c-Rel is a thermo-responsive knob that tunes fibroblast NF-κB signaling

Summary

Thermal regulation is a mechanism for warm-blooded animals to maintain homeostasis against ambient temperature stress. Also, core body temperature rises with fever, an integral part of host responses to infection. Although responses to heat or cold shock have been described, little is understood about how physiological temperature variations affect inflammatory responses. We tracked real-time responses of NF-κB RelA and c-Rel transcription factors to inflammatory and pyrogenic cytokines TNF-α or IL-1β using endogenous knock-in reporter fibroblasts in temperatures ranging from 24°C to 40°C. Quantitative live-cell imaging showed that c-Rel signaling, but not RelA, increases with rising temperature. Computational modeling and transcriptomic analysis indicate that the divergent signaling behaviors of these canonical NF-κB subunits cannot be explained solely by accelerated biochemical reactions in higher temperatures. Rather, cytokine receptor-proximal adaptors are likely to amplify c-Rel signaling for precise analog behaviors. The temperature-sensitivity of fibroblast c-Rel has implications for inflammatory cascades in elevated temperatures.

Graphical abstract

Highlights

• •Higher temperatures enhance fibroblast NF-κB signaling via induction of the c-Rel subunit
• •Most genes induced by TNF-α or IL-1β are also regulated by temperature
• •Thermoregulation of receptor-proximal genes supports temperature-sensitivity of c-Rel

Immunology; Cell biology; Systems biology

Introduction

Organisms have evolved to adapt to changing environmental temperature conditions within a tolerable range. Even under a fixed temperature, cells of different tissues in our body are maintained at different temperatures, for example, the superficial skin layer versus deep tissues such as the liver. In addition, the core body temperature fluctuates over circadian sleep-wake cycles¹ and gradually shifts downward in human aging.² In short, biology occurs in a wide range of temperatures. Yet, little data are available regarding how temperature affects various cellular processes, since most cell biological studies are conducted at 37°C.

The immune system operates across normal and febrile temperatures (37°C–41°C). Fever is considered to provide an evolutionarily conserved advantage by facilitating immune cell trafficking and cell-cell communications, which contribute to enhanced coordination of innate and adaptive immune responses to infection.³ Moreover, febrile-range temperatures may also promote homeostatic mechanisms during the resolution phase of inflammatory responses.⁴ On the other hand, cold stress has been reported to shift the immune system toward an immuno-suppressive state.^5,6 Despite the evidence of thermoregulated immune responses from animal studies, the impact of various temperatures on cell-intrinsic immune signaling and gene regulation is poorly understood for most cell types.

The skin is a barrier tissue exposed to the seasonal changes of the environment. Fibroblasts are a major component of the dermis layer of skin, which has an extensive crosstalk with other cells such as macrophages. For example, fibroblasts and macrophages can cross-regulate their growth and tissue composition through PDGF and M-CSF.⁷ Fibroblasts also produce chemokines that alert neighboring cells, which are primarily regulated by NF-κB, a major transcription factor ubiquitously involved in many immune responses. To understand how skin fibroblasts sense cytokines under different physiological temperatures, we examined thermoregulation of NF-κB signaling in ear fibroblasts responding to TNF-α and IL-1β at temperatures ranging from ambient room temperature to a febrile temperature. Using sensitive endogenous reporters and quantitative live-cell analysis, we found that the expression of the canonical c-Rel subunit of NF-κB depends on temperature, with a monotonic increase of c-Rel protein expression as temperature rises. Through computational modeling and transcriptomic analysis, we show that c-Rel-mediated thermoregulation of NF-κB is linked to cytokine receptor-proximal signaling activities. These results suggest that temperature acts as an important contextual cue shaping inflammatory signaling in skin fibroblasts.

Results

Temperature alters single-cell NF-κB dynamics in response to cytokine stimulation

To understand how temperature influences the dynamic behavior of canonical NF-κB dimers, we established a real-time live-cell imaging system capable of tracking the nuclear translocation of endogenous RelA and c-Rel across a wide temperature range (24°C–40°C). We generated immortalized ear fibroblasts from our NF-κB double-knockin (DKI) reporter mouse line,⁸ in which endogenous RelA and c-Rel are fluorescently tagged with mEGFP and mScarlet,⁹ respectively, allowing simultaneous visualization of both subunits within the same cell (Figure 1A). Before imaging, cells were pre-equilibrated for 2 h at each target incubation temperature (24°C, 34°C, 37°C, or 40°C) to ensure thermal stability. We then performed time-lapse microscopy to capture cytoplasmic to nuclear translocation of RelA and c-Rel in response to stimulation with the pro-inflammatory cytokines tumor necrosis factor-α (TNF-α) or interleukin-1β (IL-1β). Both RelA and c-Rel displayed strong and rapid nuclear translocation within the first ∼30 min after TNF-α or IL-1β treatment at 37°C, as expected (Figure 1B). However, this initial nuclear translocation was diminished at temperatures deviating from 37°C (Figure S1).

Figure 1.

Live-cell imaging of NF-κB RelA and c-Rel responses to cytokines in fibroblasts under various temperatures

(A) Experimental workflow for dual-color NF-κB live-cell imaging. Immortalized mouse ear fibroblasts were isolated and generated from double knock-in mice expressing mEGFP-RelA and mScarlet-c-Rel. Live-cell time lapse imaging was performed at the indicated temperature (24°C, 34°C, 37°C, and 40°C), following treatment with TNF-α or IL-1β.

(B) Representative live-cell fluorescence images of RelA (green) and c-Rel (magenta) after stimulation with TNF-α or IL-1β at 37°C. Scale bars, 100 μm.

(C) Single-cell trajectories of mean nuclear mEGFP-RelA and mScarlet-c-Rel intensities (arbitrary units) under different temperature conditions. Cells were stimulated at time = 0 with either TNF-α (upper block) or IL-1β (lower block). Columns correspond to incubation temperatures (24°C, 34°C, 37°C, and 40°C). Within each ligand block, the upper row shows nuclear RelA trajectories and the lower row shows nuclear c-Rel trajectories. Each figure displays single-cell nuclear fluorescence traces (gray lines). A subset of randomly selected cells is highlighted in color; matching colors in the RelA and c-Rel panels denote the same individual cells.

To resolve these dynamic patterns with greater precision, we quantified single-cell nuclear intensity trajectories for RelA and c-Rel following TNF-α or IL-1β stimulation across all four temperatures. Automated cell tracking and fluorescence quantification were performed using custom-written MATLAB code. Across all temperature conditions, RelA activation was characterized by a rapid initial nuclear surge followed by damped oscillations (Figure 1C). c-Rel displayed a similar but substantially weaker oscillatory behavior, resulting in a more muted dynamic profile. RelA signaling dynamics displayed subtle ligand-specific differences; for example, IL-1β produced a deeper trough between the first and second peaks compared with TNF-α.

Previous studies have identified the period of oscillation as a temperature-sensitive feature of NF-κB RelA dynamics.^10,11 In agreement with this, the period of RelA oscillation, defined as the time interval between the first and second nuclear peaks, became progressively shorter as temperature increased (Figure 2A), both for TNF-α stimulation (mean period: 24°C:187 min, 34°C:123 min, 37°C:110 min, 40°C:107 min) and for IL-1β (24°C: 242 min, 34°C: 156 min, 37°C: 131 min, 40°C: 114 min). In contrast to RelA, c-Rel oscillatory behavior did not show temperature-dependent acceleration and was largely temperature-insensitive (Figure 2A).

Figure 2.

Temperature alters c-Rel expression and RelA signaling responses to cytokines

(A) Violin plots of single-cell oscillation periods for RelA and c-Rel in response to TNF-α (left block) or IL-1β (right block) at 24°C, 34°C, 37°C, and 40°C.

(B) Violin plots showing the single-cell distributions of basal mean cellular intensity (a.u.) at indicated temperatures.

(C) Time series of mean cellular RelA and c-Rel intensities (a.u.) in response to real-time temperature increases during live-cell imaging. Imaging started after 4 h of equilibration at 29°C. At 2 and 4 h into the imaging session, the incubation temperature was raised to 34°C and 40°C, respectively. Colored curves indicate intensity trends averaged over the cell population.

(D) Violin plots showing the single-cell distributions of amplitude (mean nuclear intensity in a.u.) in response to TNF-α (left block) and IL-1β (right block) at indicated temperatures. Dashed lines connect medians across temperatures. Statistical comparisons between temperatures were performed using the Mann-Whitney test. ∗p < 0.05, ∗∗p < 0.01, and ∗∗∗p < 0.0001; n.s., not significant; a.u., arbitrary unit.

c-Rel is thermo-induced in higher temperatures

In addition to the period of oscillation, we extracted a panel of NF-κB signaling features from single-cell trajectories, including mean cellular intensity, peak amplitude, time to peak, area under the curve (AUC), number of cycles (NOC), duration, slope of the first rise, and attenuation rate (Figures 2 and S2A). Unexpectedly, the basal expression of c-Rel increased progressively as temperature rose, whereas basal cellular RelA intensity remained relatively stable across the physiological range (34°C–40°C) with a decrease only under the lowest temperature (24°C) (Figure 2B). We confirmed that the selective increase of basal c-Rel expression is an immediate response by raising the temperature in real time (Figure 2C). This thermoregulated basal expression of c-Rel resulted in similar patterns for peak amplitude and AUC, features defined in the same scale as the expression level (Figures 2D and S2B). Other signaling features also showed temperature-dependency, such as an increase in the number of c-Rel oscillatory cycles and a decrease in the c-Rel response duration with rising temperature for both stimuli (Figure S2B). These results suggest that c-Rel has a more graded thermal dependency than RelA.

The coupling between RelA and c-Rel is strongest at 37°C

A key advantage of our dual-reporter system is the ability to monitor RelA and c-Rel signaling simultaneously within the same cell, allowing direct comparison of their activation behaviors under identical intracellular conditions. Leveraging this, we asked how the RelA and c-Rel dimer activity patterns are interlinked under various temperatures, by calculating pairwise correlations between RelA and c-Rel signaling features defined in Figure S2A (Figure S3). For both TNF-α and IL-1β stimulations, the time to peak showed the weakest correlation between the two subunits at 37°C, while peak amplitudes and AUC of RelA and c-Rel, as well as a few cross-feature pairs, were most strongly correlated at 37°C (Figure S3, green-boxed entries). This suggests that, while the timing of nuclear translocation of RelA and c-Rel is not coordinated, the physiological temperature promotes maximal coordination of RelA and c-Rel dimers in terms of signaling strength. Given canonical NF-κB proteins exist as dimers of two subunits from RelA, c-Rel, and p50, the temperature sensitivity of the coupling between RelA and c-Rel signaling may have an impact on the subunit availability/exchange, dimer formation, and relative abundance of these dimers in the nucleus at non-physiological temperatures.

Computational models with temperature-dependent biochemical reactions do not recapitulate observed NF-κB signaling dynamics

The experimental findings above raised the question of whether acceleration of biochemical reactions with higher temperatures is sufficient to explain the observed changes in NF-κB response dynamics. To address this, we employed a deterministic mathematical model of the NF-κB signaling network to mechanistically examine how temperature modulates the dynamics of RelA and c-Rel dimers. We used a previously established model of TNF-α-induced RelA and c-Rel activation.¹² To simulate NF-κB signaling induced by IL-1β, we modified the receptor-proximal portion of the TNF-α model by incorporating the MyD88-IRAK-TRAF6 (MIT) complex, enabling comparative analysis of both cytokine pathways within a unified modeling framework (Figure 3A).

Figure 3.

Receptor-proximal components may play a role in subunit-specific thermoregulation of NF-κB signaling

(A) Schematic overview of TNF-α and IL-1β signaling modules that converge on the IKK-IκB module driving the nuclear translocation of RelA and c-Rel under different temperatures (24°C, 34°C, 37°C, and 40°C). The receptor-proximal module (orange) feeds into TAK1, which activates IKK and the downstream module (green). Illustration created with Biorender.

(B) Model simulations of single-cell nuclear NF-κB trajectories using Arrhenius temperature scaling. For each stimulus (upper: TNF-α; lower: IL-1β), paired panels show RelA and c-Rel time courses. Colored traces represent simulated time series at 24°C, 34°C, 37°C, and 40°C, as indicated in the legend.

(C and D) Partial rank correlation coefficient (PRCC) based global sensitivity analysis of NF-κB model parameters. Bar plots show Δ PRCC = |PRCC(c-Rel)|-|PRCC(RelA)|, sorted by magnitude. Parameters with stronger influence on c-Rel than RelA are shown, with receptor-proximal parameters in orange and IKK downstream parameters in green. (C) Sensitivity of total cellular concentration in the basal state (unstimulated). (D) Sensitivity of stimulus-induced AUC for TNF-α (left) and IL-1β (right) stimulation.

To model temperature-dependence of reaction rates, all rate parameters were adjusted using the Arrhenius relation, as previously described for temperature-dependent computational models of NF-κB signaling.¹⁰ Because the speed of oscillation is particularly sensitive to temperature variations, this feature provides a stringent benchmark for evaluating temperature-dependent models. When Arrhenius scaling was applied to our RelA and c-Rel signaling framework, the model predicted a progressive shortening of the period of stimulus-induced oscillations as temperature increased (Figure 3B). This recapitulated the experimentally observed compression of RelA oscillations with rising temperature (Figure 2A).

However, the Arrhenius-scaled model did not recapitulate experimental features of c-Rel responses which were progressively stronger at higher temperatures, in terms of basal expression, peak amplitude, and AUC (Figures 2B, 2D, and S2B). Instead, the Arrhenius-based model predicted decreasing AUC values and relatively stable peak amplitudes with rising temperature for both RelA and c-Rel (Figure S4A). These discrepancies expose an important limitation of the Arrhenius-scaled model and suggest that additional mechanisms must underlie the pronounced enhancement of c-Rel responses observed experimentally at elevated temperatures.

Global sensitivity analysis suggests receptor-proximal determinants of c-Rel dynamics

To gain insight into potential molecular components in the NF-κB network that might drive the selective enhancement of c-Rel activity at elevated temperatures, a global sensitivity analysis (GSA) was applied to the mathematical model of NF-κB signaling to identify parameter values influencing c-Rel signaling behavior.^13,14 Latin hypercube sampling (LHS) was combined with partial rank correlation coefficient (PRCC) analysis for efficient sampling of high-dimensional parameter space and identifying parameters with selective effects on c-Rel versus RelA features.¹⁵

Since live-cell imaging showed that basal c-Rel expression rose steadily with increasing temperature while RelA remained relatively stable range (Figure 2B), we examined the parameters producing the greatest predicted differential between RelA and c-Rel protein abundance using ΔPRCC (see “STAR Methods”). IκBε-related reactions had a dominant impact in c-Rel specific temperature dependence, with eight of the fifteen highest delta PRCC values corresponding to IκBε-related processes (Figure 3C). This is expected because IκBε preferentially inhibits c-Rel.¹⁶ In addition, parameters for receptor-proximal reactions, including “inactivation of complexed TNFR,” “activation of TAK1 by C1” (C1 represents TNFR in complex with TRADD, TRAF2, and RIP1), and “capture of IL-1,” also ranked among the top determinants of differential basal abundance of RelA and c-Rel, suggesting that variation in receptor abundance can disproportionately influence c-Rel levels compared with RelA (orange bars in Figure 3C).

When a post-stimulation dynamic feature, AUC, was used as model output in GSA and PRCC, receptor-proximal parameters had a more prominent influence on differential RelA versus c-Rel behaviors, especially for TNF-α stimulation (Figure 3D). For IL-1β stimulation, receptor-proximal factors were not as dominant as for the TNF-α, but they still ranked among the top 10 parameters (Figure 3D). A similar pattern was observed when peak amplitude was used as model output, where receptor-proximal reactions accounted for a larger fraction of the top 10 ranked parameters compared with IκB-related parameters (Figure S4B). Taken together, these findings indicate that, although IκB-mediated mechanisms remain a core determinant of NF-κB network behavior, receptor-proximal signaling may exert an outsized impact on distinct c-Rel versus RelA regulation. This shift in control architecture provides a potential mechanistic explanation for the enhanced temperature sensitivity of c-Rel observed in live-cell experiments.

RNA-seq shows thermoregulation of genes encoding receptor-proximal signaling mediators

To gain a global insight into temperature-dependent basal expression and target gene regulation, RNA-seq was performed following TNF-α or IL-1β response time courses. Gene expression was compared to the reference value at temperature 37°C, that is, as a ratio, allowing us to focus on temperature-specific deviations from the reference (Figure 4). Unsupervised clustering of 4,584 “temperature-DEGs” (genes differentially expressed in comparison to 37°C for the same ligand and time point) with a K-medoids algorithm revealed distinct temperature-dependent expression patterns. Gene ontology (GO) enrichment analysis identified biologically coherent pathways within these clusters (Figure S5). For example, “cold-induced” clusters were enriched for adipogenic pathways, including fat cell differentiation, consistent with cold-adaptive responses. This cluster includes Bard1 and Brca2 (involved in DNA repair), along with Cebpa and Cebpb (transcriptional mediators of adipocyte differentiation). Meanwhile, the “extreme-induced” cluster showed a strong upregulation at 40°C, repression at 34°C, and a mild variable induction at 24°C. This cluster prominently featured canonical heat shock genes such as Hsp90aa1 and Hspa1a, indicative of acute heat stress responses to the temperature equilibration (Figure 4, right).

Figure 4.

Transcriptomic responses to cytokines are thermoregulated

Bulk RNA-seq identified 4,584 DEGs (|log₂ FC| ≥ 0.5, FDR < 0.05) regulated by temperature. The heatmap displays log₂ [(FPKM at the indicated temperature)/(mean FPKM at 37°C for the matched ligand condition)], isolating temperature-driven effects. Genes were grouped into seven clusters by silhouette-guided clustering with a robust K-means algorithm. Selected NF-κB pathway genes linked to model parameters (Figure 3) are annotated at right (orange, receptor-proximal; green, IKK downstream; black, other).

Notably, the majority of cytokine-responsive genes were also temperature-sensitive (Figures S6A and S6B, File S1). Specifically, 84.4% of TNF-α/IL-1β-regulated DEGs, or “cytokine-DEGs,” overlapped with temperature-DEGs (Figure S6B). The functional ontology categories that are enriched in cytokine-upregulated genes were largely shared across the temperatures, except those genes from 24°C showed no enrichment (Figure S6C). In contrast, we found a number of DNA motifs matching specific transcription factor binding sequences were enriched in a temperature-dependent manner at promoter regions of the cytokine-induced DEGs (Figure S6D). Only motifs related to NF-κB or IRF3 binding were shared across the temperatures.

Given the extensive overlap between temperature-sensitive and cytokine-responsive transcriptional programs, we next examined whether temperature-DEGs included genes that may directly regulate the canonical NF-κB signaling network, based on a comprehensive list of annotated NF-κB-related genes (https://www.bu.edu/nf-kb/gene-resources/target-genes, https://geneglobe.qiagen.com/us/knowledge/pathways/nf-kappab-signaling). Moreover, we sought evidence for mechanisms suggested by modeling that may underlie the divergent thermoregulated behaviors of c-Rel from RelA, and whether temperature variations perturb the expression of receptor-proximal genes. Indeed, genes in the core NF-κB regulatory module (e.g., Rel, Nfkbia, and Tnfaip3) (green gene symbols, Figure 4), as well as genes in receptor-proximal signaling modules (orange gene symbols, Figure 4), were found among temperature-DEGs (Figure S7). The basal expression of Rel (encoding c-Rel), but not Rela (encoding RelA), increased monotonically with temperature (Figure S7), consistent with our imaging analysis (Figures 2B and 2C). Several receptor-proximal regulators within the TNF pathway, including Traf2 and Traf5 which function as scaffolds that transmit TNFR1 signals to TAK1 (orange gene symbols, Figure 4), were repressed under cold conditions. Tnfrsf1b, which encodes a high affinity receptor TNFR2 for TNF-α, also belongs to a cold-repressed cluster. Within the IL-1β pathway, receptor-proximal genes such as Irak3, a negative regulator of MyD88-dependent signaling,¹⁷ and Il1rn, encoding the IL-1 receptor antagonist, were more strongly downregulated at elevated temperatures, with the most pronounced suppression at 40°C. Notably, all of these transcriptional changes in the receptor-proximal module for either ligand were in the direction that supports a stronger signaling in higher temperatures, which our model predicted to increase c-Rel signaling with rising temperature.

Taken together, the RNA-seq data are consistent with our mathematical modeling and GSA, which pointed to receptor-proximal modules as potential contributors to temperature-sensitive differences in RelA and c-Rel signaling. While this does not establish mechanistic causation, the results support the possibility that temperature-dependent transcriptional modulation of receptor-proximal machinery may participate in shaping NF-κB pathway responses under different thermal environments.

Refining receptor-proximal parameters improves predictions of temperature-dependent models

Guided by the RNA-seq results and the sensitivity analysis, the mathematical model was next refined by selectively adjusting parameters. We sought to obtain a minimal set of adjustments needed to produce a stronger temperature dependence of c-Rel compared with RelA. The PRCC-based sensitivity analysis implicated both receptor-proximal pathway nodes and the downstream IκB-NF-κB module (Figures 3C, 3D, and S4B). The sensitivity analysis showed that several receptor-proximal reactions, including “TNFR synthesis,” “activation of complexed TNFR,” “activation of TAK1 by C1,” “activation of IKK by TAK1,” and “inactivation of TAK1,” are recurrent regulators of both amplitude and AUC for TNF-α stimulation (Figures 3D and S4B). For IL-1β stimulation, “activation of TAK1 by C2” (C2 represents IL-1R in complex with MyD88, IRAK, and TRAF6), “IL1R synthesis,” “inactivation of TAK1,” and “activation of complexed IL1R” were identified as common determinants of both amplitude and AUC (Figures 3D and S4B). As for parameters of the IκB-NF-κB module, the variation of the IκBε-c-Rel dissociation rate selectively produced an increase in basal c-Rel expression but failed to fit the other experimentally observed patterns such as AUC of RelA (Figure S8A).

Next, we compared the two temperature simulation strategies: variation of receptor-proximal parameters and variation of both receptor-proximal parameters while simultaneously increasing basal c-Rel by adjusting IκBε-c-Rel dissociation. The values of the receptor-proximal parameters were varied for a given temperature in the directions observed in the RNA-seq analysis in the Arrhenius-scaled model. For TNF-α stimulation, the five receptor-proximal parameters were perturbed by LHS (in log2 scale) around the parameter values for 37°C, while the other parameters remained fixed to the Arrhenius-scaled value for each temperature. These temperature-specific ranges were used to account for the uncertainty in the temperature dependency of the parameters (Figures 5B and S8). Unlike the Arrhenius-only simulations, which predicted a conserved peak amplitude across the temperatures (Figure 5B, solid line), the receptor-proximal-parameter-adjusted model produced increasing peak amplitudes with rising temperature especially for c-Rel (Figure 5B, violin plots). However, the model did not recapitulate the increase of basal c-Rel expression with rising temperature (Figure S8B). When the model incorporated the increase of basal c-Rel via IκBε-c-Rel dissociation rate, simulations finally fit the overall experimentally observed temperature-dependent patterns of RelA and c-Rel signaling (Figures 5 and S8C). The improved modeling of temperature-dependent changes in NF-κB responses can also be visualized through representative time course simulations for each ligand (Figure 5C).

Figure 5.

Thermoregulation of receptor-proximal genes supports temperature-dependent signaling dynamics of NF-κB in response to cytokines

(A) IκBε-c-Rel dissociate rate and receptor-proximal parameters were varied with respect to the reference 37°C parameter set, to simulate the different temperature conditions. Barplots show basal cellular concentrations of RelA and c-Rel.

(B) Comparison of predictions from an Arrhenius-only versus Arrhenius with IκBε-c-Rel dissociate rate and receptor-proximal parameter adjustments. Colored diamonds connected by a solid line show Arrhenius-only peak amplitudes at 24°C (sky-blue), 34°C (blue), 37°C (orange), and 40°C (red). Violin plots show distributions of peak amplitude obtained from simulations in which IκBε-c-Rel dissociate rate and receptor-proximal parameters were jointly varied with respect to the reference 37°C parameter set using Latin hypercube sampling (uniform in log₂ fold-change, LFC). For both TNF-α and IL-1β, eight receptor-proximal parameters were adjusted. Temperature-dependent LFC ranges were: 37°C, LFC ∈ [−0.1, +0.1]; 40°C, LFC ∈ [+0.1, +0.3]; 34°C, LFC ∈ [−0.3, −0.1]; 24°C, LFC ∈ [−0.95, −0.75]. The remaining parameters were fixed at the Arrhenius-scaled values for the given temperature. Dashed lines trace the violin-plot median peak amplitudes from simulations. Pairwise comparisons were made using the Mann-Whitney U-test. Statistical significance is denoted as ∗p < 0.05, ∗∗p < 0.01, and ∗∗∗p < 0.0001; n.s., not significant.

(C) Representative model simulations of nuclear RelA and c-Rel concentrations at indicated temperatures, incorporating Arrhenius temperature scaling with modifications to IκBε-c-Rel dissociate rate and receptor-proximal parameters. For each temperature, we selected the parameter set whose simulated peak amplitude closely matched the median peak amplitude shown in (B). The black arrows denote the trend in peak amplitude with rising temperatures.

Overall, the model with temperature-dependency of both receptor-proximal parameters and IκBε-c-Rel dissociation rate produced the best fit between simulations and experimental data. This suggests that variability in receptor-proximal processes as well as IκBε inhibition is likely involved in thermoregulation of NF-κB signaling dynamics.

Discussion

Our dual endogenous NF-κB knock-in reporter imaging analysis revealed temperature as a key modulatory factor of fibroblast inflammatory signaling and identified an unexpected role of c-Rel as a temperature-sensing subunit of NF-κB. In contrast to the constitutively expressed RelA (also known as p65) subunit of NF-κB, c-Rel expression is regulated by immune stimuli such as antigenic challenge, which is critical for driving cell proliferation and robust immune responses. However, temperature-dependent regulation of c-Rel has not been documented to our knowledge, although temperature-dependent RelA responses to inflammatory cytokines have been analyzed.^10,11,18 The amplification of c-Rel in febrile temperatures found here adds to the growing list of potential mechanisms that are associated with the functional advantages of elevated core body temperature during infections. How might c-Rel amplification affect immune responses against pathogens? Thermally increased expression of c-Rel may explain cell cycle entry, proliferation and clonal expansion of adaptive immune cells, which are enhanced at febrile temperatures.³ In innate immune cells, higher c-Rel expression may be involved in inflammatory responses which may be altered at febrile temperatures. These untested possibilities highlight the need for re-assessment of established concepts in immune responses, examining the roles of c-Rel, for all physiologically relevant temperatures and cell types.

Our work shows that both canonical subunits RelA and c-Rel respond to inflammatory cytokines across a broad temperature range (24°C–40°C) but differ sharply in how their signaling dynamics scale with temperature. RelA dynamics primarily adjusts in period of oscillation, consistent with previous findings.^10,11 In contrast, c-Rel shows graded increases in expression, peak amplitude, and AUC, suggesting that temperature affects its activation strength rather than its periodicity. These observations indicate that RelA and c-Rel dimers in fibroblasts may convey distinct temperature-dependent transcriptional programs under inflammatory conditions.

Mathematical modeling provided a quantitative framework to interpret these behaviors. Arrhenius-scaled simulations reproduced the temperature-dependent shortening of RelA oscillations but failed to capture the experimentally observed analog behavior of c-Rel responses, indicating that accelerated kinetic rates alone are insufficient to explain the graded c-Rel signaling. GSA pointed to receptor-proximal parameters, such as TNFR and IL-1R synthesis, activation of complexed TNFR and IL-1R, TAK1 activation and inactivation, and IKK activation by TAK1, in addition to IκBε-related processes, as potential contributors to thermal sensitivity of c-Rel. Both TNF-α and IL-1β pathways converge on TAK1 activation via adaptor-mediated ubiquitin scaffolds, TRAF2 and TRAF6, respectively.¹⁹ Incorporating temperature-dependent variation at these receptor-proximal nodes substantially improved the agreement between model simulations and experimental data, supporting the idea that immediate signaling events encode environmental information upstream of canonical feedback modules. In neuroblastoma cells, temperature-dependent NF-κB activation was previously reported to be influenced by the negative feedback regulator A20/TNFAIP3.¹⁰ Our mathematical model did not explicitly include A20, partly because the kinetic parameters involving A20 remain poorly characterized. However, our RNA-seq data showed that neither the basal expression nor the gene induction of A20 changes with varying temperatures, in contrast to the neuroblastoma study, which suggests that the thermal regulation of A20 may depend on the cellular context.

Transcriptomic profiling reinforced this mechanistic view. RNA-seq revealed that genes encoding receptor-proximal signaling mediators, including Traf2, Traf5, Irak3, and Il1rn, are themselves temperature-responsive. Consistent with this, Traf2, Tnfrsf1b, and Irak2 (related to our DEG Irak3) have been reported to be HSF-1 target genes (hsf1base.org).²⁰ It will be interesting to examine if the induction of the receptor-proximal genes is mediated by HSF-1 in a genetic knockout system. Although bulk RNA-seq cannot distinguish RelA- versus c-Rel-specific transcriptional outputs, the parallel temperature dependencies across signaling and gene expression support the idea that c-Rel functions as a key amplifier of thermal adaptation in fibroblasts. It will be interesting to resolve which genes are selectively regulated by c-Rel, potentially mediating the thermally regulated inflammatory responses. A transcriptomic profiling of RelA or c-Rel knockout cells and chromatin-based methods such as ChIP-seq would be required to understand how distinct NF-κB dimers differentially regulate temperature-dependent gene expression.

Together, these findings reveal that RelA and c-Rel act as interlinked but distinct thermodynamic modules within the NF-κB signaling network. Temperature not only modulates oscillatory dynamics of NF-κB but also reshapes the relative contribution of receptor-proximal processes, yielding subunit-specific signatures of signaling strength. This multi-scale integration from receptor to gene expression highlights how environmental variables such as temperature can dynamically tune the architecture of inflammatory signaling. Beyond NF-κB biology, the combined live-cell imaging, computational modeling, and transcriptomic framework presented here offers a generalizable strategy for dissecting how physical parameters manifest in immune-signaling networks, guiding future efforts to link environmental cues with immune responses.

Limitations of the study

Because our mathematical model of NF-κB signaling in fibroblasts relies on parameter fitting of a previous model for macrophages, some of the parameter values carry uncertainty which is somewhat mitigated in our global sensitivity analysis.

The role of receptor-proximal proteins in thermoregulation of c-Rel was not examined directly by genetic perturbation or induced protein degradation methods. While we focused on how receptor-proximal signaling contributes to analog thermal responses of c-Rel, there may be additional mechanisms, especially in other cell types. Given that the previous reported role of A20 was absent in our cells, more cell type-specific and conserved factors probably remain to be identified for a more comprehensive insight into thermal regulation of inflammatory responses.

Resource availability

Lead contact

Further information and requests for resources and reagents should be directed to the lead contact, Myong-Hee Sung (sungm@nih.gov).

Materials availability

This study did not generate new unique materials.

Data and code availability

• •RNA-seq data have been deposited in the NCBI Gene Expression Omnibus (GEO) under accession number GSE317459.
• •All original code has been deposited and is publicly available at https://github.com/yeuran56/Temperature_RelA_cRel_fibroblast.
• •Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Acknowledgments

We thank all members of the Sung lab for critical reading of the manuscript. All animal care and housing were provided by the Comparative Medicine Section of the NIA. We also acknowledge the NIH High Performance Computing resource, the Biowulf cluster (https://hpc.nih.gov). This research was supported by the Intramural Research Program of the National Institutes of Health (NIH). The contributions of the NIH authors are considered Works of the United States Government. The findings and conclusions presented in this paper are those of the authors and do not necessarily reflect the views of the NIH or the US Department of Health and Human Services.

Author contributions

M.-H.S. conceived and supervised the study; Y.O. performed the experimentation and computational analyses; S.M.T.R. developed custom image analysis codes and provided guidance on quantitative analysis of imaging data; K.S.O. prepared samples for RNA-seq; D.D. performed RNA-seq pre-processing; Y.O. and M.-H.S. wrote the paper with input from all the authors.

Declaration of interests

The authors declare no competing interest.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Chemicals, peptides, and recombinant proteins
PBS	Gibco	Cat# 10010049
FBS	GeminiBio	Cat# 100-500
DMEM	Gibco	Cat# 11995065
DMEM (no phenol red)	Gibco	Cat# 21063029
TNF-α	R & D Systems	Cat# 410-MT-010
IL-1β	R & D Systems	Cat# 401-ML-005/CF
Fibronectin	Fisher Scientific	Cat# FC010
SPY 650 DNA	Cytoskeleton Inc	Cat# CY-SC 501
TRIzol reagent	ThermoFisher	Cat# 15596018
Critical commercial assays
RNeasy Mini Kit	Qiagen	Cat# 74104
Deposited data
Quantified imaging data from ear fibroblasts of double knock-in Green/Red mouse	This paper	https://github.com/yeuran56/Temperature_RelA_cRel_fibroblast
RNA-seq raw data	This paper	NCBI GEO accession GSE317459
Experimental models: Cell lines
Immortalized ear fibroblast G/R cells	Rahman et al.8	N/A
Software and algorithms
MATLAB R2022b	MathWorks	N/A
ImageJ (Fiji)	National Institutes of Health	N/A
Imaging data quantification	This study	https://github.com/yeuran56/Temperature_RelA_cRel_fibroblast
R	R core Team	N/A
RStudio	Posit, PBC	N/A
NF-κB signaling network modeling and simulation code	This study	https://github.com/yeuran56/Temperature_RelA_cRel_fibroblast
Other
μ-Slide 4 Well Chamber Slide	Ibidi	Cat# 80427

Experimental model and study participant details

Mice and immortalized ear fibroblast culture

A double knock-in (DKI) NF-κB reporter mouse line expressing mEGFP-RelA and mScarlet-c-Rel was generated previously by crossing the individual mEGFP-RelA and mScarlet-c-Rel knock-in strains in our laboratory.⁸ In this DKI line, the endogenous Rela and Rel loci are fluorescently tagged at the chromosomal level, enabling simultaneous and physiological visualization of RelA and c-Rel dynamics in live cells without overexpression artifacts. All mice were housed in the National Institute on Aging (NIA) animal facility and maintained under specific pathogen-free conditions. Animal handling and procedures followed NIH guidelines and were approved by the NIA Animal Care and Use Committee under the animal study protocol 465-LMBI-2028.

Primary ear fibroblasts were isolated from adult DKI mice and cultured as previously described.⁸ Briefly, dissected ear tissue was minced, enzymatically dissociated, and plated in standard growth medium (DMEM supplemented with 10% FBS, 1% penicillin, streptomycin, and L-glutamine). Cells were maintained in phenol-red-containing medium to facilitate pH monitoring during long-term culture. To generate stable immortalized fibroblast lines, primary cultures were expanded and routinely passaged every 72 hours for approximately 20-25 passages, at which point they exhibited hallmark features of immortalized fibroblasts, including sustained proliferation and stable expression of endogenous fluorescent RelA and c-Rel reporters.

Method details

Live-cell microscopy

Immortalized fibroblasts derived from mEGFP-RelA/mScarlet-c-Rel DKI mice were plated onto fibronectin-coated glass-bottom μ-Slide 4-well imaging chambers (ibidi) in phenol red-free DMEM with 10% FBS, 1% penicillin, streptomycin, and L-glutamine. Cells were allowed to recover and adhere for 24 hours at 37 °C in humidified incubator with 5% CO₂.To ensure full thermal equilibration prior to imaging, cultures were transferred two hours before acquisition to a Zeiss LSM880 confocal microscope equipped with an environmental chamber supplying 5% CO₂ and pre-set to designated imaging temperature (24°C, 34°C, 37°C, or 40°C). For nuclear segmentation, cells were incubated with SPY650-DNA (1x-1.5x manufacturer’s recommended dilution) for one hour before imaging.

Time-lapse imaging was performed for more than 12 hours using a Zeiss LSM880 with Airyscan detection and a 40x/1.4 NA Plan-Apochromat oil immersion objective. Images were acquired every 7 minutes to capture NF-κB activation kinetics with sufficient temporal resolution while minimizing phototoxicity. Fluorescence excitation was achieved using 488 nm (mEGFP-RelA), 561 nm (mScarlet-c-Rel), and 633 nm (SPY650-DNA) laser lines, enabling simultaneous visualization of RelA, c-Rel, and nuclei. Cells were stimulated with recombinant murine TNF-α and IL-1β (10ng/mL; R&D Systems) exactly one hour after the start of imaging, ensuring accurate baseline quantification of NF-κB nuclear localization before ligand exposure. Biological replicates were obtained by ensuring that each stimulation-temperature condition was repeated in three independent imaging sessions.

Bulk RNA-seq

Immortalized mouse ear fibroblasts were seeded at medium confluence in 6-cm tissue culture dishes and cultured overnight at 37°C in a humidified incubator with 5% CO₂ to ensure recovery and stable basal transcriptional activity. The following day, cultures were transferred to incubators pre-equilibrated to 24°C, 34°C, 37°C, or 40°C and were allowed to adapt for 2 hours to establish temperature-specific steady states prior to stimulation. Cells were then either left untreated (0h) or stimulated with recombinant murine TNF-α (10ng/mL) or IL-1β (10ng/mL) for 1 hour or 3 hours. At each time point, cells were lysed directly on the plate using RLT buffer (Qiagen). Lysates were stored at -80°C until processing. Total RNA quantity and purity were assessed using a NanoDrop spectrophotometer (ThermoFisher). Three independent biological replicates were collected for each experimental condition. In total, 60 RNA-seq libraries were generated: 48 stimulated samples (4 temperatures x 2 cytokines x 2 time-points x 3 biological replicates) and 12 unstimulated baseline controls collected at 0h across the four temperatures (3 replicates each).

Mathematical modelling of NF-κB responses to TNF-α and IL-1β

A computational model describing the temporal dynamics of RelA and c-Rel activation in response to cytokine stimulation was constructed by adapting a previous model of NF-κB signaling.¹² The original model was developed for TNF-α–induced signaling in bone-marrow–derived macrophages. To tailor the model to fibroblast, parameters governing processes such as basal IκBα mRNA synthesis, induced IκBα mRNA related parameters, basal IκBε mRNA synthesis, and induced IκBε mRNA related parameters were re-estimated to fit our live-cell imaging time-series measurements of RelA and c-Rel dynamic in fibroblast.

Because the study required a unified model capable of simulating responses to both TNF-α and IL-1β, the receptor-proximal signaling module was expanded to incorporate the MyD88–IRAK–TRAF6 (MIT) axis, which mediates IL-1 receptor–dependent activation of TAK1. The receptor-proximal module of the IL-1β pathway was implemented analogously to the TNF-α module, reflecting their shared signaling architecture that converges on TAK1-mediated activation of the IKK complex. In both cytokine pathways, adaptor complexes mediate the formation of K63-linked ubiquitin scaffolds that recruit and activate TAK1 - TRAF2 and RIPK1 in the TNF-α pathway, and MyD88, IRAK, and TRAF6 in the IL-1β pathway, respectively.^19,21 The addition of IL-1β allowed parallel simulations of TNFR1- and IL-1R–initiated signaling cascades while preserving a shared downstream IKK–IκB–NF-κB regulatory architecture. Model parameters in the IL-1β modules were tuned to improve the visual fit to early responses from the time lapse imaging data. Time series data (Figure 1C) showed that IL-1β produced a deeper valley between the first and second peaks compared with TNF-α. Subsequent fitting of NF-κB dynamics for IL-1β ligand was achieved through tuning receptor-proximal module parameters. The resulting model thus supports comparison of RelA and c-Rel dynamics across two cytokine pathways within a common mechanistic framework.

Arrhenius-based modeling of temperature dependency

Following the general strategy used in prior temperature-sensitive modeling work,¹⁰ we incorporated temperature dependence into each kinetic parameter using an Arrhenius relationship. For each reaction rate parameter k_j, temperature scaling was defined as: $k_j\left(T\right)=A_j{e}^{-E_j/RT}\text{,}$ where A_j is the pre-exponential factor, E_j is the activation energy, R is the universal gas constant, and T is temperature.

To quantify the predicted temperature sensitivity of the oscillation period, we used the standard approximation: $dp/dT\propto \sum _j{c}_j{E}_j$, c_j = ∂lnp/∂lnk_j, were c_j denotes the sensitivity coefficient of the oscillation period with respect to parameter k_j. As in previous work, activation energies were assigned to maximize the experimental magnitude of |dp/dT|, providing an upper bound on possible temperature effects.

Because TNF-α and IL-1β stimulation produced comparable temperature-dependent shortening of the RelA oscillation period in our live-cell imaging experiments, we computed |dp/dT| for each ligand separately and used the mean of the two values as the effective Arrhenius-based temperature sensitivity applied to both ligands. For each rate parameter, the corresponding pre-exponential A_j was then calibrated such that the Arrhenius-scaled rate reproduced the experimentally validated parameter value at 37°C.

Global sensitivity analysis

To identify biochemical parameters that differentially influence RelA and c-Rel signaling across temperature conditions, we performed a global sensitivity analysis using Latin Hypercube Sampling (LHS) coupled with Partial Rank Correlation Coefficient (PRCC) analysis. This two-step framework is widely used in systems biology for high-dimensional nonlinear models in which many parameters are uncertain or difficult to measure experimentally.¹⁵

LHS is a statistical method that efficiently samples large parameter spaces by ensuring each parameter is explored uniformly across its range.²² The method divides the feasible range of each parameter into equally probable intervals. Extending this procedure across all parameters generates a set of parameter combinations that uniformly spans the multidimensional parameter space. In the NF-κB subunit model, LHS was applied to 61 rate parameters (non-rate constants were excluded). Several parameters in the model occur as paired biochemical processes that share the same biological scale (e.g., IκBε-RelA and IκBε-c-Rel association rates). For these parameter groups, identical LHS samples were applied to preserve biologically constrained relationships. All parameters were varied within a 10-fold range around their reference values at 37 °C. Each sampled parameter set was simulated to generate temperature-dependent NF-κB signaling features, including baseline expression, peak amplitude, and area under the curve (AUC) for both RelA and c-Rel.

PRCC provides a robust, rank-based measure of global sensitivity that captures relationships between each parameter and a model output while controlling for the influence of all other parameters. After running simulations for all LHS-sampled parameter sets, PRCC values were calculated for each biochemical parameter with respect to each signaling feature. For each signaling feature, PRCC values were computed separately for RelA and c-Rel, and we report ΔPRCC = |PRCC(c-Rel)|-|PRCC(RelA)| to highlight parameters exerting stronger influence on c-Rel relative to RelA. Positive PRCC values indicate that increasing the parameter increases the output, whereas negative values reflect an inverse relationship. The top ten parameters ranked by Δ PRCC for each condition are shown in Figures 3C, 3D, and S4B.

Quantification and statistical analysis

Quantitative analysis of live-cell imaging

Single-cell NF-κB dynamics were extracted from time-lapse fluorescent images using custom ImageJ (Fiji) macros and MATLAB scripts (R2022b). Raw multi-channel, multi-position image stacks were first separated and converted from .czi to .tif format using a custom Fiji macro for standardized processing. Nuclear segmentation was performed using the SPY650-DNA channel, which provides robust and uniform nuclear labeling across temperatures. Cytoplasmic and whole-cell regions were delineated using intensity-based masking from the mEGFP-RelA and mScarlet-c-Rel fluorescence channels, ensuring accurate segmentation of nuclear and cellular areas. Custom MATLAB pipelines were used to track individual cells over time and to compute background-subtracted fluorescence intensities for nuclear, cytoplasmic, and whole-cell compartments. Cells were excluded if they showed no detectable response to stimulation, underwent apoptosis or mitosis, displayed abnormal nuclear morphology, or moved into or out of the field of view during acquisition.

RNA-seq data analysis

Bulk RNA-seq data were analyzed using a previous analytical pipeline.²³ Stranded 151-bp paired-end reads were aligned to the mouse reference genome (GRCm39, Genome Reference Consortium (GRC) 2020) using STAR v2.7.11b.²⁴ The reads were counted per gene using RSEM v1.3.3.²⁵ Downstream processing was performed in R using edgeR.

Before statistical analysis, lowly expressed genes were filtered out to improve statistical power and reduce noise. Specifically, a gene was retained only if its expression reached FPKM ≥ 1 in at least 4 of the 60 libraries (≈7% of samples). Additionally, genes annotated on unplaced or unlocalized chromosomes were excluded. For quality assessment, FPKM values were transformed as log₂(FPKM+0.1), and their distributions were visually inspected across all samples, confirming comparable library-wide expression profiles.

Differential expression (DE) analyses were performed using the TMM-normalized count matrix (calcNormFactors). Negative binomial dispersions were estimated using estimateDisp, and gene-wise generalized linear models were fitted with glmFit. Statistical significance was assessed using likelihood ratio tests (glmLRT).

To identify temperature-responsive genes (temperature-DEGs), for each ligand and time point, gene expression at 24°C, 34°C, or 40°C was compared against the matched 37°C condition. Genes were classified as temperature-DEGs if they satisfied |log₂ FC| ≥ 0.5, and false discovery rate (FDR) < 0.05.

To identify cytokine-responsive genes (cytokine-DEGs), for each temperature condition, gene expression after TNF-α or IL-1β stimulation (1 h or 3 h) was compared to the matched no-ligand control. Genes were classified as cytokine-DEGs if they satisfied |log₂ FC| ≥ 1 and FDR < 0.05.

To visualize temperature-dependent transcriptional patterns, we generated heatmaps using log₂-transformed FPKM values. For each ligand-duration condition, temperature effects were expressed relative to the 37°C by computing: Δlog₂(FPKM) = log₂(FPKM_temp + 0.1)-log₂(meanFPKM_37°C + 0.1). Temperature-DEGs (union across all conditions) were clustered using partitioning around medoids (PAM). Heatmaps were generated using the ComplexHeatmap R package.

To visualize cytokine-dependent transcriptional responses, heatmaps were generated using log₂-transformed FPKM values, with cytokine effects computed relative to the matched no-ligand condition at each temperature: Δlog₂(FPKM) = log₂(FPKM_ligand + 0.1)-log₂(meanFPKM_no-ligand + 0.1). Cytokine-DEGs (union across all conditions) were clustered using PAM and visualized using ComplexHeatmap. Selected NF-κB target genes were annotated on the heatmap by gene symbol for visual reference.

Gene set enrichment analysis

Gene ontology (GO) enrichment analysis was performed on temperature-responsive PAM clusters using clusterProfiler R package. Gene symbols were extracted from Ensembl identifiers, converted to official mouse symbols, and mapped to Entrez IDs using org.Mm.eg.db. For each cluster, enrichment of Biological Process terms was assessed with enrichGO (Benjamini–Hochberg p-value < 0.05). For visualization, the three most significantly enriched Biological Process terms for each cluster were displayed, ranked by adjusted p value.

For the cytokine-induced DEGs, GO analysis was performed only if the number of upregulated genes (log2 FC >= 1) is at least 100. Top five categories of each gene set were combined for a dotplot.

DNA motif enrichment analysis

Motif enrichment analysis was performed using the MEME (v5.5.5).²⁶ Promoter regions were defined as 2 kb windows centered on transcription start sites (TSSs) of significantly upregulated genes (log2 FC ≥ 1) by cytokine treatment. Promoter sequences from significantly downregulated genes (log2 FC < 1) were used as the background set. Motifs were considered significantly enriched if the adjusted p value was < 0.05. Top five motifs of each gene set were combined for a dotplot.