In RNA virus–infected cells, retinoic acid–inducible gene-I–like receptors (RLRs) sense foreign RNAs and activate signaling cascades to produce IFN-α/β. However, not every infected cell produces IFN-α/β that exhibits cellular heterogeneity in antiviral immune responses. Using the IFN-β–GFP reporter system, we observed bimodal IFN-β production in the uniformly stimulated cell population with intracellular dsRNA. Mathematical simulation proposed the strength of autocrine loop via RLR as one of the contributing factor for biphasic IFN-β expression. Bimodal IFN-β production with intracellular dsRNA was disturbed by blockage of IFN-α/β secretion or by silencing of the IFN-α/β receptor. Amplification of RLRs was critical in the generation of bimodality of IFN-β production, because IFN-βhigh population expressed more RLRs than IFN-βlow population. In addition, bimodality in IFN-β production results in biphasic cellular response against infection, because IFN-βhigh population was more prone to apoptosis than IFN-βlow population. These results suggest that RLR-mediated biphasic cellular response may act to restrict the number of cells expressing IFN-β and undergoing apoptosis in the infected population.
Transcriptional activation of type I IFN (including IFN-α and IFN-β) is the critical step mediating the immune response against viral infection. Type I IFN is induced in most cell types infected with virus and transmits signals through the type I IFN receptor (IFNAR) complex to induce various IFN-stimulated genes (ISGs), which play key roles in innate and adaptive immune responses (1, 2). Beyond direct restriction of viral replication, ISGs also function to control cell proliferation and death (2, 3). For example, induced ISGs, such as dsRNA-activated protein kinase, IFN regulatory factor (IRF) 1, caspases, regulators of IFN-induced death, or Fas, sensitize host cells to apoptosis-inducing stimuli (4). Because of their protective and destructive effects, fine-tuned regulation of type I IFN production during viral infection is essential to maintain the integrity of the infected host.
It has long been reported that not every infected cell produces type I IFN; there is heterogeneity in the IFN-α/β production within populations of virus-infected immune or nonimmune cells (5–10). Heterogeneity in IFN-β expression was not due to insufficient virus infection or artifacts in the cell culture but rather was an intrinsic property of the IFN-β production system, because the percentage of the cells expressing IFN-β was subject to change (5, 8). Although it is not clear what determines the IFN-β heterogeneity or what is the physiological relevance of it, stochastic expression of RIG-I signaling components or heterogeneous chromatin status of IFN-β genomic locus recently has been suggested as possible factors controlling the IFN-β heterogeneity in the virus-infected cell population (5, 7, 10).
In addition to the stochastic nature of signaling components in signal transduction and transcriptional regulatory networks, interlinked feedback loops have been also reported to generate heterogeneous expression of target gene in the uniformly stimulated cell population by creation of two or more stable states (11–13). Genetically identical bacteria population exhibits heterogeneous phenotype depends on the feedback loop of network configurations and unimodal noise of regulatory factors (14, 15). The interlinked Cdc42/Cdc24 feedback loop controlling actin polymerization converts continuous inputs into discrete differential outputs and determines the stability of polarized “on” or “off” states (16). In summary, multiple feedback loops often generate nonhomogeneity in response to the homogeneous input signal, hence, controls various cellular response, such as cell fate decision, oscillatory behaviors, and memory of transient input stimulus (16–19).
Interestingly, RIG-I–like receptors (RLRs)–mediated IFN-β induction system is comprised with multiple feedback loops involving large number of ISGs. RNA virus-infected cells rely RLRs, which include retinoic acid-inducible genes I (RIG-I) and melanoma differentiation–associated gene 5 (MDA5), for the recognition of intracellular viral RNAs and production of type I IFNs (20–23). Both RIG-I and MDA5 are ISGs, which are strongly induced by type I IFNs and strengthened virus-induced production of type I IFN pathways (24, 25). Robust production of type I IFN requires IFN-inducible transcription factor IRF7 (26, 27), and protein stability of RLRs is controlled by posttranslational modification that are induced by type I IFN (28–30). Furthermore, type I IFN-inducible LGP2 competes with RLRs for cytosolic non–self-RNA binding and activates MAVS/IPS-1 in mitochondria, thus creating another feedback loop (24, 31, 32).
Therefore, we attempted to investigate the contribution of feedback loops in RLRs-mediated IFN-β expression system as one of the possible origin of heterogeneous IFN-β expression in the infected cell population. Using the IFN-β–GFP reporter system and mathematical modeling, we explored the characteristics of RLR–IFN-β signaling system and the fate of the infected host cells.
Materials and Methods
Construction of the IFNβ1p-GFP-IFNβ1UTR expression plasmid
A genomic fragment derived from the promoter (−3435 ∼ +1) and 5′-untranslated region (UTR) of the human IFNβ1 gene was PCR amplified using a primer set (5′-TCTAGAGCTGTGGATCATCATGGTAT-3′ and 5′-GGATCCGTTGACAACACGAACAGTG-3′) and subsequently cloned into the ApaI and BamHI sites of the pcDNA3.1 mammalian expression vector (Invitrogen). The CMV promoter in the pcDNA vector was deleted by blunt end ligation of BglII and NheI sites. A DNA fragment from the human IFNβ1-3′UTR region (+640 ∼ +3265) was PCR amplified using a primer set (5′-GCG GCC GCA GATCTCCTAGCCTGTGCC-3′ and 5′-GCCGGCCTCTGCTTGCAG-3′) and subsequently cloned into the NotI and KpnI sites of the pcDNA3.1-IFNβ1p+5′UTR plasmid. Finally, the protein coding regions of GFP were subcloned from the pEGFP-N1 (BD Clonetech) plasmid into the BamHI and NotI sites of the pcDNA3.1-IFNβ1p+5′UTR+3′UTR, thereby generating an IFNβ1p-GFP-IFNβ1UTR expression plasmid. The DNA sequences of the genomic fragment obtained by PCR were verified using automated DNA sequencing.
Generation of IFNβ1p-GFP-IFNβ1UTR stable cell lines
HepG2 cells were maintained in MEM containing 10% FBS (Hyclone, Logan, UT) and 1% penicillin–streptomycin (Life Technologies). Cells were transfected with the IFNβ1p-GFP-IFNβ1UTR expression plasmid using Lipofectamine 2000 (Invitrogen) reagent, followed by selection in G418 (2 mg/ml; Life Technologies)-containing media for 2 wk. Single colonies were picked and cultured in the media containing G418 (1 mg/ml).
Southern blot analysis
Genomic DNA (10 μg) prepared from individual clones was digested with the appropriate restriction enzymes and separated on 1% agarose gels. Fragmented DNA was transferred onto the Hybond-N+ membrane (Amersham Biosciences) under high salt conditions. The membrane was incubated with a 32P-labeled probe in hybridization solution (5× SSC, 10× Denhardt’s solution, 0.1% SDS, and 100 μg/ml salmon sperm DNA) overnight at 65°C, washed in low salt conditions, and then analyzed with a Bio-imaging analyzer (Fujifilm).
Cells were transfected with the indicated concentration of polyinosinic–polycytidylic acid (polyI:C) (Amersham Biosciences), purified bacterial genomic DNA, or salmon sperm DNA (Roche) using Lipofectamine 2000 reagent, or infected with Sendai virus (provided by Dr. P. Palese, Mount Sinai School of Medicine, New York, NY) before FACS analysis. GFP-expressing cells were detected by flow cytometry (FACSCalibur; BD Biosciences) and analyzed with FlowJo (Tree Star). To visualize intracellular polyI:C, polyI:C was labeled with Cy-5 using LabelIT nucleic acid labeling kit (Mirus) following the manufacturer’s instructions. Cy-5–labeled polyI:C was concentrated by ethanol precipitation. For cell sorting, the MoFlo XDP Cell Sorter (Beckman Coulter) was used. To observe apoptosis, cells were incubated in a staining solution containing AnnexinV-PE (BD Biosciences) and 7-aminoactinomycin D (Sigma-Aldrich) before FACS analysis.
After stimulation, cells were fixed with 4% paraformaldehyde and then permeabilized by 0.2% Trion X-100 solutions. Anti-IRF3 Ab (Santa Cruz Biotechnology), anti–RIG-I Ab (Alexis Biochemicals), anti-rabbit IgG conjugated with Alexa Fluor 633, and anti-mouse IgG conjugated with Alexa Fluor 568 (Invitrogen) were used to detect IRF3 and RIG-I, and the nucleus was stained with Hoechst 33258 (Sigma-Aldrich). Fluorescence signals were measured by fluorescence confocal microscopy (FV1000; Olympus).
Live cell imaging and analysis
Cells were transfected with 15 μg/ml polyI:C and cultured on a specimen chamber maintained at 37°C with 5% CO2 for live cell imaging. GFP expression was visualized by fluorescence confocal microscopy (FV1000; Olympus), and the intensities of fluorescence on serial images were analyzed using MetaMorph software (Universal Imaging, Philadelphia, PA). The threshold intensity of GFP to distinguish GFPhigh cells from GFPlow cells was determined by the intersection point of fluorescence distributions.
Western blot analysis
Cells were lysed in lysis buffer (25 mM Tris [pH 7.5], 150 mM NaCl, 1% Triton X-100, 0.1% SDS, and 0.5% deoxycholate with protease inhibitors), and 20 μg of total cell lysate was separated on SDS polyacrylamide gels. Proteins were transferred onto a nitrocellulose membrane and analyzed with Abs against GFP (Santa Cruz Biotechnology), RIG-I (Alexis Biochemicals), MDA5 (ProSci), poly(ADP-ribose) polymerase 1 (PARP1; Santa Cruz Biotechnology), cleaved caspase-3 (Cell Signaling Technology), and GAPDH (Chemicon). Immunoreactive signals were visualized using a LAS4000 luminescent image analyzer (Fujifilm).
Cells were transfected with the appropriate small interfering (si)RNA using Lipofectamine 2000. The following sequences were used: siControl, 5′-GAAUUUGCACGAAAACGCC-3′; siRIG-I, 5′-GAGGUGCAGUAUAUUCAGG-3′; siMDA5, Qiagen-validated siRNA (S103649037); and siIFNAR1, 5′-GGUGCAAGAGGAAGAAGAA-3′. siControl, siRIG-I, and siIFNAR1 were synthesized by the custom design service of Qiagen.
To measure cell viability, cells were incubated with 0.5 mg/ml MTT (Sigma-Aldrich) in a 1× PBS solution for 3 h. The precipitates were then dissolved in an isopropanol: DMSO (9:1) solution. The absorbance of the colored solution was quantified using a spectrophotometer.
Total RNA (1 μg), which was extracted from HepG2 cells using TRIzol reagent (Invitrogen), was reverse transcribed using the Improm-II reverse transcription system (Promega) and subjected to quantitative real-time PCR using the following primer sets: RIG-I, 5′-GCCATTACACTGTGCTTGGAGA-3′ and 5′-CCAGTTGCAATATCCTCCACCA-3′; MDA5, 5′-GAGGAGTATGCTCATAAG-3′ and 5′-CCTCATCACTAAATAAAC-3′; LGP2, 5′-GATCCTGTGGTCATCAACA-3′ and 5′-TCAGTCCAGGGAGAGGTC-3′; IRF7, 5′-AGCTGGTGGAATTCCGGG-3′ and 5′-CTAGGCGGGCTGCTCCAG-3′; IFN-β, 5′-TGCTCTCCTGTTGTGCTTCTCC-3′ and 5′-CATCTCATAGATGGTCAATGCGG-3′; IFNAR1, 5′-ATCGGTGCTCCAAAACAGTC-3′ and 5′-TTTCATCCATGGTGTGTGCT-3′; IFNAR2, 5′-AGTCCACTCCAGGACCCTTT-3′ and 5′-TCCTCTGGGTCAACCATCTC-3′; RANTES, 5′-ATGAAGGTCTCCAAAGAG-3′ and 5′-GCTCATCTCCAAAGAG-3′; ISG15, 5′-CCTCTGAGCATCCTGGT-3′ and 5′-AGGCCGTACTCCCCCAG-3′; and β-actin, 5′-TCAT GAAGTGTGACGTTGACATCCGT-3′ and 5′-CCTAGAAGCATTTGCGGTGCACGATG-3′. β-Actin was used for normalization of the expression values.
Hartigan’s DIP test (33) was used to determine the multimodality of the FACS data shown in Fig. 2A. All nonzero GFP intensity values from the original FACS data were used, and the sample distribution was compared with a random uniform distribution. Matlab codes (http://www.nicprice.net/diptest/) were used for this test.
Mechanism-driven ordinary differential equation model for RLR–IFN-β system
Our mathematical model for IFN-β production by cytosolic RLR-mediated signal transduction pathways in a single cell(i) was simplified by focusing on three components that constitute positive feedback loops via receptor (RLR: RIG-I and MDA5), active transcription factor (TF), and IFN-β (Fig. 3A). To express their mRNA and protein status, five variables sere used: IFNMi (IFN-β mRNA), IFNPi (IFNβ protein), RECMi (RLR mRNA), RECPi (RLR protein), and TFAi (active transcription factors). The RLR–IFN-β model system was described using nonlinear ordinary differential equations for the regulatory mechanism. We assumed that the synthesis of RECMi by IFN-β signaling, IFNMi by active transcription factors, and the increase of TFAi by stimulus-associated RLR followed Michaelis–Menten kinetics and was approximated by Hill equations (Supplemental Fig. 1A). Biochemical reactions shown in Supplemental Fig. 1C and 1D were expressed using following terms: TFa (Active transcription factor), IFNpro (free IFNβ gene; not activated), TFa-IFNpro (Activated IFNβ gene), IFNpro,total (total IFNβ gene), IFNP (IFNβ protein signal intensity from IFNβ protein–bound IFNAR receptor), RECpro (free receptor gene; not activated), and IFNP-RECpro (activated receptor gene by IFNP signal). All variables including time (T) were then nondimensionalized with their intrinsic synthesis rates, total amount of transcription factors, or translation rates.
Fixed point analysis
The fixed points of the system were calculated by roots-finding of analytical expanded model equations at steady states. The local stability of a fixed point was determined using linear stability analysis, involving the determination of the eigenvalues of associated Jacobian matrices. The bifurcation diagrams were expressed with various stimuli (sn) values (Supplemental Fig. 1E).
Population model simulation
To express the intrinsic noise effect among individual cells, a log-normal distribution of 1/dnI,i, the initial states of IFNMi, were given with mean of 1/dnI (dnI value indicated at Supplemental Table I) and variance 0.59, which was derived from experimentally determined noise level of basal GFP signal. Similarly, the distribution of 1/dnR,i, the initial states of RECMi, was set as a log-normal distribution with mean 1/dnR and variance 0.59. Stimuli, sni, were also given as a log-normal distribution of mean sn and variance 1.07, which was derived from experimental noise of intracellular polyI:C quantified on the FACS analysis. For simulations of the experiments involving brefeldin A treatment, each IFNPi was assumed to be reduced by β (multiply β on IFNPi), a constant reduction ratio, for the given time period. For the RLR inhibition simulation, γ, a constant number, was multiplied on dnR to increase the degradation of RLR mRNA. The dynamics of each variable were numerically calculated with the Runge-Kutta fourth method. For parameter values used in the simulations, see Supplemental Table I.
Construction of IFNβ1p-GFP-IFNβ1UTR reporter cells that mimic endogenous IFN-β expression
To investigate the RLR–IFN-β expression in a single-cell level, cellular monitoring system for IFN-β production was needed. Thus, we constructed an IFNβ1 reporter cassette with a 7.5-kb DNA fragment spanning the genomic locus of human IFNβ1 (Fig. 1A). The coding region of GFP was flanked by the promoter, 5′-UTR, and 3′-UTR sequences derived from human IFNβ1 to reflect transcriptional regulation of IFNβ1. To ensure cellular homogeneity, stable cell lines that harbor the IFNβ1p-GFP-IFNβ1UTR expression vector were then selected in the human liver carcinoma-derived HepG2 cell line, in which the RLRs-mediated antiviral signaling pathways are known to be functional (Fig. 1B) (34, 35). Because the endogenous IFNβ1 locus was not altered, the feedback loops mediated by IFN-β remained intact in these IFN-β–GFP reporter cells (Fig. 1A). The integrity and copy number of the genome-integrated IFN-β–GFP reporter constructs were verified by Southern blot analysis of individual stable clones (Fig. 1C). In the polyI:C-stimulated IFN-β–GFP reporter cells, GFP mRNA expression was similarly controlled as IFN-β mRNA expression in a stimulus-dependent manner. Production of GFP protein also follows expression dynamics of GFP mRNA (Fig. 1D), indicating that endogenous IFN-β expression can be properly monitored by GFP expression in the IFNβ1p-GFP-IFNβ1UTR reporter cells.
Stimulated cells with intracellular dsRNA exhibit bimodal IFN-β expression
Most virus harbors one or more modulators of antiviral immune system in their genome (36). Furthermore, accumulation of the intracellular viral RNA that serves as input stimulus for RLR proteins is subject to change over time during infection, depending on the strength of antiviral immune responses. Therefore, it is hard to investigate the intrinsic nature of the antiviral signaling system, when one uses intact replicating virus as an input stimulus. To focus on the intrinsic properties of host cells against intracellular non–self-RNA, we transfected cells with polyI:C, a synthetic dsRNA, to mimic the RNA virus infection, without viral interference. Using two-color FACS analysis, we simultaneously monitored the IFN-β (GFP) expression and the Cy5-labeled polyI:C stimulation at the single-cell level (Fig. 2A). In ∼90–95% of transfected cells, intracellular Cy5-polyI:C was detected after 8 h of incubation. Among cells that harbored the equal amounts of intracellular Cy5-polyI:C, only substantial portions of cells produced GFP while others did not. In the case of stable clones 4-4, for example, only 37.0% of cells were GFP-positive (GFPhigh), and the remaining 60.1% cells were GFP-negative (GFPlow) with intracellular Cy5-polyI:C presence. The same phenomenon was repeatedly observed in multiple stable clones examined (Fig. 2A). Thus, we could exclude the possibility of genetic heterogeneity at the integrated IFNβ1p-GFP-IFNβ1UTR locus. The DIP test confirmed bimodality of GFP distributions in the majority of tested clones, indicating that intracellular dsRNA signals act to generate two distinct populations of IFN-β–producing cells in the stimulated population. To ensure that the nonresponsiveness to intracellular polyI:C observed in the GFPlow cells was not due to delayed IFN-β production, we followed GFP expression of 96 individual cells for 14 consecutive hours (Fig. 2B). Cells that expressed GFP at later stage exhibited significantly higher values of fluorescence intensity than GFP nonexpressing cells at every time point examined, indicating that there are two distinct cell populations present in the population stimulated with intracellular dsRNA.
To examine whether this phenomenon is specific to intracellular dsRNA, we infected cells with Sendai virus or transfected them with foreign DNA, such as salmon sperm DNA or bacterial genomic DNA of Pseudomonas aeruginosa or Escherichia coli, and examined GFP expression (Fig. 2C). Like polyI:C stimulation, infection with Sendai virus exhibited similar bimodality in GFP expression. However, cytosolic non–self-DNA stimulation–induced unimodal IFN-β (GFP) expression, indicating that bimodal IFN-β response against intracellular RNA is a specific event. To verify the differential expression of IFNβ, endogenous IFNβ transcripts were then directly compared in the sorted GFPhigh and GFPlow cells (Fig. 2D). Sorted GFPhigh cells expressed significantly higher levels of IFN-β mRNA compared with GFPlow cells, confirming distinctive IFN-β production in the cells with the same stimulus. In addition, when the strength of input stimulus increased, the fraction of IFN-β–expressing GFPhigh population increased, and the fraction of IFN-β–nonexpressing GFPlow population decreased accordingly (Fig. 2E). If the RLR–FNβ production system is unimodal, then the GFPlow population at unstimulated condition will gradually shift to the right as the strength of stimulus increased. Instead, a discrete fraction with higher mean value appeared, and the percentage of GFPhigh population changed depending on the input stimulus, indicating that the strength of the input stimulus acts to facilitate the transition of cells from GFPlow to GFPhigh in the RLR–IFN-β expression system. On the basis of these results, we concluded that there are at least two different steady states—“on” or “off” states—present in the RLR–IFN-β expression system, and this phenomenon is specific to IFN-β responses against intracellular foreign RNA.
A model simulation of RLR–IFN system with autocrine feedback loop demonstrates bimodal IFN-β expression
Positive feedback loops in the signal or transcription regulatory networks play key roles in the generation of two or multiple stable states (11, 37), which contributes to bi- or multimodality in the uniformly stimulated cell population. Although stochastic RLR expression has been recently proposed as the source of heterogeneous IFN-β production in the virus infected cell population (10), the bimodality observed in the stimulated population proposed the existence of multiple stable states, possibly originated from positive feedback loops present in the RLR–IFN system. To evaluate the contribution of positive feedback loop in the bimodality of RLR–IFN-β expression system, we therefore set up a population model that consisted of multiple cells, each single cell carrying an IFN-β expression system with positive feedback loop mediated by RLR (Fig. 3A, Supplemental Fig. 1A, 1B). In this model system, the secreted IFN-β acts on the secreted cell itself (autocrine) or on neighboring cell (paracrine).
With estimated parameters using quantitative experimental data (Supplemental Fig. 1C, 1D, Supplemental Table I), model simulation of IFN-β dynamics in polyI:C-stimulated cell population exhibited comparable bimodality to that appeared in the experimental IFN-β kinetics (Fig. 3B). In the steady-state analysis of the single-cell system, both IFN-β and RLR showed two discrete stable states for the same input stimulus (bistability: Supplemental Fig. 1E), which contributed bimodality in the population model simulation. Because the feedback loop in the RLR–IFN system is based on the secreted IFN-β signaling, we tested whether the action mode of IFN-β, either autocrine or paracrine, matters in the maintenance of the bimodality of system. Specifically, we questioned whether strong paracrine stimulation of IFN-β dilutes the feedback effect on a single-cell level, hence disrupts the bimodal IFN-β expression in the stimulated cell population. Generally, bimodality became weakened when paracrine effect increased (Fig. 3C). However, when the system was simulated with the parameter condition exhibiting strong bistability in a single-cell level, bimodal distribution can be maintained even in the paracrine-only situation (Fig. 3C, Supplemental Fig. 1E). Therefore, the positive feedback structure of the RLR–IFN system was critical in the maintenance of the bimodal IFN-β production by intracellular dsRNA.
Secreted IFN-β–mediated amplification of RLR contributes the bimodality of IFN-β production with intracellular dsRNA
We wondered what makes bimodal population in the IFN-β production system. Because the stochastic expression of RLRs has been proposed as a leading source of IFN-β heterogeneity in the virus-infected cells (10), we directly compared the expression of RIG-I and MDA5 proteins in the sorted GFPhigh and GFPlow populations. Although expression levels of IFNAR1 and 2 were similar between GFPlow and GFPhigh cells (Supplemental Fig. 2A), induction of RIG-I and MDA5 upon stimulation were greater in the GFPhigh cells compared with GFPlow cells, and as a result, significantly higher levels of RLRs were observed in the high IFN-β–producing GFPhigh cells (Fig. 4A). Similarly, expressions of various ISGs, such as ISG15, LGP2, and RANTES in GFPhigh cells, were higher than those expressions in GFPlow cells (Supplemental Fig. 2A). To understand whether all transfected cells recognize the intracellular dsRNA similarly or its recognition is restricted to the GFPhigh cells, translocation of the activated IRF3 into the nucleus has been examined, along with RIG-I expression (Supplemental Fig. 2B). Under polyI:C or Sendai virus infection condition, nuclear IRF3 was observed in the subset of cells, which expresses relatively higher levels of RIG-I. These results clearly indicate that a subset of cells that stochastically expresses more RLRs was able to respond to stimulus, activate signaling molecules such as IRF3, and produce IFN-β, whereas other cells that stochastically express less RLRs remained poorly responsive to stimulus.
Therefore, we next evaluated the effects of RLRs and secreted IFN-β in the bimodal production of IFN-β. In the simulated condition with increased RLR degradation that mimics knockdown effect of RLR, the IFN-β–producing cell population disappeared (Fig. 4B). Similarly, when RLRs were depleted by RIG-I and MDA5-specific siRNA treatment, GFPhigh population didn’t appear, even after polyI:C stimulation (Fig. 4C). When secretion of IFN-α/β was reduced in the model simulation, diminished IFN-β–producing population was observed (Fig. 4D). In the experimental system, blockage of IFN-α/β signal was conducted in three different experimental conditions. First, we used brefeldin A, a chemical that inhibits intracellular protein transport, to block IFN-α/β secretion (38, 39). As a result, the generation of the GFPhigh population was efficiently suppressed (Fig. 4E). Second, depletion of IFNAR1, a subunit of the IFN-α/β receptor, also reduced the percentage of GFPhigh populations (Fig. 4F). Finally, blockage of type I IFNs with neutralizing Abs exhibited the same effect (Fig. 4G).
Because it is known that priming cells with low levels of IFN can enhance induction of IFN upon virus infection (40), we wondered whether the small differences in the basal IFN-β results in great differences in the IFN inducibility upon infection in our system. To answer this question, cells were treated with various dose of IFN-β, and IFN-β bimodality was determined after intracellular polyI:C stimulation (Supplemental Fig. 2C). However, bimodal expression of IFN-β was still observed, indicating that the heterogeneous IFN-β at the basal state might not be the main controlling factor of biphasic IFN-β expression in response to cytosolic non–self-RNA. Separately, we also examined whether the cell cycle or physiological state of cells affects bimodality of IFN-β. For this purpose, cells were pretreated with low (0.2% FBS) or normal (10% FBS) serum containing media, prior to polyI:C stimulation in either low or normal serum containing media. However, none of the culture condition significantly affects bimodality of IFN-β production (Supplemental Fig. 2D). These results altogether indicate that amplification of RLRs by IFN-β feedback loop might be the main contributor to the bimodal expression of IFN-β by intracellular non–self-RNA.
Biphasic determination of cell fate in populations with intracellular dsRNA
Because viral infection associated type I IFN signal not only induces antiviral response but also provokes cellular toxicity, we examined whether the RLRs–IFN system function as a biphasic switch to control cell fate upon intracellular non–self-RNA. Intracellular polyI:C mimics the behavior of RNA virus infection, because it induces IFN-β expression together with cell death in a dose-dependent manner (Supplemental Fig. 3A). The signs of apoptosis were measured by the PARP1 and procaspase-3 cleavage (Supplemental Fig. 3B) and annexin V staining in the polyI:C-transfected cells (Supplemental Fig. 3C).
The experiments were then conducted with sorted GFPhigh and GFPlow cells to investigate the fate of IFN-nonproducing and -producing cells upon intracellular dsRNA stimulation. Strikingly, most of the cells that underwent apoptosis were found in the GFPhigh population (Fig. 5A, 5B). Both PARP1 cleavage and caspase-3 cleavage were observed only in cells with IFN-β (GFP) expression. These cells were stained by annexin V, too. Live cell images taken 4–10 h after polyI:C stimulation clearly demonstrated that GFP-expressing cells selectively underwent cell death (Fig. 5C). The expression of comparable levels of GFP alone did not induce apoptosis in the absence of polyI:C stimulation, excluding the possibility of cellular toxicity by GFP expression (Supplemental Fig. 3D). These results indicate that only the IFN-βhigh-producing cells stimulated with intracellular non–self-RNA can activate the cellular switch to commit to cell death.
Because the knock-down effect of RLR expression affected the generation of IFN-βhigh-producing cells (Fig. 4C), we examined the correlation between RLR expression and apoptosis. For this, RLR expression was reduced by RIG-I and MDA5 siRNA transfection, and the cellular apoptosis induced by polyI:C transfection was measured by PARP1 cleavage (Fig.5D). Intracellular polyI:C induced substantial amount of PARP1 cleavage, which was reduced by knockdown of RIG-I and MDA5, suggesting that RLR plays key roles in the production of IFN-β as well as apoptosis.
To understand the relationship between the biphasic IFN-β production and cellular responses against infection, we measured RLR expression, PARP1 cleavage (apoptosis), and GFP (IFN-β) expression in the sorted GFPhigh and GFPlow populations, simultaneously (Fig. 5E). The result was clear, as there were statistically separable two clusters, C-0 and C-1, present; C-1 was categorized with high GFP and MDA5 expression along with high PARP1 cleavage, while C-0 exhibited low GFP and MDA5 expression as well as low PARP1 cleavage (Fig. 5F). These results indicate that cellular levels of RLR proteins, presumably caused by amplified initial states through the positive feedback of the RLRs–IFN-β system, can be positively correlated with the IFN-α/β production and eventually cell death in the cell populations stimulated with intracellular non-self RNA.
IFNs are key molecules that mediate defense against viral infection. They induce expression of multiple ISGs and function to inhibit viral replication, modulate immune-responses, and control cell survival. To monitor dynamics of IFN-β production in RLRs-mediated signal system, IFN-β–GFP reporter cells were constructed using IFNβ1p-GFP-IFNβ1UTR expression vector. When cells were exposed to cytosolic non-self RNA, two different cell populations, IFN-βhigh (GFPhigh) and IFN-βlow (GFPlow), were appeared despite the equal amount of stimulus. Both in model simulations and cellular experiments of RLR-mediated IFN-β production system, the positive feedback loop via secreted IFN-β and RLR expression was proposed as one of the factors that generate biphasic cellular responses, despite cellular variations and secreted IFN-β dispersion by paracrine effect.
In this paper, we demonstrated that two populations with different IFN-β were generated, and a high IFN-β–expressing population was more sensitive to apoptotic cell death, upon intracellular dsRNA stimulation. High IFN-β–expressing cells therefore act to transmit infection signal to neighboring cells, and subsequently suicide with cytosolic non–self-RNA within. It is interesting to observe not every infected cell produced IFN-β nor undergo apoptosis, because it is well known fact that overproduction or overreaction of type I IFN signals provokes damage to the host (4, 41). Therefore biphasic production of IFN-β might function as an effective immunity against virus infection to protect host cell system, via selective advantages for controlling the amount of IFN-α/β production in virally infected cells, in addition to limiting the number of cells that undergo apoptosis. If every infected cell responds to cytosolic non–self-RNA and expresses IFN-β, then the amount of synthesized IFN-α/β is proportional to the strength of the input stimulus. Therefore, it is very likely that excess amounts of IFN-α/β are produced, leading to potentially massive cellular apoptosis and tissue injury (Fig. 6, case 1). However, if only a subset of cells produce IFN-α/β and undergo apoptosis, whereas others do not produce IFN-α/β and remain viable, then intermediate levels of IFN-α/β are produced, and the number of apoptotic cells can be controlled (Fig. 6, case 2). Secreted IFN-α/β acts on cells that are infected but are not producing IFN-α/β and initiate antiviral immune responses to restrict viral replication.
Apoptosis of IFN-βhigh-expressing cells is adequate to stop excessive production of IFN-β as well as to block viral propagation. In addition, it also serves to propagate the infection signal and control the level of adaptive immune responses. In vivo, apoptotic cells are engulfed and processed by Ag presenting cells (APCs), which then trigger danger signals to activate adaptive immune response in a cell-extrinsic manner (42). Therefore, biphasic IFN-β expression is the optimized system to fine-balance IFN-β–mediated antiviral immune responses and prevention of massive cellular apoptosis and tissue injury.
In this research, we demonstrated heterogeneous IFN-β expression appeared as robust bimodal distribution, and this bimodality was controlled by IFN-β signaling–mediated RLR expression, the key pathway of positive feedback on the system. Toggling between two different stable states is a well-characterized behavior of physiological systems composed of multiple interlinked feedback loops (18, 43). In Xenopus, irreversible oocyte maturation with transient maturating stimuli can be achieved by the formation of positive feedback loop in the cdc2 and p42 MAPK system (12). In the budding yeast phosphate-responsive signal transduction pathway, the toggle switch between high- and low-affinity phosphate transporters is controlled by the interplay between positive and negative feedback loops, which leads to bistability in phosphate usage (13).
Previously, the contribution of initial cell-to-cell variation of RLR expression or stochasticity in the multiple steps of signaling pathway of IFN-β has been suggested as the source of heterogeneous IFN-β expression in the virus-infected cell population (44, 45). Another potential source of bimodality is stochastic chromatin states preceding transcriptional initiation, which could contribute to variability in gene expression. The bimodal expression of IL-4 in Th lymphocytes is controlled by heterogeneous chromatin states that are not affected by blockage of positive feedback within the IL-4 production network (46). For IFNβ gene expression, heterogeneous status of chromatin has been reported in virus-infected HeLa cells (5). In our case, however, the bimodal expression of IFN-β was disrupted when IFN-β secretion was blocked, indicating that cell-to-cell variability in chromatin status of IFNβ1 locus was not sufficient to establish the bimodal expression patterns of IFN-β. We suggest that an intrinsic bistable structure originating from a positive feedback loop between IFN-β and RLR is important for bimodal expression of IFN-β during viral infection.
In summary, our findings suggest that the biphasic production of IFN-β in the RLR system is an intrinsic character of cells upon intracellular non–self-RNA, and RLR-mediated positive feedback could be suggested as the restriction mechanism to control the number of cells undergoing apoptosis in the infected population, which is essential for fine-tuned antiviral response.
This work was supported by the Korea Health Technology Research and Development Project, the Ministry for Health and Welfare, Republic of Korea (Grant A103001), and the National Research Foundation of Korea grant funded by the Korea government (2012028200) awarded to J.-Y.Y.
The online version of this article contains supplemental material.
Abbreviations used in this article:
The authors have no conflicts of interest.