Single-Cytokine-Producing CD4 Memory Cells Predominate in Type 1 and Type 2 Immunity¹

CD4 memory cells are thought to occur in at least two distinct subpopulations that can play opposing roles in infectious and autoimmune diseases by producing different cytokines (1, 2). Considerable progress has been made in understanding how naive T cells differentiate into memory cells that express either IFN-γ or IL-4 (reviewed in Refs. 3 and 4). Whether other cytokines are coexpressed by these memory cells when they reencounter Ag in vivo and exactly which ones they are is not presently known. Without knowing which cytokines are produced in a linked manner in individual memory cells and which cytokines are expressed independently, no conclusions can be drawn about how many distinct types of memory T cells exist beyond the IL-4/IFN-γ dichotomy; thus, it remains unclear how precise and versatile these memory cells are in implementing the individual effector functions induced by the individual cytokines. Because of the technical limitations that had made it intractable to resolve this question by direct measurements, several indirect approaches have been used, producing conflicting results.

The first studies performed with long-term T cell clones suggested that memory T cells express two sets of cytokines in a mutually exclusive fashion; Th1 cells produce IL-2 and IFN-γ, among other cytokines, while Th2 cells secrete IL-4 and IL-5 (1). Based on these data, the mainstream model emerged, according to which naive T cells (which do not produce cytokines or produce only IL-2) first differentiate into Th0 cells that coexpress type 1 and type 2 cytokines (5, 6, 7, 8) and then further differentiate into the polarized Th1 or Th2 cells upon ongoing Ag stimulation. Subsequent studies of short-term clones showed considerable heterogeneity in cytokine profiles (9, 10), which suggested that T cells might not occur in distinct Th1 or Th2 subsets and that individual T cells can coexpress type 1 and type 2 cytokines in various combinations and ratios, a view that gave rise to the stochastic model of cytokine gene regulation (11). Studies performed later at the single-cell level using intracytoplasmic cytokine staining and dual-label cytokine hybridization also showed a high degree of heterogeneity in cytokine coexpression in cloned T cells (12, 13, 14, 15). While T cell clones have the advantage of providing defined cell populations, the extent to which they are representative of memory T cells in vivo is not known. During tissue culture, T cells are continuously driven to cycle, they undergo changes of chromatin structure, and their DNA (including their cytokine genes) becomes demethylated (16, 17, 18). After 16–35 cell divisions, T cells undergo replicative senescence in vitro, and cells that survive in culture invariably reveal severe and multiple chromosomal abnormalities (reviewed in Ref. 19), all of which can affect their cytokine gene regulation. Therefore, it remains unclear whether T cells generated under the conditions of an immune response in vivo or during chronic T cell-mediated immune pathology have cytokine expression patterns like those of in vitro-expanded cells.

A different approach to the study of cytokine coexpression in memory cells relies on TCR-transgenic models. TCR-transgenic mice themselves show little immune competence (20); therefore, the priming and subsequent differentiation of the transgenic T cells has been primarily modeled in tissue culture. Unexpectedly, during the first 7 days in culture, the TCR-transgenic cells were found to express IL-2, IL-4, IL-5, and IFN-γ in an almost completely dissociated fashion, with each T cell expressing only one of these cytokine mRNAs. However, after further propagation in vitro, this phenotype was lost, and the transgenic cells started to coexpress these cytokine genes in apparently random combinations (21, 22). When in vitro-propagated TCR-transgenic cells were studied by intracytoplasmic cytokine staining, various coexpression patterns were seen including a high degree of “Th0-like” IFN-γ/IL-4 and IL-2/IL-4 coexpression (13, 23, 24, 25). Does, therefore, the expression of only one cytokine per cell characterize the initial phase of the T cell response in vivo, while cytokine coexpression subsequently prevails, and, if so, do the memory cells coexpress type 1 and type 2 cytokines in a mutually exclusive fashion, or do they express them in random combinations?

Studies of freshly isolated, nontransgenic T cells have been primarily confined to polyclonal mitogen stimulation. Depending on the cell populations tested and activation/culturing conditions chosen, cytokine expression and coexpression patterns of various sorts were seen, which fit, respectively, the classic Th1/Th2 paradigm (26, 27, 28), the stochastic model (29, 30, 31, 32, 33), or a pattern of dissociated cytokine-expression (34, 35, 36). The reasons for these conflicting results might lie in the activation of T cells with different histories of Ag encounter and, perhaps more importantly, in the nonphysiological nature of the T cell-activating signal generated by the cross-linking of TCR by mitogens or by Abs (37, 38). During physiologic, MHC-restricted recognition of Ag, the TCR functions as a gauge for the strength of the signal (39), frequently interacting with only a few MHC-peptide complexes on the surface of the APC (40, 41); the number of TCRs engaged and the kinetics of the engagement translate into different intracellular signaling patterns (39). The strength of this signal has been shown to affect cytokine coexpression (14), and different signal strengths can induce different functions in memory T cells (42). Therefore, it remains unclear whether the cytokine expression patterns seen in mitogen-stimulated/signal-enhanced T cells will be the same as those after physiologic recognition of Ag, and how the signal strength affects the coexpression of cytokines in T cells.

An understanding of cytokine gene regulation might help to predict the coexpression of cytokines in individual T cells. The data in this field as well are still controversial. Depending on the model and the stimulation/culture conditions used, evidence supporting both stochastic gene regulation (25, 43) and precisely controlled cytokine gene activation including allelic exclusion (44) has been obtained. Although the bulk of emerging evidence seems to favor regulated expression of individual cytokine genes, predictions about the coexpression of specific type 1 and type 2 cytokines in individual T cells in vivo cannot presently be made.

To address this controversy around the cytokine signature of normal, in vivo-differentiated memory cells, we pursued the measurement of cytokine coexpression in freshly isolated CD4 cells that had been physiologically activated by the nominal Ag. Building on previous efforts (45), we have developed two-color cytokine enzyme-linked immunospot (ELISPOT)³ assays in conjunction with computer-assisted image analysis to this end. After validating this approach, we measured the coexpression of key type 1 and type 2 cytokines early and late in the course of the immune response, in polarized response types, and in chronic autoimmune stimulation. We found single-cytokine-expressing CD4 memory cells to predominate under all these conditions.

BALB/cByJ, BALB/cByJ Smn-Prkdc^scid/J, C57.BL/6J, and SJL/J mice were purchased from The Jackson Laboratory (Bar Harbor, ME) and bred at Case Western Reserve University under specific pathogen-free conditions. Female mice were injected at 6–10 wk of age. OVA was purchased from Sigma (St. Louis, MO). OVA peptide 323–339 was purchased from Princeton Biomolecules (Columbus, OH). IFA was purchased from Life Technologies (Grand Island, NY), and CFA was prepared by mixing inactivated Mycobacterium tuberculosis H37RA (Difco Laboratories, Detroit, MI) at 1 mg/ml into IFA. Ags or peptides in PBS were mixed 1:1 with adjuvant, and the specified Ag dose was injected once in 100 μl, s.c. or i.p., as specified. For the Leishmania major model, infected BALB/c mice were provided by Dr. F. Heinzel (Case Western Reserve University). The care of mice was in accordance with institutional guidelines. TCR-transgenic DO11.10 mice that are OVA_323–339 specific were obtained from Dr. M. K. Jenkins (University of Minnesota).

Intracytoplasmic staining was performed as described (15). Dual-staining for IL-2:IFN-γ and IL-4:IL-5 was achieved by combining JES6-5H4-PE/XMG1.2-FITC and TRFK5-FITC/11B1-biotin with streptavidin-PE, respectively (PharMingen, San Diego, CA). The isotype-matched control mAbs were obtained from Becton Dickinson (San Jose, CA). The samples were analyzed on a FACScan flow cytometer (Becton Dickinson). Th1 clone SH-10 (46) and the Th2 clone M33.25.6 (47) were provided by Dr. P. S. Heeger (Cleveland VA Medical Center, Cleveland, OH).

Plates (ImmunoSpot M200; Cellular Technology Limited, Cleveland, OH) were coated overnight at 4°C with the cytokine-specific capture Abs specified below. The plates were then blocked with 1% BSA in PBS for 1 h at room temperature and washed four times with PBS. Subsequently, irradiated lymph node (LN) APC from naive, syngeneic mice were added (1 × 10⁶ or 5 × 10⁵/well, as specified). Cloned T cells (in serial dilution, with the numbers specified in Fig. 1) or freshly isolated CD4 cells (obtained in >97% purity after separation on Mouse CD4 Subset Columns; R&D Systems, Minneapolis, MN) were plated in serial dilution in two to four replicate wells with or without the nominal Ag or control Ags. When used, single-cell suspensions from spinal cords were prepared according to the same procedure used for LN and spleen (48). We used serum-free HL-1 medium (BioWhittaker, Walkersville, MD) supplemented with 1 mM l-glutamine. After 24–48 h of cell culture in the incubator at 37°C, the cells were removed by washing three times with PBS and four times with PBS containing 0.05% Tween (PBST), and the two detection Abs were added simultaneously and incubated at 4°C overnight. The plates were washed three times with PBST. For the biotinylated detection mAbs, the streptavidin-alkaline phosphatase conjugate (Dako, Carpenteria, CA) was added (at 1:2000 dilution), incubated for 2 h at room temperature, and removed by washing twice with PBST and twice with PBS. The nitroblue tetrazolium (NBT)/5-bromo-4-chloro-3-indolyl phosphate (BCIP) substrate (Kirkegaard & Perry Laboratories, Gaithersburg, MD) was added first, then, after washing twice with PBS, the 3-amino-9-ethylcarbazole (AEC) substrate (Pierce, Rockford, IL) was added and left for 15–30 min for NBT/BCIP and 20–40 min for AEC. The following coating mAbs were used for IL-2, IL-3, IL-4, IL-5, and IFN-γ: JES6-1A12 (5 μg/ml), MP2-8F8 (5 μg/ml), BVD4-1D11 (2 μg/ml), TRFK5 (5 μg/ml), and R46A2 (2.5 μg/ml). The combinations of detection Abs for the IL-2:IFN-γ, IL-3:IFN-γ, IL-4:IL-5, IFN-γ:IL-5, IL-4:IFN-γ, and IL-2:IL-5 assays were: JES6-5H4-biotin:XMG1.2-HRP, MP2-43D11-biotin:XMG1.2-HRP, BVD4-24G2-biotin:TRFK4-HRP, XMG1.2-biotin:TRFK4-HRP, BVD4-24G2-biotin:XMG1.2-HRP, and JES6-5H4-biotin:TRFK4-HRP (all Abs were from PharMingen). HRP labeling of Abs was performed according to the standard method. Unlabeled TRFK4 in combination with HRP-labeled mouse anti-rat IgG2a mAb (1:300 dilution; Zymed, San Francisco, CA) was also used for IFN-γ:IL-5 and IL-4:IL-5 assays. The detection Ab concentrations were as follows: JES6-5H4-biotin (2 μg/ml), MP2-43D11-biotin (2 μg/ml), BVD4-24G2-biotin (2.5 μg/ml), TRFK4-HRP (2 μg/ml), and either XMG1.2-HRP (2 μg/ml) or XMG1.2-biotin (2 μg/ml).

We used an ImmunoSpot Image Analyzer (Cellular Technology Limited) specifically designed for two-color ELISPOT analysis. Digitized images were analyzed for the presence of areas in which color density exceeds background by a factor set on the basis of the comparison of control (containing T cells and APC without Ag) and experimental wells (containing Ag, exemplified in Fig. 2, A vs B–F). After separating spots that touch or partially overlap, additional criteria of spot size and circularity are applied to gate out speckles and noise caused by spontaneous substrate precipitation, nonspecific Ab binding. Objects that do not meet these criteria are ignored and areas that meet them are recognized as spots, counted, and highlighted. Additionally, spot-size histograms were generated reflecting the distribution of cells according to the cytokine output per cell (an example is provided in Fig. 4). Two-color ELISPOT image analysis follows the same principles except that the image analyzer detects red, blue, and double-colored spots separately by using three different threshold settings as specified in Fig. 2. Each color threshold is set in RGB mode and consists of three numbers reflecting the threshold in red, blue, and green channels. The red and blue thresholds are set by using spots from single-color assays.

The first set of experiments was done to validate the two-color cytokine ELISPOT approach for measuring cytokine coexpressed by individual T cells that occur in the low-frequency range (1:1,000–1:1,000,000), wherein CD4 memory cells usually occur and which is below the detection limit of FACS analysis. The sensitivity of two-color ELISPOT assays was tested by subjecting T cell clones to intracytoplasmic cytokine staining and ELISPOT analysis in parallel. Intracytoplasmic IFN-γ/IL-2 staining of clone SH10 showed that 48% of the cells expressed only IFN-γ and 2% only IL-2, with 23% coexpressing these cytokines (Fig. 1,A). By not producing either IL-4 or IL-5 (data not shown), SH10 qualifies as a Th1 clone. When the same cells were tested in parallel by two-color IFN-γ/IL-2 ELISPOT assays in serial dilutions keeping 1 × 10⁶ APC per well, similar results were obtained: an average of 53% of the T cells was found to produce only IFN-γ, 4% to secrete only IL-2, and 19% to secrete both cytokines (Fig. 1,B; the image analysis of such two-color spots is shown in Fig. 2). The plot of the number of cloned cells plated per well against the number of single-positive and double-positive spots was linear and passed through the origin. Therefore, every cytokine-producing cell was directly visualized, even in the presence of 1 × 10⁶ bystander cells. This close correlation between intracytoplasmic staining and ELISPOT assays was also seen in similar experiments performed on the Th2 clone M33 (Fig. 1, C and D): IL-4 and IL-5 were coexpressed by ∼50% of the cloned cells. The numbers of single-positive and double-positive cells linearly decrease with the number of CD4 cells plated, consistent with the assay having single-cell resolution. In addition, these data show that, as far as the detection of the coexpression of IFN-γ/IL-2 and of IL-4/IL-5 in single T cells is concerned, the sensitivities of intracytoplasmic staining/FACS analysis and two-color ELISPOT analysis are comparable. The color resolution of two-color ELISPOT assays based on AEC (the red substrate for HRP) and NBT/BCIP (the blue substrate for AP) is at least as good as the color resolution of FITC and PE staining in two-color FACS analysis. Last, the use of two-color cytokine ELISPOT analysis confirmed that IFN-γ and IL-2, and IL-4 and IL-5, are coexpressed in long-term cultured T cells, an observation that led to the postulate that such Th1/Th2 cells would also occur in vivo.

When we studied the size distribution of Ag-induced IFN-γ spots generated by freshly isolated CD4 cells from the BALB/c mice primed with OVA_323–339 and those produced by cloned T cells, we found them to be comparable: both types of cells produced a wide spectrum of spots of various sizes with distributions close to Gaussian and very similar to those obtained by intracytoplasmic stainings (Fig. 3, A vs B). Although the size of the spots depends on several parameters including the kinetic of cytokine secretion, it is proportional to the net cytokine output per cell over the culture period (49). Therefore, such differences in spot sizes (amount of cytokine produced) within clones and freshly isolated cells reflects cell biology. (Similar results were obtained when comparing IL-2, IL-4, and IL-5 produced by clones and freshly isolated CD4 cells, data not shown.) Because cloned cells and freshly isolated T cells produced comparable amount of cytokines, single-cell sensitivity also seems to apply for freshly isolated memory T cells, which occurred at frequencies too low for intracytoplasmic staining to measure, thereby impeding direct comparison. The single-cell resolution of these measurements was further supported by the linear decrease in spot numbers seen when CD4 LN cells primed to produce either IFN-γ or IL-5 were serially diluted with unprimed spleen cells or were mixed in different ratios (Fig. 4). These data also showed that the cognate cytokine produced by type 1 or type 2 polarized memory T cells neither induced bystander cytokine production in the APC population nor inhibited IFN-γ/IL-5 produced by the memory cells.

We chose to characterize more closely the immune response induced in BALB/c mice with a well-defined Ag, OVA, and its immunodominant peptide OVA_323–339. First, we immunized mice with the maximally immunogenic dose of OVA_323–339 peptide, 100 μg/mouse, in CFA, s.c. and tested the peptide-induced recall response of CD4 cells purified from the LN and spleens of these mice at various time points. Representative data are shown in Fig. 5 and are summarized in Table I. A classic type 1 (IFN-γ⁺, IL-2⁺, IL-3⁺, IL-4⁻, and IL-5⁻) recall response was seen. Image analysis of data obtained in three independent experiments with 4–24 mice per experiment in which cells from each mouse were tested in triplicate wells and in serial dilutions (analyzing more than 600,000 peptide-induced cytokine spots) showed that 95 ± 4% of the spots in the IFN-γ:IL-2 assay were either red or blue (single- positive). Only about 5% of the spots appeared in various shades of purple indicating that only a minor fraction of Ag-specific T cells produced both cytokines simultaneously or switched cytokine production during the assays’ 24–48 h duration. This frequency range of double-positive spots was also seen when the primed T cells were tested in serial dilutions (data not shown). We performed single-color IFN-γ and IL-2 assays in parallel to verify that we had detected all double-cytokine-producing T cells. The frequencies of cells producing IFN-γ and IL-2 in the single-color assays closely matched the sum of the frequencies of the single producers and double producers of IFN-γ/IL-2 detected in the two-color assay (within 5% error). These additional single-color data prove that the two-color analysis did not miss double-expressing cells. While we cannot exclude the possibility that some cytokine is being coexpressed below the detection limit of ELISPOT analysis, we can safely conclude that the secretion of the one cytokine detected was highly polarized.

Neither unimmunized nor control-immunized mice produced IL-2, IL-3, IL-4, IL-5, or IFN-γ when challenged with OVA or OVA peptide. Although it has been suggested that naive cells can produce IL-2 (Refs. 50 and 51 ; a notion that is not unanimously agreed upon, see Ref. 21), their frequency must have been <1/1,000,000 the detection limit of the IL-2 ELISPOT assay as performed. Only after immunization were cells producing IL-2, IL-3, or IFN-γ detectable, suggesting that these cells have undergone Ag-driven clonal expansion and differentiation. Moreover, we found that these cytokine-producing T cells resided in the L-selectin⁻, CD4 fraction (48), which corresponds to a memory phenotype. Therefore, the cytokine-producing cells that we detected after immunization appear to be memory cells that acquired their cytokine phenotype during an Ag-driven immune response in vivo.

It has been suggested that the Ag dose (the density of MHC:nominal peptide complexes on the APC defining the extent of TCR ligation and, hence, the signal strength) affects both the postthymic differentiation of naive T cells along the type 1/type 2 pathway and the pattern of cytokine expressed by differentiated memory cells (42, 52, 53). To determine whether this also applies to adjuvant-driven CD4 cells differentiation in vivo, we immunized BALB/c mice with doses of OVA_323–339 peptide ranging from 0.01 to 100 μg/mouse, in CFA, and performed recall assays on the memory T cells induced, titrating the peptide dose from 0.01 to 40 μM. The 0.01-μg immunization dose did not induce a detectable cytokine recall response; it became detectable at 0.1 μg injected per mouse and reached the plateau at 10–100 μg/mouse (Fig. 6 A). The overall cytokine signature of these recall responses was unaffected by the immunization dose: induction of IL-2, IL-3, and IFN-γ was seen, with only marginal IL-4 and no IL-5 being produced. No IL-5 and only marginal IL-4 production was seen over the full range of peptide concentrations tested at recall (data not shown).

The dose-response characteristics for the activation of cells producing IL-2, IL-3, and IFN-γ were similar (Fig. 6, B–D). The concentration of the OVA peptide at which 50% of the maximally inducible cells become activated was 50 ± 15 nM for all three cytokines; this defines the functional avidity of these T cells for OVA_323–339 (48). Whereas the frequency of cytokine-producing cells increased with the peptide concentration, the fraction of double-positive cells stayed largely constant for given immunization doses varying between 1 and 8%. These data, shown for the day 21 time point in Fig. 6, are representative for all the time points tested (days 4, 10, 21, 42, and 91, data not shown). Therefore, the dissociated production of IL-2, IL-3, and IFN-γ was neither affected by the signal strength that induced the differentiation of the naive T cells nor by the signal strength that induced cytokine production in the memory cell. Similar results were obtained when purified CD4 cells were tested on different types of APC layers, irradiated or unirradiated spleen cells of BALB/c^SCID mice, or spleen, LN, or thymic cells of naive BALB/c mice. Therefore, the presence or absence of B cells and the variation of other class II-positive cell types in these organs had no significant effect on the size and frequency of IFN-γ, IL-2, and IL-3 spots detected (data not shown).

It has been postulated that CD4 cells first differentiate into memory cells that coexpress type 1 and type 2 cytokines (Th0-type cells) in the course of the immune response and that only with chronic stimulation would these cells further differentiate into type 1 or type 2 cytokine-expressing memory cells. According to this model, the coexpression of type 1 and/or type 2 cytokines in individual T cells should be observed in vivo early in the course of response, whereas the cytokine expression pattern of the memory cells might become polarized over the course of the immune response. To address this possibility, we also studied OVA-peptide-induced CD4 memory cells soon after immunization. By the earliest time that we could detect a specific cytokine recall response, on day 4 after immunization, IL-2, IL-3, and IFN-γ single-positive cells were detected and none producing IL-4 or IL-5 were found. Therefore, at the population level, the cytokine response was type 1- polarized early on, and, at the level of individual memory cells, it was already mediated by single-cytokine-producing cells. Memory cells seem to assume this phenotype rapidly without first going through a prolonged state in which they coexpress cytokines.

Transient segregation of cytokine mRNA expression has been reported for in vitro-primed D011.10 TCR-transgenic cells when using dual-label, in situ hybridization (21, 22). Two-color cytokine ELISPOT measurements yielded comparable results when the same D011.10 cells were tested after the second cycle of in vitro stimulation (Table I). We also observed that the frequency of double-cytokine-producing cells increased with further restimulations in vitro: after 42 days of cell culture, about 16% of the cytokine-producing, transgenic cells became IL-2⁺/IFN-γ⁺ double-positive, starting to approximate the phenotype of T cell clones (Fig. 1). Moreover, also in confirmation of this previous report, we found that, with prolonged culture, the transgenic cells started to coexpress cytokine combinations like IFN-γ and IL-5 (close to 15% double-producers detected on day 42, Table I) that we have not seen directly ex vivo. This difference could be attributed to the continuous Ag/IL-2-driven proliferation that the T cells undergo in cell culture. Extensive cell cycling was shown to cause demethylation of cytokine genes and changes in chromatin structure (16), which makes cytokine genes more accessible and may favor cytokine coexpression. Our data suggest that such changes in cytokine expression may not readily occur in vivo, possibly because the T cells in vivo reach replicative senescence after 17–35 cell divisions (19).

In contrast to immunizations with CFA, which tend to induce the classic aspects of polarized type 1 immunity (IFN-γ⁺, IL-2⁺, IL-4⁻, IL-5⁻ cytokine recall, delayed-type hypersensitivity, production of specific IgG2a but no IgE Abs), protein Ag injected in IFA typically results in polarized type 2 immunity (IL-4⁺, IL-5⁺, IFN-γ⁻; IgG1⁺, IgE⁺, IgG2a⁻; no delayed-type hypersensitivity) (48, 54). The prevalence of the single-cytokine-producing T cell phenotype was also seen in the OVA:IFA-induced type 2 response (Table I and Fig. 7, A–F): ∼95% of the Ag-specific CD4 cells produced IL-4 and IL-5 or IL-2 and IL-5 mutually exclusively. Despite the vigorous Ag-specific IL-2 recall response, there was no IFN-γ induced. This dissociation of IL-2, IL-4, and IL-5, in the absence of IFN-γ, was seen over a wide range of Ag concentrations used for immunization and recall and using APC from different sources (LN, spleen, thymus of naive mice, or spleen of SCID mice, data not shown). Therefore, the overall type 2-polarized cytokine profile of the IFA-induced response was the result of different T cells individually producing the individual cytokines comprising this profile.

In addition to the OVA:IFA-injected mice shown here, we observed the induction of high-frequency IL-2-producing memory cells in the complete absence of a IFN-γ-producing memory cells after injection of seven other protein Ags in IFA into six different mouse strains (data not shown).

Unlike the immunizations with OVA protein in IFA, injections of BALB/c mice with OVA_323–339 peptide in IFA resulted in mixed “Th0-type” response with IL-2, IL-4, IL-5, and IFN-γ production (48). Two-color assay performed on CD4 cells separated from spleens of these mice 3 wk after immunization showed that each of these cytokines was produced by different cells (Table I).

To test whether dissociated cytokine expression is limited to adjuvant-induced responses or is a more general feature of the CD4 memory cells, we characterized the CD4 cells that were primed during the natural course of L. major infection in BALB/c mice (Table I). Not only did these cells show dissociated Ag-induced production of IL-2 and IFN-γ, but IL-4 production was seen in the virtual absence of IL-5 (Fig. 7, G–I). The dissociation of IL-4 and IL-5 was independent of the type of APC used; IL-4 and IFN-γ were also produced by different cells (Table I). The production of IL-4 in the absence of IL-5 was also seen when culture supernatants were studied by ELISA (data not shown). When Leishmania Ag was injected with IFA, it induced the IL-5-producing memory cells in addition to those secreting IL-2 and IL-4 but not IFN-γ (data not shown). The absence of IL-5 in Leishmania-infected animals cannot be attributed to an inhibiting effect of IFN-γ, because, based on the data from immunizations of BALB/c mice with OVA_323–339 peptide in IFA, these two cytokines are not expressed in a mutually exclusive manner on the population level (although they were at the single-cell level) either at the stage of priming (Table I) or at the stage of the recall (Fig. 4). Therefore, the presence or absence of IL-5-producing memory cells was not dictated by the nature of the Ag itself but seemed to be determined by the mode/microenvironment of the induction of the immune response.

The use of computer-assisted two-color cytokine ELISPOT overcame the limitation in measuring directly the Ag-induced cytokine coexpressed by the low-frequency, in vivo-differentiated memory cells from normal mice and avoided the pitfalls of in vivo propagation. As performed, two-color ELISPOT assays measured the production of pairs of cytokines simultaneously during the first 24 or 48 h (based on the well-established kinetics for each cytokine) after reencountering Ag, mimicking the initial effector phase after memory cell activation and before T cells start to proliferate and the memory resets itself by generating new daughter cells. By comparing cytokine ELISPOT (24 or 48 h assay) and intracytoplasmic cytokine staining/FACS analysis (6 h assay) of cloned cells (Fig. 1) or freshly isolated CD4 cells from the LN after polyclonal stimulation with anti-CD3 mAbs or Con A (data not shown), we showed that, indeed, this is the case, as within this time frame proliferation did not occur. Moreover, by measuring this early cytokine response, we wanted to ascertain that the cytokines we measure were produced by the in vivo-differentiated memory cell, as opposed to their daughter cells, which may or may not inherit the cytokine commitment of the mother cell. Because differentiation of naive cells into memory/effector cells requires longer than this 24–48 h, the cytokines measured were produced by in vivo-primed memory cells (see above).

Unlike dual-label in situ hybridization, which measures mRNA levels in the T cell at a single time point, these ELISPOT assays integrate the production of both cytokines over the entire assay period (immediately after the Ag was added to the cells), accounting for possible differences in the kinetics of cytokine production, including switching from the production of one cytokine to the another. Instead of measuring mRNA (or measuring the accumulation of cytokine in the cytoplasm of pharmacologically treated cells), the ELISPOT approach detects the immunologically relevant secretion of the cytokine, thereby accounting for posttranscriptional and posttranslational regulation. The detection limit for the minimum amount of cytokine produced per cell was found to be comparable in intracytoplasmic staining and in ELISPOT (Fig. 1) for cloned T cells. Also, the distribution of cells according to the amount of cytokine produced per cell was similar with both methods (Fig. 3). However, when it came to detecting the few Ag-specific cells within the majority of non-Ag-specific cells, only the ELISPOT assay was suited for this purpose, as it proved to be sensitive enough to detect one cytokine-producing cell within a million non-cytokine-producing cells (Fig. 1). In contrast, when the frequency of cytokine-producing cells fell below 1:1000, they became undetectable by intracytoplasmic staining; therefore, ELISPOT was 1000 times more sensitive to detect cytokine produced by rare Ag-specific cells in freshly isolated populations.

Irrespective of how the immune response was induced (by infection or immunization with adjuvant), the Ag doses used for priming and recall, or the time point tested, the cytokines measured were mostly produced by different CD4 memory cells; because very similar results were obtained in all in vivo systems tested, the cumulative data obtained in multiple independent experiments for all models studied are summarized in Fig. 8 (they represent the frequency of cytokine coexpression measured in over 1 × 10⁶ individual CD4 cells). The dissociation of IFN-γ from IL-5, and IL-4 and IL-3 from IL-5, was virtually complete. The coexpression of the “type 1” cytokines IFN-γ, IL-2, and IL-3 or the “type 2” cytokines IL-2, IL-4, and IL-5 were confined to 4–6% of cytokine-producing CD4 cells. We also found that IL-4-producing memory T cells had been engaged in the absence of IL-5 in Leishmania-induced immunity and that IL-2 had been produced in the absence of IFN-γ in IFA-induced type 2 immunity (Table I).

The level of cytokine dissociation in individual cells that we observed in vivo was striking as it approximates the precision of allelic exclusion, e.g., 4% for TCR α-chains (55); thus, it was consistent with highly regulated cytokine gene regulation in T cells physiologically stimulated by Ag. Our data can also be explained within the framework of stochastic cytokine gene expression, provided that only a small fraction of the activated memory T cells is actually induced to produce cytokine. Moreover, one would have to postulate additional very rapid and precise selection mechanisms (11) to explain why there are essentially no IL-5-producing memory cells present in Leishmania-induced immunity and why the nearly complete polarization of IL-4/IFN-γ and IL-5/IFN-γ is already seen at day 4 after immunization.

Alternatively, it is tempting to postulate that the differentiation and subsequent expansion of each of the individual cytokine-producing memory cell subpopulations is under independent instructive control. It is well-established that IL-12/IL-18 and IL-4/IL-13 are differentiation factors for the generation of memory cells expressing IFN-γ and IL-4 (3, 4). Recent evidence has emerged indicating that differentiation into IL-5-producing memory cells may have different differentiation requirements than IL-4 (56, 57), and this may apply for the other type 2 and type 1 cytokine as well. It was also demonstrated that the demethylation patterns of IFN-γ and IL-3 genes, which define their ability to be expressed upon T cell activation (16), are inherited in T cell lineages (17, 18). This observation might point toward the existence of memory cell sublineages with inherited commitment for expressing certain cytokine genes. Stringent control of the engagement and expansion of these single-cytokine-expressing memory cell lineages during the primary immune response would make the resulting composition of the memory cell effector functions finely tunable. When and where the Ag is reencountered for the second time, these memory cell lineages would express the individual cytokines to which they are precommitted. On the population level, they would create a cytokine microenvironment whose exact quality (which individual cytokine is produced and which is not) and magnitude (the population size of each single-cytokine-producing population) precisely execute the required combination of effector functions imprinted during the primary immune response.

Irrespective of the cytokine gene regulation mechanism underlying dissociated cytokine expression, the implications of these data in terms of immunobiology is that the different effector functions associated with the individual cytokines are each independently performed by the T cell system and, therefore, that cytokine-mediated effector functions of T cells are much more versatile and precise than anticipated. Dissociated expression of individual cytokines raises the repertoire of CD4 memory response types from two (IFN-γ/IL-4, Th1/Th2) to many discrete types, even within “type 1” and “type 2” immunity.

We thank R. Trezza for excellent technical assistance, Drs. P. Heeger and H. Radeke for providing the T cell clones, Dr. F. P. Heinzel for providing L. major-infected BALB/c mice and Leishmania Ag, and he and Drs. R. Fairchild and D. Kaufman for valuable discussions. The data were presented at the 1997 AAI meeting (1997 J. Alergy Clin. Immunol. 99:1490). We apologize to those authors whose work we could not cite due to limitation of space.