Research ArticleImmunology

Single-cell analysis shows that paracrine signaling by first responder cells shapes the interferon-β response to viral infection

See allHide authors and affiliations

Science Signaling  10 Feb 2015:
Vol. 8, Issue 363, pp. ra16
DOI: 10.1126/scisignal.2005728

Taking the time to respond

One of the earliest responses of cells to viral infection is expression of the gene encoding interferon-β (Ifnb1). Within a population of cells exposed to virus, the overall immune response is shaped by the percentage of cells that become infected, the proportion of those infected cells that initially express Ifnb1, and the extent of Ifnb1 expression per cell. Patil et al. used single-cell imaging to quantify the amounts of Ifnb1 and viral mRNAs in infected human dendritic cells over time. Mathematical simulations of viral infection indicated that differences in the timing of Ifnb1 induction in individual cells contributed to the dynamics of the population. Indeed, further experiments showed that the earliest responders released paracrine signals that controlled the timing of the antiviral response of the remaining cells within the population. These results have implications for understanding how individual immune cells coordinate the extent of the immune response to viral infection.

Abstract

Immune responses to viral infection are stochastic processes, which initiate in a limited number of cells that then propagate the response. A key component of the response to viral infection entails the synthesis and secretion of type I interferons (IFNs), including the early induction of the gene encoding IFN-β (Ifnb1). With single-cell analysis and mathematical modeling, we investigated the mechanisms underlying how increases in the amount of Ifnb1 mRNA per cell and in the numbers of cells expressing Ifnb1 calibrate the response to viral infection. We used single-cell, single-molecule assays to quantify the early induction of Ifnb1 expression (the Ifnb1 response) in human monocyte-derived dendritic cells infected with Newcastle disease virus, thus retaining the physiological stoichiometry of transcriptional regulators to both alleles of the Ifnb1 gene. We applied computational methods to extract the stochastic features that underlie the cell-to-cell variations in gene expression over time. Integration of simulations and experiments identified the role of paracrine signaling in increasing the number of cells that express Ifnb1 over time and in calibrating the immune response to viral infection.

INTRODUCTION

A key aspect influencing the emergent immune response to viral infection is that its early events are inherently probabilistic. A small number of host cells are initially infected, and only some of these cells initiate a detectable antiviral immune response (17). The temporal dynamics of the single-cell responses from early times after infection and the underlying mechanisms for the variations among cells are not well understood.

Interferon-β (IFN-β), a cytokine from the type I IFN family, plays a critical role in the defense against viral infection (810). After it is secreted, it binds to a cell surface receptor to activate a gene program that establishes an antiviral state and reduces the replication and spread of virus (817). The timely induction of the gene that encodes IFN-β (Ifnb1) and a fine balance of the extent of its expression are required to orchestrate effective immune responses. Dysregulation in Ifnb1 expression is associated with immune dysfunction. The excessive IFN responses observed in some infections and autoimmune diseases can cause severe toxicity (18, 19). Suppression of IFN production by viral proteins that antagonize the immune response promotes infection (2022).

The tuning of Ifnb1 expression involves a high degree of variability in its expression across individual cells, as demonstrated for viral infections of various cell types (17). Furthermore, the Ifnb1 response to viral infection is detectable in only a fraction of the infected cells. Many studies of single-cell Ifnb1 responses to viral infection have used reporter assays in transfected cells to understand the factors that contribute to variability (1, 3, 6). These reporter systems facilitate the measurement of single-cell changes in gene expression; however, they introduce tens to hundreds of copies of the Ifnb1 gene regulatory regions into each cell, a number that is greatly in excess of the two native promoters, while at the same time, the cellular amounts of transcriptional regulators are not correspondingly altered. Thus, the reporter construct approach may markedly change the concentrations and stoichiometry of transcriptional regulators at Ifnb1 regulatory domains, and the signals observed may not reflect physiological single-cell responses. In addition, reporter constructs integrate the response over many hours, making it difficult to measure the expression dynamics that occur early after exposure to infection, a time period that is crucial for understanding the processes responsible for single-cell variations in gene expression. To further our understanding of this critical immune response to virus infection, it is desirable to determine the extent of Ifnb1 gene expression in its native chromatin environment at the single-cell level while maintaining the physiological stoichiometry of its regulatory components and gene regulatory domains.

Here, we studied the single-cell distributions of Ifnb1 expression in human monocyte-derived dendritic cells (DCs) during the first few hours after infection. To achieve the necessary measurement sensitivity and accuracy, we directly counted Ifnb1 mRNA molecules by single-cell, single-molecule RNA imaging in situ. The sensitivity of this approach avoids the need for the use of reporter constructs and also eliminates amplification steps that may decrease measurement accuracy, especially at low levels of expression (2325). We obtained single-cell Ifnb1 mRNA expression distributions as a function of time from the first 10 hours after infection. We implemented a computational approach (termed stochastic feature screening) to characterize the elements responsible for the single-cell expression distributions observed over time. Our computational studies revealed the single-cell temporal dispersion of Ifnb1 expression as a characteristic property of the response to viral infection. We showed experimentally that paracrine signaling was responsible for this temporal dispersion. Our results identify an important role for paracrine signaling in driving the variability over time of single-cell Ifnb1 responses to viral infection.

RESULTS

Single-molecule imaging of Ifnb1 mRNA and viral HN mRNA in situ

To directly count the number of Ifnb1 transcripts in individual human monocyte-derived DCs infected with virus, we used single-mRNA molecule sensitivity in situ hybridization (2325). Cells were infected with Newcastle disease virus (NDV), a negative-sense, single-stranded RNA avian virus. NDV neither yields a productive infection in monocyte-derived DCs nor encodes antagonists of the human immune response, and therefore is suitable for studying the unimpeded induction of Ifnb1 expression by virus (2628). To identify infected cells, we measured expression of the gene encoding the viral hemagglutinin-neuraminidase (HN) together with that of Ifnb1 (Fig. 1). Cells were hybridized with a set of gene-specific oligonucleotides conjugated to different fluorophores, multiple stacks of images were obtained by epifluorescence microscopy, and image processing algorithms were applied to segment cells and determine exact numbers of mRNA molecules in each cell (see Materials and Methods for details).

Fig. 1 Single-cell measurement of individual mRNA molecules in situ.

Schematic representation of the measurement of single molecules of mRNA in situ. DCs were attached to precoated coverslips before being infected with NDV. The in situ hybridization assay was performed with fluorescently tagged oligonucleotide probes to measure the abundances of Ifnb1 mRNA (probes indicated in red) and HN mRNA (probes indicated in blue) simultaneously in each cell. The fluorescent, spot-like signals developed in the cell for each of the target mRNAs were imaged as z-stacks with an epifluorescence microscope. Each image (z-stack) was processed computationally to determine fluorescent spot counts and the numbers of each target mRNA per cell as described in Materials and Methods.

To evaluate the specificity and sensitivity of this technique in our experimental system, we analyzed cells infected with virus as well as uninfected control (mock) cells. We analyzed images of Ifnb1 and viral HN mRNAs in NDV-infected and uninfected cells after 12 hours (Fig. 2, A and B). The abundances of the mRNAs in a total of 670 cells were quantified in three experiments, and we prepared a scatter plot from a single representative experiment (Fig. 2C). Viral HN mRNA was detected in about 65% of the cells that underwent infection (Fig. 2D), which was to be expected for an infection with virus at a multiplicity of infection (MOI) of 1. The range of HN mRNA abundance in individual positive cells varied from 1 molecule to more than 100 (Fig. 2, C and E). None of the uninfected cells had detectable HN mRNA (Fig. 2, C and D). The absence of any false-positive HN signal in the uninfected cells together with the ability to detect a single molecule of HN mRNA in infected cells suggests that this assay is highly sensitive and specific.

Fig. 2 Measurement of Ifnb1 and HN mRNA amounts at the single-cell level after viral infection.

(A and B) Representative digital interference contrast (DIC) and high-resolution images showing the intracellular distribution of Ifnb1 mRNAs (red spots) and viral HN mRNAs (purple spots) in individual (circled) DCs (A) 12 hours after they were infected with NDV or (B) that were left uninfected (mock). The fluorescent probe images shown are flattened merges of z-stacks of deconvolved fluorescent images. Scale bars, 5 μm. (C) Scatter plot of the distribution of the numbers of Ifnb1 and HN mRNAs per cell measured 12 hours after NDV infection. Note that single-cell, single-molecule measurements yield numbers of mRNA molecules per cell and, thus, discrete values. Circles on the x and y axes represent the overlap of multiple cells, and the numbers in parentheses indicate the number of overlapping cells that are represented by the same symbol. Data are from 220 cells of a representative experiment out of three independent experiments. HN values when Ifnb1 = 0 were binned for clarity of presentation. Inset: The same scatter plot analysis was performed with uninfected control DCs (Mock) in the same experiment. (D) Percentages of DCs expressing the indicated mRNAs when left uninfected (0 hour) or 12 hours after infection with NDV. (E) Average numbers of the indicated mRNAs per cell in DCs that were left uninfected (0 hour) or that were infected for 12 hours with NDV. Data in (D) and (E) are means ± SEM of 670 cells from three independent experiments.

Most of the uninfected cells had undetectable Ifnb1 mRNA; however, fewer than 10% of these cells had detectable Ifnb1 molecules with an overall average background amount of <1 molecule per cell (Fig. 2, C and D). In contrast, at 12 hours after infection, we detected an increase in the percentage of Ifnb1-expressing cells (Fig. 2D), as well as an increase in the average mRNA abundance per cell in the infected samples to ~20 mRNA molecules per cell (Fig. 2E). This increase in the average abundance of Ifnb1 mRNA 12 hours after infection is comparable to previous findings based on real-time reverse transcription polymerase chain reaction (RT-PCR) assays (4). To further evaluate the specificity of our Ifnb1 assay, we determined the ability of unlabeled Ifnb1-specific oligonucleotides to compete for the binding of the fluorescently labeled Ifnb1 and HN probes. The unlabeled probes eliminated any detectable fluorescent signal specifically for Ifnb1 mRNA, whereas the signal for HN mRNA was unchanged (fig. S1), which further supported the specificity of our assay.

Temporal dynamics of the single-cell Ifnb1 response to viral infection

We quantified changes in the amounts of both Ifnb1 and HN mRNAs in individual DCs over time beginning at 2 hours after infection. To control for donor variability, we performed all of the experiments at least three times with cells from different donors. Furthermore, viral infections were also performed in experiments with an epithelial cell line. The percentages of Ifnb1-expressing DCs and the single-cell ranges of Ifnb1 mRNA abundance increased over time, with most of the infected cells expressing detectable Ifnb1 mRNA by the 10-hour time point (Fig. 3A). At early time points after infection, we observed an average of five Ifnb1 mRNA molecules per cell, whereas by 10 hours, there was a three- to fourfold increase in mRNA abundance, consistent with an earlier study based on biochemical assays (4). Our findings showing an increase in Ifnb1 mRNA abundance that was detectable at 2 hours after infection and the relatively high amounts of Ifnb1 mRNA detectable in infected cells differ from those of previous single-cell studies (13), which may reflect differences in experimental systems or assay sensitivity. We evaluated the single-cell correlation between HN expression and Ifnb1 expression (Fig. 3B). Whereas both genes showed a high degree of variation in expression across cells, no statistically significant correlation was observed between the amounts of HN and Ifnb1 mRNAs at any time point after infection (Pearson r2 = 0.00, 0.05, 0.04, and 0.05 for 2, 4, 6, and 10 hours, respectively). At all of the time points analyzed, some cells that expressed HN did not show Ifnb1 expression. At 2 hours after infection, about 35% of the infected (HN-expressing) cells expressed Ifnb1, the number of Ifnb1-expressing cells approximately doubled at later time points (Fig. 3C), and the average amount of Ifnb1 mRNA per cell markedly increased (Fig. 3D). Similar results for single-cell Ifnb1 responses were obtained from experiments in which cells were infected with NDV at a higher MOI. Despite having a greater percentage of cells infected, the percentage of Ifnb1-expressing cells, the average number of Ifnb1 mRNAs per cell, and the temporal dynamics did not substantially change (fig. S2).

Fig. 3 Temporal dynamics of single-cell Ifnb1 responses to viral infection.

(A) Distribution of Ifnb1-expressing DCs at the indicated times after NDV infection. Data are the percentages of DCs that had the indicated numbers of Ifnb1 mRNAs at each time point. Underlined tick labels in the x axis indicate the center of histogram bins with a bin width of 10 mRNA molecules. Inset: A similar analysis of the distribution of HN mRNA–expressing cells over time was performed. For clarity, only results at 10 hours are shown. Data are from 750 cells analyzed in a representative experiment out of three independent experiments. (B) Distribution of the numbers of Ifnb1 and HN mRNAs per cell measured at the indicated times after NDV infection. Time points are represented by the same colors used in (A). The squares of the Pearson correlation coefficients between both mRNAs (r2) at 2, 4, 6, and 10 hours after infection are 0.00, 0.05, 0.04, and 0.05, respectively. Circles on the x and y axes represent overlapping cells, and the number next to each of these circles represents the number of cells per circle. Data are from 750 cells analyzed in the representative experiment used in (A). HN values when Ifnb1 = 0 were binned for clarity of presentation. Inset: A similar distribution analysis was performed with uninfected (mock) control DCs (shown at 12 hours). (C) Percentages of Ifnb1-expressing cells (black), HN-expressing cells (light gray), and cells expressing both Ifnb1 and HN (dark gray) as a function of time after NDV infection (black line). The pink dotted line represents the mean percentage of the total cells that were infected with NDV, which was estimated on the basis of the numbers of cells that had detectable HN mRNA at 10 hours after infection. (D) Average numbers of Ifnb1 mRNAs per cell calculated only for cells expressing Ifnb1 (black line) or for all cells (gray line) at the indicated times after infection. Data in (C) and (D) are means ± SEM of 2500 cells from three independent experiments.

Distinct temporal dynamics of single-cell Ifnb1 responses induced by different stimuli

After infection, a fraction of infected cells exhibited Ifnb1 expression at early time points, whereas more cells showed Ifnb1 expression at later time points. We next compared the single-cell patterns of Ifnb1 expression in response to viral infection to those observed after stimulation of cells through a Toll-like receptor (TLR). The best-characterized inducer of Ifnb1 expression is the TLR4 ligand lipopolysaccharide (LPS), which is a component of the cell wall of Gram-negative bacteria (2931). In LPS-treated human DCs, there was a large cell-to-cell variability in the amounts of Ifnb1 mRNA molecules that we measured (Fig. 4A), as was also seen with cells infected with virus. However, the percentage of Ifnb1-expressing cells (Fig. 4B) and the average amount of Ifnb1 mRNA per cell (Fig. 4C) in response to LPS differed from the patterns observed in infected cells. Both the percentage of cells that expressed Ifnb1 after LPS exposure and the average number of Ifnb1 mRNA molecules in individual cells reached a maximum at 2 hours. The percentage of LPS-treated cells that expressed Ifnb1 was greater than that of infected cells, especially at the early time points. At 2 hours after LPS exposure, about 75% of DCs showed Ifnb1 expression (Fig. 4B), whereas at 2 hours after viral infection, about 35% of infected cells expressed Ifnb1 (Fig. 3C). Exposure to a 10-fold lower concentration of LPS induced single-cell Ifnb1 response patterns similar to those observed at the higher LPS concentration (fig. S3).

Fig. 4 The temporal dynamics of single-cell Ifnb1 responses to LPS differ from those to viral infection.

(A to C) DCs were left untreated or were treated with LPS (100 ng/ml) for the indicated times. The numbers of Ifnb1 mRNAs per cell were then measured by single-molecule imaging of RNA in situ. (A) Distribution of Ifnb1-expressing cells at the indicated times after LPS treatment. Underlined tick labels in the x axis indicate the center of histogram bins with a bin width of 10 mRNA molecules. Data are from 1000 cells analyzed in a representative experiment out of three independent experiments. (B) Percentages of Ifnb1-expressing cells quantified as a function of time after LPS treatment. (C) Average numbers of Ifnb1 mRNA molecules per cell calculated only for cells expressing Ifnb1 (black line) or for all cells (gray line) as a function of time after LPS treatment. Data in (B) and (C) are means ± SEM of 2400 cells analyzed in three independent experiments.

Single-cell Ifnb1 response pattern elicited by viral infection is characterized by temporal dispersion

We hypothesized that the dynamics of single-cell Ifnb1 expression patterns observed in response to viral infection and LPS exposure are a result of some differences in the underlying stochastic features responsible for cell-to-cell variations in gene expression. To test which combinations of stochastic features provided the best explanation for the detailed single-cell time-course distributions that we obtained, we implemented a computational methodology that we refer to as stochastic feature screening. In this approach, we generated a large set of in silico cells each characterized by a model of Ifnb1 expression (see Supplementary Materials for details). In each simulation, we incorporated different combinations of stochastic features that could contribute to the variations observed in single-cell responses. We simulated the single-cell Ifnb1 mRNA distributions over time, and then determined any similarity between the simulation and our experimental results.

We tested four stochastic features (Fig. 5A): (i) whether an individual cell is infected by virus, which is probabilistic, or exposed to LPS, which is deterministic (infectivity, I); (ii) whether an infected or LPS-treated cell mounts an Ifnb1 response, which is probabilistic (response, R); (iii) cell-to-cell variability in the rate of expression of Ifnb1 (strength, S); and (iv) cell-to-cell variability in the time of initiation of Ifnb1 expression after infection or exposure to LPS (time, T). Each combination of stochastic features yields a different pattern over time (propagation mode) of the single-cell Ifnb1 mRNA distributions. For simplicity, we refer to these propagation modes with an acronym indicating the stochastic features included. On the basis of our experimental results, we know that not all cells are infected and that not all infected cells express Ifnb1 in response to the stimulus (Figs. 2 and 3). Thus, both the stochastic features I (infectivity) and R (response) were included in the propagation mode. The stochastic feature combination IR alone resulted in identical amounts of Ifnb1 mRNA in all responding cells (fig. S4), which did not correspond to our experimental results (Fig. 3). We therefore focused on three feature combinations: IRS, IRT, and IRST. Note that including temporal dispersion (T) enforces at least a minimum level of time variation in the initiation of Ifnb1 expression across cells (see Supplementary Materials). These feature combinations yielded very different propagation modes (Fig. 5B), in which the abundance of Ifnb1 mRNA as a function of time for a set of cells is depicted. To identify which stochastic features shaped the Ifnb1 response, we studied whether the IRS, IRT, or IRST propagation modes provided a better correspondence with the experimental results (Fig. 6A). We used a cost function to quantitatively measure the agreement (goodness of it) between the experimental single-cell Ifnb1 mRNA distributions and the propagation mode single-cell distributions over time. The best fit possible (that is, the lowest cost) for each propagation mode was obtained with optimization algorithms. To compare each propagation mode, we simulated IRS, IRT, and IRST 20 times and compared their cost function values.

Fig. 5 Stochastic features and propagation modes studied to fit single-cell Ifnb1 responses.

(A) We explored four stochastic features: (i) whether an individual cell is infected by virus, which is probabilistic, or exposed to a pathogenic component, which is set to 1 for LPS (infectivity, I); (ii) whether an infected or exposed cell exhibits Ifnb1 expression (the “Ifnb1 response”), which is probabilistic (response, R); (iii) cell-to-cell variability in the rate of expression of Ifnb1 (strength, S); and (iv) cell-to-cell variability in the time of initiation of the Ifnb1 response after infection or exposure (time, T). (B) Top: Representative simulations of 250 cells for the three indicated combinations of stochastic features or propagation modes of interest: IRS, IRT, and IRST. (Bottom) Schematic for the propagation modes IRS, IRT, and IRST.

Fig. 6 Viral infection elicits a single-cell Ifnb1 response pattern that is characterized by temporal dispersion.

(A) Flow diagram describing the stochastic feature screening approach applied to analyze experimental single-cell Ifnb1 response data. (B) Characteristic propagation mode of the single-cell Ifnb1 response to viral infection. Cost function representing the goodness of fit of the IRS, IRT, and IRST modes to the single-cell Ifnb1 responses obtained experimentally from DCs infected for 2, 4, 6, and 10 hours with NDV at MOIs of 1 (left) and 4 (right). Similar results were obtained for three independent data sets. Box plots depict the result of 20 simulations for each propagation mode. Error bars represent the maximum and minimum values of all data. (C) Characteristic propagation mode of the single-cell Ifnb1 response to LPS. Cost function representing the goodness of fit of the IRS, IRT, and IRST models to the single-cell Ifnb1 responses obtained experimentally from DCs treated for 1, 2, and 4 hours with LPS at concentrations of 10 ng/ml (left) and 100 ng/ml (right). Similar results were obtained for three data sets. Box plots depict the result of 20 simulations for each propagation mode. Error bars represent the maximum and minimum values of all data. The stochastic feature time (T) in the propagation modes IRT and IRST was strictly enforced. See Supplementary Materials for the range of parameters explored for each stochastic feature. ***P < 0.001 by analysis of variance (ANOVA) on ranks, because cost function values were not normally distributed, followed by Tukey test.

We found differences between the lowest-cost propagation mode for viral infection and LPS exposure. For viral infection, addition of cell-to-cell variability in the time of initiation of Ifnb1 expression (temporal dispersion, T) to the propagation mode resulted in a better representation of the experimental data, as indicated by the reduction in cost function between modes IRS and IRT. However, the combination of the four stochastic features IRST yielded an even greater reduction in cost function, suggesting that the four stochastic features contribute to the Ifnb1 response. Similar results were obtained for infections performed at an MOI of either 1 or 4 (Fig. 6B), which suggests that this result is independent of MOI. In contrast, for single-cell LPS exposure data, the presence of obligatory temporal dispersion in the propagation mode (IRT or IRST) increased the cost function value when compared to the synchronized propagation mode IRS (see Supplementary Materials for the range of parameters for each stochastic feature). The improvement in the goodness of fit of the LPS data sets by the propagation mode IRS was comparable at two different concentrations of LPS (Fig. 6C). Furthermore, applying stochastic feature screening to single-cell measurements obtained from cells treated with LPS at earlier time points (t = 0, 15, 30, 60, 90, and 120 min) also showed an improvement in goodness of fit by the propagation mode IRS (fig. S5). This result suggests that the faster dynamics of the LPS-stimulated IFN response compared to that of viral infection was not responsible for the differences in propagation modes observed.

The stochastic feature screening method does not explicitly represent known causes of expression variation, such as the intrinsic noise that occurs as a result of the small number of molecules involved in activating the two single-gene copies in each cell. We generated a mechanistic model in which intrinsic noise was simulated with Gillespie’s algorithm. Simulations of this model gave similar fits to those obtained with the screening approach (fig. S6), suggesting that the contribution of small copy number to noise is modest (see Supplementary Materials). Overall, viral infection caused a distribution of single-cell Ifnb1 expression that is best explained by a propagation mode that shows temporal dispersion of the responses independently of the extent of infection, whereas LPS caused a distribution of responses that was most consistent with synchronous cell responses, independent of the concentration of LPS. We conclude that a temporally dispersed pattern of response is a characteristic feature of single-cell Ifnb1 responses to viral infection.

Paracrine signaling implicated in temporal dispersion of single-cell Ifnb1 responses to viral infection

Propagation modes, as defined in the stochastic feature screening approach, do not provide direct information on the specific molecular mechanisms underlying a particular response pattern. Temporal dispersion in the initiation of the response might result either from differences in the time at which the infection is initiated in each cell or from changes that occur over time in the capacity of infected cells to mount a response. If the initiation of infection were distributed over time, then shortening the duration of infection would be expected to reduce the number of infected cells. This was tested experimentally by systematically varying the time of virus incubation between 5 and 60 min. Varying the incubation period had no effect on the efficiency of infection (Fig. 7A). This result does not support a role for large variation in the time of initiation of infection as being the source of temporal dispersion in the Ifnb1 response.

Fig. 7 Paracrine signaling drives the temporal dispersion of single-cell Ifnb1 responses to viral infection.

(A) Percentage of HN-expressing DCs as a function of the time of exposure to NDV. Data are means ± SD of duplicate samples from one experiment and are representative of 1200 cells from two independent experiments. (B) Percentage of Ifnb1-expressing cells (left) and the average number of Ifnb1 mRNAs per cell (right) as a function of time after NDV infection in the presence (red lines) or absence (blue lines) of brefeldin A (5 μg/ml), which was added 40 min after exposure of the cells to virus. Data are from two independent experiments. (C) Changes in characteristic propagation mode after treatment with brefeldin A. Cost functions representing the goodness of fit of the IRS, IRT, and IRST modes to the single-cell Ifnb1 responses obtained experimentally from A549 cells infected for 1, 2, or 4 hours with NDV (MOI = 1) in the absence (left) or presence (right) of brefeldin A. Similar results were obtained for two independent data sets. Box plots depict the results of 20 simulations for each propagation mode. Error bars represent the maximum and minimum values of all data. Temporal dispersion (T) in the propagation modes IRT and IRST was strictly enforced. See Supplementary Materials for the range of parameters explored for each stochastic feature. ***P < 0.001, **P < 0.05 as determined by ANOVA on ranks followed by Tukey test. (D) Probability of finding more than one Ifnb1-expressing cell in a single field (142 × 106 μm). Black data points and the solid line indicate values for all experiments. Gray data points and the dotted line indicate the average values expected if Ifnb1-expressing cells were spatially distributed at random. Values were obtained by reshuffling the experimental data at each time point 100,000 times and randomly reassigning cells regardless to different fields as described in Materials and Methods. *P < 0.00001. (E) Percentage of Ifnb1-expressing cells as a function of time after infection with NDV under conditions in which cells were plated at 100 × 103 cells per field (gray) or 800 × 103 cells per field (black). *P < 0.05, as determined by ANOVA on ranks followed by Tukey test.

The lack of any correlation between viral HN expression and Ifnb1 expression suggests that the temporal dispersion most likely results from changes that occur in the probability of cells to respond to the virus after infection has occurred. Human monocyte-derived DCs do not secrete infective virus; however, factors secreted from infected cells, such as IFN, activate a broad gene expression program that alters the response of the cell to viral infection (3234). Therefore, if the temporal dispersion observed after virus infection resulted from cell-to-cell signaling, we hypothesized that interfering with secretion would affect the propagation mode observed.

To test this hypothesis, we studied the effect of inhibiting the process of secretion on the single-cell Ifnb1 patterns elicited by viral infection. To block protein secretion, we used brefeldin A, a drug that inhibits secretion by disrupting the transport of proteins from the endoplasmic reticulum (ER) to the Golgi (35). To avoid the potential problem of differences in the sensitivities of cells from different donors to brefeldin A, we performed these experiments with A549 cells, a cell line derived from human alveolar epithelium (27). We first established that the single-cell Ifnb1 response pattern to NDV infection in A549 cells (fig. S7) retained the same characteristics as the Ifnb1 response pattern observed in DCs (Fig. 3). We then used A549 cells to test the effect of brefeldin A to block signaling by secreted factors on the single-cell Ifnb1 responses to viral infection. Brefeldin A had no effect on infectivity or the synthesis of viral HN mRNA (fig. S8). However, when secretion was inhibited, the percentage of cells showing Ifnb1 expression did not increase after the 6-hour time point, and their response pattern diverged from that of infected control cells (Fig. 7B). The distribution in single-cell responses results from a combination of variations in both extracellular stimuli and intracellular mechanisms. To estimate the contribution of extracellular stimuli to the total cellular variation, we compared the variances in the experiments containing brefeldin A (which eliminates signaling by secreted factors) and those without brefeldin A. This analysis showed that signaling by secreted factors accounted for <10% of the total variation of single-cell responses at 2 hours. The contribution of signaling by secreted factors to total variation increased to 35% by 6 hours, and to more than 90% by 10 hours (Fig. 7B).

We then applied stochastic feature screening to isolate the features associated with the temporal distributions obtained by viral infection in the presence and absence of cell secretion by comparing the cost functions associated with the propagation modes IRS, IRT, and IRST (Fig. 7C). When secretion remained intact, the IRST propagation mode gave a better fit to the Ifnb1 data across several time points (Fig. 7C, left), consistent with our earlier experiments with DCs. However, inhibition of secretion caused the IRS propagation mode to provide the better explanation for the single-cell expression distributions obtained over time (Fig. 7C, right).

The hypothesis that signaling by factors secreted by cells that respond early mediates the recruitment of additional Ifnb1-expressing cells to the response predicts that the distance between Ifnb1-expressing cells should not be randomly distributed. If the cells influence each other, then Ifnb1-expressing cells should be closer to each other than would be expected by chance. We tested this prediction by examining how often multiple Ifnb1-expressing cells were observed in the same microscopy field at each time point. The experimental results were compared to simulations in which Ifnb1-expressing cells were assigned at random to each field. We found that there were statistically significantly more fields containing multiple Ifnb1-expressing cells than would be expected if the distances between these cells were distributed randomly (Fig. 7D). We also tested whether changes in cell density affected the kinetics of the increase in the percentage of Ifnb1-expressing cells. As would be expected if cell-to-cell signaling led to IFN responses in additional cells, we found that a reduction in the plating density from 800 × 103 to 100 × 103 cells per well resulted in a blunting of the increase in Ifnb1-expressing cells observed at later time points (Fig. 7E). These results support the hypothesis that paracrine signaling plays a substantial role in shaping the cell-to-cell variability in the Ifnb1 response to viral infection by introducing temporal dispersion in the responses of infected cells.

DISCUSSION

The homeostasis of complex organisms depends on the specialized and distributed responses and roles of different specific cell types. As researchers quantify individual cell responses, it is becoming increasingly apparent that a wide distribution of responses occurs within individual cell types as well. We quantified the early time course of single-cell Ifnb1 expression induced by viral infection of donor-derived immune cells. The variation in single-cell Ifnb1 expression was high, ranging from cells that expressed no Ifnb1 to cells that generated large amounts of Ifnb1 mRNA, and the amount of Ifnb1 mRNA in infected cells did not correlate with the abundance of viral mRNA. To elucidate the features and mechanisms responsible for the temporal dynamics of the single-cell patterns that we observed, we developed a computational framework to evaluate the sources of random variability. We found that changes in the amount of Ifnb1 mRNA at the single-cell level induced by virus, in contrast to those stimulated by LPS, are best simulated by including temporal dispersion in the initiation of Ifnb1 expression. The results from this model led to experiments that showed that this temporal dispersion depends on paracrine signaling.

Cell-to-cell variation in gene responses, a form of biological noise, is an area of study that originally focused on prokaryotic and unicellular systems, and it is less explored in mammalian systems. It is now well established that cells with identical genomes that are exposed to the same environment do not generate the same responses (36, 37), which is incompatible with deterministic models for cellular responses within a single cell type. Many studies in bacteria and yeast have demonstrated the role of noise in gene expression in determining cellular responses to perturbations (37). In systems engineered by humans, a design objective is typically to decrease variance and increase the deterministic predictability of responses. In biological systems, high instances of variation in cell-to-cell responses are unavoidable, in part as a result of the stochastic properties of reactions that occur at low numbers of molecules, including the fact that only two copies of most genes are present in each eukaryotic cell. Many mechanisms of noise reduction, such as negative feedback loops (38), have been identified in cells. Arguably, however, proper functioning of biological systems depends not only on controlling but also on optimizing the amount of noise. The dependence of single-cell organisms on the appropriate amount of genetic noise for survival is evident. Whereas bacteria, for example, must have enough fidelity in replication to retain function, their survival also depends on an error rate that enables some bacteria to overcome any severe perturbation to homeostasis that is caused by severe stresses, such as heat, changes in osmolarity, or exposure to antibiotic. Several theoretical and experimental studies have shown the adaptive roles of noise in the generation of phenotypic diversity in responses during development (39), by determining cell state in the presence of oxidative stress (40) or by improving survival in a fluctuating environment (41, 42).

The evolution of higher organisms with specialized cell subtypes enabled functions to be distributed among different cells; however, the benefits of response variation within the same cell type for shaping the successful response to perturbations, which was engineered into primordial living systems, have persisted. The immune system must generate an effective and nontoxic response to the myriad infectious challenges. Our analysis of the variation in single-cell Ifnb1 responses provides a window into the complex mechanisms underlying the optimization of cellular noise in the immune system. The induction of Ifnb1 expression must cover a wide dynamic range to accurately tune the response of the immune system to infection. There is a trade-off between providing a wide and highly regulated dynamic range of gene expression and controlling the amount of noise within the system (43, 44). Achieving a wide range of expression makes gene expression more susceptible to fluctuations in upstream factors. The expression of highly regulated genes, such as Ifnb1, is controlled by transcription factors with multiple states and is subject to noise because of the control of chromatin remodeling. Random activation and inactivation of the promoter resulting from changes in chromatin structure are particularly important if amplification is needed at the level of transcription to obtain enhanced expression.

Variability in Ifnb1 expression at the single-cell level has been explored by other groups in different systems. Identified sources of variability fall into two categories: (i) intrinsic noise, which arises from stochastic variation in chemical reactions resulting from the small number of molecules that may be involved within one cell, and (ii) extrinsic noise, which results from baseline variations in the single-cell abundances of key molecular components (37, 45). We previously measured the transcription of the two copies of Ifnb1 in each cell at late times after infection to quantify intrinsic causes of noise (5). Our results, which were corroborated by a subsequent study from other groups (7), suggested that stochasticity in the time of assembly of the transcriptional complex and architectural factors responsible for the induction of Ifnb1 expression contribute to the observed variation in Ifnb1 mRNA abundances at the single-cell level. Factors contributing to extrinsic noise in Ifnb1 expression after viral infection have also been identified. Work by Zhao et al. (1) attributed cell-to-cell variability in the Ifnb1 response to variability in the concentrations of cellular components, such as pathogen recognition receptors, adaptor molecules, and transcription factors, in individual cells. Rand et al. proposed two main causes of variability in the Ifnb1 response: variability in the signaling events downstream of viral detection by the cell that lead to the activation of transcription factors, such as nuclear factor κB (NF-κB) and IFN regulatory factors (IRFs), and stochasticity in the processes of Ifnb1 expression (3). In addition, heterogeneity in the infecting virus population or the role of defective interfering (DI) particles generated during viral replication has also been proposed as mechanisms contributing to single-cell variation in Ifnb1 expression during infection with certain viruses, including parainfluenza virus type 5 (PIV5) and Sendai virus (SeV) (1, 6); however, in one study, DI particles of NDV did not play any substantial role in single-cell variation in Ifnb1 expression in infected cells (3).

Our study of the variation in Ifnb1 expression has the measurement sensitivity and accuracy to examine responses at early time points after infection. Applying single-cell, single-molecule imaging, we found that a relatively small group of cells generated an Ifnb1 response within the first few hours after infection and that additional infected cells were activated through cell-to-cell signaling to express Ifnb1. We found a somewhat higher percentage of Ifnb1-expressing cells at early time points after infection compared to other studies (13), which may be because of differences in the cell types examined or of the assays used. Our results are consistent with the late-time results from previous studies performed in human cells (4, 6, 7).

The terms intrinsic noise and extrinsic noise, which refer to variations in single-cell responses that occur because of the small numbers of reactants or differences in initial cell states, respectively, are usually applied to fluctuations at the steady state. In contrast, we focused on characterizing the dynamics of cell-to-cell variability in Ifnb1 expression over time. Our study shows that cell-to-cell communication changes noise in a cell population dynamically over time during viral infection, presumably by affecting the amounts of pathway-specific components found in the cells, the extrinsic noise together with intrinsic noise. The net result is an additional source of variation after viral infection that is a result of extracellular signaling by secreted factors that causes a temporal variation in response initiation. To study the full innate immune response, we infected DCs with NDV, a virus that lacks effective immune antagonists for human cells. Multiscale agent-based modeling has shown that the single-cell variation in the secretion of IFN is predicted to generate a spatially heterogeneous response of IFN-inducible genes (46). The propagation of noise within one cell type and in complex organs, such as the lung, is an important area for investigation. Many pathogenic viruses in humans express proteins (antagonists), such as the NS1 protein of influenza viruses, which have evolved to suppress the host innate immune response (21, 4749). Further work will need to focus on the effects of viral factors, such as the effect of viral immune antagonists on the pattern of single-cell Ifnb1 expression.

Our study illustrates the importance of characterizing the stochastic nature of the immune response and suggests that the variability of this response at the single-cell level might be controlled differently for different stimuli. These findings reveal a role for paracrine signaling in viral infection as a mechanism to shape cell-to-cell variability in the induction of Ifnb1. Our work suggests that cell-to-cell signaling early in infection may be an important component for optimizing responses to infection while avoiding the systemic toxicity associated with excessive IFN and cytokine responses.

MATERIALS AND METHODS

Differentiation of human monocyte-derived DCs

Monocyte-derived DCs were isolated from the blood of healthy anonymous human donors (New York Blood Center, NY) according to a standard protocol described earlier (32, 48). All experiments were replicated with cells obtained from different donors. Briefly, human peripheral blood mononuclear cells (PBMCs) were isolated from buffy coats by Ficoll density gradient centrifugation (Histopaque, Sigma-Aldrich) at 490g, and CD14+ monocytes were magnetically purified with a MACS CD14 isolation kit (Miltenyi Biotec). Monocytes were then differentiated into naïve DCs by 5- to 6-day incubation at 37°C and 5% CO2 in DC growth medium, which contains RPMI 1640 medium (Invitrogen/Life Technologies) supplemented with 10% fetal calf serum (FCS; HyClone), 2 mM l-glutamine, penicillin (100 U/ml), streptomycin (100 μg/ml; Invitrogen), recombinant human granulocyte-macrophage colony-stimulating factor (GM-CSF; 500 U/ml; PeproTech), and recombinant human interleukin-4 (IL-4; 1000 U/ml; PeproTech). The monocyte-derived DCs were then plated on glass coverslips coated with poly-d-lysine (100 ng/ml; Sigma) and laminin (10 ng/ml; Sigma) and placed in six-well plates at 1 × 106 cells per coverslip. The cells were allowed to adhere on the coverslips overnight at 37°C and 5% CO2 in DC growth medium.

Viral infection and LPS treatment

The Hitchner strain of NDV (rNDV/B1) was generated in the laboratory of P. Palese (Icahn School of Medicine at Mount Sinai, New York, NY). NDV virus was grown in 9-day embryonated chicken eggs as described previously (26, 27). Eighteen hours after infection of Vero cell plates with virus, the viral titer was determined by immunofluorescence microscopy with monoclonal antibodies specific for the NDV-HN protein (Mount Sinai Hybridoma Core Facility) followed by the addition of fluorescein isothiocyanate (FITC)–conjugated anti-mouse immunoglobulin G (IgG). To infect DCs at the appropriate MOI, a cell count was first determined from three individual coverslips containing adhered DCs. Each of the coverslips with cells was subjected to trypsinization, and cells detached from each coverslip were then stained with trypan blue (0.25%). A count of viable cells was obtained for each coverslip with a cell counter (Invitrogen). The average of the three cell counts was used as the representative cell number per coverslip for MOI calculations. NDV stock virus was diluted for the desired MOI in RPMI 1640 and added directly onto DCs grown on the coverslips as described previously (32, 50). After incubation for 40 min at 37°C, infection medium was removed from each coverslip and fresh DC growth medium (without GM-CSF and IL-4) was added to the infected cells for the remainder of the infection. Uninfected control cells (mock) were subjected to the same experimental procedure as the infected DCs, but in the absence of virus. This control ensured that the mechanical manipulations performed on all of the cells were the same and would not contribute to the differences obtained in experimental readouts. In experiments with LPS treatment, LPS isolated from Salmonella enterica serovar Minnesota mutant R595 (InvivoGen) was added to the DCs adhered on the coverslips at a concentration of 100 or 10 ng/ml.

Culture of A549 cells

A549 cells were grown at 37°C in 7% CO2 in tissue culture medium: Dulbecco’s modified Eagle’s medium (DMEM; Invitrogen) supplemented with 10% FCS (HyClone), 1 mM sodium pyruvate (Invitrogen), 2 mM l-glutamine (Invitrogen), and gentamicin (50 μg/ml; Invitrogen). For experiments, A549 cells were seeded at 1 × 106 cells/ml on glass coverslips that were placed in six-well plates. Infection of A549 cells with NDV was performed by diluting the virus stock appropriately in DMEM (without FCS), which was then added to adhered cells at an MOI of 1. After incubation for 40 min at 37°C, the infection medium was removed, and complete DMEM medium was added to the cells for the remainder of the infection period.

Blocking protein secretion

To block the secretion of proteins induced in response to viral infection, A549 cells seeded on glass coverslips were infected with NDV at an MOI of 1 for 40 min at 37°C. The infection medium was then removed, and complete DMEM medium containing brefeldin A (5 μg/ml; BioLegend) was added to the cells for the remainder of the infection.

Single-molecule measurement of mRNA in situ

The amounts of Ifnb1 and HN mRNAs in individual cells were measured by single-molecule fluorescent in situ hybridization (FISH) assay as previously described (2325). Briefly, a set of oligonucleotide probes were designed that were specific for Ifnb1 (34 probes) or HN (48 probes), such that the probes would hybridize either to the entire length or to a portion of their target mRNAs. Each probe was 20 base pairs long and was labeled with a single fluorophore at the 3′ end. The Ifnb1- and HN-specific probes were labeled with Quasar 570 and Quasar 670 fluorophores, respectively, and were obtained from Biosearch Technologies. The sequences of these probe sets are listed in tables S1 and S2. In situ hybridization was performed on the basis of a previously published protocol (2325). Briefly, cells adhered on glass coverslips were washed with phosphate-buffered saline (PBS) and then were fixed in PBS containing 3.7% (v/v) paraformaldehyde for 10 min at room temperature. After fixation, cell were washed with PBS and permeabilized with 70% (v/v) alcohol for 20 min at 4°C. The cells were then washed with 10% formamide dissolved in SSC buffer (2×; Ambion), and hybridized with the probe sets for the desired target. The hybridization was performed such that the surface of the coverslip that was coated with cells was placed face down over the hybridization solution, which was placed on a sheet of parafilm. The hybridization reaction for each coverslip was performed in 50 μl of hybridization buffer that contained 10% (w/v) dextran sulfate (Sigma), Escherichia coli tRNA (transfer RNA) (1 μg/μl; Sigma), 2 mM ribonucleoside-vanadyl complex (New England Biolabs), 0.02% (w/v) ribonuclease-free bovine serum albumin (BSA; Ambion), 10% (v/v) formamide (Ambion), and 1 ng/μl of each probe set. The hybridizations were performed overnight in a water bath maintained at 37°C. After hybridization, the coverslips were washed four times with a wash buffer containing 10% formamide and 2× SSC (20 min per wash) on a shaker at 300 rpm. The coverslips were then mounted in an oxygen-depleted mounting medium consisting of a 0.4% glucose solution containing glucose oxidase and catalase.

Imaging

Cells were imaged with an upright epifluorescence microscope, Axioplan 2 (Zeiss), with a 100× oil immersion objective and a numerical aperture of 1.3. The microscope was controlled by AxioVision software and was equipped with a Zeiss AxioCam MRm camera. All of the images were acquired as z-stacks with a total stack height of 3.2 μm (17 z-sections) and with a distance of 0.2 μm between each z-section. The laser exposure time in each of the fluorescence channels ranged between 2 and 3 s. Raw images were deconvolved with AutoQuant X2 AutoDeblur software. This process reduced the out-of-focus light and substantially enhanced the quality of the signal. For density experiments, the same microscopy setup was used at a ×40 magnification. To identify Ifnb1 mRNA–containing cells, fluorescence intensity was integrated over each cell and compared to that of uninfected cells.

Image analysis to count the numbers of target mRNA molecules per cell

To identify fluorescent spots and obtain the number of the target mRNAs in each of the acquired z-stacks, a computational method was used as described previously (2325). Briefly, each z-section image was first segmented with an edge detection algorithm to create a mask for each cell in the image z-stack. The fluorescent spots associated with each cell in a z-stack were counted by normalizing the three-dimensional image for intensity, filtering for size, and finding a plateau in the threshold profile that gave us the total number of mRNAs in a given volume.

Stochastic feature screening

Every combination of stochastic features applied to the basic model was simulated in as many cells as were measured in each particular experimental data set. Parameters were optimized with multiobjective evolutionary algorithms, and a cost function based on Kolmogorov-Smirnov tests for the goodness of fit was obtained from comparing simulated distributions and experimental data across multiple time points. The summary model, which incorporates the different stochastic features considered, was first optimized and then simulated with Gillespie’s algorithm (50) to check for the effect of small copy number noise. Details of the mathematical characterization of the stochastic features, model definitions, and optimization algorithms are given in the Supplementary Materials. All modules were programmed and simulated in MATLAB (http://mathworks.com).

Statistical analysis of imaging data

The probability of finding multiple Ifnb1 mRNA–containing cells in the same microscopy field was estimated as the number of fields with >1 Ifnb1 mRNA–containing cell divided by the total number of fields. Fields with ≤2 cells were discarded. The values were normalized by dividing the number of Ifnb1 mRNA–containing cells by the total number of cells per field and multiplying by the average number of cells per field in each data set. Data sets for each time point included three to five donors. To estimate the probabilities expected if there were no cell-to-cell distance effect, for each data set, cells were reassigned randomly in fields and probabilities calculated for 100,000 simulations. For any random assignment, fields with ≤2 total cells were filtered out, the average number of cells for each field was computed, and the number of Ifnb1 mRNA–containing cells was normalized to the total number of cells per field.

SUPPLEMENTARY MATERIALS

www.sciencesignaling.org/cgi/content/full/8/363/ra16/DC1

Computational methods

Fig. S1. Validation of the single-molecule imaging of mRNA in situ.

Fig. S2.Temporal dynamics of single-cell Ifnb1 mRNA responses to viral infection at an MOI of 4.

Fig. S3. The percentages of Ifnb1-expressing cells are independent of the concentration of LPS.

Fig. S4. Temporal dispersion is a characteristic stochastic feature of the Ifnb1 response after viral infection.

Fig. S5. The faster dynamics of Ifnb1 expression in response to LPS does not change its characteristic propagation mode.

Fig. S6. Comparison between experimentally measured and simulated average numbers of Ifnb1 mRNAs per cell and the percentages of responding cells.

Fig. S7. Temporal dynamics of single-cell Ifnb1 responses in response to viral infection of A549 cells.

Fig. S8. Brefeldin A has no effect on the infectivity of A549 cells.

Table. S1. Sequences of oligonucleotide FISH probes specific for viral HN mRNA and tagged with Quasar 670.

Table. S2. Sequences of oligonucleotide FISH probes specific for Ifnb1 mRNA and tagged with Quasar 570.

Table. S3. Summary model parameter values.

REFERENCES AND NOTES

Acknowledgments: We thank C. Jayaprakash for very helpful discussions and J. Wetmur for his comments. We thank the Microscopy Shared Resource Facility, Icahn School of Medicine at Mount Sinai. Funding: This work was supported by National Institute of Allergy and Infectious Diseases contracts HHSN272201000054C and AI106036. Author contributions: S.P., F.H., and S.C.S. conceived the study; S.P. performed the experiments; M.B. and S.T. collaborated on the FISH assay; M.F. and F.H. performed the modeling; S.P., M.F., Y.G., and F.H. analyzed the data; and S.P., M.F., F.H., and S.C.S. wrote the paper, whereas the other authors reviewed and revised the paper. Competing interests: The authors declare that they have no competing interests.
View Abstract

Navigate This Article