In vivo analysis of cardiomyocyte proliferation during trabeculation

The heart is the first organ to form and function, and it faces the challenge of maintaining rhythmical contractions while undergoing multiple morphogenetic changes (Rana et al., 2013). Cardiomyocyte divisions not only lead to heart growth but also to the overall sculpting of the heart, which is the result of region-specific rates of proliferation within the organ (Moorman and Christoffels, 2003; Sedmera et al., 2003; Sedmera and Thompson, 2011).

A distinctive feature of cardiomyocytes is the presence of contractile filaments, called myofibrils, which are built by a repetitive arrangement of sarcomeres (Guan et al., 1999). It is thought that the limited proliferative capacity of adult mammalian cardiomyocytes is partially due to the presence of voluminous and stable mature myofibrils (Ahuja et al., 2004; Bersell et al., 2009; Fan et al., 2015). Consistent with this hypothesis, the regenerative capacity of adult zebrafish hearts upon injury is thought to be achieved through cardiomyocyte dedifferentiation with partial sarcomere disassembly followed by cardiomyocyte proliferation (Jopling et al., 2010; Kikuchi et al., 2010). However, most studies on cardiomyocyte proliferation have been carried out in cell culture and so our understanding of the behavior of dividing cardiomyocytes in vivo is limited.

At early stages of heart development, the increase in myocardial cell layers is mostly achieved through the formation in the ventricle of multicellular luminal projections, called trabeculae. Trabeculation is a complex process that involves crosstalk between the myocardial and endocardial layers, cardiomyocyte selection and delamination from the compact layer, and cardiomyocyte proliferation (Staudt and Stainier, 2012; Samsa et al., 2013; del Monte-Nieto et al., 2018). The disturbance of any of these processes leads to excessive or reduced trabeculation, resulting in heart failure (Zhang et al., 2013; Towbin et al., 2015; Finsterer et al., 2017). Still, the cellular and molecular mechanisms driving trabeculation are not completely understood. In mouse embryos, the growth of trabeculae is characterized by a gradient of cardiomyocyte proliferation from the base to the tip, with higher rates at the base (Sedmera and Thompson, 2011; de Boer et al., 2012). Some recent papers have also reported that, in mouse, both oriented cell divisions and migration contribute to the initial steps of trabecular formation (Li et al., 2016; Passer et al., 2016). In zebrafish, delamination of cardiomyocytes from the compact layer seems to be the main mechanism for trabecular initiation (Liu et al., 2010; Staudt et al., 2014), but the role of proliferation in this process has not been well characterized.

Multiple signaling pathways have been implicated in the control of trabecular formation, including Notch (Grego-Bessa et al., 2007; D'Amato et al., 2016), semaphorin/plexin (Toyofuku et al., 2004), bone morphogenetic proteins (BMPs) (Chen et al., 2004), transforming growth factor β (TGFβ) (Kruithof et al., 2013; Kodo et al., 2016), angiopoetin 1/Tie2 (Suri et al., 1996; Tachibana et al., 2005), and some of them also control cardiomyocyte proliferation. One of the most important pathways regulating trabeculation is the neuregulin (Nrg)/ErbB2/ErbB4 signaling pathway. Mouse embryos lacking the receptor tyrosine kinase ErbB4, the co-receptor ErbB2 or the ligand Nrg1 fail to form trabeculae (Gassmann et al., 1995; Lee et al., 1995; Meyer and Birchmeier, 1995; Grego-Bessa et al., 2007). In erbb2 zebrafish mutants, cardiomyocytes fail to delaminate and also show reduced proliferation (Liu et al., 2010; Staudt et al., 2014). In contrast, overexpression of Nrg1 or Nrg2a promotes cardiomyocyte proliferation (Gemberling et al., 2015; Rasouli and Stainier, 2017). However, the role of cardiomyocyte proliferation in the early steps of trabeculation and how the Nrg/ErbB2 pathway controls each of these processes is not completely understood.

Here, we use the well-established FUCCI system (Sakaue-Sawano et al., 2008; Zielke and Edgar, 2015) to analyze the behavior of proliferating cardiomyocytes in the developing zebrafish heart. Through time-lapse imaging of beating hearts, we were able to observe divisions parallel to the myocardial wall in both compact and trabecular layer cardiomyocytes, as well as a gradual increase in the number of dividing cardiomyocytes in the trabecular layer with time. We also observed for the first time in vivo sarcomere disassembly prior to cardiomyocyte division. By manipulating the Nrg/ErbB and TGFβ signaling pathways, we found that proliferation of compact and trabecular cardiomyocytes responds to them differently and thus may be controlled by distinct mechanisms. Altogether, our results shed light on the role of cardiomyocyte proliferation in trabecular growth in zebrafish, and provide new insights into the behavior of proliferating cardiomyocytes during development.

A FUCCI system was previously used in the zebrafish heart for a small molecule screen (Choi et al., 2013). We decided to analyze the relative distribution of zebrafish myl7:mVenus-Gmnn⁺ cardiomyocytes at different developmental stages in more detail. We crossed the Tg(myl7:mVenus-gmnn) line, which enables one to visualize cycling cardiomyocytes, to the Tg(myl7:MKATE-CAAX) line, which labels cardiomyocyte membranes. Three-dimensional reconstructions of 72, 96, 120 and 144 hpf hearts (Fig. 1A-C,E) show myl7:mVenus-Gmnn⁺ cardiomyocytes mostly localized in the outer curvature of the ventricle, close to the basal pole. At the stages analyzed, two layers can be distinguished in the zebrafish myocardium: the one cell thick compact layer at the periphery; and the trabecular layer, formed by multicellular projections growing into the lumen of the ventricle. As observed from cross-sections of hearts at 120 and 144 hpf, most myl7:mVenus-Gmnn⁺ cardiomyocytes were located in the trabeculae, with only a few myl7:mVenus-Gmnn⁺ cardiomyocytes found in the compact layer (Fig. 1D,F, arrowheads).

To confirm our observations that most proliferating cardiomyocytes are found in the trabecular layer, as observed with the Tg(myl7:mVenus-gmnn) line, we performed cell proliferation assays using EdU, a nucleotide analogue that is incorporated during S phase (Cappella et al., 2015). A short 6 h EdU incubation starting at 96 hpf labeled several trabecular cardiomyocytes (Fig. 1G, red arrowheads), whereas a longer 24 h incubation led to a high proportion of trabecular cardiomyocytes stained for EdU incorporation at 120 hpf (Fig. 1H, red arrowheads) and only a few compact layer cardiomyocytes were EdU⁺. We conclude that, during zebrafish heart development, a significant number of proliferative cardiomyocytes are located in the emerging trabeculae.

To estimate the rate of cardiomyocyte proliferation at different developmental stages, we quantified the number of myl7:mVenus-Gmnn⁺ cardiomyocytes in relation to the total number of cardiomyocytes, as assessed using the Tg(myl7:nlsDsRed) line (Fig. 1I). In parallel, we incubated Tg(myl7:nlsDsRed) zebrafish embryos and larvae with EdU for 6 h to quantify the proportion of EdU⁺ cardiomyocytes (Fig. 1J). Our results show that the rate of cardiomyocyte proliferation peaks at 96 hpf, and subsequently decreases (Fig. 1K). Next, we estimated the cell division length of cardiomyocytes by determining the time from when it starts expressing myl7:mVenus-Gmnn (beginning of S phase) until the appearance of the daughter cells. We used the dual FUCCI system, which consists of Tg(myl7:mVenus-Gmnn) expression (which marks actively proliferating cardiomyocytes) and Tg(myl7:mCherry-Cdt1) expression (which marks quiescent cardiomyocytes) in combination with lightsheet microscopy. We set up hourly imaging during which time the hearts were transiently stopped. In a 96 hpf heart, we followed a cardiomyocyte that was initially quiescent (myl7:mCherry-Cdt1⁺/myl7:mVenus-Gmnn⁻ at t=0:00, white arrowhead), appeared to enter S phase (marked by the gradual accumulation of the myl7:mVenus-Gmnn signal) and eventually divided after ∼8 h (Fig. 1M,N). Other Venus-Gmnn⁺ cardiomyocytes could also be observed dividing during the imaging (Fig. 1M, yellow arrowhead and white arrows), while one cardiomyocyte was seen losing its Venus-Gmnn signal, presumably after cytokinesis had taken place (Fig. 1M, yellow arrows). By analyzing three different hearts we observed that cardiomyocytes in different locations of the ventricle divided 7-8 h after becoming Venus-Gmnn positive (Fig. 1N). Taken together, these data show the predominantly trabecular localization of proliferative cardiomyocytes and the estimated cell cycle length of larval cardiomyocytes, supporting the use of the Tg(myl7:mVenus-gmnn) line to study cardiomyocyte proliferation in real time during zebrafish larval development.

We next wanted to examine more closely the pattern of cardiomyocyte proliferation during trabeculation, given the high number of proliferating cardiomyocytes in the trabecular layer. It has been described that the first trabecular cardiomyocytes usually arise from delamination of compact layer cardiomyocytes, rather than through oriented cell division (Liu et al., 2010; Staudt et al., 2014). We analyzed the distribution of myl7:mVenus-Gmnn⁺ cardiomyocytes in larval hearts starting shortly after the onset of trabeculation. Using Tg(myl7:MKATE-CAAX) expression to distinguish individual cardiomyocytes, we observed a few myl7:mVenus-Gmnn⁺ cardiomyocytes in the trabecular layer at 72 hpf (Fig. 2A), the number of which gradually increased as trabeculation progressed (82 and 96 hpf; Fig. 2B,C). To confirm this increase, we determined the percentage of myl7:mVenus-Gmnn⁺ cardiomyocytes in the compact and trabecular layers of the ventricle. As trabeculation proceeded, the percentage of proliferative cardiomyocytes in the trabecular layer increased, so that by 96 hpf over half of all myl7:mVenus-Gmnn⁺ cardiomyocytes were located in the trabeculae (Fig. 2D). Similarly, the total number of myl7:mVenus-Gmnn⁺ cardiomyocytes increased in the trabecular layer from 72 to 96 hpf, while the number of compact layer myl7:mVenus-Gmnn⁺ cardiomyocytes remained fairly constant (Fig. 2E).

We wanted to further examine the behavior of myl7:mVenus-Gmnn⁺ cardiomyocytes at early stages of trabeculation when cells begin to delaminate to seed the trabecular layer and performed time-lapse imaging of beating Tg(myl7:mVenus-gmnn); Tg(myl7:mCherry-CAAX) hearts. We first carried out imaging starting at 72 hpf. At this stage, we detected myl7:mVenus-Gmnn⁺ cardiomyocytes undergoing cell division within the compact layer (Fig. 2F, arrowheads; Movie 1), but not perpendicular to the myocardial wall, consistent with previous reports on the initial phase of trabeculation occurring through delamination but not oriented cell division (Liu et al., 2010; Staudt et al., 2014). Interestingly, we also observed that some cardiomyocytes in the process of delaminating from the compact layer towards the cardiac lumen were myl7:mVenus-Gmnn⁺ (Fig. 2G, Movie 2), suggesting that compact layer cardiomyocytes are capable of seeding the trabecular layer independently of their cell cycle phase.

While we detected myl7:mVenus-Gmnn⁺ cardiomyocytes undergoing delamination, we did not observe in the course of our time-lapse imaging the completion of their cell division. To observe cardiomyocyte divisions in the trabecular layer, we performed live imaging on beating hearts, starting at 82 hpf, when several myl7:mVenus-Gmnn⁺ cardiomyocytes can be found in the trabeculae. We observed divisions of trabecular cardiomyocytes parallel to the compact layer (Fig. 2H, yellow arrowheads; Movie 3). Before cytokinesis happened, we also observed diffusion of the mVenus-Gmnn signal throughout the cardiomyocyte, most probably due to the breakdown of the nuclear membrane (Fig. 2I, Movie 4 and Fig. S1). Altogether, these results address for the first time cardiomyocyte division in vivo during trabeculation and show that it is not involved in the seeding of the trabeculae, but appears to be important for trabecular growth, as indicated by the high number of proliferative cardiomyocytes in the trabecular layer.

A number of changes occur in a cell before it divides, including changes in volume and shape. In mammalian cell culture, entry into mitosis is coupled with an increase in cell volume up to 30% (Son et al., 2015; Zlotek-Zlotkiewicz et al., 2015). Another common feature for most proliferating cells is the acquisition of a spherical shape, which supports symmetrical division and correct segregation of genetic material (Stewart et al., 2011; Cadart et al., 2014; Lancaster and Baum, 2014). Cell rounding has also been reported in cardiomyocytes in culture (Yahalom-Ronen et al., 2015). We decided to use the diffuse mVenus-Gmnn signal observed throughout the dividing cardiomyocyte to assess cell volume and shape. We used Tg(myl7:BFP-CAAX) expression to label cardiomyocyte membranes and injected a myl7:mScarlet plasmid to label individual cardiomyocytes and measure their mitotic and interphase volume (Fig. 3A-C). Three-dimensional surface renderings of individual cardiomyocytes show that in the z axis, mitotic cardiomyocytes exhibit an increase in thickness and roundness compared with interphase cardiomyocytes (Fig. 3D′,E′,F′). Two types of shape changes were observed: a more homogeneous increase in thickness and overall acquisition of cylindrical shape (Fig. 3D,D′), as well as an increase in thickness only in the central part of the cardiomyocyte (Fig. 3E,E′). We also used image segmentation to measure cell volume and observed that it is increased in cardiomyocytes prior to division (Fig. 3G). Thus, similar to other cell types, before division cardiomyocytes increase their volume and undergo rearrangements in shape leading to increased roundness.

As mentioned earlier, a very distinctive feature of cardiomyocytes is the presence of contractile elements: the sarcomeres. Studies of mammalian cardiomyocyte cultures have shown that sarcomere disassembly precedes cardiomyocyte division (Ahuja et al., 2004; Engel et al., 2006; Fan et al., 2015). We wanted to address sarcomere behavior during in vivo cardiomyocyte division in the zebrafish heart. We used two reporter lines: Tg(myl7:actn3b-EGFP), in which the Z-disc anchoring protein α-actinin is labeled (Lin et al., 2012); and Tg(myl7:LA-tdTomato), in which the thin myofibrils of the sarcomeres are labeled (Reischauer et al., 2014; Fukuda et al., 2017). We first crossed the Tg(myl7:actn3b-EGFP) line with the Tg(myl7:mCherry-cdt1) line, and injected the resulting embryos with a myl7:BFP-CAAX plasmid to label individual cardiomyocytes. We observed that G1 phase cardiomyocytes, labeled by an mCherry⁺ nucleus, display continuous striations throughout the cell (Fig. S2A-B″). In contrast, an mCherry⁻ cardiomyocyte, which is potentially mitotic, shows striations only on its periphery (Fig. S2C-C″). We then performed time-lapse live imaging of beating hearts to follow the behavior of sarcomeres in dividing cardiomyocytes. Before dividing, cardiomyocytes exhibit a striated organization of their Z-bands (Fig. 3Ha,a′, Movie 5, Fig. S3A,A″ and Movie 6). However, this pattern gradually faded over time (Fig. 3Hb-c′) until only some striations were observed at the periphery of the cardiomyocytes. No clear striations were observed throughout the entire length of these cardiomyocytes (Fig. 3Hd,d′ and Fig. S3B-B″). In the next time frame, the cardiomyocyte could be seen dividing into two daughter cells, and the GFP signal remained at the border between these cells (Fig. 3He,e′ and Fig. S3C-C″). The striated pattern of the myofibrils was restored in both daughter cells over the next 1-2 h (Fig. S4 and Movie 5, Fig. S3 and Movie 6). To investigate the correlation of sarcomere disassembly with the behavior of myl7:mVenus-Gmnn⁺ cardiomyocytes, we crossed the Tg(myl7:mVenus-gmnn) line with the Tg(myl7:LA-tdTomato) line to mark sarcomeres, and then with the Tg(myl7:BFP-CAAX) line to label cardiomyocyte membranes. We analyzed the 3D reconstructions of these hearts at 96 hpf (Fig. 3I,J). As can be observed from a close-up of the ventricle, interphase cardiomyocytes (Fig. 3K,K′) display a striated pattern of Z-bands; in contrast, in a mitotic cardiomyocyte from the same heart (Fig. 3K,K″), the fluorescent signal appears more diffuse and blurry. This apparent disassembly of the sarcomeres in dividing cardiomyocytes can be also observed in the atrium where a region with diffuse fluorescent signal of LA-tdTomato corresponds to a dividing cardiomyocyte (Fig. 3L,L′). Interestingly, similar to what we previously observed with time-lapse imaging (Fig. 3He,e′), we detected daughter cells, which had not yet reassembled their sarcomeres as they underwent cytokinesis (Fig. 3M-M″). Altogether, these data show that, in vivo, cardiomyocyte division is coupled with disassembly of some of the myofibrils, although some peripheral myofibrils remain.

The Nrg/ErbB2 signaling pathway is involved in trabeculation in mouse and zebrafish, and regulates cardiomyocyte proliferation, among other processes. To further investigate its role in promoting cardiomyocyte division in more detail, we incubated Tg(myl7:mVenus-gmnn); Tg(myl7:MKATE-CAAX) larvae with the ErbB2 inhibitor PD168393 starting at 72, 96 or 120 hpf, for 12 h and quantified myl7:mVenus-Gmnn⁺ cardiomyocytes in the compact and trabecular layers 12 h later. We observed a significant reduction in the total number of proliferative cardiomyocytes at all three stages analyzed (Fig. 4A-C, Fig. S5A-C′,G-I′). Interestingly, this reduction in proliferative cardiomyocytes appears to be more pronounced in the trabecular layer, as opposed to the compact layer (Fig. 4C′, Fig. S5C′,I′), especially when the treatment was carried out from 96 to 108 hpf. In that case, the reduction in the number of proliferative cardiomyocytes was retained even 2 days after treatment, at 144 hpf (Fig. S5D-F′). These results suggest that trabecular cardiomyocytes are more sensitive to ErbB2 signaling inhibition than are compact layer cardiomyocytes. To follow the effect of this decrease in cardiomyocyte proliferation at later stages, we repeated the ErbB2 inhibition experiment at all three aforementioned stages on Tg(myl7:H2B-GFP); Tg(myl7:MKATE-CAAX) larvae, and quantified the number of cardiomyocytes in the compact and trabecular layers at 10 dpf. PD168393-treated fish did not appear to exhibit gross morphological abnormalities or body length phenotype compared with control (Fig. 5I). However, the trabeculae of these larvae were shorter and exhibited a simpler morphology (Fig. 4D-F″), and the number of trabecular cardiomyocytes was significantly reduced (Fig. 4G,H). Interestingly, treatment with the ErbB2 inhibitor starting at 72 or 120 hpf did not significantly affect the number of cardiomyocytes (Fig. S6A-H). We conclude that cardiomyocyte proliferation within the trabecular network relies on Nrg/ErbB signaling, and that it is particularly sensitive to its absence in a narrow time window (∼96 hpf).

We recently reported that Nrg2a is the main ligand for the ErbB2 signaling pathway during cardiac trabeculation in zebrafish and that its ectopic overexpression in cardiomyocytes leads to a significant increase in their proliferation (Rasouli and Stainier, 2017). We performed phospho-histone 3 (PH3) and EdU staining on Tg(myl7:nrg2a-p2a-tdTomato) fish, which ectopically express Nrg2a in cardiomyocytes, at different developmental stages (Fig. 4J-S). Interestingly, we observed a higher number of PH3⁺ cells already at 48 hpf, and some perpendicular divisions from the compact layer towards the lumen [Fig. 4J,K, inset; seven control embryos (0 to 1 PH3⁺ cardiomyocytes per embryo), 11 nrg2a-overexpressing embryos (0 to 3 PH3⁺ cardiomyocytes per embryo)]. At 72 and 96 hpf, we observed a higher number of EdU⁺ cardiomyocytes in nrg2a-overexpressing animals (Fig. 4L-N,P-R), which were seen in both the compact (outer) layer and the non-compact (inner) layers of the heart (Fig. 4O,S). At these later stages, we also observed in the nrg2a-overexpressing animals some perpendicular cell divisions from the compact layer towards the lumen, which are never observed in controls (Fig. 4L,M, inset). Altogether, our results indicate a crucial role for the Nrg/ErbB2 signaling pathway in regulating trabecular growth and cardiomyocyte division in both myocardial layers.

Recently, two TGFβ ligands, Inhbaa and Mstnb, have been shown to modulate cardiomyocyte proliferation in opposing manners during cardiac regeneration, promoting and inhibiting it, respectively (Dogra et al., 2017). To gain further insights into the potential role of Inhbaa in cardiomyocyte proliferation, we first performed 24 h EdU labeling on inhbaa-overexpressing larvae, using the Tg(myl7:inhbaa-2A-H2B-EGFP) line, from 72 and 96 hpf, and observed an increase in the total number of proliferative cardiomyocytes (Fig. 5A-H). Strikingly, a significant increase in the number of EdU⁺ cardiomyocytes was observed only in the trabecular, not the compact, layer (Fig. 5D,H). We took a closer look at the cardiac phenotype of Tg(myl7:inhbaa-2A-H2B-EGFP) fish at 96, 120 hpf and 10 dpf (Fig. S7 and Fig. 5I-J″). At 96 hpf, several larvae already exhibited a more dense trabecular network (Fig. S5A-A′,C-C′), a phenotype that became more evident at 120 hpf (Fig. S7D-E′). At 10 dpf, we observed a higher number of cardiomyocytes at the base of the trabeculae (Fig. 5I-J′) and the trabecular mesh looked denser with less space between the trabecular projections (Fig. 5I″-J″). These observations are in agreement with previous data showing that adult inhbaa-overexpressing fish exhibit hypertrabeculation but a grossly normal compact myocardial layer (Dogra et al., 2017). We conclude that inhbaa overexpression promotes proliferation specifically in the trabecular layer, but not in the compact layer, from early larval stages.

We next checked cardiomyocyte proliferation in mstnb mutants, which display an increase in cardiomyocyte numbers in both compact and trabecular layers in adulthood (Dogra et al., 2017). Although the mutant larvae did not exhibit gross morphological phenotypes, we observed a slight increase in the number of EdU⁺ cardiomyocytes, due to an increase in proliferation in both compact and trabecular layers (Fig. S8A), suggesting that the TGFβ ligand Mstnb functions to inhibit cardiomyocyte division in both layers.

Additionally, we wanted to check the activation of the TGFβ pathway in the different myocardial layers and performed immunostaining for phosphorylated Smad3, which has been shown to be downstream of Inhbaa signaling during heart regeneration. We observed a significantly higher proportion of pSmad3⁺ cardiomyocytes in the trabecular layer compared with the compact layer (Fig. S8B-D). Thus, activation of the TGFβ signal transducer Smad3 is higher in the trabecular layer and may contribute to the higher proliferative capacity of these cardiomyocytes. Altogether, these results suggest that the compact and trabecular myocardial layers possess different proliferative capacities that require tight regulation through distinct signaling pathways.

In the present study, we used the FUCCI system to investigate cardiomyocyte division in the developing zebrafish heart. A previous study (Choi et al., 2013) using a zebrafish cardiomyocyte FUCCI system reported lower proliferation rates than the rate we observed with our line. Two major differences between the two myl7:mVenus-Gmnn lines may explain the discrepancy. First, Choi et al. used human Geminin, whereas we used zebrafish Geminin. Both genes have been described to be reliable markers of proliferation in zebrafish embryos, but the numbers of Gmnn-expressing cells were not compared (Sugiyama et al., 2009). Additionally, different myl7 promoters were used in the two lines: a 5.1 kb region of the zebrafish myl7 promoter (Rottbauer et al., 2002) was used in Choi et al., whereas a shorter, 250 bp region of the zebrafish myl7 (Huang et al., 2003) was used in our study. Different myl7 promoters have been shown to promote different levels of transgene expression (Huang et al., 2003). Furthermore, we also noticed that after 5 dpf, larvae from some of the founders silence the transgene and thus the lower number of Gmnn⁺ cardiomyocytes reported at 6 dpf by Choi et al. might be a reflection of partial silencing.

We observed two distinctive features in the distribution of proliferative cardiomyocytes in the ventricle: they are mostly located in the outer curvature of the chamber, and, as trabeculae emerge, more dividing cardiomyocytes are observed in the newly formed trabecular layer. A higher proliferative rate in the outer curvature, a region of chamber development and ballooning, is a conserved feature in a range of organisms from birds to humans (Soufan et al., 2006; Sizarov et al., 2011; de Boer et al., 2012). Our results indicate that this mechanism of chamber growth is also present in zebrafish.

The higher proportion of dividing cardiomyocytes in the trabeculae differs from what has been described in mouse, where higher cardiomyocyte proliferation is observed at the base of the trabeculae and is gradually lower towards the trabecular tip (de Boer et al., 2012; Park et al., 2013). Currently, it remains unclear whether a gradient of cell proliferation is present within the zebrafish trabeculae as trabeculation proceeds. The presence of slowly proliferating cardiomyocytes in the trabeculae of mouse embryonic hearts is related to the presence of terminally differentiated cardiomyocytes of the conduction system (Sedmera et al., 2003; van Weerd and Christoffels, 2016). Zebrafish hearts, as well as those of other ectothermic species, do not have anatomically distinguishable conduction system components, and in adulthood their ventricular walls remain trabeculated and may function as a homogenous conduction network (Chi et al., 2008; Jensen et al., 2012; van Weerd and Christoffels, 2016). This observation may at least partially explain the difference in the distribution patterns of proliferating cardiomyocytes between zebrafish and mouse. Additional studies are needed to identify potential heterogeneity in cardiomyocyte proliferation throughout the zebrafish trabecular network.

Our time-lapse imaging is the first study that allows the observation of cardiomyocyte divisions in real time during trabeculation. In agreement with previous reports, we observed only parallel divisions in the compact layer, supporting the notion that the initial seeding of trabeculae in zebrafish mainly occurs through delamination (Liu et al., 2010; Staudt et al., 2014). This process may be different in mouse, where both oriented perpendicular divisions and delamination seem to be involved in the initial stages of trabecular formation (Li et al., 2016; Passer et al., 2016). However, a more recent report shows that trabeculation in mouse initiates earlier than was described previously (del Monte-Nieto et al., 2018), so additional studies in mouse embryos, involving time-lapse imaging in vivo and ex vivo, would be a valuable addition to the existing data to characterize the initiation of murine trabeculation. Regarding trabecular growth, as mentioned above, zebrafish trabecular cardiomyocytes appear highly proliferative compared with compact layer cardiomyocytes, in contrast to the situation in mouse where proliferation is higher in compact layer compared with the trabecular layer (de Boer et al., 2012; Hashimoto et al., 2014; Park et al., 2013). These results suggest that once the first cardiomyocytes enter the trabecular layer, the mechanisms of the growth and formation of the trabecular network may differ between these two species. However, lineage tracing data with Mesp1-Cre showed that murine trabeculae had a polyclonal origin (Chabab et al., 2016), similar to zebrafish larval trabeculae (Gupta and Poss, 2012), pointing to potential evolutionary conservation in this process and indicating the need for further studies.

Interestingly, in contrast to the exclusively parallel divisions observed in wild-type fish, overexpression of nrg2a in cardiomyocytes is capable of promoting divisions from the compact layer towards the lumen, but these cells arrange themselves in layers instead of forming the complex morphology of trabeculae. It is still not completely understood how cells are selected from the compact layer to seed the trabecular layer, but it is reasonable to think that, once selected, these cells activate different pathways to behave like trabecular cells. Whether these mechanisms are altered when nrg2a is overexpressed and whether the Nrg2a-induced extra cardiomyocytes continue behaving like compact layer ones require further research. Additionally, our study only addresses early stages of trabeculation and we cannot exclude the possibility that oriented divisions take place within the compact layer and contribute to the trabecular layer at later developmental time points, as has been described in mouse embryos (Li et al., 2016). We also observed that the first divisions in trabecular layer cardiomyocytes occur parallel to the compact layer and not towards the lumen. It will be interesting to analyze the molecular regulation of this cardiomyocyte division orientation as well as the post-division behavior of the daughter cells. Previous work has shown that trabecular myofibrils are formed by clonally unrelated cardiomyocytes at least until 10 dpf (Gupta and Poss, 2012), raising the possibility that newly divided cardiomyocytes actually move apart to contribute to different trabeculae. Additional studies, including lineage tracing experiments and clonal analysis, will be needed to investigate these later events.

Notably, we found that cardiomyocyte proliferation appears to be regulated by different signaling pathways in the compact and trabecular myocardial layers. Our data with ErbB2 signaling inhibition suggest that an increase in the number of trabecular cardiomyocytes at a narrow time window (at 4 dpf) is needed to sustain trabecular growth at later time points. However, ErbB2 signaling is known to regulate different aspects of cardiomyocyte behavior, including delamination and myofibril assembly (Reischauer et al., 2014; Staudt et al., 2014). Thus, results from perturbing ErbB2 signaling should be interpreted with caution, as it affects multiple aspects of trabeculation. Indeed, treatment with ErbB2 inhibitor at 3 and 5 dpf did not have a significant effect on the number of trabecular cardiomyocytes at later stages, but the expansion of the trabecular network was disturbed compared with control. It will be important to identify Nrg2a/ErbB2 targets at different developmental stages and in the two different myocardial layers to explain the complex effects this pathway exerts on trabecular formation.

The differential effect of the TGFβ ligands on compact and trabecular cardiomyocyte proliferation is another example of how proliferation is controlled distinctly in the two myocardial layers. Whether this effect is achieved through layer-specific expression of receptors or downstream effectors requires further analysis. In support of this idea, a recent report identified the Inhbaa receptor gene acvr2ab as highly enriched in the trabecular Tbx5a⁺ layer (Sánchez-Iranzo et al., 2018). It will also be interesting to investigate whether any other pathways exhibit different effects on compact and trabecular layer cardiomyocytes, and the time point at which trabecular cardiomyocytes start expressing genes that enable them to respond to distinct stimuli compared with compact layer cardiomyocytes.

Mature mammalian cardiomyocytes possess very low proliferative capacity, which ultimately results in serious health issues in the event of heart damage. For the benefit of individuals with heart disease, much effort has been put into developing means to promote cardiomyocyte proliferation (Tzahor and Poss, 2017). Because major cellular reorganization events in coordination with cell cycle phase changes occur in dividing cells, studying these processes in cardiomyocytes may shed light on the reasons behind their limited capacity to divide. Some of the most crucial changes required in most dividing eukaryotic cells include increases in cell volume and cell reshaping to a spherical architecture (Champion et al., 2017). Previous studies suggest that, in contrast to most other cell types, cardiomyocytes do not completely round up before mitosis (Ahuja et al., 2004; van Amerongen and Engel, 2008). Because cardiomyocyte division proceeds in conjunction with constant contraction, it is reasonable to speculate that heart development has evolved such that cardiomyocyte division interferes as minimally as possible with tissue integrity. Our observations in live dividing zebrafish cardiomyocytes show that, although these cells do not completely round up, they do undergo a change from being flat in shape to increasing in thickness and hence become more rounded than interphase cardiomyocytes. Rounding of cells is required for different aspects of cell division, including chromosome capture, spindle stability and positioning (Lancaster et al., 2013; Cadart et al., 2014). An interesting question is whether cardiomyocyte rounding also takes place in the multilayered myocardium in mammalian embryos or even in adult zebrafish following injury.

An additional level of complexity in cardiomyocyte division comes from the presence of voluminous myofibrils. Here, we show that, similar to previous observations in cell culture and on mouse heart sections (Bersell et al., 2009; Porrello et al., 2011), cardiomyocytes disassemble most of their sarcomeres during cell division in vivo. These results raise additional questions as to whether sarcomere disassembly is a prerequisite for cardiomyocyte division and the mechanisms behind this process. Differences were observed between embryonic and neonatal cardiomyocytes in culture in terms of the disassembly sequence of sarcomere components and the potential mechanisms involved (protein degradation in the case of embryonic cardiomyocytes and protein phosphorylation in neonatal cardiomyocytes) (Ahuja et al., 2004; Fan et al., 2015). The specific mechanisms by which zebrafish cardiomyocytes disassemble their sarcomeres during development, and whether the mechanisms used at embryonic stages, or alternative mechanisms, are employed for sarcomere disassembly during heart regeneration are interesting issues that merit further research.

All zebrafish husbandry was performed under standard conditions in accordance with institutional (Max Planck Institute), and national ethical and animal welfare guidelines.

Phenylthiourea (PTU) was added at 24 hpf to prevent pigmentation. Embryonic stages are given as hours post-fertilization at 28.5°C. Transgenic and mutant lines used are as follows: Tg(myl7:mVenus-gmnn)^ncv43Tg (Jiménez-Amilburu et al., 2016), abbreviated as Tg(myl7:mVenus-gmnn); and Tg(myl7:mCherry-cdt1)^ncv68Tg (this work), abbreviated as Tg(myl7:mCherry-cdt1) [both lines generated similarly to previously described FUCCI lines (Fukuhara et al., 2014)]; Tg(myl7:mKATE-CAAX)^sd11 (Lin et al., 2012), abbreviated as Tg(myl7:MKATE-CAAX); Tg(myl7:H2B-EGFP)^zf521Tg (Mickoleit et al., 2014), abbreviated as Tg(myl7:H2B-EGFP); Tg(myl7:mCherry-CAAX)^bns7 (this work), abbreviated as Tg(myl7:mCherry-CAAX); Tg(-0.8myl7:nlsDsRedExpress)^hsc4 (Takeuchi et al., 2011), abbreviated as Tg(myl7:nlsDsRed); Tg(Myl7:BFP-CAAX)^bns193 (Guerra et al., 2018), abbreviated as Tg(myl7:BFP-CAAX); Tg(myl7:nrg2a-p2a-tdTomato)^bns140 (Rasouli and Stainier, 2017), abbreviated as Tg(myl7:nrg2a-p2a-tdTomato); Tg(myl7:inhbaa-2A-H2B-EGFP)^bns146 (Dogra et al., 2017), abbreviated as Tg(myl7:inhbaa-2A-H2B-EGFP); Tg(myl7:actn3b-EGFP)^sd10 (Lin et al., 2012), abbreviated as Tg(myl7:actn3b-EGFP); Tg(myl7:LIFEACT-Tomato)^bns141 (Fukuda et al., 2017), abbreviated as Tg(myl7:LIFEACT-Tomato). The previously published mstnb^bns5 mutant line (abbreviated as mstnb^−/−) (Dogra et al., 2017) was used, and larvae were obtained by crossing mstnb^+/− to mstnb^−/− fish. High-resolution melting amplification was used for genotyping with the following primers: 5′-GTGTATTAATTGCATGTGGTCCAG-3′ and 5′-GAACACTGCTCGCTTTCCTC-3′.

The 0.2myl7:mScarlet plasmid was generated by cloning the mScarlet fluorescent protein into the iSce-I plasmid under the 0.2myl7 promoter. The same plasmid was used for cloning the BFP fluorescent protein tagged with a CAAX membrane-targeting motif (Lin et al., 2012).

Whole-mount immunochemistry was performed according to standard protocols. Embryos and larvae were anaesthetized with 0.4% (w/v) tricaine and fixed in 4% PFA overnight at 4°C. The next day they were treated with a permeabilizing solution for 2 h, left for 1 h in blocking solution and incubated overnight at 4°C with primary antibodies: myosin heavy chain MF20 (Developmental Studies Hybridoma Bank, clone MF20; 1:20), anti-DsRed (Clontech, 632496; 1:300), anti-phospho-Histone3 (Millipore, 06-570; 1:200) and anti-pSmad3 (Abcam, 52903; 1:200). Over the next day, the samples were washed five times and incubated overnight with secondary antibodies [anti-rabbit Alexa488, anti-rabbit Alexa 561, anti-mouse Alexa 488 (Live Technologies; 1:500)] and DAPI (Sigma; 1:1000). The next day, samples were washed five times and stored in washing solution, and confocal images were acquired using a 40× (1.1 numerical aperture) water-immersion objective on a Zeiss LSM800 upright microscope.

Embryos and larvae at different stages were incubated with 1 mM EdU for 6, 12 or 24 h in egg water containing PTU and 1% DMSO to facilitate EdU solubilization. After treatment, larvae were rinsed twice with water containing PTU, anaesthetized with 0.4% (w/v) tricaine for 5 min and fixed overnight in 4% PFA. The CLICK-IT reaction for EdU labeling was performed as described by the manufacturer (Invitrogen).

Larvae were mounted in 0.8% low-melting agarose containing 0.2 mg/ml tricaine on glass-bottom dishes and kept in PTU water containing 0.2 mg/ml tricaine throughout the experiment. Timelapse of individual planes of beating zebrafish hearts were imaged at 90 frames per second (fps) using a Cell Observer Spinning disc microscope (Zeiss, CSU-X1 Yokogawa spinning disk) with a 40×/1.1 water-immersion objective or a 63×/1.3 water-immersion objective. Movies were recorded with a Hamamatsu ORCAflash 4.0 sCMOS camera and the image was binned 4×4 to achieve a calculated pixel resolution of 0.7 µm (40× objective) or 0.4 µm (63× objective). For individual developmental time points, movies of two or three heart beats were acquired for 20-40 z planes 1 µm apart from each other. Developmental time points were collected each 30 min or 1 h, depending on the experiment, for 10 h. GFP and mCherry channels were aligned for pixel shifts using a standard sample. Data were analyzed in Fiji.

A lightsheet microscope (Lightsheet Z.1 Zeiss) was used for live 4D imaging of temporarily stopped hearts. Larvae were mounted in 0.5% agarose on a cobweb holder and kept for the entire experiment in a water chamber maintained at 28°C (Strobl et al., 2017). For stopped heart imaging, larvae were immersed in 1.6 mg/ml tricaine in PTU-containing water until the heartbeat stopped (between 5 and 12 min, depending on the thickness of the 0.5% low melting point agarose). The images were acquired with a Plan APO 20×/1.0 and CMOS camera with a 2× zoom. Embryos were revived by washing with 0.008 mg/ml tricaine in PTU-containing water and larvae were left in this solution for 30 min or 1 h until the next time-lapse point. Images were analyzed with the ZEN software (Zeiss) and Imaris x64 (Bitplane, 7.9.0).

Embryos and larvae were mounted in 1.5% agarose containing 1.6 mg/ml tricaine on glass-bottom dishes and then imaged using a spinning disk confocal microscope (Zeiss, CSU-X1 Yokogawa) or Zeiss LSM800 confocal microscope using a 40×/1.1 water-immersion objective. Three-dimensional images were processed using Imaris x64 (Bitplane, 7.9.0). Images were prepared using Adobe Photoshop.

Larvae were exposed to the ErbB2 signaling inhibitor PD168393 (Calbiochem) at 10 µM for 12 h as previously described (Jiménez-Amilburu et al., 2016). Egg water containing 1% DMSO was used to facilitate solubilization and control larvae were incubated in 1% DMSO. After the treatment, wells were rinsed twice with water containing PTU and kept for another 12 h at 28.5°C before imaging. Two independent incubations were performed. Alternatively, animals were kept in a 28.5°C incubator until 5 dpf and transferred to the system (16 larvae per tank) and kept in normal breeding conditions until 10 dpf, when they were collected for imaging. Three independent inhibitor treatment experiments were carried out.

Animals were kept in PTU water (0.03%) to prevent pigmentation from 24 until 120 hpf, at which time they were transferred to the system where they were raised until 10 dpf. In total, 16 larvae were grown per tank. A Zeiss LSM800 observer was used with a LD C-Apochromat 40×/1.1 W objective and a z-stack was acquired every 1 µm. Larvae in which the heart appeared collapsed and the trabecular and compact layers could not be clearly distinguished were discarded. After imaging the heart, the larvae were photographed in glass-bottom dishes using a Nikon SMZ25 high-end stereoscopic microscope. Body length was measured using Fiji. To quantify the total number of trabecular and compact cardiomyocytes, seven z-stacks separated by 10 stacks each were examined per heart. The membrane marker was used to distinguish compact from trabecular cardiomyocytes.

We thank Rashmi Priya, Chi-Chung Wu and Sebastian Gauvrit for valuable comments on the manuscript, Albert Wang for editing. Kenny Mattonet for suggesting the use of a cobweb holder in the lightsheet microscope, and Rita Retzloff and team for excellent zebrafish care.