Generation of Diversity in the Innate Immune System: Macrophage Heterogeneity Arises from Gene-Autonomous Transcriptional Probability of Individual Inducible Genes¹

Macrophages respond to activation by microbial products with induction of an enormous diversity of products that are required for innate immunity and priming of the acquired immune response (1). The macrophage activation response is experimentally tractable because relatively pure primary cell populations and macrophage cell lines respond to well-defined agonists such as LPS in vitro (1, 2). Since the pioneering studies of the response of fibroblasts to serum stimulation (3), there has been a very rapid escalation of the use of cDNA microarrays to study patterns of inducible gene expression in mammalian cells. In a recent study, Rosenberger et al. (4) compared the response of a murine macrophage cell line, RAW264, to LPS and live Salmonella using a small commercial nylon array, demonstrating common patterns of induction of many known inflammatory cytokines and chemokines. The apparent coordination of these gene induction phenomena does not imply that all individual cells in a macrophage population respond by inducing the same set of genes. Macrophages are the first line of defense against potential pathogens, so variations in their function at a single cell level could be an important mechanism ensuring that, as a population, they present a complex target to a microorganism. RAW264 macrophage cells have been very widely used in studies of LPS action and signaling, in part because they are one of the few macrophage cell lines that can be readily transfected (5, 6). We and others have found that it is possible to isolate subclones of the RAW264 cell line with quite divergent patterns of LPS-inducible gene expression (7). The same kind of phenotypic heterogeneity is evident in LPS-stimulated primary macrophages where, for example, only a subset of cells express the IL-6, IL-12, and inducible NO synthase (iNOS)³ genes (7, 8, 9). These observations raise the question of whether the very large suite of genes induced by LPS in RAW264 cells, and by extension in primary macrophages, is induced in every cell in the population and if it is not, whether there is a pattern to the variation (a clustering of regulated genes) or each gene behaves autonomously. In fact, macrophage heterogeneity is very well documented. Heterogeneity probably arises as macrophages differentiate from their progenitors, and studies of different mouse strains reveal that the extent and nature of the heterogeneity is strongly influenced by genetic background (10, 11, 12). Most explanations of macrophage heterogeneity are determinist in their underlying rationale, implying that specialized subpopulations differentiate in response to specific stimuli.

In this study, we have used cDNA microarrays to examine the patterns of LPS-inducible genes in subclones of RAW264 cells. The results indicate a remarkable degree of heterogeneity, and also a degree of order, that has profound implications for our understanding of innate immunity and the nature of gene regulation.

RNA was isolated from RAW264 cell clones using the TRIzol reagent kit (Life Technologies, Rockville, MD). Thirty micrograms of total RNA was blotted onto nylon membranes (Millipore, Bedford. MA) using a slot-blot apparatus under vacuum. The RNA was fixed by UV irradiation, and was prehybridized in Denhardt’s with 100 μg/ml of sheared salmon sperm DNA for 1 h at 55°C. Radiolabeled cDNA probes were prepared using PCR-amplified full-length cDNA as template, with the RedyPrime kit (Pharmacia, Peapack, NJ). Hybridization was conducted for 12 h at 55°C in the same buffer. The blots were washed once with 0.1% SDS, 2× SSC, twice with 0.01% SDS, 2× SSC, and then once with 2× SSC. Bound radioactivity was detected using a Bio-Rad Phosphorimager (Bio-Rad, Hercules, CA). Loading was determined by probing with a radiolabeled cDNA directed against the 16S mouse ribosomal RNA subunit. Multiple independent blots were made, and each one was used for up to four separate cDNA hybridizations plus the rRNA control.

We designed a 3700-element mouse microarray enriched for macrophage-expressed genes. Seventeen hundred mouse UniGenes (Research Genetics, Huntsville, AL) were sequence verified, and a panel of known macrophage-expressed and developmentally regulated control genes was added from our own laboratory resources and colleagues within the Institute for Molecular Biosciences (University of Queensland). Additionally, we produced a subtracted library from RAW264 cells that had been stimulated for 4 h with LPS. The Clontech PCR-Select cDNA subtraction kit (Clontech Laboratories, Palo Alto, CA) was used exactly as per the manufacturer’s instructions, and mRNA from unstimulated RAW264 cells was used to produce cDNA for subtraction. Subtracted library clones were plated and picked individually. From an initial set of around 400, 50 were sequenced. We used a pool of multiple hit cDNAs to reprobe the library, and we selected an additional 1300 clones to array. The cDNA inserts were PCR-amplified, column purified, and cDNA arrays were printed onto polylysine-coated glass microscope slides using an Affimetrix 418 gridder at the Queensland Institute for Medical Research. The methods are described elsewhere (13).

Total RNA from RAW264 cells was harvested using Qiagen RNeasy mini-prep columns (Qiagen, Valencia, CA). The integrity of each RNA preparation was checked on a 1.2% agarose/Tris-acetate EDTA gel, and A260/A280 was measured. RNA was concentrated to 2.5 μg/μl in a microcon-30 column (Millipore). Fluorescently labeled cDNA was transcribed from each 50-μg aliquot by incorporating Cy3-dCTP or Cy5-dCTP into a modified superscript reverse transcriptase (Life Technologies) protocol. Hybridization and washing of slides was conducted as described previously (13).

Slides were scanned on an Affimetrix 417 scanner. Images were analyzed using ImaGene 4.1 (Biodiscovery, Los Angeles, CA) and GeneSpring V3.2.11 (Silicon Genetics, Redwood City, CA). Because the large majority of elements on the arrays were induced by LPS, the Cy3 and Cy5 fluorescence distributions were quite distinct, with the former (LPS stimulated) substantially skewed to higher values (not shown).

The human IL-1β promoter clones into the luciferase vector pGL3 was a gift from Dr. Matt Fenton (14). We subcloned the promoter fragment into the corresponding sites of the Promega Renilla luciferase vector. The IL-12 reporter was constructed by us. The murine IL-12 p40 promoter region from −349 to +56 (15) was generated by PCR and was cloned into the pGL2-basic KpnI/XhoI site. The endothelial leukocyte adhesion molecule (ELAM) luciferase reporter is as previously described (16).

To avoid phenotypic drift in cell culture (7), all experiments were conducted using RAW264 cells that had been recently obtained from the American Type Culture Collection (ATCC, Manassas, VA), expanded immediately, and then frozen in aliquots. Cells were maintained in culture for no more than 4–6 wk. Subclones of RAW264 cells were produced by random limiting dilution cloning in 96-well plates. After 3 days of culture, individual wells were inspected to identify those with a single small focus of cells, and such wells were selected and grown for further study.

A total of 5 × 10⁶ RAW264 cells were transfected by electroporation as described (6). All cell culture, including the electroporation procedure, was conducted in RPMI 1640 medium with 10 mM HEPES and 10% FBS. The inclusion of HEPES for transfections is a departure from previous studies (17) and appears to increase survival and transfection efficiency. Ten micrograms of the desired firefly luciferase reporter gene, 10 μg of the IL-1β promoter-Renilla luciferase plasmid, and 2 μg of the selective marker plasmid pNeoTak, which also directs expression of the Tet repressor (18), were added to each transfection. In this system, only the Neo resistance cassette is relevant, but the Tet repressor could be used subsequently for studies using inducible modifier cassettes. After transfection, the cells were placed in a 100-mm square bacteriological petri dish (Sterilin, Teddington, U.K.) with ∼25 ml of medium. Following overnight incubation, most viable cells adhere weakly. The medium was changed, and geneticin (G418, 200 μg/ml) was added. The cells were left in culture with one change of medium to remove dying cells around days 4 and 5. By days 7–10, several hundred individual foci of stably transfected cells were evident in each dish. These were removed by washing of the surface using a syringe with an 18-guage needle (the cells are weakly adherent to bacteriological plastic), expanded, and frozen in aliquots for future experiments. After around 3 wk in culture, the stable transfectants were cloned by limiting dilution. Time course, dose response, and agonist sensitivity of these lines will be described in detail elsewhere (33). For this study, stable transfectant clones were plated at 2 × 10⁵ cells/well and cultured overnight. LPS (100 ng/ml of salmonella R595 LPS from Sigma-Aldrich, St. Louis, MO) was added, and cells were harvested after 8 h for determination of firefly and Renilla luciferase activity using the Promega Dual Luciferase system. The dose of LPS and time were predetermined maxima.

The probes were made and labeled using digoxygenin (DIG) RNA labeling mix according to the manufacturer’s instructions (Boehringer Mannheim, Mannheim, Germany), using T7 and SP6 polymerases from Promega. Mouse 800-bp GLI3 probe in the pGEM vector was linearized with HindIII (anti-sense) and XbaI (sense). Thioglycollate-elicited peritoneal macrophages were plated at 1 × 10⁵ cells/chamber on 16-chamber glass slides (Nuc: Lab Tek) and were cultured overnight in RPMI 1640 medium and 10% FBS. LPS (100 ng/ml) was added where required. After 4 h, cells were fixed in 4% (w/v) paraformaldehyde overnight, and were dehydrated with ethanol/glacial acetic acid (95:5). After rehydration, they were permeabilized with proteinase K (1 μg/ml) for 15 min. Cells were prehybridized for 3 h in Denhardt’s buffer at 44°C and were then hybridized overnight in Denhardt’s buffer containing 10 ng/μl of labeled probe. Posthybridization washes were also performed at 44°C as follows: two times for 15 min with 2×, one time for 15 min with 0.2× SSC, and two times for 15 min with 0.1× SSC. Washed cells were blocked using 10% sheep serum and 2% BSA in 1% blocking reagent (Boehringer Mannheim) for the last 30′ and they were then incubated overnight with anti-DIG F(ab′)₂ conjugated to alkaline phosphatase (Boehringer Mannheim). Alkaline phosphatase activity was localized using 4-nitroblue tetrazolium chloride and 5-bromo-4-vhloro-3-indolyl-phosphate (Boehringer Mannheim).

The murine macrophage cell line RAW264 was originally obtained from an Abelson leukemia virus-induced tumor in mice, and was cloned (RAW264.7) and deposited in the ATCC. We obtained the line from the ATCC, expanded it, and froze aliquots immediately so that the line used herein is as close as possible to the original stock. Limiting dilution subcloning of the original RAW264.7 cells gives sublines with divergent patterns of gene expression in response to LPS. For example, individual subclones do, or do not, express iNOS or plasminogen activator type 2 (PAI-2) in response to LPS (7, 9). To gain a greater insight into the extent of this diversity, we selected a set of 16 known LPS-inducible genes as well as control genes that were expected to be constitutively expressed and conducted a limited array analysis using slot blots. Analysis was performed on 25 subclones that were selected randomly based upon limiting dilution subcloning. The result is shown in Fig. 1. Only IL-6 was induced in all of these clones, demonstrating that they are all LPS-responsive. Even the TNF-α gene, an archetypal LPS responsive gene that is implicated in LPS-induced shock in vivo (2), was induced in only 72% of the clones, whereas most other genes were inducible in significantly fewer. One surprise was that the GAPDH gene, an intended “control,” was induced in the majority of clones. The GAPDH gene was used as a normalization control in the previous study on an RAW264 population (4). Fig. 1 shows the mean and variation of induction in the clones in which the gene was induced, from which it is clear that each gene was either induced greatly or not at all. Pairwise analysis revealed no absolute correlation between the induction of this limited set of genes in individual clones (not shown).

It seemed almost inconceivable that there would be no coregulation of sets of genes given that some must share transcriptional regulatory elements. To analyze the regulatory diversity in greater detail, we decided to use cDNA microarrays. For this purpose, we used a 1.7-K mouse Unigene set. To expand the representation of LPS-inducible genes, we prepared a subtracted, normalized cDNA library from RAW264 cells stimulated for 4 h with LPS (see Materials and Methods). After selecting 400 clones, we sequenced 50. Among these, macrophage-inflammatory protein (MIP)-2, RANTES, virus-like 30S retrotransposon, PGH synthase (cyclooxygenase-2), and glucocorticoid-attenuated response gene-39, all known LPS-inducible genes, were represented more than once. We probed the library to identify and remove from consideration these abundant clones, then selected an additional 1300 cDNAs. Additional to these sets, we included 400 known macrophage-expressed cDNAs and other developmentally regulated genes available in our laboratory. The total set of ∼3700 cDNAs was arrayed on glass slides. We selected the five most divergent RAW264 clones from the original analysis to assay in detail, using the 4-h time point (the same time as used in the library construction).

To normalize the data for relative labeling efficiency with Cy3/Cy5 and between clones, we selected a pool of 30 relatively abundant genes that were expressed consistently relative to each other in the induced and uninduced states over all the clones and that include several commonly used loading controls (e.g. cyclophilin). The Cy3/Cy5 ratio for this set of genes was relatively consistent in different clones regardless of LPS stimulation.

Fig. 2 shows a global comparison of the induction or repression by LPS of the individual elements of the array in the five clonal lines. There are several features that warrant comment in the following paragraphs..

Gene induction and repression was substantial and unequivocal. As observed using nylon arrays (4), many LPS target genes were induced >100-fold. Particular examples are shown in Fig. 3. The heterogeneous gene expression observed using the slot blots in Fig. 1 for genes such as MIP-1β, RANTES, and TNF-α and previously reported for PAI-2 and iNOS was confirmed and validated using the microarrays (Fig. 3).

Not all genes varied between clones. In Fig. 3, the lower four panels show four genes that were not used in normalization; mitochondrial aspartate-amino transferase, ubiquitin conjugating enzyme, elongation factor 1A, and propidium iodide-3-kinase. Each was induced marginally if at all by LPS, but uniformly among the five clones. These controls validate the normalization. By contrast, again validating the data in Fig. 1, β-actin and GAPDH, previously used as normalization controls in uncloned RAW264 cells (4) were clearly inducible in some clones.

The five lines tested can also be clustered hierarchically into three groups (Fig. 2) based upon the number of inducible genes that they share. Table I lists some of the known genes that are coregulated in the three different regulatory “patterns.” This observation suggests that certain sets of genes are, indeed, more likely to be coexpressed. Each clone is still unique and can be distinguished from even the most closely related clone by the lack of coexpression of many individual genes. To assess the probability of coexpression of any two genes, a much larger set of individual subclones would need to be assessed.

There are at least as many genes that were repressed as were induced by LPS. In fact, the impression is that the overall hybridization signal remains similar following LPS addition, suggesting that there may be some kind of balance between induction and repression.

The genes that were induced reproducibly by LPS in some subclones include some very unexpected genes. Among the Unigene set they include the muscle-specific transcription factor MyoD (19), the key regulator of cardiac development Nkx2.5 (20, 21), and the patched hedgehog pathway target gene, Gli-3 (22). We have confirmed that each of these genes is also induced in primary bone marrow-derived macrophages stimulated with LPS (see below). The array results formed the basis for prioritizing additional sequencing of the clones from the subtracted library. Some of the most highly inducible sequenced genes, and the numbers of clones in which they were detected, are shown in Table II. Several are uncharacterized expressed sequence tags.

The data in Fig. 1 and the array data suggest that each inducible gene in macrophages has its own intrinsic probability of being activated in response to LPS challenge in any RAW264 subclone, presumably reflecting a stable difference in chromatin structure at the level of individual alleles. If the probability of being in an inducible state is allele autonomous, we might expect the same stochastic pattern to be manifest with reporter genes. To address this issue in a novel way, we created stably transfected RAW264 lines in which different LPS-responsive promoters driving either Renilla luciferase or firefly luciferase were introduced into the same line. A pool of stable transfectants was produced in which either the ELAM promoter (a commonly used indicator of NF-κB-dependent transcription) or the IL-12 promoter were coupled to firefly luciferase, and the IL-1β promoter was coupled to Renilla luciferase. The IL-12 and IL-1β genes did not cosegregate in apparent clusters in RAW264 clones (not shown). The pools were subcloned by limiting dilution cloning to generate 24 individual lines. Reporter gene activities in the subclones are shown in Table III. The ELAM promoter, which is a relatively short fragment, produced quite divergent basal activity in subclones, ranging over at least 3 orders of magnitude. Despite the baseline variation, activity was induced at least 20-fold in 23 of 24 lines by LPS. By contrast, basal activity of the IL-1β promoter was remarkably consistent between subclones, suggesting that is not subject to effects of position of integration. Despite the consistent baseline, only 8 of 24 clones exhibited induction by LPS. In the combination of the IL-12 and IL-1 promoters, approximately the same frequency of clones showed inducible IL-1β promoter activity. The IL-12 promoter was also expressed quite consistently in the basal state, but like the IL-1β promoter, it was activated by LPS in only a subset of clones (5 of 24). As would be expected if the inducibility of the integrated IL-1 and IL-12 promoters segregates independently and randomly, only two of these clones also exhibited LPS-activated IL-1β promoter activity. These findings show unequivocally that all clones are LPS responsive. In addition, the activation of the transcription factor NF-κB, which is strongly implicated in many LPS actions and underlies induction of the ELAM promoter (16), is not sufficient to ensure induction of all LPS target genes (although it may well contribute to an increased probability). For the purpose of this study, the results reinforce the concept that each gene has an intrinsic probability of being activated in individual subclones of RAW264 cells and demonstrate that the determinant of probability lies within the cis-acting elements of the individual gene promoter.

A systematic validation of the expression of all of the LPS-inducible genes discovered in this study is beyond the scope of the current paper. As noted in the introduction, single cell heterogeneity in primary macrophages has already been demonstrated for the iNOS, PAI-2 IL-6, and IL-12 genes (7, 8, 9). To extend this evidence, and to cross-validate array data from RAW264 cells to primary macrophages, we focused on a newly discovered LPS-inducible gene (see above), the transcription factor, Gli-3 (22). Fig. 4 shows detection of Gli-3 in primary thioglycollate-elicited peritoneal macrophages. In the unstimulated state in adherent macrophages, Gli-3 mRNA was detected at low levels in a subset of cells (around 50%). Upon LPS-stimulation, Gli-3 mRNA was clearly induced. Both the proportion of expressing cells and the level of expression per cell was increased, but even in the LPS-stimulated state, not all cells expressed the gene.

There is a growing recognition that transcription initiation is controlled in a probabilistic manner (23). The promoters of individual genes can exist in either active or inactive chromatin, and they may or may not successfully assemble a transcription initiation complex. The probability of each of these binary decisions is influenced by transcriptional regulatory elements and the transcriptional activators that bind them. When a macrophage cell line proliferates in culture, or macrophages expand from their progenitors in bone marrow, there are mechanisms that ensure that methylation patterns and chromatin structure are reproduced at mitosis. In the case of constitutively expressed genes that are absolutely required for cell function and survival, the mechanisms that ensure appropriate reassembly and nucleosome phasing must be sufficiently reproducible to ensure that gene expression is retained in at least one allele in most or all cells. Conversely, for inducible genes, there must be a mechanism that ensures that despite being in inactive chromatin, the gene is in an inducible state that is susceptible to remodeling. The IL-12 p40 promoter has been studied specifically in macrophages. Activation by LPS was found to cause rapid disruption of a nucleosome overlying the proximal promoter region (24). The fact that this reassembly can be detected in a population of macrophages, but only a subpopulation expresses the gene, implies that it does not determine transcriptional outcome. Rather, it increases the probability that the gene will be expressed. This finding provides an insight into the growing number of examples of monoallelic expression of inducible genes in other hematopoietic cells, for example the IL-2 and 4 genes in T cells (reviewed in Ref. 23). Indeed, in the data shown in Fig. 4, we would suggest that the increased level of expression of Gli-3 mRNA individual macrophages in LPS-stimulated compared with control cells reflects a transition from monoallelic to biallelic expression.

Although the IL-12 and IL-1β promoters, and the cytokine and chemokine genes in Fig. 1, showed no absolute coregulation, the data in Fig. 2 suggest that there are sets of genes that are commonly coexpressed in macrophages. Because of the probabilistic basis of the variation, such order only becomes evident if one examines many different genes or many different cells. There is a clear analogy with studies of the Th1/Th2 dichotomy in stimulated T helper cells where sets of lymphokines tend to be expressed together in a cell population, but at a single cell level, the archetypal Th1 (IFN-γ) and Th2 (IL-4) can clearly be expressed in the same cell (see Ref. 23). Clusters of genes that tend to be induced together can define a population phenotype even if individual cells cannot be classified based upon analysis with any one marker. The phenotypes that emerge among RAW264 cell clones are consistent with the proposed existence of alternative activation phenotypes in macrophages (25, 26, 27) in which the balance between pro-inflammatory and anti-inflammatory secretory products is altered. These proposed phenotypes, like the corresponding Th1 and Th2 T cell phenotypes, were based upon differences in the patterns of inducible cytokines between inbred mouse strains, and indeed the term M-1/M-2 has been proposed (10). In this study, we have show that distinguishable macrophage activation phenotypes can arise on a single mouse genetic background in a cell line.

Coexistence of macrophages with different phenotypes in a single inflammatory site may be an important part of host defense. For example, the data in Table I shows that the natural resistance-associated macrophage protein genes (NRAMP1 and NRAMP2) may tend to be coinduced in the related lines Clone 14 and Clone 30 along with high levels of TNF-α, MIP-1α, MIP-1β, and IL-6. The NRAMP 1 gene is mutated in mouse strains that display wide-ranging sensitivity to intracellular pathogens such as salmonella and mycobacteria (28). The observed induction of nRAMP2 confirms a recent report (29). We may speculate that in wild-type (NRAMP1⁺) mice the nRAMP1/2 genes are coinduced by pathogen in only a subset of macrophages that provide a nonpermissive environment as a consequence. Those macrophages that fail to induce NRAMP1 and NRAMP2 to help resist the pathogen will eventually be killed, but will also provide a reservoir of Ag and much higher amounts of inducible cytokines (30, 31) to permit priming of the acquired immune system. Given that NRAMP1 and NRAMP2 have distinct functions in moving iron and other divalent cations across the endosomal and plasma membranes (32), the intracellular environment may be different again in cells that induce one gene but not the other.

Where it has been examined, primary macrophages display the same heterogeneity in inducible gene expression as RAW264 cells (7, 8, 9). We have extended this evidence to include the newly discovered LPS-inducible transcription factor, Gli-3. Heterogeneous expression of transcription factors like Gli-3 clearly indicates that both trans-acting and cis-acting mechanisms might determine transcription probability. Heterogeneity in primary macrophages was shown most clearly by Witsell and Schook (12) who reported that individual macrophage colonies derived by cultivation of mouse bone marrow cells in CSF-1 are heterogeneous in terms of their ability to express IL-1β and several other inducible cytokines in response to LPS. These authors favored a hierarchical determinist model in which the ability to produce particular cytokines arises as different subsets of macrophages develop and mature. If the underlying mechanism is actually probabilistic, as is demonstrated clearly in this work, the progeny of each committed macrophage progenitor are unique and the innate immune system must present an infinitely complex team of foes to resist an invading pathogen.