Inflammation runs on a transcription factor called NF-κB, which sits caged in the cell's cytoplasm until danger signals release it to enter the nucleus and switch on inflammatory genes. Two facts about this system are now well established and, taken together, define an unsolved engineering problem. First, the timing of NF-κB's nuclear presence determines cell fate: an early, brief pulse (under an hour) delivers pro-survival genes, while sustained retention (hours) drives inflammatory death, as shown by injecting bacterial effector proteins into cancer cells (Re-engineering of TNF α -NF- κ B signalling dynamics in cancer cells using pathogenic E. coli effectors). Second, blanket suppression of the pathway is clinically lethal: in gram-negative septic shock, unconditioned corticosteroids raised 28-day mortality from 23% to 44% (Log-rank P = 0.028), with harm concentrated in patients whose baseline inflammation was low (Association between corticosteroid use and 28-day mortality in septic shock patients with gram-negative bacterial infection: a retrospective study). The pathway must therefore be suppressed conditionally and in a time-aware way, yet every available tool is either static (drugs, knockdowns) or biologically exotic (live bacterial needles).
This proposal builds a programmable nucleic-acid logic circuit that performs exactly that missing computation: it senses endogenous inflammatory transcripts, integrates their presence over time, and releases an NF-κB inhibitor only when the signal has persisted long enough to indicate pathological, sustained activation rather than a physiological pulse. Molecular logic gates made of DNA and RNA already execute Boolean reasoning with high dynamic range and single-molecule sensitivity, but only as equilibrium devices that report a color or classify a sample (Multi-arm RNA junctions encoding molecular logic unconstrained by input sequence for versatile cell-free diagnostics; Deep Molecular Programming: A Natural Implementation of Binary-Weight ReLU Neural Networks). None of them discriminates signal duration, and none actuates a therapeutic output inside a mammalian cell. We propose to close both gaps at once.
Core Claim
A synthetic nucleic-acid circuit can implement duration-gated suppression of NF-κB: it stays silent during a transient inflammatory pulse, and fires an inhibitory payload only after the inflammatory input has been sustained past a programmable time threshold. Formally, with input concentration $u(t)$ (an inflammatory mRNA, sensed by the circuit) and output $y(t)$ (the released inhibitor), the circuit computes
$$y(t) = 1 \iff \int_0^t u(\tau)\,d\tau > \theta,$$
where $\theta$ is a threshold set by circuit kinetics — a molecular leaky integrator. The integral form is what a threshold cascade with slow turnover naturally implements in a strand-displacement architecture (a DNA computation scheme where one strand binds and peels another off a partially complementary duplex), and it is exactly the computation that rate-independent designs deliberately discard: Deep Molecular Programming proves its chemical networks converge to the same steady state regardless of kinetic rates, which is robust but structurally blind to timing. We reintroduce rate dependence as a feature, not a bug.
This differs from the closest existing work in three ways. Multi-arm RNA junctions provides our sensor architecture — loop-initiated RNA activators that detect a target RNA regardless of its sequence — but its gates are static AND/OR devices whose output is a reporter enzyme in cell-free paper strips and E. coli; the authors explicitly leave mammalian endogenous-gene regulation as untested future work, and their gates cannot distinguish a one-hour input from a six-hour input. Re-engineering of TNF α -NF- κ B signalling dynamics proves the biology that motivates us — early versus late nuclear NF-κB have opposite consequences — but achieves temporal control with natural virulence proteins delivered by a live bacterial type III secretion system, with the intentionally cytotoxic goal of killing 95% of cancer cells. We use de novo programmable nucleic-acid kinetics, delivered as nucleic acids, with the opposite goal: preserving viable tissue. Actin networks modulate heterogenous NF-κB dynamics in response to TNFα supplies the measurement platform and a caution: it maps an endogenous F-actin–dependent feedback loop (through NUAK2 and ARHGAP31, two of 236 strictly RELA-dependent genes) that already gates translocation in single cells, and it shows that cell-to-cell heterogeneity is large — so a useful circuit must gate per cell, not per population.
Research Program
The build has three modules. The sensing module rewrites the loop-initiated RNA activators of Ma et al. so that the input strands are endogenous cytokine transcripts — TNF-α and IL-6 mRNAs — rather than viral RNA; the multi-arm junction architecture was chosen precisely because it decouples the sensed input sequence from the output, letting us retarget it without redesigning the downstream cascade. The timing module is the new ingredient: a cascade of strand-displacement reactions with deliberately slow, tuned turnover, in which a fraction of the intermediate species decays back per unit time, so that only sustained input drives the accumulated intermediate past the firing threshold. The actuator module releases a double-stranded decoy oligonucleotide carrying NF-κB binding sites that sequesters free p65 (RelA), an approach that is reversible — unlike the NleC protease route, which digests RelA outright and fails when the cell needs a baseline pool of it (Re-engineering of TNF α -NF-κ B signalling dynamics).
Design and tuning happen first in chemical-reaction-network simulation using ReaDISH, sweeping kinetic rates to place the duration threshold in the biologically motivated window: silent for pulses under one hour, firing for retention beyond four to six hours, matching the early/late boundary established in Butera et al.. We will then test in mammalian cell-free lysate, and finally in living cells, using two assay systems from the literature: BV2 microglia challenged with LPS at 1 µg/mL (the protocol of Aucubin promoted neuron functional recovery by suppressing inflammation and neuronal apoptosis in a spinal cord injury model) and the CRISPR-tagged RELA-GFP pancreatic cell lines with automated confocal imaging over ten hours, the platform that generated the ~130,000-cell single-cell atlas in Butera et al.. Delivery will use lipid nanoparticles, since the circuit is entirely nucleic-acid.
Baselines
The naive reference point is the untreated, LPS-challenged cell: no circuit, no drug. This defines the untreated inflammatory tone (RELA nuclear fraction, TNF-α and IL-6 secretion, iNOS and COX-2 expression) against which any suppression is measured; for the in vivo extension, the untreated spinal-cord-injury rat reaches a mean Basso, Beattie and Bresnahan locomotor score of approximately 6 at day 28 (Xiao et al.). No reported result found on Lacuna for quantitative cytokine output in LPS-stimulated BV2 cells, so we will establish those numbers ourselves in the naive arm.
The strongest existing competitor is blanket static suppression with a corticosteroid (dexamethasone), because it is the clinical standard and the most potent readily available NF-κB brake; its documented failure mode is exactly what we claim to fix, since steroids raised 28-day mortality to 44% versus 23% without them, and severe kidney injury to 78% versus 58% (P = 0.003), with harm concentrated in the low-inflammation stratum (Neutrophil-to-Lymphocyte Ratio < 20) (Dong et al.). To keep the comparison fair, dexamethasone dose will be titrated to match the circuit's peak suppression of TNF-α secretion in the same cells, same timeline, and the circuit will receive no more design-tuning iterations than the dose optimization receives conditions. The circuit beats this baseline if it matches late-phase cytokine suppression while preserving the early pro-survival transcriptional program and a host-defense readout (LPS-induced interferon-stimulated genes, and phagocytic clearance of heat-killed bacteria) that the static drug suppresses.
The key ablation is the same circuit without the delay line: a static two-input AND gate, built in the multi-arm junction architecture of Ma et al., sensing the same TNF-α and IL-6 mRNAs and releasing the same decoy, but firing immediately at threshold concentration. This isolates duration discrimination as the sole difference. For gate engineering quality we hold ourselves to the published benchmark of that architecture: ON/OFF ratios above 50-fold (achieved by 16 of 24 designs there, with a best of roughly 350-fold), and if we include a compiled chemical-network controller variant as an exploratory arm, its multi-input state-classification accuracy should be judged against the 95.20% that a rate-independent chemical network reached on human viral-expression profiles in simulation (Vasic et al.).
Why not just…?
Why not just dose steroids only to patients with high inflammation? This is the most serious alternative, and Dong et al. genuinely supports it at the population level: the harm they measured was confined to patients below Neutrophil-to-Lymphocyte Ratio 20. But bedside biomarker gating is population-level and static, while the single-cell atlas of Butera et al. shows NF-κB dynamics vary enormously cell to cell within one tissue, driven by cytoskeletal state rather than anything a plasma biomarker can see. A patient can be both hyper-inflammatory overall and harbor large cell populations mid-pulse that blanket dosing would wrongly silence. Our single-cell imaging experiment — measuring whether the duration-gated circuit spares transient-RELA cells while suppressing sustained-RELA cells in the same field — is precisely the test that would settle whether population gating leaves measurable damage on the table; if per-cell and population gating give indistinguishable outcomes, this alternative wins and we will say so.
Why not just knock down RelA with siRNA? Unconditional silencing cannot tell a pulse from a siege. Butera et al. showed the first hour of nuclear NF-κB carries anti-apoptotic, pro-survival transcription; a static knockdown removes that along with everything else, which is why their NleC protease (which destroys RelA indiscriminately) was counterproductive in combination therapy. The head-to-head arm of our study — siRelA versus duration-gated circuit, both challenged with a brief TNF-α pulse followed by sustained LPS — directly measures the viability cost of erasing the early pulse. We expect siRelA to lose viability in the pulse condition; if it does not, the duration-gating premise is wrong and the project should stop.
Why not just use the bacterial effectors that already work? Tir and NleE achieved the most impressive temporal rewiring in the literature, eliminating 95% of target cells — but that is the point against them: the mechanism is caspase-8-mediated cell death, delivered by a live enteropathogenic needle (Butera et al.), the polar opposite of preserving viable parenchyma in sepsis or reperfusion injury, and imaging revealed resistant high-baseline-NF-κB subpopulations that a threshold-free protein cannot accommodate. Our decoy payload is reversible and our threshold is tunable per design, which is exactly how we plan to handle those resistant subpopulations rather than destroy the shared signaling intermediate.
Proposed Experiments and Decisive Checks
The program proceeds in four gated stages. Stage one is simulation: encode the leaky-integrator cascade in ReaDISH and sweep kinetic parameters under synthetic input profiles (a 30-minute pulse, a 1-hour pulse, a 6-hour plateau), requiring the circuit to stay off for the pulses and fire by hour four of the plateau, with a leak rate below what would cause false firing during physiological oscillation. Stage two is cell-free mammalian lysate: measure ON/OFF ratio for the retargeted cytokine sensors against the >50-fold benchmark of Ma et al., and verify duration discrimination against the same input time courses. Stage three is the cellular decision experiment in BV2 microglia and RELA-GFP lines: four arms (untreated, dexamethasone, static AND gate, duration-gated circuit), read out as RELA nuclear-translocation trajectories per cell over ten hours of imaging, secreted TNF-α and IL-6, iNOS and COX-2 expression, IL-10, viability, and the host-defense surrogates. The decisive check is a two-by-two of input regime (pulse versus sustained) by treatment: the duration-gated circuit should suppress only in the sustained column, while dexamethasone and the static gate suppress in both. Stage four, only if stage three succeeds, extends to the neonatal-rat-cardiomyocyte hypoxia/reoxygenation model (two hours hypoxia, four hours reoxygenation) of C1q/TNF‑related protein 9 decreases cardiomyocyte hypoxia/reoxygenation‑induced inflammation by inhibiting the TLR4/MyD88/NF‑κB signaling pathway, measuring LDH, CK and CK-MB release and cell viability; no reported result found on Lacuna for quantitative viability values in that system, so we will establish matched untreated controls ourselves. The single result that would falsify the core claim is stage three showing that per-cell duration gating confers no advantage over the static AND gate on any viability or host-defense endpoint — in that case the timing biology of Butera et al. does not translate into a therapeutic window addressable by molecular circuits, and the finding would itself be worth reporting.
References
- Re-engineering of TNF α -NF- κ B signalling dynamics in cancer cells using pathogenic E. coli effectors
- Association between corticosteroid use and 28-day mortality in septic shock patients with gram-negative bacterial infection: a retrospective study
- Multi-arm RNA junctions encoding molecular logic unconstrained by input sequence for versatile cell-free diagnostics
- Deep Molecular Programming: A Natural Implementation of Binary-Weight ReLU Neural Networks
- Actin networks modulate heterogenous NF-κB dynamics in response to TNFα
- Aucubin promoted neuron functional recovery by suppressing inflammation and neuronal apoptosis in a spinal cord injury model
- C1q/TNF‑related protein 9 decreases cardiomyocyte hypoxia/reoxygenation‑induced inflammation by inhibiting the TLR4/MyD88/NF‑κB signaling pathway