Abstract
Concentration of transcription factors (TFs) and their cell-to-cell protein variability are important functional determinants in development, yet how variability is controlled remains poorly understood. Using Fluorescence Correlation Spectroscopy (FCS), we characterized in live Drosophila imaginal discs the concentration and cell-to-cell variability of 14 endogenously tagged TFs. We found that the Hox TF Antennapedia (Antp) transitioned from a low concentration/high variability state early in development to a high concentration/low variability state later in development. Using FCS and temporally resolved genetic studies, we uncovered that Antp is necessary and sufficient to drive a developmental regulatory switch from auto-activation to auto-repression, thereby reducing variability. This switch is controlled by a progressive change in relative concentrations of preferentially activating and repressing Antp protein isoforms, which bind to chromatin with different affinities. We derived a simple mathematical model, confirming that the Antp auto-regulatory circuit would suffice to increase protein concentration while suppressing variability over time.
Introduction
In order to understand the mechanisms that control pattern formation and cell fate specification in developing organisms, the intranuclear concentration, DNA-binding kinetics and cell-to-cell variability of relevant TFs need to be quantitatively characterized. TF concentration variability at the tissue level is thought to arise from diverse processes, including mRNA transcription, protein production and degradation. For example, gene transcription in a given tissue is a noisy process. The noise is due to stochastic binding and interactions of proteins involved in transcriptional activation of the specific gene (intrinsic noise) (Blake et al., 2003; Elowitz et al., 2002) and also due to differences among cells in terms of abundance of the transcriptional and post-transcriptional cellular machinery which affects the efficiency of transcriptional activation in general (extrinsic noise) (Swain et al., 2002). These can influence the expression and production of functional protein, resulting in protein concentration that exhibits variability among cells.
In undifferentiated tissue or cells, TF cell-to-cell variability can be the driving force for differentiation. For example, progressive establishment of a Nanog salt-and-pepper expression pattern leads to the formation of primitive endoderm in the mouse preimplantation embryo, whereas loss of the variability results in embryos lacking primitive endoderm entirely (Kang et al., 2013). In Drosophila, the Senseless (Sens) TF is required for the establishment of proper number of sensory organ precursors in the ectodermal proneural cluster of cells and unequal concentration among cells is required for their specification (Li et al., 2006). Moreover, variability in concentration (rather than its overall average concentration) of the Yan TF drives the transition of developing photoreceptor cells to a differentiated state during Drosophila eye development (Pelaez et al., 2015).
Conversely, in already differentiated tissue or cells, TF expression variability among cells may need to be counteracted to ensure homogeneity of gene expression patterns and robustness of commitment to a certain transcriptional regime. Such homogenization of expression levels has been identified for the Snail (Sna) TF, which is required for the invagination of the mesoderm during Drosophila gastrulation (Boettiger and Levine, 2013), or the Bicoid (Bcd) and Hunchback (Hb) TFs during early embryogenesis (Gregor et al., 2007a; Gregor et al., 2007b; Little et al., 2013). These studies have quantified the tolerable degrees of concentration variability that allow establishment of gene expression territories with remarkable precision in the developing embryo.
In addition, differential fates within the same developmental territory may be specified by TFs deploying different DNA-binding dynamics despite the existence of very similar concentrations (i.e. low variability). For example, studies on the Oct4 TF in early mouse embryos have shown that differential kinetic behavior of DNA binding, despite equal Oct4 concentration among blastomeres, ultimately dictates an early developmental bias towards lineage segregation (Kaur et al., 2013; Plachta et al., 2011).
So far, gene expression variability studies have focused predominantly on monitoring the noise of mRNA production (Holloway et al., 2011; Holloway and Spirov, 2015; Little et al., 2013; Lucas et al., 2013; Pare et al., 2009). Little information exists about TF variability at the protein level within a tissue, since studies of this sort would require the use of quantitative methods with single-molecule sensitivity.
We have previously used Fluorescence Microscopy Imaging and FCS, to study Hox TF interactions with nuclear DNA in living salivary gland cells (Papadopoulos et al., 2015; Vukojevic et al., 2010). FCS has also been instrumental for the quantification of TF dynamics in living cells or tissue in several recent studies (Clark et al., 2016; Kaur et al., 2013; Lam et al., 2012; Mistri et al., 2015; Papadopoulos et al., 2015; Perez-Camps et al., 2016; Szaloki et al., 2015; Tiwari et al., 2013; Tsutsumi et al., 2016). Yet, in these studies, protein mobility has been measured in overexpressing systems. However, to understand TF behavior in vivo, proteins need to be quantified at endogenous levels (Lo et al., 2015).
In this study, we take advantage of the availability of fly toolkits, in which TFs have been endogenously tagged by different methodologies: fosmid, BAC, FlyTrap and MiMIC lines (Buszczak et al., 2007; Ejsmont et al., 2011; Ejsmont et al., 2009; Kelso et al., 2004; Morin et al., 2001; Quinones-Coello et al., 2007; Sarov et al., 2016; Venken et al., 2011), to measure the intranuclear concentration of various TFs in vivo by FCS, and their cell-to-cell variability in several fly imaginal discs. The imaginal discs are flat, single-layered epithelia comprised of small diploid cells and many TFs are expressed in defined regions within these tissues during development. In this system, we found that Antp, a well-characterized homeotic selector gene, responsible for specification of thoracic identity in fly tissues, displayed high cell-to-cell variability during early disc development and this variability was suppressed at later developmental stages. Through a combination of genetics, single-molecule measurements of TF dynamics by FCS and computational modeling, we uncovered a mechanism that controls Antp concentration and variability in developing discs.
Results
Characterization of average protein concentrations and cell-to-cell variability of Drosophila TFs
Average concentrations of TFs in neighboring nuclei of third instar imaginal discs were measured by FCS (Figure 1 A-J and Figure 1 – figure supplement 1 A-P). FCS is a non-invasive method with single molecule sensitivity, in which a confocal arrangement of optical elements is used to generate a small (sub-femtoliter) volume inside living cells, from which fluorescence is being detected (Figure 1 C and D, green ellipsoid). Fluorescent molecules diffuse through this observation volume, yielding fluorescence intensity fluctuations that are recorded over time by detectors with single-photon sensitivity (Figure 1 E). These fluctuations are subsequently subjected to temporal autocorrelation analysis, yielding temporal autocorrelation curves (henceforth referred to as FCS curves, Figure 1 F), which are then fitted with selected models to extract quantitative information about the dynamic processes underlying the generation of the recorded fluctuations. In the case of molecular movement of TFs (Supplement 1), information can be obtained regarding: a) the absolute concentrations of TFs (Figure 1 F), (b) TF dynamic properties, such as: diffusion times, differences in their interactions with chromatin and fractions of free-diffusing versus chromatin-bound TFs (Figure 1 G); and c) TF cell-to-cell concentration variability (Figure 1 H).
For the 14 selected TFs, we measured average concentrations ranging about two orders of magnitude among different TFs, from ∼30 nM to ∼1.1 μM (Figure 1 I and Figure 1 – figure supplement 1 A-Q). We also obtained various diffusion times and fractions of slow and fast diffusing TF molecules (Figure 1 J), indicating differential mobility and degree of DNA-binding among different TFs (Vukojevic et al., 2010). Comparison of the y-axis amplitudes at the smallest lag time of the FCS curves (the points at which the FCS curves cross the y-axis), which are inversely proportional to the concentration of fluorescent molecules (Figure 1 F), gives information about concentration variability (heterogeneity) among different cell nuclei, i.e. reflects heterogeneity of protein concentration at the tissue level (Figure 1 H). We measured the variability of all 14 TFs in our dataset (expressed as the variance over the mean squared, to be in the range 7 − 37% (Figure 1 K and Figure 1 – figure supplement 1 Q). These numbers are consistent with previous observations (Sanchez et al., 2011). We then used this dataset as a starting point for studying the control of variability during imaginal disc development.
Since low variability at the tissue level is likely to be achieved through some active mechanism that controls it, we searched for TFs that exhibited low variability and relatively high concentrations. One TF, the Hox gene Antp, had comparatively lower variability (CV2 < 0.2) for its high average concentrations (Figure 1 K), in particular in the leg disc. This distinction prompted us to measure variability of Antp at different concentrations by examining clusters of neighboring cells from across the disc displaying different average expression levels (Figure 1 L). Because FCS performs best at low to moderate expression levels, we performed this analysis in the wing disc (Figure 1 L). We established that the observed fluorescence intensity fluctuations were caused by diffusion of TF molecules through the illuminated confocal volume (Figure 1 – figure supplement 2 and Supplement 1). Our data showed that Antp cell-to-cell variability decreased with increasing Antp concentration (Figure 1 M), suggesting complex transcriptional regulatory processes (Franz et al., 2011; Smolander et al., 2011) that we further investigated using the powerful Drosophila genetic toolkit.
Control of Antp concentration by transcriptional auto-regulation
One mechanism by which genes control their expression level variability is auto-regulation (Becskei and Serrano, 2000; Dublanche et al., 2006; Gronlund et al., 2013; Nevozhay et al., 2009; Shimoga et al., 2013; Thattai and van Oudenaarden, 2001). To test whether Antp can regulate its own protein levels, we monitored the concentration of endogenous Antp protein upon overexpression of Antp from a transgene. To distinguish between overexpressed and endogenous protein, we used synthetic Antp (SynthAntp) transgenes fused to eGFP (SynthAntp-eGFP). These transgenes encode the Antp homeodomain, the conserved YPWM motif and the C terminus (but lack the long and non-conserved N terminus of the protein, against which Antp antibodies have been raised) and they harbor Antp-specific homeotic function (Papadopoulos et al., 2011). Clonal overexpression of SynthAntp-eGFP in the wing disc notum (Figure 2 A-B’ and controls in Figure 2 – figure supplement 1 D and D’) repressed the endogenous Antp protein, indicating that Antp is indeed able to regulate its own protein levels.
Since Antp is a TF, we next asked whether the auto-repression indeed occurs at the transcriptional level. The Antp locus is subject to complex transcriptional regulation, involving a distal and a proximal promoter (P1 and P2 promoters, respectively), spanning more than 100 kb of regulatory sequences. We established that the P1 promoter (rather than the P2 promoter) is predominantly required to drive expression of Antp in the wing disc notum (Figure 2 – figure supplement 1 A-C’), in line with previous observations ((Engstrom et al., 1992; Jorgensen and Garber, 1987; Zink et al., 1991) and Materials and Methods). Moreover, mitotic recombination experiments in regions of the wing disc unique to P2 transcription have shown no function of the P2 promoter transcripts in wing disc development (Abbott and Kaufman, 1986). Thus, the P1 Antp reporter serves as a suitable reporter of the Antp locus transcriptional activity in this context.
Clonal overexpression of SynthAntp-eGFP in the wing disc repressed the Antp P1 transcriptional reporter (Figure 2 C and C’ and controls in Figure 2 – figure supplement 1 E and E’). To rule out putative dominant negative activity of the small synthetic Antp-eGFP peptide, we also performed these experiments with the full-length Antp protein (Figure 2 – figure supplement 1 F and F’). We conclude that the Antp protein is able to repress its own transcription from the P1 promoter, suggesting a possible mechanism of suppressing cell-to-cell variability of Antp expression levels (Figure 2 D).
In the course of these experiments, we noticed that ectopic overexpression of SynthAntp-eGFP or the full-length Antp protein from the Distal-less (Dll) (MD23) enhancer surprisingly resulted in activation of the Antp P1 reporter in distal compartments of the wing disc, such as the wing pouch, where Antp is normally not detected (Figure 2 E-F’ and controls in figure 2 – figure supplement 1 G-H’). This finding suggests that next to its auto-repressing function, Antp is also capable to activate its own transcription (Figure 2 G).
To exclude that the auto-activation and repression of Antp are artifacts of overexpression, we measured by FCS the concentration of Antp triggered by different Gal4-drivers and found it to directly correlate with the degree of its homeotic transformation capacity (Figure 2 – figure supplement 2 A-I). Importantly, we observed indistinguishable DNA-binding behavior not only across the whole concentration range examined (Figure 2 – figure supplement 2 J), but also between endogenous and overexpressed Antp (Figure 2 – figure supplement 3 A-B). Importantly, the auto-activating and auto-repressing capacity of Antp was preserved even with the weak Gal4-driver 69B (Figure 2 – figure supplement 2 K-P) that triggered concentrations of Antp slightly lower than its normal concentration in the wing disc (473 nM versus 501 nM), indicating that auto-activation and auto-repression of Antp take place at endogenous concentrations.
We conclude that, Antp is able to repress and activate its own transcription (Figure 2 D and G) and hypothesize that this auto-regulatory circuit somehow sets the correct concentration of Antp protein in imaginal discs.
A temporal switch controls the transition of Antp from a state of auto-activation to a state of auto-repression
To further investigate the mechanism by which the Antp auto-regulatory circuit sets the Antp expression levels precisely, we next asked whether the seemingly opposing auto-regulatory activities of Antp are separated in time during development. To that end, we induced gain-of-function clones of full-length untagged Antp either at 26 h (first larval instar – henceforth referred to as “early” stage) or at 60 h (late second larval instar – henceforth referred to as “late” stage) of development and analyzed the clones in late third instar wing imaginal discs (Figure 3). As a pre-requisite for this analysis, we established that the Antp-eGFP homozygous viable MiMIC allele recapitulates the endogenous Antp pattern in the embryo and all thoracic imaginal discs and therefore can be used to monitor endogenous Antp protein (Figure 3 – figure supplement 1). Clonal induction of full-length untagged Antp in early development triggered strong auto-activation of Antp-eGFP (Figure 3 A, B and B’ and controls in Figure 3 – figure supplement 2 A-C’). As before, we confirmed that early auto-activation of Antp is transcriptional and similar for both full-length and SynthAntp proteins (Figure 3 – figure supplement 2 D-E’ and controls in F-G’). Early auto-activation was further supported by a loss-of-function experiment, where RNAi-mediated early knockdown of Antp resulted in downregulation of the Antp reporter (Figure 3 C and C’ and controls in Figure 3 – figure supplement 2 H and H’). The loss and gain-of-function analysis together suggest that during early disc development Antp is required for sustaining its own expression.
In contrast, clonal induction during the late second instar stage (Figure 3 F) repressed Antp-eGFP (Figure 3 G and G’) and, reciprocally, the clonal knockdown by RNAi triggered auto-activation of Antp transcription (Figure 3 H and H’). Hence, in contrast to early development, Antp represses its own expression in third instar discs.
While the gain-of-function experiments show that Antp is sufficient to execute auto-regulation, loss-of-function analysis indicates that it is also necessary for both repression and activation at the transcriptional level.
Together, these results revealed the existence of a switch in Antp auto-regulatory capacity on its own transcription during development. Starting from a preferentially auto-activating state early in development (Figure 3 D), it changes into an auto-inhibitory mode at later developmental stages (Figure 3 I).
Antp switches from a low-concentration/high-variability to a high-concentration/low-variability state
If the Antp auto-repressive state limits the variability of Antp protein concentration among neighboring cells late in development, we expected that the variability would be higher during earlier stages, when auto-repression does not operate. We, therefore, monitored the endogenous expression levels and cell-to-cell variability of Antp nuclear concentration in second instar wing and leg discs by FCS. We observed significantly lower average concentrations of Antp protein in second versus third instar wing and leg discs and the inverse was true for concentration variability (Figure 3 E and Figure 3 – figure supplement 3 A, A’ and C), indicating that the developmental increase in concentration is accompanied by suppression of concentration variability. In addition, FCS analysis revealed a notable change in Antp characteristic decay times (signifying molecular diffusion, limited by chromatin-binding) at early versus late stages (Figure 3 E and Figure 3 – figure supplement 3 B). This behavior indicates that endogenous Antp is initially fast moving in the nucleus and undergoes considerably fewer interactions with chromatin, compared to later stages where its interactions with chromatin are more frequent and longer lasting.
Taken together, our FCS measurements show that Antp is expressed at relatively low and highly variable levels in early developing discs, when genetic evidence indicates auto-activation capacity on its own transcription. Later in development, when Antp has reached a state of higher average concentrations, auto-repression kicks in, resulting in considerably lower variability among neighboring cells.
Dynamic control of Antp auto-regulation by different Antp isoforms
The changing binding behavior of Antp on chromatin from second to third instar discs and the developmental transition from an auto-activating to an auto-repressing state suggested a causal relationship between the two phenomena. We, therefore, sought to identify molecular mechanisms that could link the observed changes in Antp chromatin-binding to Antp auto-activation and repression. It is well established that the Antp mRNA contains an alternative splice site in exon 7 immediately upstream of the homeobox-containing exon 8, and generates Antp isoforms differing in as little as 4 amino acids in the linker between the YPWM motif (a cofactor-interacting motif) and the homeodomain (Figure 4 A) (Stroeher et al., 1988). Our previous observation that long linker isoforms favor transcriptional activation of Antp target genes, whereas short linker ones favor repression of Antp targets (Papadopoulos et al., 2011), prompted us to examine whether the linker length is also responsible for differences in auto-regulation.
Ectopic expression of SynthAntp-eGFP peptides featuring a long linker displayed significantly weaker repression capacity on endogenous Antp, as compared to their short linker counterparts (Figure 4 B, B’, F and F’ and quantified in D and H, see also Materials and Methods). We confirmed that, also in this case, the repression was at the transcriptional level (Figure 4 – figure supplement 1 I-J’). Inversely, long linker Antp isoforms exhibited stronger activation of Antp reporter, as compared to short linker ones (Figure 4 C, C’, G, G’ and quantified in D and H, see also Materials and Methods). We, additionally, validated that short linker isoforms encoded by full-length or SynthAntp cDNAs behaved as weaker auto-activating and stronger auto-repressing Antp species in all our previous experiments of endogenous Antp protein and P1 reporter (Figure 4 – figure supplement 1 A-H’). We conclude that, also in the case of Antp auto-regulation, short linker isoforms function as more potent repressors, whereas long linker ones operate as more potent activators.
Since the Antp P1 promoter unit changes its configuration from preferential auto-activation to auto-repression, and short and long linker Antp isoforms function as preferential auto-repressors and auto-activators, it appeared possible that the switch in Antp regulation is executed at the level of transcript variant abundance of these isoforms. Therefore, we next quantified the relative abundance of long and short linker transcript variants in the embryo, second and third instar discs (Figure 4 D and H). We found that the concentration of the long linker variant decreased, whereas the concentration of the short linker variant increased over time in development, in line with previous observations (Stroeher et al., 1988). As hypothesized, this finding suggested that relative transcript variant abundance may underlie the switch between auto-activation and auto-repression (without excluding additional mechanisms).
Relative changes in Antp transcript variant concentration (Figure 4 D and H), differential efficiency of their encoding isoforms to repress or activate the Antp gene (Figure 4 B-D and F-H), the developmental switch of the Antp gene from auto-activation to repression (Figure 3) and the different mobilities of Antp between second and third instar imaginal discs (Figure 3 E) all pointed towards the hypothesis that the two isoforms have different properties in their modes of interaction with chromatin. To investigate this, we expressed the two isoforms in third instar wing and antennal discs from the 69B enhancer, which we established to result in Antp concentrations close to (if not below) endogenous levels (Figure 2 – figure supplement 2 A-J). FCS measurements revealed that the short linker isoform displayed longer characteristic decay times and higher fraction of DNA-bound molecules, suggesting stronger and more pronounced binding to chromatin than its long linker counterpart (Figure 4 D and H and Figure 4 – figure supplement 2 A-B). With chromatin (and therefore Antp binding sites configuration) being identical between the two instances (short and long linker isoforms examined in third instar wing and antennal imaginal discs of the same age), we were able to directly compare the apparent equilibrium dissociation constants for the two isoforms (Supplement 3). We found that the affinity of binding to chromatin (Kd−1) of the repressing short linker isoform is at least 2.3 times higher compared to the activating long linker isoform (Figure 4 D and H and Figure 4 – figure supplement 2 C-D’). We, additionally, validated the different affinities of short and long isoforms by gel-shift experiments using different Antp binding sequences and obtained from two- to eightfold higher affinity of the short linker isoform compared to the long linker isoform in binding of previously characterized Antp and homeodomain binding sites (Figure 4 D and H and Figure 4 – figure supplement 3). Collectively, our experiments support the notion that differences in Antp regulation during disc development can be largely attributed to differences in the affinity of the investigated Antp isoforms.
Taken together, the switch of Antp from an auto-activating to an auto-repressing state and the alteration of its DNA-binding behavior during disc development can be largely explained by a temporal developmental regulation of the relative concentrations of preferentially auto-activating and auto-repressing Antp protein isoforms, which themselves display distinct properties in their modes of interaction with chromatin (Figure 4 E and I).
Robustness of Antp auto-regulation
In order to further substantiate the qualitative model of Antp auto-regulation suggested by our experimental findings, we developed a mathematical model of stochastic Antp expression. This model tests whether the identified interplay between positive and negative auto-regulation of Antp through distinct isoforms is sufficient to explain the increase in protein concentration and decrease in nucleus-to-nucleus variability from early to late stages. The model consists of a dynamic promoter, which drives transcription of Antp followed by a splicing step, leading to the expression of either the auto-repressing or the auto-activating isoform of Antp. In line with our finding that the repressing isoform has higher concentration at later stages, we assumed that splicing is more likely to generate this isoform than the activating isoform. The initial imbalance of Antp towards the activating isoform (Figure 4 D and H) is modeled through appropriate initial concentrations of each isoform.
Since Antp copy numbers per nucleus are in the thousands at both early and late stages, intrinsic noise of gene expression is likely to explain only a certain portion of the overall variability in Antp concentrations (Elowitz et al., 2002; Taniguchi et al., 2010). The remaining part (termed extrinsic variability) is due to cell-to-cell differences in certain factors affecting gene expression such as the ribosomal or ATP abundances. To check whether extrinsic variability significantly affects Antp expression, we expressed nuclear RFP constitutively, alongside with endogenous Antp and measured the abundances of green-labeled Antp and RFP. Since extrinsic factors are expected to affect both genes in a similar way, one should observe a correlation between the concentration of nuclear RFP and Antp-eGFP. Our data showed a statistically significant correlation between RFP and Antp (Figure 5 – figure supplement 1, r = 0.524 and p = 9.77 • 10−5). Correspondingly, we accounted for extrinsic variability also in our model by allowing gene expression rates to randomly vary between cells (Zechner et al., 2012).
The promoter itself is modeled as a Markov chain with three distinct transcriptional states. In the absence of Antp, the promoter is inactive and transcription cannot take place (state “U” in Figure 5 A). From there, the promoter can switch into a highly expressing state “A” at a rate that is assumed to be proportional to the concentration of the long-linker, auto-activating isoform. This resembles the positive auto-regulatory function of Antp. Conversely, the promoter can be repressed by recruitment of the short-linker, auto-repressing isoform, corresponding to state “R” in the model (Figure 5 A). To account for potential leakiness of the promoter, this rate is not assumed to be zero, but significantly lower than that of state “A”. Since the auto-repressing isoform of Antp can also activate the promoter, albeit significantly weaker than the auto-activating isoform, and vice versa, we allow the promoter to switch between states “A” and “R”.
While this promoter model resembles the dual-feedback structure of Antp locus inferred from experiments, it is unclear whether the two isoforms compete for the same binding sites on the P1 promoter or if auto-repression can take place regardless of whether an activating isoform is already bound to the promoter. In the former case, an increase in concentration of repressing Antp species enhances the probability to reach state “R” only if the promoter is in state “U” (Figure 5 B and B”). In the latter case, also the rate of switching between “A” and “R” depends on the concentration of repressing isoforms of Antp (Figure 5 A-A’’). We analyzed both model variants by forward simulation and found that both of them can explain the increase in average Antp concentration between early and late stages (Figure 5 A’’ and B’’), as well as the relative fraction of repressing and activating isoforms (Figure 5 D and D’). However, only the non-competitive binding model (Figure 5 A) can explain the substantial reduction of total Antp variability between early and late stages (Figure 5 A’), whereas in the competitive model variability is not reduced (Figure 5 B’). We additionally established that the negative feedback is required for suppression of variability (Figure 5 C-C’’), since without this, no suppression of variability is conferred (Figure 5 C’). Thus, our model suggested that auto-repression is required and it is possible also if an activating isoform of Antp is already bound to the P1 promoter. Correspondingly, we use the non-competitive promoter model for further analyses.
To further validate our model, we first examined its predictions on variability by comparing the variability values predicted by the model to the ones generated by our experimental measurements. Next to the CV2, the Fano factor (expressed as the variance over the mean, in concentration units) is another commonly used index to quantify variability in biological systems. Fano factor values that increase with average concentrations indicate that the underlying transcriptional processes cannot be sufficiently explained by a simple one-step promoter configuration with purely intrinsic Poissonian noise and that extrinsic noise is likely to contribute significantly to the overall variability (Newman et al., 2006; Schwanhausser et al., 2011; Taniguchi et al., 2010). Our model predicted a decrease in variability as a function of total Antp concentration and an increase in the Fano factor. These findings are in good agreement with our experimental data (compare Figure 5 E to E’ and F to F’).
We next analyzed the model behavior under different genetic perturbations. We found that overexpression of both auto-activating and auto-repressing isoforms leads to an increase of the total Antp concentration (Figure 5 G’ and H’), but there is no negative effect on the noise suppressing property of the circuit (Figure 5 G and H). In fact, the variability is even further decreased, which can be explained by the characteristic inverse relation between intrinsic noise and average concentration (Paulsson, 2004). In line with this prediction, flies expressing roughly eightfold concentration of either SynthAntp auto-activating or auto-repressing isoform in distal appendages (Figure 5 I and J, Figure 2 – figure supplement 2 A and I) or the notum (Figure 5 K and Figure 5 – figure supplement 2 A and A’) displayed the wild type morphology, indicative of normal Antp function.
However, overexpression of an exogenous repressor, such as Sex combs reduced (Scr), which can only repress Antp at the transcriptional level, but can neither activate it nor activate its own transcription (Figure 5 – figure supplement 2 B-G’), was predicted by the model to block transcription almost entirely and to have correspondingly severe effects on Antp dynamics (Figure 5 – figure L and L’), leading to high variability even at late stages and to very low expression levels. As expected, flies overexpressing SynthScr in the distal appendages or the notum, displayed severe malformations of both leg and notum development (Figure 5 M-O).
Taken together, the minimal model of Antp auto-regulatory genetic circuit is able to explain the experimentally observed differences in Antp concentration and cell-to-cell variability at early and late developmental stages.
Discussion
In this work, we have characterized the endogenous molecular numbers (concentration) and cell-to-cell variability in concentration of 14 TFs in Drosophila imaginal discs by FCS. We have identified Antp as a TF displaying, for its high average concentrations, considerably lower variability among cells. We used a combination of genetics, FCS and mathematical modeling to quantitatively characterize Antp behavior in live imaginal discs and identified a kinetic mechanism responsible for the suppression of variability in third instar discs compared to earlier developmental stages. We found that Antp can auto-regulate its expression levels during the course of development, starting from a preferentially auto-activating state early in development and transitioning to a preferentially auto-repressing state later. The early state is characterized by lower average Antp concentrations and high variability, whereas the opposite is true for the later repressing state. Without excluding other mechanisms, such as chromatin configuration and accessibility of Hox binding sites to Antp, we showed that differential expression of Antp isoforms is one contributing mechanism for the observed regulatory switch. These isoforms have preferentially activating or repressing activities on the Antp promoter, bind chromatin with different affinities and are themselves expressed in different relative amounts during development. Finally, based on this data, we have derived a simple kinetic model of Antp auto-regulation and confirmed its predictions by introducing genetic perturbations.
Negative auto-regulation has been identified as a frequently deployed mechanism for the reduction of noise (cell-to-cell variability) and the increase of regulatory robustness in various systems (Becskei and Serrano, 2000; Dublanche et al., 2006; Gronlund et al., 2013; Nevozhay et al., 2009; Shimoga et al., 2013; Thattai and van Oudenaarden, 2001). Auto-repression has been described for the Hox gene Ultrabithorax (Ubx) in haltere specification and as a mechanism of controlling Ubx levels against genetic variation (Crickmore et al., 2009; Garaulet et al., 2008), as well as in Ubx promoter regulation in Drosophila S2 cells (Krasnow et al., 1989). In contrast, an auto-activating mechanism is responsible for the maintenance of Deformed expression in the embryo (Kuziora and McGinnis, 1988). Moreover, global auto-regulation of Hox gene complexes has been shown to be in effect also in mammalian limb development (Sheth et al., 2014). These experiments point to evolutionarily conserved mechanisms for establishing (auto-activation) or limiting (auto-repression) Hox TF levels and variability in different developmental contexts.
Our data suggest that the developmental switch from auto-activation to auto-repression is, at least in part, mediated by molecularly distinct Antp linker isoforms. Differences in affinities of different Hox TF isoforms, based on their linker between the YPWM motif and the homeodomain, have also been identified for the Hox TF Ubx. Interestingly, its linker is also subject to alternative splicing at the RNA level (Reed et al., 2010). In a similar way to Antp, the long linker Ubx isoform displays 4-5fold lower affinity of DNA binding, as compared to short linker isoforms, and the two isoforms are not functionally interchangeable in in vivo assays. Finally, the Ubx linker also affects the strength of its interaction with the Hox cofactor Extradenticle (Exd), underscoring the functional importance of linker length in Hox TF function (Saadaoui et al., 2011). Thus, protein isoform control might represent a common regulatory mechanism of Hox-mediated transcriptional regulation.
Our model predicted that the Antp auto-regulatory circuit is robust with respect to initial conditions and extrinsic noise by being able to suppress cell-to-cell concentration variability even at very high concentrations of the auto-repressing or the auto-activating Antp isoform. This “buffering” capacity on cell-to-cell variability is reflected in the ability of flies to tolerate up to 7-fold overexpression of Antp without exhibiting abnormal phenotypes. Therefore, two different isoforms produced from the same gene with opposing roles in transcriptional regulation and different auto-regulatory binding sites on the gene’s promoter seem to suffice to create a robust gene expression circuit that is able to “buffer” perturbations of the starting conditions. So far, we have only been able to indiscriminately increase or decrease Antp concentration at the tissue level and record the phenotypic outcome of these perturbations. It will be interesting to test whether controlled perturbations of TF variability at the tissue level (making TF concentration patterns less or more noisy among neighboring cells) lead to abnormal phenotypes, however the technology for such manipulation is currently not readily accessible in flies.
While our study has focused on the quantitative analysis of the influence of Antp on its own expression, it may also have important implications for the regulation of Antp target genes. In particular, the repression and activation of genes through different isoforms of the same TF represents a plausible design principle to achieve differential expression of target genes. This could be achieved either through temporal developmental control of isoform abundance, or by spatial control, for example through the formation of different nuclear microenvironments.
In the case of temporal control, since target genes harbor different amounts of binding sites for activating and repressing isoforms, transcriptional programs of cells could be easily switched by changes in relative concentration of TF isoforms. In the case of Antp, for instance, targets that allow binding of only the short-linker, repressing isoform, could be highly expressed initially, but expression would be shut down at later stages as soon as this isoform becomes dominant. Conversely, the opposite behavior would be observed for targets featuring binding sites for only the activating isoform. We previously established that Antp target genes are activated or repressed with different efficiencies by Antp isoforms (Papadopoulos et al., 2011). Such a mechanism would allow distinct sets of target genes to be differentially regulated by the same Hox TF at different developmental stages. Evidence from genome-wide investigation of Hox target sites in the wing disc during development (third instar larval, prepupal and pupal wing discs) has shown that target gene batteries change dramatically in development (Pavlopoulos and Akam, 2011).
In the case of spatial control, enhancer binding sites of similar affinity could cluster in topologically distinct regions in the nucleus, according to the highest concentrations of either the preferentially activating or the repressing Antp isoform, thus creating microenvironments, capable of favoring transcriptional activation or repression. In this case, developmental control of the abundances of these two isoforms would change the efficiencies of activation and repression of their targets by enrichment or shrinkage of the microenvironments and/or departure/arrival of enhancers to pre-existing microenvironments. We have already observed that Antp nuclear distribution is not even, but features sites of accumulation (Figure 5 – figure supplement 1 B). While this scenario remains to be examined in detail for Antp, recent work on the Ubx-dependent expression of the shavenbaby (svb) gene identified that this mechanism of generation of functionally distinct nuclear microenvironments not only exists, but also allows robust expression of the svb locus from low-affinity enhancers (Crocker et al., 2015; Crocker et al., 2016; Crocker et al., 2017).
In general, while this work has increased our understanding of how developmentally important TFs secure their own regulatory robustness, it will be interesting to investigate whether the design principles of auto-regulatory circuits extend also to target genes.
Materials and Methods
Fly stocks used
The Antp-eGFP MiMIC line has been a kind gift from Hugo J. Bellen. The atonal (VDRC ID 318959), brinker (VDRC ID 318246), spalt major (VDRC ID 318068), yorkie (VDRC ID 318237), senseless (VDRC ID 318017) and Sex combs reduced (VDRC ID 318441) fosmid lines are available from the Vienna Drosophila Resource Center (VDRC) and have been generated recently in our laboratory (Sarov et al., 2016). The fork head (stock 43951), grainy head (stock 42272), Abdominal B (stock 38625), eyeless, (stock 42271), spineless (transcript variant A, stock 42289), and grain (stock 58483) tagged BACs were generated by Rebecca Spokony and Kevin P. White and are available at the Bloomington Stock Center. For the scalloped gene, a GFP-trap line was used (Buszczak et al., 2007), a kind gift from Allan C. Spradling laboratory (line CA07575), with which genome-wide chromatin immunoprecipitation experiments have been performed (Slattery et al., 2013). For the spineless gene, Bloomington stock 42676, which tags isoforms C and D of the Spineless protein has been also tried in fluorescence imaging and FCS experiments, but did not yield detectable fluorescence in the antennal disc, rendering it inappropriate to be used in our analysis. Therefore, we resided to stock 42289, which tags the A isoform of the protein. For the eyeless gene, the FlyFos015860(pRedFlp-Hgr)(ey13630::2XTY1-SGFP-V5-preTEV-BLRP-3XFLAG)dFRT line (VDRC ID 318018) has been tried also in fluorescence imaging and FCS experiments, but did not yield detectable fluorescence in the eye disc for it to be used in our analysis. The act5C-FRT-yellow-FRT-Gal4 (Ay-Gal4) line used for clonal overexpression or RNAi knockdown has been described (Ito et al., 1997). The UAS-Antp lines (synthetic and full-length), as well as UAS-SynthScr constructs have been previously described (Papadopoulos et al., 2011; Papadopoulos et al., 2010). The Dll-Gal4 (MD23) line has been a kind gift of Ginés Morata (Calleja et al., 1996). 69B-Gal4 and ptc-Gal4 have been obtained from the Bloomington Stock Center. The Antp P1-lacZ and P2-lacZ have been previously described (Engstrom et al., 1992; Zink et al., 1991). The P1 reporter construct spans the region between 9.4 kb upstream of the P1 promoter transcription initiation site and 7.8 kb downstream into the first intron, including the first exon sequences and thus comprising 17.2 kb of Antp regulatory sequences (pAPT 1.8). The line used has been an insertion of the pAPT 1.8 vector bearing the P1 promoter regulatory sequences upstream of an actin-lacZ cytoplasmic reporter and has been inserted in cytogenetic location 99F on the right chromosomal arm of chromosome 3. The Antp-RNAi line has been from VDRC, line KK101774. UAS-eGFP stock was a kind gift of Konrad Basler. We are indebted to Sebastian Dunst for generating the ubi-FRT-mCherry(stop)-FRT-Gal4(VK37)/CyO line, which drives clonal overexpression upon flippase excision, while simultaneously marking cells by the loss of mCherry. For red-color labeling of clones the act5C-FRT-CD2-FRT-Gal4, UAS-mRFP1(NLS)/TM3 stock 30558 from the Bloomington Stock Center has been used. For marking the ectopic expression domain of untagged Antp proteins the UAS-mRFP1(NLS)/TM3 stock 31417 from the Bloomington Stock Center has been used. The MS243-Gal4; UAS-GFP/CyO line was a kind gift from the laboratory of Ernesto Sánchez-Herrero.
Fly genotypes corresponding to fluorescence images
Figure 1 – figure supplement 1 A: FlyFos018487(pRedFlp-Hgr)(ato37785::2XTY1-SGFP-V5-preTEV-BLRP-3XFLAG)dFRT
Figure 1 – figure supplement 1 B: FlyFos024884(pRedFlp-Hgr)(brk25146::2XTY1-SGFP-V5-preTEV-BLRP-3XFLAG)dFRT
Figure 1 – figure supplement 1 C: FlyFos030836(pRedFlp-Hgr)(salm30926::2XTY1-SGFP-V5-preTEV-BLRP-3XFLAG)dFRT
Figure 1 – figure supplement 1 D: FlyFos029681(pRedFlp-Hgr)(yki19975::2XTY1-SGFP-V5-preTEV-BLRP-3XFLAG)dFRT
Figure 1 – figure supplement 1 E: w1118; PBac(fkh-GFP.FPTB)VK00037/SM5
Figure 1 – figure supplement 1 F: sd-eGFP (FlyTrap, homozygous)
Figure 1 – figure supplement 1 G: w1118; PBac(grh-GFP.FPTB)VK00033
Figure 1 – figure supplement 1 H: FlyFos018974(pRedFlp-Hgr)(Scr19370::2XTY1-SGFP-V5-preTEV-BLRP-3XFLAG)dFRT
Figure 1 – figure supplement 1 I: FlyFos015942(pRedFlp-Hgr)(sens31022::2XTY1-SGFP-V5-preTEV-BLRP-3XFLAG)dFRT
Figure 1 – figure supplement 1 J and K: Antp-eGFP (MiMIC) homozygous (line MI02272, converted to an artificial exon)
Figure 1 – figure supplement 1 L: w1118; PBac(Abd-B-EGFP.S)VK00037/SM5
Figure 1 – figure supplement 1 M: w1118; PBac(ey-GFP.FPTB)VK00033
Figure 1 – figure supplement 1 N: w1118; PBac(ss-GFP.A.FPTB)VK00037
Figure 1 – figure supplement 1 O and P: w1118; PBac(grn-GFP.FPTB)VK00037
Figure 2 B and B’: hs-flp/+; act5C-FRT-yellow-FRT-Gal4/+; UAS-SynthAntp long linker-eGFP/+
Figure 2 C and C’: hs-flp/+; act5C-FRT-yellow-FRT-Gal4, UAS-eGFP/+; UAS-Antp long linker (full-length, untagged)/+
Figure 2 F and F’: Dll-Gal4 (MD23)/+; UAS-SynthAntp-eGFP/ Antp P1-lacZ
Figure 2 – figure supplement 1 A and A’: Antp P1-lacZ/TM3
Figure 2 – figure supplement 1 B and B’: Antp P2-lacZ/CyO
Figure 2 – figure supplement 1 C and C’: wild type
Figure 2 – figure supplement 1 D and D’: hs-flp; act5C-FRT-yellow-FRT-Gal4, UAS-eGFP
Figure 2 – figure supplement 1 E and E’: hs-flp/+; act5C-FRT-yellow-FRT-Gal4, UAS-eGFP/+; Antp P1-lacZ/+
Figure 2 – figure supplement 1 F and F’: hs-flp/+; act5C-FRT-yellow-FRT-Gal4, UAS-eGFP/+; UAS-Antp long linker (full-length, untagged)/ Antp P1-lacZ
Figure 2 – figure supplement 1 G and G’: Dll-Gal4 (MD23)/+; UAS-Antp long linker (full-length, untagged), UAS-mRFP1(NLS)/ Antp P1-lacZ
Figure 2 – figure supplement 1 H and H’: Dll-Gal4 (MD23)/+; UAS-mRFP1(NLS)/ Antp P1-lacZ
Figure 2 – figure supplement 2 A and E: Dll-Gal4 (MD23)/+; UAS-SynthAntp long linker-eGFP/+
Fig 2 – figure supplement 2 B and F: ptc-Gal4/+; UAS-SynthAntp long linker-eGFP/+
Figure 2 – figure supplement 2 C and G: Dll-Gal4 (MD713)/+; UAS-SynthAntp long linker-eGFP/+
Figure 2 – figure supplement 2 D, H and K, L, O: 69B-Gal4/UAS-SynthAntp long linker-eGFP
Figure 2 – figure supplement 2 M, N and P: 69B-Gal4/UAS-eGFP
Figure 3 B, B’, G and G’: hs-flp/+; ubi-FRT-mChery-FRT-Gal4/+; Antp-eGFP (MiMIC)/UAS-Antp long linker (full-length, untagged)
Figure 3 C and C’: hs-flp/+; UAS-AntpRNAi/+; Antp P1-lacZ/act5C-FRT-CD2-FRT-Gal4, UAS-mRFP1(NLS)
Figure 3 H and H’: hs-flp/+; UAS-AntpRNAi/act5C-FRT-yellow-FRT-Gal4, UAS-eGFP; Antp P1-lacZ/+
Figure 3 – figure supplement 1 B: Antp P1-lacZ/ TM6B
Figure 3 – figure supplement 2 A and A’: hs-flp/+; ubi-FRT-mChery-FRT-Gal4/+; Antp-eGFP (MiMIC)/UAS-Antp long linker (full-length, untagged)
Figure 3 – figure supplement 2 B-C’: hs-flp/+; ubi-FRT-mChery-FRT-Gal4/+; Antp-eGFP (MiMIC)/+
Figure 3 – figure supplement 2 D and D’: hs-flp/+; act5C-FRT-yellow-FRT-Gal4, UAS-eGFP/+; Antp P1-lacZ/UAS-Antp long linker (full-length, untagged)
Figure 3 – figure supplement 2 E and E’: hs-flp/+; act5C-FRT-yellow-FRT-Gal4/+; UAS-SynthAntp long linker-eGFP/+
Figure 3 – figure supplement 2 F and F’: hs-flp/+; act5C-FRT-yellow-FRT-Gal4, UAS-eGFP/+
Figure 3 – figure supplement 2 G and G’: hs-flp/+; act5C-FRT-yellow-FRT-Gal4, UAS-eGFP/+; Antp P1-lacZ/+
Figure 3 – figure supplement 2 H and H’: hs-flp/+; UAS-AntpRNAi/+; Antp-eGFP (MiMIC)/act5C-FRT-CD2-FRT-Gal4, UAS-mRFP1(NLS)
Figure 4 B and B’: ptc-Gal4/+; UAS-SynthAntp long linker-eGFP/+
Figure 4 C and C’: Dll-Gal4 (MD23)/+; UAS-SynthAntp long linker-eGFP/ Antp P1-lacZ
Figure 4 F and F’: ptc-Gal4/+; UAS-SynthAntp long linker-eGFP/+
Figure 4 G and G’: Dll-Gal4 (MD23)/+; UAS-SynthAntp short linker-eGFP/ Antp P1-lacZ
Figure 4 – figure supplement 1 A and A’: hs-flp/+; act5C-FRT-yellow-FRT-Gal4/+; UAS-SynthAntp short linker-eGFP/+
Figure 4 – figure supplement 1 B, B’, G and G’: hs-flp/+; act5C-FRT-yellow-FRT-Gal4/+; UAS-SynthAntp short linker-eGFP/ Antp P1-lacZ
Figure 4 – figure supplement 1 C, C’, H and H’: hs-flp/+; act5C-FRT-yellow-FRT-Gal4/+; UAS-Antp short linker (full-length, untagged)/ Antp P1-lacZ
Figure 4 – figure supplement 1 D and D’: hs-flp/+; Dll-Gal4 (MD23)/+; UAS-Antp short linker (full-length, untagged), UAS-mRFP1(NLS)/ Antp P1-lacZ
Figure 4 – figure supplement 1 E-F’: hs-flp/+; ubi-FRT-mChery-FRT-Gal4/+; Antp-eGFP (MiMIC)/UAS-Antp short linker (full-length, untagged)
Figure 4 – figure supplement 1 I and I’: ptc-Gal4/+; UAS-SynthAntp long linker-eGFP/ Antp P1-lacZ
Figure 4 – figure supplement 1 J and J’: ptc-Gal4/+; UAS-SynthAntp short linker-eGFP/ Antp P1-lacZ
Figure 5 I and J: Dll-Gal4 (MD23)/+; UAS-SynthAntp long linker-eGFP/+ or Dll-Gal4 (MD23)/+; UAS-SynthAntp short linker-eGFP/+
Figure 5 K: MS243-Gal4/+; UAS-SynthAntp long linker-eGFP/Dr or MS243-Gal4/+; UAS-SynthAntp long linker-eGFP/Dr
Figure 5 M and N: Dll-Gal4 (MD23)/+; UAS-mCitrine-SynthScr/+
Figure 5 O: MS243-Gal4/+; UAS-mCitrine-SynthScr/+
Figure 5 P: MS243-Gal4/+; UAS-mCitrine-SynthScr/+
Figure 5 – figure supplement 1 A-B’: ubi-mRFP1(NLS)/+ or y; Antp-eGFP (MiMIC)/+
Figure 5 – figure supplement 2 A and A’: MS243-Gal4/+; UAS-SynthAntp long linker-eGFP/Dr or MS243-Gal4/+; UAS-SynthAntp long linker-eGFP/Dr
Figure 5 – figure supplement 2 B and B’: ptc-Gal4/+; UAS-mCitrine-SynthScr/+
Figure 5 – figure supplement 2 C and C’: MS243-Gal4/+; UAS-mCitrine-SynthScr/+
Figure 5 – figure supplement 2 D and D’: ptc-Gal4/+; UAS-mCitrine-SynthScr/ Antp P1-lacZ
Figure 5 – figure supplement 2 E and E’: Dll-Gal4 (MD23)/+; UAS-mCitrine-SynthScr/ Antp P1-lacZ
Figure 5 – figure supplement 2 F and F’: Dll-Gal4 (MD23)/+; UAS-mCitrine-SynthScr/+
Figure 5 – figure supplement 2 G and G’: MS243-Gal4/+; UAS-eGFP/+
Preparation of second and third instar imaginal discs for FCS measurements
For FCS measurements, imaginal discs (eye-antennal, wing, leg, humeral and genital) and salivary glands were dissected from third instar wandering larvae, or wing and leg discs from second instar larvae, in Grace’s insect tissue culture medium (ThermoFisher Scientific, 11595030) and transferred to 8-well chambered coverglass (Nunc® Lab-Tek™, 155411) containing PBS just prior to imaging or FCS measurements. Floating imaginal discs or salivary glands were sunk to the bottom of the well using forceps.
Immunostainings in larval imaginal discs
Larval imaginal discs were stained according to (Papadopoulos et al., 2010). Stainings for the endogenous Antp protein have been performed using a mouse anti-Antp antibody (Developmental Studies Hybridoma Bank, University of Iowa, anti-Antp 4C3) in a dilution of 1:250 for embryos and 1:500 for imaginal discs. eGFP, or eGFP-tagged proteins have been stained using mouse or rabbit anti-GFP antibodies from ThermoFisher Scientific in a dilution of 1:500 in imaginal discs and 1:250 in embryos. mRFP1 was stained using a Chromotek rat anti-RFP antibody. For Antp P1 promoter stainings in imaginal discs we used the mouse anti-β-galactosidase 40-1a antibody from Developmental Studies Hybridoma Bank, University of Iowa in a dilution of 1:50. The rabbit anti-Scr antibody was used in a dilution of 1:300 (LeMotte et al., 1989). Confocal images of antibody stainings represent predominatly Z-projections and Zeiss LSM510, Zeiss LSM700 or Zeiss LSM880 Airyscan confocal laser scanning microscopy systems with an inverted stand Axio Observer microscope were used for imaging. Image processing and quantifications have been performed in Fiji (Schindelin et al., 2012). For optimal spectral separation, secondary antibodies coupled to Alexa405, Alexa488, Alexa594 and Cy5 (ThermoFischer Scientific) were used.
Colocalization of wild type and eGFP-tagged MiMIC Antp alleles in imaginal discs
To examine whether the pattern of the MiMIC Antp-eGFP fusion protein recapitulates the Antp wild type expression pattern in both embryo and larval imaginal discs, we performed immunostainings of heterozygous Antp-eGFP and wild type flies to visualize the embryonic (stage 13) and larval expression of Antp and eGFP. In this experiment, we 1) visualized the overlap between eGFP and Antp (the eGFP pattern reflects the protein encoded by the MiMIC allele, whereas the Antp pattern reflects the sum of protein produced by the MiMIC allele and the allele of the balancer chromosome) and 2) compared the eGFP expression pattern to the Antp expression pattern in wild type discs and embryos.
Induction of early and late overexpression and RNAi-knockdown clones in imaginal discs
Genetic crosses with approximately 100 virgin female and 100 male flies were set up in bottles and the flies were allowed to mate for 2 days. Then, they were transferred to new bottles and embryos were collected for 6 hours at 25°C. Flies were then transferred to fresh bottles and kept until the next collection at 18°C. To asses Antp auto-activation, the collected eggs were allowed to grow at 25°C for 26 h from the midpoint of collection, when they were subjected to heat-shock by submersion to a water-bath of 38°C for 30 min and then placed back at 25°C until they reached the stage of third instar wandering larvae, when they were collected for dissection, fixation and staining with antibodies. To assess Antp auto-repression, the same procedure was followed, except that the heat-shock was performed at 60 h of development after the midpoint of embryo collection. Whenever necessary, larval genotypes were selected under a dissection stereomicroscope with green and red fluorescence filters on the basis of deformed (dfd)-YFP bearing balancer chromosomes (Le et al., 2006) and visual inspection of fluorescence in imaginal discs.
Measurement of Antp transcript variant abundance
The linker between the Antp YPWM motif and the homeodomain contains the sequence RSQFGKCQE. Short linker isoforms encode the sequence RSQFE, whereas long linker isoforms are generated by alternative splicing of a 12 base pair sequence encoding the four amino acid sequence GKCQ into the mRNA. We initially designed primer pairs for RT-qPCR experiments to distinguish between the short and long linker mRNA variants. For the short linker variant, we used nucleotide sequences corresponding to RSQFERKR (with RKR being the first 3 amino acids of the homeodomain). For detection of the long linker variant we designed primers either corresponding to the RSQFGKCQ sequence, or GKCQERKR. We observed in control PCRs (using plasmid DNA harboring either a long or a short linker cDNA) that primers designed for the short linker variant still amplified the long linker one. Moreover, with linker sequences differing in only four amino acids, encoded by 12 base pars, primer pairs flanking the linker could also not be used, since, due to very similar sizes, both variants would be amplified in RT-qPCR experiments with almost equal efficiencies. Therefore, we used primer pairs flanking the linker region to indiscriminately amplify short and long linker variants, using non-saturating PCR (18 cycles) on total cDNA generated from total RNA. We then resolved and assessed the relative amounts of long and short linker amplicons in a second step using Fragment Analyzer (Advanced Analytical). RNA was extracted from stage 13 embryos, second instar larvae at 60 h of development, and leg or wing discs from third instar wandering larvae using the Trizol® reagent (ThermoFischer Scientific), following the manufacturer’s instructions. Total RNA amounts were measured by NanoDrop and equal amounts were used to synthesize cDNA using High-Capacity RNA-to-cDNA™ Kit (ThermoFischer Scientific), following the manufacturer’s instructions. Total cDNA yields were measured by NanoDrop and equal amounts were used in PCR, using in-house produced Taq polymerase. 10 ng of plasmid DNA, bearing either a long or a short transcript cDNA were used as a control. PCR product abundance was analyzed both by agarose gel electrophoresis and using Fragment Analyzer (Advanced Analytical).
The quantification of the transcript variant concentration (Figure 4 D and H) has been made considering 100% (value equal to 1 on the y axis) as the sum of long and short isoforms at each developmental stage, whereas the quantification of the relative activation and repression efficiency has been performed considering the short linker variant as having 100% repression and the long linker variant as having 100% activation (values equal to 1 on the y-axis) efficiency.
Quantification of the relative repressing and activating efficiencies of different Antp isoforms
Quantification of the relative efficiency of Antp activating and repressing isoforms (Figure 4 D and H) were performed in Fiji (Schindelin et al., 2012) by selection of the total region of repression or activation of Antp protein or P1 reporter staining and quantification of the relative fluorescence intensity of the selected regions. 5-7 imaginal disc images per investigated genotype were used for analysis. For the repression assay the obtained values have been normalized over the intensity of Antp protein calculated in the region of overlap between eGFP and Antp (negative control). In both cases (repression and activation), the highest efficiency per transcript variant (for repression, the short linker isoform; for activation the long linker isoform) have been set to 100%.
Gel-shifts (Electrophoretic Mobility Shift Assays – EMSAs)
Full-length Antp short and long linker variants (transcript variants RM and RN), encoding activating and repressing Antp isoforms, respectively, were cloned into the pET21b(+) vector (Novagen), which features a C-terminal 6xHis tag, and expressed in RosettaTM 2 cells (Novagen), following the manufacturer’s standard protocol. The two proteins were then Ni-column purified and subjected to gelfiltration. The concentrations of purified proteins were then compared by Western blotting, using the anti-Antp 4C3 antibody (Developmental Studies Hybridoma Bank, University of Iowa), and equal starting concentrations were used in successive twofold serial dilutions in gel-shift experiments. The BS1and BS2 binding sites have been identified ∼2 kb upstream of the engrailed gene promoter and characterized for Antp binding previously (Affolter et al., 1990). The HB1 binding site has been described previously (Keegan et al., 1997) and is a binding site found in the intron of the mouse Hoxa-4 gene. The D4 probe has been characterized previously (Duncan et al., 2010) as a functional element in the spineless gene.
Fluorescence Microscopy Imaging of live imaginal discs and FCS
Fluorescence imaging and FCS measurements were performed on two uniquely modified confocal laser scanning microscopy systems, both comprised of the ConfoCor3 system (Carl Zeiss, Jena, Germany) and consisting of either an inverted microscope for transmitted light and epifluorescence (Axiovert 200 M); a VIS-laser module comprising the Ar/ArKr (458, 477, 488 and 514 nm), HeNe 543 nm and HeNe 633 nm lasers and the scanning module LSM510 META; or a Zeiss LSM780 inverted setup, comprising Diode 405 nm, Ar multiline 458, 488 and 514 nm, DPSS 561 nm and HeNe 633 nm lasers. Both instruments were modified to enable detection using silicon Avalanche Photo Detectors (SPCM-AQR-1X; PerkinElmer, USA) for imaging and FCS. Images were recorded at a 512X512 pixel resolution. C-Apochromat 40x/1.2 W UV-VIS-IR objectives were used throughout. Fluorescence intensity fluctuations were recorded in arrays of 10 consecutive measurements, each measurement lasting 10 s. Averaged curves were analyzed using the software for online data analysis or exported and fitted offline using the OriginPro 8 data analysis software (OriginLab Corporation, Northampton, MA). In either case, the nonlinear least square fitting of the autocorrelation curve was performed using the Levenberg–Marquardt algorithm. Quality of the fitting was evaluated by visual inspection and by residuals analysis. Control FCS measurements to asses the detection volume were routinely performed prior to data acquisition, using dilute solutions of known concentration of Rhodamine 6G and Alexa488 dyes. The variability between independent measurements reflects variabilitys between cells, rather than imprecision of FCS measurements. For more details on Fluorescence Microscopy Imaging and FCS, refer to Supplement 1.
Sample size, biological and technical replicates
For the measurement of TF molecular numbers and variability (Figure 1 and Figure 1 – figure supplement 1), 7-10 larvae of each fly strain were dissected, yielding at least 15 imaginal discs, which were used in FCS analysis. For the Fkh TF, 7 pairs of salivary glands were analyzed and for AbdB, 12 genital discs were dissected from 12 larvae. More than 50 FCS measurements were performed in patches of neighboring cells of these dissected discs, in the regions of expression indicated in Figure 1 by arrows. Imaginal discs from the same fly strain (expressing a given endogenously-tagged TF) were analyzed on at least 3 independent instances (FCS sessions), taking place on different days (biological replicates) and for Antp, which was further analyzed in this study, more than 20 independent FCS sessions were used. As routinely done with FCS measurements in live cells, these measurements were evaluated during acquisition and subsequent analysis and, based on their quality (high counts per molecule and second, low photobleaching), were included in the calculation of concentration and variability. In Figure 1 – figure supplement 1 Q, n denotes the number of FCS measurements included in the calculations.
For experiments involving immunostainings in imaginal discs to investigate the auto-regulatory behavior of Antp (Figures 2-5 and supplements thereof, except for the temporally-resolved auto-activating and repressing study of Antp in Figure 3, as discussed above), 14-20 male and female flies were mated in bottles and 10 larvae were selected by means of fluorescent balancers and processed downstream. Up to 20 imaginal discs were visualized by fluorescence microscopy and high resolution Z-stacks were acquired for 3-5 representative discs or disc regions of interest per experiment. All experiments were performed in triplicate, except for the temporal analysis of Antp auto-regulatory behavior in Figure 3 (and figure supplements thereof), which was performed 6 times and the quantification of repression efficiency of short and long linker Antp isoforms in Figure 5 (and figure supplements thereof), which was performed 5 times.
For the quantification of transcript variant abundance in Figure 4 D and H, RNA and thus cDNA was prepared from each stage 3 independent times (biological replicates) and the transcript abundance per RNA/cDNA sample was also analyzed 3 times.
For the experiments involving perturbations in Antp expression whereby the proper development of the leg and the notum have been assessed in Figure 5, more than 100 adult flies have been analyzed and this experiment has been performed more than 10 times independently.
Acknowledgements
We are deeply saddened by the unexpected passing of Prof. Walter J. Gehring, at the very inception of this work, when the project was still in the planning and preliminary data gathering stage. Prof. Gehring was an extraordinary human being and a scientific giant, whose work will continue to educate and inspire generations to come. The authors are indepted to Sonal Nagarkar Jaiswal, Paolo Mangahas, and Hugo J. Bellen for creating and sharing with us the Antp-eGFP line. DKP has been supported by a long-term fellowship from the Swiss National Science Foundation (PBBSP-138700) and a long-term fellowship from the Federation of European Biochemical Societies (FEBS) at initial stages of this project. VV has been supported by the Knut and Alice Wallenberg foundation and Karolinska Institute Research Funds. DKP would like to express his gratitude to PT for outstanding scientific, and uninterrupted financial, support. DKP is indebted to Markus Burkhardt, head of the imaging platform at the Center for Regenerative Therapies Dresden (CRTD), for help and discussions regarding FCS experiments in Dresden; Sylke Winkler and the DNA sequencing facility of the Max-Planck Institute of Molecular Cell Biology and Genetics (MPI-CBG) for providing assistance with Antp transcript variant quantification; Aliona Bogdanova and the Protein expression and purification facility of the MPI-CBG for purification of the repressing and activating Antp isoforms; as well as the Light Microscopy facility of MPI-CBG. DKP is also grateful to KS for numerous discussions and support throughout the implementation of this work and to Konstantinos Papadopoulos for advice on the mathematical analysis of the relative binding constants of repressing and activating Antp isoforms. The authors would like to acknowledge Jan Brugues and Thomas M. Schultheiss for their critical reading and insightful comments on the manuscript.