST-HSCs, MPPs) — a regime the authors call biologically unlikely, since near-perfect epigenetic transmission is not expected; (2) adding positive feedback mediated by immune-cell-secreted cytokines — IFN-γ-driven upregulation of IL-15 availability in bystander activation memory; an interleukin-6 (IL-6) loop in which macrophages in the epigenetic memory state secrete more IL-6, and IL-6-driven activation of the transcription factor STAT3 in the same or neighboring macrophages maintains that open-chromatin state; or a loop in which monocytes with the altered state secrete more interleukin-1β (IL-1β), and IL-1β promotes epigenetic reprogramming of progenitors — produces a circuit with four dynamical regimes. In the physiological strong-feedback regime, an acute challenge triggers a switch-like excitable jump followed by return to baseline along a different path (hysteresis, the path-dependent behavior characteristic of positive-feedback systems), so memory decays slowly at a rate set by the feedback strength; durability becomes tunable and no longer requires near-perfect epigenetic transmission fidelity, because memory is now encoded in the state of the circuit rather than in individual cells. The paper's own limitations section states that the contribution of positive feedback to memory durability in vivo remains to be assessed experimentally and that mechanisms other than positive feedback cannot be ruled out.

Problem Statement

Determine, in vivo, whether cytokine-mediated positive-feedback circuits maintain and tune the durability of antigen-agnostic immune memory — both bystander T cell activation memory and trained immunity (peripheral and central) — and quantify how tunable that durability is. Concretely, building on the model and predictions of Tripathi et al. (bioRxiv 2025, doi:10.1101/2025.07.18.663399): (1) test whether blocking signaling of the hypothesized feedback cytokine (e.g., IFN-γ signaling inhibition for bystander activation memory, or IL-6 or IL-1β pathway inhibition for trained immunity) after the inciting inflammatory challenge has resolved abolishes or substantially weakens the corresponding antigen-agnostic memory — for trained immunity, the epigenetic memory state and functional hyper-responsiveness of myeloid effectors — as the feedback hypothesis predicts, while leaving induction intact; the inhibition must be withdrawn before the probing (second) challenge is introduced, so that the recall response is not assessed while the inhibitor is still on board, and because redundant cytokine circuits can compensate, a null result for a single blocked pathway does not by itself establish the absence of feedback-based maintenance; (2) establish whether durability depends on feedback strength (yielding the predicted hysteresis-like slow relaxation and excitable acute response) rather than only on cell-intrinsic epigenetic transmission fidelity pdivisionp_{division} and pdifferentiationp_{differentiation}, whose effective values in vivo the modeling framework can infer from measured durability time courses; and (3) determine whether mechanisms other than cytokine positive feedback contribute to tunable memory durability. These aims are not jointly mandatory for resolving the feedback-maintenance question: aims (1) and (2) together carry it — the hypothesis is supported if post-resolution blockade of the feedback cytokine weakens memory while leaving induction intact and if measured durability varies with feedback strength — whereas, because redundant cytokine circuits can compensate, a null result against any single pathway is uninformative about the absence of feedback, so the hypothesis is refuted only by insensitivity of durability to feedback-circuit operation across the relevant pathways. Aim (3) is a separate, open-ended completeness aim: it asks whether non-feedback mechanisms also contribute and may remain unresolved even after (1) and (2) are answered.

Operational conventions follow the source. Durability is measured as the duration, after inflammatory-challenge resolution, for which the memory readout remains elevated above its pre-challenge baseline: for peripheral trained immunity the source defines durability as the duration for which the fraction of tissue macrophages with an epigenetic memory state is at least 10% higher than the pre-challenge baseline, and for bystander activation memory as the duration for which memory T cell counts remain above baseline; the source's 'long-lasting' trained-immunity phenotype corresponds to memory durability greater than 180 days. Feedback strength is the model's positive-feedback parameter: in trained immunity, λEpiSig\lambda_{EpiSig}, defined as the maximum fold-change in the epigenetic reprogramming rate that the immune-cell-secreted feedback cytokine can induce, together with λSigEpi\lambda_{SigEpi}, the fold-increase in cytokine secretion by cells in the memory state; in bystander activation memory, the corresponding maximum fold-change in IL-15 availability that T cell-secreted IFN-γ can cause. For in-vivo assessment the source proposes longitudinal durability measurement — probing the memory response at multiple time points after the initial challenge, with different animals at each time point — with simultaneous tracking of circuit components (relevant cell counts and tissue IFN-γ, IL-6, and IL-1β levels) and their decay kinetics; whether these components decay at timescales similar to the memory response is the source's criterion for their role in maintenance, and provides the in-vivo counterpart of the model's feedback-strength parameter.

References

  1. Tripathi S, Liu C, Lau WW, Sparks R, Tsang JS (2025). Constraints and tunability of antigen-agnostic memory durability. bioRxiv preprint, posted 2025-07-19, doi:10.1101/2025.07.18.663399, https://doi.org/10.1101/2025.07.18.663399
  2. Netea MG, Quintin J, van der Meer JWM (2011). Trained immunity: a memory for innate host defense. Cell Host & Microbe 9:355-361, doi:10.1016/j.chom.2011.04.006, https://doi.org/10.1016/j.chom.2011.04.006
  3. Kaufmann E, Sanz J, Dunn JL, Khan N, Mendonça LE, Pacis A, Tzelepis F, Pernet E, Dumaine A, Grenier JC, Mailhot-Léonard F, Ahmed E, Belle J, Besla R, Mazer B, King IL, Nijnik A, Robbins CS, Barreiro LB, Divangahi M (2018). BCG Educates Hematopoietic Stem Cells to Generate Protective Innate Immunity against Tuberculosis. Cell 172:176-190.e19, doi:10.1016/j.cell.2017.12.031, https://doi.org/10.1016/j.cell.2017.12.031