miRNA-mediated feedback inhibition of JAK/STAT morphogen signaling establishes a cell fate threshold

Patterns of cell fates generated by morphogens are critically important for normal development, yet the mechanisms by which graded morphogen signals are converted into all-or-none cell fate responses are incompletely understood. In the Drosophila ovary, high and sustained levels of the secreted morphogen Unpaired (Upd) specify the migratory border cell population by activating the Signal Transducer and Activator of Transcription (STAT)^1,2. Lower and transient STAT activity specifies a non-migratory population of follicle cells^3,4. Here we identify miR-279 as a component of a feedback pathway that further dampens the response in cells with low JAK/STAT activity. miR-279 directly repressed STAT, whereas loss of miR-279 mimicked STAT gain-of-function or loss of Apontic (Apt), a known feedback inhibitor of STAT. Apt was essential for miR-279 expression in non-migratory follicle cells whereas another STAT target, Ken and Barbie (Ken), down-regulated miR-279 in border cells. Mathematical modeling and simulations of this regulatory circuit including miR-279, Apt, and Ken supported key roles for miR-279 and Apt in generating threshold responses to the Upd gradient.

Morphogens emanate from a localized source, generate a gradient, and pattern gene expression during development. Target gene responses are not necessarily graded. In fact normal pattern formation frequently requires target gene expression to differ dramatically between neighboring cells, even if there is only a small morphogen concentration difference. Known mechanisms that convert graded signals to all-or-none responses include cooperative binding of Bicoid to the hunchback enhancer⁵, as well as positive transcriptional feedback^{6, 7} and mutual repression of target genes⁸.

In the Drosophila ovary, STAT functions as a morphogen that patterns follicle cell fates³. Fruit fly ovaries are composed of egg chambers, each of which produces an egg. Egg chambers contain 16 germline cells (15 nurse cells and 1 oocyte) enveloped by a monolayer of epithelial follicle cells. At each egg chamber pole, a pair of polar cells develops and secretes Upd, a cytokine. Upd diffuses to form an extracellular gradient⁹ and specifies distinct cell fates at different concentrations^{3, 4}. Border cells differentiate immediately adjacent to the anterior polar cell pair, where the morphogen concentration is highest. STAT activity is indispensable for the specification and migration of border cells^{1, 10}. Although initially Upd activates STAT in a gradient across ~12 cells, only 4–6 cells differentiate as migratory border cells and retain high levels of STAT activity, due to a negative feedback circuit that includes the Apontic (Apt) protein⁴. In contrast to graded STAT activity, Apt is expressed relatively uniformly in anterior follicle cells. Those cells in which STAT activity exceeds Apt maintain high STAT, differentiate as border cells, invade the neighboring nurse cells and migrate, carrying the polar cells with them⁴. Cells in which Apt inhibition exceeds STAT activation, differentiate as squamous follicle cells and remain within the epithelium⁴. It is unclear however by what molecular mechanism Apt antagonizes STAT.

We investigated the possibility that one or more microRNAs (miRNAs) might function in patterning follicle cell fates and STAT activity. MiRNAs are non-coding 22–24 nucleotide RNAs that repress gene expression post-transcriptionally by partial pairing with the 3’ UTR of specific mRNAs¹¹. MiRNAs can fine-tune target gene expression levels¹². To test whether a miRNA might modulate morphogen gradient responses, we searched for miRNAs predicted to bind the 3’UTRs of core genes in the JAK/STAT pathway. We used the following target prediction programs - miRanda¹³, PicTar¹⁴, and TargetScan¹⁵, as well as a database¹⁶. Of the four components – Upd, Domeless (Dome), Hopscotch (hop), and STAT92E examined, only STAT92E and UPD 3’UTRs contain putative miRNA binding sites. The miRNAs predicted to bind the STAT 3’UTR were miR-279, miR-277, miR-280; miR-284, miR-92a; in addition, the Upd 3’ UTR contained one predicted miR-279 binding site. The candidate miRNAs were then overexpressed in S2 cells. Only miR-279 repressed expression of a reporter gene fused to the STAT 3’UTR (Fig. 1a). Overexpression of miR-279 did not repress expression of a mutant STAT 3’UTR reporter gene lacking the miR-279 seed-binding site (Fig. 1b, c). Furthermore, knock-down of endogenous miR-279 using a 2’-O-methyl miR-279 antagomir increased STAT 3’UTR reporter activity in S2 cells, whereas control antagomirs did not (Fig. 1d). Thus, miR-279 directly targeted STAT via its 3’UTR. Although the Upd 3’UTR contained one putative miR-279 site, the Upd 3’UTR reporter did not respond to miR-279 overexpression (Fig. 1e). Thus, miR-279 targets the JAK/STAT signaling pathway by repressing STAT.

To determine whether miR-279 was expressed in the Drosophila egg chamber, we examined transgenic flies containing a transcriptional reporter miR-279-Gal4 and UAS-GFP¹⁷. We detected expression in most follicle cells beginning at stage 8 (Fig. 1f–i). However, miR-279 expression was undetectable in polar cells (Fig. 1g, i, k, m) and was somewhat lower and more variable in border cells than in non-migratory anterior follicle cells (Fig. 1f–u).

If miR-279 normally inhibited JAK/STAT signaling, loss-of-function of miR-279 might cause phenotypes similar to gain-of-function of STAT. Consistent with this hypothesis, extra cells invaded in between the nurse cells in egg chambers containing miR-279 mutant clones (Fig. 2b), compared to controls (Fig. 2a), a phenotype also observed following ectopic activation of STAT^1,4,10. We then expressed a miR-279 “sponge” (an RNA containing 3 copies of the miR-279 binding site) using miR-279-Gal4, as an alternative method to reduce miR-279 function^{18, 19}. This treatment also resulted in ectopic invasive cells (Fig. 2c). Ectopic cells or clusters were observed in approximately 50% of egg chambers following the induction of clones using three different miR-279 alleles, a phenotype that was ameliorated by the addition of a wild-type miR-279 transgene to the genetic background (Fig. 2g). Ectopic cells or clusters were also observed in 30% of egg chambers expressing two copies of the miR-279 sponge (Fig. 2h). To determine whether the abnormal cells came from dissociation of the original border cell cluster or from abnormal invasion of follicle cells that normally remain within the epithelium, we counted the number of cells in the main cell cluster, which contains the polar cells. In all miR-279 mutant clones, we found that the number of border cells within the main cluster was normal (Fig. 2i).

The level of STAT protein expression in border cells is critical. In fact, in stat/+ heterozygous females border cells fail to complete their migration by stage 10 in 10% of egg chambers^1,10. This is highly unusual as most mutations that affect border cells are fully recessive. Over-expression of STAT also impaired border cell migration (Fig. 2d, j), as does STAT hyperactivation¹. Similarly, mutation of miR-279 caused frequent border cell migration defects (Fig. 2e), in contrast to wild-type (Fig. 2a). Approximately 50% of border cells mutant for any one of the three alleles, failed to migrate normally, a phenotype that was rescued by the wild-type miR-279 transgene (Fig. 2k). Knock down of miR-279 function by expressing the miR-279 sponge in border cells using slbo-Gal4 similarly resulted in migration defects (Fig. 2f, l). Together these results show that loss of miR-279 mimics both phenotypes associated with STAT gain-of- function^1,4,10.

To confirm that STAT is a target of miR-279 in vivo, we assessed nuclear STAT protein levels using an antibody. STAT was enriched ~1.6 fold in miR-279 mutant border cells compared to adjacent wild-type cells in mosaic clusters (Fig. 3a; Supplementary information, Fig. S1). In addition, we measured STAT activity using a reporter construct containing 10 STAT binding sites upstream of GFP (10XSTAT92E-GFP)²⁰. In wild-type stage 10 egg chambers, STAT activity is predominantly localized in border cells (Fig. 3b, c; Supplementary information, Movie S1). Reducing miR-279 function using the miR-279-Gal4 driver to express the miR-279 sponge, dramatically elevated STAT activity in other follicle cells (Fig. 3d, e; Supplementary information, Movie S2), consistent with the relatively high level of miR-279 Gal4 expression in those cells. Taken together, these results demonstrated that miR-279 repressed STAT protein expression and activity in vivo.

To determine the functional significance of STAT as a miR-279 target, we tested for genetic interactions. If the miR-279 phenotypes were primarily caused by excess STAT expression, reducing the level of STAT might ameliorate them. Remarkably, reducing STAT expression using a homozygous hypomorphic, P-element insertion allele (stat^ep3391) rescued the ectopic invasive cell phenotype (Fig. 3f) and the border cell migration defect in miR-279 mutant clones (Fig. 3g) to a similar extent as the wild-type miR-279 transgene. Reducing STAT expression with stat^ep3391/+ suppressed the miR-279 sponge phenotype (Fig. 3h). Overexpression of miR-279 also rescued border cell migration defects caused by overexpression of a STAT cDNA, which contained both the protein coding sequence and the normal 3‘UTR (Fig. 3i). Whereas only 30% of border cells over-expressing STAT completed migration normally, nearly 80% of border cells over-expressing STAT together with miR-279 completed migration. In contrast, over-expression of miR-279 did not rescue the phenotype caused by over-expression of STAT with a mutated 3’UTR in which the miR-279 seed sequence was deleted. In the presence or absence of miR-279, approximately 40% of clusters completed migration normally (Fig. 3i). Taken together, these results indicate that STAT is a functional target of miR-279 in vivo.

A genetic regulatory circuit consisting of STAT, Apt, and Slow Border Cells (SLBO) was previously shown to convert the initially graded Upd/JAK/STAT signal into “on-off” states of STAT activation^4,21. Therefore, we decided to investigate how miR-279 was related to the other components of the circuit. Apt functions as a feedback repressor of STAT whereas SLBO amplifies STAT activity by antagonizing Apt function^4,21. However, the mechanism by which Apt negatively regulates STAT is unknown.

Since miR-279 causes phenotypes very similar to apt, and Apt is a nuclear protein in follicle cells, we tested whether Apt affects miR-279 expression. In striking contrast to wild-type (Fig. 4a, b), miR-279 expression was undetectable in apt mutant egg chambers (apt¹⁶⁷/ apt^KG05830) (Fig. 4c, d). We confirmed this result using a miR-279 activity sensor, which expresses GFP under the control of the tubulin promoter and contains six miR-279 binding sites in the 3’UTR, rendering GFP expression sensitive to miR-279. Consistent with loss of miR-279 expression in apt mutant egg chambers, expression of the sensor was elevated in (apt¹⁶⁷/ apt^KG05830) compared to the wild-type control (Supplementary information, Fig. S2a). A control “sensor” which lacked the critical miR-279 seed sequence showed no difference between wild-type and the apt mutant. (Supplementary information, Fig. S2b). Therefore, Apt is essential for miR-279 expression and activity in follicle cells.

Genetic interactions between apt and miR-279 were consistent with the observation that miR-279 expression was undetectable in apt mutant egg chambers. Knock-down of miR-279 enhanced the extra invasive cell phenotype in females heterozygous for an apt null allele (apt¹⁶⁷/+) (Fig. 4e), but did not enhance the extra invasive cell phenotype in homozygous mutants (apt¹⁶⁷/ apt^KG05830) (Fig. 4e). 75–80% of apt¹⁶⁷/ apt^KG05830 mutant egg chambers contain extra invasive cells⁴ (Fig. 4e) compared to 45–50% of miR-279 mutants (Fig. 2g), and STAT expression is elevated three-fold in apt mutant border cells⁴ compared to 1.6-fold for miR-279 (Fig. 3a). Therefore, although there are likely additional Apt targets, miR-279 is an important one in repressing STAT activity and anterior follicle cell invasion.

Apt is expressed in most follicle cells up to stage 8⁴. During stage 9, Apt is expressed across the anterior field of follicle cells in a broad and shallow gradient (Fig. 4g). However unlike Apt, miR-279 is repressed in cells immediately adjacent to the polar cells, which become border cells (Fig. 4a, b). Therefore we wondered which gene(s) might repress miR-279 in border cells. Ken & Barbie (Ken) is the Drosophila homologue of human BCL6, a BTB/POZ domain-containing transcriptional repressor²². Ken expression is border cell-enriched based on a microarray analysis and in situ hybridization²³, which we confirmed using a ken enhancer trap insertion^{24, 25} (Fig. 4h, i). To test the effect of Ken on miR-279 expression, we crossed the miR-279 reporter into the ken^k11035/ ken¹ mutant background. In wild-type egg chambers, miR-279-Gal4 expression in border cells was about half of that of anterior non-migratory border cells, whereas in ken^k11035/ ken¹ mutants, the ratio was close to 1 (Fig. 4j). Over-expression of Ken also decreased mir-279 promoter activity in S2 cells (Supplementary information, Fig. S3). In a microarray analysis, Ken mRNA was significantly upregulated in response to over-expression of Upd in follicle cells (Wang, X. and Montell, D. J., unpublished data), which is also true in eye imaginal discs²⁶. Ken expression is also highest in border cells (Fig. 4h, i), where STAT activity is highest, suggesting ken could be a STAT target. Consistent with this inference, in egg chambers mutant for a temperature-sensitive stat allele (stat³⁹⁷/stat^ts) incubated at the restrictive temperature, Ken expression was reduced in border cells (but not polar cells which have little or no active STAT) compared to control (Fig. 4k–o). Therefore in border cells, ken is a target of STAT.

Together these data suggest a model for the conversion of the graded Upd signal into migratory border cell and stationary epithelial cell fates (Fig. 5a,b). We previously developed a mathematical model and used simulations of the patterning process to show that a gene regulatory circuit consisting of STAT, Apt, and Slbo was sufficient⁴. However, it was unclear how Apt exerted its effect on STAT. The results presented here suggest that miR-279 is a major target of Apt that directly targets STAT (Fig. 5a, b). We used a set of differential equations to approximate the relative concentration of each component across the field of the epithelium in computer simulations (see methods). Multiple iterations of these computations led to evolving patterns of gene expression, which accurately reproduced the observed patterns of miR-279, STAT, Apt, Ken, and Slbo over time (Fig. 5c–f). In the simulations loss of miR-279, like loss of apt, resulted in failure of the proper pattern of JAK/STAT activity to form (Fig. 5g, h). We note that the deterministic (rather than probabilistic) nature of the model does not permit simulation of incomplete penetrance and thus does not reproduce the difference in penetrance between these two (80% for apt vs. 50% for miR-279).

Interestingly, in simulations, loss of ken still permitted development of “on” and “off” states (Fig. 5i). Experimentally, we also found that loss of ken in mosaic clone analyses did not impede border cell migration, and forced expression of UAS-miR-279 with slbo-Gal4>UAS-miR-279 was similarly benign (not shown). The simulations show that the mutual repression between Slbo and Apt is sufficient to ensure relatively robust patterning, even in the absence of ken. The modeling and genetic results both show that this regulatory circuit is more sensitive to loss than gain of miR-279.

Complex genetic networks are required for a variety of biological phenomena, however their complexity often challenges our intuitive understanding. Mathematical modeling can be useful for probing the behavior of such complex systems²⁷. Here the mathematical model revealed non-obvious features of the morphogen interpretation system. One interesting feature of this system is that non-linear positive auto-regulation of the JAK/STAT pathway is not responsible for creating the threshold response of target gene expression. Even though such auto-regulation can produce threshold responses, and auto-regulation is a property of JAK/STAT signaling, by itself it does not produce the initially graded pattern observed here. The multiple positive and negative feedback loops shown here can produce a more complex response, including a transient response to a low level of morphogen, which is not characteristic of a response dominated by non-linear auto-regulation.

In this work we describe a key role for miR-279 in shaping threshold responses to the Upd morphogen gradient. Apt-mediated expression of miR-279 is critical to repress STAT in anterior follicle cells destined to remain within the epithelium. Since miR-279 provides post-transcriptional control of STAT, it can repress the function of pre-existing mRNA and in principle lead to a swifter and more decisive cellular response than feedback at the level of transcription alone. Thus miRNA-mediated feedback may be particularly important in tissues undergoing rapid development. In addition, this mechanism creates not only a lower level response but also a transient response. In some morphogen systems, the length of time that a cell experiences the signal is important for generating correct cell fates²⁷. Border cells represent such a system. Border cells that express a temperature-sensitive form of STAT and are shifted to the non-permissive temperature part way through their migration, turn on expression of at least one anterior follicle cell marker after 2.5 hours. Therefore a transient STAT signal may specify anterior cell fate even if it first reaches a level high enough to promote border cell fate and migration¹⁰. miR-279 is part of the mechanism that terminates STAT signaling and thus specifies anterior follicle cell fate. Another tissue in which sustained versus transient STAT signaling is important is in stem cell fate specification in the Drosophila testis. It will be of interest to determine if miR-279 also contributes to patterning fates in this context as well.