To address whether the size difference between AB and P₁ that derives from the unequal division of the C. elegans zygote is important for subsequent development, we took advantage of the temperature-sensitive lin-5(ev571) mutant allele (Lorson et al., 2000). We discovered that shifting lin-5(ev571) zygotes from 17°C to 27°C for ~5 min during mitosis results in the spindle remaining in the cell center, yielding AB and P₁ cells of more similar sizes than normal (Figure 1A–C, Figure 1—figure supplement 1, Videos 1–3). Such embryos are referred to as ‘upshifted lin-5(ev571)’ hereafter. There was a slight variability in the size of AB relative to that of the entire zygote in these embryos (51.2%±2.8 standard deviation [SD], N = 257). Moreover, we found that inactivation of function was readily reversible since returning upshifted lin-5(ev571) embryos to 17°C after the ~5 min upshift rapidly restored LIN-5 function, allowing subsequent cell divisions to proceed (see below). Furthermore, to control for potential effects of the lin-5(ev571) mutation unrelated to AB and P₁ equalization, early 2-cell stage lin-5(ev571) embryos were shifted from 17°C to 27°C for ~5 min; such embryos are referred to as ‘controls’ throughout the article.

Figure 1 with 3 supplements see all

Download asset Open asset

Video 1

Download asset

Video 2

Download asset

Video 3

Download asset

Severe or complete depletion of lin-5 function does not impair A-P polarity (Srinivasan et al., 2003), and we found the same to be true upon acute LIN-5 inactivation in upshifted lin-5(ev571) embryos, as evidenced by the normal distribution of the posterior polarity protein PAR-2 (Figure 1—figure supplement 2A, B). Moreover, the polarity mediator MEX-5, which is enriched in the anterior cytoplasm of the wild-type zygote in response to A-P polarity cues (Cuenca, 2003; Schubert et al., 2000), was localized similarly in control and upshifted lin-5(ev571) embryos, resulting in MEX-5 enrichment in AB following first division (Figure 1—figure supplement 2C). In addition, as anticipated from the more centrally located first cleavage furrow, slightly more MEX-5 was present in P₁ in upshifted lin-5(ev571) embryos than in the control (Figure 1—figure supplement 2D). Taken together, these observations indicate that upshifted lin-5(ev571) zygotes exhibit normal A-P polarity, providing us with an experimental means to address specifically the impact of cell size asymmetry.

In search for a complementary method to equalize first division, we reasoned that optogenetic recruitment of LIN-5 to the anterior cortex could counteract the larger posterior forces normally acting on astral microtubules, thus maintaining the spindle in the center (Fielmich et al., 2018). We used an extant worm strain expressing both LOV::PH::GFP bound to the plasma membrane and endogenously tagged LIN-5::ePDZ::mCherry, which can thus be recruited to the desired portion of the plasma membrane by transient local induction of LOV<-> ePDZ interaction using 488 nm laser light (Fielmich et al., 2018; Figure 1D, E). As anticipated, we found that acute recruitment of LIN-5::ePDZ::mCherry to a small region of the anterior cortex can maintain the spindle in the center of the zygote, producing similarly sized AB and P₁ cells (Figure 1F, Video 4).

Video 4

Download asset

Overall, we conclude that two means of balancing anterior and posterior pulling forces during mitosis can serve to equalize the first division.

To assess the potential importance of the size difference between AB and P₁ for subsequent development, we utilized time-lapse microscopy to analyze upshifted lin-5(ev571) embryos and optogenetically manipulated embryos following return to their respective permissive conditions, scoring embryonic lethality at the motile 3-fold stage or at hatching (Materials and methods). We found that those upshifted lin-5(ev571) embryos that died invariably failed to elongate into a worm shape towards the end of embryogenesis, reflecting a failure in ventral closure of the epidermis that was accompanied by extrusion of endodermal and mesodermal tissues as muscle contraction began (Figure 1C, Figure 1—figure supplement 3, Video 5, Video 6; n = 92). The terminal embryonic phenotype of optogenetically manipulated embryos that died was likewise characterized by a failure in ventral epidermal closure, with the exception of one embryo that died as an unhatched larva (n = 26). Moreover, we found several post-embryonic phenotypes among upshifted lin-5(ev571) animals that hatched, including death during larval stages and reduced adult fertility (Figure 1—figure supplement 3C), demonstrating that equalizing the first division can result in lethality beyond embryogenesis.

Video 5

Download asset

Video 6

Download asset

We took advantage of the slight variability in relative AB size following equalization to ask whether this trait correlated with the incidence of embryonic lethality. As shown in Figure 2A, B, we observed that whereas control embryos exhibited nearly identical lethality compared to lin-5(ev571) embryos maintained at 17°C (~7%), lethality of upshifted lin-5(ev571) embryos gradually increased as AB and P₁ became closer in size, from ~24% when relative AB size was (56–54]% to ~75% when it was (50–48]% (refer to Supplementary file 6 for statistical analyses throughout the article). We found a similar gradual increase of embryonic lethality when AB and P₁ became closer in size in optogenetically manipulated embryos (Figure 2C, D), which however exhibited ~23% lethality even when not exposed to 488 nm laser light. In addition, we found in both experimental settings that embryos with a relative AB size smaller than 48% invariably died (Figure 2B, D; ‘inverted embryos’ hereafter).

Figure 2 with 1 supplement see all

Download asset Open asset

Figure 2—source code 1

Cell size manipulation and lethality analysis in lin-5(ev571) and optogenetic embryos.

https://cdn.elifesciences.org/articles/61714/elife-61714-fig2-code1-v3.zip

Download elife-61714-fig2-code1-v3.zip

Figure 2—source data 1

Cell size measurements and division timing in lin-5(ev571) and optogenetic embryos.

https://cdn.elifesciences.org/articles/61714/elife-61714-fig2-data1-v3.xlsx

Download elife-61714-fig2-data1-v3.xlsx

Taken together, these findings establish that first cleavage equalization can be tolerated to some extent since a subset of embryos with similar AB and P₁ sizes survive. However, other such embryos die, suggesting that the unequal first cleavage is required for invariably successful embryogenesis. Furthermore, our findings reveal a size asymmetry threshold at the 2-cell stage below which embryogenesis always fails.

We investigated further the consequences of rendering AB and P₁ similar in size. We utilized the lin-5(ev571) upshift setting for all experiments described hereafter because the optogenetic strain exhibited higher background lethality and was not amenable to investigation with fluorescently labeled markers since both GFP and mCherry channels are already occupied.

We first report our analysis of cell division timing. In the wild type, AB divides before P₁, in part owing to a size difference between the two cells (Brauchle et al., 2003). We addressed whether such a reduction in asynchrony was recapitulated in upshifted lin-5(ev571) embryos, using H2B::mCherry to precisely time anaphase onset. As shown in Figure 2E, we indeed found a strong correlation between relative AB size and extent of division asynchrony, confirming that the size difference between AB and P₁ contributes to their division asynchrony. Moreover, inverted embryos retained residual asynchrony (Figure 2E), verifying that cell size is only one of the factors dictating such asynchrony (Budirahardja and Gönczy, 2008; Rivers et al., 2008).

Could the reduction in division asynchrony between AB and P₁ cause the lethality of upshifted lin-5(ev571) embryos? We reasoned that if this were the case, then survival should be higher in upshifted lin-5(ev571) embryos in which asynchrony between the two cells was experimentally restored to the wild-type timing. Therefore, we set out to specifically slow down the cell cycle of P₁ in upshifted lin-5(ev571) embryos by directing a 405 nm laser onto its nucleus, adapting a method used previously for endodermal cells in the worm (Lee et al., 2006). We determined a non-lethal irradiation regime that retarded P₁ division timing in the wild type by 132 ± 94 s on average (n = 7), but did not cause lethality (Materials and methods). We found that a similar treatment in upshifted lin-5(ev571) embryos restored the extent of asynchrony between AB and P₁ to that of control embryos (Figure 2F). Despite this, however, viability was not rescued (Figure 2F), leading us to conclude that reduction in division asynchrony between AB and P₁ per se is not responsible for the lethality of upshifted lin-5(ev571) embryos.

Interestingly, in addition, we found a severe phenotype arising at the 4-cell stage in a minor fraction of upshifted lin-5(ev571) embryos, which we reasoned might be related to reduced division asynchrony between AB and P₁. In the wild type, division asynchrony results in the AB spindle elongating before the P₁ spindle, and thereby to an oblique positioning of the AB daughter cells due to the geometrical constraints imposed by the eggshell (Figure 2—figure supplement 1A). This leads to a characteristic rhomboid arrangement at the 4-cell stage, with the P₁ daughter P₂ contacting the AB daughter ABp (Figure 2—figure supplement 1A). P₂ expresses the Delta-like ligand APX-1, which interacts with the Notch receptor GLP-1 present on ABp, thereby instructing ABp descendants to adopt neural and epidermal fates (Mango et al., 1994b; Mello et al., 1994; Shelton and Bowerman, 1996; Figure 2—figure supplement 1B). We found that ~4% of upshifted lin-5(ev571) embryos adopted an abnormal T-like arrangement at the 4-cell stage (T-arrangement thereafter) (Figure 2—figure supplement 1C; Video 7; n = 257). In such embryos, P₂ cannot contact ABp and, by inference, Delta/Notch signaling cannot take place between the two cells (Figure 2—figure supplement 1D). As anticipated from such lack of signaling, all embryos with a T-arrangement at the 4-cell stage later died (n = 11).

Video 7

Download asset

Overall, we conclude that reduced division asynchrony between AB and P₁ per se is not responsible for the lethality of most upshifted lin-5(ev571) embryos. Moreover, we find that a minor fraction of upshifted lin-5(ev571) embryos lack contact between P₂ and ABp and, thereby, a critical Delta/Notch induction event, illustrating how the size difference between AB and P₁ can contribute to successful fate acquisition.

To reach a more comprehensive understanding of potential deviations from normal development following first division equalization, we conducted systematic 4D lineage tracing using nuclear H2B::mCherry as a proxy for cell position (Bao et al., 2006; Jelier et al., 2016; Krüger et al., 2015). We collected data up to the ~120 cell stage, assessing 13 features for each cell, including division timing, cell position, and division orientation, resulting in up to ~2000 data points per embryo, although not all stages or features could be scored in every instance (see Supplementary file 1 for information regarding each lineaged embryo and Supplementary file 2 for individual cell data). The entire lineaging data set is available at https://github.com/UPGON/worm-rules-eLife; (Jankele, 2021; copy archived at swh:1:rev:069c5e3147b7721885b5824282f342cac8a4de5b). The lineaging data set comprised wild-type embryos (wild-type, n = 10), lin-5(ev571) embryos shifted early in the 2- cell stage (control, n = 18), as well as upshifted lin-5(ev571) embryos that were equalized (empirically defined as having a relative AB size of 48–53%), and that either lived (equalized alive, n = 21) or died (equalized dead, n = 28). There was no significant difference in relative AB size between these two equalized groups (50.6% in both; p=0.99), so that their distinct fate later in embryogenesis cannot be attributed to a size difference at the 2-cell stage. In addition, we lineaged lin-5(ev571) embryos with a <48% relative AB size, all of which died (inverted, n = 7). Given their severe and early departure from normality, we did not lineage the minor fraction of lin-5(ev571) embryos with a 4-cell stage T-arrangement.

We set out to mine this rich data set to uncover deviations from normal development following AB and P₁ equalization, comparing wild-type and control embryos with equalized lin-5(ev571) and inverted lin-5(ev571) embryos. In addition, we investigated whether differences could be unveiled between equalized alive and equalized dead lin-5(ev571) embryos to identify features that may cause lethality in the latter group.

We first report our analysis of temporal aspects in the lineaging data. Plotting separately the proliferation of cells descending from AB and P₁ (hereafter referred to as AB and P₁ lineage, respectively), we observed no statistically significant difference in division timing in the AB lineage between wild-type or control and equalized or inverted lin-5(ev571) embryos (Figure 3A – overlapping confidence intervals, Supplementary file 2). In stark contrast, we found a clear difference in the P₁ lineage, where cell number increased more rapidly in equalized and in inverted embryos compared to wild-type or control embryos (Figure 3B). Furthermore, cell cycle duration in equalized lin-5(ev571) embryos was rarely correlated with relative AB size for AB descendants, but highly correlated for most P₁ descendants (Figure 3—figure supplement 1, Supplementary file 3).

Figure 3 with 1 supplement see all

Download asset Open asset

Investigating cell cycle duration in individual blastomeres revealed that the germline precursor P₄ was most affected, with the cell cycle being ~33% shorter in equalized embryos compared to the control (Figure 3C, p<0.001 with Welch’s two-sample t-test, which is used also hereafter unless stated otherwise). Interestingly, we also uncovered a deviation from the normal division pattern in the P₄ lineage suggestive of fate transformation. Whereas P₄ normally undergoes a single division (Figure 3D, top, Figure 3E), we found that it underwent at least one additional division in ~35% of equalized lin-5(ev571) embryos (n = 36), and in all four inverted lin-5(ev571) embryos scored for this trait (Figure 3D, bottom, Figure 3E). Overall, we conclude that following first division equalization the cell cycle is shortened most in the germline precursor P₄, the fate of which is also imparted less faithfully.

We found in addition that faster cycling in the P₁ lineage of equalized embryos eventually changed the relative order of cell division during development (Figure 3F). Notably, the division of the P₁ descendants Ea/Ep, D, Da, and P₄ occurred before that of the corresponding AB lineage cells (Figure 3F, red arrows). Could such an altered sequence of events in the AB versus the P₁ lineage contribute to the lethality of equalized lin-5(ev571) embryos? To address this question, we investigated whether there was a statistically significant difference in cell cycle duration or division timing comparing individual cells in equalized alive and equalized dead embryos, but found none (Supplementary file 2, Welch’s t-tests with Benjamini–Hochberg correction for multiple tests). Taken together, our data demonstrate that equalizing the first division results in a global increase in cell cycle pace in the P₁ lineage, leading to an altered sequence of divisions, but that this alone does not explain why some equalized embryos die whereas others do not.

We were also interested in exploring the variability of cell cycle duration in the lineaging data set. We determined the coefficient of variation (CV) for each cell and found that the overall average CV of cell cycle duration was 4.4% ± 1.8 in the wild-type and 5.3% ± 1.8 in lin-5(ev571) control embryos (Figure 3G, p<0.001), indicative of increased variability already in the latter group. In addition, we found that whereas the CV in cell cycle duration in equalized alive lin-5(ev571) embryos was comparable to that of control embryos (5.2% ± 2.4, p=0.75), it was significantly larger than that in equalized dead and inverted lin-5(ev571) embryos, a difference due strictly to the P₁ lineage (Figure 3G, 6.6% ± 2.7% and 6.5% ± 4.2, respectively, p<0.0001). We conclude that equalized dead and inverted lin-5(ev571) embryos exhibit less stereotyped cell cycle durations.

We next report our analysis of spatial aspects in the lineaging data. In particular, we monitored cell position and division orientation along the three embryonic axes, comparing their averages to a reference model of spatially and temporally aligned control lin-5(ev571) embryos (Jelier et al., 2016) (Materials and methods).

We found that equalized alive lin-5(ev571) embryos exhibited a mild increase in the average positional deviation per cell from the control starting at the 4-cell stage; this deviation remained essentially constant thereafter (Figure 4A, compare blue and yellow). By contrast, cell positions in equalized dead lin-5(ev571) embryos diverged increasingly from the control over time (Figure 4A, compare blue and red). For inverted embryos, cell positions deviated already early on and diverged further thereafter (Figure 4A, gray). Furthermore, we found that division orientation of individual cells exhibited a mild increase in angular deviation in equalized alive lin-5(ev571) embryos compared to the control, and that this deviation was more pronounced in equalized dead and inverted lin-5(ev571) embryos (Figure 4B). Together, these findings establish that embryos that die following first division equalization diverge more from the norm in terms of cell positioning and division orientation than those that live.

Figure 4 with 1 supplement see all

Download asset Open asset

Investigating spatial features in the lineaging data for individual blastomeres revealed a striking skew in the division axis of EMS at the 6-cell stage in a subset of embryos. In the wild type, the EMS spindle is oriented along the A-P axis owing to Wnt/Src signaling emanating from P₂ (Bei et al., 2002; Liro and Rose, 2016). We found that EMS division orientation was skewed by >35° in ~19% equalized and ~43% inverted lin-5(ev571) embryos (Figure 4F; n = 48 and 7). In all eight embryos with such a skew that later died, this alteration led to striking cell mispositioning at the 8-cell stage, with E and MS assuming abnormal oblique positions (Figure 4G, top), a defect that propagated subsequently (Figure 4G, bottom). By contrast, in the four equalized embryos that lived despite an initially aberrant EMS division angle (Figure 4F), the skew was corrected in late anaphase, resulting in essentially normally positioned E and MS. These observations indicate that, unless corrected, a skew in the EMS division axis results in embryonic lethality.

To uncover further differences between equalized alive and equalized dead lin-5(ev571) embryos, excluding embryos with a T-arrangement at the 4-cell stage and those with an EMS skew at the 6-cell stage, we compared all spatial features between the two sets of embryos from the 4- to the 15-cell stage (Figure 4—figure supplement 1A, Supplementary file 4). This analysis uncovered 10 features that differed significantly between the two groups and that were highly correlated with each other (Figure 4—figure supplement 1A). Among these features, we found notably that the division orientation of MS deviated approximately three times more with respect to the A-P axis in equalized dead lin-5(ev571) embryos than in equalized alive ones (Figure 4H). This led to increased mispositioning of the MS daughter cells MSa and MSp (Figure 4I), which were frequently inverted along the L-R axis, a phenotype significantly associated with death (Figure 4J, K). However, such L-R inversion was also observed in some equalized alive lin-5(ev571) embryos (Figure 4I), indicating that it can be compatible with successful embryogenesis, presumably owing to subsequent compensatory movements. Taken together, these observations lead us to conclude that severe mispositioning of E and MS, as well as of MSa and MSp, can occur in equalized dead lin-5(ev571) embryos and contribute to lethality.

Like for the temporal data, we explored the overall variability of cell positioning and division orientation in the lineaging data set, comparing the SDs of all individual cells. As shown in Figure 4C, we found a slight increase in overall variability of cell positioning among equalized alive lin-5(ev571) embryos compared to the control, as well as a further significant increase among equalized dead and inverted lin-5(ev571) embryos (Figure 4D). Together, these results indicate that cell positioning and division orientation defects in equalized embryos are also variable rather than stereotyped, especially in embryos that die.

Serendipitously, we found that 6 of the 10 features identified above correlate with the extent of compression the embryo experienced from the coverslip during time-lapse imaging, suggesting that this external factor affects cell positioning, division angles, and cell migration differently in equalized alive and dead lin-5(ev571) embryos. Furthermore, equalized dead lin-5(ev571) embryos tended to be significantly more compressed than equalized alive ones (Figure 4E, 19.8 ± 2.1 µm versus 21.9 ± 2.2 µm, p<0.002), whereas a comparable degree of compression had no detrimental effect on control embryos (Figure 4E). These observations led us to investigate directly whether equalized embryos die less frequently in the absence of compression. Therefore, we imaged uncompressed embryos using 45 µm diameter beads or removing the coverslip after first division equalization, finding decreased lethality of equalized embryos compared to imaging under mild compressed condition with 20–25 µm beads or in the continued presence of the coverslip (33% versus 55%, n = 45 and 174, p=0.03, Fisher’s exact test). Together, these findings indicate that equalized embryos tolerate less well mechanical compression, which may perturb cell positioning, division orientation, and cell–cell contacts.

We investigated whether defects in cell fate might occur following first division equalization, focusing our analysis on the endodermal and pharyngeal lineages.

In the wild-type endodermal lineage, the P₁-derived Ea/Ep cells exhibit a nearly twofold lengthening of the cell cycle compared to the MS lineage sister cells as they ingress into the blastocoel (Figure 5A, B, Sulston et al., 1983). Interestingly, we found that Ea/Ep divided earlier and ingressed less in equalized dead lin-5(ev571) embryos than in either control or equalized alive lin-5(ev571) embryos, proportionally to relative AB size (Figure 5C, D, Figure 5—figure supplement 1A, B, Video 8). We noted also that Ea/Ep tended to divide while still in contact with the eggshell in most equalized embryos (Figure 5E, green nuclei and dashed cell contour). Overall, we conclude that endodermal cells divide precociously and ingress only partially in equalized lin-5(ev571) embryos, particularly in the subset that later dies.

Figure 5 with 1 supplement see all

Download asset Open asset

Figure 5—source code 1

Analysis of Ea/Ep migration, cell cycle duration, and END-3::GFP expresion in lin-5(ev571) and control embryos.

https://cdn.elifesciences.org/articles/61714/elife-61714-fig5-code1-v3.zip

Download elife-61714-fig5-code1-v3.zip

Figure 5—source data 1

Ea/Ep migration, cell cycle duration, and END-3::GFP expresion in individual lin-5(ev571) and control embryos.

https://cdn.elifesciences.org/articles/61714/elife-61714-fig5-data1-v3.xlsx

Download elife-61714-fig5-data1-v3.xlsx

Video 8

Download asset

We set out to assess whether these alterations in Ea/Ep behavior are accompanied by improper endodermal fate specification. Normally, such specification is imparted by a redundant cascade of GATA-family transcription factors, including END-3/END-1, which peak in Ea/Ep and activate ELT-7/ELT-2, with ELT-2 remaining present throughout embryogenesis (reviewed in Maduro, 2015; McGhee, 2013). Suggestively, we found that END-3::GFP in equalized lin-5(ev571) embryos was expressed at lower levels than in controls (~62 ± 19%, p<0.001), correlating with relative AB size and Ea/Ep cell cycle duration (Figure 5E, F; n = 18 and 19, respectively, Figure 5—figure supplement 1C, Video 8). Furthermore, ELT-2::GFP, which is normally expressed in eight intestinal progenitors at the ~100-cell stage, was often expressed in only 2–6 cells in equalized lin-5(ev571) embryos (Figure 5G, H). Together, these findings indicate that endodermal fate is acquired less faithfully following first division equalization, although altered ELT-2::GFP distribution is not significantly associated with embryonic lethality (p=0.69, Fisher’s exact test), in line with the notion that even embryos with a partial gut can survive (Choi et al., 2017).

We also investigated fate specification of the pharyngeal lineage. In the wild type, the anterior pharynx is induced in ABalp and ABara through Delta/Notch signaling stemming from MS/MSa/MSp (reviewed in Mango, 2009). Although all four ABa descendants express the GLP-1 Notch receptor, normally ABala and ABarp are not induced since they do not contact the MS/MSa/MSp cells that express the Delta ligand (Figure 5—figure supplement 1D). We analyzed embryos expressing the pharyngeal marker PHA-4::GFP, which is detectable at the ~100-cell stage in the lineages of MS, ABalp, and ABara, labeling 18 cells in total (Horner et al., 1998; Mango et al., 1994a; Murray et al., 2008). We found that equalized alive lin-5(ev571) embryos systematically harbored PHA-4::GFP in 18 nuclei (Figure 5I, J; n = 5). By contrast, abnormal PHA-4::GFP expression was observed in ~47% of equalized dead and inverted lin-5(ev571) embryos, usually with >18 positive nuclei (Figure 5J, Figure 5—figure supplement 1E; n = 15). Ectopic PHA-4::GFP was present in some cells derived from ABala in five such embryos and in ABarp in another one (Figure 5—figure supplement 1E), suggesting that MS/MSa/MSp formed illegitimate contacts in these cases, leading to aberrant pharynx induction in their descendants.

Overall, we conclude that equalized lin-5(ev571) embryos exhibit defective fate acquisition at the least in the endodermal and pharyngeal lineages.

We explored whether the above described defects occurred simultaneously in equalized embryos. However, we found no significant association between any pair of qualitative features described thus far in equalized embryos in the entire lineaging data set (Fisher’s exact test), including between additional P₄ division, EMS skew, MSa/MSp inversion, aberrant ELT-2, or PHA-4 expression, suggesting that these defects frequently arise independently of each other.

Despite such widespread phenotypic variability among equalized lin-5(ev571) embryos, we wondered whether the ultimate fate of equalized embryos could be predicted early in development. Again excluding embryos with a T-arrangement at the 4-cell stage or with an EMS skew at the 6-cell stage, we utilized a machine learning approach to identify a small number of features that hold predictive value for distinguishing the two sets of embryos. To this end, we utilized Lasso (Least absolute shrinkage and selection operator), which is a penalized regression method that iteratively eliminates features that contribute the least to the predictive performance of a statistical model, eventually converging on a simpler solution (Tibshirani, 1996). We conducted this analysis at the 4-, 8-, 15-, and 28- cell stage, using as input all available features in each case for equalized lin-5(ev571) embryos (n = 31 embryos scored for all features at all four stages, Supplementary file 7). To obtain a robust classifier despite the relatively small sample size, we used repeated fivefold cross-validation, which randomly splits the data into five subsets of similar sizes, four of which are used for training the model, which is then evaluated against the remaining subset for goodness of fit. Performing 250 repetitions of cross-validation in each case, we found that features extracted at the 15-cell and 28-cell stage hold similarly high predictive value, with an overall model accuracy of ~88 ± 6% SD and ~85 ± 10% SD, respectively (Figure 6A). Remarkably, we found in particular that ~89% of the models selected at the 15-cell stage contained just three predictive features (Figure 6B, C): ABara dorsoventral (D-V) position, ABara positional deviation (pOV), and Ca net movement (Ca.netdis). The best performing model at the 15-cell stage achieved 100% sensitivity – meaning that all dead embryos were correctly classified – and 86% specificity – meaning that a small fraction of alive embryos were misclassified (Figure 6D).

Figure 6 with 3 supplements see all

Download asset Open asset

A striking example of physically unequal asymmetric division is that of the female meiotic divisions, where minute polar bodies are generated in addition to the much larger oocyte. There is little doubt that the unequal size of daughter cells is of paramount importance in this case to ensure proper allocation of maternal components to the oocyte and future embryo. Physically unequal divisions, albeit less striking in terms of size inequality, are also frequent during early embryogenesis of metazoan organisms. One possibility is that this ensures swift and proper allocation of components to daughter cells destined to yield different numbers and types of descendants in early embryos that rely on maternally contributed components and undergo rapid cell cycles. This may be particularly important in holoblastic embryos, which cleave within the constraints of an eggshell or an analogous structure, precluding compensation for potential variability in blastomere volumes through differential cell growth.

A paradigmatic example of a physically unequal division in early embryos is that of the C. elegans zygote, which normally yields a larger AB daughter and a smaller P₁ daughter. Previously, the specific role of size inequality had not been addressed independently from fate asymmetry. Here, we addressed this question by altering the function of the spindle positioning protein LIN-5 to equalize first cleavage without affecting A-P polarity. We discovered that first division equalization is tolerated to some extent, as evidenced by approximately half of equalized embryos completing development. Moreover, we discovered that embryos with a relative AB size smaller than 48% always die, revealing a size asymmetry threshold for viability.

Prior work established that the size difference between AB and P₁ contributes to their division asynchrony in the wild type because the DNA replication checkpoint is engaged preferentially in P₁ compared to AB (Brauchle et al., 2003; Stevens et al., 2016). In addition to this mechanism, the positive regulators of mitotic entry polo-like-kinase I (PLK-1), cyclin B3, and CDC-25 phosphatase are enriched in AB compared to P₁, thereby also contributing to division asynchrony, in a size-independent manner (Budirahardja and Gönczy, 2008; Michael, 2016; Rivers et al., 2008). As anticipated from these studies, we found here that division asynchrony decreases following first division equalization, in a manner that depends on AB size. However, this asynchrony alone does not cause subsequent death since equalized embryos with experimentally restored asynchrony die at a similar rate. This is in line with the viability of embryos depleted of DNA replication checkpoint components, which also exhibit decreased asynchrony between AB and P₁ (Brauchle et al., 2003). Likewise, the majority of embryos in which asynchrony between the two cells is abrogated entirely through local infrared treatment of P₁ also live (Choi et al., 2020). Furthermore, we uncovered here that cell cycle progression is not set merely by size in all C. elegans embryonic blastomeres since diminishing the size of AB and its descendants does not slow their cell cycle, perhaps because the abovementioned positive mitotic regulators are present in excess in AB.

We found also that not only P₁, but also many of its descendants, progress faster through the cell cycle upon first division equalization. It will be interesting to address whether these later differences reflect a sustained excess of positive regulators such as PLK-1, cyclin B3, and CDC-25 inherited following first division equalization. Regardless of the underlying cause, faster cell cycle progression in the P₁ lineage might have repercussions also for fate acquisition, which can be reliant on precisely timed expression of transcription factors during C. elegans embryogenesis (Murray et al., 2012; Nair et al., 2013; Sarov et al., 2012).

Although not detrimental to viability per se, faster cell cycle progression in the P₁ lineage, but not in the AB lineage, results over time in an altered sequence of cell division in equalized embryos, which can lead to missing or aberrant interactions between cells. A particularly striking example occurs in a minor fraction of equalized embryos that exhibit an aberrant T-arrangement at the 4-cell stage, leading to a lack of inductive Delta/Notch signaling. Suggestively, we found that equalized embryos with such a T-arrangement tend to be elongated (see legend of Figure 2—figure supplement 1). Compatible with embryo length playing a role, T-arrangements also occur in lon-1 mutant embryos, which are very elongated but where AB and P₁ have normal relative sizes (Yamamoto and Kimura, 2017). Interestingly, a similar T-arrangement occurs naturally in other Nematode clades, reflecting plasticity in early division patterns that likely requires adaptations of fate determination networks (Brauchle et al., 2009; Schulze and Schierenberg, 2011; Schulze and Schierenberg, 2009).

Missing or aberrant interactions between cells can also derive from defective division orientation, leading to daughter cell mispositioning, as exemplified with the skewed EMS and MS division orientation. Interestingly, an EMS skew is observed also upon impaired Wnt signaling from P₂ (Bei et al., 2002; Schlesinger et al., 1999; Thorpe et al., 1997). More generally, a posterior signaling center that relies on a Wnt-dependent relay mechanism originating from P₂/P₃ polarizes multiple cells during C. elegans embryogenesis (Bischoff and Schnabel, 2006), and our findings are compatible with the possibility that this signaling center is compromised upon first cleavage equalization. Furthermore, the EMS skew might affect cell–cell contacts and therefore alter the instructive Delta/Notch signaling that normally occurs between MS and two adjacent ABa-derived cells, which are thereby triggered to differentiate towards pharyngeal fates (reviewed in Mango, 2009). We indeed observed inappropriate induction of pharyngeal fate in several ABa cells, raising the possibility that aberrant contacts occurred between MS/MSa/MSp, which express a Delta ligand, and ABala or ABarp, which express a Notch receptor, but that are normally shielded from ligand binding owing to their distant location. Moreover, we noted that the intestinal progenitors Ea/Ep expressed less END-3::GFP in equalized embryos, perhaps reflecting insufficient induction by maternal factors, followed by stochastic expression of the downstream component ELT-2. Compatible with this view, maternally provided SKN-1, as well as its targets END-1 and END-3, must reach a critical threshold for complete activation of ELT-2 (Maduro, 2015; Raj et al., 2010).

Regarding MS position, a physical computational model predicted that abolishing both size asymmetry and asynchrony between AB and P₁ lineages would impact the positions of MS and E daughters at the 24- cell stage in ~25% of embryos (Fickentscher et al., 2016). Accordingly, we observed instances of MSa/MSp mispositioning and incomplete ingression of E daughters in equalized lin-5(ev571) embryos. Thus, these defects in equalized embryos could be at least in part explained by abnormal forces experienced by cells with altered volumes and division timing.

Detailed lineage analysis enabled us to uncover that in many equalized embryos the germline progenitor P₄ undergoes continued divisions in a manner that resembles the behavior of its cousin D. Such a defect in germline fate specification could help explain the reduced fertility observed in some adults that derive from embryos that survived equalization. Lineage transformation of P₄ into D also occurs in embryos derived from mes-1 mutant animals, in which P₂ and P₃ lack proper polarity (Berkowitz and Strome, 2000). By extension, such polarity defects may also be present following first division equalization. In the wild type, P₄ divides symmetrically and its daughters remain quiescent until hatching (Sulston et al., 1983). A size-dependent switch from asymmetric to symmetric division occurs in P₄ as it reaches a size too small to allow formation of the reciprocal PAR protein gradient (Hubatsch et al., 2019). Our findings lead us to speculate that P₄ cell size reduction might be required not only for this switch of division mode but also to maintain germline progenitors quiescent until hatching.

Wild-type C. elegans embryogenesis is highly stereotyped, exhibiting little variability between embryos in terms of division timing, cell positioning, division orientation, and fate acquisition, ultimately yielding 558 cells at hatching in the hermaphrodite (Richards et al., 2013; Sulston et al., 1983). The array of incompletely penetrant and variable defects revealed here implies that a physically unequal first division is critical for the stereotypy of C. elegans embryogenesis.

Despite important phenotypic variability upon first division equalization, we found that three features at the 15-cell stage together hold strong predictive power in predicting the ultimate fate of equalized embryos: ABara D-V position, ABara positional deviation, and Ca net displacement. Closer analysis of cell positions in the lineaging data set lends further support to the importance of these three features. Indeed, several AB descendants on the right side of embryo, including ABar and its progeny, were significantly shifted dorsally as early as the 8-cell stage solely in dying embryos (Figure 6—figure supplement 2, Supplementary file 2). Although the mechanisms responsible for such dorsally shifted position or for increased Ca movement in equalized dead lin-5(ev571) embryos remain to be fully understood, we conclude that these three features provide a shared early signature among most equalized embryos that then diverges into more variable phenotypic manifestations.

We also wondered whether having differently sized AB and P₁ blastomeres contributes to robust embryogenesis. Developmental systems are deemed to be robust if they remain unchanged or else change in a reproducible manner in the face of a given perturbation. The loss of robustness is manifested by increased variability upon perturbation without a change in the mean (reviewed in Félix and Barkoulas, 2015). Applying these criteria to features monitored in our data set, we found only 23 of the 1608 analyzed in which the mean remained the same but the variability increased when comparing equalized lin-5(ev571) and control embryos (Figure 6—figure supplement 3). This indicates that these features in particular are less robust to perturbation by first division equalization. Moreover, our findings suggest that having differently sized AB and P₁ blastomeres provides resilience against mechanical stress, as evidenced by the increased lethality of equalized lin-5(ev571) embryos upon compression. Normally, compressed C. elegans embryos can correct specific defects in cell positioning through the concerted migration of several AB descendants (Jelier et al., 2016). Presumably compressed equalized embryos cannot undertake such corrective movements for steric reasons or because their fate is compromised.

How general is the importance of having differently sized blastomeres in early embryogenesis? Physically asymmetric cleavage of the zygote is present across the Rhabditida order (Brauchle et al., 2009; Valfort et al., 2018), suggesting functional importance over substantial evolutionary times. Moreover, unequal cleavages are present in early development of other systems, including annelid, ascidian, and echinoderm embryos (reviewed in Moorhouse and Burgess, 2014; Negishi and Nishida, 2017; Shimizu et al., 1998). Conversely, generating cells of equal sizes seems of importance during early human development since unevenly sized blastomeres frequently correlate with aneuploidy and lower rates of implantation and pregnancy (Hardarson, 2001). Overall, generating appropriately proportioned daughter cells is widespread in metazoan organisms, and our work demonstrates the importance of physical asymmetry of the first division in C. elegans, which is fundamental for ensuring invariably successful development.

C. elegans strains used in this study are listed in Supplementary file 5 and were maintained on standard NGM plates with OP50 Escherichia coli as a food source. Temperature-sensitive strains were maintained at 16°C, other strains at 20°C. Strains expressing the desired fluorescent markers were crossed with the lin-5(ev571) carrying strain, and homozygous progeny selected based on temperature sensitivity and fluorescent marker expression. After establishing a given strain, worms were maintained for at least one generation prior to analysis. For embryo imaging, 1–2-day-old adult lin-5(ev571) hermaphrodites were dissected in M9 buffer chilled at 15°C, and one-cell stage embryos with visible pronuclei collected using a mouth pipet. Embryos were then mounted either on a 2% agarose pad for DIC imaging with oil immersion objectives or into a bead slurry containing ~20 µm polystyrene beads (Polysciences, #18329-5) in M9 + 0.5% (w/v) methylcellulose for water immersion objectives or 20% iodixanol (Optiprep, Sigma Aldrich) for glycerol objectives to raise the refractive index of the medium (Boothe et al., 2017). Two methods were utilized to image embryos without compression. First, we mounted embryos in a viscous M9 medium containing + 0.5% (w/v) methylcellulose and 45 µm polystyrene beads. Second, embryos were mounted on a 2% agarose pad as usual, but the coverslip removed after the temperature shift, and the embryo then deposited with the pad onto an NGM plate. Non-compressed embryos obtained by either method were analyzed jointly for the extent of embryonic lethality.

Rapid temperature shifts between 17°C and 27°C were performed on the microscope stage with the CherryTemp fluidic temperature controller (Cherry Biotech, France). The CherryTemp device is equipped with two thermalization chambers that were set to 17°C and 27°C. By changing the chamber through which the thermalization solution flowed, the sample was rapidly (~15 s) shifted to the desired temperature. To generate equally dividing lin-5(ev571) one-cell stage embryos, we performed the upshift right after nuclear envelope breakdown (NEBD) for embryos imaged by DIC and at metaphase for embryos expressing fluorescently labeled histone, and kept them at the restrictive temperature of 27°C until the completion of cytokinesis, ~5 min thereafter. Thereafter, the sample was shifted back to the permissive temperature of 17°C for the rest of embryogenesis. Embryos imaged only for a fraction of embryogenesis were moved to an incubator set to 17°C and survival scored the next day. In all experiments, we scored as ‘alive’ embryos that reached a normal looking, motile, 3-fold pre-larval stage, or that hatched, and as ‘dead’ embryos with an abnormal morphology at that stage or unable to hatch. Note that the lin-5(ev571) strain used for the majority of experiments expressed in addition the plasma membrane marker GFP::PH and the chromatin marker mCherry::H2B; this strain exhibited a lethality on plates of 4.7% (n = 829), close to that of lin-5(ev571) embryos (2.5%, n = 394).

DIC time-lapse microscopy was performed on a Zeiss Axioscope 2 equipped with DIC optics and a 100 × 1.25 NA Achrostigmat objective, recording with a USB 3.0 1.3 MP monochrome CMOS camera (Ximea, model MQ013MG-E2, Slovakia), controlled by the open-source μManager software (Edelstein et al., 2014).

Combined DIC and epifluorescence time-lapse microscopy was performed on a motorized Zeiss Axio Observer D1 using a 63 × 1.2 NA C-Apochromat water immersion objective, equipped with an Andor Zyla 4.2 sCMOS camera, a piezo-controlled Z-stage (Ludl Electronic Products), and an LED light source (Lumencor SOLA II). The setup was controlled by μManager.

The lineaging data set was acquired on a Leica SP8 laser scanning confocal microscope equipped with a 60× HC PL APO 1.3 NA glycerol immersion objective, HyD detectors set to 100% sensitivity, and a tunable Chameleon laser (Coherent). Time-lapse recordings were acquired at 2.5 min intervals, capturing 35 slices 0.75 µ apart, typically for 100 frames. The microscope was set to 8 kHz resonance scanning mode to reduce phototoxicity (Richards et al., 2013), 4 × line averaging plus 2 × frame accumulation, with a pixel size of 140 nm, a pixel dwell time of 50–70 ns, and the pinhole set to 1.2 Airy units. To compensate for the signal loss deeper in the sample, the excitation light was ramped across the Z range from 2% to 15% for the 488 nm laser line and from 2% to 18% for the 585 nm laser line. Embryos were mounted in M9–20% iodixanol (Optiprep, Sigma Aldrich) to match the sample’s refractive index (Boothe et al., 2017) and placed in a sandwich between two #1.5 coverslips (40 × 22 mm and 18 × 18 mm) separated by ~20 µm polystyrene beads and sealed with melted VALAP (vaseline, lanolin, paraffin, 1:1:1). Sample temperature was maintained at 17°C using the CherryTemp temperature controller. The coverslip sample sandwich was attached to the CherryTemp thermalization chip glass surface through a thin layer of water.

Worms expressing LIN-5::EPDZ::mCherry, PH::EGFP::LOV, and GFP::TBB-2 in the embryo (a kind gift from Sander van den Heuvel) were dissected in a dark room under a red light to prevent premature activation of PH::LOV and mounted for imaging on 2% agarose pads (Fielmich et al., 2018). Imaging was performed on a Leica SP8 in the Live Data Mode of the LAS X software, allowing to toggle laser lines on/off during acquisition. The microscope settings were the same as above with the following modifications: no Z-compensation of exposure, time-lapse acquired at 10 s interval, capturing 11 slices with 1.25 µm spacing, and this for 80–100 frames (from 1- to 4-cell stage); moreover, the temperature was maintained at 22°C. LIN-5::EPDZ::mCherry distribution was monitored with a 585 nm laser; 585 nm transmitted light images were collected on an additional PMT detector to follow NEBD for analysis of AB/P₁ mitotic asynchrony. Interaction of LIN-5::LOV with PH::EPDZ was induced by activating the 488 nm laser at 3–5% intensity only during acquisition in a small rectangular region (~10 × 5 µm) at the anterior cortex, from NEBD until cytokinesis completion.

A pulsed 405 nm diode laser (at 70% output, 700 mW/cm² at 100%) was continuously scanned over the entire P₁ nucleus early in the 2-cell stage for 4.5 min at 8 kHz using a 70 ns dwell time on the SP8 confocal setup described above. This induces photodamage, likely in the form of thymidine dimers in the DNA, which might cause replication fork stalling and activation of the DNA replication checkpoint. Using otherwise wild-type embryos expressing mCherry::H2B, we determined experimentally an optimal non-lethal duration (4.5 min) of continuous laser scanning that induces and enhances the delay between AB and P₁ mitoses from the normal ~3.5 ± 0.34 min to ~6 ± 1.6 min at 17°C.

All images were rotated, Z-projected, and adjusted in Fiji (ImageJ) for display (Schindelin et al., 2012). The 2D surface of the AB and P₁ cells was determined manually from DIC images or from the GFP::PH plasma membrane signal in strains expressing this marker. This was achieved by tracing cell outlines with a Fiji polygon tool at the mid-plane, with both nuclei in focus, when the interface between AB and P₁ was perfectly straight, ~5 min after cytokinesis. The size of AB was then expressed relative to the embryo cross-sectional area, corresponding to the sum of the P₁ and AB surfaces. To assess how well the 2D mid-plane cell area measurements matched the corresponding 3D cell volumes, we analyzed embryos expressing GFP::PH, segmenting the full 3D volumes of AB and P₁ using a watershed-based segmentation (MorphoLibJ plugin in Fiji).

Embryos expressing endogenously tagged PAR-2::GFP and mCherry::MEX-5 were imaged on a wide-field Zeiss Axio-Observer microscope as described above. Images were taken every 30 s until the 4-cell stage, capturing 11 planes with 1.5 µm spacing. For signal quantification, we determined the mean intensity in the mid-plane at the 2-cell stage, including the area covering the nucleus and the entire cortex. MEX-5 intensity was normalized to the mean AB intensity in control embryos from the corresponding day because MEX-5 intensity differed significantly between experiments.

Transgenic embryos expressing END-3::GFP or ELT-2::GFP in a lin-5(ev571) background were imaged using combined DIC and fluorescence time-lapse microscopy as described above, with 2 min intervals and 1 µm optical slicing, capturing a 25 µm stack. Gastrulation movements of Ea/Ep cells were tracked in 3D with the TrackMate Fiji plugin (Tinevez et al., 2017), utilizing the nuclear END-3::GFP signal, which is expressed from the beginning of the Ea/Ep cell cycle (Maduro et al., 2005). We then quantified the peak GFP intensity, that is, the mean voxel intensity in the nuclear volume obtained from the sphere detection macro in TrackMate. The ELT-2::GFP expression pattern was scored visually for the number of GFP-positive cells at the E8 stage. Analysis of pharyngeal differentiation was based on the PHA-4::GFP expression pattern in embryos co-expressing the pan-nuclear mCherry::H2B marker (Sarov et al., 2012). Embryos were imaged with the SP8 confocal microscope as described above, followed by cell lineaging as described below, allowing us to determine the identity of GFP-positive cells at the ~100-cell stage.

Box and whisker plots shown throughout this article were generated in R using the boxplot function. Briefly, each box contains 50% of all data points between the first and third quartile (interquartile range [IQR]) and its center lies at the mean along the Y-axis, with the median indicated by a thick line in the box. Whiskers extend to ±1.58 IQR/sqrt(n). Data points out of this range are shown as individual points and can be considered as outliers, except for plots of variability, where outliers are not displayed for simplicity.

Statistical comparisons were performed in R using Welch’s two-sample t-test with Benjamini–Hochberg correction for multiple comparisons unless indicated otherwise in the figure legends. Simultaneous comparisons of multiple groups for difference in their means were performed using Tukey’s honest significant difference test.

Three-dimensional time-lapse recordings of embryos expressing mCherry::H2B in either wild-type or lin-5(ev571) background were first preprocessed to enhance the nuclear signal and remove noise with the Noise2Void/CARE machine learning pipeline (Krull et al., 2019). Thereafter, a custom Fiji macro was used to correct the drift using the first polar body as a bright fiducial marker. The lineage was then traced using a level-set image segmentation and model evolution implemented in MATLAB (MathWorks, USA) as described previously (Dzyubachyk et al., 2010; Krüger et al., 2015), and corrected in the WormDeLux Java-based lineage editor (Jelier et al., 2016). Results were exported in the StarryNite format (Bao and Murray, 2011). Cells were then automatically named in the lineage editor according to the canonical lineage (Sulston et al., 1983), and manually checked afterward for possible errors, especially in mutant embryos that often substantially deviated in division orientation and cell positioning from the wild-type model used for naming.

As mentioned also in the results section, all lineage files and the source code are available at https://github.com/UPGON/worm-rules-eLife. Note that we excluded embryos with the 4-cell stage T-arrangement from the lineage analysis because they are known to exhibit specific defects stemming from the lack of ABp induction (Priess and Thomson, 1987), which would introduce additional variability and thus mask potential novel phenotypes. Note that some embryos were imaged only up to the 26 cell stage; therefore, the number of embryos used for statistical comparison differs for individual cells and in different analyses reported in this article; Supplementary file 2 reports the number of embryos/cells used for each comparison.

Lineages in the StarryNite format were imported into R version 3.6.1 (R Development Core Team, 2014). The growth curves of all embryos were aligned in time, and a specific scaling factor for each embryo was deployed to match the mean pace of wild-type development at 17°C based on the maximum correlation between the wild type and each experimental curve (Supplementary file 1). Differential pace of development could be caused by slight variations in temperature or by inherent embryo variability. We then set ABa division time as time 0 because time-lapse recording for the lineaging experiments typically started at the 4-cell stage, after the temperature upshift was performed during the first division, or in the early 2-cell stage for controls.

Embryos were aligned in space with respect to their inferred A-P and D-V axes using a custom Java-based script (Jelier et al., 2016). First, nuclear positions were normalized to length, width, and height of each embryo along each imaging axis. Then, the alignment method first determines the A-P axis coinciding with the first principal component of the PCA calculated from all nuclear positions within first 100 frames of development. The D-V axis is orthogonal to the A-P axis and proceeds through the ventral surface of the embryo defined as the average position of the MS lineage cells. Finally, the left–right (L-R) axis is orthogonal to both the A-P and D-V axes.

We then computed 13 features (variables) using this aligned data set for each cell in every embryo. First, we computed the time of cell birth (variable name [v]: StartTime), and cell cycle duration (v: LifeTime), as well as the time of division (v: EndTime). Next, we recorded 3D positions of each cell along the three principal axes at the last frame before anaphase (variables: pAP, pLR, pDV), as well as the total trajectory distance (totdis) and the net displacement (netdis) of each cell as the Euclidean distance between nuclear 3D coordinates at birth and at the end of the cell cycle. Further, division angles (v: aAP, aLR, aDV) were computed with respect to the inferred embryonic axes at the first frame after anaphase onset. The last two variables, that is, angular (v: aMean) and positional deviation (v: pOV), required a comparison of a given embryo to the lin-5(ev571) control model that was determined by aligning embryos in time and space using the General Procrustes Analysis (Gower, 1975) and then calculating average positions of individual cells. Using this model, overall angular deviation (v: aMean) was computed as an angle between the reference division vector and the observed spindle orientation vector in a given embryo for each cell (Jelier et al., 2016). Similarly, positional deviation was calculated as the Euclidean distance between the position of a given nucleus at metaphase from that in the control model. Supplementary file 2 contains complete statistics for all 13 features in experimental groups of embryos.

We filtered variables according to effect size and statistical significance comparing equalized dead and alive lin-5(ev571) embryos at the 15-cell stage. We chose an arbitrary effect threshold of 15% and calculated Welch’s unequal variances t-test for each variable. The p-value cut-off α = 0.008 was determined for a false discovery rate of 10% using scrambled experimental data with shuffled group labels (100 repetitions). We then plotted effect size versus p-value as a volcano plot.

To assess variability within each group of embryos, we calculated the standard deviation and CV for each cell and every feature. In particular, the CV was used to compare cell cycle variability among different cells that have lineage and stage-specific differences in average cell cycle duration. To assess the mean and the variance (Figure 6—figure supplement 3), we required each feature to be scored in at least five embryos in both alive and dead equalized groups, thereby decreasing the total number of analyzed features to 1608.

We searched the parameter space of variables at 4-, 8-, 15-, and 28-cell stage for features that would predict lethality in equalized embryos (n = 31 embryos scored for all features at all these stages). Data preprocessing before feeding data into the model building pipeline included removing those variables with low overall variance (empirically set at <2.5%, n = 24 variables), those that would have more than 25% of missing value (n = 12 variables), and also replacing missing values with the group mean (128/18,228 values), and scaling each variable within a 0–1 range. We used the penalized Lasso regression to generate a binary classification model (glmnet R package). We employed a machine learning strategy to obtain a robust value of the shrinkage coefficient lambda by performing fivefold cross-validation (cv) 250 times and choosing a model with minimal cross-validation mean squared error. We report average classification accuracy for best models selected over 250 cv iterations. To evaluate the stability of solutions thus obtained, we scored the number of chosen predictors and their identity over 250 iterations.

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

We are grateful to Zoltan Spiró for his initial observation of acute temperature sensitivity of lin-5(ev571), as well as to Sander van den Heuvel, Joel Rothman, and Geraldine Seydoux for their generous sharing of strains. Some strains were provided by the Caenorhabditis Genetics Center (CGC), which is funded by the NIH Office of Research Infrastructure Programs (P40 OD010440). We thank Alessandro Berto, Alexandra Bezler, Nils Kalbfus, and Fabian Schneider for constructive comments on the manuscript, as well as Marco Mina for advice on statistics. For help with image acquisition and processing, we thank Olivier Burri and Nicolas Chiaruttini from the Bioimaging and Optics platform of the School of Life Sciences (EPFL, Lausanne).

1. Received: August 3, 2020
2. Accepted: February 22, 2021
3. Accepted Manuscript published: February 23, 2021 (version 1)
4. Accepted Manuscript updated: February 25, 2021 (version 2)
5. Version of Record published: March 18, 2021 (version 3)

© 2021, Jankele 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.