HLH-2/E2A Expression Links Stochastic and Deterministic Elements of a Cell Fate Decision during C. elegans Gonadogenesis

SUMMARY

Stochastic mechanisms diversify cell fate in organisms ranging from bacteria to humans [1–4]. In the anchor cell/ventral uterine precursor cell (AC/VU) fate decision during C. elegans gonadogenesis, two ‘‘a cells,’’ each with equal potential to be an AC or a VU, interact via LIN-12/Notch and its ligand LAG-2/DSL [5, 6]. This LIN-12/Notch-mediated interaction engages feedback mechanisms that amplify a stochastic initial difference between the two a cells, ensuring that the cell with higher lin-12 activity becomes the VU while the other becomes the AC [7–9]. The initial difference between the α cells was originally envisaged as a random imbalance from ‘‘noise’’ in lin-12 expression/activity [6]. However, subsequent evidence that the relative birth order of the α cells biases their fates suggested other factors may be operating [7]. Here, we investigate the nature of the initial difference using high-throughput lineage analysis [10]; GFP-tagged endogenous LIN-12, LAG-2, and HLH-2, a conserved transcription factor that orchestrates AC/VU development [7, 11]; and tissue-specific hlh-2 null alleles. We identify two stochastic elements: relative birth order, which largely originates at the beginning of the somatic gonad lineage three generations earlier, and onset of HLH-2 expression, such that the α cell whose parent expressed HLH-2 first is biased toward the VU fate. We find that these elements are interrelated, because initiation of HLH-2 expression is linked to the birth of the parent cell. Finally, we provide a potential deterministic mechanism for the HLH-2 expression bias by showing that hlh-2 is required for LIN-12 expression in the α cells.

Graphical Abstract

In Brief

Two bipotential α cells interact via LIN-12/Notch and its ligand LAG-2/DSL, engaging feedback mechanisms that amplify a stochastic initial difference between them. Attner et al. identify two interrelated stochastic elements and provide a potential deterministic mechanism by showing that hlh-2 is required for LIN-12 expression in the α cells.

RESULTS AND DISCUSSION

Large-Scale Lineage Analysis Refines the Relationship between Birth Order and Cell Fate

The anchor cell (AC) is a unique cell in the proximal region of the developing gonad that serves as the signaling nexus for uterine and vulval patterning and in connecting the uterus and the vulva [12, 13]; its correct specification is therefore critical for maximizing reproductive success. Initially, four cells in the developing somatic primordium have the potential to be the AC (Figure 1A). As described further below, the conserved transcription factor HLH-2 is required to endow these cells with AC potential. The ‘‘β cells’’ soon lose AC potential and always become ventral uterine precursor cells (VUs); the two α cells maintain AC potential and interact via LIN-12/Notch to resolve which will become the AC and which will become another VU [6, 14–16].

Figure 1. Long-Term Imaging Refines the Relationship between Birth Order and Cell Fate.

(A) Early gonadogenesis. The somatic gonad precursor cells Z1 and Z4 divide in the first larval stage (L1); three rounds of division (Div 1–3) generate the twelve cells of the somatic gonad primordium, which forms in the L2 stage. Z1.ppp and Z4.aaa, the ‘‘α cells,’’ and their sisters, the ‘‘β cells,’’ have anchor cell (AC) potential. The β cells lose this potential quickly and invariably become ventral uterine precursor cells (VUs). The α cells undergo the AC/VU decision in the L2 stage. As described herein, the parents of the α cells, Z1.pp and Z4.aa, are born at different times. Depending on the time interval, the AC/VU decision is biased or stochastic with respect to birth order. Shown here, Z1.pp divides before Z4.aa and Z4.aaa becomes the AC, but it is equally likely that Z4.aa will divide first and Z1.ppp will become the AC.

(B) Representative maximum-z projections of deconvolved z stacks of mCherry::H2B-marked somatic gonad nuclei. The α cells Z1.ppp and Z4.aaa, generated after Division 3, have not yet undergone the AC/VU decision (Videos S1 and S2). Scale bar, 5 µm.

(C) Relationship between birth order and cell fate from long-term imaging at 25°C. The time difference between the births of Z1.ppp and Z4.aaa (minutes) and fate of the first-born cell (blue square, VU; red triangle, AC) are shown for 64 individuals. We observed no significant preference for either Z1.ppp or Z4.aaa to be the first-born cell (prob[Z1.ppp first-born]: p = 0.48 ± 0.1). At birth-order time differences ≤24 min, fate outcomes were statistically indistinguishable from chance (prob[first-born cell AC | Δt < 24 min]: p = 0.35 ± 0.18). At birth-order time differences >24 min, the first-born cell is highly biased to become the VU (prob[cell first born j cell fate AC]: p = 0.09 ± 0.26). Confidence intervals for probabilities are 95% based on z value of a normal distribution.

(D) The relative division order of Z1 and Z4 is generally maintained throughout their lineages. Each column represents one animal. Each row represents the division order in the lineage. Animals in which all three divisions were captured in the imaging experiment with birth-order differences <105 min are shown (n = 42). The marked interval encompasses individuals with ≤24 min (short) differences in birth order at Div3. Columns are sorted by the relative birth order of the α cells. Top: for each column, orange, firstborn descended from Z1; gray, firstborn from Z4; white, no difference at that time point. Bottom: lighter shading indicates less time between division of corresponding cells from the Z1 and Z4 lineages; switching of relative order only occurred when birth-order differences were short. Asterisks, the first-born became the AC.

See also Figure S1 and Videos S1 and S2.

The α and β cells are descended from two progenitors, Z1 and Z4, which divide during the first larval (L1) stage; their 12 descendants form the somatic primordium in the early L2 stage [17] (Figure 1A). The 12-cell set is completed after Z1.pp and Z4.aa, also referred to herein as ‘‘the parents,’’ each divide to produce an α and a β cell. Although in a population of worms, the AC/VU decision appears stochastic—in half of the individuals, the α cell Z1.ppp becomes the AC, whereas in the other half, the α cell Z4.aaa becomes the AC [17]—traditional lineage analysis of thirteen individuals suggested that the first-born α cell is biased to become a VU [7].

To investigate the origins of this bias, we first performed high-throughput lineage analysis in a microfluidic device [10]. We labeled all somatic gonad cells with mCherry::H2B (STAR Methods) and gathered 70 lineages at 20°C and 64 lineages at 25°C (Figure 1B; Videos S1 and S2). This analysis revealed that (1) a bias is only evident at relatively longer differences in time of birth (>30 min at 20°C and >24 min at 25°C), whereas for smaller time differences, the AC/VU decision is random with respect to birth order (Figures 1C, S1A, and S1B), and (2) the relative birth order of the α cells is highly correlated with the relative order of division of Z1 and Z4 at the top of the lineage tree (Figures 1D and S1C), i.e., the relative birth order set at the top of the lineage tree is generally maintained throughout the lineages. This high correlation arises because (1) Z1 and Z4 in a given animal divide, on average, relatively far apart in time (~35 min, 25°C) and (2) subsequent cell-cycle durations between corresponding cells are strongly correlated (e.g., Z1.p versus Z4.a, r = 0.89; p < 2.81e—12; average cell-cycle duration difference, 10.8 min; Figure S1D). As discussed further below, the relative division order of Z1 and Z4 is one stochastic element impacting the AC/VU decision.

To gain insight into the molecular events impacting, and impacted by, birth order, we analyzed the temporal patterns of HLH-2, LIN-12, and LAG-2 expression. Importantly, we visualized fluorescently tagged endogenous proteins in live worms, so as to reflect the integration of all modes of regulation that control protein abundance at each developmental stage, avoid potential transgene artifacts, and obviate difficulties of spatiotemporal resolution and cell identification inherent in methods requiring fixed specimens.

Long-Term Imaging of GFP::HLH-2 Reveals that the Relative Timing of GFP::HLH-2 Onset in the Parents Predicts Subsequent α Cell Fate

hlh-2 encodes the sole C. elegans E transcription factor [18] and endows the α and β cells with the potential to be an AC; if this ‘‘proAC’’ role is compromised, hermaphrodites do not have an AC, and both α cells become VUs ([11] (considered further below). For this proAC role, as in its later roles in AC specification and function, HLH-2 functions as a homodimer [8]. We used CRISPR/Cas9 to create hlh-2(ar623), which encodes a fully functional GFP-tagged form of HLH-2 [GFP::HLH-2] (STAR Methods). We imaged gonadogenesis in the microfluidic device, collecting most data at 16 min intervals to minimize phototoxicity from repeated imaging (STAR Methods). GFP::HLH-2 was visible in both parents and in the α and β cells before becoming restricted to the presumptive AC (Figure S2A), as previously described for endogenous, untagged HLH-2 using antibody staining [7, 11, 18] and for transgene-driven GFP::HLH-2 [8]. Visualizing expression of GFP::HLH-2 over time revealed two remarkable and unexpected features.

First, GFP::HLH-2 appeared in each parent (Z1.pp or Z4.aa) in a narrow time window, 103 ± 18 min after its birth, regardless of how far apart the two parents had been born, the length of their cell cycles, or the eventual fate of its daughter α cell (Figure 2A; STAR Methods). Within a given animal, the parents were born at variable times with respect to each other, displaying a relatively large range in the difference in time of birth (47 min on average) (Figure S2E), but the cell-cycle lengths of Z1.pp and Z4.aa in an individual differed by an average of only 17 min (Figure 2A). The onset of HLH-2 expression within a small window after the birth of a parent cell could reflect a mechanistic link to the parent’s cell-cycle progression [19] or to a distinct timing mechanism initiated with the birth of the parent (e.g., [20]). This relatively precise onset after birth is observed in each parent even when the two parents are born at greatly different times (Figure S2E), suggesting that HLH-2 expression is not directly triggered by a pulse of an external hormonal signal (e.g., [21–23]).

Figure 2. The Relative Timing of GFP::HLH-2 Onset in the Parents Predicts Subsequent α Cell Fate.

(A) GFP::HLH-2 expression in all 34 individuals in which the lineage was captured from the birth of the parents, Z1.pp and Z4.aa, through specification of α cell fate. GS9062 contains hlh-2(ar623) [GFP::HLH-2] and gonad cell marker transgenes [ckb-3p::mCherry::H2B] (STAR Methods). For this strain and imaging condition, the time threshold for when the AC/VU decision displayed the birth order bias is >40 min (STAR Methods). Bars representing parent cells are paired per animal; green indicates GFP::HLH-2 expression in the first parent to divide, and gold indicates expression for the second parent. If both parent cells in an individual divided at the same time, both bars are green. The birth of each parent is set to time = 0, highlighting the narrow time window of GFP::HLH-2 onset; non-normalized data are in Figure S2E. Asterisks, individuals in which the first-born α cell became an AC.

(B–D) GFP::HLH-2 expression in all 36 individuals in which the onset of GFP::HLH-2 in one parent before another was captured. Patterns and quantification of GFP::HLH-2 fluorescence intensity (STAR Methods) of representative individuals for each case are shown (in arbitrary units, a.u.). Note that GFP::HLH-2 becomes restricted to the specified AC. In an additional 6 individuals, GFP::HLH-2 appeared in the same time point (see text).

See also Figures S2 and S3.

Second, the relative timing of GFP::HLH-2 appearance in the parents is remarkably predictive of the outcome of the AC/VU decision. Because onset of GFP::HLH-2 is linked to the birth of the parent, and parent cells are typically born at different times, GFP::HLH-2 usually appears in one of the parents before the other (n = 36/42 individuals) (Figures 2 and S2). In toto, 35/36 individuals displayed a GFP::HLH-2 onset bias, such that GFP::HLH-2 appeared in one parent first and its α cell daughter became the VU.

We further subdivide these individuals into three cases (Figures 2B–2D). In Case 1, where the interval between the birth of the α cells is long, both the birth-order bias and the GFP:: HLH-2 onset bias were seen (n = 22/22). In Cases 2 and 3, the interval between the birth of the α cells is short, so the AC/VU decision appears random with respect to birth order; in these cases, the onset of GFP::HLH-2 expression in the parent highly predicted subsequent α cell fate (n = 13/14), even when the first-born α cell became the AC.

For the 35/36 individuals in which GFP::HLH-2 was expressed first in the parent of the VU, it appears that the stochastic event(s) resolved by the AC/VU decision does not occur in the α cells; instead, there is a deterministic element provided by hlh-2 activity. Furthermore, the different cases (Figure 2) revealed two stochastic events that occur prior to the birth of the α cells. The first stochastic event, evident from Case 1 individuals and the lineage analysis described above, is the relative timing of the Z1 and Z4 division (Div 1; Figure 1): HLH-2 expression in the parent cell is tied to its birth, and the relative birth order at each step is tied to the division time at the previous step. When the interval between the division of Z1 and Z4 is long, then one parent cell is likely to be born before the other and express GFP::HLH-2 first. The second stochastic event is evident from the individuals of Cases 2 and 3: there is a small window of variability in the onset of HLH-2 expression after the birth of the parents, so when the interval between the birth of the parent cells is short, the first parent to express GFP::HLH-2 is highly likely to give rise to the α cell that becomes the VU.

Finally, we note that in 6/42 individuals, GFP::HLH-2 onset in both parents was observed at the same time point, followed by correct resolution of the AC/VU decision. There may be a difference in HLH-2 level or activity between the parent cells that we cannot detect using our methods (Figures S2F and S2G; STAR Methods), or there may be other factors contributing to differences in the α cells, including ‘‘noise’’ in expression [24] of any key components in the parents or the α cells themselves. Nevertheless, our analysis indicates that the HLH-2 expression bias is strong, and in the next section, we provide a potential mechanistic explanation for it.

Tissue-Specific Knockout of hlh-2 in the Proximal Gonad Reveals a Link between HLH-2 and LIN-12 Expression

GFP::HLH-2 appears first in the parent of the α cell that becomes the VU and second in the parent of the AC. At first glance, this observation seems paradoxical because hlh-2 has been shown to endow α cells with the potential for AC fate and is not required for VU fate ([11]; Figure S3). However, once an α cell has the potential to become an AC, lin-12 activity is required for it to adopt the VU fate, so we wondered if the missing link might be a role for hlh-2 in promoting LIN-12 expression. We therefore tested the effect of removing hlh-2 activity on expression of endogenous LIN-12::GFP [lin-12(ar624)] (STAR Methods).

To do so, we generated hlh-2(Δprox) [hlh-2(ar614)] a proximal gonad-specific null allele (STAR Methods; Figures 3A and S3). In an hlh-2(+) background, LIN-12::GFP was not detected in the parents, but was present at the cell surface and internally in the α and β cells before they are specified (Figure 3A); LIN-12::GFP became brighter in the VUs as they were specified and was lost in the AC, consistent with described feedback mechanisms [7–9]. LIN-12::GFP expression in the α cells was abolished in hlh-2(Δprox), indicating that HLH-2 is required for LIN-12 expression in conjunction with endowing AC potential.

Figure 3. hlh-2 Is Required for LAG-2 and LIN-12 Expression in the α and β Cells.

For all cases, the fluorescent image (left) is also shown in black on white (right), including a schematic indicating cells expressing lag-2(ar635) [LAG-2::GFP] (red outline), lin-12(ar624) [LIN-12::GFP] (blue outline), or neither (black outline). All somatic gonad cells are marked by arTi145[ckb-3p::mCherry::H2B]. hlh-2(Δprox) = hlh-2(ar614) is a proximal gonad-specific null allele (see text). All photomicrographs are maximum-intensity z-projections to allow visualization of the α and β cells simultaneously. Yellow arrow indicates expression in an α or β cell. Scale bars, 5 μm.

(A) LIN-12::GFP is visible in the α and β cells in a wild-type background after Div3. In this individual, a dashed yellow arrow indicates that LIN-12::GFP is already lower in the presumptive AC compared to the presumptive VUs. LIN-12::GFP is visible in one or both α cells in hlh-2(+)(n = 11/14) but is never visible in α cells in hlh-2(Δprox) (n = 0/18). Before Div3, LIN-12::GFP was not visible in parent cells in hlh-2(+) (n = 0/6).

(B) Photomicrograph shows LAG-2::GFP in a parent, Z1.pp (open triangle), at a time when Z4.aa has already divided to produce an α and β pair (filled triangles). Before Div 3, LAG-2::GFP expression was evident in at least one parent cell (6/6).

(C) LAG-2::GFP is present in the α and β cells in all larvae scored after Div3, when the early Z1 and Z4 lineages are complete (n = 11/11), but not in the background of hlh-2(Δprox) at the same stage (n = 0/17). The somatic gonad is indicated by a dashed line, and the α and β cells are encircled with a solid line here and in (A).

See also Figure S3.

Previous work indicated that lag-2 is a direct transcriptional target of HLH-2 during the AC/VU decision and in the differentiated AC [7]. In an hlh-2(+) background, we observed endogenous LAG-2::GFP [lag-2(ar635)] (STAR Methods) in the parent cells as well as in the α and β cells (Figure 3). All LAG-2::GFP expression was lost in the proximal gonad in the hlh-2(Δprox) background, indicating that LAG-2 expression also occurs in conjunction with the endowment of AC potential by hlh-2. The presence of surface LAG-2::GFP suggests that they are poised to activate LIN-12 in neighboring cells.

A New Model for the AC/VU Decision: How a Stochastic Event Creates a Bias to Ensure a Reproducible and Robust Outcome

Our lineage analysis revealed that when the interval between the birth of the α cells is long, there is a birth-order bias: the first-born α cell is strongly biased toward the VU fate. However, when the interval between the birth of the α cells is short, the AC/VU decision appears random with respect to birth order. We also identified a more proximate cause of bias in α cell fate: an HLH-2 expression bias, such that the first parent (Z1.pp or Z4.aa) to express HLH-2 gives rise to the α cell that adopts the VU fate. Finally, we found that hlh-2 activity is required for initial LIN-12::GFP expression in cells that are also endowed with AC potential by hlh-2. These findings lead us to propose a new model for the AC/VU decision (Figure 4).

Figure 4. A New Model for the AC/VU Decision.

The apparently stochastic division order of the somatic progenitors Z1 or Z4 is generally propagated through their lineages. GFP::HLH-2 appears in each parent, Z1.pp or Z4.aa, within a narrow time window after its birth. When the interval between the parents’ births is long, the parent cell that is born first (here, Z1.pp) generally also expresses GFP::HLH-2 first. The parent that expresses GFP::HLH-2 first gives its α cell daughter the edge in activating LIN-12, which we propose is because hlh-2 is required for LIN-12 expression.

Initially, HLH-2 may be required for LIN-12 expression in conjunction with its proAC role, whereas later, activated LIN-12 maintains lin-12 transcription through autoregulation [9] while also promoting HLH-2 degradation [7, 8] in the presumptive VU. As they undergo the AC/VU decision, the presumptive AC expresses HLH-2 (green nucleus) and LAG-2 (red arrows), whereas presumptive VUs express only LIN-12 (blue outline); this reciprocal change in expression occurs prior to commitment [9]. For long birth-order differences, the first-born α cell can receive LAG-2 signal from either its sister β cell or the undivided parent of the other α cell (shown), whereas at short birth-order differences, the first-born α cell can receive LAG-2 signal from its sister β cell or the other α cell.

In this model, two stochastic elements affect the AC/VU decision prior to the birth of the α cells. First, relative birth order: a random initial difference in when the precursors Z1 and Z4 divide generally is maintained throughout the lineage, and is the basis of the birth-order bias seen at long birth time differences. Second, the onset of HLH-2 expression in the parent cells: the first parent that expresses HLH-2 gives rise to the α cell that becomes the VU, and is the basis of the ‘‘HLH-2 expression bias’’ that is evident when birth time differences are short. These two stochastic elements are interrelated: the onset of HLH-2 expression in each parent occurs within a narrow time window after its birth. At short birth time differences, the HLH-2 expression bias appears much more predictive of α cell fate than the relative birth order of the α cells themselves. How the timing of HLH-2 expression becomes determinative is suggested by our finding that HLH-2 is required for expression of LIN-12 in the α and β cells when they have AC potential. We envisage that earlier expression of HLH-2 in a parent might give its a daughter an edge in LIN-12 activation by providing a legacy of lin-12 transcript, or HLH-2 protein to promote initial lin-12 transcription or pioneer function in opening chromatin for lin-12 expression, directly or indirectly [25, 26].

Such an edge requires ligand to be available to activate LIN-12 in cells with AC potential. Indeed, we observed LAG-2 in cells that are in contact with the α cell: for longer birth time differences, the parent of the other α cell; for shorter birth time differences, the other α cell; and for both, its sister β cell. Over time, the initial edge is amplified by positive autoregulation of lin-12 transcription, which is required for VU fate and separable from initial lin-12 expression [9]. Such autoregulation will continue to promote lin-12 transcription after HLH-2 is degraded as part of the feedback mechanism to downregulate lag-2 expression in the presumptive VU [7, 8].

In a review of systems in which stochastic mechanisms diversify cell fates, Johnston and Desplan [1] point out that the nature of stochastic events has often remained elusive. They also draw a distinction between paradigms with variable outcomes, such as random expression of different Rhodopsin genes to diversify photoreceptors in Drosophila [1, 27–30], versus those with reproducible outcomes, such as the AC/VU decision, which always produces a single AC. Our findings suggest that even though the α cells are uncommitted when they are born and need to interact via LIN-12/Notch to resolve their fates, events in their ancestors have essentially already determined their fates. Perhaps for other paradigms with a reproducible outcome, what has been considered ‘‘noise’’ may just appear so because a deterministic element has not been identified.

STAR*METHODS

LEAD CONTACT AND MATERIALS AVAILABILITY

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Iva Greenwald (isg4@columbia.edu). Strains GS8513 and GS8949 will be available via the Caenorhabditis Genetics Center; all other strains will be available upon request.

EXPERIMENTAL MODEL AND SUBJECT DETAILS

C. elegans alleles and transgenes

See Key Resources Table for the full list of strains. Strains were maintained at 20°C, Experiments were performed at 25°C, unless otherwise indicated, because fluorescent proteins appeared brighter than at 20°C and the lineages progressed faster, making it easier to capture the progression of early gonadogenesis in long-term imaging experiments.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Chemicals, Peptides, and Recombinant Proteins
BD BactoAgar	Fisher	DF0140-07-4
Calcium chloride, anhydrous	VWR	97062-59
Cholesterol	MP	101382
Magnesium Sulfate, anhydrous	Avantor-Fisher	2506-01
Potassium phosphate, dibasic	Avantor-Fisher	4012-05
Tetramisole hydrochloride (levamisole)	Sigma-Aldrich	L9756-10G
Experimental Models: Organisms/Strains
GS7434: arIs51 IV	This paper	N/A
GS8513: arTi145 [ckb-3p::mCherry::his-58::unc-54 3ʹUTR] II	This paper	N/A
GS8592: arTi145 II, cross	This paper	N/A
GS8645: hlh-2(ar614) I	This paper	N/A
GS8675: ccIs4251 I; arTi145 II; arTi112 [ckb-3p::mCherry::his-58:: unc-54 3ʹUTR] V	This paper	N/A
GS8686: ccIs4251 I; arTi145 II; arTi112 V cross	This paper	N/A
GS8812: arTi145 II; arIs131; arTi112, cross	This paper	N/A
GS8949: hlh-2(ar623) I	This paper	N/A
GS9013: hlh-2(ar623ar629) I	This paper	N/A
GS9046: hlh-2(ar623) I; pha-1(e2123) III	This paper	N/A
GS9062: hlh-2(ar623) I; arTi145 II; arTi112 V	This paper	N/A
GS9127: hlh-2(ar623) I; arTi145 II	This paper	N/A
GS9128: hlh-2(ar623ar629) I; arTi145 II	This paper	N/A
GS9129: arTi145 II; lin-12(ar624) III	This paper	N/A
GS9130: hlh-2(ar614) I; arTi145 II; lin-12(ar624) III	This paper	N/A
GS9221: arTi145 II; lag-2(ar635) V	This paper	N/A
GS9223: hlh-2(ar614) I; arTi145 II; lag-2(ar635) V	This paper	N/A
GS9232: hlh-2(ar614) I; arIs51 IV	This paper	N/A
Software and Algorithms
MATLAB	MathWorks	R2015a
Micromanager	[31]	RRID: SCR_000415
ImageJ	[32]	https://imagej.nih.gov/ij/
Other
Resource website	This paper	https://github.com/wolfgangkeil/Attner_Keil_et_al_2019_code

Three hlh-2 alleles were generated using CRISPR/Cas9 as described in Method Details below. hlh-2(ar623) encodes GFP::HLH-2; the tag was added to the N terminus of HLH-2 as addition at the C terminus would potentially compromise its dimerization and function [8, 33]. hlh-2(ar614) contains a deletion in a regulatory element, hlh-2prox [8]. hlh-2(ar623ar629) is a derivative of hlh-2(ar623) in which the same deletion in hlh-2prox as in hlh-2(ar614) was generated using CRISPR/Cas9.

The alleles lag-2(ar635) [LAG-2::GFP], made by Catherine O’Keeffe and Lindsay Florek, and lin-12(ar624) [LIN-12::GFP], made by Jessica Chan, were kindly provided to us by them.

The following transgenes were used to label somatic gonad nuclei or facilitate scoring or cell fate determination: arTi112[ckb-3p::mCherry::H2B] V is a single-copy insertion transgene that marks all cells of the somatic gonad primordium and was described in [34], and was based on the ckb-3p described in [35]. We made arTi145[ckb-3p::mCherry::H2B] II, a second single-copy insertion of the same ckb-3p::mCherry::H2B marker. Using both transgenes together improved the brightness of the marker for longitudinal imaging studies.

arIs51[cdh-3p::gfp] [7] is expressed in the late L2 and L3, and was used as an AC marker to assess the phenotype of hlh-2(ar614). arIs131[lag-2p::2XNLS::YFP] isa lag-2 transcriptional reporter [36].

ccIs4251[myo-3::2XNLS::GFP] [37] was used in some experiments to facilitate z stack alignment. Originally obtained in strain SD1546, we backcrossed ccIs4251 to wild-type N2 ten times to generate GS8611.

METHOD DETAILS

Generating hlh-2(ar614) [hlh-2(Δprox)]

hlh-2prox is a regulatory element that is sufficient to drive expression in the α and β cells and their parents [8]. To delete this element, we used a method [38] in which Cas9-generated cleavage events are repaired precisely by homology-dependent repair (HDR) using a single-strand oligodeoxynucleotide (ssODN) as a repair template combined with repair of pha-1(e2123) as a co-CRISPR selection strategy as described in [39]. The ssODN repair template, oJB78 (50 ng/μL), was injected with two hlh-2 directed sgRNA plasmids pJB28 & pJB30 (25 ng/μL each), a pha-1 directed sgRNA and Cas9-expressing plasmid pJW1285 (50 ng/μL), and a pha-1 repair ssODN (50 ng/μL) [39] into the germline of pha-1(e2123) hermaphrodites. Injected hermaphrodites were placed at 25°C. 3–4 days after injection, surviving progeny of the injected P0s were singled out onto NGM plates. At this point, we noticed a recessive Vulvaless phenotype exhibited by the progeny of a surviving F1, as expected for a proximal gonad-specific null allele created by deleting the hlh-2prox element. We established strain GS8645, amplified the hlh-2prox region by PCR, and sequenced it. Although the anticipated homology-directed repair event did not occur, a deletion of 207 base pairs of the element were deleted in what we presume was a nonhomologous end-joining event. This deletion allele was named hlh-2(ar614) and is referred to herein as hlh-2(Δprox).

hlh-2 GFP-tagged alleles

We used the self-excising cassette (SEC) method of generating fluorescent protein knock-ins as described in [40]. With the help of Hannah Dayton, we amplified two 1.5kb-long homology arms from N2 genomic DNA and used Gibson cloning to insert them into a pDD282 backbone, which contains the GFP tag and the self-excising selection cassette. The resulting repair template was named pJB43. To generate hlh-2(ar623), we injected the repair template pJB43 (10 ng/μL), two hlh-2 directed sgRNA plasmids pJB44 & pJB47 (50 ng/μL each), a Cas9 expressing plasmid ‘‘Peft-3::cas9-SV40_NLS::tbb-2 3⁰UTR’’ (50 ng/μL) [41], and the coinjection markers pGH8 (rab-3p::mCherry, 2.5 ng/μL) and pCFJ90 (myo2p::mCherry, 10 ng/μL) [42] into the germline of wild-type strain N2 hermaphrodites. After the injection, we placed injected hermaphrodites at 25°C; after two days, we selected for integrants by treating plates with hygromycin (5 mg/mL) and picking animals with a Roller (Rol) phenotype as described in [40]. We then excised the SEC of both alleles by heat shocking in a 34°C water bath for 4 hours and picking non-Rol animals.

The resulting hlh-2(ar623) allele does not affect viability or fertility, and lays eggs copiously, indicating that hlh-2 function has not been significantly compromised. hlh-2(ar623) encodes GFP inserted in frame at the amino terminus, and the GFP-HLH-2 expression pattern agrees with previous analyses using an antibody directed to HLH-2 [40] and single-and multicopy transgenes that express GFP::HLH-2 [8]. We note that the hlh-2(ar623) allele contains a small indel upstream of the Cas9 target site about 1 kb upstream of the hlh-2 ATG (CCACCCTCG C → CAAAAAGAAG), which we believe is the result of more than one cut at the site by Cas9 (because the PAM site was not mutated in the repair template). We also note that excision of the SEC also removed one FLAG-tag, resulting in a 2xFLAG instead of a 3xFLAG from the pDD282 cassette [40].

To generate a deletion identical to the one present in hlh-2(ar614) in the hlh-2(ar623) background, we used a similar method as the one used to generate hlh-2(ar614), except that we used guides designed specifically to replicate the ar614 lesion. The ssODN repair template, oJB208 (50 ng/μL), two hlh-2 directed sgRNA plasmids pJB59 & pJB30 (25 ng/μL each), a pha-1 directed sgRNA and Cas9-expressing plasmid pJW1285 (50 ng/μL), and a pha-1 repair ssODN (50 ng/μL) were injected into the germline of GS9046 [hlh-2(ar623); pha-1(e2123)] hermaphrodites. We established the strain GS9013 from a surviving, egg-laying defective individual, and confirmed that the resulting deletion allele, ar629, was identical to the deletion present in hlh-2(ar614).

Microfluidics and long-term imaging

Synchronized populations of early/mid L1 animals were generated through a 1–2 hour timed egg-lay the night before an experiment, and placed at the temperature at which the experiment would be conducted. Early/mid L1 animals were picked and mounted into a microfluidic device that allows reversible immobilization of the animals as previously described [10]. Initial experiments were performed on the original setup, courtesy of Shai Shaham (Rockefeller University); subsequent experiments were performed on an equivalent setup constructed in the Greenwald lab.

In order to improve confinement of the small L1 larvae within the microfluidic chip, we reduced the size of the micro-channels of the worm imaging chamber from 4.88 μm [10] to 4 μm. During imaging, animals were fed a constant flow of NA22 E. coli in S medium. For cell lineaging in Figures 1 and S1 (GS8513, GS8592, GS8675, GS8686, and GS8812), animals were imaged every 8 min from L1 through the AC/VU decision until they reached the L3 stage (~20–24 hours). To visualize GFP::HLH-2 together with cell lineages (strain GS9062), animals were imaged every 16 min to minimize phototoxicity, with the exception of 9 animals imaged at 8 min intervals. We note that, although the 9 animals included developed normally, some animals were not included in the analysis because they did not grow when imaged at 8 min intervals, so we did not continue using that condition.

Long-term imaging was performed on a Zeiss AxioObserver.Z1 inverted microscope with a 40X, 1.4NA oil immersion objective, a 26mm WD, 0.55NA condenser and an XCite 120LED for epifluorescence microscopy illumination. Images were captured with a Hamamatsu Orca flash 4.0 LT+ CMOS camera. We positioned a 0.63x demagnifying adaptor in front of the camera, allowing us to capture the entire chamber in a single field of view. To operate electro-pneumatic regulators (SMC) and electric valves (Pneumadyne), a replicate of the setup developed in [10] was used. Custom-written MATLAB scripts combined with the open source platform mManager [43] were used to orchestrate microfluidics pressures/flows with imaging.

During long-term imaging, the device temperature was kept constant to ± 0.2°C by controlling and monitoring temperature of both objective and stage inset hosting the microfluidic device. The objective was water-cooled/heated using a custom-built aluminum ring with integrated temperature probe (https://www.ovenind.com/). Water temperature was regulated by a circulating water bath (https://www.coleparmer.com/). The microfluidic chip was mounted on the microscope using a custom-built aluminum stage inset (see [10]), whose temperature was independently controlled and monitored via a Peltier device.

Scoring lineage and fate in long-term imaging

Strains GS8675 and GS8686 contain ckb-3p::mCherry::H2B transgenes to mark the nuclei of Z1 and Z4 and of their descendants, and ccIs4251 [myo-3p::2XNLS::GFP], which was used to develop automated scoring capabilities. GS8675 was maintained by selfing; GS8686 was derived from GS8675 and maintained as a cross. Both strains showed the same birth-order bias and the data were combined from the two strains in Figure 1.

Lineages were scored manually using µManager and ImageJ software. A single worm was viewed at each time point, and also in each slice within a time point. At each time point, the worm is in a different orientation and position within the chamber. The mCherry::H2B marker was used to mark interphase nuclei and also divisions, as it is associated with chromosomes and clearly identifies cells in metaphase and anaphase.

When the somatic primordium forms, cells arrange in the ‘‘5R’’ configuration if Z1.ppp becomes the AC (Video S1) or the ‘‘5L’’ configuration if Z4.aaa becomes the AC [8] (Video S2). These configurations can be reliably scored based on the nuclear positions of the AC and VUs, and provided the basis for scoring the outcome of the AC/VU decision.

For Figures 1 and S1, if the division timing of Z1.ppp and Z4.aaa was below the temporal resolution of our imaging, the animal was excluded from further analysis of the birth order bias because there was no difference to analyze (n = 4 at 25°C; n = 5 at 20°C).

For Figures 2A and S2E, we show the analysis of GFP::HLH-2 expression onset in all individuals in which the birth of the parents through the fate of the α cells was captured. For Figures 2B–2D, we focus on individuals in which GFP::HLH-2 was evident in one parent before the other (n = 34 for Figures 2A and S2E; n = 36 for Figures 2B–2D). GFP::HLH-2 onset was scored manually using mManager and ImageJ software. Brightness and contrast were increased by setting the upper clipping on the fluorescence histogram to 1500 counts.

In six animals, the onset timing difference of GFP::HLH-2 was the same in both parents. To see if there was a difference in GFP::HLH-2 expression dynamics or level in these animals, we quantified GFP::HLH-2 onset dynamics for all subsequent frames in the parents of these animals. GFP::HLH-2 dynamics were similar between the parent cells in all cases, and the low signal-to-noise ratio around GFP::HLH-2 onset precluded any inference of the order of GFP::HLH-2 expression onset.

Scoring expression of GFP-tagged proteins

We imaged lag-2(ar635) and lin-12(ar624) in a wild-type and hlh-2(Δprox) background (strains GS9921, GS9129, GS9923, GS9130) using a Zeiss AxioObserver Z1 inverted microscope with a 63X, 1.4NA oil immersion objective equipped with a spinning disk, CSU-X1A, and laser bench rack. A 488nm, 100mW laser was used to excite GFP and a 561nm, 75mW laser was used to excite mCherry. Images were captured with a Photometrics Evolve EMCCD camera.

arTi145[ckb-3p::mCherry::H2B] was present in all strains to mark all cells of the somatic gonad to allow for unambiguous identification of the developmental progress of the somatic gonad primordium. To synchronize a population of worms and enrich for the relevant developmental stage, we prepared eggs by treating gravid hermaphrodites with bleach to release eggs using a standard protocol [44], and depositing the resulting eggs onto seeded NGM plates. After 13–17 hours, we mounted larvae onto 4% agarose pads and immobilized them with 10 mM levamisole. Z stacks of GFP (50% laser power, 500 ms exposure) and mCherry (20% laser power, 300 ms exposure) fluorescence were collected from each larva scored using a Zeiss spinning disk confocal dual camera system. Each z-stack contained the entire somatic gonad primordium as determined by mCherry expression, and slices were taken at 260 nm intervals. Using the z-stacks, we observed LAG-2::GFP or LIN-12::GFP expression and categorized each animal by the number and position of cells in the somatic gonad primordium (identified by mCherry signal).

We imaged GFP::HLH-2 in a wild-type (GS9127) and the hlh-2 Δprox background (GS9128) as above except using 30% laser power and 300 ms exposure for GFP.

QUANTIFICATION AND STATISTICAL ANALYSIS

Fluorescence intensity quantitation

GFP::HLH-2 fluorescence intensities were quantified using custom-written MATLAB scripts. Epifluorescence micrographs were median-filtered with a filter size of three pixels to reduce camera shot noise. To obtain fluorescence intensity of a nucleus, we selected the z-slice in which it was most in-focus, manually outlined the nucleus based on the DIC and the fluorescence channel, and extracted the mean intensity level I_n in that region. In the same z-slice, we also outlined a background region within the germline and extracted its mean intensity I_b. Fluorescence intensity of the nucleus was calculated as I_n – I_b. Fluorescence time series (Figures 2B–2D, S2C, S2D, S2F, and S2G) were filtered with a Gaussian function with standard deviation of 10 min. Experiments with 16 min imaging interval were linearly interpolated to 8 min intervals for presentation.

Time lapse videos

To generate time lapse videos of early gonadogenesis (Videos S1 and S2), we performed dual channel imaging, visualizing somatic gonad cell divisions with ckb-3p::mCherry::H2B and overall worm morphology with Nomarski microscopy (DIC). We straightened each three-dimensional image stack using a previously published algorithm [10, 45] based on a worm backbone which was manually outlined using the DIC channel. The resulting straightened worm z-stacks were then cropped in xy to a region of interest surrounding the somatic gonad. To deconvolve the straightened, cropped ckb-3p::mCherry::H2B z stacks, we measured the point-spread function (PSF) of our optical setup using red (580/605nm) fluorescent 200nm beads (ThermoFisher Scientific). Using the obtained PSFs, we deconvolved stacks using the classic maximum likelihood estimation (CMLE) algorithm with standard parameters (refractive index of imaging medium: 1.338) in the Huygens Essential software (Huygens Essential 3.7.1) by Scientific Volume Imaging (SVI). Finally, time lapse videos for ckb-3p::mCherry::H2B z stacks (Figure 1B; Videos S1 and S2) were generated by aligning the centers-of-mass of fluorescence of all straightened, cropped and deconvolved z-stacks and calculating maximum-z projections for each frame. Frames were removed from the videos whenever residual animal movement excessively blurred the images.

Statistical analyses

All statistical analyses were performed in MATLAB, either using built-in functions (Pearson correlation coefficient, Student’s t tests) or custom-written scripts (bootstrap permutation test). Differences between two groups are judged to be statistically significant when p < 0.05 by Student’s t test or bootstrap permutation test, where appropriate.

Statistical analysis of GFP::HLH-2 onset timing and cell-cycle duration

As discussed in the main text (cf. Figure 2A), the onset of GFP::HLH-2 expression in each parent cell occurs within a narrow time window after its birth. We found no correlation between the overall length of the parent cell cycle and the time from parent birth to onset of GFP::HLH-2 expression (r = 0.16, p > 0.18; n = 34), and no significant difference in GFP::HLH-2 onset timing between parents that gave rise to the AC or the VU (p > 0.2; two-sample t test; n = 34). We note that the time threshold for when the AC/VU decision displayed the birth order bias was slightly longer, likely because the overall developmental rate, as measured by cell-cycle lengths, was ~15% slower in GFP::HLH-2 +marker than in the marker-only experiments (Figures 1 and S1), attributable to mild phototoxicity effects of repeated illumination at the wavelength for visualizing GFP (e.g., average Z1.pp cc-duration 201min versus 172min). Animals developed normally otherwise, and the AC/VU decision correctly resolved in all imaged individuals.

We also found no significant difference in time from parent birth to onset of GFP::HLH-2 expression between the first-born or second-born parent (105min ± 6min versus 100.5min ± 5 min, p > 0.23; n = 34). Similarly, we found no significant difference in time from parent birth to onset of GFP::HLH-2 expression between a parent that gave rise to the first-born alpha-cell or second-born alpha cell (104min ± 5min versus 102min ± 6min, p > 0.63). Finally, there was also no correlation between the difference in birth timing of the parent cells and when the first-born parent or the second-born parent initiated GFP::HLH-2 expression (r = —0.16, p > 0.23). These observations are consistent with a cell-intrinsic mechanism for initiating HLH-2 expression.

Highlights.

• LIN-12/Notch resolves a stochastic cell fate decision of two equipotential α cells

• Two stochastic events bias their fates: relative birth order and HLH-2 expression

• Onset of HLH-2 expression is linked to the birth of the parents of the α cells

• LIN-12 expression requires HLH-2: a potential deterministic decision mechanism