1. Developmental Biology

Embryonic geometry underlies phenotypic variation in decanalized conditions

  1. Anqi Huang
  2. Jean-François Rupprecht
  3. Timothy E Saunders  Is a corresponding author
  1. Mechanobiology Institute, National University of Singapore, Singapore
  2. CNRS and Turing Center for Living Systems, Centre de Physique Théorique, Aix-Marseille Université, France
  3. Department of Biological Sciences, National University of Singapore, Singapore
  4. Institute of Molecular and Cell Biology, Proteos, A*Star, Singapore
https://doi.org/10.7554/eLife.47380
Share this article
Cite this article
  1. Anqi Huang
  2. Jean-François Rupprecht
  3. Timothy E Saunders
(2020)
Embryonic geometry underlies phenotypic variation in decanalized conditions
eLife 9:e47380.
https://doi.org/10.7554/eLife.47380
  2. Download BibTeX
  3. Download .RIS

Abstract

During development, many mutations cause increased variation in phenotypic outcomes, a phenomenon termed decanalization. Phenotypic discordance is often observed in the absence of genetic and environmental variations, but the mechanisms underlying such inter-individual phenotypic discordance remain elusive. Here, using the anterior-posterior (AP) patterning of the Drosophila embryo, we identified embryonic geometry as a key factor predetermining patterning outcomes under decanalizing mutations. With the wild-type AP patterning network, we found that AP patterning is robust to variations in embryonic geometry; segmentation gene expression remains reproducible even when the embryo aspect ratio is artificially reduced by more than twofold. In contrast, embryonic geometry is highly predictive of individual patterning defects under decanalized conditions of either increased bicoid (bcd) dosage or bcd knockout. We showed that the phenotypic discordance can be traced back to variations in the gap gene expression, which is rendered sensitive to the geometry of the embryo under mutations.

Introduction

The phenomenon of canalization describes the constancy in developmental outcomes between different individuals within a wild-type species growing in their native environments (Flatt, 2005; Hallgrimsson et al., 2019; Waddington, 1942). To better understand how canalized a developmental process is, we need to quantitatively measure the molecular profiles of the developmental regulatory genes in multi-cellular organisms. In Drosophila, the highly reproducible body patterning in adult flies originates from both the reproducible setup of the instructive morphogen gradients and the precise downstream transcriptional readouts early in embryogenesis (Bollenbach et al., 2008; Gregor et al., 2007a; Petkova et al., 2014). In particular, the inter-individual variation of the positional information conferred by gene expression can be less than the width of a single cell (Dubuis et al., 2013; Gregor et al., 2007a); that is these processes are highly canalized. Given the ubiquity of canalization in nature, such highly reproducible developmental processes are likely not exclusive to insect development.

The developmental canalization that we see in contemporary species is the product of evolution, either as the consequence of stabilizing selection (Gibson and Wagner, 2000) or the manifestation of the intrinsic properties of the underlying complex gene regulatory networks (Siegal and Bergman, 2002). Canalization can break down in individuals subjected to aberrant genetic mutations or extreme environmental conditions (Imasheva et al., 1997; Rutherford and Lindquist, 1998). Such individuals in decanalized conditions become sensitive to variations in their genetic background and external environments, which are otherwise neutral to developmental outcomes. This leads to significantly increased inter-individual variation in phenotypic outcomes.

It is important to characterize the sources of variation in order to understand what canalization is actually buffering against. Interestingly, significant phenotypic variation remains in laboratory animals with isogenic genomes, exposed to homogeneous environments (Gärtner, 1990). This indicates that under decanalized conditions, other components besides genetic and environmental variation may also cause phenotypic discordance. Previous work has proposed that stochastic expression of redundant genes predicts the developmental outcome of the mutant individuals (Burga et al., 2011). However, in many other cases, it remains elusive as to why mutation increases inter-individual phenotypic variation and what alternative components underlie such variation (Janssens et al., 2013; Surkova et al., 2013).

To identify potential sources of variation that govern phenotypic discordance, we utilized the highly canalized process of early Drosophila embryonic AP patterning. We proposed that the inter-individual variation of the embryonic geometry is one of the factors that predetermine individual patterning outcomes under decanalizing mutations. We found that the patterning system in genetically intact embryos has a striking capacity to buffer against variations in embryonic geometry. When we artificially reduced the aspect ratio of the embryos by more than twofold, the relative boundary positions of segmentation genes showed only a minor shift compared to that of the wild type. The preservation of the boundary positions can be, at least partially, attributed to changes in the Bcd gradient profile due to the rounder embryonic shape in the shorter embryos. Importantly, the inter-individual variations of boundary positions remained comparable to that of the wild type. In contrast, when we introduced decanalizing mutations such as increased maternal bcd dosage or bcd knockout, the embryonic geometry becomes predictive of patterning outcomes of individuals. In both scenarios, the inter-individual phenotypic discordance can be traced back to variations in gap gene expression patterns. This suggests that under decanalized conditions particular gene interactions are preferentially affected by geometric constraints. Taken together, our study reveals embryo geometry, or more generally the physical domain in which patterning occurs, is a significant source of variation that can account for phenotypic discordance under decanalized conditions.

Results

Embryos with dramatically reduced embryonic length still proceed with normal development

Scaling is one of the most astonishing features of embryonic patterning in dipteran insects. The expression boundaries of the early patterning genes reside at conserved relative positions among closely related dipteran species, although they differ dramatically in egg size (Gregor et al., 2005). Such inter-species scaling can be attributed to the Bcd gradient of a conserved characteristic length (Gregor et al., 2008). Within the Drosophila species, individual embryos of different size also exhibit scaled pattern as a consequence of adapted Bcd production and degradation rate (Cheung et al., 2011; Cheung et al., 2014). We propose here that another macroscopic variable that the patterning system needs to adapt to in order to ensure scaling is the geometry, or shape of the egg. Given a defined embryonic volume, the embryos can vary significantly in their aspect ratio (AR = embryonic length/embryonic width, Figure 1A). Since Bcd molecules diffuse mainly at the cortical region of the embryo (Gregor et al., 2007b), varying embryonic geometry affects the effective diffusion area, and thus the length and time scales of the morphogen gradient formation (Grimm et al., 2010). Therefore, we first examined how altering embryonic geometry affects embryonic patterning.

Figure 1 with 1 supplement see all
Download asset Open asset
The Bcd gradient in embryos with differing geometry is consistent with the SDD model.

(A) Distribution of embryonic length in wild type (n = 239) and fat2RNAi (n = 364) embryos. (B) Distribution of embryonic width in wild type (n = 239) and fat2RNAi (n = 364) embryos. (C) Aspect ratio (EL/EW) against embryonic length for each embryo. Black dots denote expected aspect ratio if embryonic volume is conserved (assuming ellipsoidal geometry). (D) Number of nuclei along the AP axis plotted against embryonic length in wild type and fat2RNAi embryos. Line indicates linear regression of all data. (E) En expression in fat2RNAi embryos showing defective dorsal closure (left) or normal morphogenesis (right). Arrows indicate locations of defects. Green dots indicate En stripes. (F) Midsagittal plane of ctrl (top) and fat2RNAi (bottom) embryos expressing eGFP::Bcd in mid n.c. 14. (G) eGFP::Bcd profiles of both ctrl and fat2RNAi embryos plotted as a function of absolute distance from the anterior pole (left) or scaled AP position (right). Each dot represents the average concentration in a single nucleus. Colormap indicates the absolute AP length of each individual. (H) Mean and standard deviation of nuclear intensity within each 2% EL were computed for group of embryos longer (red, n = 27) and shorter (blue, n = 17) than 450 µm. Inset is close-up of profile near embryo midpoint to show the intersection of the two curves. (I) Representative fluorescent in situ hybridization (FISH) against bcd mRNA in wild type (top) and fat2RNAi (bottom) in n.c. 11 embryos. Scale bar, 50 μm. (J) Fluorescent intensity profile of FISH assay along AP axis in n.c. 4 (top) and n.c. 11 (bottom). Normalization to measured fluorescence signal in the region 120 μm from anterior. n = 5, 2 (n.c. 4) and n = 2,2 (n.c. 11) for wild type (red) and fat2RNAi (blue) respectively. Error bars show standard deviation. (K) Fitting of SDD model to experimentally measured Bcd gradient. All parameters, as outlined in Materials and methods, are kept constant, with only change being embryonic geometry. See Materials and methods for details. Lower panel shows intensity difference in experimental measurements and predicted profiles along the AP-axis.

The geometry of each embryo is predetermined during oogenesis when the follicle cells surrounding the egg chamber transform the developing egg from a sphere to an ellipse (Haigo and Bilder, 2011; He et al., 2010b). This process is mediated by the planar cell polarity of the follicle cells and the elliptical shape of the embryos remains unchanged throughout embryogenesis. Here, we used maternal ShRNA to knockdown one of the planar cell polarity core components, atypical cadherin Fat2 specifically in the follicle cells (Horne-Badovinac et al., 2012), and we henceforth refer to these as fat2RNAi embryos. As Fat2 expression is only inhibited within the somatic cells of the egg chamber using traffic jam (tj) >Gal4 driver, the fertilized eggs produced by this perturbation remain genetically intact. This reduces the embryonic length from 510 ± 17 (s.d.) µm in wild type to 432 ± 40 µm in fat2RNAi embryos (Figure 1A), with some embryos as short as 320 μm. Meanwhile, the perturbed embryos show an increased embryonic width (EW) along the dorso-ventral (DV) axis (Figure 1B, 196 ± 5 µm in wild type and 202 ± 11 µm in fat2RNAi). Together, these geometrical variations lead to only a slight reduction of ~8% in the embryonic volume (assuming an ellipsoidal geometry) compared to wild type embryos (Figure 1C; Figure 1—figure supplement 1A–C). The round eggs are fertilizable and continue with proper embryogenesis (Figure 1—figure supplement 1D; Video 1).

Video 1
Download asset
Embryogenesis of different geometry.

Wide field movies of wild type (top) and fat2RNAi (bottom) embryos from the onset of gastrulation until hatching.

We examined the nuclear distribution in the blastoderm embryos, as nuclei are the basic units interpreting positional information, and an altered nuclear distribution may affect patterning outcomes. We found that nuclear number along the AP axis decreases proportionally to embryonic length, leaving the inter-nuclear distance unchanged (Figure 1D; Figure 1—figure supplement 1E). In other words, the number of nuclei to interpret AP positional information reduces from 85 ± 4 in wild type to 65 ± 5 in fat2RNAi embryos.

Despite the significant changes in embryonic geometry, the fat2RNAi embryos show invariably eleven Engrailed (En) stripes, demarcating the posterior boundary of each body segment including three thoracic (T1 ~T3) and eight abdominal segments (A1 ~A8; Figure 1E). We noticed that fat2RNAi embryos shorter than 400 µm developed morphological defects in late embryogenesis, where abnormal dorsal closure leads to mismatch between the two lateral sides of the ectoderm (Figure 1E, arrows; Figure 1—figure supplement 1D). However, such local morphological abnormality is likely due to defective tissue morphogenesis as a consequence of limited physical space, rather than patterning errors (Video 2).

Video 2
Download asset
Defective morphogenesis in late embryo development due to extreme embryonic geometry.

Confocal movies of fat2RNAi embryos expressing en >mCD8::GFP from germband retraction to dorsal closure stages. Two embryos in the movie represent individuals with EL shorter (top) and longer (bottom) than 400 µm. The dashed box indicates mismatch between opposing ectodermal tissue during dorsal closure. The numbers indicate the EL of each embryo.

Developmental reproducibility is preserved with a minor impact on scaling under geometrical perturbations

Before tackling the consequences of decanalization, we next ask how the AP patterning network, from the initial morphogen gradient formation to gap gene readout, is altered under large geometric perturbations. First, we focus on Bcd, which provides AP positional information in the early embryo. To compare the spatial Bcd concentration profile in different embryonic geometry, we live-imaged eGFP-Bcd fusion protein in both control and fat2RNAi backgrounds, and measured the nuclear Bcd intensity along the AP axis at mid nuclear cycle (n.c.) 14 (Figure 1F). We found that the absolute Bcd concentration is lower in the anterior half of the embryo in fat2RNAi individuals compared to that of control (Figure 1G and H) (p<10−3, two-sided t-test comparing intensity in region 5–15% EL). Further, Bcd profiles from different individuals intersect near the mid region of the embryo, with shorter embryos showing higher Bcd concentration in the posterior region (Figure 1G and H) (p<10−3, two-sided t-test comparing intensity in region 65–75% EL).

The formation of the Bcd gradient depends on the amount and the spatial distribution of bcd mRNA molecules (Little et al., 2011; Cheung et al., 2014). To examine bcd mRNA distribution under geometrical perturbations, we performed fluorescent in situ hybridization (FISH) against bcd mRNA in both wild type and fat2RNAi embryos (Figure 1I; Figure 1—figure supplement 1F). We found that at all stages prior to n.c. 12, the majority of the bcd mRNA is localized within the anterior 50 µm of the embryo, regardless of the embryonic geometry (Figure 1J). Due to the difficulty in quantifying FISH data, we note that we cannot discount possible variation in the bcd mRNA amount between wild type and fat2RNAi embryos. However, such variations are likely small and do not substantially alter the overall distribution of bcd mRNA. These results are consistent with previous work (Cheung et al., 2011), which suggested that embryo volume was a key determinant of bcd mRNA levels; in the fat2RNAi embryos, volume does not substantially change. Importantly, the geometrical alterations induced in fat2RNAi embryos does not change the bcd mRNA distribution on an absolute scale.

Given that the bcd mRNA distribution does not appear substantially different between the wild type and fat2RNAi embryos, what is the cause of the spatial shift in the Bcd profile? We used theoretical modeling to test whether the geometrical changes alone (Umulis and Othmer, 2012) are sufficient to explain the altered Bcd profile. The synthesis , diffusion , degradation (SDD) model of Bcd gradient formation provides an excellent description of the Bcd gradient (Gregor et al., 2007b; Durrieu et al., 2018; Little et al., 2011). A one-dimensional description generally works well for modelling the Bcd gradient (Gregor et al., 2007b; Durrieu et al., 2018). However, the full three-dimensional geometry of the embryo can be important (Mogilner and Odde, 2011), such as in interpreting FRAP (Castle et al., 2011) measurements and explaining differences in the gradient between dorsal and ventral sides of the embryo (He et al., 2010a; Hengenius et al., 2011). Here, we perform our modeling in three-dimensions as our experimental perturbation affects the three-dimension geometry of embryos.

Considering the Bcd concentration [Bcd], Bcd diffusion (D), Bcd degradation rate (μ) and Bcd production rate (J), the SDD model is described by

∂tBcd=D∇2Bcd-μBcd+Jfx

We use Bcd dynamic parameters D = 4 μm2s−1 and μ = 0.0005 s−1 consistent with the most recent estimates for Bcd dynamics (Durrieu et al., 2018). Note these parameters are based on an one-dimensional SDD model, as such models are typically a reasonable approximation to the formation of the Bcd gradient (Little et al., 2011). f(x) is defined such that little Bcd production occurs greater than 50 μm from the anterior pole (Materials and methods). We solve for [Bcd] on the surface of an ellipsoid, as Bcd transport through the yolk appears to be limited (Gregor et al., 2007b). For each simulation, we keep all dynamic parameters fixed and only change the embryo geometry. In particular, we assume that the Bcd production rate is unchanged in fat2RNAi embryos. Since the embryo volume does not markedly change, this means that the total amount of Bcd is similar in the model for both wild-type and fat2RNAi embryos. We change the width to maintain the experimentally measured aspect ratios (Figure 1C) and account for eGFP folding time (~50 min) (Liu et al., 2013). See Materials and methods for detailed description.

The SDD model can fit the observed Bcd gradient profiles in different embryonic geometries without requiring any change to the Bcd dynamics or production (Figure 1K). As a consequence, we can explain the above observations. First, the lower Bcd intensity near the anterior pole in fat2RNAi embryos is because of dilution due to the larger DV extent in rounder embryos. If the total Bcd molecule number is similar at a particular AP position, the measured Bcd nuclear concentration (which depends on both total protein number and the local volume) at that position is lower in fat2RNAi embryos. Second, since the diffusion coefficient appears to remain unchanged, there is greater accumulation of Bcd in the posterior region of fat2RNAi embryos as the total travel distance is reduced in the rounder, shorter embryos, even when accounting for the higher curvature in fat2RNAi embryos. The ability of the SDD model to explain these differences in the Bcd profile simply by accounting for embryo geometry provides further support to the SDD model being an excellent biophysical description of the process of Bcd gradient formation. It is worth noting that the nuclei respond to the local concentration of mature Bcd protein regardless of the folding state of the tagged eGFP. Although we cannot infer the precise profile of mature Bcd from our experimentally measured eGFP profile as the Bcd protein folding rate is unknown, our model can test the effects of different protein folding time on the functional gradient in different geometries. From the model shown in Figure 1K, we can infer the total Bcd concentration, where we assume Bcd folds at a much faster rate than eGFP (Figure 1—figure supplement 1G). The predicted total Bcd concentration displays qualitatively similar behavior in varying embryonic geometries to our experiments; that is our results are unlikely an artefact of protein folding time differences between Bcd and eGFP.

Does the change of the Bcd concentration profile in fat2RNAi embryos impact on the scaling of downstream patterning genes? To address this question, we measured Hunchback (Hb) expression in mid n.c. 14 using live imaging of hb >LlamaTag (Bothma et al., 2018), (Figure 2A). In control embryos, the Hb expression boundary locates at 49.0% EL with variation of 1.3% EL, consistent with previous reports (Houchmandzadeh et al., 2002). Comparatively, the Hb boundary shows a posterization in fat2RNAi embryos (52.9% EL). Further, we found an increased variation of 2.3% EL in the Hb boundary position (Figure 2B). However, considering the absolute length of fat2RNAi embryos, the variation of Hb boundary positions (~8 µm) between different individuals is still less than the average distance between neighboring nuclei.

Figure 2
Download asset Open asset
Embryonic patterning is robust to perturbation of embryonic geometry.

(A) Max projection of ctrl (top) and fat2RNAi (bottom) embryos expressing maternally loaded eGFP and hb >LlamaTag. (B) Scaled Hb boundary positions plotted against EL in two genotypes. (C–F) Comparison of gap gene expression between wild type and fat2RNAi embryos. Profiles normalized to max intensity and the computed mean and s.d. plotted against scaled AP length. Boundary positions of each individual is plotted to show the distribution in two genotypes (WT, red and fat2RNAi, blue). Scale bars, 100 µm.

Using immunofluorescence, we next investigated the impact of geometrical perturbations on the expression domains of other gap genes. In agreement with our results for Hb, the boundary positions of all four gap genes displayed slight shifts in the posterior direction in fat2RNAi embryos (Figure 2C–F). However, the inter-individual variation remains comparable to that of the wild type (Figure 2C–F, bottom row). We conclude that when we manipulate the embryonic geometry to an extent beyond that naturally observed, the reproducibility of the patterning outcomes is preserved. Therefore, the intact early embryonic patterning network is highly robust to variations in embryonic geometry.

Embryonic length predetermines patterning outcomes in decanalized conditions of increased bcd dosage

We have demonstrated that fat2RNAi embryos provide an excellent system for testing the role of geometry on patterning networks. Taking advantage of the availability of genetic manipulations in Drosophila, we now use this tool to explore the role of geometric constraints in determining phenotypic outcomes in decanalized conditions: first, in embryos with increased maternal bcd gene dosage; and second in embryos with depleted maternal bcd.

Phenotypic discordance has been observed previously as a consequence of artificially altered maternal bcd dosage. Gradual increase of the maternal bcd gene copies leads to a larger proportion of individuals showing defective patterning (Namba et al., 1997). We wanted to test our hypothesis that the embryonic geometry predetermines patterning outcomes of these individuals. To efficiently increase the Bcd gradient amplitude, we generated a tandem bcd construct, where two copies of the bcd gene are linked by the P2A self-cleaving peptide (Figure 3A). Two transgenic fly lines with two and four genomic insertions of this construct deposit bcd mRNA into embryos at ~3 (6x bcd) and ~5 (10x bcd) fold wild-type amounts, respectively (Figure 3B; Figure 3—figure supplement 1A–C). As Bcd protein counts scale linearly with that of its mRNA, the corresponding amplitude of the Bcd gradient are expected to show the same fold changes (Petkova et al., 2014), as manifested by the posterior displacement of cephalic furrow position (Figure 3B).

Figure 3 with 1 supplement see all
Download asset Open asset
Defects due to decanalization by Bcd over-expression are length-dependent.

(A) Schematic illustration of tandem bcd construct. (B) Embryos expressing 2x (wild type), 6x and 10x of maternal bcd fixed at the onset of gastrulation and stained with Phalloidin. Green arrowheads indicate the positions of cephalic furrow (CF) formation. (C) Distribution of defective cuticular segments in non-hatched 10xbcd (top), fat2RNAi, 6xbcd (mid) and fat2RNAi (bottom) embryos. Colormap indicates the frequency of defects in each segmental region. (D) Bar plots showing the distribution of defective cuticular segments in three genotypes. (E) En expression patterns in 4x (top), 6x (mid) and 10x (bottom) bcd embryos with different EL. Green dots, normal En stripes; red dots, defective En stripes; A, anterior; P, posterior. (F) Representative 6xbcd embryos with different EL expressing en >mCD8::GFP. Numbers indicate En stripe identities and red mask indicate defective segmental regions. Scale bar, 100 µm.

The ~5 fold bcd overexpression compromises viability to adulthood (Figure 1—figure supplement 1D) and the non-hatched embryos displayed a plethora of defective patterning phenotypes (Figure 3C; Figure 3—figure supplement 1D). Individuals with mild defects frequently displayed missing or fused denticle belts in the fourth abdominal (A4) segment (Figure 3C–D), a positional bias that has been reported previously (Namba et al., 1997). More severe phenotypes showed defects in a spreading region centered about the A4 segment. Meanwhile, embryos show high rate of mouth defects as a consequence of significantly increased local Bcd concentration in the most anterior region (Figure 3C–D; Figure 3—figure supplement 1D). Patterning defects were rarely seen in 3-fold bcd over-expression embryos unless fat2RNAi knockdown is further introduced into this genetic background (Figure 3C–D; Figure 3—figure supplement 1E). A large percentage of these individuals showed abdominal patterning defects, with A4 showing the highest defective frequency (Figure 3C–D). Interestingly, a similar distribution of defective abdominal segments is also seen in the small proportion of non-hatched fat2RNAi individuals (Figure 3C–D).

To understand if embryonic geometry predetermines the severity of phenotypic defects in individuals, we characterized the patterning outcomes using En expression in embryos with various bcd copy number and embryonic length. Individuals with 2-fold bcd overexpression (single insertion of tandem bcd), within the natural range of embryonic geometry, showed intact En expression. However, shorter embryos (<450 µm) frequently presented patterning defects, most commonly in the 6th En stripe (Figure 3E, top panel). Interestingly, this position corresponds to the A4 segment in the cuticle pattern. Comparatively, patterning defects become more pervasive in 3-fold bcd over-expression individuals when embryonic length reduced below 470 µm. The range of defective segments gradually expands from the 6th En stripe to both anterior and posterior regions with decreasing embryonic length (Figure 3E, middle panel and 3F). Further, increasing bcd dosage 5-fold renders patterning processes exceedingly susceptible to reduced embryonic length. Defects are observed in comparatively shorter individuals within the natural range and recurringly the 6th En stripe is the most frequent breaking point in the patterning (Figure 3E, bottom panel).

The A4 segment is a weak point in the gap gene network due to repression of eve stripe five in short embryos

The defective abdominal patterning that we observe here is an intuitive result, as both the posterization of gap gene boundaries due to increased bcd dosage and reduced embryonic length lead to reduced number of nuclei along the AP axis in the trunk region. When the number of nuclei falls short of the minimal requirement to fulfill all the different cell identities along the AP axis, certain cell fates become lost. It is surprising, however, that this defect originates at and expands from the same position in all defective embryos, the A4 segment. This positional bias is also reflected in the segmentation gene pattern at the blastoderm stages. While the gap gene boundaries remain roughly at the same scaled positions across different geometries (Figure 4A), the absolute distance between neighboring gap gene expression peaks decreases in response to reduced embryonic length (Figure 4B). This in turn changes the combinatorial inputs to activate downstream pair-rule genes, for example even-skipped (eve). The expression peaks of Kni and Gt are brought into proximity with gradually reducing embryonic length (Figure 4B–C). As Kni and Gt confine the boundaries of eve stripe 5 (Fujioka et al., 1999), the expression of this eve stripe is over-repressed (Figure 4C, asterisk). This results in the loss of correct cell fate at this position, corresponding to the future A4 segment. With further reduced embryonic length, a larger percentage of individuals fail to activate Kni and Gt in the trunk region (Figure 4B, green crosses; Figure 4C, arrowhead), leading to defects across a broader range.

Figure 4
Download asset Open asset
Embryonic patterning breaks down at A4 segment with bcd overexpression.

(A–B) Embryonic length of 6xbcd individuals plotted against the scaled (A) or absolute (B) AP position of four gap gene boundaries (Hb, blue; Gt, magenta; Kr, red and Kni, green). Data from every 30 µm EL interval were binned to compute mean and s.d. and the colored areas are generated by connecting mean values of different EL ranges. Dashed line indicates posterior boundary of individuals; green and magenta crosses overlapping the dashed line indicate individuals with corresponding EL not expressing Kni and Gt, respectively. (C) Representative segmentation gene expression in 6xbcd embryos with different EL. Asterisk indicates repressed Eve stripe five and arrowhead indicates failed activation of Kni. Scale bar, 100 µm.

AP patterning of bcd mutants correlates with embryonic length

We have shown that embryonic geometry predicts individual patterning outcomes under increased bcd dosage. To understand if embryonic geometry is a general factor underlying phenotypic discordance in decanalized conditions, we asked how geometrical perturbation influences phenotypic outputs in the absence of Bcd inputs. This question is motivated by the significant phenotypic variation observed among embryos derived from females carrying the same bcd null allele (Frohnhöfer and Nüsslein-Volhard, 1986; Frohnhofer and Nusslein-Volhard, 1987). For example, the individuals derived from bcdE1 homozygous females show highly variable pattern in the perspective A1 to A5 segments, manifesting in either fusion or depletion of various number of denticle belts.

To systematically understand the inter-individual phenotypic variation among bcd mutants, we utilized an allele generated by the CRISPR-MiMIC method (Huang et al., 2017; Venken et al., 2011). The MiMIC transposon carrying stop codons in all three reading frames is targeted by CRISPR to insert into the first intron of the endogenous bcd gene. Therefore, no functional Bcd protein is produced by this knockout allele (annotated bcdKO). Moreover, the MiMIC construct contains an eGFP marker, which facilitates further genetic manipulation, such as recombination, carried out in this study.

The cuticular pattern of the bcdKO allele qualitatively recapitulates that of bcdE1 (Figure 5—figure supplement 1A–C). While the anterior embryonic patterning is entirely defective, the patterning defects in posterior regions are more variable. The number of normal abdominal denticle belts in each embryo ranges from three (A6 - A8) to seven (A2 - A8), with four intact abdominal segments (A5 –A8) being the most frequently observed phenotype (Figure 5A). Structures indicating partially differentiated abdominal segments can be observed in the anterior regions of the bcdKO embryos. However, these structures do not recapitulate any of the wild-type denticle belts (Figure 5—figure supplement 1B). Further, we observed and classified the variable phenotypes of the duplicated posterior spiracles according to the completeness of the organ morphogenesis (Figure 5B; Figure 5—figure supplement 1D). Interestingly, the fully developed ectopic spiracle organ (Figure 5—figure supplement 1D (i)) is only observed in embryos showing less than six intact abdominal segments, while most frequently observed in individuals with three intact abdominal segments (Figure 5A, dashed bar area). This indicates correlation between the patterning outcomes of the abdominal regions and the ectopic spiracles.

Figure 5 with 3 supplements see all
Download asset Open asset
bcd mutant phenotypes correlate with embryonic length.

(A) Phenotypic frequency showing different number of normal abdominal segments in bcdKO mutant individuals (n = 202). Dashed area indicates proportion of embryos showing fully developed ectopic spiracles (see Figure 5—figure supplement 1D(i)). (B) Phenotypic frequency of different ectopic spiracle morphology as shown in Figure 5—figure supplement 1D (n = 168). (C) Variation of gap gene boundary positions in wild type vs. bcdKO embryos. Error bars are computed by bootstrapping with data shown in Figure 5—figure supplement 1E–H. (D) Representative cuticle patterns of fat2RNAi, bcdKO embryos within different ranges of embryonic length, from left to right, 330–360, 360–390, 390–420, 420–450, and 450–480 µm. Stars indicate normal abdominal segments. (E) Number of normal abdominal segments plotted against EL range of each individual. (F) En expression in wild type embryo. Green dots indicate En stripes. (G) Representative images of fat2RNAi, bcdKO embryos showing different number of En stripes. Magenta dots indicate En stripes. Scale bar, 100 µm. (H) Number of En stripes plotted vs. EL in individuals from three genotypes. Magenta triangles indicate individuals showing defective morphogenesis at the end of dorsal closure; cyan triangles indicate normal morphogenesis.

The wide spectrum of bcdKO phenotypes can be attributed to variations in patterning gene expression during the blastoderm stage. By the end of the blastoderm stage, the relative boundary positions of all gap gene expression domains show significant anterior shift compared to those in the wild-type embryos (Figure 5—figure supplement 1E–H). Importantly, the absence of Bcd activity results in significantly increased variation in gap gene boundary positions (Figure 5C, p<0.01 for all measured boundaries to have increased error randomly). Occasionally we detected no Krüppel (Kr) nuclear intensity in the presumed expression region in bcdKO individuals (2 out of 15 individuals), indicating the failure of Kr gene activation in these embryos (Figure 5—figure supplement 2A). The anterior Gt domain shows similar inter-individual variation, with the majority of the embryos failing to properly activate anterior Gt expression (8 out of 10 individuals). Instead only a thin stripe of diminished cytoplasmic signal can be detected (Figure 5—figure supplement 1H, asterisk; Figure 5—figure supplement 2B). It is noteworthy that embryos derived from a single pair of bcdKO parents raised in constant environmental conditions show equivalent phenotypic variation (Figure 5—figure supplement 2C–D), suggesting that the inter-individual variation observed cannot be attributed to differences in either environment or genetic background.

Next, we introduced the fat2RNAi knockdown into a bcdKO background to see how embryonic patterning is affected. The cuticle pattern of embryos derived from fat2RNAi, bcdKO females resembles that of bcdKO alone, but the number of properly patterned abdominal denticle belts reduces with decreasing embryonic length. Moreover, novel phenotypes showing only one or two abdominal segments were observed when the embryonic length drops beyond the natural range (Figure 5D and E). All of the fat2RNAi, bcdKO embryos showed duplicated spiracles with fully developed morphology (see Figure 5—figure supplement 1D (i)), consistent with our previous result that such spiracles prevail in embryos with shorter AP length. We observed similar behavior in the pattern of En expression. In contrast to invariable eleven En stripes in control embryos (Figure 5F), the number of En stripes shows positive linear correlation with embryonic length in bcdKO embryos, both in individuals with natural embryonic geometry (Figure 5—figure supplement 3; Video 3) and fat2RNAi knockdown (Figure 5G and H).

Video 3
Download asset
Engrailed expression in bcdKO mutant individuals.

Live imaging of bcdKO embryogenesis from onset of gastrulation to the end of dorsal closure. Embryos express H2Av::mCh (red) and en>mCD8::GFP (green). Dots indicate En stripes and the numbers on top of the embryos indicate the AP length of each individual.

Embryonic length dictates gap gene expression patterns in the absence of bcd

What underlies the correlation between embryonic length and phenotypes of bcdKO embryos? We next focus on variation in gap gene expression patterns in bcdKO embryos of varying geometry. Figure 6A shows representative expression patterns of four gap genes in bcdKO and bcdKO, fat2RNAi mutants. The gene network shows qualitative differences in behavior within different ranges of embryonic length. Without the long-range gradient of Bcd, zygotic hb transcription is activated by the termini system mediated by the terminal gap gene, tailless (Margolis et al., 1995; Figure 6—figure supplement 1A). As a result, two Hb stripes form near the anterior and posterior poles of the embryo, spanning a width of ~10 and 15 nuclei, respectively. In embryos with extremely large aspect ratio (range 1, EL within 330–360 µm), the two Hb expression domains are in close proximity. This inhibits the expression of Kni, which is strongly repressed by Hb, in the central region of the embryo (Hülskamp et al., 1990). Meanwhile, Gt is activated by uniformly distributed maternal Cad protein, and in turn inhibits Kr expression (Kraut and Levine, 1991; Rivera-Pomar et al., 1995; Figure 6A–C, range 1).

Figure 6 with 2 supplements see all
Download asset Open asset
Phenotypic discordance can be traced back to variations in gap gene expression in bcdKO individuals.

(A) Representative gap gene expression patterns within different range of EL. Range1, 330–360 µm; range2, 390–420 µm; and range3, 510–540 µm. (B) Embryonic length of bcdKO individual plotted vs. boundary position of four gap genes shown as absolute distance from the anterior pole (Hb, blue; Gt, magenta; Kr, red and Kni, green). Data from every 30 µm EL interval were binned to compute mean and s.d. and the colored areas are generated by connecting mean values of different EL ranges. Dashed boxes indicate ranges of EL corresponding to (A). Dashed line indicates posterior boundary of individual embryos; red and green crosses overlapping the dashed line indicate individuals with corresponding EL not expressing Kr and Kni, respectively. (C) Schematic illustration of positional information transfer from maternal systems to gap gene expression in bcdKO embryos within different range of embryonic length. Scale bar, 100 µm.

In individuals with moderately increased embryonic length (range 2, EL within 390–420 µm), Hb stripes in the terminal regions separate further apart, permitting Kni expression in the middle region (Figure 6A–C, range 2). This Kni stripe is sandwiched by two Gt expression domains, a thin anterior stripe and a wider posterior one. The anterior Gt stripe in bcd mutants has been observed before (Staller et al., 2015) but its regulatory interactions remain elusive. Potentially it is activated by the remnant anterior determinants such as the maternal Hb, distributed in the anterior half of the embryo (Irish et al., 1989). Comparatively shorter embryos show phenotypically higher degree of symmetry in both cuticle and gene expression patterns, conceivably due to stronger repression of maternal Hb in shorter individuals (Figure 6—figure supplement 1B–D; Figure 6—figure supplement 2).

Looking more closely at individuals within the natural range of embryonic geometry (range 3, EL within 510–540 µm), the sufficient physical space between two Hb stripes permits the expression of Kr, Kni and Gt, arranged in spatial order that is conserved as in wild-type embryos (Figure 6A–C, range 3). In summary, as a consequence of gradually increasing embryonic length, a continuously increasing variety of gap gene expression domains are activated along the AP axis, which is in turn, translated into increased number of body segments, as manifested by the pair-rule gene expression pattern (Figure 6—figure supplement 1E).

Discussion

Individuals of the same species often display a certain level of morphological and behavioral differences, such as in animal color patterns and human facial features (Beldade et al., 2002; Sheehan and Nachman, 2014). This reflects inter-individual variation in genetic composition and life-history environmental exposure (Dall et al., 2012). Such intraspecific individuality may have significant ecological and social impacts on the population (Forsman and Wennersten, 2016). Equally, these same genetic and environmental variations pose challenges to fundamental developmental processes as they try to generate invariant developmental outcomes. Multiple lines of evidence suggest that organisms have evolved canalization mechanisms that render developmental processes insensitive to such sources of variation (Félix and Barkoulas, 2015; Flatt, 2005). Early Drosophila embryonic patterning provides an excellent example of a canalized developmental process – the boundaries of segmentation gene expression remain highly reproducible amongst individuals in the face of heterozygous mutations (Lehmann and Nüsslein-Volhard, 1987; Wieschaus et al., 1984), genetic variations (Lott et al., 2007) and temperature perturbations (Houchmandzadeh et al., 2002; Lucchetta et al., 2005). These studies suggest that mechanisms including epistasis, genotype-environment interactions and canalizing gene regulatory networks (Manu et al., 2009) work together to ensure precise patterning outputs.

In this study, we have identified embryonic geometry as an additional source of variation that patterning processes have evolved to buffer against. The geometry, or in other words, the aspect ratio of each ellipsoid-shaped embryo is determined during oogenesis, and this parameter varies by ±10% in the population of the wild-type strain OreR. The variable geometry in turn increases the variation in embryonic length given the natural range of embryonic geometry. Previous studies have shown that patterning outcomes are highly reproducible and remain scaled to embryonic length (Antonetti et al., 2018; Lott et al., 2007). What roles do Bcd and the gap gene interactions have in this scaling (Wu et al., 2015)? Combining results from fat2RNAi embryos with theoretical modeling may reveal exciting new insights into the underlying scaling mechanisms (Bieler et al., 2011; Jaeger et al., 2004). Correspondingly, we found that under decanalized conditions, either by depleting maternal bcd inputs or artificially increasing the bcd dosage, the patterning process loses its capacity to buffer embryonic length variations. Consequently, the length of an individual embryo predetermines its patterning outcomes. The predictive power becomes stronger when we artificially increase the variation of the embryonic geometry. The aspect ratio of the fat2RNAi embryos differs by ±30% while the average embryonic volume is only slightly reduced by ~8%. These results further support embryonic geometry as a major source of variation that accounts for inter-individual phenotypic variation under decanalized conditions.

Both embryonic volume and embryonic length are inheritable traits and therefore adaptive to artificial selection or environmental changes (Azevedo et al., 1997; Fox and Czesak, 2000; Lott et al., 2007; Miles et al., 2011). It will be interesting to understand if the aspect ratio of the embryo shape is also a genetically variable trait so that the population can be selected to produce progenies with a biased geometry. If this is the case, embryonic geometry may be involved in the complex interplay between environment, genetic components and developmental processes during the course of evolution. When a population confronts selection towards a new phenotypic optimum, for example, larger egg volume due to decreasing temperature (Azevedo et al., 1997), such directional selection may result in decanalizing effects on the patterning processes (Miles et al., 2011; Wagner et al., 1997). Meanwhile, the naturally variable embryonic geometry - together with other sources of variation - generates a spectrum of patterning outcomes in different individuals. As a result, a different range of embryonic geometry will be favored and selected as they maintain the patterning outcomes of the parental lines. Conceivably, this may be one of the reasons why eggs of closely related Dipteran species differ not only in size (Gregor et al., 2008; Markow et al., 2009) but also in geometry, and such geometrical differences can also be observed in different laboratory lines carrying different genetic background.

Our quantitative analysis of segmentation gene expression demonstrates how embryonic geometry affects individual patterning outcomes under two decanalizing conditions. In the case of the maternal bcd null mutant, we have shown that the signaling centers located at both poles of the embryo initiate the hierarchical gene expression along the AP axis in a non-scaled manner. This explains, mechanistically, how patterning processes incorporate information of the embryonic geometry to account for the final outputs. It remains unclear, however, in the case of increased bcd dosage, what determines the breaking point (the fourth abdominal segment) of the final pattern. One possibility is that the susceptibility of this position reflects the strength of the regulatory interactions between the segmentation genes (Jaeger, 2011). Systematic comparisons among different Drosophila species have shown that the regulatory sequences of the segmentation genes are rapidly evolving and thus substantially diverged (Ludwig et al., 2000). Interestingly, the spatio-temporal dynamics of the segmentation gene expression patterns are highly conserved between species, suggesting that the co-evolution of modular transcription binding sites compensate for each other to keep the patterning outcomes unchanged (Gregor et al., 2005; Ludwig et al., 2005; Ludwig et al., 2000). Such an inter-species canalization phenomenon is also observed among more distally related species within the sub-taxon Cyclorrhapha, which involved more dramatic rewiring of the regulatory network (Crombach et al., 2016; Wotton et al., 2015). If the breaking point of patterning processes under decanalized conditions truly depends on the system parameters of the underlying network (Jaeger and Crombach, 2012), we expect to see different susceptible points in different network structures. This can be tested by characterizing decanalizing phenotypes in related species.

A longstanding question in patterning is how do gene regulatory networks downstream of morphogens incorporate information about the macroscopic geometrical parameters of each individual to give rise to scaled patterning outputs? While our results do not provide mechanistic insight into the scaling mechanisms of the gap gene network, they define the physical boundaries where such scaling breaks down. The emergence of scaled expression boundaries is closely linked to embryo geometry, and future models will hopefully more rigorously test the role of geometry in the scaling. Further, our results directly show that the Bcd gradient shape depends on the embryo geometry; in our case, by roughly conserving embryo volume but reducing the aspect ratio, the Bcd gradient extends further into the posterior, resulting in a posterior shift in gap gene expression boundaries.

In conclusion, embryonic geometry was identified as a source of variation in addition to environmental and genetic factors that predetermines phenotypic outcomes in mutant conditions. We think that embryonic or more generally, tissue geometry may play an important role in other decanalizing conditions by affecting patterning outputs, such as other segmentation gene mutants (Janssens et al., 2013; Surkova et al., 2013), or in vitro induction of patterning systems (Lancaster and Knoblich, 2014; Simunovic and Brivanlou, 2017), both of which show significant inter-individual phenotypic variations. Our work highlights that care must be taken when taking a system out of its native environment – for example organoids – as the system boundaries affect the operation of signaling networks. Characterizing the influence of the geometrical parameters will help us to have a more complete understanding of decanalization, and in turn, canalizing phenomenon.

Materials and methods

Fly stocks and genetics

Request a detailed protocol

The bcd knockout allele (bcdKO) used in this study was generated by CRISPR-mediated insertion of a MiMIC cassette into the first intron of the bcd gene (Huang et al., 2017; Venken et al., 2011) The cuticle phenotype of bcdKO was compared to that of the classic bcdE1 allele (Frohnhöfer and Nüsslein-Volhard, 1986) To generate embryos with artificially reduced aspect ratio, we expressed RNA interference against the fat2 gene using a maternal traffic jam (tj) >Gal4 driver (Barlan et al., 2017) Both UAS > fat2 RNAi and tj >Gal4 were either crossed to or recombined with the bcdKO allele, so that the females carrying all three alleles produce bcd null embryos with wide range of aspect ratio.

The tandem bcd construct was generated by replacing the eGFP sequence in the pCaSpeR4-egfp-bcd vector (Gregor et al., 2007b) by bcd protein coding sequence. First, the vector was digested with NheI and SphI to remove the eGFP. Next the bcd coding sequence was amplified by PCR from the vector using primer pairs

  • 5’-cggagtgtttggggctagcaaagatggcgcaaccgccg-3’ and

  • 5’-gttagtagctccgcttccattgaagcagtaggcaaactgcgagtt-3’.

Further the P2A self-cleaving peptide with the GSG linker was synthesized as oligo pairs

  • 5’-tttgcctactgcttcaatggaagcggagctactaacttcagcctgctgaagcaggctggagacgtggaggagaaccct ggacctgcatgcatggcgcaaccgc-3’ and

  • 5’ggcggttgcgccagcatgcaggtccagggttctcctccacgtctccagcctgcttcagcaggctgaagttagtagctccgcttccattgaagcagtaggcaaa-3’.

These two fragments and the digested vector were then assembled using Gibson Assembly strategy (NEB). The final construct was injected (BestGene Inc) and two insertions on 2nd (tdBcd(II)) and 3rd (tdBcd(III)) chromosome, respectively, were established and used for this study. Consequently, hetero- or homo-zygous tdBcd(III) females produce embryos with 4x and 6x of maternally loaded Bcd protein, respectively (compared to 2xbcd in the wild-type); and homozygous tdBcd(II);tdBcd(III) females generate 10xbcd embryos. Finally, females homozygous for tdBcd(III) which also carry tj >Gal4 and UAS >fat2 RNAi generate 6xbcd embryos with reduced aspect ratio.

Other fly lines used in this study include a laboratory OreR strain raised in 25°C (the wild-type control); en >mCD8 GFP (to visualize dynamic en expression pattern); egfp-bcd line (for quantification of Bcd gradient profile); mat >eGFP; hb >LlamaTag (gift from Hernan Garcia’s lab); Df(3R)tllg (BL#2599).

Measurement of embryonic geometry

Request a detailed protocol

To compare the geometrical parameters between OreR, bcdKO and fat2RNAi populations, embryos were dechorionated and aligned laterally on an agar plate and imaged under a stereoscope (Nikon SMZ18). Images were then segmented to extract the embryo contour and fitted to elliptic shapes. The long and short axes of fitted ellipses were taken as the measurement of embryonic length and width, respectively. The approximate embryonic volume is calculated using the measured length and width, assuming embryos are ellipsoids in shape. For each strain, more than 200 individuals were measured.

For confocal live imaging, embryonic length was measured as the longest distance between the anterior and posterior poles. The immunostaining procedures result in an isotropic shrinkage of embryonic volume. To measure the geometrical parameters of the fixed embryos, we first carried out a linear fit between aspect ratio and embryonic length using stereoscope data. Further we measured the aspect ratio of each fixed embryo and estimate its embryonic length and width using the same linear fit equation.

We checked our assumption of isotropic shrinkage upon fixation (Figure 1C and Figure 1—figure supplement 1A). Comparing embryos from live imaging and fixed embryos, we did not observe significant anisotropy in the embryo geometry (Figure 1—figure supplement 1B–C).

Immunostaining

Request a detailed protocol

Embryos at desired stages were dechorionated by household bleach and fixed in heptane saturated by 37% paraformaldehyde (PFA) for 1 hr. The vitelline membrane was subsequently manually removed. Prior to incubation with primary antibodies, embryos were blocked with 10% BSA in PBS. Antibodies used were guinea pig anti-Hb (1:2000), rabbit anti-Gt (1:800), guinea pig anti-Kr (1:800), guinea pig anti-Kni (1:800), guinea pig anti-Eve (1:800). Primary antibodies were detected with Alexa Fluor-labelled secondary antibodies (1:500; LifeTech). Embryos were co-stained with Phalloidin conjugated with Alexa Fluor for staging purpose or visualizing cephalic furrow position. Short incubation of Dapi dye was carried out during the last wash prior to mounting to visualize presyncytial nuclei. Embryos were mounted in AquaMount (PolySciences, Inc) and imaged on a Zeiss LSM710 microscope with a C-Apochromat 40x/1.2 NA water-immersion objective. Hb, Gt, Kr, Kni and Eve antibodies were gifts from Johannes Jaeger.

Cuticle preparation

Request a detailed protocol

Embryos of various genotypes were collected during the blastoderm stages and allowed to develop at 25°C until the end of embryogenesis. The embryos were then dechorionated, fixed, devitellinized and incubated into a mixture of Hoyer’s medium and Lactic acid in a 1:1 ratio at 65°C between an imaging slide and a cover slip. For an exhaustive description of the method used see Alexandre (2008).

Measurement of bcd profile

Request a detailed protocol

For measurement of Bcd gradient profile, we followed the protocols detailed in Gregor et al. (2007a). Embryos expressing eGFP-Bcd either with or without fat2RNAi were dechorionated and mounted laterally on a confocal microscope (Zeiss LSM710). The images were acquired at the midsagittal plane of embryos at early n.c. 14. Data acquired in Figure 1F–H for different individuals (including eGFP:Bcd and eGFP:Bcd; tj >Gal4,UAS > fat2 RNAi) were co-mounted on the same glass-bottom dish and taken with identical microscope settings. For each image, nuclear centers along the dorsal edge of the embryo were manually selected and the corresponding circular area was used to compute the average fluorescent intensity. Nuclear intensity was then plotted against either absolute distance from the anterior or scaled AP position. To compare average profiles between control and fat2RNAi embryos, all nuclei from embryos either longer or shorter than 450 µm are binned in 50 bins along the scaled AP axis over which the mean and standard deviation were computed.

Fluorescence in situ hybridization

Probe synthesis

Request a detailed protocol

A 471 bp region of bcd transcript was amplified from early embryo cDNA using primer pairs:

5’-cccggatccCTCAAATAGCAGAGCTGGAGC-3’ and 5’-cccggtaccGCTGCTGCTGGAAGAACTG-3’ and subcloned into the pSP18 vector. Further, the vector was linearized by BamHI restriction digestion and DIG-labeled ‘anti-sense’ RNA was synthesized with T7 RNA polymerase using DIG RNA Labeling Kit [Roche-11175025910]. Synthesized RNA probes were then precipitated by adding 75 µl pre-chilled ethanol, 1.3 µl LiCl (7.5 M) and 1 µl yeast tRNA (25 mg/ml) to the labeling reaction. Precipitation takes overnight at −20°C. After centrifuging at 4°C full speed for 30 min, the precipitated RNA probes were washed with 70% ethanol (in DEPC water) and re-suspended in 100 µl Hyb-A buffer (50% formamide, 5XSSC buffer, 100 µg/ml salmon DNA [BDL F012], 50 µg/ml heparin [Sigma-Aldrich H4784], 0.1% tween-20 in DEPC water). The synthesized RNA probes were stored in −20°C.

Pre-hybridization

Request a detailed protocol

Embryos at stages within 2 hr after fertilization were collected, dechorionated and fixed in 1:1 solution of heptane and 4% formaldehyde in PBS for 20 min. After that, lower phase of the solution was replaced by methanol with the same volume. The solution was vortexed for 20 s to remove the vitelline membrane, and the embryos at the bottom of the lower phase were collected and rinsed three times with methanol. Embryos were then rehydrated by washing them in sequentially increasing percentage of PBT (PBS+0.1%Tween-20) in methanol. Next, embryos were post-fixed in 4% formaldehyde in PBT for 20 min and washed 5 times in PBT. Embryos were then washed in Hyb-B buffer (50% formamide, 5XSSC in DEPC water). Finally, pre-hybridization was carried out by placing the embryos in Hyb-A buffer at 65°C for 3.5 hr.

Hybridization

Request a detailed protocol

5 µl of RNA probes was diluted in 250 µl Hyb-A buffer. The probe mix was heated at 80°C for 10 min and placed on ice for 5 min. The probe mix was added to the embryos and hybridization was carried out at 65°C for 18 hr. The embryos were washed at 65°C in Hyb-B buffer for 6 times, 30 min each wash. After, the embryos were washed sequentially in the following solution at room temperature: 4:1 Hyb-B:PBT, 1:1 Hyb-B:PBT, 1:4 Hyb-B:PBT and PBT, 5 min each wash. Embryos were incubated with blocking solution (4:1 PBT:Western blocking reagent (Roche- 11921673001)) for 1 hr. For primary staining against DIG-labeled RNA probes, embryos were incubated with sheep anti-DIG [Sigma-Aldrich 11333089001] with 1:400 dilution in the blocking solution at 4°C overnight. After washing in PBT for 6 times with 20 min each wash, embryos were stained with the secondary antibody donkey anti-sheep Alexa 555 [Thermo Fisher Scientific A21436] with 1:500 dilution in PBT for 1 hr at room temperature. Finally, after washing in PBT for 6 times, embryos were mounted on microscopy slides with Aqua-Poly/Mount [Polysciences, Inc- 18606].

Microscopy imaging and image analysis

Request a detailed protocol

Slides were imaged on a Zeiss confocal LSM880 using 40X water immersion lens. For each embryo, Z-stack images were taken covering the region from surface to the midsagittal plane, and the neighbouring planes were separated by 3 µm. Following that, a projected maximum intensity image was generated. The outside of the embryo was segmented using thresholding in Matlab after using rolling ball background subtraction. A 10μm-thick mask was defined around the embryo perimeter in order to avoid effects from yolk autofluorescence. Finally, the average signal intensity at each position along the AP axis was taken from the mask. The signal from each embryo varies significantly, making a rigorous quantitative comparison challenging. The signal from each embryo was normalized by the intensity around 200 μm from the anterior pole.

Simulation of SDD (Synthesis, Diffusion, Degradation) model

Request a detailed protocol

We use COMSOL Multiphysics 5.3 to simulate diffusion along the surface of prolate ellipsoids of varying sizes to account for the geometries of wild type and fat2RNAi embryos.

We consider steady state equations for the concentration of Bcd in the unfolded (U) and folded (F) states. These take the form:

∂tU=D∇2U-(μ+α)U+Jfx.

∂tF=D∇2F+αU-μF,

where: ∇2 represents the Laplacian on the ellipsoid surface; α is the GFP folding rate; µ is the decay rate of Bcd in both the unfolded and folded forms; and Jf(x) corresponds to a production rate of unfolded Bicoid, which we suppose to depend only on the distance to anterior (x-axis corresponds to AP direction). We consider a step-like production function:

fx=1-tanh⁡x-λδ,

where λ is the AP-axis extension of a region of approximatively constant production; δ is a smoothing parameter. We have checked that the simulations weakly depend on the value of δ as long as δ<< λ.

Physical parameters were fixed based on previous literature, see Table below. Quantitative fits were obtained by using a L2-measure with a window-type weight function for data-points within the rescaled length range x/L∈[0.2,0.8].

The solution for folded Bcd is not flat is the vicinity of z = 0 (i.e. near the head); this is due to the projection into the Cartesian coordinates, leading to a singular integration measure near its extremal value z = 0. Similarly, we disregard the experimentally measured values of Bcd intensity within the first microns form the anterior pole as these are prone to large errors.

ParameterValuesReference
DDiffusion coefficient of folded and unfolded bcd4 µm2s−1Durrieu et al., 2018
λProduction region50 µmin situ data in
this study and Little et al., 2011
δWidth of decay of the production region5 µmin situ data in this study
µProtein decay rate1/ (35*60) sDurrieu et al., 2018
αFolding rate1/ (50*60) sDurrieu et al., 2018 and Liu et al., 2013
aWTWT type length, dorso-ventral axis100 µmSee Figure 1
cWTWT type length, anterio-posterior axis250 µmSee Figure 1
afat2fat2RNAi type length, dorso-ventral axis115 µmSee Figure 1
cfat2fat2RNAi type length, anterio-posterior axis175 µmSee Figure 1

As the equations are solved on a closed surface, there are no explicit boundary conditions.

Gap gene boundary quantification

Request a detailed protocol

Confocal Z-stack images were Z-projected (maximum intensity) in Fiji (RRID:SCR_002285) for further analysis. The images of laterally oriented embryos were rotated so that the anterior is to the left and dorsal to the up. A line with the width of 100 pixels crossing the center of the embryo was drawn to extract average intensity along the AP axis. The intensity profile was shown as a function of percent embryonic length (%EL). We determined the background intensity Imin and maximum intensity Imax of each embryo; Imin was subtracted from the intensity profile and the resulting curve was rescaled by 1/(Imax-Imin). The boundary position is defined where rescaled intensity profile equals to 0.5.

To estimate the variation of boundary position, we performed bootstrapping using the Matlab function bootstrp. We performed 1000 simulation runs to infer the variability on the boundary precision. In each run, a random data set was generated by sampling from the known experimental data. In this procedure, each data point can be included multiple times within each random sample. As the number of samples per boundary was small, we did not test for significant changes in the precision of boundary specification for a single boundary between wild-type and bcdKO embryos. However, pooling the data from the different boundaries, we observe that precision in all the measured boundaries decreases (i.e. the error increases) in bcdKO embryos. We calculated the p-value using the paired sample t-test across all boundaries.

Quantification of maternal bcd transcripts

Request a detailed protocol

To compare the relative amount of maternally loaded bcd transcripts in different genotypes, we extracted total mRNA from presyncytial embryos (within 1 hr after egg deposition) generated by OreR, fat2RNAi, tdBcd(III) or tdBcd(II);tdBcd(III) females and reverse transcribed to cDNA. We performed qRT-PCR with bcd-specific primer pair using SYBR Green (Thermo Fisher) protocol and the housekeeping gene rpl32 was used as internal reference. The relative bcd mRNA amount was normalized to that of OreR. Three independent measurements were carried out over which the mean and standard deviation was calculated.

Data availability

All data generated or analysed during this study are included in the manuscript and supplementary files.

References

    1. Fujioka M
    2. Emi-Sarker Y
    3. Yusibova GL
    4. Goto T
    5. Jaynes JB
    (1999)
    Analysis of an even-skipped rescue transgene reveals both composite and discrete neuronal and early blastoderm enhancers, and multi-stripe positioning by gap gene repressor gradients
    Development 126:2527–2538.
    1. Jaeger J
    (2011) The gap gene network
    Cellular and Molecular Life Sciences 68:243–274.
    https://doi.org/10.1007/s00018-010-0536-y
  44. Book
    1. Jaeger J
    2. Crombach A
    (2012) Life’s Attractors
    In: Soyer OS, editors. Evolutionary Systems Biology. New York: Springer. pp. 93–119.
    https://doi.org/10.1007/978-1-4614-3567-9
    1. Kraut R
    2. Levine M
    (1991)
    Mutually repressive interactions between the gap genes giant and kruppel define middle body regions of the Drosophila embryo
    Development 111:611–621.
    1. Margolis JS
    2. Borowsky ML
    3. Steingrímsson E
    4. Shim CW
    5. Lengyel JA
    6. Posakony JW
    (1995)
    Posterior stripe expression of hunchback is driven from two promoters by a common enhancer element
    Development 121:3067–3077.
    1. Namba R
    2. Pazdera TM
    3. Cerrone RL
    4. Minden JS
    (1997)
    Drosophila embryonic pattern repair: how embryos respond to bicoid dosage alteration
    Development 124:1393–1403.

Decision letter

  1. Naama Barkai
    Senior and Reviewing Editor; Weizmann Institute of Science, Israel

In the interests of transparency, eLife publishes the most substantive revision requests and the accompanying author responses.

Thank you for submitting your article "Embryonic geometry underlies phenotypic variation in decanalized conditions" for consideration by eLife. Your article has been reviewed by two peer reviewers, and the evaluation has been overseen by a Reviewing Editor and Naama Barkai as the Senior Editor. The reviewers have opted to remain anonymous.

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

Summary:

Huang and Saunders investigate phenotypic variability of bicoid mutant Drosophila embryos. Remarkably, they find that in mutant embryos, the number of engrailed stripes and the number of abdominal segments is determined by the length of the embryo. This rule applies to embryos with abnormally short AP axes generated by knockdown of fat2. In contrast, in the presence of Bcd, the correct number of segments always arise, regardless of embryo length, even in fat2 KD embryos. Thus, Bcd is required for scaling of expression patterns to accommodate variation in embryo length. Measurements of Bcd-GFP strongly suggests that this scaling is achieved by reducing the amplitude of the Bcd gradient concomitant with the reduction in egg length. By examining gap gene expression, they show that in short bcd mutant eggs, the distance between anterior and posterior poles must be sufficiently large to permit the expression of the genes Kruppel and knirps. Their data are consistent with a view of the bicoid mutant phenotype in which Torso signaling is not scaled in accord with embryo length, so that terminal signaling in short embryos overrides factor(s) that would otherwise permit the expression of Kr and knirps. Finally, the authors show that excess Bcd also disrupts patterning by interfering with specification in the vicinity of the fourth abdominal segment, with additional segments affected in shorter embryos.

The work is exciting because it articulates an explanation for phenotypic variability in bcd mutants. The manuscript is significant in showing that the Bcd gradient itself provides a buffer against variation in egg length.

Essential revisions:

The paper will benefit from re-structuring, to better explain the main findings. This is detailed in the individual reports, but in the discussion, the reviewers reached an agreement on what they find to be promising approach, which is described below. In addition, to fully establish the main findings, the following additional data are requested:

1) Please explore further how bcd mRNA and protein distributions are affected by fat2 KD, since this may help explain how bcd provides scaling. If the simulation can be used to address this point, the manuscript will be stronger. However, the authors must provide much more detail on how the simulation was implemented if the reader is to believe any conclusion drawn from it. As it stands now, the simulation is a distraction from the results.

2) Please perform an in situ hybridization and quantification of bcd mRNA, as suggested in the review

3) Please include an experimental verification of embryo shrinking.

Suggestion for revising the writing:

The opening question can remain the same ("what is canalization buffering against"); the authors show buffering against a wide range of embryo sizes, even well outside the normal range; this sets up the question of what AP patterning factors are required for canalization; then authors show the requirement of Bcd: without bcd, aspects of the resulting mutant phenotype (like the number of En stripes) now correlate with embryo size. I also suggest to keep the Bcd overexpression data at the end of the manuscript. Then at the very least, the authors should discuss ways / propose models to explain how Bcd might provide scaling, possibly through altered mRNA expression, localization, and/or translation, with references to Jun Ma's set of work on this very subject. The manuscript would be greatly strengthened if the authors could provide evidence, experimental and/or using their (currently rather poorly explained) model, that one or more of these processes are different in embryos of different sizes. This would help the field understand how the Bcd protein gradient is reshaped (as they claim) and in turn how this reshaping confers scaling.

Reviewer #1:

Executive summary: Overall, I believe this paper contains enough material to support and justify an interesting point using a clever method. However, I also find that, as written, its order of presentation does the narrative a great disservice. Below, I advocate for a rather serious reorganization of presentation (which incidentally should also make the paper shorter). But provided the presentation is modified, I do believe this paper can be made acceptable for eLife without additional experimental work (except for one small point – see moderate point #3).

Details:

Embryonic development in the fruit fly is said to exhibit "canalization", generating reproducible phenotypic outcomes despite inter-individual differences in genetic background and environmental conditions, within a certain range. Outside this range ("decanalized conditions"), individuals become sensitive to variations that would otherwise be neutral. In this manuscript, Huang & Saunders ask exactly what sources of variation canalization is buffering against. They argue that, in addition to genetic and environmental perturbations, another previously unidentified source of variation is the "naturally variable embryonic geometry". [But see "minor point 2"]

To address this question ("what sources of variation is canalization buffering against?"), the authors designed a clever way to modify geometry of the embryo, plausibly without changing very much else [but see "minor point 3"]. Using their clever tool, the authors can effectively increase the variability in embryo geometry – thus testing the ability of the developmental system to buffer against such perturbations.

In the absence of other complicating factors, the embryo demonstrates a truly remarkable capacity to deal with perturbation of its geometry. Frankly, to me this is perhaps the most striking finding reported here, but this is not what the authors focus on.

Now, in my mind, the way for the authors to make their point would be to apply a pressure towards decanalization of development, and observe that the ability of the embryo to buffer the geometric perturbations is reduced. With no other problems, embryos are basically normal (Figure 3—figure supplement 1, panels EFGH are frankly amazing and deserve to be in the main text.) With some perturbation (e.g. temperature, or increased dosage of Bcd like here) the embryos are able to buffer small variations of geometry but not big ones. With an even larger perturbation, embryos with non-normal aspect ratio fail to develop. This is the approach I would have expected, given the authors' stated goals.

But the authors instead apply a truly dramatic perturbation, removing maternal bcd. This is a lethal mutant. They then observe that sometimes the embryo is short, and everything is broken completely. At other times the embryo is longer and everything is broken partially. (A few structures do form in a way that resembles normal – but the entire first half of the embryo is still gone.) Thus, observed results subsection 2, the degree to which the embryo is broken is correlated with embryo length. And if we make the embryo's life even more difficult by compounding the already lethal bcd knockout with a further perturbation of geometry, things get even worse. (Results subsection 4).

I do not understand how we can make claims about what perturbations a system had evolved to buffer by studying it in a completely dysfunctional regime, which is by definition irrelevant for development. The authors demonstrated that in a family of lethal mutants shorter embryos are even more screwed up than longer ones. But all of them are lethal, at which point counting "just how badly lethal" seems irrelevant.

The authors then go on to apply another set of dramatic perturbations – rather than increasing Bcd a little, they overexpress it 6x or 10-fold, the latter again being lethal.

By this point I personally was not buying the story at all – but finally, the paper gets to Figure 5 and subsection “The gene network breaks at susceptible point in decanalized conditions” onwards, which finally turned everything around. I found that result very strong and interesting. But if I were a reader rather than a reviewer, I would not have made it that far.

My proposal: if the point made in the Introduction/Discussion section is indeed the intended focus of the narrative, I believe the paper could make it point much more persuasively, while also being shorter:

• Introduce the fat2RNAi trick that allows generating embryos with a much more varied geometry than normal, and explain how that offers a fantastic window into studying canalization

• Observe that, absent other perturbations, the development is capable of buffering that "geometric" variability. Move supplementary panels from Figure 3 supplement to the main text.

• Stress the system by increasing bcd dosage (4x, 6x), and proceed through the argument at the end of the Results section.

This is the narrative that actually makes the point set up in the Introduction, and this point is a strong one: the construct developed by the authors allows directly probing the ability of a system to buffer variations – in a very clean, controlled setting, of which I know few. It is a remarkable setting and beautiful result. Only after this point is made, can the response to bcd removal really be seen as further reinforcing the authors' point – not before, because such dramatic perturbation is way beyond anything relevant for normal development and therefore canalization; on its own, this line of evidence alone would be questionable. E.g. last paragraph of subsection “Embryonic length dictates segmentation gene pattern in the absence of bcd” – these discussions are useful for mapping out who activates who in a pathway, but entirely irrelevant for normal development and "what canalization evolved for", surely! I would in fact consider largely moving all the discussions of non-viable mutants to the supplement.

Other specific issues:

Moderate points:

1) Paragraph two of subsection “Embryonic length dictates segmentation gene pattern in the absence of bcd” unexpectedly bring up the question of expression pattern scaling and I'm not sure whether this is a good move. Exactly what is the paper contributing to that conversation? It seems like a distraction. I suggest removing.

2) "Genetic network breaks at susceptible point" – what does this sentence mean? How does a genetic network "break"? What is a "susceptible point"? Is this a result, or some intuition the authors are trying to communicate? The strongest point in this paper is introduced by the weakest/meaningless header. Similarly, when the Abstract is revised, I would suggest removing the last sentence referring to "vulnerable points in the network" which again sounds like some intuition (belonging to Discussion section) rather than a finding.

3) Second paragraph of subsection “Measurement of embryonic geometry”: I found this hard to parse (particularly the use of the term "further"). My interpretation is that EL pre-fixation is inferred from EL post-fixation assuming that aspect ratio is unchanged – correct? This should be clarified. But even more importantly – was this assumption of isotropic shrinkage verified? In my experience a harsh fixation protocol can lead to extensive deformations, their isotropic nature is not obvious, and given the role played by the aspect ratio in this paper, this seems an important point to demonstrate. [This is the one place where additional data may be required, but shouldn't be too onerous to obtain]

4) SDD simulation: I am very confused.

a) If I use an cylinder to approximate an ellipsoid, then geometrically, the "best approximation" cylinder is surely narrower than the embryo. Not wider. It seems like width is used as a free fitting parameter to make the data fit better, but the text tries to make it sound like this larger-than-actual radius was somehow expected, and I don't understand the argument.

b) Further, I believe the diffusion parameters, both in the bulk and at the surface, for a small molecule like Bcd or Hb, were previously measured in the embryo specifically. So it sounds like this could be a zero-parameter fit. How bad is it? Why the discrepancy?

c) And in particular, if using an elliptical geometry truly does not significantly alter the results, why not use it, given that the parameters of the actual geometry are, again, known?

d) Does the SDD simulation also reproduce the features that the plot of Figure 3G specifically highlights, namely the larger concentration in the posterior and the crossing point? The panel D in Figure 3 supplement does not allow to see this.

Reviewer #2:

Huang and Saunders investigate phenotypic variability of bicoid mutant Drosophila embryos. Remarkably, they find that in mutant embryos, the number of engrailed stripes and the number of abdominal segments is determined by the length of the embryo. This rule applies to embryos with abnormally short AP axes generated by knockdown of fat2. In contrast, in the presence of Bcd, the correct number of segments always arise, regardless of embryo length, even in fat2 KD embryos. Thus, Bcd is required for scaling of expression patterns to accommodate variation in embryo length. Measurements of Bcd-GFP strongly suggests that this scaling is achieved by reducing the amplitude of the Bcd gradient concomitant with the reduction in egg length. By examining gap gene expression, they show that in short bcd mutant eggs, the distance between anterior and posterior poles must be sufficiently large to permit the expression of the genes Kruppel and knirps. Their data are consistent with a view of the bicoid mutant phenotype in which Torso signaling is not scaled in accord with embryo length, so that terminal signaling in short embryos overrides factor(s) that would otherwise permit the expression of Kr and knirps. Finally, the authors show that excess Bcd also disrupts patterning by interfering with specification in the vicinity of the fourth abdominal segment, with additional segments affected in shorter embryos.

The work is exciting because it articulates an explanation for phenotypic variability in bcd mutants. The manuscript is significant in showing that the Bcd gradient itself provides a buffer against variation in egg length. The main finding, at least to this reviewer, is that Bcd itself is required for length scaling. The authors should draw greater attention to this result. This result seems consistent with the work of Jun Ma, whose publications should be cited (e.g., doi: 10.1242/dev.064402). Interestingly, scaling occurs in fat2-depleted embryos despite the paradoxical finding of more bcd mRNA upon fat2 KD (Figure 5—figure supplement 1). This suggests alteration in Bcd translation and/or localization of mRNA in short eggs. The manuscript would be strengthened if the authors could comment further about the nature of Bcd-mediated canalization, particularly to explain why more mRNA generates less protein. Along these lines, more detailed description of the simulation would be helpful. In particular, it would strengthen the manuscript to know whether bcd mRNA localization might play a role in canalization. What geometry of bcd mRNA was used in the simulation? It should not be beyond the technical limits of the lab to perform a semi-quantitative in situ hybridization for bcd mRNA to address the question of whether bcd mRNA distribution is changed by fat2 KD in a way that helps explain Bcd-mediated scaling.

As a final major comment, the authors document variability in gap gene boundaries in control embryos, fat2 KD, and bcd mutants (Figure 1 J-M, Figure 1—figure supplement 1, and Figure 3—figure supplement 1). The authors should address how much of this variability arises from variability in egg length. The LlamaTag experiment suggests that at some of the variability originates from differences in egg length (this is clear from the anticorrelation in Figure 3I). If egg length determines boundary positions, then there should be a relationship between egg length and gap gene boundary position in bcd mutants. The degree of correlation might be even stronger in bcd mutants due to the absence of the canalizing activity of Bcd. The authors should plot boundary position for each gap gene boundary as a function of egg length, similar to Figure 3I.

[Editors' note: further revisions were suggested prior to acceptance, as described below.]

Thank you for resubmitting your work entitled "Embryonic geometry underlies phenotypic variation in decanalized conditions" for further consideration by eLife. Your revised article has been evaluated by Naama Barkai (Senior Editor) and a Reviewing Editor.

The manuscript has been improved but there are some remaining issues that need to be addressed before acceptance. please address the remaining comments, as specified below

Reviewer #1:

I believe the revisions improved the manuscript significantly.

Reviewer #2:

The revised manuscript will be satisfactory for publication if the authors can clarify the following points listed below regarding the simulation of Bcd-GFP production. The main point of Figure 1F-K is to address how the Bcd gradient changes in embryos with altered geometry. My concern is that the presentation of the simulation in the text may appear at first glance incongruent with the figures. In the end I believe the figures are consistent with the text, but I would like some clarification. I imagine what I am requesting will be trivial for the authors to provide, but at the same time very useful and insightful for readers.

1) The authors state that the same parameter values describe the distribution of fluorescent Bcd-GFP regardless of embryo geometry. Unless I am mistaken, this means that, at steady-state imposed by the model (and using the finding that bcd mRNA doesn't change very much in terms of its total amount), the total amount of Bcd-GFP embryo-wide is equivalent in all embryos regardless of their geometry. Further, the total amounts of fluorescent and non-fluorescent Bcd-GFP are also equivalent. It would be helpful to explicitly state this in the text for those not immediately familiar with modeling and/or may not be aware that the maturation time of GFP is non-negligible in this context.

2) Assuming that Bcd-GFP amount is the same in all embryos, a reader will then examine Figure 1G and note that the summed nuclear intensity (or in some sense, the area under the curve) does not appear to be the same in embryos of different sizes. Instead it appears that Bcd-GFP scales with embryo length. This appears to contradict the idea that total BcdGFP is unchanged. However, the authors say that this is due to "dilution" of Bcd-GFP. I interpret this to mean that if one considers an embryo as a set of "salami slices," then, due to the altered aspect ratio, a slice at some absolute distance from the anterior pole in fat2 KD (or, more generally, in short) embryos has a larger volume compared to a slice at the same distance in wild-type. The apparent reduction in "nuclear concentration" of Bcd-GFP occurs because whatever amount of Bcd-GFP is available at that distance is roughly the same between long and short embryos. This amount distributes into the available volume, so that the apparent concentration of BcdGFP appears lower in fat2 KD. Could the authors please state this more explicitly, if my interpretation is correct. I believe that splitting the long sentence the fifth paragraph of subsection “Developmental reproducibility is preserved with a minor impact on scaling under geometrical perturbations” into two or three sentences will suffice, devoting a sentence or two to each of the two differences (low anterior, high posterior) in the gradients of long and short eggs. This will help the reader to understand why the 1D plot of Bcd-GFP vs AP does not accurately represent the total amount of Bcd and results in the reduction in nuclear intensity in the anterior.

3) Figure 1 presents data and simulation regarding only fluorescent Bcd-GFP. A powerful aspect of the simulation is that it can estimate the quantity that we (and the embryo) actually care about, the distribution of _total_ Bcd. Given the sizable maturation time of 50 minutes used in the simulation, my intuition is that the invisible proportion likely extends over a significantly larger fraction of egg length in short embryos compared to long. However my intuition may well be wrong. Could the authors please present and discuss the estimated _total_ Bcd-GFP distribution as a function of both absolute and normalized distance in embryos of various aspect ratios (lengths)? To my thinking, this is the only way to address the question of how the actual gradient (not just the visible one) changes with changing geometry. For example, after estimating the gradients of total Bcd, could the authors now show the difference in total Bcd concentration, as in Figure 1K lower panel, except now for total Bcd. (Could the authors also include error bars on the difference in 1K and in a new plot regarding total concentration?) Moreover, the difference in total Bcd can be converted into a difference in fractional egg length ("positional error") as a function of normalized AP position, i.e., the difference in the positions at which the same Bcd concentration is found in short eggs compared to long. This difference is zero near the midpoint of the AP axis where the curves cross and becomes larger moving toward either pole. This will give the reader a sense of magnitude of the mistakes in assigning positional identity that we would expect in short embryos if Bcd were the only patterning cue. In turn, these differences can be compared to the changes in the placement of gap gene expression boundaries presented in Figure 2. Are the changes in the Bcd gradient predictive of the changes in boundary placement (in other words, do gap gene boundaries arise at the same Bcd concentration in long and short eggs)? Do gap gene boundaries shift more, less, or the same amount as predicted by the changes in the Bcd gradient? The answers will help address how the altered Bcd gradient in short embryos generates (or works against!) scaling, and how much of the scaling phenomenon must arise independently of the Bcd gradient.

https://doi.org/10.7554/eLife.47380.sa1

Author response

Essential revisions:

The paper will benefit from re-structuring, to better explain the main findings. This is detailed in the individual reports, but in the discussion, the reviewers reached an agreement on what they find to be promising approach, which is described below.

We agree that the structure of the original submission was not clear. We have now reworked presentation of the data such that: (1) we first characterize the fat2RNAi embryos regarding the problem of scaling and reproducibility of segmentation gene expression; (2) then we characterize the effects of embryo geometry under the decanalized condition of Bcd over-expression; (3) and finally we characterize the effects of embryo geometry under the decanalized condition of bcd mutant. We outline specific changes below in our detailed response to specific points raised.

In addition, to fully establish the main findings, the following additional data are requested:

1) Please explore further how bcd mRNA and protein distributions are affected by fat2 KD, since this may help explain how bcd provides scaling. If the simulation can be used to address this point, the manuscript will be stronger. However, the authors must provide much more detail on how the simulation was implemented if the reader is to believe any conclusion drawn from it. As it stands now, the simulation is a distraction from the results.

We have carried out fluorescent in situ hybridization against bcd mRNA in both wild type and fat2 KD embryos. We found that, on an absolute scale, the bcd mRNA in fat2 KD distributes to a similar extent compared to that of the wild type. We have presented these data in Figure 1I-J and subsection “Developmental reproducibility is preserved with a minor impact on scaling under geometrical perturbations”. Further, we have included better simulations that use more realistic embryo geometry. This new work was done with a theorist, Jean-Francois Rupprecht, who performed the new simulations in more realistic geometries. We outline specific changes below in our detailed response to specific points raised.

2) Please perform an in situ hybridization and quantification of bcd mRNA, as suggested in the review

We have performed these experiments. We outline specific changes below. Briefly, these experiments (a) validate that the 10x bcd line significantly increases the levels of bcd mRNA present; and (b) the fat2RNAi line does not have noticeably altered bcd mRNA levels compared to wild-type embryos. This is consistent with a previously reported volume dependence on bcd mRNA levels (Cheung et al., 2011), as our fat2RNAi embryos do not have substantially reduced embryo volume.

3) Please include an experimental verification of embryo shrinking.

We have now performed additional quantifications to test the impact of fixation on the embryo geometry. We do not observe major shifts in the aspect ratio of the embryos before and after fixation, supporting the assumption of isotropic embryo shrinkage. We now present this data in Figure 1—figure supplement 1B-C. We discuss these results further below.

Suggestion for revising the writing:

The opening question can remain the same ("what is canalization buffering against"); the authors show buffering against a wide range of embryo sizes, even well outside the normal range; this sets up the question of what AP patterning factors are required for canalization; then authors show the requirement of Bcd: without bcd, aspects of the resulting mutant phenotype (like the number of En stripes) now correlate with embryo size. I also suggest to keep the Bcd overexpression data at the end of the manuscript. Then at the very least, the authors should discuss ways / propose models to explain how Bcd might provide scaling, possibly through altered mRNA expression, localization, and/or translation, with references to Jun Ma's set of work on this very subject. The manuscript would be greatly strengthened if the authors could provide evidence, experimental and/or using their (currently rather poorly explained) model, that one or more of these processes are different in embryos of different sizes. This would help the field understand how the Bcd protein gradient is reshaped (as they claim) and in turn how this reshaping confers scaling.

We thank the reviewers for detailing their suggested revised structure. We broadly agree with the above reworking. We now detail the behaviour of the fat2RNAi embryos first under normal Bcd expression levels. We describe in more depth the effects of embryo size change on Bcd and the gap genes. We then introduce the effects of geometry on patterning when Bcd is over-expressed. We finish with discussing the effects of geometry on patterning when Bcd is depleted. We believe the last section is important – important mechanistic insights can be made from mutant embryos, and we clarify the point in the text.

Reviewer #1:

Executive summary: Overall, I believe this paper contains enough material to support and justify an interesting point using a clever method. However, I also find that, as written, its order of presentation does the narrative a great disservice. Below, I advocate for a rather serious reorganization of presentation (which incidentally should also make the paper shorter). But provided the presentation is modified, I do believe this paper can be made acceptable for eLife without additional experimental work (except for one small point – see moderate point #3).

With regard to writing, please see the above replies.

Details:

Embryonic development in the fruit fly is said to exhibit "canalization", generating reproducible phenotypic outcomes despite inter-individual differences in genetic background and environmental conditions, within a certain range. Outside this range ("decanalized conditions"), individuals become sensitive to variations that would otherwise be neutral. In this manuscript, Huang & Saunders ask exactly what sources of variation canalization is buffering against. They argue that, in addition to genetic and environmental perturbations, another previously unidentified source of variation is the "naturally variable embryonic geometry". [But see "minor point 2"]

To address this question ("what sources of variation is canalization buffering against?"), the authors designed a clever way to modify geometry of the embryo, plausibly without changing very much else [but see "minor point 3"]. Using their clever tool, the authors can effectively increase the variability in embryo geometry – thus testing the ability of the developmental system to buffer against such perturbations.

In the absence of other complicating factors, the embryo demonstrates a truly remarkable capacity to deal with perturbation of its geometry. Frankly, to me this is perhaps the most striking finding reported here, but this is not what the authors focus on.

We agree this observation is truly remarkable. We have restructured the paper such that this result is now apparent in Figures 1-2 of the revised manuscript.

Now, in my mind, the way for the authors to make their point would be to apply a pressure towards decanalization of development, and observe that the ability of the embryo to buffer the geometric perturbations is reduced. With no other problems, embryos are basically normal (Figure 3—figure supplement 1, panels EFGH are frankly amazing and deserve to be in the main text.)

We agree with this suggestion. We have restructured the manuscript and moved the suggested images to revised Figure 2.

With some perturbation (e.g. temperature, or increased dosage of Bcd like here) the embryos are able to buffer small variations of geometry but not big ones. With an even larger perturbation, embryos with non-normal aspect ratio fail to develop. This is the approach I would have expected, given the authors' stated goals.

But the authors instead apply a truly dramatic perturbation, removing maternal bcd. This is a lethal mutant. They then observe that sometimes the embryo is short, and everything is broken completely. At other times the embryo is longer and everything is broken partially. (A few structures do form in a way that resembles normal – but the entire first half of the embryo is still gone.) Thus, observed results subsection 2, the degree to which the embryo is broken is correlated with embryo length. And if we make the embryo's life even more difficult by compounding the already lethal bcd knockout with a further perturbation of geometry, things get even worse. (Results subsection 4).

I do not understand how we can make claims about what perturbations a system had evolved to buffer by studying it in a completely dysfunctional regime, which is by definition irrelevant for development. The authors demonstrated that in a family of lethal mutants shorter embryos are even more screwed up than longer ones. But all of them are lethal, at which point counting "just how badly lethal" seems irrelevant.

The use of mutant embryos to understand gene interactions has been a very powerful tool, especially in Drosophila. As the reviewer highlights, a bcd null mutant is a severe perturbation, yet we still observe correlation in phenotype with embryo length. The large variability in bcd null mutant phenotypes has been discussed previously, but crucially, here, we demonstrate that this phenotypic variability is not random. Instead, the severity of phenotypes directly correlates with embryo length. This suggests that the ability of the gap gene network – which is largely still there in the absence of Bcd – to scale to embryo size is lost under Bcd depletion.

Our results explain a longstanding problem, noticed in papers from the 1980’s: there is large phenotypic variability in bcd null mutants, and this variability is, at least partially, explainable by embryo geometry. This result may well be applicable to other systems, and therefore we believe it should be made clearly in the manuscript. We discuss this point in subsection “AP Patterning of bcd mutants correlates with embryonic length” and in the Discussion.

The authors then go on to apply another set of dramatic perturbations – rather than increasing Bcd a little, they overexpress it 6x or 10-fold, the latter again being lethal.

The Drosophila embryo is known to be remarkably robust to increases in Bcd dosage. Namba et al.,1997 reported that embryos developed into larvae even with 6x- Bcd levels. Here, we apply 4x, 6x and 10x perturbations. Given the robustness of the system, we can characterize these as mild, middle, and strong perturbations respectively. Therefore, we are exploring a reasonable range of perturbation. For example, in revised Figure 3E we clearly show the relationship between Bcd overexpression and embryo geometry on pattering outcomes.

By this point I personally was not buying the story at all – but finally, the paper gets to Figure 5 and subsection “The gene network breaks at susceptible point in decanalized conditions” onwards, which finally turned everything around. I found that result very strong and interesting. But if I were a reader rather than a reviewer, I would not have made it that far.

We have improved the discussion of the above points to make their scientific importance more clear. Secondly, we have moved the material from Figure 5 into revised Figures 3 and 4. We have given greater prominence to these results and explained their significance in more detail.

My proposal: if the point made in the Introduction/Discussion section is indeed the intended focus of the narrative, I believe the paper could make it point much more persuasively, while also being shorter:

• Introduce the fat2RNAi trick that allows generating embryos with a much more varied geometry than normal, and explain how that offers a fantastic window into studying canalization

We have reordered the manuscript to make the fat2RNAi results with normal Bcd levels more obvious – this now constitutes revised Figures 1 and 2.

• Observe that, absent other perturbations, the development is capable of buffering that "geometric" variability. Move supplementary panels from Figure 3 supplement to the main text.

We have included this data in revised Figure 2.

• Stress the system by increasing bcd dosage (4x, 6x), and proceed through the argument at the end of the Results section.

We have now brought these results forward, and they form the basis of revised Figures 3 and 4.

This is the narrative that actually makes the point set up in the Introduction, and this point is a strong one: the construct developed by the authors allows directly probing the ability of a system to buffer variations – in a very clean, controlled setting, of which I know few. It is a remarkable setting and beautiful result. Only after this point is made, can the response to bcd removal really be seen as further reinforcing the authors' point – not before, because such dramatic perturbation is way beyond anything relevant for normal development and therefore canalization; on its own, this line of evidence alone would be questionable. E.g. last paragraph of subsection “Embryonic length dictates segmentation gene pattern in the absence of bcd” – these discussions are useful for mapping out who activates who in a pathway, but entirely irrelevant for normal development and "what canalization evolved for", surely! I would in fact consider largely moving all the discussions of non-viable mutants to the supplement.

We understand the reviewers point, but we do feel that the depleted Bcd condition provides important insights that cannot be gleaned from the over-expression mutants. For example, as stated above, these results provide an important insight into how the large phenotypic variability often seen in mutant conditions of key patterning genes may have (at least to some degree) a systematic cause related to embryo size. In the revised manuscript, these results now constitute revised Figures 5 and 6.

Other specific issues:

Moderate points:

1) Paragraph two of subsection “Embryonic length dictates segmentation gene pattern in the absence of bcd” unexpectedly bring up the question of expression pattern scaling and I'm not sure whether this is a good move. Exactly what is the paper contributing to that conversation? It seems like a distraction. I suggest removing.

We agree that where this point was raised in the original manuscript was confusing. We do feel that an important point here is being discussed regarding how scaling occurs. Therefore, we have moved this part in to the Discussion and reworked it to highlight the key points.

2) "Genetic network breaks at susceptible point" – what does this sentence mean? How does a genetic network "break"? What is a "susceptible point"? Is this a result, or some intuition the authors are trying to communicate? The strongest point in this paper is introduced by the weakest/meaningless header. Similarly, when the Abstract is revised, I would suggest removing the last sentence referring to "vulnerable points in the network" which again sounds like some intuition (belonging to Discussion section) rather than a finding.

We have reworked this section, including the header. By “susceptible”, we meant the point in the patterning network where we most often see patterning defects under perturbation (either to embryo size or Bcd expression levels, or both). However, we agree that the original wording was not clear. As suggested, we have removed these words from the Abstract and Results, and elaborated these concepts in the Discussion.

3) Second paragraph of subsection “Measurement of embryonic geometry”: I found this hard to parse (particularly the use of the term "further"). My interpretation is that EL pre-fixation is inferred from EL post-fixation assuming that aspect ratio is unchanged – correct? This should be clarified. But even more importantly – was this assumption of isotropic shrinkage verified? In my experience a harsh fixation protocol can lead to extensive deformations, their isotropic nature is not obvious, and given the role played by the aspect ratio in this paper, this seems an important point to demonstrate. [This is the one place where additional data may be required, but shouldn't be too onerous to obtain]

We thank the reviewer for raising this important point which was insufficiently explained in the original submission. As described in the Materials and methods section, we fix the embryos in heptane saturated by 37% paraformaldehyde for 1 hr, after which the embryos are devitellinized manually. As this fixation method does not involve any vortex or heat shock step, it generally preserves embryo morphology. The fixed embryos only shrink during mounting, when the mounting medium is added (Aqua-Poly/Mount).

In the revised manuscript, we have carried out quantifications to verify that such shrinkage is close to isotropic by nature. We measured the length and width of OreR and fat2RNAi embryos, in both live and fixed conditions (n = 239 each condition). We found that the aspect ratio of the OreR embryos remains comparable prior and post fixation (Figure 1—figure supplement 1B-C). This suggests isotropic shrinkage and supports our approach to use aspect ratio of post-fixation embryos to estimate their original embryonic length.

We do see a smaller aspect ratio of fat2RNAi embryos post fixation. However, this can be explained by the differential collection of embryos. While the live condition measurements are carried out on a stereoscope, where we collected fat2RNAi embryos spanning all geometrical range, the fix condition measurements are obtained using confocal staining data, where we focused on imaging fat2RNAi individuals of extremely short length.

4) SDD simulation: I am very confused.

a) If I use an cylinder to approximate an ellipsoid, then geometrically, the "best approximation" cylinder is surely narrower than the embryo. Not wider. It seems like width is used as a free fitting parameter to make the data fit better, but the text tries to make it sound like this larger-than-actual radius was somehow expected, and I don't understand the argument.

In the original submission we tried to keep the model as simple as possible, while including 3D effects of embryo size. However, it is clear that our approximation was unhelpful. In the revised manuscript we now perform our simulations on the surface of an ellipsoid – a much better approximation to the embryo geometry.

b) Further, I believe the diffusion parameters, both in the bulk and at the surface, for a small molecule like Bcd or Hb, were previously measured in the embryo specifically. So it sounds like this could be a zero-parameter fit. How bad is it? Why the discrepancy?

The reviewer is correct; effectively this is a zero parameter fit. Taking the measured diffusion coefficient and degradation time from Durrieu et al.,2018 and accounting for the eGFP folding time (~50 minutes), we are able to fit reasonably well both the Bcd concentration profile in n.c. 13 for both wild-type and fat2RNAi embryos by simply altering the embryo geometry and no other parameters.

We now present this data more clearly in Figure 1, making assessment of fit quality easier.

c) And in particular, if using an elliptical geometry truly does not significantly alter the results, why not use it, given that the parameters of the actual geometry are, again, known?

We have now included the full elliptical solution.

d) Does the SDD simulation also reproduce the features that the plot of Figure 3G specifically highlights, namely the larger concentration in the posterior and the crossing point? The panel D in Figure 3 supplement does not allow to see this.

We still observe the crossing point in concentrations between wild type and fat2RNAi embryos (when the AP axis is scaled), and the higher Bcd concentration in the posterior for fat2RNAi embryos. In the revised Figure 1 this is clearer.

Reviewer #2:

Huang and Saunders investigate phenotypic variability of bicoid mutant Drosophila embryos. Remarkably, they find that in mutant embryos, the number of engrailed stripes and the number of abdominal segments is determined by the length of the embryo. This rule applies to embryos with abnormally short AP axes generated by knockdown of fat2. In contrast, in the presence of Bcd, the correct number of segments always arise, regardless of embryo length, even in fat2 KD embryos. Thus, Bcd is required for scaling of expression patterns to accommodate variation in embryo length. Measurements of Bcd-GFP strongly suggests that this scaling is achieved by reducing the amplitude of the Bcd gradient concomitant with the reduction in egg length. By examining gap gene expression, they show that in short bcd mutant eggs, the distance between anterior and posterior poles must be sufficiently large to permit the expression of the genes Kruppel and knirps. Their data are consistent with a view of the bicoid mutant phenotype in which Torso signaling is not scaled in accord with embryo length, so that terminal signaling in short embryos overrides factor(s) that would otherwise permit the expression of Kr and knirps. Finally, the authors show that excess Bcd also disrupts patterning by interfering with specification in the vicinity of the fourth abdominal segment, with additional segments affected in shorter embryos.

The work is exciting because it articulates an explanation for phenotypic variability in bcd mutants. The manuscript is significant in showing that the Bcd gradient itself provides a buffer against variation in egg length. The main finding, at least to this reviewer, is that Bcd itself is required for length scaling. The authors should draw greater attention to this result. This result seems consistent with the work of Jun Ma, whose publications should be cited (e.g., doi: 10.1242/dev.064402).

Our results are consistent with previous studies (Cheung et al., 2011, 2014) showing that the scaling of embryonic patterns can be traced back to the scaling of the Bcd gradient itself. Although the phenomena are similar, the mechanisms underlying the adaptive Bcd gradient formation are different. This is because the macroscopic variable that the patterning system is buffering against is different (volume and geometry, respectively). Embryos adapt to volume variation by altering the amount and distribution of mRNA deposition (Cheung et al., 2011). In our case, geometry changes but not (substantially) volume. In our case, we found that SDD model is sufficient to generate different Bcd profiles in embryos of different geometry without altering any other dynamic parameters. We have improved discussion of this point, and it is included within Figure 1.

Interestingly, scaling occurs in fat2-depleted embryos despite the paradoxical finding of more bcd mRNA upon fat2 KD (Figure 5—figure supplement 1). This suggests alteration in Bcd translation and/or localization of mRNA in short eggs.

In our new FISH data, we do not see a significant change in the distribution of bcd mRNA between wild-type and fat2RNAi embryos. Though it is challenging to quantify FISH data, our results do not reveal an obvious change in the bcd mRNA levels in fat2RNAi embryos.

Further, in the revised manuscript we are able to fit the differing Bcd profiles between wild-type and fat2RNAi embryos by simply accounting for embryo geometry – i.e. no change in production, diffusion or degradation of Bcd. Therefore, our evidence currently suggests that the translation and/or localization of Bcd is not significantly altered in fat2RNAi embryos.

These results are broadly consistent with Cheung et al.,2011. In that paper, they show that Bcd concentration is related to embryo volume. In our case, even the shortest embryos are still ~90% of the volume of wild-type embryos (assuming ellipsoidal geometry). Therefore, our observed stability of the Bcd production in the different embryos is consistent.

The manuscript would be strengthened if the authors could comment further about the nature of Bcd-mediated canalization, particularly to explain why more mRNA generates less protein.

See answer above and also below regarding the new FISH data.

Along these lines, more detailed description of the simulation would be helpful. In particular, it would strengthen the manuscript to know whether bcd mRNA localization might play a role in canalization. What geometry of bcd mRNA was used in the simulation?

We have improved the simulations. We now use an ellipsoidal geometry. From the FISH data, we do not see a significant shift in the bcd mRNA profile. Therefore, in the simulations we use the same production domain for both wild-type and fat2RNAi embryos.

It should not be beyond the technical limits of the lab to perform a semi-quantitative in situ hybridization for bcd mRNA to address the question of whether bcd mRNA distribution is changed by fat2 KD in a way that helps explain Bcd-mediated scaling.

We agree with the Reviewer and have undertaken this experiment. The fixation protocol required optimization, which took a few months. We performed FISH on wild-type, fat2RNAi and 10x-bcd embryos. In the latter case, we see a clear increase in both the magnitude and extent of bcd mRNA. For the other two scenarios the profile of bcd mRNA is comparable.

As the Reviewer rightfully highlights, this analysis is only semi-quantitative. To compare embryos, we normalized signal by the background signal apparent in embryos ~120μm from the embryo anterior pole. We cannot definitively state that the wild-type and fat2RNAi embryos have the same bcd mRNA profiles. However, it is unlikely that any variations are large; which contrasts with the 10x-bcd line, which has clear change in both magnitude and profile extent.

This data is now included in Figure 1, Figure 1—figure supplement 1 and Figure 3—figure supplement 1.

As a final major comment, the authors document variability in gap gene boundaries in control embryos, fat2 KD, and bcd mutants (Figure 1 J-M, Figure 1—figure supplement 1, and Figure 3—figure supplement 1). The authors should address how much of this variability arises from variability in egg length. The LlamaTag experiment suggests that at some of the variability originates from differences in egg length (this is clear from the anticorrelation in Figure 3I). If egg length determines boundary positions, then there should be a relationship between egg length and gap gene boundary position in bcd mutants. The degree of correlation might be even stronger in bcd mutants due to the absence of the canalizing activity of Bcd. The authors should plot boundary position for each gap gene boundary as a function of egg length, similar to Figure 3I.

We have plotted the relative boundary positions against embryonic length for all four gap genes in bcdKO and bcdKO, fat2RNAi embryos (Figure 6—figure supplement 2). Most of the relative boundary positions show strong correlation with embryonic length, demonstrating the loss of scaling.

[Editors' note: further revisions were suggested prior to acceptance, as described below.]

Reviewer #2:

The revised manuscript will be satisfactory for publication if the authors can clarify the following points listed below regarding the simulation of Bcd-GFP production. The main point of Figure 1F-K is to address how the Bcd gradient changes in embryos with altered geometry. My concern is that the presentation of the simulation in the text may appear at first glance incongruent with the figures. In the end I believe the figures are consistent with the text, but I would like some clarification. I imagine what I am requesting will be trivial for the authors to provide, but at the same time very useful and insightful for readers.

1) The authors state that the same parameter values describe the distribution of fluorescent Bcd-GFP regardless of embryo geometry. Unless I am mistaken, this means that, at steady-state imposed by the model (and using the finding that bcd mRNA doesn't change very much in terms of its total amount), the total amount of Bcd-GFP embryo-wide is equivalent in all embryos regardless of their geometry. Further, the total amounts of fluorescent and non-fluorescent Bcd-GFP are also equivalent. It would be helpful to explicitly state this in the text for those not immediately familiar with modeling and/or may not be aware that the maturation time of GFP is non-negligible in this context.

Yes, the reviewer interpretation is correct. Because the fat2RNAi embryos are bigger in embryonic width than wild-type embryos, the change in volume is not large despite the change in the anterior-posterior length. Therefore, from our model, the total amount of Bcd in the embryos does not change substantially. We have clarified this in the text during discussion of the modelling (subsection “Developmental reproducibility is preserved with a minor impact on scaling under geometrical perturbations”).

2) Assuming that Bcd-GFP amount is the same in all embryos, a reader will then examine Figure 1G and note that the summed nuclear intensity (or in some sense, the area under the curve) does not appear to be the same in embryos of different sizes. Instead it appears that Bcd-GFP scales with embryo length. This appears to contradict the idea that total BcdGFP is unchanged. However, the authors say that this is due to "dilution" of Bcd-GFP. I interpret this to mean that if one considers an embryo as a set of "salami slices," then, due to the altered aspect ratio, a slice at some absolute distance from the anterior pole in fat2 KD (or, more generally, in short) embryos has a larger volume compared to a slice at the same distance in wild-type. The apparent reduction in "nuclear concentration" of Bcd-GFP occurs because whatever amount of Bcd-GFP is available at that distance is roughly the same between long and short embryos. This amount distributes into the available volume, so that the apparent concentration of BcdGFP appears lower in fat2 KD. Could the authors please state this more explicitly, if my interpretation is correct. I believe that splitting the long sentence the fifth paragraph of subsection “Developmental reproducibility is preserved with a minor impact on scaling under geometrical perturbations” into two or three sentences will suffice, devoting a sentence or two to each of the two differences (low anterior, high posterior) in the gradients of long and short eggs. This will help the reader to understand why the 1D plot of Bcd-GFP vs AP does not accurately represent the total amount of Bcd and results in the reduction in nuclear intensity in the anterior.

Yes, the reviewer interpretation is correct. We appreciate the advice on improving the description and we have clarified in the text (subsection “Developmental reproducibility is preserved with a minor impact on scaling under geometrical perturbations”).

3) Figure 1 presents data and simulation regarding only fluorescent Bcd-GFP. A powerful aspect of the simulation is that it can estimate the quantity that we (and the embryo) actually care about, the distribution of _total_ Bcd. Given the sizable maturation time of 50 minutes used in the simulation, my intuition is that the invisible proportion likely extends over a significantly larger fraction of egg length in short embryos compared to long. However my intuition may well be wrong. Could the authors please present and discuss the estimated _total_ Bcd-GFP distribution as a function of both absolute and normalized distance in embryos of various aspect ratios (lengths)? To my thinking, this is the only way to address the question of how the actual gradient (not just the visible one) changes with changing geometry.

We agree that the relevant readout biologically is the local concentration of mature Bcd protein regardless of the folding state of the tagged GFP. However, we do not have data on the folding time of Bcd protein itself and so it is not straightforward to infer the functional Bcd concentration from the experimental data. In the modelling we now test two different scenarios. First, we assume Bcd folds at the same rate as eGFP, as shown in Figure 1. We also include in Figure 1—figure supplement 1G, the total nuclear Bcd concentration (assuming immediate folding of Bcd) as a function of absolute or normalized embryonic length. The real functional Bcd gradient lies between these two curves.

Interestingly, the profiles of long and short embryos intersect at a relatively more anterior position when we consider the total Bcd concentration (disregarding protein folding time). Intersection of the profiles now occurs around 25% EL vs. ~40% EL in our observed experimental profiles (which are affected by protein folding rates). To double-check our simulations, we used a simpler cylindrical geometry where we could analytically solve the SDD model to further validate this observation. Taking diffusion to only occur on the body of cylinders of equal surface area (ignoring ends) but different lengths (350um and 500um), we also see an anterior shift in the crossover position when considering the total Bcd concentration, Author response image 1. We discuss these results in paragraph five of subsection “Developmental reproducibility is preserved with a minor impact on scaling under geometrical perturbations”.

Author response image 1
Download asset Open asset
Bcd gradient at steady-state using SDD model on surface of a cylinder (blue: L = 500μm, r = 100μm; red: L =350μm, r = 143μm).

Bcd is introduced at rate J at the anterior end of the cylinder. We exclude the ends of the cylinder. (A) Bcd gradient assuming immediate protein folding (total Bcd). (B) Bcd gradient accounting for protein folding time. Dashed lines highlight how the crossover point shifts due to accounting for folding..

For example, after estimating the gradients of total Bcd, could the authors now show the difference in total Bcd concentration, as in Figure 1K lower panel, except now for total Bcd. (Could the authors also include error bars on the difference in 1K and in a new plot regarding total concentration?)

We have replotted Figure 1K to include the experimental data with error bars. We have included a similar plot for the theory in Figure 1—figure supplement 1G. In summary, the parameter of protein folding time does not affect the relative shape of the gradient in long vs short embryos but does alter the relative position where curves intersect.

Moreover, the difference in total Bcd can be converted into a difference in fractional egg length ("positional error") as a function of normalized AP position, i.e., the difference in the positions at which the same Bcd concentration is found in short eggs compared to long. This difference is zero near the midpoint of the AP axis where the curves cross and becomes larger moving toward either pole. This will give the reader a sense of magnitude of the mistakes in assigning positional identity that we would expect in short embryos if Bcd were the only patterning cue. In turn, these differences can be compared to the changes in the placement of gap gene expression boundaries presented in Figure 2. Are the changes in the Bcd gradient predictive of the changes in boundary placement (in other words, do gap gene boundaries arise at the same Bcd concentration in long and short eggs)? Do gap gene boundaries shift more, less, or the same amount as predicted by the changes in the Bcd gradient? The answers will help address how the altered Bcd gradient in short embryos generates (or works against!) scaling, and how much of the scaling phenomenon must arise independently of the Bcd gradient.

This is a challenging question to answer. As shown by a number of labs, the positional information transfer from the Bcd gradient to gap gene expression is a dynamic and history dependent process. Liu et al.,2013, have shown that the gap gene boundary positions only correlate with absolute local Bcd concentration at early stages (10 min into n.c. 14), while in late stages (50 min into n.c. 14) the boundary positions depend on combinatorial inputs from maternal gradients and the gap genes themselves. Here Liu et al.perturb the Bcd gradient by altering maternal dosage, which changes mainly the amplitude but not the characteristic length of the gradient. However, how gap gene expression temporally integrates a gradient of different characteristic length (as in this study) remain unclear. Our results (Figure 2) only demonstrate that at late stages, the gap gene boundaries largely recapitulate the wild-type scenario, even with an altered Bcd gradient due to geometrical perturbation. But, providing precise predictions for the gap gene shifts due to Bcd perturbation is difficult without a full model of gap gene formation. We have in fact tried tackling this problem using the model published by Bieler et al., 2011. However, there are many biological and technical challenges in fully modelling such interactions and this represents a new piece of work. Indeed, we have had discussions with modelling groups, such as Manu, to explore this question further. We have added a sentence to the Discussion to highlight this interesting point.

https://doi.org/10.7554/eLife.47380.sa2

Article and author information

Author details

  1. Anqi Huang

    Mechanobiology Institute, National University of Singapore, Singapore, Singapore
    Contribution
    Conceptualization, Formal analysis, Validation, Investigation, Visualization, Methodology, Writing - original draft, Writing - review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-0551-1160

  2. Jean-François Rupprecht

    1. Mechanobiology Institute, National University of Singapore, Singapore, Singapore
    2. CNRS and Turing Center for Living Systems, Centre de Physique Théorique, Aix-Marseille Université, Marseille, France
    Contribution
    Formal analysis, Investigation, Writing - review and editing
    Competing interests
    No competing interests declared

  3. Timothy E Saunders

    1. Mechanobiology Institute, National University of Singapore, Singapore, Singapore
    2. Department of Biological Sciences, National University of Singapore, Singapore, Singapore
    3. Institute of Molecular and Cell Biology, Proteos, A*Star, Singapore, Singapore
    Contribution
    Conceptualization, Resources, Formal analysis, Supervision, Funding acquisition, Visualization, Writing - original draft, Writing - review and editing
    For correspondence
    dbsste@nus.edu.sg
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-5755-0060

Funding

National Research Foundation Singapore (NRF2012NRF-NRFF001-094)

  • Timothy E Saunders

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

We thank Sally Horne-Badovinac and Hernan Garcia for fly lines. We thank Alexis Kerh for help with fly work. We thank Jean-Paul Vincent for support in revising the manuscript. This work was supported by a National Research Foundation Singapore Fellowship awarded to TES (NRF2012NRF-NRFF001-094) and funding from the Mechanobiology Institute, National University of Singapore, Singapore.

Senior and Reviewing Editor

  1. Naama Barkai, Weizmann Institute of Science, Israel

Version history

  1. Received: April 3, 2019
  2. Accepted: February 11, 2020
  3. Accepted Manuscript published: February 12, 2020 (version 1)
  4. Version of Record published: February 20, 2020 (version 2)

Copyright

© 2020, Huang et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 1,964
    Page views
  • 254
    Downloads
  • 19
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Open citations (links to open the citations from this article in various online reference manager services)

Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)

  1. Anqi Huang
  2. Jean-François Rupprecht
  3. Timothy E Saunders
(2020)
Embryonic geometry underlies phenotypic variation in decanalized conditions
eLife 9:e47380.
https://doi.org/10.7554/eLife.47380

Categories and tags

Research organism

Further reading

    1. Developmental Biology

    Dynamic compartmentalization of the pro-invasive transcription factor NHR-67 reveals a role for Groucho in regulating a proliferative-invasive cellular switch in C. elegans

    Taylor N Medwig-Kinney, Brian A Kinney ... David Q Matus
    Research Article

    A growing body of evidence suggests that cell division and basement membrane invasion are mutually exclusive cellular behaviors. How cells switch between proliferative and invasive states is not well understood. Here, we investigated this dichotomy in vivo by examining two cell types in the developing Caenorhabditis elegans somatic gonad that derive from equipotent progenitors, but exhibit distinct cell behaviors: the post-mitotic, invasive anchor cell and the neighboring proliferative, non-invasive ventral uterine (VU) cells. We show that the fates of these cells post-specification are more plastic than previously appreciated and that levels of NHR-67 are important for discriminating between invasive and proliferative behavior. Transcription of NHR-67 is downregulated following post-translational degradation of its direct upstream regulator, HLH-2 (E/Daughterless) in VU cells. In the nuclei of VU cells, residual NHR-67 protein is compartmentalized into discrete punctae that are dynamic over the cell cycle and exhibit liquid-like properties. By screening for proteins that colocalize with NHR-67 punctae, we identified new regulators of uterine cell fate maintenance: homologs of the transcriptional co-repressor Groucho (UNC-37 and LSY-22), as well as the TCF/LEF homolog POP-1. We propose a model in which the association of NHR-67 with the Groucho/TCF complex suppresses the default invasive state in non-invasive cells, which complements transcriptional regulation to add robustness to the proliferative-invasive cellular switch in vivo.

    1. Chromosomes and Gene Expression
    2. Developmental Biology

    Transcription: A hub of activity

    Virginia L Pimmett, Mounia Lagha
    Insight

    Imaging experiments reveal the complex and dynamic nature of the transcriptional hubs associated with Notch signaling.

    1. Cell Biology
    2. Developmental Biology

    Cylicins are a structural component of the sperm calyx being indispensable for male fertility in mice and human

    Simon Schneider, Andjela Kovacevic ... Hubert Schorle
    Research Article

    Cylicins are testis-specific proteins, which are exclusively expressed during spermiogenesis. In mice and humans, two Cylicins, the gonosomal X-linked Cylicin 1 (Cylc1/CYLC1) and the autosomal Cylicin 2 (Cylc2/CYLC2) genes, have been identified. Cylicins are cytoskeletal proteins with an overall positive charge due to lysine-rich repeats. While Cylicins have been localized in the acrosomal region of round spermatids, they resemble a major component of the calyx within the perinuclear theca at the posterior part of mature sperm nuclei. However, the role of Cylicins during spermiogenesis has not yet been investigated. Here, we applied CRISPR/Cas9-mediated gene editing in zygotes to establish Cylc1- and Cylc2-deficient mouse lines as a model to study the function of these proteins. Cylc1 deficiency resulted in male subfertility, whereas Cylc2-/-, Cylc1-/yCylc2+/-, and Cylc1-/yCylc2-/- males were infertile. Phenotypical characterization revealed that loss of Cylicins prevents proper calyx assembly during spermiogenesis. This results in decreased epididymal sperm counts, impaired shedding of excess cytoplasm, and severe structural malformations, ultimately resulting in impaired sperm motility. Furthermore, exome sequencing identified an infertile man with a hemizygous variant in CYLC1 and a heterozygous variant in CYLC2, displaying morphological abnormalities of the sperm including the absence of the acrosome. Thus, our study highlights the relevance and importance of Cylicins for spermiogenic remodeling and male fertility in human and mouse, and provides the basis for further studies on unraveling the complex molecular interactions between perinuclear theca proteins required during spermiogenesis.