Blood stem cells emerge from aortic endothelium by a novel type of cell transition

Vol 464 | 4 March 2010 | doi:10.1038/nature08761

LETTERS Blood stem cells emerge from aortic endothelium by a novel type of cell transition Karima Kissa1 & Philippe Herbomel1

The ontogeny of haematopoietic stem cells (HSCs) during embryonic development is still highly debated, especially their possible lineage relationship to vascular endothelial cells1,2. The first anatomical site from which cells with long-term HSC potential have been isolated is the aorta-gonad-mesonephros (AGM), more specifically the vicinity of the dorsal aortic floor3. But although some authors have presented evidence that HSCs may arise directly from the aortic floor into the dorsal aortic lumen4, others support the notion that HSCs first emerge within the underlying mesenchyme5. Here we show by non-invasive, high-resolution imaging of live zebrafish embryos, that HSCs emerge directly from the aortic floor, through a stereotyped process that does not involve cell division but a strong bending then egress of single endothelial cells from the aortic ventral wall into the sub-aortic space, and their concomitant transformation into haematopoietic cells. The process is polarized not only in the dorso-ventral but also in the rostro-caudal versus medio-lateral direction, and depends on Runx1 expression: in Runx1-deficient embryos, the exit events are initially similar, but much rarer, and abort into violent death of the exiting cell. These results demonstrate that the aortic floor is haemogenic and that HSCs emerge from it into the sub-aortic space, not by asymmetric cell division but through a new type of cell behaviour, which we call an endothelial haematopoietic transition. We6,7 and others8,9 have previously shown that in zebrafish embryos, the very thin space that initially separates the dorsal aorta (DA) from the underlying axial vein (AV) all along the trunk is the anatomical and functional homologue of the AGM in mammals, that is, it contains the multipotent haematopoietic stem/progenitor cells (HSPCs) that will enter circulation to colonize, expand and differentiate in the subsequent haematopoietic organs, first the caudal haematopoietic tissue (CHT)6, which plays a haematopoietic role similar to the fetal liver in mammals, and then the definitive haematopoietic organs—the thymus and kidney marrow in fish. In mammalian and avian ontogeny, CD41 has been the earliest marker distinguishing blood progenitors from endothelial cells/progenitors, notably in the AGM5. Accordingly, in transgenic CD41– GFP (green fluorescent protein) zebrafish embryos, we reported that HSPCs can be first detected in the AGM of live zebrafish embryos as CD41–GFPlow cells, that appear asynchronously there from 33 hours post-fertilization (h.p.f.) until past 54 h.p.f. (ref. 7). Each new CD41– GFPlow cell then enters circulation through the underlying axial vein within the next 2–3 h, to almost immediately seed the CHT, expand there, and later seed the thymus and kidney7. In the present work, we investigated whether these HSPCs might arise from the aortic endothelium. We took advantage of the KDR– GFP transgenic line, in which the promoter of the Kdrl gene10 drives vascular-specific GFP expression from the angioblast stage11, and we followed up systematically the behaviour of aortic cells in live embryos by time-lapse fluorescence confocal microscopy, from the 1

stage of aorta formation (18 h.p.f.) to 100 h.p.f. (well beyond the peak of HSPC generation in the AGM and seeding of the CHT). Starting from about 30 h.p.f., the imaging revealed a high frequency of endothelial cells from the aortic floor that underwent lasting contraction then bending towards the sub-aortic space, and remained in this strongly bent configuration for typically 1–2 h (Figs 1d, e, k, 2h and Supplementary Fig. 2). Then a further contraction of the bent DA floor cell along the medio-lateral axis brought its two lateral (left and right) endothelial neighbours in contact with each other (Figs 1f, l, 2i, arrowheads, Supplementary Figs 1, 2e, g and Supplementary Movie 3). The cell then released its contact with its now joined left and right neighbours, but still maintained strong focal contact with both its rostral and its caudal neighbours (Fig. 1g, h, m, arrowheads, and Supplementary Fig. 2), while rounding up and already manifesting motility, leading to its oscillatory motion along the vessel’s axis (see for example cell 9 in Fig. 1m, n and Supplementary Movie 1). Then these distal contacts also dissolved, and the now free cell started to move in the sub-aortic space, with a typical haematopoietic progenitor morphology (Supplementary Movies 1–4). The successive steps that we could discern in this stereotyped sequence are recapitulated in Supplementary Fig. 1. They reveal how an endothelial cell that is part of the vascular floor is able to leave it without compromising the vessel’s integrity. We name this new type of cell transition an endothelial haematopoietic transition, or EHT. In about half of the cases, the resulting haematopoietic cell underwent apparently symmetrical division within the next 2 h (Fig. 1o, p–s and Supplementary Movies 1 and 2), then it entered a ‘microstroma’ made by the reticulation of the dorsal wall of the underlying axial vein (Figs 1r–u, 2a, Supplementary Movies 1, 2 and 4 and data not shown), and from there entered the vein lumen and blood circulation (Fig. 1u–v and Supplementary Movies 1 and 4), as we had previously documented with CD41–GFP embryos7. In the two-somite long segment of the DA observed and analysed in Fig. 1 and Supplementary Movie 1, two pairs of cells (1 to 4 in Fig. 1b) were already round and off the DA floor at the onset of the movie, and six were seen undergoing the entire EHT sequence during the recording session, that is, between 36 and 52 h.p.f. (cells 5 to 10), showing that most if not all cells forming the DA floor at 35 h.p.f. eventually undergo EHT. Four different embryos similarly imaged over 10–16 h in the same 32–56 h.p.f. period displayed a similar frequency of EHT events: 0.19, 0.25, 0.27 and 0.30 per somite per hour. As the AGM extends over about 12 somites in length (somites 6–17, ref. 6), this would mean an average of three cells undergoing EHT per AGM per hour, about half of which will divide before leaving, hence about 4.5 cells per hour entering the blood to seed the caudal haematopoietic tissue (CHT), in very good agreement with our previous estimations of CHT seeding rate by CD41–GFPlow HSPCs from the AGM (2.5 and 5 cells per hour at 35 and 48 h.p.f., respectively7).

Institut Pasteur, Unite´ Macrophages et De´veloppement de l’Immunite´ and CNRS: URA2578, 25 rue du Dr Roux, F-75724 Paris CEDEX 15, France.

112 ©2010 Macmillan Publishers Limited. All rights reserved

LETTERS

NATURE | Vol 464 | 4 March 2010

a

b

04:05 d

c

07:49 e 4

08:24 f

09:30

09:19 g

09:34

h

4

7

1

1 2

6 2

7 7

7

7

8

8

8

9

8

DA

8

7 8

AV

i 10

j

k

10

l 10

3

m

n

o

10

10

10

6

6

6

10

4 5 7 8 00:00

6

6

6

6

9 9

06:39

p

9

9

9 09:34 6b

12:34

r

13:25

13:40

s

t

9a 9b

9

13:50

15:30

u

v

12:44

q

6b 6a

6b

6b

6b 6a

12:44

6a

6a 13:40

13:50

6a 14:15

6a 14:20

14:25

Figure 1 | Endothelial cells from the aortic floor exit into the sub-aortic space to become haematopoietic cells. a, Global view of a KDR–GFP embryo at 26 h.p.f.; GFP fluorescence highlights the whole vasculature, the boxed area was imaged by time-lapse confocal microscopy from 36 to 52 h.p.f. (b–v) at 14 depth (z) planes (Z1 to Z14) spaced by 2 mm, every 5 min. White numbers indicate recording time in hours and minutes. Aortic cells undergoing EHT are numbered in red; cells 6a, 6b and 9a, 9b are daughters of cells 6 and 9, respectively. Each image is from one confocal plane, mostly Z10 (b–o); inset in b is from Z4 to show the already exited cells 1 and 2; p, Z6,

q, Z11, r–v, Z7. Arrows in b show the sense of circulation in the aorta (DA) and axial vein (AV). Arrowheads in f and l show the thin line evidencing the joining of the lateral neighbours of the exiting cell, whereas arrowheads in g, h and m point at its remaining focal attachment sites to rostral and caudal neighbours. A dotted line in s–v shows the trajectory of cell 6b from the subaortic space into a small niche within the AV microstroma (arrowhead in r, s), and its disappearance from there (v) as it entered the blood stream. Scale bars, 25 mm (b), 10 mm (c–v). See also Supplementary Movie 1.

Consistently, the first haematopoietic HSPCs that seed the CHT by 35 h.p.f.7, previously shown to originate from the AGM7, are KDR– GFP1 (Fig. 2b). By 3 days post fertilization (d.p.f.), KDR–GFP1 cells then heavily colonize the thymus (Fig. 2c), just as the CD41–GFPlow HSPCs did in our previous study7, but KDR–GFP expression appears to stay longer than CD41–GFP and still illuminates many thymocytes in the early thymus by 6 d.p.f. (data not shown). It was not possible to directly image potential KDR–GFP1 HSPCs in the kidney by 5 d.p.f., owing to the deep location of the tissue and dense KDR–GFP1 vascular network there. We therefore injected a plasmid-borne KDR– dTomato transgene in CD41–GFP embryos, so as to obtain its mosaic expression in vascular cells, and then searched for KDR–dTomato1 free cells in the kidney area. Most of those that we found did coincide with CD41–GFP1 HSPCs (Fig. 2d–g, arrowheads). To visualize the filiation between the aortic endothelial cells undergoing EHT and the initial emergence of CD41–GFPlow HSPCs directly, we used double transgenic (Lmo2–Dsred12 and CD41–GFP) embryos. Although Lmo2–Dsred expression in vascular cells is weaker than KDR–GFP, we could detect the egress of Lmo2–Dsred1 cells from the DA floor into the sub-aortic space, accompanied by the onset of detectable CD41–GFP expression, then a rise in the latter marking the resulting haematopoietic cells, as shown previously7 (Fig. 2h–n). We then addressed more closely the circumstances of the EHT events. We first noticed a temporal correlation between EHT events and previously undescribed phases of the development and behaviour of the DA as a whole. Figure 3 and Supplementary Movie 5

show the evolution of the DA area in a single embryo followed from 23 to 100 h.p.f. Up to 37 h.p.f., the DA diameter is steadily enlarging, from 18 to 37 mm, with a steep increase between 30 and 37 h.p.f. (Fig. 3a–d, j). During this period, DA endothelial cells undergo rearrangements relative to each other, some of which are quite extensive, bringing endothelial cells from the DA lateral or even dorsal wall to the DA floor (Fig. 3a–c, arrowheads and Supplementary Movie 5). Cells start undergoing EHT from the DA floor, then from 40 to 52 h.p.f., corresponding to the peak of EHT events (Fig. 3d–g and Supplementary Movie 5) and CHT seeding7, the DA diameter now steadily decreases from 37 to 20 mm (Fig. 3j). Then as EHT events become rarer, the DA diameter decreases more slowly (from 20 to 17.3 mm between 52 and 65 h.p.f.; Fig. 3h–j). No more EHT events were detected in the AGM past 60 h.p.f. A similar DA evolution was observed in four different embryos, with the DA width always peaking by 36–40 h.p.f. The temporal correlation between cell egress from the DA floor and gradual reduction of the DA diameter is probably meaningful. Indeed, the EHT process revealed by our imaging involves the joining of the lateral (rather than rostral and caudal) neighbours of the cell undergoing EHT, which by itself tends to reduce the vessel’s diameter significantly (Supplementary Fig. 2), for the DA circumference consists in no more than four cells. Finally, since the emergence of HSCs in the embryo of both amniotes and zebrafish is known to require the Runx1 transcription factor7,8,13,14, we suppressed specifically its expression in the developing embryos with a morpholino, as we did in our previous 113

©2010 Macmillan Publishers Limited. All rights reserved

LETTERS

NATURE | Vol 464 | 4 March 2010

a

b

c

a

KDR–GFP

AGM DA

KDR–dTomato CD41–GFP

31.3 h.p.f.

32.5 h.p.f.

36.5 h.p.f.

d

e

f

40.5 h.p.f.

42 h.p.f.

48.5 h.p.f.

g

h

i

54.8 h.p.f.

59 h.p.f.

91 h.p.f.

s

t DA

58 h.p.f.

35 h.p.f.

e

k

c

ph

AV

52 h.p.f.

d

b

oc

g

f

g

AV

oc

DA

h

i

33 h.p.f.

j

33.3 h.p.f.

0 mm 33.3 h.p.f.

3 mm

CD41–GFP

DIC

Lmo2–Dsred

7 d.p.f.

k

l

m

n

j

40 µm

c de

Lmo2–Dsred CD41–GFP

35

DA

b f

30

a

g

25

h

20

42 h.p.f.

43 h.p.f.

43.5 h.p.f.

44 h.p.f.

Figure 2 | The haematopoietic cells emerging by EHT of aortic floor endothelial cells are HSPCs that seed the successive haematopoietic organs. a–c, Live KDR–GFP embryos and larvae. a, Three KDR–GFP1 haematopoietic cells beneath the DA, one along the DA floor and two within the reticulated dorsal wall of the AV, one of which will soon enter the vein lumen (see Supplementary Movie 2). b, The first HSPCs that seed the CHT by 35 h.p.f. (black arrowhead on the video-enhanced DIC image to the left) are KDR–GFP1 (white arrowhead). c, The HSPCs that seed the thymus rudiment (t) by 58 h.p.f. are KDR–GFP1; phs, primary head sinus; oc, otic capsule. d–g, Live CD41–GFP larva injected at the one-cell stage with KDR–dTomato plasmid, imaged at 7 d.p.f.; the kidney (k) area, boxed in d, is shown enlarged in e–g; four of the five dTomato1 cells present in this field (e, arrowheads) are also GFP1 (f, g); g, gallbladder (autofluorescent). h–n, Live (Lmo2–Dsred; CD41–GFP) double transgenic embryos. h–j, A bending, Dsred1 cell of the DA floor already expresses CD41–GFP detectably (h), although less than already emerged HSPCs (blue asterisks); 20 min later (i, j), the joining of its endothelial lateral neighbours is visible (i, black arrowheads), and the next focal plane (j) reveals the cell beginning to round up. Consistently, before the joining (h), circulating erythrocytes in the DA, appearing as red stripes due to the scanning process, still come in contact with the bending cell (h, white arrowhead), whereas they no longer do after the joining (i). k–n, Progressive rise in CD41–GFP expression in a Lmo2–Dsred1 DA floor cell (white asterisk) as it undergoes EHT; blue asterisks, CD41–GFP1 HSPC, dividing between time points m and n. Scale bars,10 mm (a, b and h–j), 50 mm (c and e–g), 100 mm (d), 25 mm (k–n).

study7. We found that in these Runx1 morphants, once the DA is formed, its cells express the KDR–GFP transgene as in control embryos (Fig. 4a). Then from 32 h.p.f. onwards, some EHT events are initiated, but much more rarely than in normal embryos, and they are abortive: the endothelial cell contracts, initiates EHT, but then bursts into pieces as it rounds up to become a haematopoietic cell

i

15 10 18

28

38

48

58

68

78

88

98

108 h.p.f.

Figure 3 | EHT events correlate with a phase of DA radial shrinking, preceded by a phase of DA radial expansion. a–i, Time-lapse imaging of GFP fluorescence in a KDR–GFP embryo/larva from 28 to 100 h.p.f. Maximum projection from 18 z-planes spaced by 3 mm, colour-coded such that cells change colours as they move from one z-plane to another. a–c, Between 30 and 37 h.p.f., the DA diameter expands rapidly, and cell rearrangements occur; arrowheads here follow two DA cells that move respectively from the DA lateral wall and roof to reach its floor before the peak of EHT events. d–h, Rise, peak then decrease of EHT events (asterisks showing some) correlate with a gradual reduction of the DA diameter, which then remains stable (i). j, Graph showing the temporal evolution of the DA diameter in this embryo (green line), measured every 2 h, and its correspondence with panels a–i. The DA diameter in the 18–28 h.p.f. interval (blue dotted line) was measured from another embryo. Scale bar, 25 mm. See also Supplementary Movie 5.

(Fig. 4b–g and Supplementary Movie 6). Consequently, the CHT and then thymus are not seeded (ref. 7 and data not shown). On the basis of these data, we propose that Runx1 expression is required in the endothelial cell13 to achieve the EHT successfully. It seems very likely that the EHT described here for the zebrafish will also apply to the AGM of mammals and birds, and more generally to any haemogenic endothelium. Recent in vitro data using embryonic stem cell-derived cultures brought evidence that mouse endothelial cells may give rise to blood cells14,15 by a process that seemed not to involve asymmetric cell division15, like the EHT described here. At a more general level, the EHT is a novel type of cell transition by which a cell that belongs to a squamous epithelium, the blood vessel, can leave it to become a free cell without compromising the vessel’s

114 ©2010 Macmillan Publishers Limited. All rights reserved

LETTERS

NATURE | Vol 464 | 4 March 2010

a

08:17 b

00:00 c

02:23 d

1

1

02:36

1 2

Full Methods and any associated references are available in the online version of the paper at www.nature.com/nature. Received 6 July; accepted 15 December 2009. Published online 14 February 2010.

1

1. 2 AV

DA

29 h.p.f.

e

2 08:17 f

2 09:29 g

2. 09:47

3.

4. 3

3

3

3 37.3 h.p.f.

5. 37.3 h.p.f.

Figure 4 | Runx1 expression is required for a successful EHT. Time-lapse imaging between 29 and 39 h.p.f. of a KDR–GFP embryo injected with an anti-Runx1 morpholino at the one-cell stage. Numbers in red indicate recording time in hours and minutes. The three numbered DA floor cells initiate EHT, but then burst into pieces (arrowheads). Scale bars, 20 mm (a), 10 mm (b–g). See also Supplementary Movie 6. A similar outcome with only rare, abortive EHT events was observed in four out of four Runx1 morphant embryos.

6. 7. 8. 9. 10. 11.

integrity. As with all epithelia, the arterial endothelium displays a structural and functional apical–basal polarity, with the apical side facing the vessel’s lumen. Therefore, we suggest that the always basalwards direction of the endothelial cell bending and subsequent EHT in zebrafish is embedded in the apical–basal polarity of the endothelium. This would predict that in all cases and species, HSCs will be generated from haemogenic arteries into the abluminal space first, and will only subsequently enter circulation by intravasating through a vessel in the vicinity—the axial vein in the zebrafish AGM, the aorta itself in the AGM of amniotes. Thus, the two major and competing models in the field are both partly true: the aortic endothelium is indeed haemogenic, but owing to the inherent polarity of the EHT that makes it so, the resulting HSCs first emerge in the sub-aortic space. METHODS SUMMARY Zebrafish husbandry and embryo treatments. Zebrafish embryos were raised and staged as described previously6. The KDR–GFP, Lmo2–Dsred and CD41– GFP transgenic lines used here have been described11,12,16. Runx1 morpholinomediated knockdown was performed as in our previous study7. In vivo imaging. Embryos were anaesthetized and mounted for time-lapse confocal fluorescence microscopy as described previously7. Combined videoenhanced differential interference contrast (DIC) and fluorescence wide-field imaging was performed as previously described7.

12. 13.

14. 15. 16.

Godin, I. & Cumano, A. The hare and the tortoise: an embryonic haematopoietic race. Nature Rev. Immunol. 2, 593–604 (2002). Yoshimoto, M. & Yoder, M. C. Developmental biology: birth of the blood cell. Nature 457, 801–803 (2009). Taoudi, S. & Medvinsky, A. Functional identification of the hematopoietic stem cell niche in the ventral domain of the embryonic dorsal aorta. Proc. Natl Acad. Sci. USA 104, 9399–9403 (2007). Dieterlen-Lie`vre, F., Pouget, C., Bollerot, K. & Jaffredo, T. Are intra-aortic hemopoietic cells derived from endothelial cells during ontogeny? Trends Cardiovasc. Med. 16, 128–139 (2006). Bertrand, J. Y. et al. Characterization of purified intraembryonic hematopoietic stem cells as a tool to define their site of origin. Proc. Natl Acad. Sci. USA 102, 134–139 (2005). Murayama, E. et al. Tracing hematopoietic precursor migration to successive hematopoietic organs during zebrafish development. Immunity 25, 963–975 (2006). Kissa, K. et al. Live imaging of emerging hematopoietic stem cells and early thymus colonization. Blood 111, 1147–1156 (2008). Gering, M. & Patient, R. Hedgehog signaling is required for adult blood stem cell formation in zebrafish embryos. Dev. Cell 8, 389–400 (2005). Jin, H., Xu, J. & Wen, Z. Migratory path of definitive hematopoietic stem/ progenitor cells during zebrafish development. Blood 109, 5208–5214 (2007). Bussmann, J., Lawson, N., Zon, L. & Schulte-Merker, S. Zebrafish VEGF receptors: a guideline to nomenclature. PLoS Genet. 4, e1000064 (2008). Jin, S. W., Beis, D., Mitchell, T., Chen, J. N. & Stainier, D. Y. Cellular and molecular analyses of vascular tube and lumen formation in zebrafish. Development 132, 5199–5209 (2005). Zhu, H. et al. Regulation of the lmo2 promoter during hematopoietic and vascular development in zebrafish. Dev. Biol. 281, 256–269 (2005). Chen, M. J., Yokomizo, T., Zeigler, B. M., Dzierzak, E. & Speck, N. A. Runx1 is required for the endothelial to haematopoietic cell transition but not thereafter. Nature 457, 887–891 (2009). Lancrin, C. et al. The haemangioblast generates haematopoietic cells through a haemogenic endothelium stage. Nature 457, 892–895 (2009). Eilken, H. M., Nishikawa, S. & Schroeder, T. Continuous single-cell imaging of blood generation from haemogenic endothelium. Nature 457, 896–900 (2009). Lin, H. F. et al. Analysis of thrombocyte development in CD41-GFP transgenic zebrafish. Blood 106, 3803–3810 (2005).

Supplementary Information is linked to the online version of the paper at www.nature.com/nature. Acknowledgements We thank the Zebrafish International Resource Center at the University of Oregon and L. Zon for providing KDR–GFP and Lmo2–Dsred transgenic zebrafish, respectively, Y. Blum and M. Affolter for the KDR–dTomato plasmid, and C. Herbomel and O. Bihan-Poudec for graphic artwork. Author Contributions K.K. performed the confocal fluorescence imaging, data analysis, and morpholino or plasmid microinjections; P.H. performed the video-enhanced DIC imaging and wrote the manuscript with input from K.K. Author Information Reprints and permissions information is available at www.nature.com/reprints. The authors declare no competing financial interests. Correspondence and requests for materials should be addressed to P.H. ([email protected]).

115 ©2010 Macmillan Publishers Limited. All rights reserved

doi:10.1038/nature08761

METHODS Zebrafish husbandry and embryo treatments. Zebrafish embryos were raised and staged as described17. For imaging, embryos were grown in the presence of 0.003% 1-phenyl-2-thiourea to prevent melanin pigment formation17. The KDR–GFP transgenic line, previously named flk–GFP11, was obtained from the Zebrafish International Resource Center (University of Oregon) under the name Tg(kdrl:EGFP)s843. To obtain mosaic expression of a red fluorescent reporter in the vascular cells of embryos expressing the CD41–GFP transgene haematopoietic marker, as in Fig. 2d–g, we used a plasmid bearing a KDR–dTomato reporter gene18,19 and one I-Sce1 site (gift from Y. Blum and M. Affolter). An 18 ml premix was prepared containing 22 ng of this plasmid, 5 mM MgCl2 and 2 ml of I-SceI 103 buffer (Promega); 2 ml of I-Sce1 meganuclease (5 units ml21, Promega) were added just before the injection session; the resulting mix was kept on ice, and used within the next 15 min to inject transgenic CD41–GFP embryos at the one-cell stage (1 nl per embryo) into the blastodisc, as described20. Suppression of Runx1 expression was obtained as in our previous study7, by using an antisense morpholino designed against the splice donor at the end of the second exon (59-AGCGCTCTTACCGTATTTGGCGTCC-39) (ref. 8). One nanolitre of a 1 mM morpholino solution was injected in one- to four-cell embryos, just beneath the blastoderm. Vital counterstaining. Just before the imaging session, a BODIPY TR methyl ester dye solution (5 mM in DMSO, Invitrogen)7,21 was diluted 1:100 into fish water. KDR–GFP embryos were dechorionated and incubated for 1 h in this staining solution, then rinsed twice for 15 min in fish water before imaging. Mounting of zebrafish larvae for time-lapse imaging. Embryos were anaesthetized with 160 mg ml21 tricaine, then oriented and immobilized in 1% low melting point agarose in 35-mm glass bottom dishes (Iwaki), then covered with 2 ml fish water containing tricaine. Time-lapse confocal fluorescence imaging of live zebrafish embryos and larvae. Confocal microscopy was performed at 26 uC (room temperature) on two Leica SPE microscopes equipped with solid state lasers, one in upright configuration (Fig. 3 and Supplementary Movies 2, 3 and 5), the other in inverted configuration (Figs 1, 2, 4 and Supplementary Movies 1, 4 and 6), with a 340 oil (Figs 1b–v, 2a, h–j and 4), 316 oil (Fig. 2c, k–n), or 320 water immersion (Fig. 3) objective. Imaging conditions were adjusted for each sample to minimize embryo illumination. We typically used the following parameters in our systems: the 488-nm laser diode set at 50% power, with gain at 1,000 and zero offset, and image size at 5123512 pixels. The spacing between successive confocal planes was typically 2 mm, or 0.6 mm when higher resolution was required to produce a

satisfactory orthogonal projection, as in Supplementary Fig. 1. Confocal z-stacks were acquired every 5–6 min over several hours (up to 108 h in Fig. 3). We noted that for long time-lapse recordings, decreasing the time interval between successive acquisitions to below 5 min led to some phototoxicity, notably manifested by a decrease in the frequency of observable EHT events from the aorta. Processing confocal images. Following the time-lapse recording session, we first generated either a maximum or a colour-coded projection of the z-stacks for each time point so as to visualize the overall dynamics of the aorta and associated haematopoietic progenitors. Confocal planes were then analysed individually through time to select the most relevant ones. To visualize the behaviour of endothelial cells from the aortic ventral wall in transverse view, confocal images were acquired in lateral view through a 340 oil immersion objective with a z-spacing of 0.6 mm, then the data from the z-stacks were visualized through an orthogonal projection along the y axis: images in each z-stack were cropped down to a slice of 2 mm along the y axis passing through the cell of interest (double arrow in Supplementary Fig. 2a), and this was followed by a maximal projection of all remaining pixels along the y axis (Supplementary Fig. 2c, e, g and i). Maximum or colour-coded projection of z-stacks, as well as red/green/DIC signal overlay, aorta diameter measurements, and export of time-lapse series as avi movies were performed with the Leica LAS software driving the confocal microscopes. avi movies were cropped and annotated with the ImageJ software, then compressed and converted into QuickTime movies with the QuickTime Pro software. Wide-field microscopy. Combined video-enhanced DIC and fluorescence wide-field imaging (Figs 1a and 2b) was performed on a Nikon 90i microscope as described previously7. 17. Westerfield, M. The Zebrafish Book: a Guide for the Laboratory Use of Zebrafish Danio rerio 4th edn (Univ. of Oregon, 2000) Æhttp://zfin.org/zf_info/zfbook/ zfbk.htmlæ. 18. Shaner, N. C. et al. Improved monomeric red, orange and yellow fluorescent proteins derived from Discosoma sp. red fluorescent protein. Nature Biotechnol. 22, 1567–1572 (2004). 19. Blum, Y. et al. Complex cell rearrangements during intersegmental vessel sprouting and vessel fusion in the zebrafish embryo. Dev. Biol. 316, 312–322 (2008). 20. Soroldoni, D., Hogan, B. M. & Oates, A. C. Simple and efficient transgenesis with meganuclease constructs in zebrafish. Methods Mol. Biol. 546, 117–130 (2009). 21. Cooper, M. S. et al. Visualizing morphogenesis in transgenic zebrafish embryos using BODIPY TR methyl ester dye as a vital counterstain for GFP. Dev. Dyn. 232, 359–368 (2005).

©2010 Macmillan Publishers Limited. All rights reserved