Stimulus-selective crosstalk via the NF-κB signaling system reinforces innate immune response to alleviate gut infection

Tissue microenvironment functions as an important determinant of the inflammatory response elicited by the resident cells. Yet, the underlying molecular mechanisms remain obscure. Our systems-level analyses identified a duration code that instructs stimulus specific crosstalk between TLR4-activated canonical NF-κB pathway and lymphotoxin-β receptor (LTβR) induced non-canonical NF-κB signaling. Indeed, LTβR costimulation synergistically enhanced the late RelA/NF-κB response to TLR4 prolonging NF-κB target gene-expressions. Concomitant LTβR signal targeted TLR4-induced newly synthesized p100, encoded by Nfkb2, for processing into p52 that not only neutralized p100 mediated inhibitions, but potently generated RelA:p52/NF-κB activity in a positive feedback loop. Finally, Nfkb2 connected lymphotoxin signal within the intestinal niche in reinforcing epithelial innate inflammatory RelA/NF-κB response to Citrobacter rodentium infection, while Nfkb2−/− mice succumbed to gut infections owing to stromal defects. In sum, our results suggest that signal integration via the pleiotropic NF-κB system enables tissue microenvironment derived cues in calibrating physiological responses. DOI: http://dx.doi.org/10.7554/eLife.05648.001


Introduction
Tight regulation of inflammatory responses is important; uncontrolled inflammation underlies various human ailments, while insufficient responses limit host defense to pathogens. Tissue-resident cells those that participate in inflammatory immune activation also exhibit functional differences by adapting to the repertoire of cell-differentiating cues present in distinct microenvironments. Indeed, macrophages and dendritic cells present in different anatomic niche display heterogeneity in inflammatory signatures (Iwasaki and Kelsall, 1999;Stout and Suttles, 2004). Likewise, a requirement for CD40, primarily involved in B-cell maturation, in inflammatory gene expressions in endothelial cells was documented (Pluvinet et al., 2008). Similarly, lymph node inducing lymphotoxin-β receptor (LTβR) was shown to be critical for innate immune responses (Spahn et al., 2004;Wang et al., 2010). Yet, the cellular circuitry that functions at the intersection of tissue microenvironment derived signals and those impinged upon by pro-inflammatory cytokines or pathogen-derived substances remains obscure.
The NF-κB family of transcription factors plays an essential role in activating pathogenresponsive gene-expression program in tissue-resident cells. In the canonical NF-κB pathway, inflammatory cues engage NEMO-IKK2 (NEMO-IKKβ) kinase complex to phosphorylate inhibitory IκB proteins, the major isoform being IκBα, bound to the cytoplasmic RelA:p50 NF-κB dimers. Signal-induced phosphorylation leads to proteasomal degradation of IκBs and release of RelA:p50 dimers into the nucleus. The nuclear RelA dimers activate the expressions of pro-inflammatory chemokine and cytokine genes as well as its own inhibitor IκBα, which ensures proper attenuation of inflammatory responses in a negative feedback loop. In contrast to canonical signaling, the non-canonical pathway transduces signals from cell-differentiating cues those engage BAFFR, CD40, or LTβR. Non-canonical signaling involves NIK and IKK1 (NIK-IKKα) mediated phosphorylation of Nfkb2 encoded precursor p100 bound to RelB (Sun, 2012). Subsequent proteasomal processing removes the C-terminal inhibitory domain of p100 from RelB:p100 complex to generate RelB:p52 NF-κB dimer, which mediates the expressions of organogenic chemokine genes in the nucleus (Bonizzi et al., 2004).
Molecular interaction between the non-canonical signal transducer p100 and RelA has also been charted. In its homo-oligomeric form, termed IκBδ, p100 was shown to utilize its inhibitory domain to sequester a subpopulation of the RelA:p50 dimer (Basak et al., 2007;Savinova et al., 2009). LTβR through non-canonical NIK-IKK1 signal inactivates IκBδ to induce a weak yet sustained RelA: p50 activity. Conversely, RelA-induced synthesis of p100 and consequent accumulation of inhibitory IκBδ was shown to exert negative feedback limiting canonical RelA activity (de Wit et al., 1998;Legarda-Addison and Ting, 2007;Shih et al., 2009). In addition, an alternate RelA: p52 dimer has been reported which is thought to constitute a minor kappaB DNA binding activity (Hoffmann et al., 2003). Crosstalk between apparently distinct cell signaling pathways is known to offer diversity in cellular responses. Despite these connectivities, a plausible role of signal integration via the NF-κB system in regulating inflammatory RelA NF-κB responses has not been investigated.
In a multidisciplinary study combining biochemistry, genetics, and mathematical modeling, here, we characterized a duration code that determines stimulus-specific crosstalk between canonical and non-canonical signaling in fine-tuning inflammatory RelA NF-κB activity. Through such crosstalk, LTβR sustained TLR4 triggered RelA NF-κB responses by supplementing RelA:p52 NF-κB dimer in a positive feedback loop. Finally, we established the physiological significance of crosstalk control of RelA in intestinal epithelial cells (IECs), where, the NF-κB system integrates gut microenvironment derived lymphotoxin signals through Nfkb2 to calibrate innate immune responses to Citrobacter rodentium. eLife digest The innate immune system is the body's first line of defense against infection and disease. Innate immune cells are found in every tissue type, poised to respond immediately to damaged, stressed, or infected host cells. When innate immune cells recognize any injury or infection, one of the first things they do is trigger the inflammatory response. Fluid and other immune cells then move from the blood into the injured tissues. This movement can cause redness and swelling. But the response helps to establish a physical barrier against the spread of infection, promotes the elimination of both invading microbes and damaged host cells, and encourages the repair of the tissue.
Inflammation is tightly controlled. If the response is too weak, it could leave an individual prone to serious infection. On the other hand, excessive inflammation can severely damage healthy cells and tissues. Inflammation is regulated differently in different tissue types, and the environment within the tissue itself influences the activity of local innate immune cells and the inflammatory response. However, the molecular mechanisms responsible for receiving and interpreting the signals derived from the host tissue remain unknown. Now, Banoth et al., have revealed that the integration of inflammation-provoking signals, such as injury or infection and cues from the tissue environment occurs via the so-called 'NF-κB signaling system'. NF-κB is a protein found in almost all cell types, and when activated it is able to switch on the expression of many different genes. Banoth et al. explain that signal integration via the NF-κB system enables cues from the tissue environment to tune a cell's responses. Further experiments confirmed the importance of this signal integration by showing how a signal coming from intestinal tissue can influence the activity of innate immune cells and inflammation in the gut.
These findings suggest that a breakdown in the NF-κB signaling system's ability to integrate multiple signals, including those derived from the tissue environment, may be responsible for many inflammatory disorders, and in particular those that involve the gut. Future work is now needed to explore this possibility.

Results
A duration code controlling crosstalk between canonical and noncanonical NF-κB signaling Given the interconnectedness of the canonical and non-canonical arms (see Introduction and Figure 1A), we asked if signal integration via the NF-κB system would allow cell-differentiating cues to modulate inflammatory RelA NF-κB responses. Mathematical reconstruction of dynamic networks illuminates emergent properties, such as crosstalk (Basak et al., 2012). To explore crosstalk control, we developed a mathematical model, which we termed the NF-κB Systems Model v1.0 (Appendix-1), basing on previously published single NF-κB dimer model versions (Hoffmann et al., 2002;Basak et al., 2007). In our mathematical model, however, we depicted nuclear activation of both the major RelA:p50 dimer and RelA:p52 dimer, which is thought to constitute a minor RelA NF-κB activity. As described in the preceding single dimer models (Hoffmann et al., 2002;Basak et al., 2007), signal-responsive degradation and resynthesis of IκBα, IκBβ, IκBε, and inhibitory p100/IκBδ dynamically controlled RelA activity. The model was parameterized based on literature, our own measurements (Appendix-1, Appendix figures 1-5, Supplementary files 1-3), and fitting procedures. Simulating individual TNFR or LTβR regime, we could recapitulate experimentally observed strong, but temporally controlled, activation of RelA NF-κB complexes during canonical IKK2 signaling or the weak induction of RelA dimers during non-canonical NIK-IKK1 signaling, respectively ( Figure  Next, we examined potential crosstalk between IKK2 and NIK-IKK1 inputs in augmenting RelA NF-κB response in silico. To this end, we generated a theoretical library (Shih et al., 2009) of 356 kinase activity profiles, where each member possesses distinct peak onset time, peak amplitude, and duration ( Figure 1C, Figure 1-figure supplement 2A). To screen for permissive crosstalk conditions, we fed this library into the model through IKK2 or NIK-IKK1 or both the arms and iteratively simulated respective RelA activities. Then, we computed RelA responses in the co-treatment regime relative to individual cell stimulations to assign crosstalk indexes to different IKK2 and NIK-IKK1 combinations ( Figure 1D). Plotting the dynamic features of the crosstalk-proficient kinase inputs, we could reveal a critical duration threshold; where IKK2 activities sustained for more than 2 hr were more likely to engage into crosstalk for varied peak amplitudes and inputs with shorter duration were crosstalk inefficient ( Figure 1E and Figure 1-figure supplement 2B). Illustrating a similar but more elaborate duration control, NIK-IKK1 activities longer than 8 hr selectively participated into crosstalk with the canonical pathway.
Inflammatory mediators activate canonical IKK2 with disparate temporal controls. Consistent to the prior report (Werner et al., 2008), our kinase assay ('Materials and methods') revealed that IL-1β, an important pro-inflammatory cytokine, only transiently activates IKK2 in mouse embryonic fibroblasts (MEFs) (left panel, Figure 1F). In contrast, bacterial LPS through TLR4-induced IKK2 activity that persisted above the basal level even at 24 hr post-stimulation (right panel, Figure 1F, Figure 1-figure supplement 3B) (Covert et al., 2005). Mimicking prolonged signaling during cell-differentiation processes, LTβR engagement using agonistic αLTβR antibody led to sustained activation of the non-canonical NIK-IKK1 ( Figure 1G and Figure 1-figure supplement 3A,B). Using these experimental kinase activities as inputs, our computational simulations revealed insulation of IL-1R signaling from LTβR-mediated crosstalk (left panel, Figure 1H), but robust crosstalk between TLR4 and LTβR that amplified late RelA response upon costimulation (right panel, Figure 1H). Therefore, our mathematical analyses predicted that a duration code selectively engages long lasting canonical kinase activities into crosstalk with LTβR induced NIK signal to impart stimulus specificity.
Stimulus-specific crosstalk allows LTβR signal to prolong TLR4 induced RelA NF-κB response To experimentally verify stimulus specificity of crosstalk control, we measured nuclear RelA activities induced in MEFs by canonical or non-canonical inducers or co-treatment regime that concomitantly activated both the pathways. IL-1R signal, in parallel to transient IKK2 activation, elicited strong RelA activity at 30 min in EMSA that was largely attenuated within 1 hr, whereas, non-canonical LTβR signal only weakly induced RelA and RelB dimers those persisted even at 24 hr ( Figure 2A). Indeed, we were unable to detect any significant enhancement of RelA activity, relative to IL-1 induced peak, upon The kinase inputs were fed into the model through the IKK2 or NIK-IKK1 or both the arms. The RelA NF-κB responses, quantified as baseline corrected total area under the respective activity curves, were used for computing crosstalk indexes. (E) Based on their respective crosstalk indexes, top 10% combinations of theoretical IKK2 and NIK1 activity profiles were identified and duration as well as amplitude of the associated crosstalk-proficient IKK2 (top) or NIK-IKK1 (bottom) profiles were plotted. (F and G) The IKK2 (F) or NIK-IKK1 (G) activities were monitored by incubating GST-IκBα or GST-IκBδ with NEMO or NIK co-immunoprecipitates derived from MEFs treated with IL-1β, LPS (F, left and right panels), or αLTβR (G), respectively. For IKK2 assay, co immunoprecipitated IKK1 and for NIK-IKK1 assay, actin present in the cell extracts was used as loading controls. (H) Computational simulations predicting augmented RelA activity in LPS+αLTβR (right panel) and a lack of crosstalk in IL-1+αLTβR (left) co-treatment regimes. DOI: 10.7554/eLife.05648.003 The following figure supplements are available for figure 1:  (Figure 2A). In comparison, canonical TLR4 induced a temporally distinct RelA NF-κB activity with an early peak at 1 hr, subsequent descend and a progressively weakened late phase between 8 hr and 24 hr ( Figure 2B). Corroborating our mathematical prediction, concomitant LTβR signal sustained NF-κB response triggered by TLR4 ( Figure 2B). Signal integration via the NF-κB system synergistically enhanced TLR4-induced late RelA activity at 24 hr in the costimulation regime with mostly unaltered RelB response relative to solitary LTβR engagement ( Figure 2B, quantification and Figure 2C). Sequentially engaging MyD88 and Trif, TLR4 was shown to produce extended IKK2 activity (Covert et al., 2005). To determine if the observed stimulus specificity of crosstalk is indeed due to the duration of IKK2, we utilized Trif-deficient MEFs that only transiently activated IKK2 upon LPS treatment ( Figure 2D). Despite a functional non-canonical pathway (Figure 2-figure supplement 1), LTβR was restricted from crosstalk with TLR4 in Trif-deficient cells ( Figure 2E), thereby, suggesting a Trif-dependent mechanism that relies on the duration of canonical IKK2 in imparting stimulus specificity of crosstalk control.
Signal integration via the NF-κB system amplifies the late expressions of TLR4-induced pro-inflammatory genes Long-lasting kinase activities are expected to elicit sustained RelA responses. Then what might be the significance of signal integration via the NF-κB system? Interestingly, computational simulations demonstrated only a muted increment in the RelA activity with increasing duration of IKK2 ( Figure 3A). LTβR induced NIK-IKK1 signal relieved this saturation to fully unravel the NF-κB activation potential of long duration IKK2 signals upon crosstalk. We postulated that the difference in the RelA responses induced by long-lasting IKK2 in the presence or absence of non-canonical signal would decode into differential gene activities.
To evaluate the potential gene-effect of crosstalk, we measured the expressions of several known RelA target chemokine and cytokine genes using quantitative RT-PCR. Our analyses revealed that IL-1 treatment rapidly induces the expressions of TNF, IP-10, and MIP-1α mRNAs within 1 hr with residual expressions at 24 hr post-stimulation ( Figure 3B). The levels of IL-1β and RANTES mRNAs were insensitive to IL-1 treatment. Also, weak LTβR signal alone did not significantly induce the expressions of these pro-inflammatory genes in MEFs. Indeed, LTβR costimulation was ineffective in augmenting IL-1 induced early or late expressions of the chemokine and cytokine genes ( Figure 3B). Solitary LPS treatment not only robustly induced the expressions of TNF, IP-10, and MIP-1α mRNAs, but also led to late accumulation of IL-1β and RANTES mRNAs at 24 hr ( Figure 3C). Consistent to our hypothesis, LTβR costimulation prolonged TLR4-induced gene expressions with further augmented late, but not early, expressions of IL-1β, IP-10, MIP-1α, and RANTES mRNAs. TNF mRNA levels were insensitive to crosstalk regulation ( Figure 3C).
Furthermore, we used microarray analyses to compare global gene expressions activated by TLR4 or LTβR or both at 24 hr post-stimulation. Estimating normalized crosstalk scores (bottom panel, Figure 3D, Figure 3-source data 1, 'Materials and methods'), we could reveal a synergistic effect of LTβR on TLR4 stimulated late gene expressions in WT MEFs. Out of 943 LPS induced genes, however, a select set of 114 genes was further upregulated upon costimulation. Strikingly, gene set enrichment analysis (GSEA) (Subramanian et al., 2005) (Figure 3-source data 2, 'Materials and methods') demonstrated an enrichment of NF-κB targets among genes positively controlled through crosstalk (middle and top, Figure 3D). We have also noted downregulation of several LPS-induced genes in the costimulation regime those appeared less likely to be NF-κB targets in GSEA. Taken together, these analyses substantiated an important function of prolonged RelA activity in crosstalk-amplification of TLR4-induced late expressions of NF-κB target genes, particularly those encode pro-inflammatory chemokines and cytokines. . LTβR signal sustains TLR4, but not IL-1R, induced RelA NF-κB response. (A) Nuclear NF-κB activities induced in MEFs by IL-1β or αLTβR or co-treatment were resolved in EMSA using a κB site containing DNA probe. The faster migrating complex, indicated with an arrowhead, consists of RelB and the slower migrating complex activated by both canonical or non-canonical signaling, denoted with an arrow, consists of RelA dimers. The compositions of the DNA binding complexes were ascertained in Figure 2C. Right, signal corresponding to RelA NF-κB activities were quantified and graphed relative to the respective IL-1 induced peak value. Data were expressed as mean of 3 quantified biological replicates ± SEM. (B) EMSA result, representative of three independent biological replicates, revealing augmented late NF-κB activities in the co-treatment regime as compared to cell treatment with LPS or αLTβR alone. Right, signal corresponding to RelA NF-κB activities were similarly quantified and graphed relative to LPS induced peak value. Note, late RelA activities in the LPS+αLTβR cotreatment regime were significantly augmented from of the LPS induced activities.  Non-canonical signal transducer Nfkb2 supplements RelA:p52 dimer to sustain canonical RelA NF-κB responses To determine the mechanism underlying signal integration via the NF-κB system, we individually perturbed 105 model parameters and quantified relative changes in the crosstalk index ('Materials and methods'). Our parameter sensitivity analysis identified the rate constant associated with the NF-κB-induced transcription of Nfkb2 as the most critical parameter underlying crosstalk control of RelA NF-κB activity ( Figure 4A). As such, Nfkb2 encodes for both NF-κB inhibitor and NF-κB precursor functions. Computationally simulating individual RelA:p50 and RelA:p52 nuclear activities in the LPS, αLTβR, or costimulation regimes, we could further suggest that the Nfkb2 precursor function in generating RelA:p52 dimer is important for augmenting RelA NF-κB activity during crosstalk in WT system ( Figure 4B). Consistently, our modeling analyses predicted complete abrogation of crosstalk in Nfkb2 −/− cells ( Figure 4B).
Our biochemical studies revealed that LPS induces RelA NF-κB-driven transcription of Nfkb2 to produce p100 (top panel, Figure 4C and Figure 4-figure supplement 1A), which was shown to oligomerize as NF-κB inhibitory IκBδ Shih et al., 2009;Tao et al., 2014). Concomitant LTβR signal instead utilized TLR4-induced, newly synthesized p100 to potently generate p52, thereby, neutralizing the inhibitory p100 function ( Figure 4C). p50 levels were not discernibly altered in these stimulation regimes in our experiments. Our immunoprecipitation based analysis demonstrated that NIK-IKK1 signal relieves RelA from p100/IκBδ-mediated inhibition during crosstalk ( Figure 4D). Intriguingly, LTβR costimulation of MEFs for 24 hr also produced ∼fourfold more RelA:p52 NF-κB dimer as compared to solitary LPS treatments ( Figure 4D). Limited transcriptional up-regulation of Nfkb2 by weak LTβR signal was correlated with only the modest p52 and RelA:p52 generation ( Figure 4C,D). In supershift assay, we could ascertain that RelA:p52 dimer generated upon LTβR costimulation appears as a strong nuclear DNA binding activity at 24 hr ( Figure 4E) to supplement to the TLR4-induced RelA NF-κB responses. An important role of dimerization in stabilizing NF-κB monomers from degradation has been reported earlier (Fusco et al., 2008). Interestingly, RelA protein rapidly accumulated in cells upon proteasome inhibition ( Figure 4F) suggesting a robust constitutive degradation mechanism that offsets basal synthesis of RelA monomers in maintaining the steady-state level. Our study indicated that this enduring flux ensures copious supply of RelA to bind to de novo generated p52, produced from the newly synthesized p100 during crosstalk.
Next, our genetic analyses revealed that canonical RelA:p50 response to TLR4 signal, primarily controlled through classical IκBs, is largely intact in Nfkb2 −/− with early induction and diminished late activities comparable to WT MEFs ( Figure 4G and Figure 4-figure supplement 1B). Consistent to the prediction based on computational modeling studies, a lack of RelA:p52 dimer generation in Nfkb2 −/− cells, however, ablated LTβR-mediated enhancement of TLR4-induced late RelA DNA binding activity ( Figure 4G) as well as crosstalk amplification of RelA target pro-inflammatory gene expressions (bottom panel, Figure 4H). LTβR costimulation not only enhanced TLR4 induced RelA DNA binding but also activated RelB dimers through the non-canonical pathway. Prior reports have indicated cell-type specific inhibitory as well as activating role of RelB in chemokine gene expressions (Weih et al., 1996;Shih et al., 2012). Importantly, costimulation of Relb −/− MEFs led to similar hyperactivation of LPS-induced late expressions of IL-1β, IP-10, and RANTES mRNAs as in WT cells (compare top panel, Figure 4H with Figure 3C). Although, the crosstalk effect on MIP-1α expressions was somewhat muted owing to prolonged expression of this gene in Relb −/− MEFs in response to solitary LPS treatment. Therefore, our analyses confirmed that the precursor function encoded by Nfkb2 in generating RelA:p52 NF-κB dimer is critical for integrating lymphotoxin derived signals to the pro-inflammatory RelA NF-κB pathway. Our analyses also suggested that LTβR costimulation led to the hyperactivation of LPS-induced expressions of chemokine and cytokine genes in an Nfkb2dependent manner with only a minor, if any, role for Relb.

Inducible synthesis of Nfkb2 by canonical signal triggers a positive feedback loop during crosstalk
Given the computational prediction for an important role of NF-κB-induced transcription of Nfkb2, we compared the inducible expression of the crosstalk mediator Nfkb2 in response to LPS or IL-1 to understand the molecular basis of stimulus-specific control. As opposed to rapid expression of Nfkbia mRNA, which encodes IκBα, LPS induced Nfkb2 mRNA with a delay (top panel, Figure 5A). Similarly, chronic TNF treatment induced Nfkb2 mRNA in WT MEFs with an explicit 1 hr delay ( Figure 5B) that was also observed earlier and incorporated in both the previous (Basak et al., 2007) as well as the current mathematical model versions. Analogous time lags were observed in the expression of several inflammatory genes those require additional chromatin modifications for the initiation of RelA-induced transcription (Natoli et al., 2005). When Nfkb2 transgene was stably expressed in Nfkb2 −/− MEFs from an exogenous NF-κB responsive promoter, Nfkb2 mRNA was readily induced by TNF without a delay ( Figure 5B). Remarkably, IL-1 treatment was ineffective in activating the expression of Nfkb2 mRNA in  Figure 5A). Only upon eliminating the transcriptional delay, our mathematical model could simulate Nfkb2 mRNA induction by IL-1 treatment ( Figure 5C). Indeed, we could also experimentally rescue the defect in Nfkb2 mRNA induction by IL-1 treatment in the engineered Nfkb2 −/− cell line, which expresses Nfkb2 transgene from the NF-κBresponsive promoter without the delay ( Figure 5D). Consistent to our computational identification that NF-κB inducible transcription of Nfkb2 is important, disruption of NF-κB-inducible synthesis by expressing p100 from a constitutive promoter in Nfkb2 −/− MEFs abrogated crosstalk amplification of TLR4-induced late NF-κB activity by concomitant LTβR signal ( Figure 5E and Our results confirmed that signal induction of Nfkb2 is important for crosstalk and suggested that a promoter intrinsic delay necessitates persistent canonical signal for RelA-mediated induction of prosynergistic Nfkb2. Such delay encoding insulated IL-1R signaling, which transiently activates IKK2 and RelA by restricting Nfkb2 mRNA expressions, and accounted for the abrogated crosstalk in Trifdeficient MEFs that transiently activated the NF-κB pathway. Our studies also explained the requirement for the long-duration NIK-IKK1 signals in targeting this late acting p100 Nfkb2 feedback for RelA:p52 dimer generation. In contrast, IL-1 signal led to early induction of Nfkb2 mRNA expressions in the engineered cells, which inducibly express Nfkb2 transgene without the delay ( Figure 5D). Pretreatment of these engineered cells with αLTβR for 8 hr and subsequent IL-1 stimulation that effectively converged the non-canonical signal to IL-1-induced Nfkb2 feedback, potentiated p52 production ( Figure 5-figure supplement 1C) and prolonged IL-1-induced RelA response ( Figure 5G and Figure 5-figure supplement 1D). Correlating with the early onset of Nfkb2 mRNA induction in response to IL-1 treatment, observed crosstalk effects were indeed obvious within 1 hr of IL-1 treatment in these engineered cells.
In sum, we elucidate a crosstalk mechanism that discriminates between TLR4 engagement and concomitant cell activation through TLR4 and LTβR ( Figure 5H). Negative feedbacks by IκBα and p100/ IκBδ coordinately terminate canonical TLR4 response. But, Nfkb2 functions pro-synergistically upon costimulation; in a positive feedback loop, non-canonical LTβR signal targets the newly synthesized p100, abundantly produced by TLR4, to potently generate p52 and RelA:p52 dimers in sustaining inflammatory RelA NF-κB responses. Importantly, RelA:p50 and RelA:p52 heterodimers were shown to share DNA binding and gene-expression specificities (Siggers et al., 2012;Zhao et al., 2014). Our experimental data also indicated that RelA:p52 dimer has comparable efficiency in inducing the expression of Nfkb2 mRNA as the RelA:p50 dimer (Appendix-1, Appendix figure 4C). Although, emergent crosstalk is expected to be controlled by several biochemical constrains, the transcriptional delay intrinsic to the Nfkb2 promoter appears to be critical for the duration code and thereby the stimulus specificity.
Nfkb2 integrates lymphotoxin signal within intestinal niche to reinforce epithelial NF-κB responses to C. rodentium In addition to its role in lymph node development during embryogenesis, recent studies have illustrated a requirement for LTβR in innate immune responses in adult mice. Disruption of LTβR signal using LTβR-Ig fusion protein was shown to compromise innate immune responses upon subsequent infection with C. rodentium, a natural mouse enteric pathogen that led to mortality (Spahn et al., 2004;Wang et al., 2010). IEC-specific deletion of LTβR similarly obliterated bacterial clearance (Wang et al., 2010). The engagement of ligand-expressing RORγt + innate lymphoid cell is thought to  (Upadhyay and Fu, 2013). A requirement of epithelial RelA activity in the chemokine gene expressions has been documented earlier (Alcamo et al., 2001). Given our identification of a costimulatory function of LTβR in inflammatory RelA activation, we asked if signal integration via the NF-κB system could explain the epithelial requirement of LTβR in innate immune responses in vivo.
First, a functional non-canonical pathway downstream of LTβR ( Figure 6-figure supplement 1A) augmented the RelA activity induced by pathogen sensing TLR4 in otherwise hypo-responsive MSIE colon epithelial cell-line ( Figure 6A). Next, we biochemically analyzed NF-κB activation in IECs derived from WT mice intraperitoneally injected with antagonistic LTβR-Ig or a control-Ig 1 day prior to oral infection with C. rodentium. Upon colonization, Citrobacter initially triggered epithelial accumulation of p100 that was fully processed into p52 by day5 ( Figure 6B) generating RelA:p52 dimer ( Figure 6C) in control-Ig, but not LTβR-Ig, treated mice. Bacterial infection elicited RelA DNA binding activity in IECs that gradually accumulated in the nucleus ( Figure 6D,E) with substantial contribution from RelA:p52 dimer along with RelA:p50 dimer at day5 post-infection ( Figure 6-figure supplement 1B). Our supershift analyses further confirmed complete absence of RelB containing NF-κB DNA binding activity in IECs derived from infected mice ( Figure 6E). Perturbing LTβR signal attenuated RelA NF-κB activation with more obvious defects at day5 ( Figure 6D). Likewise, pathogen-responsive RelA activation in IECs derived from Nfkb2 −/− mice was severely weakened at day5 ( Figure 6F) that led to significantly reduced expressions of the RelA target chemokines encoding KC and MIP-2α as compared to WT mice ( Figure 6G). Indeed, infected Nfkb2 −/− mice exhibited diminished neutrophil recruitment in the lamina propria, as revealed by anti-myeloperoxidase immunostaining of the colon sections ( Figure 6H). Sustained epithelial RelA activity that relies on LTβR mediated processing of pathogen-induced p100 into p52, therefore, mirrored our MEF-based analyses depicting crosstalk between canonical and noncanonical signaling. Collectively, our results connected the previously reported epithelial requirement of LTβR (Wang et al., 2010) and NIK (Shui et al., 2012) in innate immune response to the NF-κB system in reinforcing RelA activity through Nfkb2 mediated crosstalk control. Subdued epithelial NF-κB activation, and not hyper-induction, in IECs from infected Nfkb2 −/− mice also suggested that a dominant precursor function of p100 supplying RelA:p52 dimer prolongs RelA response within the intestinal niche.

Stromal expression of Nfkb2 is required for limiting C. rodentium infection
While WT mice efficiently eliminated infections, increased fecal excretion of bacteria at day10 postinfection in Nfkb2 −/− mice ( Figure 7A) indicated an inadequacy in limiting local infection, thereby, correlating with the observed defects in the early innate inflammatory response in this knockout ( Figure 6). Histological analysis of the shrunken colons (Figure 7-figure supplement 1A,B) derived from the infected Nfkb2 −/− mice further revealed exacerbated damage with signatures of submucosal leukocyte infiltration ( Figure 7B). Breach in the intestinal barrier was accompanied by systemic bacterial dissemination with increased count in blood (left panel, Figure 7C) and liver (right, Figure 7C). Finally, bacterial colitis induced in Nfkb2 −/− mice resulted in significant body weight loss ( Figure 7D) and onset of mortality as early as day10 post-infection ( Figure 7E). Next, we performed reciprocal bone marrow transfer experiments between WT and Nfkb2 −/− mice to ensure that the observed sensitivity was not due to previously reported hematopoietic defects in Nfkb2 −/− mice (Caamaño et al., 1998). WT bone marrow cells in Nfkb2 −/− recipients (Figure 7-figure supplement  1C) were unable to prevent the infection-related colon pathology ( Figure 7F), reductions in the    Figure 7G) and morality ( Figure 7H). In contrast, WT recipients receiving either WT or Nfkb2 −/− bone marrow resolved infections with comparable efficiencies (Figure 7F-H).
In addition to modulating RelA activity through crosstalk, Nfkb2 also mediates RelB:p52 activation in response to singular non-canonical stimuli. Heightened RelB:p52 activation in B-cells was shown to strengthen host defense in Otu7b −/− mice (Hu et al., 2013). Our inability to rescue the infection-related mortality in Nfkb2 −/− mice using WT hematopoietic cells indicated that the protective function of Nfkb2 lies within stromal cells. Despite the presence of mesenteric lymph nodes (Lo et al., 2006), additional stromal defects in Peyer's patches in Nfkb2 −/− mice (Yilmaz et al., 2003) could impair mucosal IgA responses. Although IgM levels were comparable, very low IgA levels at day7, prior to the onset of mortality in Nfkb2 −/− recipients, precluded bona fide comparisons in the radiation chimeras. Interestingly, adult mice, similar to those utilized in our study, lacking IgA or its transporter pIgR efficiently resolved Citrobacter infections . μMT mice depleted of peripheral B-cells presented few mucosal changes in the first 2 weeks of infection and were completely free of mortality (Simmons et al., 2003). Finally, Rag1 −/− mice succumbed to Citrobacter only late during infections in the 4th week (Wang et al., 2010). Though, our study does not rule out crosstalk independent engagement of the Nfkb2 pathway in activating RelB:p52 dimer in the hematopoietic compartment at a later stage of infection, epithelial RelA NF-κB activation defects coupled to aggravated early colon pathology and early onset of mortality in Nfkb2 −/− mice suggested that the stromal requirement of Nfkb2, at least in part, lies within the intestinal epithelial cells in the initial events controlling early innate immunity and involves crosstalk regulation of RelA NF-κB activity ( Figure 7I, also discussed later).

Discussion
The NF-κB system transduces signals from a variety of cell-activating stimuli. Our study suggested that such pleiotropic system enables tuning of cellular response to an instructive signal by microenvironment-derived additional costimulatory signals. However, a duration code selectively integrates costimulatory LTβR signal with the TLR4 pathway, insulating transient cytokine signaling, secondary to microbial infections, from crosstalk amplifications. Recent studies have identified important positive feedback regulations underlying dose threshold control of NF-κB response in Bcells (Shinohara et al., 2014) or in myeloid cells (Sung et al., 2014). Our crosstalk study illuminated a role of a positive feedback loop in sustaining NF-κB response; where non-canonical LTβR signal prolonged canonical TLR4 response by targeting RelA induced p100 for the generation of RelA:p52 NF-κB dimer. Indeed, a dominant precursor function of p100 in producing RelA:p52 led to ablated crosstalk in Nfkb2 −/− cells. As such, non-canonical signal generates RelB:p52 by removing the Cterminal domain of p100 present in the preexisting RelB:p100 dimeric complex (Sun, 2012) or liberates RelA:p50 dimer from the p100/IκBδ-inhibited complexes (Basak et al., 2007). Signal generation of the RelA:p52 dimer requires both canonical induction of p100/Nfkb2 expressions and concomitant processing of the newly synthesized p100 into p52 by non-canonical signal. Previous report demonstrating that p100 could be efficiently processed cotranslationally (Mordmüller et al., 2003) explained the requirement of canonical signals in amply synthesizing nascent p100 as a substrate for the abundant production of p52 subunit during crosstalk. More so, mature p52 could readily dimerize with RelA, despite a preference of full-length p100 for RelB binding (Fusco et al., 2008). These observations along with our current study, thereby, elaborated a requirement of convergence of canonical and non-canonical signals in synergistically generating and activating RelA: p52 dimer. Nevertheless, consistent to the observed gene-effect of crosstalk in our study, a significant overlap between RelA:p50 and RelA:p52 dimers in DNA binding (Siggers et al., 2012) and proinflammatory gene expressions (Hoffmann et al., 2003) was reported earlier. Of note, our results rely on bulk measurements of signaling intermediates and deterministic modeling approaches. Given studies documenting cell-to-cell variations in signal-induced NF-κB responses (Lee et al., 2009), it would be interesting to examine the potential implication of signal integration at the single cell level in determining cellular heterogeneity.
TLR4 activation of epithelial RelA was implicated in the chemokine gene expressions and neutrophil recruitment upon bacterial infections (Khan et al., 2006). Yet, epithelial LTβR (Wang et al., 2010) was also important for effective innate immune responses to Citrobacter. In our proposed model ( Figure 7I), we could clarify that LTβR provides a critical costimulatory signal through Nfkb2 to sustain RelA NF-κB response to pathogens in otherwise hyporesponsive colonic epithelial cells. Such signal integration ameliorated innate immune functions by enhancing pro-inflammatory gene expressions. Interestingly, p100 −/− mice, which lacked the expression of p100, but aberrantly produced p52, revealed hyperplasia of gastric epithelial cells and elevated expressions of RelA target genes (Ishikawa et al., 1997). In Nlrp12 −/− mice, robust p52 generation in stromal cells through the non-canonical pathway led to colon cancer associated inflammation (Allen et al., 2012), a hallmark for aberrant RelA activity. These studies indeed support a possible role of Nfkb2 in mucosal epithelial cells in strengthening RelA activity. LTβR engagement in the dendritic cells within colonic patches was shown to trigger IL-22 production by innate lymphoid cells involving cell-cell communications to potentiate gut immunity (Tumanov et al., 2011). Although, a defect in the colonic patches in Nfkb2 −/− mice could impair IL-22 mediated protective responses, our cell-intrinsic crosstalk model explained that the reported epithelial requirement of LTβR (Wang et al., 2010) is in sustaining RelA NF-κB response during bacterial infection. In future, tissue-specific knockouts may help to further distinguish between innate immune functions of Nfkb2 in different cell types.
The underlying mechanism and biological functions of RelB:p52 dimer activated by non-canonical signal is well established. From the perspective of signaling crosstalk, our study offers a significant revision of our understanding of non-canonical signaling in amplifying canonical RelA responses. It extends the intriguing possibility that cell-differentiating cues present in the tissue microenvironment may play a more direct role, separate from merely determining the differentiation states of the resident cells, in calibrating innate immune responses by engaging into cell-autonomous signaling crosstalks. Future studies ought to further examine the signal integration via Nfkb2 in potentiating immune responses against other microbial pathogens. More so, the potential involvement of the deregulated crosstalk control in the pathophysiology of inflammatory disorders, particularly those involving gut, remains to be addressed.

Mice, cells and recombinant DNA
Wild-type or gene-deficient C57BL/6 mice were housed at NII small animal facility and used in accordance with the IAEC guidelines. Primary MEFs were generated from E12.5-14.5 embryos. Late passage Trif-deficient and NIK-deficient MEFs have been described (Basak et al., 2007). MSIE cell line was a gift from R. Whitehead, Ludwig Cancer Research. Mouse Nfkb2 was stably expressed from a promoter containing five tandem kappaB sites from HRS.puro or from a constitutive promoter from pBabe.puro retroviral constructs.

Gene expression studies
Total RNA was isolated using RNeasy Kit (Qiagen, Venlo, Netherlands). For microarray analysis, labeling, hybridization of RNA samples to the Illumina MouseRef-8 v2.0 Expression BeadChip, data processing and quantile normalization was performed by Sandor Pvt. Ltd (Hyderabad, India). We have considered genes that are induced at least 1.3 fold by LPS at 24 hr in representative data sets and has a detection p-value < 0.05 for LPS, αLTβR and co-treatment regimes. Next, LPS response genes in WT MEFs were ranked based on the merit of their normalized crosstalk scores, which is defined below and also described earlier (Zhu et al., 2006).
Crosstalk score = [(Δco-treatment − (ΔLPS + ΔαLTβR))/0h_int], where 0h_int indicates the signal intensity of a given gene in untreated cells and Δtreatment signifies the differences in signal intensities between treated and untreated cells, Normalized crosstalk score = Crosstalk Score * [{(ΔL + ΔB))/0h_int/|(ΔL + ΔB))/0h_int|], As implied, positive crosstalk scores signify hyperactivation, whereas negative crosstalk scores imply diminished gene expressions in the co-treatment regime as compared to cell treatment with the individual stimuli. The ordered gene set was examined in GSEA v2.0.12 (Broad Institute at MIT) (Subramanian et al., 2005). The MIAME version of the microarray data set discussed in this publication are available on NCBI Gene Expression Omnibus (accession number GSE62301). For quantitative RT-PCR, total RNA was reverse transcribed with Transcriptor cDNA kit and amplified using Sybr Green PCR mix (Roche, Mannheim, Germany) in ABI7500 cycler. The relative gene expressions were quantified using ΔΔCT method upon normalizing to β-actin mRNA level. Absolute quantification was done using plasmid DNA constructs encoding respective genes as standards and normalized to express as gene/actin mRNA level.

Murine infection model
Sex matched, 8 to 10 week old mice, fasting for 8 hr, were orally gavaged with 1.2 × 10 10 cfu of C. rodentium strain DBS100 (ATCC 51459). In certain instances, mice were intraperitoneally injected with 200 μg of murine LTβR-IgG1 fusion protein or MOPC21 isotype control (Biogen Idec) 1 day prior to infection, as described (Wang et al., 2010). IECs, isolated following published procedure (Greten et al., 2004), were utilized for biochemical analyses. For histology, dissected colons were fixed in 10% neutral buffered formalin. Paraffin-embedded tissue sections were stained with anti-myeloperoxidase antibody (Pierce, Waltham, Massachusetts, USA) for neutrophil recruitment or with Hematoxylin and Eosin (H&E) for tissue pathology evaluation. Fecal samples were weighed, homogenized, and serially diluted homogenates were plated on MacConkey agar (HiMedia, Mumbai, India) to score for C. rodentium. Similarly, spleens and livers were aseptically removed and assessed for bacterial load. For bone marrow chimera experiment, recipient WT or Nfkb2 −/− mice were lethally irradiated and marrow cells from the indicated donor mice were transferred. After 6-8 weeks, mice were infected.

Computational modeling
The NF-κB Systems Model v1.0 was simulated in Matlab (v. 2012b, Mathworks, Natick, MA, USA) using the ode15 s (Basak et al., 2007). A detailed description of the model has been provided in the Appendix-1. To estimate crosstalk sensitivity, each parameter values were individually increased and decreased by 10%, euclidean distances were used to determine the resultant changes in the crosstalk indexes as compared to the unperturbed system, averaged for a given parameter and normalized to nominal crosstalk index.

Statistical analysis
Data are expressed as mean of 3-5 quantified biological replicates ± SEM. Statistical significance was calculated by two-tailed Student's t-test. For survival curves, log rank (Mantel-Cox) test was conducted.

Additional files
This article additionally contains (i) 10 figure supplements associated with the main text, (ii) three Supplementary tables (Supplementary files 1-3) and five Appendix figures associated with the an Appendix file (Appendix-1), which provides a detailed description of the mathematical model, as well as a file describing Matlab source codes. The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Additional information
Author contributions BB, Acquisition of data, Analysis and interpretation of data, Drafting or revising the article; BC, BV, MVRP, PR, Acquisition of data, Analysis and interpretation of data; SB, Conception and design, Analysis and interpretation of data, Drafting or revising the article

Ethics
Animal experimentation: Wild-type or gene-deficient C57BL/6 mice were housed at NII small animal facility and used strictly in accordance with the Institutional Animal Ethics Committee guidelines of the institute. The protocol was approved by the committee with the approved protocol no: IAEC#258/11 (for embryonic fibroblast cell collection) and IAEC#313/13 (for infection related studies).

Major dataset
The following dataset was generated: Publicly available at the NCBI Gene Expression Omnibus (GSE62301).
Standard used to collect data: The MIAME version of the microarray data set discussed in this publication are available on NCBI Gene Expression Omnibus (accession number GSE62301).