Multi-enhancer transcriptional hubs confer phenotypic robustness

Thank you for sending your article entitled "Multi-enhancer transcriptional hubs confer phenotypic robustness" for peer review at eLife. Your article is being evaluated by three peer reviewers, and the evaluation is being overseen by a Reviewing Editor and Jessica Tyler as the Senior Editor.

Given the list of essential revisions, including new experiments, the editors and reviewers invite you to respond within the next two weeks with an action plan and timetable for the completion of the additional work. We plan to share your responses with the reviewers and then issue a binding recommendation.

Essential points to address are the following:

1) The connection between enhancers, Ubx protein concentration, transcriptional output, and phenotype. Although different concentrations of Ubx are measured at the deleted vs. wild-type locus, an important point that the three reviewers agree upon is that gene expression levels and Ubx concentrations are not well correlated. There appear to be other unexplained factors that are influencing the expression of svb that do not correlate with the TF concentration. Thus, it is unclear whether the experiments described here support or reject the hypothesis laid out in the "multi-enhancer "hubs" improve robustness by increasing transcription factor retention near transcription sites."

The authors must consider further whether the global measurements shown in Figure 3 are the most effective measure of this proposed correlation, or whether a more fine-grained analysis might support the idea that boosted Ubx levels from complexed enhancers are driving transcription. As noted below, the authors might show scatter plots of Ubx concentration vs. svb transcription within each condition. A positive correlation would support their hypothesis.

The trichome phenotype is also not directly correlated to the transcriptional outputs and Ubx concentrations, and as mentioned in the Discussion, further layers of regulation are likely in play. This idea needs to be more fully fleshed out.

A finding related to "enhancer hubs boosting local concentration" is the finding that an ectopic BAC appears not to influence Ubx concentration in a wild-type svb background, only in a deletion background. Does this finding indicate that such hub formation is saturable? And if so, does this indicate that the wild-type svb locus forms a local hub with enhancers in the immediate vicinity, and trans-complementation is not a normal feature of svb function?

2) A second point raised concerned the exact cis regulatory regions in play. A minimal DG3 enhancer drives gene expression in ventral abdominal stripes, but does not rescue Ubx concentrations from a trans setting. A larger deletion that includes DG3 as well as additional Ubx binding regions (that are not sufficient for, but may be part of, ventral DG3-related activity) impacts transcription, trichome development, and robustness. A yet larger cis-regulatory domain on a BAC rescues some aspects of gene expression and Ubx concentration. The interpretation conflates DG3 with the function of the deleted region; reviewers noted that a more careful interpretation would differentiate results from each of these different cis elements. For instance, the lack of trans-rescue by the DG3 enhancer alone may be due to the inability of a short segment to transvect effectively. The interpretation should explicitly take into account known properties of transvecting regulatory loci in Drosophila.

3) Several aspects can be addressed by better justification and explanation of methods and data presentation, including

– Why sometimes either A1 or A2 trichomes are quantitatively assessed, depending on the figure;

– The use of one-tailed (vs. two-tailed) T-tests for statistical relevance;

– Recommendations for inclusion of both 25C and 32C phenotypes and data uniformly, and;

– Clarification of exact role of Ubx in T1-T3 regulation, as the conclusions drawn about DG3 and Ubx roles are difficult to know based on the single images shown;

– The number of data points in Figures 3 and 4 are limited; is there a technical limitation to more extensive sampling of the transcriptional readouts and Ubx intensities?

4) In some cases, the actual experimental approach was unclear to the reviewers:

– there was some confusion about whether the DG3 deletion mutants were homozygous, so there would be no wild-type copy of the gene in these embryos – is that indeed the case?

– Figure 4 shows svb, but not dsRed mRNA expression; is that correct? Imaging of dsRed only used to score a svb locus as "overlapped" vs. "not overlapped"?

– Choice of pixel size for ROI.

– Signal to noise for Ubx across the nucleus, as well as variation in average Ubx from sample to sample.

Reviewer #1:

This paper addresses a timely topic, namely how multiple enhancers may coordinate to regulate gene expression. This is being actively investigated in a number of systems using both case studies and genome wide methods, and the field is fueled by new capabilities in high resolution microscopy. Thus, interest in the topic is high. However, I have substantial concerns about how this work contributes to this topic.

The central hypothesis of this work is that increased levels of Ubx, provided by micro-environments created through coalescing multiple Ubx-responsive enhancers, increases the level of transcription of shavenbaby, svb; this increased level of transcription allows svb expression to be robust to high temperature variation. However, the evidence presented in this paper does not consistently support this hypothesis.

Let's first consider the set of experiments correlating the presence of the enhancer deletion (via a deficiency), the concentration of Ubx and the level of svb transcription.

– Deleting an svb enhancer (DG3) decreases resilience to high temperature stress, resulting in smaller numbers of trichomes in A1.

– In the A1 segment, svb enhancers localize to a microenvironment where Ubx concentrations are high compared to background.

– Deleting an svb enhancer decreases the concentration of Ubx in these microenvironments moderately at high temperature, but has no effect at low temperature.

– At high temperatures, transcription of svb decreases at both WT and deficiency alleles.

– At low temperatures, transcription of svb decreases under low temperature stress for the deficiency allele.

These results do not support the hypothesis because svb transcription is not consistently correlated with Ubx levels in the microenvironment.

A second concern is the interpretation of the rescue experiment, where a BAC containing the full svb locus is integrated onto another chromosome. This partially rescues the trichome phenotype of the deficiency allele. However, the Ubx levels in microenvironments were unchanged from WT. If the extra svb regulatory region associates with the endogenous svb transcription site, this should either result in above WT levels of Ubx, or there is some additional layer of regulation. This was not discussed.

Reviewer #2:

In this paper, Tsai et al. studied the general problem of how low affinity binding sites, which promote enhancer specificity, also permit robust activation. They focus on ventral "shadow" enhancers for shavenbaby/svb.

The authors start by focusing on the DG3 enhancer, which was less studied before, and responds to Ubx in A1. Indeed, ChIP-seq data shows Ubx binding in a DG3 element that drive reporter gene expression in trichome forming territories, and this reporter responds to perturbations of Ubx.

A deletion, Df(x)svb¹⁰⁸, removes DG3, as well as several other Ubx-bound regions, and causes trichome formation defects only upon heat stress, and primarily in the region of DG3 activity.

State-of-the-art imaging of svb transcription and Ubx protein distribution show a reduction of Ubx concentration in the vicinity of active svb loci, specifically in Df(X)svb¹⁰⁸mutant cells under heat stress, and a reduction of transcriptional output upon heat stress, even in wild-type embryos.

Remarkably, exogenous svb cis-regulatory DNA could rescue the formation of a Ubx-rich "hub" containing the endogenous svb locus, and svb transcription, in Df(X)svb¹⁰⁸ mutants, but only when the two loci colocalized.

Taken together, the data presented is consistent with the notion that multiple enhancers form a Ubx-rich microenvironment necessary for robust svb expression and trichome formation. While the specifics are rather specialized, the problem of transcriptional hub formation and robust gene expression is a general problem of importance.

Nevertheless, there are still specific issues with the main conclusions based on the work presented, which seems somewhat preliminary.

According to Figure 2, Df(X)svb¹⁰⁸removes more than DG3, including elements that clearly bind Ubx as per Figure 1—figure supplement 1, so it is not clear if the observed defects can be attributed to loss of DG3. Ideally, CRISPR/Cas9 should be used to remove Ubx-bound regions outside DG3 (e.g. Ubx-bound DG2) and to remove DG3 specifically, in order to support the conclusions that the observed effects of Df(X)svb¹⁰⁸ are due to a loss of DG3.

DG3 remains active in the most dorso-lateral cells in Ubx mutant (Figure 1), in a domain that seems to be specifically marked by DG3 reporters (Figure 1). This domain is portrayed as primarily affected by reduction of trichome formation in Df(X)svb¹⁰⁸ mutants, but then it is not clear that this defect is due to a lack of Ubx activity on DG3 (both because DG3-LacZ seems to remain active and because the region removed by Df(X)svb¹⁰⁸binds Ubx outside DG3).

In Figure 3, the discrepancy between robust accumulation of Ubx, except in Df(X)svb¹⁰⁸embryo under stress, and the impact of heat stress on transcription even in wild-type embryos further indicates that changes in Ubx accumulation is only part of the explanation for both the Df(X)svb¹⁰⁸ phenotype and the response to heat stress.

The rescue presented in Figure 4 is a strong result but also difficult to interpret: for instance, it seems likely that the "hub" forms independently of Ubx, which presumably cannot accumulate on the Df(X)svb¹⁰⁸chromosome, and then help place the svb locus in a Ubx-rich microenvironment. This is implied but not explicitly stated, or clearly demonstrated by the data.

For instance, Figure 5 adds to the notion that regulatory inputs other than DG3 and Ubx govern the formation of the svb containing micro-environment and svb transcription. As acknowledged also in the Discussion, robustness of trichome formation also involves mechanisms downstream of svb transcription, making the link between Ubx hubs and robust trichome formation even more complex.

Taken together, a few of the criticisms above question whether the accumulation of Ubx, or the lack thereof, in microenvironments containing the svb locus explains the phenotypes described for Df(X)svb¹⁰⁸. Rescue experiments more directly combining loss of DG3 function with gain of Ubx function should help resolve these issues.

Reviewer #3:

In their previous work, Tsai, et al. observed that, during fruit fly embryogenesis, actively transcribing shavenbaby (svb) loci preferentially colocalized with regions of elevated concentration for the transcription factor Ubx, a known regulator of svb activity. Moreover, they observed that other Ubx target genes exhibited a tendency to colocalize to these same Ubx-enriched "microenvironments", even when located on different chromosomes. Based on these findings, they hypothesized that the spatial overlap between multiple distal enhancers and transcription factor microenvironments could result in increased transcriptional output and could be a mechanism for redundancy in the face of environmental stresses. In the present work, Tsai, et al. test this hypothesis directly by deleting the DG3 enhancer from the svb cis-regulatory region and assessing the effects on microenvironment formation, svb transcription, and trichome number (a phenotype tied to svb expression) under normal (25C) and heat-stress (32C) conditions. This manuscript constitutes an intriguing extension of the findings presented in the authors' earlier work on Ubx microenvironments in the vicinity of svb loci. They find that the deletion mutant results in a significant depletion in Ubx concentration around the mutated svb locus at 32C, and that this depletion coincides with a reduction both in svb transcription and trichome number. Furthermore, they find that the addition of the full cis-regulatory region of svb on a different chromosome leads to preferential colocalization between the added region and the original mutated svb locus. The authors report that local concentration of Ubx in the vicinity of these colocalized enhancer regions recovers to wild-type levels under heat shock. They argue that, consistent with their initial hypothesis, the coincident partial recovery of svb transcription and trichome number indicates that this colocalization of multiple Ubx targets to a single microenvironment is indeed functioning to counteract the embryo's mutation-induced sensitivity to heat shock conditions.

Overall, the new experimental results presented in this work are deeply intriguing and hold the potential to offer significant new insights regarding the physical nature of transcription factor hubs, as well as their functional role in the spatiotemporal control of transcriptional activity. Nonetheless, several key questions remain regarding the interpretation of the experimental data and the proposed functional role of transcriptional microenvironments in facilitating phenotypic robustness in the face of genotypic and environmental perturbations.

Perhaps most fundamentally, the authors' conclusions hinge upon the assumption that the Ubx enrichment signal they observe at active loci play a causal role in modulating transcriptional activity, yet they present no data to directly support this assumption. Indeed, the box plots in Figure 3 C and D seem to show that, while both Ubx concentration and transcription decrease in response to heat, DG3 deletion, or both, the relative impacts of these perturbations on transcription and Ubx concentration differ significantly in scale. This leaves open the possibility that the perturbations affect each feature separately, and that the correlation between transcription and local Ubx concentration are spurious, not causal. A simple but important check for this would be to show scatter plots of Ubx concentration vs. svb transcription within each condition. If the input concentration and output rate of transcription for a given locus are positively correlated when other conditions are held constant, this would constitute a solid basis for the authors' subsequent arguments. If, on the other hand, no correlation is evident, then this would fundamentally alter how the manuscripts' results are interpreted. This question can be resolved with data already in-hand and, in this reviewer's opinion, it will help clarify the significance of the reported experimental observations. Along similar lines, the strength of the work would benefit from the addition of plots that clarify not only how different variables (concentration, locus colocalization, transcription rate) change across conditions, but also with respect to one another.

On a more conceptual note, the authors' proposal that the colocalization of enhancers in transcription factor microenvironments could serve as an added layer of redundancy (akin to multiple TF binding sites and multiple enhancers) to increase the transcriptional robustness is thought-provoking. The concept of robustness is used throughout biology with different meanings, and further discussion of the implications and nuances of this proposal would also help establish what exactly the authors mean by it. For instance, what the authors cast as "robustness" could equally well be termed "desensitization". While robustness to environmental changes could be desirable, the fast pace of Drosophila development also necessitates rapid, precise transcriptional responses to changes in transcription factor concentrations. Thus, naively at least, if robustness leads to desensitization of the response of an enhancer to changes in the concentrations of inputs transcription factors, one might expect locus colocalization to actually be deleterious in the case of genes that must respond rapidly to time-varying transcription factor inputs. Augmenting the existing commentary to address this point would help clarify the implications of the present work. Does this trade-off make predictions regarding where and when this kind of clustering might manifest over the course of development (and for what genes)?

General Comments:

Can the authors comment on why the box plots in Figure 3 and Figure 4 contain so few data points? All contain 52 or fewer, yet one would expect there to be many more active svb loci per embryo at this point in development. If the full set of svb loci was not used for the analyses presented, can the authors comment on how the analysis subset was selected. Also, given that many of the effects presented are relatively subtle, the addition of data points (to the extent that it is feasible) would greatly enhance the robustness of the analysis and might lead to the discovery of additional trends within the data not evident in such small sample sizes. Given that the authors get multiple nuclei per embryos, is there a fundamental limitation to how much data they can present such as, for example, their data analysis pipeline?

As mentioned above, this work leans heavily upon the presumed causal relationship between local Ubx concentration and transcriptional output. The work would be greatly strengthened by the addition of bivariate analyses that rigorously test the relationship between Ubx concentration and transcription within each of the four conditions shown in Figure 3 C and D.

The boxplots in Figure 4C show that Ubx concentration at loci where the mutated svb site is colocalized with the svbBAC more or less recovers Ubx concentration at the endogenous (unperturbed) svb locus, yet the authors do not address whether the resulting signal is less than, equal to, or greater than the sum of the signal at each locus (Df(X)svb and svbBAC) when they are not colocalized. Showing how the enrichment signal at colocalized loci compares to the signal at each separately would indicate whether and to what degree some sort of synergy is at play, or whether the increased signal is merely a result of having more Ubx binding sites in region.

While the trends in Ubx intensity, radial separation, and transcriptional activity shown in Figure 4C, E, and F are interesting taken individually, much more could be learned from the data by making bivariate plots. How does transcriptional output at the mutated svb locus vary with the radial separation between loci? What about the Ubx intensity?