Predictive modeling reveals that higher-order cooperativity drives transcriptional repression in a synthetic developmental enhancer

  1. Yang Joon Kim
  2. Kaitlin Rhee
  3. Jonathan Liu
  4. Selene Jeammet
  5. Meghan A Turner
  6. Stephen J Small
  7. Hernan G Garcia

  • Curated by eLife

    eLife logo

    eLife assessment

    The work by Kim et al., used synthetic constructs in Drosophila to examine the relationship between regulators (activator/repressor) and transcription initiation. By measuring regulator concentrations and the corresponding RNA polymerase initiation rates in different synthetic constructs and using a thermodynamic model, the authors concluded that that higher-order cooperativities between the repressor on adjacent binding sites, and that between the repressor and RNA polymerase are needed to explain the observed response curves in RNA polymerase loading rate. This work targets a challenging question in eukaryotic transcription regulation, where higher-order cooperativity between different molecular components, in addition to simple transcription factor binding and unbinding, is often necessary to account for observed promoter behaviors when multiple elements (repressors, mediators, activators) exist.

Reviewed by

Read the full article See related articles

Listed in

Log in to save this article

Abstract

A challenge in quantitative biology is to predict output patterns of gene expression from knowledge of input transcription factor patterns and from the arrangement of binding sites for these transcription factors on regulatory DNA. We tested whether widespread thermodynamic models could be used to infer parameters describing simple regulatory architectures that inform parameter-free predictions of more complex enhancers in the context of transcriptional repression by Runt in the early fruit fly embryo. By modulating the number and placement of Runt binding sites within an enhancer, and quantifying the resulting transcriptional activity using live imaging, we discovered that thermodynamic models call for higher-order cooperativity between multiple molecular players. This higher-order cooperativity captures the combinatorial complexity underlying eukaryotic transcriptional regulation and cannot be determined from simpler regulatory architectures, highlighting the challenges in reaching a predictive understanding of transcriptional regulation in eukaryotes and calling for approaches that quantitatively dissect their molecular nature.

Article activity feed

  1. Version published to 10.7554/elife.73395 on eLife
  2. eLife

    Author Response

    Reviewer #1 (Public Review):

    In this manuscript by Kim et al., the authors use live-cell imaging of transcription in the Drosophila blastoderm to motivate quantitative models of gene regulation. Specifically, they focus on the role of repressors and use a 'thermodynamic' model as the conceptual framework for understanding the addition and placement of the repressor Runt, i.e. synthetic insertion of Runt repressor sites into the Bicoid-activated hunchback P2 enhancer. Coupled with kinetic modeling and live-cell imaging, this study is a sort of mathematical enhancer bashing experiment. The overarching theme is measuring the input/output relationship between an activator (bicoid), repressor (runt), and mRNA synthesis. Transcriptional repression is understudied in my opinion. One finding is that the inclusion of cooperativity between trans-acting factors is necessary for understanding transcriptional regulation. Most, if not all, of the tools used in this paper have been published elsewhere, but the real contribution is a deep, quantitative dissection of transcriptional regulation during development. As such, the only real questions for this referee are whether the modeling was done rigorously to produce some general biological conclusions. By and large, I think the answer is yes.

    We thank the reviewer for this thoughtful evaluation of our work. We agree with the reviewer’s assessment that transcriptional repression, especially the quantitative dissection of transcriptional repression, is understudied compared to transcriptional activation.

    Comments:

    Fig. 6 was the most striking figure for this referee, specifically that different placements of Runt molecules on the enhancer lead to distinct higher order interactions. I am wondering if the middle data column in Fig. 6 represents a real difference from the other two, and if so, it seems that the positioning - as opposed to simply the stoichiometry - is essential in cooperativity. This conclusion implies that transcriptional regulation is more precise than what some claim is just a mushy ball of factors close to a promoter. In other words, orientation may matter. Proximity may matter. Interactions in trans matter.

    We thank the reviewer for pointing out a feature of our data that we did not emphasize enough originally. Indeed, the construct in the middle column, which we termed [101], could be better recapitulated with the simplest model of zero free parameters than the other two constructs. As the reviewer pointed out, this raises an interesting question about the “grammar” of an enhancer: the placement and orientation of binding sites for transcription factors might matter yet we do not have a clear understanding of the logic. We have now incorporated a discussion of this topic in the Discussion section.

    There needs to be at least one prediction which is validated at the level of smFISH / mRNA in the embryo. Without detracting from the effort the authors have expended in looking directly at transcription, if the effects can't be felt by the blastoderm at the level of mRNA/cell, it becomes difficult to argue for the relevance to development. Also, I feel there is little chance that these measurements can be quantitatively replicated unless translated to the level of total protein or mRNA. Such a measurement (orthogonal quantitative confirmation of the repressor cooperativity result) would also assuage my concern about the time averaging as shown in Fig. S3.

    Our study focused on predicting the initial rate of transcription because it is the measurable quantity that most directly relates to the binding and action of the transcriptional activators and repressors used in this study. We argue that the action of transcription factors would be more accurately assessed by monitoring the rate of transcription, rather than the accumulated mRNA, which could be confounded by the dynamics of the whole transcription cycle—initiation, elongation and termination—as well as nuclear export, diffusion and degradation of transcripts. We are, of course, excited to eventually be able to predict a whole pattern of cytoplasmic mRNA over space and time from knowledge of the enhancer sequence. However, if we cannot predict the initial rate of RNA polymerase loading dictated by an enhancer, we argue that there is little hope in predicting such cytoplasmic patterns. We emphasized this point in the Discussion (Line XX-YY). Regardless, to assuage the reviewer’s concern, we have performed additional analyses to assess the effect of repression at the level of accumulated mRNA.

    First, we have quantified the accumulated mRNA during nuclear cycle 14, which is the time window that we have focused on in this study. To make this possible, we have integrated the area under the curve of MS2 time traces which has been already shown to be a reporter of the total amount of mRNA produced by FISH (Garcia et al., Current Biology 23:2140, 2013;Lammers et al., PNAS 17:836, 2020). This integration reporting on accumulated mRNA is now shown for all constructs in the presence and absence of Runt protein in the new Figure S17. This figure clearly shows that the consequences of repression are present in the blastoderm, not just at the level of transcriptional initiation, but also at the level of accumulated mRNA.

    We then compared the accumulated mRNA profiles shown in Figure S17 to the initial rate of RNAP loading at each position of the embryo along the anterior-posterior axis for all constructs in the presence and absence of Runt protein. These new results are shown in a new figure, Figure S19. Interestingly, we saw a good correlation (Pearson correlation coefficient of 0.90) between these two metrics. Thus, we argue that our conclusion that higher-order cooperativity is necessary to account for the initial rate of RNA polymerase loading would still hold for predicting the accumulated mRNA.

    Reviewer #3 (Public Review):

    The authors have presented results from carefully planned and executed experiments that probe enhancer-drive expression patterns in varying cellular conditions (of the early Drosophila embryo) and test whether standard models of cis-regulatory encoding suffice to explain the data. They show that this is not the case, and propose a mechanistic aspect (higher order cooperativity) that ought to be explored more carefully in future studies. The presentation (especially the figures and schematics) are excellent, and the narrative is crisp and well organized. The work is significant because it challenges our current understanding of how enhancers encode the combinatorial action of multiple transcription factors through multiple binding sites. The work will motivate additional modeling of the presented data, and experimental follow-up studies to explore the proposed mechanisms of higher order cooperativity. The work is an excellent example of iterative experimentation and quantitative modeling in the context of cis-regulatory grammar. At the same time, the work as it stands currently raises some doubts regarding the statistical interpretation of results and modeling, as outlined below.

    We thank the reviewer for noting the significance of our work. We tried our best to address the concerns of the reviewer regarding the statistical interpretation of results and theoretical modeling throughout our responses below.

    The results presented in Figure 5 are used to claim that the data support (i) an unchanging K_R regardless of the position of the Runt site in the enhancer and (ii) an \omega_RP that decreases as the site goes further away from the promoter, as might be expected from a direct repression model. This claim is based on only testing the specific model that the authors hypothesize and no alternative model is compared. For instance, are the fits significantly worse if \omega_RP is kept constant and the K_R allowed to vary across the three sites. If different placements of the Runt site can result in puzzling differences in RNAP-promoter interaction, it seems entirely possible that the different site placements might result in different K_R, perhaps due to unmodeled interference from bicoid binding. Due to these considerations, it is not clear if the data indeed argue for a fixed K_R and distance-dependent \omega_RP.

    We apologize for the lack of justification in assuming that Kr remains constant and wrp varies depending on the position of the Runt binding sites. Following the reviewer’s suggestion, we tested the alternative scenarios where we either fix or vary different combinations of wrp and Kr for our one-Runt binding site constructs. The result is now shown in a new figure, Figure S16. In short, as reported by the Akaike Information Criterion (AIC) in Figure S16F, the MCMC fit explains the data best in the scenario of fixed Kr and different wrp values for one-Runt binding site constructs. Furthermore, we also performed the MCMC inference in the case where we varied both Kr and wrp values across constructs. From this analysis, we obtained similar values of Kr while having different values of wrp across constructs as shown in Figure S16G. Overall, we believe that this evidence strongly supports our assumption of having consistent Kr values but different wrp values for the one-Runt binding site constructs.

    Results presented in Figure 6 make the case that higher order cooperativity involving two DNA-bound molecules of Runt and the RNAP is sufficient to explain the data. The trained values of such cooperativity in the three tested enhancers appear orders of magnitude different. As a result, it is hard to assess the evidence (from model fits) in a statistical sense. Indeed, if all of the assumptions of the model are correct, then using the high-order cooperativity is better than not using it. To some extent, this sounds statistically uninteresting (one additional parameter, better fits). It is not the case that the new parameter explains the data perfectly, so some form of statistical assessment is essential.

    The inferred cooperativity values are indeed orders of magnitude different. However, the cooperativity terms can be also written as “w = exp(-E/(kBT))”, where the E is the interaction energy, kB is the Boltzmann constant, and T is the temperature. As a result, we should compare the magnitude of the different cooperativities on a log-scale. In brief, the interaction energies wrr from the three two-Runt binding site constructs range between 0 and 1kBT, and the higher-order cooperativity wrrp has an energy between -2 and 4kBT. Interestingly, these energies are of the same order of magnitude as the interaction energies typically reported for bacterial transcription factors (e.g., Dodd et al., Genes and Development 18:344-54, 2004). It is important to note that our inferred interaction energies could be either positive or negative, suggesting that both cooperativity and anti-cooperativity can be at play depending on the architecture of the two Runt binding sites. We now report on these interactions in the language of energies Table S1 and elaborate on this in the Discussion section (Line XX-YY).

    Finally, following the reviewer’s suggestion on statistical assessment of whether addition of parameters indeed explains the data better, we adopted the Akaike Information Criterion (AIC) as a metric to compare different models used in Figure 6 and now show the results in a new panel, panel G. Briefly, AIC is calculated by assessing the model’s ability to explain the data while penalizing for having more parameters. The smaller the AIC value is, the better the model explains the data. As we have claimed, the AIC showed a dramatic decrease when adopting the higher-order cooperativity as shown in Figure 6G. Thus we argue that the addition of higher-order cooperativity, while not being able to completely explain the data, is indeed capable of increasing the agreement between experiments and theory across all our two-Runt site constructs.

    Moreover, it is not the case that the model structure being tested is the only obvious biophysics-driven choice: since this is the first time that such higher order effects are being tested, one has to be careful about testing alternative model structures, e.g., repression models that go beyond direct repression and pairwise cooperativity that goes beyond the traditional approach of a single (pseudo)energy term.

    We agree with the reviewer that alternative models with different mechanisms of repression should be mentioned. We have clarified this point further in Discussion (Line XX -YY). In summary, we tested both “competition” and “quenching” models of repression as proposed in Gray et al, (Genes and Development 8:1829, 1994). Interestingly, Figure S5 shows that the “competition” model gives a worse fit compared to the “direct repression” and “quenching” models for the one-Runt binding site cases. We further tried to test these alternative models in the case of two-Runt binding sites constructs. The result is shown in Figure S7 (competition) and S8 (quenching). These figures also reveal that the “competition” model underperformed compared to the “direct repression” or “quenching” models. For the “quenching” model to fit the data, we also had to invoke higher-order cooperativity that is beyond pairwise cooperativity. Thus, we believe that the requirement of higher-order cooperativity holds regardless of the choice of the specific model. Of course, our models of repression are very likely an oversimplification of how repressors actually work. However, given that these simple models have been a prevalent choice of proposed mechanisms for repression in the field of transcriptional repression for the past decades, we believe that the significance of our work lies in the fact that we challenged these models by turning them into precise mathematical statements (in the form of widespread thermodynamics models) and confronting them with quantitative data.

    The general theme seen in Figure 6 is seen again in Figure 7, when a 3-site construct is tested: model complexities inferred from all of the previous analyses are insufficient at explaining the new data, and new parameters have to be trained to explain the results. The authors do not seem to claim that the higher order cooperativity terms (two parameters) explain the data, rather that such terms may be useful.

    We agree that our previous approach was confusing. Figure 7A indeed incorporated all inferred parameters from the previous rounds of inference (Kb, wbp, p, R, as well as Kr, wrp, wrr, and wrrp). However, it is clear that this set of parameters, even including the higher-order cooperativity from two-Runt binding sites cases, was not enough to explain the data from three-Runt binding sites case. Thus, we had to invoke another free parameter, which we termed wrrrp, to explain the data. We have revised Figure 7B such that it is now showing the “best” MCMC fit which explains the data quite well (instead of just showing the “improvement” of fits).

  3. eLife

    eLife assessment

    The work by Kim et al., used synthetic constructs in Drosophila to examine the relationship between regulators (activator/repressor) and transcription initiation. By measuring regulator concentrations and the corresponding RNA polymerase initiation rates in different synthetic constructs and using a thermodynamic model, the authors concluded that that higher-order cooperativities between the repressor on adjacent binding sites, and that between the repressor and RNA polymerase are needed to explain the observed response curves in RNA polymerase loading rate. This work targets a challenging question in eukaryotic transcription regulation, where higher-order cooperativity between different molecular components, in addition to simple transcription factor binding and unbinding, is often necessary to account for observed promoter behaviors when multiple elements (repressors, mediators, activators) exist.

  4. eLife

    Reviewer #1 (Public Review):

    In this manuscript by Kim et al., the authors use live-cell imaging of transcription in the Drosophila blastoderm to motivate quantitative models of gene regulation. Specifically, they focus on the role of repressors and use a 'thermodynamic' model as the conceptual framework for understanding the addition and placement of the repressor Runt, i.e. synthetic insertion of Runt repressor sites into the Bicoid-activated hunchback P2 enhancer. Coupled with kinetic modeling and live-cell imaging, this study is a sort of mathematical enhancer bashing experiment. The overarching theme is measuring the input/output relationship between an activator (bicoid), repressor (runt), and mRNA synthesis. Transcriptional repression is understudied in my opinion. One finding is that the inclusion of cooperativity between trans-acting factors is necessary for understanding transcriptional regulation. Most, if not all, of the tools used in this paper have been published elsewhere, but the real contribution is a deep, quantitative dissection of transcriptional regulation during development. As such, the only real questions for this referee are whether the modeling was done rigorously to produce some general biological conclusions. By and large, I think the answer is yes.

    Comments:

    Fig. 6 was the most striking figure for this referee, specifically that different placements of Runt molecules on the enhancer lead to distinct higher order interactions. I am wondering if the middle data column in Fig. 6 represents a real difference from the other two, and if so, it seems that the positioning - as opposed to simply the stoichiometry - is essential in cooperativity. This conclusion implies that transcriptional regulation is more precise than what some claim is just a mushy ball of factors close to a promoter. In other words, orientation may matter. Proximity may matter. Interactions in trans matter.

    There needs to be at least one prediction which is validated at the level of smFISH / mRNA in the embryo. Without detracting from the effort the authors have expended in looking directly at transcription, if the effects can't be felt by the blastoderm at the level of mRNA/cell, it become difficult to argue for the relevance to development. Also, I feel there is little chance that these measurements can be quantitatively replicated unless translated to the level of total protein or mRNA. Such a measurement (orthogonal quantitative confirmation of the repressor cooperativity result) would also assuage my concern about the time averaging as shown in Fig. S3.

  5. eLife

    Reviewer #2 (Public Review):

    In this paper, eGFP: LlamaTag-Runt was inserted into Drosophila embryo cells by CRISPR-Cas9 technology, and quantitative gene expression and time-lapse measurements were performed. The molecular mechanism was modeled and analyzed by thermodynamic model, the experimental data were fitted by MCMC, and the necessity of cooperation was given.

  6. eLife

    Reviewer #3 (Public Review):

    The authors have presented results from carefully planned and executed experiments that probe enhancer-drive expression patterns in varying cellular conditions (of the early Drosophila embryo) and test whether standard models of cis-regulatory encoding suffice to explain the data. They show that this is not the case, and propose a mechanistic aspect (higher order cooperativity) that ought to be explored more carefully in future studies. The presentation (especially the figures and schematics) are excellent, and the narrative is crisp and well organized. The work is significant because it challenges our current understanding of how enhancers encode the combinatorial action of multiple transcription factors through multiple binding sites. The work will motivate additional modeling of the presented data, and experimental follow-up studies to explore the proposed mechanisms of higher order cooperativity. The work is an excellent example of iterative experimentation and quantitative modeling in the context of cis-regulatory grammar. At the same time, the work as it stands currently raises some doubts regarding the statistical interpretation of results and modeling, as outlined below.

    The results presented in Figure 5 are used to claim that the data support (i) an unchanging K_R regardless of the position of the Runt site in the enhancer and (ii) an \omega_RP that decreases as the site goes further away from the promoter, as might be expected from a direct repression model. This claim is based on only testing the specific model that the authors hypothesize and no alternative model is compared. For instance, are the fits significantly worse if \omega_RP is kept constant and the K_R allowed to vary across the three sites. If different placements of the Runt site can result in puzzling differences in RNAP-promoter interaction, it seems entirely possible that the different site placements might result in different K_R, perhaps due to unmodeled interference from bicoid binding. Due to these considerations, it is not clear if the data indeed argue for a fixed K_R and distance-dependent \omega_RP.

    Results presented in Figure 6 make the case that higher order cooperativity involving two DNA-bound molecules of Runt and the RNAP is sufficient to explain the data. The trained values of such cooperativity in the three tested enhancers appear orders of magnitude different. As a result, it is hard to assess the evidence (from model fits) in a statistical sense. Indeed, if all of the assumptions of the model are correct, then using the high-order cooperativity is better than not using it. To some extent, this sounds statistically uninteresting (one additional parameter, better fits). It is not the case that the new parameter explains the data perfectly, so some form of statistical assessment is essential. Moreover, it is not the case that the model structure being tested is the only obvious biophysics-driven choice: since this is the first time that such higher order effects are being tested, one has to be careful about testing alternative model structures, e.g., repression models that go beyond direct repression and pairwise cooperativity that goes beyond the traditional approach of a single (pseudo)energy term.

    The general theme seen in Figure 6 is seen again in Figure 7, when a 3-site construct is tested: model complexities inferred from all of the previous analyses are insufficient at explaining the new data, and new parameters have to be trained to explain the results. The authors do not seem to claim that the higher order cooperativity terms (two parameters) explain the data, rather that such terms may be useful.

  7. Version published to 10.1101/2021.07.28.454075v1 on bioRxiv