Report

Systems Properties and S

patiotemporal Regulationof Cell Position Variability during Embryogenesis

Graphical Abstract

Highlights

d Quantification of cell position variability during C. elegans

embryogenesis

d Positional variability is highly deterministic and temporally

dynamic

d Positional variability is associated with lineage identity,

location, and symmetry

d Fate specification and cell-cell junctions contribute to

reduction of variability

Li et al., 2019, Cell Reports 26, 313–321January 8, 2019 ª 2018 The Author(s).https://doi.org/10.1016/j.celrep.2018.12.052

Authors

Xiaoyu Li, Zhiguang Zhao, Weina Xu,

Rong Fan, Long Xiao, Xuehua Ma,

Zhuo Du

Correspondencezdu@genetics.ac.cn

In Brief

How do developmental processes

achieve high reproducibility? Using live

imaging and single-cell analyses, Li et al.

show that the variability of cell position

during C. elegans embryogenesis is

highly deterministic and temporally

dynamic, regulated by intrinsic and

extrinsic mechanisms. Fate specification

and cell-cell junctions contribute to the

reduction of variability.

Cell Reports

Report

Systems Properties and SpatiotemporalRegulation of Cell Position Variabilityduring EmbryogenesisXiaoyu Li,1,2,3 Zhiguang Zhao,1,2,3 Weina Xu,1,2 Rong Fan,1,2 Long Xiao,1,2 Xuehua Ma,1 and Zhuo Du1,2,4,*1State Key Laboratory of Molecular Developmental Biology, Institute of Genetics and Developmental Biology, Chinese Academy of Sciences,Beijing 100101, China2University of Chinese Academy of Sciences, Beijing 100049, China3These authors contributed equally4Lead Contact*Correspondence: zdu@genetics.ac.cn

https://doi.org/10.1016/j.celrep.2018.12.052

SUMMARY

An intriguing question in developmental biology ishow do developmental processes achieve highreproducibility among individuals? An in-depth anal-ysis of information contained in phenotypic vari-ability provides an important perspective to addressthis question. In this work, we present a quantitativeand functional analysis of cell position variability dur-ing Caenorhabditis elegans embryogenesis. We findthat cell position variability is highly deterministicand regulated by intrinsic and extrinsic mechanisms.Positional variability is determined by cell lineageidentity and is coupled to diverse developmentalproperties of cells, including embryonic localization,cell contact, and left-right symmetry. Temporaldynamics of cell position variability are highlyconcordant, and fate specification contributes to asystems-wide reduction of variability that could pro-vide a buffering strategy. Positional variability isstringently regulated throughout embryogenesisand cell-cell junctions function to restrict variability.Our results provide insight into systems propertiesand spatiotemporal control of cellular variability dur-ing development.

INTRODUCTION

Development is a highly reproducible and robust process by

which individuals of a given species grow into morphologically

and functionally equivalent organisms. An intriguing question is

how is developmental reproducibility achieved and maintained

in the presence of extensive molecular, cellular, and environ-

mental variation? Most studies in developmental biology have

focused on deciphering molecular and cellular mechanisms

that regulate the average behavior of developmental pro-

cesses, with less attention paid to variability in these pro-

cesses. However, more and more evidence shows that

developmental processes exhibit considerable intrinsic vari-

CeThis is an open access article under the CC BY-N

ability at the molecular and cellular levels, such as in gene

expression (Burga et al., 2011; Raj et al., 2010), cell proliferation

(Fang and Yuste, 2017), growth (Hong et al., 2016), differentia-

tion, and morphogenesis (Haas et al., 2018). Systems-level

analysis of the sources, properties, and implications of devel-

opmental variability will provide an important perspective to un-

derstand the establishment and maintenance of developmental

reproducibility.

Caenorhabditis elegans (C. elegans) is an excellent model

system to use for investigating developmental reproducibility

and variability at the single-cell level in metazoans. C. elegans

development follows an invariant cell lineage in which different

animals generate the same set of cells in exactly the same way

(Sulston et al., 1983). This reproducibility allows the straightfor-

ward identification of equivalent cells between individuals and

cell-by-cell comparisons of phenotypes. In addition, the devel-

opmental behaviors of individual C. elegans cells are highly

stereotypical (Bao et al., 2008; Du et al., 2014; Moore et al.,

2013; Richards et al., 2013; Schnabel et al., 2006). This predict-

ability makes the analysis of variability in cellular behaviors

particularly meaningful and biologically relevant. Previous

studies quantified diverse cellular behaviors across different

animals during normal C. elegans embryogenesis and reported

that, despite being reproducible, cells do exhibit considerable

variability (Bao et al., 2008; Richards et al., 2013; Schnabel

et al., 1997). However, explicit investigation of the general

properties and regulation of cellular variability has not yet

been performed. As an integral part of developmental pro-

cesses, phenotypic variability contains information on the gen-

eral principles of developmental control of variability. Many key

questions about cellular variability remain to be answered,

such as the following: is cellular variability stochastic or deter-

ministic? Is variability coupled to the developmental properties

and functions of cells? What regulates cellular variability

spatially and temporally?

In this work, we used live imaging and quantitative single-cell

analysis to investigate the variability of three-dimensional (3D)

cell position in developing C. elegans embryos. Cell position is

a highly regulated and important developmental phenotype

that underlies many key developmental processes. Our results

reveal properties, implications, and developmental control of

ll Reports 26, 313–321, January 8, 2019 ª 2018 The Author(s). 313C-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

Figure 1. Quantification of Cell Position Variability during C. elegans Embryogenesis

(A) 3D time-lapse imaging of embryogenesis. Micrographs show 3D maximum projections of an embryo at representative stages. Scale bar, 5 mm.

(B) Automated cell identification and lineage tracing. The nucleus-localized, ubiquitously expressed mCherry signal is used to identify (circles) and trace (arrows)

all of the nuclei over time. Scale bar, 5 mm.

(C) Determination of the position of each nucleus at each time point.

(D) Quantification of relative cell position as a vector of geometrical distances (color gradient) between the centroid of the nucleus of a target cell (red) and that of

all of the other cells (white).

(E) Determination of cell position variability (PV). PV is quantified by comparing distance vectors of equivalent cells between embryos. The root-mean-square

deviation (RMSD) values between distance vectors for all pairwise comparisons are measured and then averaged to quantify PV.

(F) Tree visualization of PV levels during embryogenesis. Cells and their lineage relations are visualized as a tree structure, with vertical lines indicating cells and

horizontal lines indicating cell divisions. Cell placement on the tree is according to Sulston nomenclature. See also Figure S1 and Table S1.

cellular variability during embryogenesis, contributing to under-

standing the cellular basis of developmental reproducibility.

RESULTS

Quantification of Cell Position and Variability duringC. elegans EmbryogenesisCell position variability (hereafter referred to as PV) is defined as

the variability in the 3D position of a cell among wild-type em-

bryos. To quantify PV, we first determined the 3D position of

each cell at minute temporal resolution and then measured the

variability between wild-type embryos grown under identical

conditions. To visualize cells, we used ubiquitously expressed

mCherry to label every cell nucleus and performed 3D time-lapse

imaging to record embryogenesis for 6 hr at 75-s intervals (Fig-

314 Cell Reports 26, 313–321, January 8, 2019

ure 1A). Based on the fluorescence signal, we identified the 3D

position of each cell at each time point and traced them over

time to reconstruct the entire cell lineage out to the 350-cell

stage using an established method (Bao et al., 2006; Santella

et al., 2010, 2014) (Figures 1B and 1C). By the 350-cell stage,

the embryos have completed all but the last round of cell

division and contain >50% of all cells (1,341) produced during

embryogenesis.

To compare cell positions between embryos, we adopted a

relative approach for representing the position of a given target

cell as a vector of geometrical distances from the centroid of

its nucleus to those of all other co-existing cells (reference cells)

(Figure 1D). We then took PV as the differences in that vector be-

tween 28 wild-type embryos (Figure 1E). To reliably quantify

PV, we normalized the size differences between embryos and

Figure 2. Deterministic Nature of Cell Position Variability(A) Comparison of PV levels of leaf cells derived from 12 sub-lineages. See also Table S2.

(B) PV-changing points. Green circles denote 49 PV-changing points with a significantly higher PV of cells from the daughter cell lineage placed on the left side

(p < 0.05); red circles denote the opposite situation. Asterisks indicate 13 PV-changing points, with p value < 0.001. Numbered rectangles indicate 20 sub-lineage

groups classified by the 13 highly significant changing points. See also Table S2.

(C and D) Changes of PV levels in glp-1 (C, n = 8) and skn-1 (D, n = 11) mutant embryos. For each genotype, PV levels of cells from both sub-lineages are

normalized to the mean PV of cells from the ABala (glp-1) or the C sub-lineage (skn-1). See also STAR Methods and Figure S2.

(E) Comparison of PV levels between cell types. PV levels of cells are normalized by the average PV for their sub-lineages. See also Table S2.

(F andG) Comparison of PV levels of cells within each localization cluster. (F) Clusters of cells with similar localization. (G) PV levels of cells within each cluster. See

also Table S2.

(H and I) Correlation of PV between pairs of contact cells (H, n = 1,111) and between randomly selected mock contact cells (I, n = 1,111). Because cell lineage

influences PV, the lineage relation is controlled when selecting mock contact cells. For each pair of contact cells (A and B), a mock contact cell B whose lineage

(legend continued on next page)

Cell Reports 26, 313–321, January 8, 2019 315

aligned the developmental times of individual cells (STAR

Methods). Consequently, we quantified PV at a high temporal

resolution across an extended period of embryogenesis, yielding

>23,000 data points for 714 cells (Figure 1F; Table S1).

We validated the reproducibility, sensitivity, and robustness of

PV measurement, including the accuracy of position assignment

(STAR Methods); the robustness and sensitivity of the relative

quantification of PV (Figures S1A and S1B); the accuracy of

time alignment (Figure S1C); the sufficiency of the sample size

used (Figure S1D); the stability of PV measurement (Figure S1E);

and the consistency of PV across different embryo mounting

methods (Figure S1F).

Cell Lineage Identity Determines Cell PositionVariabilityWe first addressed the question: is PV stochastic or determin-

istic? If deterministic, then PV levels would be significantly asso-

ciated with the developmental properties of cells. The vast

knowledge available concerning C. elegans cells allowed us to

correlate PV with diverse cell properties including cell lineage,

fate, 3D localization, and symmetry.

Cell lineage describes the developmental origin and history of

all cells (Figure 1F). We evaluated whether the lineage history of

cells affects PV. We classified leaf cells (the traced terminal cells

at the 350-cell stage) into 12 groups based on the sub-lineages

they derived from and found that the levels of PV were sub-line-

age specific (Figure 2A; Table S2). For example, cells from MS

and P3 sub-lineages exhibited higher levels of PV, while cells

from the ABpra and C sub-lineages exhibited lower levels. To

further elaborate the relation between PV and cell lineage, we

searched for PV-changing points, defined as ancestral cells

that give rise to two sub-lineages with significantly different

levels of PV. Many PV-changing points were identified (49 points

at p < 0.05 and 13 points at p < 0.001,Wilcoxon signed-rank test;

Figure 2B; Table S2), suggesting that PV is strongly associated

with lineage identity. These PV-changing points partitioned the

entire cell lineage line into sub-lineageswhich exhibited relatively

homogeneous levels of PV. Figure 2B shows the 20 sub-lineage

groups partitioned by the 13 highly significant points. General

linear regression revealed that these sub-lineage groups explain

50.1% of the variance in PV levels.

To test whether lineage identity plays a causal role, we exam-

ined how PV changes when lineage identities were switched.

This question was first addressed using the ABala and ABara

sub-lineages, which exhibited significantly different levels of

PV (p = 1.18e�7; Figure 2A). These lineages are distinguished

by Notch signaling, and the loss of function of Notch induces a

switch of ABara lineage identity to that of ABala (Du et al.,

2014; Hutter and Schnabel, 1994; Moskowitz et al., 1994). If line-

distance to cell A equaled that between cell A and cell B is selected as a control. Lin

that separate the two cells from their lowest common ancestor. See also Figure

(J) Schematic representation of left-right symmetric organization of cells.

(K and L) Correlation of PV levels between pairs of left-right symmetric cells (K, n =

the lineage relation is controlled when selecting mock symmetric cells. See also

(M) Top: red and green represent cells that will localize on the left or right side f

coefficient of PV between left-right symmetric cells (blue) and mock symmetric c

stage (90 pairs), and AB256 stage (164 pairs). The p value was calculated by t te

316 Cell Reports 26, 313–321, January 8, 2019

age identity underlies differential PV, then we expected that PV

levels of the ABala and ABara sub-lineages in Notch mutants

would be indistinguishable because the identities of the two

sub-lineages are identical. In glp-1/Notch embryos (n = 8), we

confirmed by tissue marker expression pattern and character-

istic programmed cell death (Figure S2A) that the ABara sub-

lineage switched to ABala. In glp-1(e2142), we concomitantly

detected a significant reduction in the relative PV levels of cells

from the ABara sub-lineage as compared to that of wild-type

(p = 1.19e�9; Figure 2C). In addition, the difference in PV be-

tween the ABala and ABara sub-lineages in glp-1 embryos

(p = 0.097) was significantly reduced as compared to the wild-

type (p = 1.18e�7).

We also tested other two sub-lineages (MS and C) exhibiting

significantly different levels of PV (p = 1.34e�10; Figure 2A)

and again validated that the PV level changes with lineage iden-

tity. The loss of function of skn-1/NRF2 induces a lineage identity

switch from MS to C (Bowerman et al., 1992; Du et al., 2014). In

skn-1(zu67), we found that the relative PV of the MS sub-lineage

was significantly reduced (n = 11, p = 1.30e�8) compared to that

of wild-type (Figure 2D), becoming indistinguishable from that of

C (p = 0.93). We confirmed that the MS lineage in skn-1(zu67)

was transformed to that of C based on cell lineage topology (Fig-

ure S2B). The above results confirmed the important role of line-

age identity in determining PV level.

Cell fate is another important developmental property related

to the function of cells. However, we found that the PV of

different cell types did not differ significantly (Figure 2E). Hence,

at least by the 350-cell stage, cell fate does not significantly influ-

ence PV. The above correlative and perturbation results reveal a

deterministic nature for PV dictated by an intrinsic mechanism

that relies on cell lineage but not cell fate.

Cells Similarly Localized and in Physical Contact ExhibitSimilar Cell Position VariabilityAnother important developmental property of cells is their 3D

localization. Does the localization of cells affect PV? To address

this,we clustered cells basedon similarity in their relative position

and compared PV levels between clusters. We identified 15

clusters of cells with similar localization and they exhibited char-

acteristic levels of PV (Figures 2F and 2G; Table S2). The intra-

cluster divergency of PV was significantly lower than expected

(p = 6.55e�100; Figure S2C). General linear regression revealed

that cell localization explains 61.7% of the variance in PV.

Because both lineage identity and localization were associ-

ated with PV and lineage identity could regulate localization,

we examined the influence of each factor while controlling for

the influence of the other. The results excluded the possibility

that lineage identity or cell localization was a cofounder of

eage distance between two cells is defined as the total number of cell divisions

S2.

320) and pairs of randomly selectedmock symmetric cells (L, n = 320). As in (I),

Figure S2 and Table S2.

ollowing embryogenesis, respectively. Bottom: comparison of the correlation

ells (gray) at developmental stages before AB128 stage (66 cell pairs), AB128

st. See also Figure S2 and Table S2.

each other (STAR Methods; Figures S2D and S2E). Lineage and

localization jointly explained 76.8% of the variance and analysis

of their relative importance in a linearmodel showed that in all ex-

plained variance, the contributions of lineage and localization

were generally orthogonal (45% versus 55%).

To further validate the influence of cell localization on PV, we

applied an alternative strategy for identifying similarly localized

cells. In a previous study, a Voronoi algorithm was used to infer

a reference map describing cell-cell contact (Chen et al., 2018).

By definition, contact cells must be similarly localized. We

compared PV levels between contact cells and found that they

were significantly correlated (R = 0.743, p = 3.76e�195; Fig-

ure 2H; Table S2). The correlation coefficient was significantly

higher (p = 4.37e�67) than that obtained for mock cells (Fig-

ure 2I). In addition, the divergency of PV values between

contact cells was significantly lower than between mock cells

(p = 4.43e�59; Figure S2F). This result independently confirmed

that cell localization influences PV.

In addition to having similar localization, contact cells must

also physically interact. We therefore examined whether the

presence of physical contact further reduced divergency in PV.

We integrated the cell localization clusters described above

with the cell contact data and within each localization cluster

compared the divergency of PV between cell pairs with or

without cell contact (Table S2). We found that the consistency

of PV levels between contact cells was significantly higher than

in those without contact (p = 3.40e�4; Figure S2G), suggesting

that in addition to similar localization, physical contact further in-

fluences PV.

The above results suggest a potential extrinsic regulation of

PV in which cells with similar 3D localization exhibit similar levels

of PV, especially for those with physical contact.

Left-Right Symmetric Cells Exhibit Similar Cell PositionVariabilityAs a bilateral animal, the C. elegans body plan is generally left-

right symmetric (Figure 2J). Following embryogenesis, cells

from 30 pairs of sub-lineages (Sulston et al., 1983) are generally

organized in a left-right symmetric pattern (Table S2). We found

that the PV levels of corresponding left and right cells (n = 320)

were highly correlated (R = 0.739, p = 1.96e�56; Figure 2K).

The correlation coefficient was significantly higher (t test,

p = 4.82e�22) than that obtained for mock cells (Figure 2L).

Moreover, the divergency of PV values between left-right

symmetric cells was significantly lower than in mock cells

(p = 5.55e�12; Figure S2H).

Because the establishment of left-right symmetry is progres-

sive (Pohl and Bao, 2010), we examined whether the correlated

PV levels were specifically coupled to a stage in which symmetry

was established. As exemplified in Figure 2M, the establishment

of left-right symmetry is time dependent. We classified all of the

symmetric cell pairs into three categories according to develop-

mental stage (based on the number of AB lineage-derived cells;

Table S2): before AB128 stage (symmetry not established),

AB128 stage (partially established), and AB256 stage (well es-

tablished). The highest correlation coefficient was obtained for

cell pairs at the AB256 cell stage (Figure 2M). In contrast, at

the developmental stages before AB128, the similarity of PV

between pairs was not significantly different from that of mock

symmetric cells (Figures 2M and S2I). These results suggest

that the morphogenic organization of cells in the embryo influ-

ences PV.

A Concordant Low-High-Low Dynamic Pattern of CellPosition VariabilityWe next sought to examine the temporal dynamics of PV during

embryogenesis. Quantification of global PV (average PV across

all cells) at each embryogenesis time point revealed a smooth

and dynamic temporal profile (Table S3). As shown in Figure 3A,

this profile follows a low-high-low pattern: that is, during early

embryogenesis, PV progressively increases with time until mid-

embryogenesis; thereafter, the PV level gradually reduces, and

by the 350-cell stage, it reaches a level that is comparable to

the beginning of embryogenesis. To test whether this observed

global pattern was an artifact of averaging PV levels across

many cells, we also quantified PV changes in individual cell

developmental tracks. A cell track is defined as following tempo-

rally aligned cells within a cell lineage from the ancestral cell to a

leaf cell (Figure S3A). We found that individual cells also exhibit a

highly concordant pattern of PV change over time (R = 0.719;

Figure S3A).

An important morphogenesis process that occurs during the

analyzed time interval is gastrulation, during which a large

number of cells move from the embryo surface into the interior

(internalization or ingression) (Harrell and Goldstein, 2011). In

addition, a recent study revealed that certain cells also migrate

from the interior to the surface (egression) (Jelier et al., 2016).

Do these characteristic cell movements influence PV levels

and their dynamics? We compared the PV of each ingressing

or egressing cell to those of stage and sub-lineagematched con-

trol cells and found that in the majority of cases, PV levels were

comparable to those of control cells (Figure S3B). In addition, the

temporal dynamics of PV in these cells were similar to those of

control cells (Figure S3C). These results suggest that ingression

and egression movements during gastrulation do not signifi-

cantly affect PV and its temporal dynamics.

It should be noted that the accumulative nature of cell position

determines that PV levels over time are not independent values.

For example, the PV of a previous time point will be propagated

to the next and that of a mother cell to its two daughter cells after

cell division. Null models in which the direction of cell displace-

mentwas randomized consistently showed amonotone increase

of PV over time (Figure 3B). Therefore, the observed global down-

regulation of PV levels was a consequence of active control.

Fate Specification Contributes to GlobalDownregulation of Cell Position VariabilityWhat underlies the global downregulation of PV? A previous

study proposed a model in which the blastomere fates could

provide spatial codes to their descendant cells and showed

that changes of fate accordingly induce the re-localization of

cells (Schnabel et al., 2006).We testedwhether fate specification

underlies PV downregulation. Specifically, we quantified the sin-

gle-cell protein expression of four well-characterized tissue fate

determinants as a readout of the timing of fate specification. In

most cases, the progenitor cells with a stable expression of

Cell Reports 26, 313–321, January 8, 2019 317

Figure 3. Temporal Regulation of Cell Position Variability

(A) Boxplot shows PV levels across all cells at each embryogenesis time point with vertical lines indicating five characteristic developmental stages. See also

Figure S3 and Table S3.

(B) Simulation of PV level changes over time (28 embryos, 100 cells, 150 time points) with various cell displacement distances in one time interval. D is equivalent

to the median displacement distance observed in real embryos. The initial cell positions are identical (PV = 0), and for each time interval all of the cells in each

embryo are allowed to move randomly over a given distance but within the predefined embryonic space.

(C) Expression of fate determinants correspondswith PV downregulation. Color-coded trees represent the expression of fate determinants in corresponding sub-

lineages. Red lines indicate progenitor cells that give rise to corresponding cell types with downregulated PV. See also Figure S3.

(D) Abolishment of gut fate specification in end-1;end-3 embryos. Color-coded trees represent the lineage pattern and single-cell expression of the gut-specific

reporter ELT-2. Overproliferation of E sub-lineage cells (32) in end-1;end-3 compared to the wild-type (16) is due to a loss of original cell fate.

(E) Boxplots show the change of global PV during embryogenesis for cells from E sub-lineage in wild-type (top, 28 embryos) and end-1;end-3 (bottom, 9

embryos). Vertical lines indicate developmental stages based on the number of E-derived cells. See also Figure S3.

fate determinants showed PV downregulation in the same cell or

in its daughter cells (Figures 3C and S3D). In only 2 of 101 cases

was the expression of the fate determinants later than PV down-

regulation. These results suggest that downregulation of PV

accompanies fate specification.

If fate specification plays an important role in reducing PV, then

we expected that perturbation of fate specificationwould abolish

this downregulation. We tested this possibility in gut cells, in

which the regulation of fate specification is well characterized.

All gut cells exclusively differentiate from a single progenitor

cell called E, and this process is regulated by end-1 and end-3;

loss of function of both genes specifically abolishes gut fate

specification in the E sub-lineage (Maduro et al., 2005). We con-

318 Cell Reports 26, 313–321, January 8, 2019

structed an end-1;end-3 double mutant and verified that fate

specification was abolished in the mutant embryos, as deter-

mined by the absence of the characteristic cell lineage topology

and of the expression of the gut marker ELT-2 (Figure 3D). In the

mutant (n = 10), we found that PV downregulation was eliminated

for cells from the E lineage and that PV levels accumulated over

time (Figure 3E). ThePVdynamic pattern in other cellswas largely

unaffected (Figure S3E). This result suggests that cell fate spec-

ification contributes to PV downregulation.

The above results reveal general properties of the temporal

changes of PV. First, cell position is highly consistent in early

and later stages, and PV dynamics are highly concordant. Sec-

ond, cell fate specification contributes to a more reproducible

Figure 4. Developmental Control of Cell Po-

sition Variability

(A) Strategy for simulating expected PV levels. The

red arrow indicates the direction and amount of

cell displacement observed in a real embryo; black

arrows represent simulated cell displacements.

(B and C) Comparison of observed PV level (x axis)

to expected (y axis) for every cell (B, n = 714) and

for every time interval (n = 23,486) during

embryogenesis (C). The diagonal line indicates

equal PV levels.

(D) Comparison of the ratio of observed to ex-

pected PV at different times within a cell cycle.

(E–J) Scatter plots compare PV levels (E for hmp-

2, F for jac-1, and G for inx-2) and velocity of cell

displacement (H for hmp-2, I for jac-1, and J for

inx-2) in the mutant embryos to that of wild-type in

equivalent cells (n = 714). Velocity is measured as

the average distance of cell displacement in one

time interval. Diagonal lines represent equal PV or

velocity. Percentages indicate the frequency of

cells showing higher PV or velocity of cell

displacement in mutant embryos compared to the

wild-type. p values were calculated by the Wil-

coxon signed-rank test. See also Figure S4 and

Table S4.

cell position, likely through promoting equivalent cells in individ-

ual embryos to localize in specific positions, hence reducing

variability.

Cell Position Variability Is Tightly Controlled throughoutEmbryogenesisBecause high variability could be detrimental to the stability of

embryogenesis in a population, we investigated the develop-

mental control of PV. We first tested whether PV was subject

to tight constraint by comparing observed PV levels to expected

values.We simulated expected PV by randomizing the directions

of cell displacements while keeping the total distance in a cell cy-

cle identical (Figure 4A). The resulting expected PV levels were

significantly higher than observed (p = 1.49e�118; Figure 4B),

suggesting that PV is under stringent constraint. We further

applied the simulation to all cells in two successive time points.

Once again, the expected PV was higher than the observed (Fig-

ure 4C), suggesting that cell displacement is directional

throughout embryogenesis. We detected significantly higher

constraint on PV during cell division (Figures 4C and 4D), sug-

gesting that the positioning of daughter cells is tightly regulated

during the division of most mother cells.

Cell Adhesion and Gap Junction Function to RestrictCell Position VariabilityTo elucidate the mechanism underlying the restriction of PV, we

analyzed PV levels in mutants of genes that are potentially impli-

Cell

cated in variability control. We reasoned

that molecules mediating the cell-cell

junctions would be important for restrict-

ing random cell displacement, hence,

reducing variability. We quantified PV in

the mutants of three genes that are important for cell-cell

adhesion and gap junctions: hmp-2/b-catenin and jac-1/p120

catenin, which are components of the cadherin-catenin complex

that is required for cell-cell adhesion (Pettitt, 2005), and inx-2/

innexin, which encodes a gap junction protein known to be ex-

pressed during embryogenesis (Altun et al., 2009). In all of the

mutants, we detected significant increase in PV levels for thema-

jority of cells (Figures 4E–4G; Table S4), suggesting that these

genes contribute to PV restriction. Concomitantly, we observed

in all of the mutants that the velocity of cell displacement

(average amount cell displacement in a time interval) was signif-

icantly increased as compared to that of wild-type (Figures 4H–

4J; Table S4). These results suggest that cell adhesion and gap

junctions restrict PV potentially by restraining the capacity for

random cell displacements. However, the characteristic tempo-

ral pattern of PV dynamics was largely preserved in all three mu-

tants (Figure S4A), suggesting that these genes do not play

essential roles in regulating PV dynamics.

DISCUSSION

In this study, we focused on cell position, an important cellular

phenotype that describes the systems behavior of embryogen-

esis, and performed a quantitative analysis of cellular variability

during in vivo development. Our findings extend the understand-

ing of the properties and developmental control of cellular

variability.

Reports 26, 313–321, January 8, 2019 319

First, we systematically quantified and revealed important

properties of PV. Through live imaging and cell tracing, we

have accurately quantified 3D cell position and its variability at

high temporal resolution throughout an extended portion of

embryogenesis (Figure 1). From the data, we show that PV is

determined by cell lineage identity and is coupled tomany devel-

opmental properties of cells (Figure 2). These results reveal an

unexpected deterministic nature of PV. It indicates that cellular

variability is highly biologically relevant and could be a specific

target of regulation. In the present study, we traced cell positions

until the 350-cell stage. Cell positions were approximated using

nucleus centroids, as the nuclei are generally localized in the

centers of cells. The analysis of cell positions and their variability

in later embryonic stages remains challenging due to the diffi-

culty of accurately tracing cells and the complexity of reliably

defining cell positions. It was reported that in very late embryos,

while the position of certain neuronal cells in neuropils is highly

reproducible, their cell-body placements exhibit considerable

variability (Insley and Shaham, 2018). Therefore, to study cell po-

sitions and their variability in late embryos, it is important to

consider cell shape and structure, especially for those cells

with specialized morphology.

Second, we reveal highly concordant low-high-low temporal

dynamics of PV across diverse cells (Figure 3). It seems that em-

bryos can tolerate relatively high systems-wide variability, which

is later relieved by a global downregulation of variability. Our re-

sults suggest that cell fate specification reduces PV, based on

two observations: (1) expression of fate determinants coincides

with variability reduction (Figure 3C) and (2) perturbation of gut

cell fate concomitantly abolishes the downregulation of vari-

ability in E cells (Figure 3E) as compared to other cells (Fig-

ure S3E). However, this result does not rule out the possibility

that some unknown factors downstream of cell fate, such as

change of cell position or cell number, could be responsible for

this observation. Thus, whether fate specification per se plays

a causal role in the process requires further investigation. We hy-

pothesize that this regulated downregulation of variability level

could function as a correction mechanism to ensure reproduc-

ible cell positions while also conferring plasticity. We found in

two mutants (hmp-2 and inx-2) that increased PV relative to

wild-type at earlier stages was partially relieved by this global

downregulation (Figure S4B). A previous study proposed that

cell fate provides a positional code that guides cells to localize

to specific regions for pattern formation (Schnabel et al., 2006).

Here, our data suggest that a related mechanism could be

invoked to reduce and buffer variability.

Finally, our findings shed light on the spatiotemporal control

of PV. We show that PV is tightly controlled throughout

embryogenesis to maintain a reproducible cellular configuration

in embryos (Figure 4). It is well known that in certain early blas-

tomeres such as EMS (Bei et al., 2002; Sugioka and Bower-

man, 2018) and ABar (Walston et al., 2004), the mitotic spindle

is tightly regulated during cell division. Our data show that cell

movements within a cell cycle are also subject to constraint,

although to a lesser extent compared to cell division. In addi-

tion, our data predict the existence of cell lineage-based

intrinsic and cell localization-based extrinsic mechanisms in

determining the variability level. As an initial step for character-

320 Cell Reports 26, 313–321, January 8, 2019

izing regulators of PV, we identified that proteins mediating cell-

cell junctions contribute to reducing PV (Figure 4). It will be

interesting to test further whether biophysical properties of em-

bryos, distribution of morphogen gradients, or local cell-cell

signaling plays a role in specifying region-specific variability

levels.

STAR+METHODS

Detailed methods are provided in the online version of this paper

and include the following:

d KEY RESOURCES TABLE

d CONTACT FOR REAGENT AND RESOURCE SHARING

d EXPERIMENTAL MODEL AND SUBJECT DETAILS

B C. elegans Strains and Culture

d METHOD DETAILS

B Embryo Mounting

B 3D Time-lapse Imaging

B CRISPR/Cas9-based Gene Knockout and Knockin

d QUANTIFICATION AND STATISTICAL ANALYSIS

B Cell Identification, Tracing, and Cell Lineage Construc-

tion

B Measurement of PV in Wild-type Embryos

B Assessment of the Accuracy and Reliability of PVMea-

surement

B Measurement of PV of Mutant Embryos

B Comparison of PV between Different Cell Types

B Identification of Cells with Similar Localization

B Confounder Analysis

B Quantification of Fate Determinant Expression in Sin-

gle Cells

B Simulations

B Statistics

d DATA AND SOFTWARE AVAILABILITY

SUPPLEMENTAL INFORMATION

Supplemental Information includes four figures and four tables and can be

found with this article online at https://doi.org/10.1016/j.celrep.2018.12.052.

ACKNOWLEDGMENTS

This work was supported by grants from the National Natural Science Founda-

tion of China (31771598 to Z.D.), and the Strategic Priority Research Program

of the Chinese Academy of Sciences (XDB19000000 to Z.D.). Some strains

were provided by the Caenorhabditis Genetics Cancer Center (CGC), which

is funded by the NIH Office of Research Infrastructure Programs (P40

OD010440).

AUTHOR CONTRIBUTIONS

X.L. and Z.D. conceived the project and designed the study; W.X., R.F., L.X.,

and X.M. conducted the experiments and generated the data; X.L., Z.Z., and

Z.D. performed the data analysis and statistics; Z.D. wrote the manuscript,

with input from all of the authors.

DECLARATION OF INTERESTS

The authors declare no competing interests.

Received: September 9, 2018

Revised: November 6, 2018

Accepted: December 11, 2018

Published: January 8, 2019

REFERENCES

Altun, Z.F., Chen, B., Wang, Z.W., and Hall, D.H. (2009). High resolution map of

Caenorhabditis elegans gap junction proteins. Dev. Dyn. 238, 1936–1950.

Bao, Z., Murray, J.I., Boyle, T., Ooi, S.L., Sandel, M.J., and Waterston, R.H.

(2006). Automated cell lineage tracing in Caenorhabditis elegans. Proc. Natl.

Acad. Sci. USA 103, 2707–2712.

Bao, Z., Zhao, Z., Boyle, T.J., Murray, J.I., and Waterston, R.H. (2008). Control

of cell cycle timing during C. elegans embryogenesis. Dev. Biol. 318, 65–72.

Bei, Y., Hogan, J., Berkowitz, L.A., Soto, M., Rocheleau, C.E., Pang, K.M.,

Collins, J., and Mello, C.C. (2002). SRC-1 and Wnt signaling act together to

specify endoderm and to control cleavage orientation in early C. elegans em-

bryos. Dev. Cell 3, 113–125.

Bowerman, B., Eaton, B.A., and Priess, J.R. (1992). skn-1, a maternally ex-

pressed gene required to specify the fate of ventral blastomeres in the early

C. elegans embryo. Cell 68, 1061–1075.

Burga, A., Casanueva, M.O., and Lehner, B. (2011). Predicting mutation

outcome from early stochastic variation in genetic interaction partners. Nature

480, 250–253.

Chen, L., Ho, V.W.S., Wong, M.K., Huang, X., Chan, L.Y., Ng, H.C.K., Ren, X.,

Yan, H., and Zhao, Z. (2018). Establishment of Signaling Interactions with

Cellular Resolution for Every Cell Cycle of Embryogenesis. Genetics 209,

37–49.

Du, Z., Santella, A., He, F., Tiongson, M., and Bao, Z. (2014). De novo inference

of systems-level mechanistic models of development from live-imaging-based

phenotype analysis. Cell 156, 359–372.

Du, Z., Santella, A., He, F., Shah, P.K., Kamikawa, Y., and Bao, Z. (2015). The

Regulatory Landscape of Lineage Differentiation in a Metazoan Embryo. Dev.

Cell 34, 592–607.

Fang, W.Q., and Yuste, R. (2017). Overproduction of Neurons Is Correlated

with Enhanced Cortical Ensembles and Increased Perceptual Discrimination.

Cell Rep. 21, 381–392.

Frøkjær-Jensen, C., Davis, M.W., Sarov,M., Taylor, J., Flibotte, S., LaBella,M.,

Pozniakovsky, A., Moerman, D.G., and Jorgensen, E.M. (2014). Random and

targeted transgene insertion in Caenorhabditis elegans using a modified

Mos1 transposon. Nat. Methods 11, 529–534.

Haas, P.A., Hohn, S.S.M.H., Honerkamp-Smith, A.R., Kirkegaard, J.B., and

Goldstein, R.E. (2018). The noisy basis of morphogenesis: mechanisms and

mechanics of cell sheet folding inferred from developmental variability. PLoS

Biol. 16, e2005536.

Harrell, J.R., and Goldstein, B. (2011). Internalization of multiple cells during C.

elegans gastrulation depends on common cytoskeletal mechanisms but

different cell polarity and cell fate regulators. Dev. Biol. 350, 1–12.

Hong, L., Dumond, M., Tsugawa, S., Sapala, A., Routier-Kierzkowska, A.L.,

Zhou, Y., Chen, C., Kiss, A., Zhu, M., Hamant, O., et al. (2016). Variable Cell

Growth Yields Reproducible Organ Development through Spatiotemporal

Averaging. Dev. Cell 38, 15–32.

Hostettler, L., Grundy, L., Kaser-Pebernard, S., Wicky, C., Schafer, W.R., and

Glauser, D.A. (2017). The Bright Fluorescent Protein mNeonGreen Facilitates

Protein Expression Analysis In Vivo. G3 (Bethesda) 7, 607–615.

Hutter, H., and Schnabel, R. (1994). glp-1 and inductions establishing embry-

onic axes in C. elegans. Development 120, 2051–2064.

Insley, P., and Shaham, S. (2018). Automated C. elegans embryo alignments

reveal brain neuropil position invariance despite lax cell body placement.

PLoS One 13, e0194861.

Jelier, R., Kruger, A., Swoger, J., Zimmermann, T., and Lehner, B. (2016).

Compensatory Cell Movements Confer Robustness to Mechanical Deforma-

tion during Embryonic Development. Cell Syst. 3, 160–171.

Katzman, B., Tang, D., Santella, A., and Bao, Z. (2018). AceTree: a major up-

date and case study in the long term maintenance of open-source scientific

software. BMC Bioinformatics 19, 121.

Maduro, M.F., Hill, R.J., Heid, P.J., Newman-Smith, E.D., Zhu, J., Priess, J.R.,

and Rothman, J.H. (2005). Genetic redundancy in endoderm specification

within the genus Caenorhabditis. Dev. Biol. 284, 509–522.

Mello, C.C., Kramer, J.M., Stinchcomb, D., and Ambros, V. (1991). Efficient

gene transfer in C. elegans: extrachromosomal maintenance and integration

of transforming sequences. EMBO J. 10, 3959–3970.

Moore, J.L., Du, Z., and Bao, Z. (2013). Systematic quantification of develop-

mental phenotypes at single-cell resolution during embryogenesis. Develop-

ment 140, 3266–3274.

Moskowitz, I.P., Gendreau, S.B., and Rothman, J.H. (1994). Combinatorial

specification of blastomere identity by glp-1-dependent cellular interactions

in the nematode Caenorhabditis elegans. Development 120, 3325–3338.

Murray, J.I., Bao, Z., Boyle, T.J., Boeck, M.E., Mericle, B.L., Nicholas, T.J.,

Zhao, Z., Sandel, M.J., andWaterston, R.H. (2008). Automated analysis of em-

bryonic gene expression with cellular resolution in C. elegans. Nat. Methods 5,

703–709.

Paix, A., Wang, Y., Smith, H.E., Lee, C.Y., Calidas, D., Lu, T., Smith, J.,

Schmidt, H., Krause, M.W., and Seydoux, G. (2014). Scalable and versatile

genome editing using linear DNAs with microhomology to Cas9 Sites in Cae-

norhabditis elegans. Genetics 198, 1347–1356.

Pettitt, J. (2005). The cadherin superfamily. In Wormbook, The C. elegans

Research Community (WormBook). https://doi.org/10.1895/wormbook.1.50.1.

Pohl, C., and Bao, Z. (2010). Chiral forces organize left-right patterning in C.

elegans by uncoupling midline and anteroposterior axis. Dev. Cell 19,

402–412.

Raj, A., Rifkin, S.A., Andersen, E., and van Oudenaarden, A. (2010). Variability

in gene expression underlies incomplete penetrance. Nature 463, 913–918.

Richards, J.L., Zacharias, A.L., Walton, T., Burdick, J.T., and Murray, J.I.

(2013). A quantitative model of normal Caenorhabditis elegans embryogenesis

and its disruption after stress. Dev. Biol. 374, 12–23.

Santella, A., Du, Z., Nowotschin, S., Hadjantonakis, A.K., and Bao, Z. (2010). A

hybrid blob-slice model for accurate and efficient detection of fluorescence

labeled nuclei in 3D. BMC Bioinformatics 11, 580.

Santella, A., Du, Z., and Bao, Z. (2014). A semi-local neighborhood-based

framework for probabilistic cell lineage tracing. BMC Bioinformatics 15, 217.

Schnabel, R., Hutter, H., Moerman, D., and Schnabel, H. (1997). Assessing

normal embryogenesis in Caenorhabditis elegans using a 4D microscope:

variability of development and regional specification. Dev. Biol. 184, 234–265.

Schnabel, R., Bischoff, M., Hintze, A., Schulz, A.K., Hejnol, A., Meinhardt, H.,

and Hutter, H. (2006). Global cell sorting in the C. elegans embryo defines a

new mechanism for pattern formation. Dev. Biol. 294, 418–431.

Sugioka, K., and Bowerman, B. (2018). Combinatorial Contact Cues Specify

Cell Division Orientation by Directing Cortical Myosin Flows. Dev. Cell 46,

257–270.e5.

Sulston, J.E., Schierenberg, E., White, J.G., and Thomson, J.N. (1983). The

embryonic cell lineage of the nematode Caenorhabditis elegans. Dev. Biol.

100, 64–119.

Walston, T., Tuskey, C., Edgar, L., Hawkins, N., Ellis, G., Bowerman, B.,Wood,

W., and Hardin, J. (2004). Multiple Wnt signaling pathways converge to orient

the mitotic spindle in early C. elegans embryos. Dev. Cell 7, 831–841.

Cell Reports 26, 313–321, January 8, 2019 321

STAR+METHODS

KEY RESOURCES TABLE

REAGENT or RESOURCE SOURCE IDENTIFIER

Bacterial and Virus Strains

E. coli: OP50 Caenorhabditis Genetics Center OP50

Deposited Data

4D images, cell lineages and positions, and PV results This paper (available at: https://

usegalaxy.org/u/zhuodu/h/

dataset-cell-position-variability)

N/A

Experimental Models: Organisms/Strains

C. elegans: Strain RW10434: itIs37 [pie-1p::mCherry::H2B::pie-1

30UTR + unc-119(+)] IV. stIs10116 [his-72p::his-24::mCherry::let-858

30UTR + unc-119(+)]. stIs10394 [cnd-1::TGF(3C3)::GFP::TY1::3xFLAG].

Used to quantify cell position and variability

Caenorhabditis Genetics Center RW10434

C. elegans: Strain BV137: glp-1(e2142); itIs37 [pie-1p::mCherry::H2B::pie-1

30UTR + unc-119(+)] IV. stIs10116 [his-72p::his-24::mCherry::let-858 30UTR +

unc-119(+)]. Used to determine the influence of lineage identity on PV

This paper N/A

C. elegans: Strain SYS403: skn-1(zu67) IV/oxTi994 [eft-3p::GFP::2xNLS::tbb-2

30UTR + PuroR] IV; ujIs113 ([pie-1prom::mCherry::H2B; nhr-2prom::mCherry::

HIS-24-let858UTR;unc-119(+)]) II. Used to determine the influence of lineage

identity on PV

This paper N/A

C. elegans: Strain RW10425: itIs37 [pie-1p::mCherry::H2B::pie-1 30UTR +

unc-119(+)] IV. stIs10116 [his-72p::his-24::mCherry::let-858 30UTR + unc-

119(+)]. stIs10389 [pha-4::TGF(3E3)::GFP::TY1::3xFLAG]. Used to analyze

pharynx fate specifier PHA-4 expression

Caenorhabditis Genetics Center RW10425

C. elegans: Strain SYS121: wgIs354 [elt-1::TY1::EGFP::3xFLAG + unc-

119(+)];ujIs113 ([pie-1prom::mCherry::H2B; nhr-2prom::mCherry::HIS-24-

let858UTR;unc-119(+)]) II. Used to analyze hypodermal fate specifier

ELT-1 expression

This paper N/A

C. elegans: Strain SYS167: wgIs659 [unc-120::TY1::EGFP::3xFLAG + unc-

119(+)];ujIs113 ([pie-1prom::mCherry::H2B; nhr-2prom::mCherry::HIS-24-

let858UTR;unc-119(+)]) II. Used to analyze muscle fate specifier UNC-120

expression

This paper N/A

C. elegans: Strain SYS549: end-1(dev162[end-1:: mNeonGreen] V; ujIs113

([pie-1prom::mCherry::H2B; nhr-2prom::mCherry::HIS-24-let858UTR;unc-

119(+)]) II. Used to analyze gut fate specifier END-1 expression

This paper N/A

C. elegans: Strain SYS412: elt-2(dev99[mNeonGreen::elt-2]) X; ujIs113

([pie-1prom::mCherry::H2B; nhr-2prom::mCherry::HIS-24-let858UTR;unc-

119(+)]) II. Used to determine the influence of end-1;end-3 on gut fate

specification

This paper N/A

C. elegans: Strain SYS691: end-1&ric-7(ok558); end-3(dev85) V/oxTi991 [eft-

3p::GFP::2xNLS::tbb-2 30UTR + HygroR] V; elt-2(dev99[mNeonGreen::elt-2])

X; ujIs113 ([pie-1prom::mCherry::H2B; nhr-2prom::mCherry::HIS-24-let858UTR;

unc-119(+)]) II. Used to determine the influence of end-1;end-3 on gut fate

specification

This paper N/A

C. elegans: Strain SYS713: end-1&ric-7(ok558); end-3(dev85) V/oxTi991 [eft-3p::

GFP::2xNLS::tbb-2 30UTR + HygroR] V; ujIs113 ([pie-1prom::mCherry::H2B;

nhr-2prom::mCherry::HIS-24-let858UTR;unc-119(+)]) II. Used to determine

the influence of gut fate specification on PV dynamics

This paper N/A

(Continued on next page)

e1 Cell Reports 26, 313–321.e1–e7, January 8, 2019

Continued

REAGENT or RESOURCE SOURCE IDENTIFIER

C. elegans: Strain SYS112: hmp-2(qm39) I; itIs37 [pie-1p::mCherry::H2B::pie-1

30UTR + unc-119(+)] IV. stIs10116 [his-72p::his-24::mCherry::let-858 30UTR +

unc-119(+)]. stIs10394 [cnd 1::TGF(3C3)::GFP::TY1::3xFLAG]. Used to determine

the function of hmp-2 in PV regulation

This paper N/A

C. elegans: Strain SYS123: jac-1(ok3000) IV; itIs37 [pie-1p::mCherry::H2B::pie-1

30UTR + unc-119(+)] IV. stIs10116 [his-72p::his-24::mCherry::let-858 30UTR +

unc-119(+)]. stIs10394 [cnd-1::TGF(3C3)::GFP::TY1::3xFLAG]. Used to determine

the function of jac-1 in PV regulation

This paper N/A

C. elegans: Strain SYS124: inx-2(ok376) X; itIs37 [pie-1p::mCherry::H2B::pie-

1 30UTR + unc-119(+)] IV. stIs10116 [his-72p::his-24::mCherry::let-858 30UTR +

unc-119(+)]. stIs10394 [cnd-1::TGF(3C3)::GFP::TY1::3xFLAG]. Used to determine

the function of inx-2 in PV regulation

This paper N/A

C. elegans: Strain VC271: end-1&ric-7(ok558) V. Used to generate end-1;

end-3 _underdouble mutant

Caenorhabditis Genetics Center VC271

Oligonucleotides

single-guide RNA (sgRNA) targeting end-3 gene for gene knockout: end-3-KO-

N-SgRNA: AATTCAATGAAGAAGGCTAC

This paper N/A

single-guide RNA (sgRNA) targeting end-3 gene for gene knockout: end-3-KO-

C-SgRNA: AACTGGACGGCGTCTTCTCT

This paper N/A

Primers to detect end-3 gene knockout: end-3-ko-F: 50-GTATGTTCCAAGTGT

TCCGATG-30This paper N/A

Primers to detect end-3 gene knockout: end-3-ko-R: 50-TTATTATGGAAACCA

GCAGAGG-30This paper N/A

Synthetic oligo as 50 arm for knocking in mNeonGreen in end-1 gene This paper N/A

GCATGCCTACTCACAACAGTATCCCCAAATTGGACAAGACTATCAACAACA

GGAAATCGAAAATATTCCACCAGTCTCTACAAATCGCAAAATTGTGAACAAA

AAGGTTTGGTTTTCTTCACTTTCAGAATCATTTTTATTTTATCATTTTCAGCCC

TCAACGTTCCACACGAACTCTGTCTGCTCCAATCCAAATTGTAGAACTCGTG

AGACAACTCTATGGAGAAGAACTGATTCAGGAGCTATTGAGTGCAAGTGAGT

TGAGGTTTTTGAAATTAAAAATGAATAAGATAGGTAGTTTTTATTTCAAAATAG

AAAAAAAAATCAGTTTTGAATAATTCACTATTCTATTTTTTCAAGAAATTTAGTT

GTTCCAAAAAAAATTCTCGAAAATTTTTTAACGACGTCCGTAATCTTAGCGGA

GCCCCTGAGAGAAACATCATTTGTTTTTGAAGGCTAGTTCAGGCGAGGTGA

GACGGGGGTTAGACGCCTGTTTGAAACCTGCCATGAATTCATAAATTTATCA

TTTTATGAAAATTAAATAAAGAAATGTGAAAAAATTCCGGTGTATTTTAGGAG

AAGGCGTGGGTGAAGGCACGAGGCAGGCGGTGGTCGGCTTAAGGCCGGG

CAGGCAGGCAGGGGTTTTGGAAGCTCTACTTTCTTAATGTTCTTGATTATTC

TTCCTCCGAGACGTTTTCCAGGCGTCACCCCTAGGAACTCTCGTCCCAAAT

AAAATAAAAGATTTCAAAAATCTGTAAAAACTGAAAATACAATTGAGCGCATT

TACAAACACACCACATTTAAAATTAATTTCAGCGGTTGCTCCCTGTACTTCC

GAAAAAATGGGATTCAAAGACCAGCCGAGTTGTGCCGTAAAACAATAATGA

AAAGAAACCGACGCCCACGTGCCGAAGTACAATCTCCAACACCAGAAGAC

TCGAAACTATGTCATAATACTACTCCACTGCAAAATATCCCATCGCAACATT

TCTCTGGTACC

(Continued on next page)

Cell Reports 26, 313–321.e1–e7, January 8, 2019 e2

Continued

REAGENT or RESOURCE SOURCE IDENTIFIER

Synthetic oligo as 30 arm for knocking in mNeonGreen in end-1 gene This paper N/A

CTGCAGATTTATCTTATCTTTTTATGTGTATGCTCGTTTGTTCCTTTT

GATGATCTCTAAATGAATGTGACAGTTCTACTGATAAGCTCTTTTTG

TGTGCTACCAATCAGGGGAACTCTAACTAAAGTTTTCTGTCTCTGG

GGATTTCTCTAGTATAAAAACATTATTAAATTTAAAATTTCTAATCATT

CATCAAAAAAATTGATAAACTTGATAAAAATCTTTCAAGAAATCATTA

ACAAATTGTCAAGAAAGCAGTCATAAAATGTGAGAAAAAATAATGAT

AACATGAATTGCCTATGAATTCAGAAAAAAAGCGCTACGAGACCTAA

AATTCTACAAAAAATACAGAAACTTTTGAACCAGCGAGAATGGAGGG

AAAACCTACGGAAAGTGGGGGTGAATTGGCACCGGGAACATGAAA

AATTTGGAAAGTTTGGGAGAAAAATTGAACAAAACTCGATATTTCTTA

GAGCTCCTCGAGTTGAAAAGAGCCTCCTACACTGGGATGGGGTTGC

AGGATCGAACAGGTGAAGGGAAACACATAAAAAATTTTGTTGAGGAA

ATTATTGAATTAGAGTTTTGATTACCCGATATAAATACAGAAAAATTGA

GTATTTTCAAGATGACAAAAGAAGGTGAAAATTATAAGCCCGTTTTAC

AAGTTACGAATAACAAAAAGAGGGTCCTAGTTATTATTCATCGTCCGG

TATGTAGAACAGGGGAACCGGACGTTTTCCAAGGCGTAGCCATGTTA

GACAAGGCGCAGATATAGGTGATGCTGATGCAGAAAAACGATTGATT

TTTGGATGTTCGGATAGTCCACGGGCATCAATGATTTCGATGGAGAC

GCCACTGATTCGGCAGAGAATGTTCCAGAGAATCTGAAATGAGAATA

TGGTAAACTATGAAGCAAGTTCCATTTTGAAATTTTGTTTCTAGTTAAA

GGCTTGTATAGACTGAAACTGACACAAAACGGCTGAATTTTGTAATG

ACTAGT

sgRNA targeting end-1 gene for mNeonGreen knockin end-1-CKI-

SgRNA1: TGTGATGGAATGTTCTGTAG end-1-CKI-SgRNA2: GAG

ATCATCAAAAGGAACAA

This paper N/A

Primers used to detect mNeonGreen knockin end-1-CKI-F: 50- CCAGCACAACAATCGACACC �30 end-1-CKI-R: 50- GTGACAAGAGCA

CATTGGCAA �30

This paper N/A

Recombinant DNA

Plasmid: pDD162 Addgene # 47549

Plasmid: pRF4 (Mello et al., 1991) N/A

Plasmid: pPD95.77 Addgene # 1495

Plasmid: end-3sg This paper N/A

Plasmid: end-1sg This paper N/A

Plasmid: end-1HR This paper N/A

Software and Algorithms

StarryNite https://github.com/zhirongbaolab/

StarryNite (Santella et al., 2010, 2014)

N//A

AceTree https://github.com/zhirongbaolab/

AceTree (Katzman et al., 2018)

N/A

R https://www.r-project.org/ Version 3.2.5

Python https://www.python.org/ Version 3.6.4

MeV (Multiple Experiment Viewer) http://mev.tm4.org/ Version 4.9.0

CONTACT FOR REAGENT AND RESOURCE SHARING

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Zhuo Du

(zdu@genetics.ac.cn).

EXPERIMENTAL MODEL AND SUBJECT DETAILS

C. elegans Strains and CultureSome C. elegans strains used in this study came from the Caenorhabditis Genetics Center. All strains were cultured under standard

laboratory conditions on nematode growth media plates seeded with OP50 bacteria in incubators at a constant temperature. Most

e3 Cell Reports 26, 313–321.e1–e7, January 8, 2019

strains were maintained at 21�C; temperature-sensitive mutants were maintained at 15�C and switched to 25�C for at least eight

hours before the experiment.

METHOD DETAILS

Embryo MountingEmbryo collection and mounting were performed as previously described (Du et al., 2014, 2015). Briefly, 4-6 gravid adult worms with

one row of eggs were transferred into a droplet of M9 buffer on a Multitest slide (MP Biomedicals) and cut open under the dissecting

microscope (Nikon SMZ745). Two- to four-cell stage embryoswere transferred using an aspirator tube assembly (Sigma-Aldrich) into

a 2.5 mL droplet of egg buffer containing around thirty 20 mmpolystyrenemicrospheres (Polyscience) on a 24 X 60mmcoverslip (Fish-

erbrand). Embryo positions were adjusted using an eyelash to produce clusters of two or three embryos, after which an 18 X 18 mm

coverslip was laid on top of the droplet, followed by sealing withmelted Vaseline. The same procedure was used tomount embryos in

an uncompressedmanner, except that 45 mmpolystyrenemicrospheres (Polyscience) were used to ensure the spacing between two

coverslips was larger than the thickness of embryos (�30 mm).

3D Time-lapse Imaging3D time-lapse imaging was performed using spinning disk confocal microscopy (Revolution XD) with an Olympus IX73 inverted mi-

croscope, a Yokogawa CSU-X1 spinning-disk unit, an ASI PZ-2150-XYZ stage with Piezo-Z positioning, and an Andor iXon Ultra 897

EMCCD camera. Images were taken using MetaMorph acquisition software under 60X magnification (PLAPON 60XO N.A. 1.42

objective) with 30 Z slices (1 mm spacing) for six hours at 75 s intervals. The resulting voxel size at each time point was 0.22 X

0.22 X 1.00 mm (X, Y, Z). Imaging parameters (laser power and exposure time) were optimized tominimize photo-damage while main-

taining a high signal to noise ratio, and all wild-type embryos hatched after long-term imaging.

CRISPR/Cas9-based Gene Knockout and KnockinKnockout of end-3 gene: The end-1 and end-3 genes are tightly linked on chromosome V. To bypass the inherent difficulty of

crossing, we constructed the end-1;end-3 double mutant using CRISPR/Cas9-based gene editing to knockout the end-3 gene in

end-1 mutants. Two single-guide RNA (sgRNA) sequences targeting the N- and C-terminals of end-3 were designed using the

CRISPR Design Tool (http://zlab.bio/guide-design-resources) and cloned into a pDD162 plasmid by circular PCR using a previ-

ously-published protocol (Paix et al., 2014) to construct the Cas9-sgRNA plasmid (end-3sg). This plasmid and a reporter plasmid

pRF4 (inducing a dominant roller phenotype) were purified with the PureLink PCR Micro kit (Invitrogen) and injected into the gonads

of worms at a final concentration of 50 ng/mL for each plasmid. Single-wormPCRwas performed on individual F1 animals with a roller

phenotype to detect gene knockout events. Progeny of worms with a large gene deletion were selected to obtain homozygous

knockout animals. We obtained a knockout strain with a 931 bp deletion (chr V: 14,014,806-14,015,796, WS235/cel11) and a

9 bp insertion (TTTCAATGA) in the end-3 gene.

Knockin of mNeonGreen in the end-1 gene: To visualize endogenous END-1 protein expression, we fused a codon-optimized cod-

ing sequence of mNeonGreen (Hostettler et al., 2017) to the C-terminal of the end-1 gene. The CRISPR/Cas9 protocol used to

perform knockin was the same as that for knockout except an additional plasmid with a repair template (end-1HR) was constructed

and used. The repair template sequence included a �1000 bp arm homologous to the end-1 gene sequence on both ends of the

mNeonGreen coding sequence, and was cloned into the pPD95.77 plasmid between the AgeI and PstI restriction sites. Animals

with mNeonGreen knockin were detected by PCR and confirmed by sequencing.

QUANTIFICATION AND STATISTICAL ANALYSIS

Cell Identification, Tracing, and Cell Lineage Construction4D images were processed with the StarryNite software (Bao et al., 2006; Santella et al., 2010, 2014) for automated cell identification,

tracing, and cell lineage reconstruction, and the results were visualized andmanually curated using AceTree software (Katzman et al.,

2018). Detailed procedures for cell identification, lineage tracing, error detection, error correction, and quality control were described

in a previous paper (Du et al., 2015). First, a nucleus-localized, ubiquitously expressed mCherry fluorescence signal was used to

segment nuclei at each time point using a hybrid blob-detection algorithm (Santella et al., 2010). Second, nuclei were traced over

time using a semi-local neighborhood-based framework (Santella et al., 2014). Third, the raw results from automated nuclei identi-

fication and tracing were subjected to systematic manual inspection and error correction using AceTree (Katzman et al., 2018).

AceTree allows visualization of cell lineages and provides a convenient interface for linking lineage tracing results to the original

4D images for efficient detection and correction of errors. Finally, a unique name was assigned to each identified cell according

to the Sulston nomenclature (Sulston et al., 1983). Briefly, cell identity (ABa, ABp, EMS, and P2) and body axis were first determined

based on the characteristic arrangement of cells in four-cell stage embryo. Then, names were assigned to daughter cells based on

the mother cell name and cell position relative to the body axis following cell division. For example, ABal is the daughter cell of ABa

that is located on the left side following ABa cell division; ABala is the daughter cell of ABal that is located on the anterior side following

ABal cell division. The names of most cells follow this general rule, excepting several early progenitor cells designated with special

Cell Reports 26, 313–321.e1–e7, January 8, 2019 e4

names (e.g., MS, E, C, D). More information on the naming scheme is given elsewhere (Bao et al., 2006; Santella et al., 2010; Sulston

et al., 1983). After lineage tracing, cells present at each time point and associated phenotypic information including cell name; nu-

cleus diameter; X, Y, Z positions; and fluorescence intensity were systematically digitized for further analysis.

Measurement of PV in Wild-type EmbryosDefinition of relative cell position: We used the 3D coordinates of the centroid of the nucleus as a proxy for cell position and defined

relative cell position as a vector of geometrical distance from the centroid of the nucleus of the cell of interest (target cell) to that of all

other cells (reference cells) present at the same time point (Figure 1D). Because the imaging resolutions are different for XY and Z

(pixel size 0.22 versus 1), the value of Z wasmultiplied by 4.5 tomatch the XY scale when calculating geometrical distance. To control

for embryo size variation, distances were normalized by the mean distance of all pairwise cell-cell distances in the embryo.

Comparison of relative cell position: PV was measured by comparing distance vectors of equivalent cells between wild-type em-

bryos. The root-mean-square deviation (RMSD) values between two distance vectors for all pairwise comparisons were measured

and then averaged to quantify PV.

Time alignment: Due to variability in cell cycle length, the present time for a specific cell in different embryos could differ. To ensure

accurate comparisons of cell position at a given time, wemapped the absolute time of each cell to a relative scale and calculated PV

at a comparable relative time across all wild-type embryos (Figure S1C). For a given target Cell-X, we first screened the same cell in all

28 embryos to identify the cell with the longest present time and used it as the reference time point for equivalent cells in other em-

bryos. Next, all time points for Cell-X in other embryos were scaled and mapped onto the closest reference time point. The median

value of present time for all cells was 32; our definition of comparable time thus on average has a difference less than 1.56% of the

total cell cycle length (0.5*1/32). Whenever the PV of a cell was used for analysis, its PV at all time points was averaged.

Selection of reference cells: Variations in cell cycle length also result in a differential composition of reference cells at a given time

point in different embryos. PV calculations for pairwise comparisons only included reference cells that were present in both embryos.

Assessment of the Accuracy and Reliability of PV MeasurementWe validated the accuracy and reliability of PVmeasurements in several ways. First, we confirmed that cell position at each time point

was accurately determined. Cell identification and tracing were performed automatically using a highly effective algorithm with an

accuracy of over 99% (Du et al., 2015). The raw results were further systematically inspected and corrected for residual errors. After

the manual curation of cell identification and tracing results in the 28 embryos used for PV measurements, 125 cells were randomly

selected to evaluate the correctness of cell position determinations. Each cell was manually traced for ten consecutive time points,

totaling 1250 instances, to whether cell position assignments were consistent with the corresponding 3D images. We found all but

one instance was correct, leading to an evaluated accuracy of 0.999 (1249/1250).

Second, we validated the robustness of the relative method for defining cell position as a measure of PV. We applied this relative

approach, quantifying a cell’s position as the distance to reference cells (Figure 1D), in order to circumvent requirements for embryo

rotation and positional alignment, which could introduce bias. However, a caveat with this approach is that a cell’s position and PV

measurement could be significantly affected by those of its reference cells. We measured PV levels using subsets (10%, 20%, and

50%) of reference cells and showed that they were highly correlated with those obtained using all reference cells (Figure S1A), con-

firming that PV measurement is insensitive to the constituency of reference cells. We also validated that the PV levels of reference

cells have only minor influence on the PV levels of target cells. We estimated this by computationally altering a cell’s 3D position

to introduce in silico variability and then determined its impact on PV measurement. This manipulation caused significantly larger

changes in PVwhen the cell was treated as a target cell as compared to when it was treated as a reference cell (Figure S1B,Wilcoxon

signed-rank test, p = 7.38e-45). These results suggest that although the positions of other cells are used as references, the relative

method is most sensitive to the PV of the cell of interest.

Third, we determined the positions of cells across multiple time points and ensured that their PV levels were measured at compa-

rable times. We aligned cells by the relative time between embryos and compared cell positions at comparable relative times (Fig-

ure S1C). This treatment ensured that temporal measurements of PV were precise. However, a small-scale time misalignment was

inevitable after alignment, due to temporal resolution of the imaging. We modeled the influence of this artifact by artificially misalign-

ing cell times for one time interval (75 s, equivalent to 3.4% of the average cell cycle length). We found PV levels with or without this

misalignment were highly correlated (Pearson correlation, R = 0.985, p < 2.22e-308), with an average coefficient of variation of 0.01.

This excludes the possibility that small-scale time misalignment affects PV measurement.

Fourth, we confirmed that the number of embryos analyzed was sufficient to represent PV.We simulated the effects of sample size

on PV levels by sampling smaller numbers of embryos (n = 5, 10, 15, and 20) and comparing to the total 28 embryos.We found that the

coefficient of variation of PV levels was less than 0.05 at an embryo number of ten (Figure S1D). Moreover, we found that PV levels in

embryo sampling replicates were highly correlated (Figure S1E), suggesting PV is a stable trait of cells.

Finally, we confirmed that the mounting method used for live imaging does not dramatically affect PV. Two types of mounting

methods are widely used for C. elegans live imaging, differentiated by whether the embryos are compressed (Bao et al., 2006) or

uncompressed (Jelier et al., 2016) during imaging. We found that PV levels for the two methods were highly correlated (Figure S1F,

Spearman rank correlation, r = 0.79, p = 1.38e-155). However, PV levels for compressed embryos were significantly lower than that

of uncompressed (Wilcoxon Signed-Rank test, p = 7.27e-96), suggesting cell position is more reproducible when the embryos are

e5 Cell Reports 26, 313–321.e1–e7, January 8, 2019

compressed. This is consistent with a previous observation that slight compression allows the embryo to rotate to a stereotypical

orientation (Sulston et al., 1983).

Collectively, the above analysis of the reproducibility, sensitivity, and robustness of PV measurement systematically validated the

accuracy and reliability of ourmeasurement at high temporal resolution. In this study, we used PV data from compressed embryos for

all analyses.

Measurement of PV of Mutant Embryosglp-1(e2142) is a temperature sensitive allele; these worms were maintained at 15�C and switched to 25�C for eight hours before

collecting embryos for imaging. skn-1(zu67) is a maternal effect recessive allele causing 100% embryonic lethality (Bowerman

et al., 1992). We balanced skn-1(zu67) using an ubiquitously expressed GFP transgene (oxTi994, Integration site: IV, 2.45) inserted

close to the skn-1 locus (Frøkjær-Jensen et al., 2014) and collected embryos from skn-1(zu67) homozygote worms (determined by

animals without GFP expression) for imaging. Similarly, because end-1&ric-7(ok558); end-3(dev85) mutants have 100% embryonic

or larval lethality, that double mutant was balanced by a ubiquitously expressed GFP transgene (oxTi991, Integration site: V, 5.59)

inserted near the end-1 and end-3 loci (Frøkjær-Jensen et al., 2014). Embryos from heterozygote adult worms were collected for im-

aging, and embryos that were homozygous for end-1 and end-3 (determined by embryos without GFP expression) were used for

further analysis.

Measurement of PV in mutant embryos was performed as in the wild-type except that for certain mutants, the reference cells used

were carefully selected. Some mutations induced widespread, incompletely penetrant changes in cell lineage identity which could

cause ambiguity in PVmeasurement because the positions of cells with and without the phenotypes are compared. For these cases,

we only used reference cells that were deemed to be unaffected. For glp-1(e2142) mutants, cells from the ABala sub-lineage were

used as reference cells to calculate and compare PV in both mutants and wild-type embryos (Figure 2C). For skn-1(zu67) mutants,

cells from the C sub-lineage were used as reference cells (Figure 2D). For end-1;end-3 double mutants, all cells except gut cells were

used as reference cells (Figure 3E).

Comparison of PV between Different Cell TypesDue to technical difficulties, we did not trace the cell positions through the terminal stage of embryogenesis, but only to the 350-cell

stage; nevertheless, by that point, 28% (100) of cells had exited embryonic mitosis and differentiated into specific cell types

(Table S2). We focused on these cells and compared PV levels between cell types after controlling for lineage identity.

Identification of Cells with Similar LocalizationHierarchical clustering was applied to the pairwise cell-cell distance matrix for all leaf cells to identify clusters of cells with similar

localization using MeV software with the following parameters: Euclidian distance, average linkage clustering, optimize leaf order.

A tree cutoff of five was used to identify cell clusters.

Confounder AnalysisBecause both lineage identity and localization were associated with PV and lineage identity could regulate localization, we examined

the influence of each factor while controlling for the influence of the other.We first grouped cells by lineage (Figure 2B) and normalized

the PV level of each cell according to its localization cluster (Figure 2G). If cell localization was a confounder of lineage identity, we

expected normalization to extinguish the characteristic pattern of PV in each lineage group. On the contrary, we found that the char-

acteristic levels of PV were largely preserved (Figure S2D), and PV values before and after normalization were significantly correlated

(Pearson correlation, R = 0.613, p = 3.70e-38). These results suggest that PV is associated with cell lineage even after controlling

localization. Similarly, we found continued association of PV with cell localization after controlling for the influence of lineage identity

(Figure S2E, Pearson correlation, R = 0.700, p = 1.25e-53).

Quantification of Fate Determinant Expression in Single CellsDual color fluorescent reporter strains were constructed to profile single-cell expression of fate determinants during embryogenesis.

In each strain, nucleus-localized, ubiquitously expressed mCherry was used to identify and trace nuclei, and GFP fused with the pro-

tein sequence of the fate determinant was used to simultaneously quantify protein expression level. Live imaging, cell identification

and tracing, cell lineage construction, and quality control were performed as for PV measurement. All analyzed fate determinants

were transcription factors with nuclear localization; this facilitates straightforward quantification of GFP intensity once the nuclei

were identified and traced. Quantification of GFP expression within the nucleus was performed using a previously reported method

(Murray et al., 2008). Briefly, GFP intensity was first measured within each nucleus at each time point, and then with the local back-

ground intensity subtracted. Background intensity was estimated by measuring the GFP intensity within 1.2 r to 2 r from the nuclear

centroid for each nucleus (r being the radius of the nucleus). Empirically, an intensity of more than 2,000 A.U. is considered expres-

sion. Tree visualizations of single-cell protein expressions were generated by AceTree.

Cell Reports 26, 313–321.e1–e7, January 8, 2019 e6

SimulationsSimulation of the sensitivity of PV measurement: To introduce variability, we randomly changed the 3D position of a cell in one of the

28 embryos by less than 5% of the embryo’s long axis. After 100 simulations, we then calculated the PV levels of manipulated cells

and other unaffected cells to determine the influence of this manipulation (Figure S1B).

Simulation of PV dynamics: To model PV dynamics over time (Figure 3B), we simulated the cell positions for multiple cells (n = 100)

in 28 embryos over 150 time points. The initial cell positions were identical (PV = 0), and for each time interval we allowed all cells in

each embryo to move randomly over a given distance but within the predefined embryonic space. PV levels at each time point were

calculated as for real data. Three distances were modeled, 0.5D, 1D, and 2D, where D is the median distance of cell displacement

over one time interval in real embryos.

Simulation of the expected PV level: To model expected PV levels, we randomized the direction of cell displacement but kept its

distance identical to that of real embryos (Figure 4B and 4C). To simulate the expected PV level for a cell, we first determined in the

real embryo the displacement distance of the cell from its birth time to its division time (Dc). We then allowed the cell at Dc to move in a

random direction from the position of its birth time. This simulation was applied to all cells in all embryos. The same strategy was used

for simulating the expected PV level of a time interval, except the cell displacement distance within a time interval (Dt) was used.

StatisticsWilcoxon Signed-Rank test was performed to identify PV-changing points (Figure 2B), to compare the influence of cell position

changes on the PV levels of position-switched and unaffected cells (Figure S1B), to compare observed PV level to that of expected

(Figure 4B), and to compare PV levels (Figure 4E-G) and velocity of cell displacement (Figure 4H-J) between wild-type and mutant

embryos in equivalent cells. General linear regression was used to assess the contribution of cell lineage identity and cell localization

to the variance in PV levels. Mann-Whitney U test was performed to evaluate the differences in sub-lineage PV levels between wild-

type and mutant embryos (Figure 2C and 2D), to evaluate the differences in the divergency of PV levels between real and mock cell

clusters with similar localization (Figure S2C), between cell pairs from real and mock contact cells (Figure S2F), between real and

mock left-right symmetric pairs (Figure S2H), and between real and mock symmetric cells at different developmental stages (Fig-

ure S2I). Mood’s Median Test was used to assess the differences in PV levels between different cell types (Figure 2E). Pearson cor-

relation was performed to calculate the correlation coefficient of PV between contact cells (Figure 2H), between left-right symmetric

cells (Figure 2K), to calculate the correlation of temporal dynamics of PV between cell tracks (Figure S3A). Spearman rank correlation

was performed to calculate the correlation coefficient of PV level measured using all reference cells to that using a subset of reference

cells (Figure S1A), and to calculate correlation coefficient of PV levels between compressed and uncompressed embryos (Fig-

ure S1F). Sign test was performed to compare PV levels of ingressing or egressing cells to that of control cells (Figure S3B). All sta-

tistical methods and corresponding p values were described in the main text, figures or figure legends. Statistical analyses were per-

formed using Python and Minitab.

DATA AND SOFTWARE AVAILABILITY

All data used to analyze PV including 4D images, cell lineages and positions, and PV results are available at http://dulab.genetics.ac.

cn/data_sharing.html and https://usegalaxy.org/u/zhuodu/h/dataset-cell-position-variability.

e7 Cell Reports 26, 313–321.e1–e7, January 8, 2019

Cell Reports, Volume 26

Supplemental Information

Systems Properties and Spatiotemporal

Regulation of Cell Position Variability

during Embryogenesis

Xiaoyu Li, Zhiguang Zhao, Weina Xu, Rong Fan, Long Xiao, Xuehua Ma, and Zhuo Du

  

Figure S1. Quality Assessment of Positional Variability Measurement. Related to Figure 1.

(A) Influence of the composition of reference cells on PV levels. Scatter plots show the correlation of PV level calculated using all reference cells to that using a subset of reference cells.

(B) Assessment of the sensitivity of PV measurement. Box plot compares the influence of cell position changes on the PV levels of position-switched and unaffected cells.

(C) Strategy for time alignment. Details are described in the STAR Methods.

(D) Influence of embryo number on PV levels. Box plot shows the coefficient of variation of PV levels for calculations using a small number of embryos and that using all embryos.

(E) Stability of PV levels. Box plot shows the correlation coefficient between PV levels in replicate samples using different numbers of embryos.

(F) Influence of embryo mounting methods on PV levels. Scatter plot shows correlation of PV measurements between compressed (X-axis, 28 embryos) and uncompressed embryos (Y-axis, 10 embryos).

  

  

Figure S2. Cell Position Variability is Associated with Developmental Properties of Cells. Related to Figure 2.

(A) Switch of ABara sub-lineage identity to that of ABala in glp-1. The color coded tree shows the traced cell lineage and cellular expression pattern of an ABara-sub-lineage marker (PHA-4::GFP). Arrows indicate programmed cell death.

(B) Switch of MS sub-lineage identity to that of C in skn-1.

(C) Cells with similar localization exhibit smaller divergency in PV level. Box plot shows the divergency of PV levels (measured as interquartile range/median) in real and mock localization clusters.

(D) Influence of cell lineage identity on PV after considering localization. Box plot shows unnormalized (blue) and normalized (red) relative PV levels for cells from different lineage groups (Figure 2B). Relative PV is normalized via dividing by the average PV of its cell lineage group.

(E) Influence of cell localization on PV after considering cell lineage identity. Box plot shows unnormalized (blue) and normalized (red) relative PV levels of cells from different localization clusters (Figure 2G). Relative PV is calculated by dividing the PV of each cell by the mean PV level. Cell PV was normalized via dividing by the average PV of its localization group.

(F) Contact cells exhibit smaller divergency in PV levels. Box plot shows the coefficient of variation of PV levels between cell pairs from real and mock contact cells.

(G) Comparison of divergency of PV levels for cells in each cell cluster with (red) or without (blue) cell-cell contact. Divergency is measured as the coefficient of variation.

(H) Left-right symmetric cells exhibit smaller variation in PV levels. Box plot shows the coefficient of variation of PV levels in real and mock left-right symmetric pairs.

(I) Correlation of PV between left-right symmetric cells is associated with the establishment of symmetry. Box plot compares the deviation of PV levels between real and mock symmetric cells at different developmental stages.

  

  

Figure S3. Temporal Dynamics of Cell Position Variability during Embryogenesis. Related to Figure 3.

(A) PV level dynamics in each cell track during embryogenesis. Top panel illustrates the definition of cell track. Tree structure represents a sub-lineage with 16 leaf cells generated from an ancestral cell; the blue line represents a cell track that generates a leaf cell. Bottom panel shows dynamics of relative PV levels in all cell tracks (n=361) contributing to a set of 361 leaf cells at different stages of embryogenesis. Relative PV is calculated by dividing cell PV level at each time point by the mean PV of its cell track.

(B) Cells undergoing ingression and egression exhibit PV levels similar to other cells. Scatter plots compare PV levels of ingressing or egressing cells (Y-axis) to those of corresponding control cells (X-axis). Diagonal lines represent equal PV, and red dots indicate cells showing significant difference (Sign test, Benjamini–Hochberg corrected p < 0.01). Ingressing cells (n=66 and n=50, respectively) were defined by Harrell (Harrell and Goldstein, 2011) and by Jelier (Jelier et al., 2016); egressing cells (n=18) were defined by Jelier (Jelier et al., 2016). For each cell, other cells belonging to the same sub-lineage (Figure 2B) and at the same developmental stage are used as controls.

(C) Temporal dynamics of PV in ingressing and egressing cells are similar to other cells. Figure organization is as in (A); red lines indicate PV dynamics of corresponding ingressing or egressing cells.

(D). Changes in PV levels during the cell cycles of individual cells with different cell fates. Heatmap shows changes in PV level during different relative times within a cell cycle for those cells showing PV down-regulation. Color gradient from blue to red represents PV levels from low to high.

(E) Dynamics of global PV levels of non-E-sub-lineage cells in end-1;end-3 embryos. Characteristic developmental stages (based on the number of cells in the embryo) are indicated on the X-axis.

  

Figure S4. Dynamics of Positional Variability in Cell Adhesion and Gap Junction Mutant Embryos. Related to Figure 4.

(A) Dynamics of global PV in hmp-2, jac-1, and inx-2 embryos. Box plots show average PV across all cells at each developmental time during embryogenesis.

(B) Change of PV over time in hmp-2 and inx-2 embryos relative to wild type. For ease of comparison, a cell’s average PV level was used to calculate its relative PV (measured as PV of mutant/PV of WT for equivalent cells) at different times in embryogenesis.