STAT6 tunes maximum T cell IL‐4 production from stochastically regulated Il4 alleles

T helper 2 (Th2) cells stochastically express from the Il4 locus but it has not been determined whether allelic expression is linked or independent. Here, we provide evidence that alleles are independently activated and inactivated. We compared Il4 locus expression in T cells from hemizygous IL‐4 reporter mice in culture and in vivo following exposure to type 2 immunogens. In culture, Il4 alleles had independent, heritable expression probabilities. Modeling showed that in co‐expressors, dual allele transcription occurs for only short periods, limiting per‐cell mRNA variation in individual cells within a population of Th2 cells. In vivo profiles suggested that early in the immune response, IL‐4 output was derived predominantly from single alleles, but co‐expression became more frequent over time and were tuned by STAT6, supporting the probabilistic regulation of Il4 alleles in vivo among committed IL‐4 producers. We suggest an imprinted probability of expression from individual alleles with a short transcriptional shutoff time controls the magnitude of T cell IL‐4 output, but the amount produced per allele is amplified by STAT6 signaling. This form of regulation may be a relevant general mechanism governing cytokine expression.


INTRODUCTION
Not all equivalently differentiated Th2 cell clones express from their Il4 locus following synchronous activation.A body of work suggests that the expression is stochastic, and relates to an underlying probabilistic regulation mechanism, resulting in all-or-none allele transcription and translation upon activation, 1-3 referred to as an expression probability (P EXP ).The GATA3 expression amounts are equivalent in the expressing and nonexpressing cells of a clone, [4][5][6] and both expressing and non-expressing cells exhibit similar nuclear abundances of nuclear factor and activator of transcription (NFAT) c2. 6 EXP is related to the extent of demethylation across the Il4 locus, 4,7 and expression per se associated with 2-3-fold higher accessibility at the Il4 hypersensitivity site (HS) V A , 6,8 as well as the amount of NFATc2 and associated transcription factors bound to that site.9 The TCR stimulation strength impacts locus P EXP , 1,2 but neither enforced STAT-6 over-expression nor multiple rounds of IL-4 exposure alter the probability of all-or-none expression in clones.4 It is currently unknown whether stochastic activation simultaneously activates both Il4 alleles, or operates on the alleles independently.][11][12][13] Mariani et al. 5 also elegantly demonstrated that accessibility of the Il4 locus in IL-4 producers was not a linked phenomenon and that allelic competence and expression were distinctly regulated events.
However, other data suggest that IL-4 production instead derives from synchronously expressed alleles.Activated Th2 cells produce a uniform log-normal IL-4 expression peak, rather than three resolvable peaks for non-expressing, single-allele expressing and co-expressing Th2 cells, which has been suggested as evidence against independent regulation. 9Further contention with independence arises from the essentially biallelic expression pattern of the 4get IL-4 reporter, 14 and the co-expression of human CD2 with GFP in > 85% of hCD2 + cells in 4get/KN2 Th2 cultures, 15,16 suggesting the expression of alleles is associated, if not linked, in these cases.In a view that goes some way towards accommodating the collective observations, William Paul argued that for Il4, Th2 cells gain transcriptional competence on both alleles, but then express alleles with independent and low probabilities upon stimulationallele co-expression arises as a function of the independent probabilities. 3,11Thus, most contention in the means of regulation relates to the inferred likelihood of expression in Th cells being high or low when activated to express, and the degree to which expression of one allele predicts expression of the other.Whether in vivo allelic use profiles are consistent with the independent regulation is also not established.
We sought to clarify whether the regulation of individual Il4 alleles was relevant to the biology underlying Il4 expression.We applied Th cell differentiation protocols and type 2 immunogen immunizations to profile expression of the murine Il4 locus in hemizygous crosses of the native Il4 allele with the IL-4-substituting KN2 16 and G4 3 reporter alleles (Figure 1).Both reporters are under the control of the murine Il4 promoter, with KN2 replacing the first two Il4 exons plus the intervening intronic region with the gene encoding a non-functional human CD2 linked to the murine CD4 enhancer (hCD2; Mohrs et al., 16 Sawada et al. 17 ).The G4 allele replaces exon 1 and 178 nucleotides of the first Il4 intronic region with the gene encoding enhanced green fluorescent protein (GFP). 3As such, both reporters lack a GATA3 binding site in the first Il4 intron 7 but leave remaining Il4 regulatory regions unmodified.In vitro, we confirmed independent, heritable, expression of the Il4 locus in committed Th2 cells.Our in vivo data were consistent with probabilistic locus expression operating on alleles independently as well, but impacted by STAT6 signaling.The independent regulation is hypothesized to constrain the amount of mRNA transcribed by the average cell following stimulation.We suggest this means of regulation is key to Il4 locus expression and IL-4 production in the context of type 2 immunity, and might be relevant for the regulation of expression of other cytokines.

Reporters driven by the Il4 promoter reflect Il4 locus expression
To determine whether Il4 alleles are expressed synchronously after stimulation we examined IL-4 expression in wild type (wt) mice, Il4 hemizygous mice with one allele replaced with one of two reporters under control of the Il4 promoter, G4 and KN2, and IL-4 null G4/KN2 mice.We first examined how G4 and KN2 reflected the behavior of the native Il4 locus during in vitro Th2 priming.CD3-stimulated KN2/G4 T cells, at 20 h, exhibited 3.7-fold more cells expressing hCD2 than at T = 0, while GFP had increased 2.5-fold (Figure 2a).hCD2 expression reached a maximum of ≈40% of the cultured cells by 36 h and remained largely unchanged thereafter (Figure 2a, b).GFP similarly reached its maximum at 36 h, and maintained its plateau at a level of ≈15% (Figure 2a, b).At each measured time, the proportional shifts between baseline and maximal expression of that reporter were similar for hCD2 and GFP (Figure 2b).Thus, the two reporter alleles exhibited different magnitudes of expression, but similar kinetics of induction toward those maximum values.
In cultures continued over several days, and including additional cultures from IL-4 null KN2/KN2 and G4/G4 mice, we observed that resting caused the proportion of hCD2 + and GFP + cells to diminish (Figure 2c, d).We note here that while G4/G4 mice exhibited approximately twice as many GFP + cells as either the KN2/G4 or G4/+ single-allele-reporter mice, the difference in hCD2 + fractions between KN2/KN2 and the relevant KN2/+ and KN2/G4 strains was less marked (Figure 2c, d).During the rest phase, surface hCD2 positivity diminished more rapidly than did intracellular GFP, with a calculated decay T 1/2 of 23.99 h for hCD2 and 40.42 h for GFP (Figure 2e).The GFP reporter thus exhibited a longer half-life than the hCD2 reporter, but the expression of both reporters was diminished, and diminished at set rates, upon the removal of IL-4 and TCR stimulation.
Thus far, we have observed several differences in the expression of hCD2 and GFP under Th2 differentiated conditions, indicating differences in the behaviors of the   reporters that might confound our investigations of locus regulation; namely, (i) the absolute expression frequencies differed by more than 2-fold in Th2 cultures, (ii) the decay rate of the protein reporters differed by approximately 2-fold, and (iii) the difference between hCD2 expression in KN2/KN2 and KN2/G4 cultures was only 10%, whereas the difference in GFP was 40-50% in G4/G4 and KN2/G4.The difference in maximal expression under Th2 conditions suggests the two alleles were differentially IL-4 responsive, but, contingent on both being co-expressed as the multiple of the independent probabilities, would still fit with independent regulation of alleles.Therefore, we designed a series of studies to ask (i) whether expression profiles were heritable, (ii) whether expression profiles were consistent with independent allelic regulation, and (iii)   whether the native Il4 allele behaved differently with the two different counterpart alleles.As long-term Th2 clones exhibit allelic biasing, with clones expressing alleles with equivalent probabilities as the parental line, 3 we tested whether expression profiles of KN2 and G4 exhibited distributions reflecting heritable P EXP upon activation.Populations sorted on their allelic expression profile (Figure 3a) were rested and restimulated (Figure 3b-d).
The restimulated cells fell in four relatively discrete populations (Figure 3b).Sorted and restimulated hCD2and GFP-expressing populations had essentially reverted to the same hCD2 and GFP expression profile as the pre-sorted cells, rather than retaining their sorted profile (Figure 3c).However, there was a slight preference for re-expression of the previously expressed alleles.This can be accounted for by a short-term allelic 'memory' based on transcriptional permissivity of a formerly expressed IL-4 allele.With a previously demonstrated T 1/2 of 2.9 days, 5 this would estimate approximately 5% of the sorted cells should retain a re-expression advantage for their previously expressed allele(s), which likely explains these outcomes.Relative to the sorted locus-expressing cells, the hCD2 neg GFP neg cells had about 40% lower frequencies of cells expressing from the Il4 locus at all upon restimulation (Figure 3d), but those that did had approximately the expression profile of the other cultures, fitting with a heritable P EXP upon activation and showing that those that did express were similar to those in the fractions sorted as expressors (Figure 3c).The double negative-sorted fraction probably contained a less-Th2-differentiated population, in addition to those that are strongly Th2-committed but not expressing at the time of sorting, which may explain the maintenance of doublenegative cells observed after restimulation of this population relative to the others.It is also possible that these cells (and single-positive sorted cells) had expressed the locus transiently before the sort occurred, or that single-expressors preferentially grew out in the cultures, confounding the outcome from the restimulations, although the half-lives of the reporters make this an unlikely scenario for the overall result.Thus, both G4 and KN2 exhibited heritable, probabilistic expressions in vitro.The heritable aspect was the probability of expression being higher in the Th2 cultured cells than at baseline, and the probabilistic component was the specific allele combination to be expressed upon restimulation.

Il4 alleles are expressed at distinct but predictable frequencies in differentiated Th2 cells
To evaluate the allelic Il4 expression regulation mechanism, we considered that inherent allele coexpression upon all-or-none locus activation predicts the average wt T cell to express 2× as much IL-4 as the average hemizygous T cell.To test this, we measured the expression of IL-4 protein in cells from wt, KN2/+, G4/+ and KN2/G4 mice, differentiated, rested and then restimulated.At various times, cells were harvested and stained for hCD2 and intracytoplasmic IL-4.
The IL-4 expression histograms of hemizygotes overlaid each other (Figure 4a and Supplementary figure 1a), indicating the native Il4 allele indeed behaved similarly in the presence of the two different reporters.Plots depicting hCD2, GFP and IL-4 expression in different heterozygote strains before and 6 h after restimulation are shown in the data supplement (Supplementary figure 1b).In these restimulation assays, GFP expression profiles in G4/+ and KN2/G4 were also similar, as were hCD2 expression profiles from KN2/+ and KN2/G4 mice (Supplementary figure 1c, d).In the KN2/G4 line, the observed hCD2 and GFP co-expression frequencies were close to those calculated for independently regulated alleles (Supplementary figure 1e), again consistent with probabilistic expression.Th0 activated CD4 + T cells expressed IL-4 predominantly from individual alleles, as reflected in the differential in GFP and hCD2 positivity in respective hemizygous and homozygous G4 and KN2 reporter strains (Supplementary figure 1f), suggesting the difference in behavior of the alleles was an IL-4-driven phenomenon.Further, CD8 + T cells cultured alongside differentiated, reactivated Th2 cells were largely restricted to expressing only single reporters upon activation (Supplementary figure 1g).Thus, studies with the G4 and KN2 reporters overwhelmingly supported independent regulation of IL4 alleles, with co-expression only becoming frequent in differentiated Th2 cells.We acknowledge here that both reporters lack a GATA3 binding site associated with recall IL-4 response 'memory' in Th2 lines in vitro, 7 so the reporters only reflect Il4 locus expression in part.However, as the IL-4 protein expression appeared similar in both hemizygotes, it seemed reasonable to us to infer locus behavior by comparing expression profiles of IL-4 in wild type and hemizygote Th2 cells in the cultures.
Overton's subtraction is a method by which a negative distribution of cells can be subtracted from a mixed negative and positive population. 18This method accommodates lowly positive values that do not separate above the negative population distribution, but rather are seen as a 'shift' in the average expression intensity, allowing for more accurate determination of the expression frequency than gating above the top of the negative control distribution. 18At 6 h post-stimulation, by Overton's subtraction, 18 ≈50% of wt cells were IL-4 + , and ≈38% were IL-4 + in each of the hemizygous strains (Figure 4b), with approximately that differential maintained over time (Figure 4b).While hemizygotes expressed similar IL-4 amounts as each other, small fractions of wt cells attained expression amounts in excess of those possible from the hemizygotes, revealed in both the proportion of cells expressing amounts equivalent to the 99 th percentile of the wt distribution (Figure 4a), and the MFI of the top 1% of each strains' distribution (Figure 4c).When we focused only on IL-4-expressing cells, removing non-expressing cell distributions by removing negative peak distributions in a manner approximating the subtractive method outlined by Overton, 18 the geometric mean fluorescence (geo medFI) of the wt IL-4 + cell distribution was approximately 20% higher than for the hemizygote IL-4 + cell distribution, and not 2-fold lower as would be predicted from an equivalent proportion of cells activating to express but expressing different quanta based on one or two alleles (Figure 4d, e).The normalization and subtraction process is depicted in Figure 4d.We note that the 'background' fluorescence is not accommodated in the result, but the values were quite removed from the background, so we think this does not confound these analyses.These data suggest that cells possessing two functional Il4 alleles can produce more IL-4 than those carrying only one, but do not produce exactly 2× the amount as the average hemizygous cell.

Stochastic activation and cessation of transcription accounts for IL-4 production variations
We examined whether a regulation model in which allele expression is stochastic could accommodate such an expression profile.We applied a simple model using previously established expression parameters. 5In this modeling (Figure 5a), stimulation recruits NFAT1 and associated transcription factors to 100% of alleles.While NFAT1 and the transcription complex is in place (TF ON ), an imprinted initiation probability permits all-or-none transcription and translation of the locus.When that complex is inactivated or dissociates from the allele (TF OFF ), neither transcription initiation nor continuing transcription occur. 5The window for active transcription may be more constrained than we have allowed in our model, and reconfiguration of the transcriptional machinery could conceivably shut down both alleles synchronously in a narrow time window, 19 which would impact these mechanics.However, to let the model operate with a minimal set of assumptions, we set the transcription shutoff as independent for the two alleles.Overall, the model is essentially a set of intersecting probabilities: that of the Pexp after TF ON , and then that of the TF OFF probability after activation at the P EXP (Figure 5a).The figure (Figure 5a) depicts a single allele, but the model accounts for the same probabilities applying for both alleles, and the outcome is the intersection of the dual probabilities.All other parameters proceed at constants defined by Mariani et al., highlighted (Figure 5a) and outlined again in the Methods section.We modeled T cells with low P EXP of 0.05/h, resembling weakly polarized Th2 cells; those approximating 2-week polarized Th2 cells with heavily remodeled II4 locus chromatin (0.21/h); and those at (0.35/h) or beyond (0.6/h) presumed maximum P EXP in T cells.In a given stimulation round, the fraction of cells expressing Il4 Allele One is related to the P EXP , with more than 50% of strongly Th2biased cell populations predicted to transcribe from that allele in 6 h (Figure 5b).Because the modeled transcription is dependent on ongoing NFAT occupancy, which is lost at the TF OFF , at most, 40% of cells are predicted to actively transcribe from Allele One at any given point in time (Figure 5c).Allele co-transcription at any point in time is also considerably less frequent than the multiple of the two P EXP , estimated at 5-15% of actively transcribing cells (Figure 5d), as it requires the temporal intersection of independent transcription events to occur.Co-transcription should revert to single-allele transcription at twice the rate of the TF OFF T 1/2 (i.e.39 0 here [Figure 5e]).The implication is that cells attaining co-expression transcribe more than single-expressors, but their total mRNA and protein accrual is constrained relative to a situation of linked, co-activated expression of both alleles.Accordingly, Il4 mRNA accumulated by the average Il4-transcribing wt cell is predicted to be no more than 25% in excess of the average hemizygous cell (Figure 5f).Thus, the lack of discrete expression peaks for wt versus hemizygous cells may be explained by temporal dissociation of independently activated and inactivated events.We note here that there are additional mRNA regulation, translation and unknown post-translational mechanisms that may also impact the profile to lead to these results of a non 0, 1 or two allele expression profile observed in wt stimulated Th2 cells, so acknowledge that this is not the only possible explanation, but see this as a plausible mechanism underlying locus opening, transcription and translation of cytokines that fits with prior observations of locus behavior.

Il4 expression profiles in allele-marked IL4 reporter mice in vivo
We extended our evaluations to the in vivo setting.To do this, we injected s.c.house dust mite (HDM) into ear pinnae and compared the kinetics of Il4 locus expression in draining lymph node CD3 + CD4 + T cells through the surrogate KN2 and G4 alleles (Figure 6a).Three days after immunization, the GFP+ cells in G4/+ and KN2/G4 strains were at approximately half the frequency observed in the G4/G4 strain (Figure 6b).Likewise, hCD2+ cell proportions in KN2/+ and KN2/G4 mice were similar, and approximately half those of KN2/KN2 mice (Figure 6c).On day 7, KN2/G4 mice generated fewer GFP or hCD2 expressing cells than the relevant G4/G4 and KN2/KN2 strains (Figure 6a-c).However, we observed an IL-4-dependent increase in reporter + cells in both reporter backgrounds on day 7 (Figure 6a-c).The cells coexpressing hCD2 and GFP in the KN2/G4 strain, at all timepoints, expressed more hCD2 and more GFP per allele than the hCD2-and GFP-only expressing KN2/G4 cells, with these MFI sitting either side of the MFI of the respective single reporter-carrying G4/+ and KN2/+ strains (Figure 6d, e).In the presence of IL-4 (in G4/+ and KN2/+ lines), the MFIs were also shifted to be close to those observed in the dual-reporter homozygotes (G4/G4 and KN2/KN2 lines), suggesting IL-4 can influence the amounts of expression of the alleles (Figure 6d, e).In the CD4 + T cells from HDM-immunized KN2/G4 mice, the co-expression of hCD2 and GFP increased from 10% of locus-expressing cells on day 3 to about 20% by day 7 (Figure 6f), suggesting a shift toward allele co-expression over time.This 20% value was stable to day 10 in the KN2/G4 mice.These data suggest that as the cells progressed through the immune response, there was a shift toward allele co-expression, with slightly elevated output from each allele occurring in those co-expressing cells.
At this point, we highlight that the comparisons between KN2/G4 and G4/G4 or KN2/KN2 (i.e.IL-4 null ) strains report one or two alleles reporting for a single construct, and these are generally maintained at about a 60% differential, implying most cells are expressing from single alleles in these strains.Even so, the co-expression of reporters in some 20% of KN2/G4 cells would suggest the P EXP of each allele in the expressing cells is reasonably high.The comparisons in the KN2/+ and G4/+ strains show elevations in hCD2 and GFP representations among CD4 + T cells, implying IL-4dependent expansion of the Th2-lineage cells in these cases.That the fluorescence amounts of the reporters reach the approximate levels observed in the homozygotes points to IL-4 also influencing expression amount, as otherwise the alleles would remain at lower fluorescences than the G4/G4 and KN2/KN2 strains.With this in mind, we sought to clarify whether IL-4-signaling via STAT6 could affect allelic Il4 expression amounts.

Th cell Il4 output can be tuned
We sought to address the question of whether Il4 output from a T cell can be regulated.It has been previously argued that probabilistic expression could confer on Th cell populations a likelihood of expression proportional to the strength of the Th2-inducing stimulus, with a set amount of production occurring upon all-or-none individual allele transcription. 11That is, the expectation was that irrespective of the magnitude of the Th2 response, the expressing cells should produce similar IL-4 amounts from individual alleles.Therefore, we drew on prior knowledge that helminth parasites have a major IL-4-committed cell-inducing bias that distinguishes them from more 'normal' allergens such as HDM and ovalbumin (OVA), 20 and compared the potential for Il4 induction by two distinct immunogens, dead L3 N. brasiliensis larvae, 20 and OVA admixed with alum adjuvant.We used OVA/alum rather than HDM in this case to generate a similarly-sized total Th response relative to the Nb parasites, such that the magnitude of the response was less confounding for the analyses.
As expected draining lymph nodes from the immunized mice carried similar CD4 + T cell numbers (Figure 7a), but dead Nb generated 5-fold more GFP-expressing cells than did OVA/alum (Figure 7b).This was associated with threefold more abundant CXCR5 + PD1 hi follicular T helper cells in response to the N. brasiliensis larvae (Figure 7c), and significantly more Il4 locus-expressing cells among the TFH of the dead Nb immunized mice (Figure 7d).Unexpectedly, although perhaps consistent with the previous HDM analysis in the KN2/+ and G4/+ lines relative to KN2/KN2 and G4/G4 lines (Figure 6d, e), GFP + CXCR5 + PD1 hi follicular T helper cells generated in response to dead N. brasiliensis expressed higher GFP MFIs than those generated by OVA/alum immunization (Figure 7e, f).Rather than reflecting increased transcription of the allele, this probably reflects a reduced rate of transcript degradation. 21he finding that per-allele IL-4 outputs may be tunable was surprising to us, as the strength of the Th2-inducing stimulus typically results in more IL-4-expressing cells rather than greatly modulating average expression amounts in Th2 lines. 3We were therefore interested to understand whether there were qualitative means by which IL-4 expression amounts could be increased.STAT6 phosphorylation downstream of IL-4Rα promotes Th2 cell IL-4 production in vivo, 22 and thus might be the means by which IL-4-driven increases in GFP and hCD2 expression amounts in the G4/+ and KN2/+ lines relative to the G4/G4 and KN2/KN2 lines (Figure 6d, e) might occur.We therefore assessed whether STAT6 affected Il-4 locus expression frequency or the amount of transcription of an individual allele in the G4/+ line.We observed that both the proportion and GFP amount per cell from the G4 allele was enhanced by STAT6 signaling (Figure 7g-j).These data demonstrated that both the frequency of locus expression and the total translational output could be modulated; TFH per-cell IL-4 output is thus a tunable event, but this occurs in addition to the underlying regulatory mechanism being independent activation and cessation of transcription on alleles individually.

DISCUSSION
We provide evidence that IL-4 expression profiles in vitro and in vivo can be accounted for by the independent activation and cessation of transcription of Il4 alleles.We conclude that both native Il4 and reporter alleles carry intrinsic P EXP .These probabilities increase as cells become Th2 in vitro and increase in vivo over time as the Th2type follicular T helper cell population matures.This increase in allelic P EXP is also associated with a STAT6dependent increase in per allele output, making Il4 expression a tunable but probabilistic event.
The relevance of this form of regulation to IL-4 biology is that bursts of IL-4 production following stimulation arise stochastically but are temporally constrained, such that the inheritance of two different alleles with slightly different P EXP can have less impact on IL-4 production at the cellpopulation level.With this model, the total IL-4 output from wt mice is predicted to be ≈2-fold above that from hemizygotes, but that total output is distributed across more cells, and the estimated constraint of co-expression in time limits average wt Il4 mRNA transcript accumulation to be in the range of only 20-25% higher than the average hemizygous cell in the most highly Th2-genic contexts.This would explain the comparatively restricted ≈20% increase in mean per-cell Il4 transcript numbers determined by in situ hybridization in differentiated Th2 cells relative to naı ¨ve Il4 transcribing cells, 23 alongside the ≈2-fold higher maximal transcript abundance in those differentiated Th2 cells. 23Interestingly, our data, here (Figure 7i, j) suggest that the IL-4 expression amount, per allele, is increased by STAT6 signaling.This suggests that IL4 regulation is more complicated than on/off signaling induction alone, as it is susceptible to modulation as cells become more Th2 committed in vivo.All-or-none expression is thus not the only parameter to consider in understanding the expression of the Il4 locus in vivo, although tunability is probably less important than P EXP for overall expression amounts, given that mean per-cell Il4 amounts increase only 20% comparing naı ¨ve to differentiated Th2 cells. 23ccording to the model put forward by Mariani et al., 5 it is the locus accessibility (i.e. a chromatin opening step) that restricts expression from an allele.By inference, once the chromatin is accessible, restimulation would permit immediate transcription during any period of TF ON , implying that serially stimulated cells would eventually coexpress with each stimulation.We cannot discount this outcome, but an earlier report with the KN2/+ hemizygote shows that despite similar proportions of IL-4 + and hCD2 + lung Th2 cells after ex vivo restimulation, IL-4-expressing cells comprise but a minor fraction of those expressing hCD2. 24We would accommodate this within the model by suggesting that the retention of re-phosphorylated NFAT1, or another pre-transcriptional but probabilistic restraint on expression limits locus expression in every stimulation round.That is, probabilistic expression occurs subsequent to chromatin remodeling and establishment of transcriptional competence on the individual alleles.
If a difference in the expression amounts between single-and dual-allele-expressing cells can arise, should it matter for the IL-4-driven response outcome?We would suggest that the immune response has instead evolved to limit such potential, with stochastic expression acting to restrict cell population-level variation in production amounts.Numerous aspects of IL-4 biology would support this position.For example, IL-4 is secreted nondirectionally rather than synaptically, 25 suggesting that proximity to an IL-4-secreting cell should permit capture of reasonable IL-4 amounts.The affinity of IL-4 for the IL-4Rα is of the order of 10 À10 M, 26 making comparatively low local IL-4 concentrations saturating.8][29] Thus, competition of occupancy of the common γ-chain is likely a key event affecting the outcomes of T cell:antigen-presenting cell interactions.Many biological processes affected by IL-4 are elicited by comparatively small IL-4 amounts.For example, IL-4 is known to promote IgG1 class-switch recombination at 1000-fold lower concentrations than it promotes IgE production. 30Given that IgE class-switch recombination occurs rarely or not at all in the sites at which most lymphoid IL-4-producing cells are located, germinal centers, 31,32 a small amount of IL-4 is likely sufficient for most IL-4-sensitive germinal center processes and germinal center responses probably unperturbed by reasonably large differences in IL-4 amounts.However, we also acknowledge that our major observations of the biallelic expression profile of the Il4 locus came from Il4 null dual allele reporter mice, so there is the caveat that IL-4 allele co-expression, driven by IL-4, may become higher than the 20% value we observed for the KN2/G4 response, which might provide a larger differential in expression for cells interacting with dual allele expressors.
Il4 is not the only cytokine reported to have each allele regulated independently, although it is perhaps the most consistently reported as so (e.g.Bayley et al. 12 ).Initial reports demonstrated IL-2 as monoallelically regulated, based among other findings on hemizygous and homozygous reporter mice to exhibit ≈2-fold abundance differentials in frequencies of cells transcribing from the locus, 33 similar to our general observations with the KN2 and G4 hemizygotes.Subsequent work suggested that many IL-2-expressing cells do so biallelically and that the gene is not monoallelically expressed. 34,35That Il2 alleles are not independently regulated is more compatible with a graded change in the amount of transcription as well as the frequency of transcription observed in activated Th cells, 1 whereas Il4 and Ifng expression levels are less modulatable in T cells. 1 However, if the P EXP for transcription is high in a given context, independently regulated alleles (or genes) will appear to be linked, and this may be the case for Il2.For example, studies indicate that in Th1 clones, the Ifng P EXP is high, 36 predicating  common biallelic transcriptionany gene where the P EXP exceeds 0.67 will exhibit more than 1/2 of the expressing cells co-transcribing, and we see this as the probable case for expression of Ifng in Th1 cells.This alone, however, would not account for the 'tunability' of IL-2 affecting the amount that activated T cells express, based on the amount of co-stimulation applied. 1 An additional candidate for probabilistic transcriptional regulation is the gene Il21.While scant data exist on its means of regulation, we espouse probabilistic regulation for three main reasons: (i) in vivo, approximately 1/3 of TFH express an IL-21 reporter at any given time after nitrophenyl-OVA immunization 37 ; (ii) even after presumably strong stimulation, reasonably fixed proportions of TFH exhibit expression of reporters for IL-4, IL-21 or both in predictable profiles 38 ; and (iii) the genes for IL-21R and IL-4Rα are in close proximity, centered around a common super-enhancer site, 39 potentially linking for evolutionary selection the fate of B cells in environments rich in IL-4, IL-21 or in co-producing TFH.The provision of IL-21 to B cells with or without IL-4 may result in distinct outcomes for the differentiation unfolding from the germinal center.We note that IL-4 and IL-21 co-expression in vivo has been monitored (e.g.Shulman et al., 38 Weinstein et al. 40 ), but adapting the reporter resolutions to approximate actual periods of expression would be required to appropriately establish the potential importance of individual versus co-cytokine expression patterns on interaction outcomes in vivo.
In conclusion, we find that Il4 locus expression variation in vitro can be accounted for by the independent regulation of alleles; that in vivo expression profiles are most compatible with this form of regulation; and that time is a key parameter to consider in addressing how probabilistic regulation may 'tone' the output for expressing cells.Stochastic regulation in this manner may be important for the expression of other cytokines, which we suggest include IL-21, and genes exhibiting allelic biasing in general.

METHODS Mice
BALB/c mice were from the Wellington School of Medicine maintained at the Malaghan Institute of Medical Research Biomedical Research Unit (BRU; Victoria University of Wellington, Wellington, New Zealand).STAT6 À/À breeding pairs were obtained from mice originally from Kaplan et al. 41 KN2 mice 16 were donated by Richard Locksley (University of California at San Francisco, CA, USA).G4 mice 3 were donated by William Paul (National Institutes of Health, Bethesda, MD, USA).All strains were backcrossed to a BALB/c background prior to arrival at the BRU.Experimental mice were maintained in a 14/10 hour light-dark cycle at the BRU.Mice were euthanized by either cervical dislocation or carbon dioxide asphyxiation.Mice were used in accordance with the permissions of the Victoria University of Wellington Animal Ethics Committee.

Antigens and immunizations
Crushed Dermatophagoides pteronyssinus house dust mite extract (HDM) was purchased from Greer laboratories (Lenoir, NC, USA; XPB82D3A2.5).Albumin from chicken egg white (OVA) was purchased from Sigma (St Louis, MO, USA).Nippostrongylus brasiliensis larvae were maintained at the Malaghan Institute of Medical Research as described. 42Dead L3 larvae were killed by three freeze-thaw cycles at À80°C.For ear injections, after anesthesia with 200 μL anesthetic solution (100 mg/kg ketamine [Ceva Animal Health Pty Ltd, Glenorie, Australia], 3 mg/kg xylazine [Phoenix Pharm, Auckland, New Zealand]), mice were intradermally injected in the outer ear pinna, generally the left ear, with 30 μL of immunogen in PBS.The draining auricular lymph node was later excised and examined for live cells by trypan blue exclusion and flow cytometry.

Trypan blue exclusion
Samples were diluted 1:1 with 0.4% trypan blue (Invitrogen, Waltham, USA).Between 30 and 200 cells were counted and cell numbers back-calculated from frequencies determined by flow cytometry.

Spleen isolation
After excision, spleens were placed on a 70-μm cell strainer and dissociated with pressure against the strainer using the rubber end of a 1-mL tuberculin syringe (BD Biosciences, Franklin Lakes, USA) in Iscove's modified Dulbecco's medium + Glutamax (IMDM; Invitrogen).Cells were spun down and resuspended in 2 mL red blood cell lysis buffer for 2 min at room temperature.The reaction was quenched with 13 mL complete media (cIMDM; 5-10% fetal calf serum (Sigma)), 100 U mL À1 penicillin/100 μg mL À1 streptomycin (Invitrogen), 55 μM β-mercaptoethanol (Invitrogen) and resuspended in either cIMDM for culture or in flow buffer (0.01% NaN 3 , 2% fetal calf serum in PBS) for flow cytometry.Tubes were spun down for 4 min at rcf max = 419.Spleen cells were then resuspended in 10 mL buffer for counting.The cells were respun and resuspended for future use.

Cell culture and cell culture reagents
During preparation, the cells were handled in IMDM and cultured in cIMDM.Standard incubation conditions of 37°C humidified in 5% CO 2 in air were used.Recombinant murine IL-2 was from the X62-IL-2 cell line used from aliquoted supernatant stored at -80°C and was used for Figure 1 cultures.The remaining cultures used recombinant human interleukin-2 (Proleukin; Novartis Vaccines & Diagnostics, Basel, Switzerland) which was stored lyophilized at 4°C.It was made up in IMDM at a concentration of 105 U mL À1 , and single-use aliquots stored at À20°C.Chinese Hamster Ovary cell line CHO-mIL-4 was cultured to produce murine IL-4 protein.Supernatants containing IL-4 were tested using IL-4 dependent cell lines (CT.4S) to determine bioactivity.The aliquots were stored at À80°C.

Th cell differentiation
Cultures used either fractionated CD4 + T cells or total splenocytes without enriching for naı ¨ve cells.For culture, 24-well plates (BD) were coated overnight at 4°C, or for 2 h at 37°C with anti-CD3 in PBS (2C11, 1 μg mL À1 , 300 μL/well; prepared in house).On the day of the assay, the plates were washed twice with 500 μL IMDM and 1 × 10 6 splenocytes from naive mice of the specified genotype were added to each well in a 1 mL culture volume.Cultures were left untouched for 3 days, at which point the wells were resuspended and half of the cells were removed and discarded and 500 μL of 2× conditioning medium was again added to wells.On day 4, 500 μL was removed without resuspension and cells were fed as on day 3.For restimulation assays, cells were removed from wells on day 5, washed twice in IMDM and replated in wells containing 100 U mL À1 IL-2 and 10 μg mL À1 anti-IL-4 (11B11; prepared in house) as the final culture conditions.During the rest period, half of the medium was replaced every day and if the medium reached pH = 7.0 or less in any culture, determined by a change in phenol red coloring, the contents of the wells were split in two.Separate wells were maintained for individual mice from the start of culture throughout the process, such that data points represent biological replicates.For initial conditioning, Th0 medium contained 100 U mL À1 IL-2, 10 μg mL À1 anti-IL-4 and 10 μg mL À1 anti-IFNγ (AN.18; prepared in house).Th2 medium contained 100 U mL À1 IL-2 and 1000 U mL À1 CHO-mIL-4 (IL-4 activity was compared by ELISA using standards originally calibrated using the CT.4S cell line as described. 43One unit is equivalent to approximately 0.5 pg IL-4 mL À143 ).During primary stimulation, all cultures contained 20 μL mL À1 anti-CD28 (37.51; grown and titrated in house without purification) that was not replaced during subsequent feeds.

In vitro restimulation
After conditioning, Th cells were transferred to wells coated with anti-CD3 at 10 μg mL À1 . 44Restimulation medium included 100 U mL À1 IL-2 and 20 μL mL À1 anti-CD28.For the detection of intracellular cytokines, 4 μL Golgistop containing monensin (BD) was combined with 16 μL IMDM and 20 μL was added to each well 2 h before the cells were harvested.
For surface-only staining 1.25 × 10 6 cells were transferred to 5-mL flow-compatible tubes through 70-μm gauze.Flow buffer containing 50 μL anti-CD16/32 antibody (24G2) made in house was added to each tube for 5 min which was then incubated with 50 μL containing the primary antibody mix.Samples were incubated for 30 min at 4°C and then washed with 4 mL flow buffer and resuspended for flow cytometry.For secondary antibody incubation, cells were instead resuspended in 100 μL containing optimally diluted secondary reagent.

Intracellular staining
Cells were harvested and transferred to flow-compatible tubes containing 3 mL 4°C PBS.Tubes were spun and resuspended in 1 mL PBS for counting.Cells were respun and incubated with 1 mL 1/1000 Live/dead fixable blue in PBS for 30 min at 4°C in the dark.Cells were washed with 3 mL flow buffer and were subjected to normal surface staining methods.After staining, cells were fixed and permeabilized using the Fix & Perm cell permeabilization kit (Invitrogen) according to the manufacturer's instructions with the following modifications.Cells were fixed for 15 min using 'Reagent A' at room temperature, followed by quenching with 4 mL room temperature flow buffer.Samples were split in two at this stage and spun down at 1400 rpm for 4 min at room temperature.Intracellular stains for analysis were added diluted to 100 μL volumes of reagent B. After spinning down the cells, intracellular stains or isotypes were added to samples and incubated for 20 min at room temperature.The staining was stopped by addition of flow buffer and samples were spun down then resuspended for flow cytometry.Samples were analyzed on an LSR-II special order product (BD

Sort experiments
Cells were handled in 0.002% DNAse I (Roche, Basel, Switzerland) and 2% fetal calf serum in PBS.For Figure 2, 5 days cultured KN2/G4 splenocytes were stained with antibodies against CD3, CD4 and hCD2 and sorted as CD3 + CD4 + and then by hCD2 and GFP into four populations as depicted (Figure 2a) on a FACSVantage or FACSARIA III cell sorter using DIVA software (BD).

Statistical analysis and modeling
Data were graphed and analyzed in Graphpad Prism.Statistical analyses used are stated in the figure captions.With the exception of MFI determinations (Figure 4d, e), flow cytometry data were analyzed in FlowJo software (Treestar).For the subtraction used to determine the expression amounts in the positive fractions, negative distributions for IL-4 were concatenates of n = 3 biological replicates of stimulated KN2/G4 splenic CD4 + cells cultures at each time point, evaluating these as baselines with data distributed into the default 300 bins.For hCD2 and GFP negative distributions (Supplementary figure 1), stimulated wild type cells were used.Expected co-expression probabilities for ever-expressors were determined as the multiple of the independent hCD2 + and GFP + fractions from the proportions determined by Overton's subtraction, while hCD2 + GFP + fractions were estimated by standard cytometric gating.To determine the mean fluorescence values in positive fractions, compensated IL-4 MFI for individual events were exported from Flowjo as .csvfiles.Data for all files were converted to integers using the ROUND function in excel, and then histogram dot-plots were generated in Graphpad prism by binning integer count data by IL-4 fluorescence bins of 5. Data were normalized to the maxima for each strain and then the KN2/G4 distribution was removed from the wild type and IL4 hemizygote data sets.Peak subtractions giving values < 0 were set to 0. Positive peak distributions were then log-transformed and the best fit log-normal distribution resolved in Prism.Resolved geometric means were derived from those curves and graphed.

Modeling
A model using the basic constraints set by Mariani et al. was applied. 5Transcription was modeled to occur for a cell activated to transcribe at T = 0 during TF ON until a transcription complex dissociates with the single-phase linear decay TF off T 1/2 = 1.3 h.We applied the estimates for mRNA birth/death of 40/min and the transcript decay T 1/2 of 60-90 min, a translational rate of four IL-4 proteins/mRNA/minute and an IL-4 secretion T 1/2 of 45 0 from Mariani et al. 5 At any given time n, the number of IL-4 protein molecules (#prot.) the average cell will have is given by #prot: ¼ ∑ where mp min is the proportion of IL-4 retained after translation each minute, and mT min the proportion of mRNA transcripts remaining after 1 min.In the model, mRNA will accrue until time TF OFF , at which time transcription ceases and protein output decreases proportionally as a function of the mT min and mp min decay rates.
An expression probability (P EXP ) component was introduced to accommodate varying probabilities of initiating transcription upon stimulation.In the model, once an allele is transcribed, it can shut off transcription (with P = TF OFF ), but cannot initiate transcription in the stimulation round.Over time, the proportion of cells that can transcribe is determined by the fraction that remains untranscribed but has TF ON , while the proportion actively transcribing is the fraction with TF ON that has been triggered to express (at the cell's allelic P EXP ).We set P EXP rates during occupancy (TF ON ) as equivalent for alleles One and Two.P EXP were modulated between 0.05 h À1 and 0.6 h -1 .IL-4 accumulation in cells stochastically achieving transcription during TF ON for allele One at T act will be at the rate of #prot: ¼ ∑

À Á
For the modeling, Microsoft Excel was used to set an array table for each minute.T 1/2 s from Mariani et al., 5 were converted to per minute rates for (i) the fraction of generated mRNA persisting to the next minute as #mRNA*P, where P = 0.992327946 if mRNA T 1/2 = 90 0 and P = 0.98851402 if T 1/2 = 60 0 ; (ii) for IL-4 secretion with a T 1/2 = 45 0 , the fraction of intracellular IL-4 molecules (#IL4) persisting to the next minute as #IL4*P, where P = 0.984714753; (iii) a TF OFF T 1/2 = 1.3 h, then of transcribing alleles the fraction remaining TF ON /min = #transcribing*0.991152866;(iv) the fraction activated to express/minute = #TF ON + untranscribed *P, where P = 0.0035/min if P EXP = 0.21 h À1 , P = 0.000833333 if P EXP = 0.05 h À1 , P = 0.005833333 if P EXP = 0.35 h À1 and P = 0.01 if P EXP = 0.6 h À1 .During TF ON , for transcribing cells mRNA synthesis was set at 40/min/transcribed allele, and translation was set at 4*#mRNA.Equal probabilities for TF OFF and P EXP were assumed for both alleles.The adjustment to make NFAT1 and transcription factor inactivation (TF OFF ) linear was made a priori to simplify modeling, but is consistent with NFAT1 remaining in apposition to the il4 locus at 4 h, 6 while near-complete NFAT1 re-phosphorylation occurs by 5 h 19 implying much of the Il4-localized NFAT1 becomes rephosphorylated by this time.

Figure 1 .
Figure1.A map of Il4 locus and modification of locus in reporter mouse strains used in this work.The native Il4 allele has four exons, while the KN2 and G4 alleles disrupt the first intron (KN2) or its first 178 nucleotides (G4).There is a GATA3 binding motif associated with recall Il4 responses in this intronic region therefore missing in both reporter lines.7

Figure 2 .
Figure 2. In vitro induction and decay kinetics of hCD2 and GFP.(a, b) hCD2 and GFP from the respective KN2 and G4 reporter alleles were monitored in CD4 + splenocytes during Th2 priming in vitro, and proportional expression monitored over time by flow cytometry.During periods of stimulation and after transfer to rest medium, (c) hCD2 and (d) GFP expressions were monitored for 5 days and (e) decay rates compared.hCD2 and GFP decay were fitted by non-linear regression in Graphad Prism with a plateau set at 0.1%.T = 0 values were 47.5% hCD2 + and 22.4% GFP + .Data points show means of (b) n = 3 biological replicate KN2/G4 cultures from one experiment, or (c-e) two biological replicates from one of three similar experiments with n = 2 or 3 biological replicates each.In (e) lines show non-linear regression decay curves.*P < 0.05 by (b) one-way repeated measures ANOVA against the T = 0 expression for both GFP and hCD2 expression profiles or (e) by two-way repeated measures ANOVA against hCD2 decay.

Figure 3 .
Figure 3. Heritable probabilities of expression in differentiated KN2/G4 Th2 cells.Five-day-differentiated KN2/G4 CD4 + Th2 cells were sorted based on GFP and hCD2 expression, rested in IL-2-containing media for 12 days, restimulated for 6 h and evaluated for hCD2 and GFP expression after restimulation.(a) The expression profile of CD4 + CD3 + T cells prior to and after sorting.(b) The 12-day rested, sorted populations were restimulated and after 6 h, hCD2 and GFP expression profiles determined.The reporter expression frequencies among (c) the total locusexpressing cells, and (d) among total CD4 + T cells were quantified.Data shown are from one of two similar experiments.

Figure 4 .
Figure 4. IL-4 output from wild type and Il4 hemizygous reporter strains.Th2 cells were differentiated for 5 days under Th2 conditions, rested for 6 days without IL-4, and restimulated for varying times.(a) Th2 cells from wild type, KN2/+ and G4/+ hemizygous mice, or KN2/G4 mice at 6 h of restimulation were examined for intracellular IL-4 (filled histograms).Dashed lines show cells from each strain rested in medium without restimulation for those 6 h.Inset numbers show the percentage of cells expressing above the top 1% of the wild type mouse expression distribution by standard flow cytometric gating.Left and right italicized numbers show the median fluorescence intensity of unrestimulated and restimulated cells, respectively.(b) The frequency of IL-4 + events in the indicated strains over time.Inset numbers (b) show relative % of wt for the KN2/+ (top number) and G4/+ (lower number) strains.(c) The median expression of the gated top 1% of each strain's IL-4 expression distribution was determined.(d) Depiction of the peak subtraction method used to quantify IL-4 expression amounts among positive expressors relative to wild type IL-4 amounts (e).Data shown are concatenated from n = 6 hemizygotes (n = 3 G4/+ and n = 3 KN2/+ mice) and n = 3 wt biological replicates from one of three similar experiments.Data symbols (b) show means from individual experiments and lines connect grand means from one to three separate experiments for each timepoint with n = 3 biological replicates/strain/experiment.Data symbols (c) show individual biological replicates and lines show mean from one of three similar experiments with n = 3 biological replicates/strain.*P < 0.05 by one-way ANOVA with Tukey's post-test.

Figure 5 .
Figure 5. Independent activation and shutdown of transcription of alleles is predicted to constrain per-cell mRNA amounts and IL-4 outputs.(a) Graphic depiction of the parameters modeled (b-f).(b) A mathematical model based on5 was used to calculate expected fractions of cells transcribing from allele One at all in a 6 h time period, or (c) actively transcribing at an instant of time over that period.This relates to both the P EXP and the TF OFF , with only cells in the TF ON state able to transcribe, and populations initiating transcription according to the P EXP .(d) Based on various P EXP and a TF OFF T 1/2 = 1.3 h (78 min), shown is the fraction of cells simultaneously transcribing from alleles One and Two over time if T ACT and TF OFF are independent for the two alleles.(e) Of cells activated to transcribe both alleles One and Two at T = 0, the fraction continuing to transcribe from both alleles synchronously thereafter was modeled for the instance that both alleles carry a TF OFF T 1/2 = 1.3 h.(f) Based on the frequency of cells transcribing allele One, allele Two, or co-transcribing in wild type mice, the average accumulated transcript abundance in wt cells relative to hemizygous cells was estimated for populations with alleles with the stated P EXP .

Figure 6 .
Figure 6.In vivo expression of hCD2 and GFP in reporter mice.(a) Representative flow cytometry plots of hCD2 and GFP expression in indicated mouse strain CD4 + lymphoid Th cells 7 days after immunization with 100 μg HDM.Fractional reporter expression 3, 7 and 10 days after s.c.HDM immunization of (b) GFP and (c) hCD2 in the indicated strains.(d) Expression MFIs of GFP and (e) hCD2 in the indicated populations.(f) Kinetic profile of hCD2 and GFP expression and allele co-expression (hCD2 + GFP + cell) frequencies in the KN2/G4 expressing cells after D. pteronyssinus house dust mite extract immunization.Bars show arithmetic mean + SD (b, c, f), and symbols show individual mice (d, e).Data shown are pooled (b, c), or from one representative (d-f) experiment from 4 to 6 experiments with n = 3-5 mice per group per timepoint.MFIs across days (d-f) are not directly comparable.Total mouse numbers are shown above the bars (b, c).Data were analyzed by (b, c) one-way ANOVA with Bonferroni's post-test.*P < 0.05.

Figure 7 .
Figure 7. Impact of Th2 strength on in vivo Th cell IL-4 expression.(a) Seven days after s.c.immunization, draining LN CD4 + T cells were counted and examined for (b) GFP expression.(c) CXCR5 + PD1 hi TFH cell frequencies were similarly enumerated and (d) examined for GFP expression.(e) Example of GFP expression in total TFH, and (f) average GFP MFIs among the indicated Th populations are shown.Seven days after s.c.HDM immunization, wild type, G4/+ and G4/+-STAT6 À/À mouse draining LN CD4 + Th cells were evaluated for (g, h) GFP expression frequencies, and (i, j) GFP expression amounts.Data were analyzed by (a-d, g, h) One-way ANOVA with Tukey's post-test or (f, j) two-way repeated measures ANOVA with Bonferroni's post-test.*P < 0.05.
). Fluorophores were detected as specified.Numbers indicate filter and the letter denotes the position of the detector in the flow cube.Ultraviolet laser (355 nm): DAPI 450/50B, Live/Dead fixable blue 450/50B.In some experiments a BD FACSCalibur was used.For these experiments 7-AAD was used as the live/dead stain in place of DAPI.The configuration for these experiments was as follows: Blue laser (488 nm): One of FITC, Alexafluor-488 or GFP on the 530/30 FL1; PE 585/42 FL2; 7-AAD 670LP FL3; APC 661/16 FL4.