Embryonic geometry underlies phenotypic variation in decanalized conditions

During development, many mutations cause increased variation in phenotypic outcomes, a phenomenon termed decanalization. Such variations can often be attributed to genetic and environmental perturbations. However, phenotypic discordance remains even in isogenic model organisms raised in homogeneous environments. To understand the mechanisms underlying phenotypic variation, we used as a model the highly precise anterior-posterior (AP) patterning of the early Drosophila embryo. We decanalized the system by depleting the maternal bcd product and found that in contrast to the highly scaled patterning in the wild-type, the segmentation gene boundaries shift away from the scaled positions according to the total embryonic length. Embryonic geometry is hence a key factor predetermining patterning outcomes in such decanalized conditions. Embryonic geometry was also found to predict individual patterning outcomes under bcd overexpression, another decanalizing condition. Further analysis of the gene regulatory network acting downstream of the morphogen identified vulnerable points in the networks due to limitations in the available physical space.


Introduction
The phenomenon of canalization describes the constancy in developmental outcomes between different individuals within a wild-type species growing in their native environments [1][2][3]. 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 [4][5][6]. In particular, the inter-individual variation of the positional information conferred by gene expression is below the width of a single cell [4,7]. This means that these developmental 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 [8] or the manifestation of the intrinsic properties of the underlying complex gene regulatory networks [9]. Canalization can break down in individuals subjected to aberrant genetic mutations or extreme environmental conditions [10,11]. 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 [12]. This indicates that other components besides genetic and environmental variation cause phenotypic discordance under decanalized conditions. Previous work has proposed that stochastic expression of redundant genes predicts the developmental outcome of the mutant individuals [13]. However, in many other cases, it remains elusive as to why mutation increases inter-individual phenotypic variation and what alternative components underlie such variation [14,15]. Therefore, to identify the potential sources of variation that govern phenotypic variation, we utilized early Drosophila embryonic patterning. The developmental process was decanalized by either knocking out the maternal bicoid (bcd) gene or introducing aberrantly high amplitude of the Bcd morphogen gradient, both resulting in increased inter-individual variation of patterning outcomes [16,17].
We found that the naturally variable embryonic geometry acts as a previously unidentified source of variation that predetermines the individual phenotype.
Further, we found that specific segmentation identities are preferentially affected in decanalized conditions, which is informative of the epigenetic interactions of the underlying developmental genes.

Lack of maternal bcd results in increased variation in patterning outcomes
The cuticle phenotype of maternal bcd mutants was previously characterized using a series of different bcd alleles, with increasing allelic strength showing defects reaching further into posterior regions of the embryonic patterning [16,18].
Phenotypic variation was also observed among embryos derived from females carrying the same mutant allele [16,18]. For example, the amorphic bcd E1 allele gives rise to embryonic patterning lacking the head and thoracic structures with 100% penetrance and replaced by duplicated posterior spiracles. Comparatively, the patterning outcomes ranging from abdominal segments 1 (A1) to A5 are highly variable among different individuals, manifesting in either fusion or depletion of various number of denticle belts. Meanwhile, segments posterior to A6 (inclusive) always remain intact.
To systematically understand the inter-individual phenotypic variation among bcd mutants, we utilized an allele generated by the CRISPR-MiMIC method [19,20]. 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 bcd KO ).
Moreover, the MiMIC construct contains an eGFP reporter, which facilitates further genetic manipulation, such as recombination, carried out in this study.
The cuticular pattern of the bcd KO allele qualitatively recapitulates that of bcd E1 ( Fig   1A-F, Fig S1A). 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 (Fig 1G). Structures indicating partially differentiated abdominal segments can be observed in the anterior regions of the bcd KO embryos. However, these structures do not recapitulate any of the wild-type denticle belts (Fig 1A-F).
Further, we observed and classified the variable phenotypes of the duplicated posterior spiracles according to the completeness of the organ morphogenesis ( Fig   1H-I). Interestingly, the fully developed ectopic spiracle organ (Fig 1H(i)) is only observed in embryos showing less than six intact abdominal segments, while most frequently observed in individuals with three intact abdominal segments (Fig 1G, dashed bar area). This indicates correlation between the patterning outcomes of the abdominal regions and the ectopic spiracles.
The wide spectrum of bcd KO cuticle phenotypes can be attributed to the variation of patterning gene expression during the blastoderm stage. In the absence of a maternal Bcd gradient, the antero-posterior (AP) axis of the embryos is patterned by the remnant maternal positional cues including: the anterior Hunchback gradient; the uniformly distributed Caudal; the posteriorly localized Nanos; and the Torso pathway activated at both terminal regions [21,22]. 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 (Fig 1J-M). Importantly, in the wild-type condition, the inter-individual variation of most boundary positions is at the range of 1% embryonic length (EL), with an exception being the posterior boundaries of Giant (Gt) and Knirps (Kni), which have positional variation slightly over 1.5 % EL [7]. In comparison, the absence of Bcd activity results in significantly increased variation in gap gene boundary positions (Fig S1B, p<0.01 for all measured boundaries to have increased error randomly), with the anterior Kni boundary being the most variable, showing variation above 5% EL (Fig 1J-M). Occasionally we detected no Krüppel (Kr) nuclear intensity in the presumed expression region in bcd KO individuals (2 out of 15 individuals), indicating the failure of Kr gene activation in these embryos (Fig S1C). 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 (Fig 1M, asterisk; Fig S1D). It is noteworthy that embryos derived from a single pair of bcd KO parents raised in constant environmental conditions show equivalent phenotypic variation (Fig S1E and F), suggesting that the inter-individual variation observed cannot be attributed to differences in either environment or genetic background.

AP Patterning of bcd mutants correlates with embryonic length
We next wanted to understand the causes of reduced developmental reproducibility in the absence of Bcd-instructed positional information and identify potential sources of variation that predetermine patterning outcomes of each mutant individual. We monitored the dynamic expression of the segment-polarity gene engrailed (en) in different bcd KO individuals. All the control embryos show invariably eleven En stripes, demarcating the posterior boundary of each body segment including three thoracic (T1~T3) and eight abdominal segments (A1~A8; Fig 2B). In contrast, different numbers of En stripes are activated in bcd KO embryos, consistent with the variable number of intact denticle belts that we observed previously. This number remains unchanged throughout embryogenesis. We also measured the geometrical parameters of each embryo (Fig 2A). Interestingly, we found that the number of En stripes positively correlates with the embryonic length along the AP axis (Fig 2C-G, Video   1). These results led us to hypothesize that, in the absence of maternal bcd activity, the number of body segments generated during pattern formation is dependent on the embryo absolute length. Therefore, embryonic geometry could be a previously unrecognized source of variation that underlies phenotypic variation in patterning outcomes of mutant individuals.

Developmental reproducibility is preserved in a wide range of embryonic geometry
To test our hypothesis, we first asked if altering embryonic geometry alone can alter embryonic pattern formation. 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 [23,24]. This process is mediated by the planar cell polarity (PCP) 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 PCP core components, atypical cadherin Fat2 in the follicle cells [25], and we henceforth refer to these as fat2RNAi embryos. This reduces the embryonic length from 510 ± 17 (s.d.) µm in wild type to 432 ± 40 µm in fat2RNAi embryos ( Fig 3A). Meanwhile, the perturbed embryos show an increased embryonic width (EW) along the dorso-ventral axis (Fig S2A and B, 196 ± 5 µm in wild type and 202 ± 11 µm in fat2RNAi). Together, these geometrical variations lead to only a slight reduction ~8 % in the embryonic size (assuming an ellipsoidal geometry) compared to wild type embryos ( Fig 3B). Depletion of the bcd gene product does not affect embryonic geometry (Fig 3A and B). The round eggs of fat2RNAi embryos are fertilizable and continue with proper embryogenesis (Video 2).
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 ( Fig 3C; Fig S2C). 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.
Next, we investigated how establishment of the Bcd morphogen gradient is influenced by the altered embryonic geometry in fat2RNAi embryos. 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 ( Fig 3D). We found that the absolute Bcd concentration is lower in the anterior half of the embryo in fat2RNAi individuals compared to that of control (Fig 3E-F). 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 ( Fig 3G). These observed Bcd gradient profiles in different embryonic geometry are consistent with the SDD model [26] of Bcd gradient formation in different geometries (Methods and Fig S2D).
We predicted that the 'flattening' of the Bcd profile in response to decreasing aspect ratio of the embryo should result in a slight but measurable shift in the downstream gap gene expression domains. We first measured Hunchback (Hb) expression boundary at mid n.c. 14 using live imaging of hb>LlamaTag [27]. In control embryos, the Hb expression boundary locates at 49.0% EL with variation of 1.3% EL, consistent with previous reports [28]. Comparatively, the Hb boundary shows a These results suggest that, when we manipulate the embryonic geometry to an extent beyond that naturally observed, the reproducibility of the patterning outcomes is preserved as compared to that of the wild-type scenario. Therefore, the intact early embryonic patterning network is highly robust to variations in embryonic geometry.

Embryonic length dictates segmentation gene pattern in the absence of bcd
We now return to our hypothesis that embryonic geometry is a potential source of variation that predetermines phenotypic outcomes of mutant individuals. We introduced the fat2RNAi knockdown into a bcd KO background to see how embryonic patterning is affected accordingly. The cuticle pattern of embryos derived from fat2RNAi, bcd KO females resembles that of bcd KO 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 (Fig 4A and B). In individuals with 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 ( Fig 4F-H, 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 [33] 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 [22]. 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 (Fig S3B-D).
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 (Fig 4 F-H, 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 pairrule gene expression pattern ( Fig S3E).

The gene network breaks at susceptible point in decanalized conditions
Phenotypic discordance has been previously observed as a consequence of artificially altered maternal bcd dosage, with an increasing number of bcd gene insertions leading to a larger proportion of individuals showing defective patterning [17]. Hence, we wanted to know if embryonic geometry also underlies inter-individual variation under decanalized conditions of bcd overexpression. To effectively 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 ( Fig 5A). 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 that of the wild-type amount, respectively ( Fig 5B; Fig S4A). 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 [6], as manifested by the posterior displacement of cephalic furrow position ( Fig 5B).
The ~5-fold bcd overexpression compromises viability to adulthood and the nonhatched embryos displayed a plethora of defective patterning phenotypes (Fig 5C; Fig   S4B). Individuals with mild defects frequently displayed missing or fused denticle belts in A4 segment (Fig 5C-D), a positional bias that has been reported previously [17]. 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 (Fig 5C-D; Fig S4C). Patterning defects were rarely seen in 6x bcd embryos unless fat2RNAi knockdown is further introduced into this genetic background (Fig 5C-D; Fig S4D). A large percentage of these individuals showed abdominal patterning defects, with A4 being the most susceptible position (Fig 5C-D; Fig S4B). Interestingly, a similar distribution of defective abdominal segments is also seen in the small proportion of non-hatched fat2RNAi individuals (Fig 5C-D).
To understand if embryonic geometry predetermines the severity of phenotypic defects in individuals with bcd overexpression, we characterized the patterning outcomes using En expression in embryos with various bcd copy number and embryonic length. Individuals with 4x bcd (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 (Fig 5E, top  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 the position of lost cell fates is not stochastic, but originates at and expands from 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 geometry (Fig S4E), the absolute distance between neighboring gap gene expression peaks decreases in response to reduced embryonic length. This in turn changes the combinatorial inputs to activate downstream pair-rule genes, e.g. even-skipped (eve). Fig 5G-H illustrate that the expression peaks of Kni and Gt are brought into proximity with gradually reducing embryonic length. As Kni and Gt confine the boundaries of eve stripe 5 [34], the expression of this eve stripe is over-repressed (Fig 5H, 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 (Fig 5G, green crosses; Fig 5H, arrowhead), leading to defects across a broader range.

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 [35,36]. This reflects inter-individual variation in genetic composition and life-history environmental exposure [37]. Such intraspecific individuality may have significant ecological and social impacts on the population [38]. Equally, these same genetic and environmental variations pose challenges to the fundamental developmental processes, as they are to generate invariant developmental outcomes. Multiple evidences suggest that organisms have evolved canalization mechanisms that render developmental processes insensitive to such sources of variation [2,39]. 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 [40,41], genetic variations [42] and temperature perturbations [28,43]. These studies suggest that mechanisms including epistasis, genotype-environment interactions and canalizing gene regulatory networks [44] 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 size. Previous studies have shown that patterning outcomes are highly reproducible and remain scaled to embryonic length [42,45].
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 size is only slightly reduced by ~8 %. These results further support embryonic geometry as a major source of variation that accounts for interindividual phenotypic variation under decanalized conditions. Both embryonic size and embryonic length are inheritable traits and therefore adaptive to artificial selection or environmental changes [42,[46][47][48]. 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 size due to decreasing temperature [47], such directional selection may result in decanalizing effects on the patterning processes [46,49]. 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 [50,51] 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 [52].
Systematic comparisons among different Drosophila species have shown that the regulatory sequences of the segmentation genes are rapidly evolving and thus substantially diverged [53]. 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 [53][54][55]. 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 [56,57]. If the breaking point of patterning processes under decanalized conditions truly depends on the system parameters of the underlying network [58], we expect to see different susceptible points in different network structures. This can be tested by characterizing decanalizing phenotypes in related species.
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 [14,15], or in vitro induction of patterning systems [59,60], 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 environmente.g. 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.

Fly stocks and genetics
The bcd knockout allele (bcd KO ) used in this study was generated by CRISPRmediated insertion of a MiMIC cassette into the first intron of the bcd gene [19,20] The cuticle phenotype of bcd KO was compared to that of the classic bcd E1 allele [61] To generate embryos with artificially reduced aspect ratio, we expressed RNA interference against the fat2 gene using a maternal traffic jam (tj)>Gal4 driver [62] Both UAS>fat2RNAi and tj>Gal4 were either crossed to or recombined with the bcd KO 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 [26] 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 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>fat2RNAi generate 6xbcd embryos with reduced aspect ratio.

Measurement of embryonic geometry
To compare the geometrical parameters between OreR, bcd KO

Cuticle preparation
Embryos of various genotypes were collected during the blastoderm stages and allowed to develop at 25ºC 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
For measurement of Bcd gradient profile, we followed the protocols detailed in Gregor et al. [4]. 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. Images

Simulation of SDD model
We simulated using Matlab the SDD model in steady-state for morphogen concentration, !(!, !) at position x and time t: where D is the diffusion coefficient and α the degradation rate. We approximated the Drosophila embryo as a cylinder, with different lengths (500µm, 400µm) and radii (90µm, 130µm) for the wild-type and fat2RNAi embryos respectively. The effective radius in the simulation for fat2RNAi embryos is slightly larger than measured experimentally (~110µm), reflecting the rounder geometry in these embryos (hence the cylindrical approximation of embryo shape is less accurate). Using an elliptical geometry does not significantly alter these results (data not shown). We account for eGFP folding time (~50 minutes) [63], and the results plotted in Figure S2D are for the folded population of Bcd::eGFP.

Gap gene boundary quantification
Confocal Z-stack images were Z-projected (maximum intensity) in Fiji performed 1000 simulation runs to infer the variability on the boundary precision. 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 bcd KO 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 bcd KO embryos. We calculated the p-value using the paired sample t-test across all boundaries.

Quantification of maternal bcd transcripts
To compare the relative amount of maternally loaded bcd transcripts in different genotypes, we extracted total mRNA from presyncytial embryos (within 1h 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.