Research Article

MicroRNAs Regulate Brain Morphogenesis in Zebrafish

See allHide authors and affiliations

Science  06 May 2005:
Vol. 308, Issue 5723, pp. 833-838
DOI: 10.1126/science.1109020


MicroRNAs (miRNAs) are small RNAs that regulate gene expression posttranscriptionally. To block all miRNA formation in zebrafish, we generated maternal-zygotic dicer (MZdicer) mutants that disrupt the Dicer ribonuclease III and double-stranded RNA-binding domains. Mutant embryos do not process precursor miRNAs into mature miRNAs, but injection of preprocessed miRNAs restores gene silencing, indicating that the disrupted domains are dispensable for later steps in silencing. MZdicer mutants undergo axis formation and differentiate multiple cell types but display abnormal morphogenesis during gastrulation, brain formation, somitogenesis, and heart development. Injection of miR-430 miRNAs rescues the brain defects in MZdicer mutants, revealing essential roles for miRNAs during morphogenesis.

MicroRNAs are evolutionarily conserved small non-protein-coding RNA gene products that regulate gene expression at the posttranscriptional level (1-3). In animals, mature miRNAs are ∼22 nucleotides (nt) long and are generated from a primary transcript (termed pri-miRNA) through sequential processing by nucleases belonging to the ribonuclease III (RNaseIII) family. Initially, Drosha cleaves the pri-miRNA and excises a stem-loop precursor of ∼70 nt (termed pre-miRNA), which is then cleaved by Dicer (4-7). One strand of the processed duplex is incorporated into a silencing complex and guides it to target sequences (1, 3). This results in the cleavage of target mRNAs and/or the inhibition of their productive translation (1-3).

Several hundred vertebrate miRNAs and several thousand miRNA targets have been predicted or identified, but little is known about miRNA function during development (1, 2, 8, 9). Clues to vertebrate miRNA function have come from several approaches, including expression analyses (1-3, 10-12), computational prediction of miRNA targets (8, 13-15), experimental support of predicted targets (13, 14, 16, 17), cell culture studies (16), and gain-of-function approaches (18). These studies have led to the suggestions that vertebrate miRNAs might be involved in processes such as stem cell maintenance (12, 19) or cell fate determination (17, 18, 20); however, no loss-of-function analysis has assigned a role for a particular miRNA or miRNA family in vivo, and it has been unclear how widespread the role of miRNAs is during vertebrate embryogenesis.

One approach to reveal the global role of vertebrate miRNAs is to abolish the generation of mature miRNAs with the use of dicer mutants. For example, dicer mutant embryonic stem cells fail to differentiate in vivo and in vitro (20), and dicer mutant mice die before axis formation (19), suggesting that mature miRNAs (or other Dicer products) are essential for early mammalian development. In zebrafish, maternal dicer activity has hampered the analysis of the single dicer gene. Mutants for the zygotic function of dicer (Zdicer) retain pre-miRNA processing activity up to 10 days postfertilization, presumably because of maternally contributed dicer (21). Zdicer mutants have no obvious defects other than a developmental delay at 7 to 10 days postfertilization, a stage when embryogenesis and major steps of organogenesis have been achieved (21). Hence, the global role of miRNAs during vertebrate embryogenesis is unknown. In light of these observations, we decided to generate zebrafish embryos that lack both maternal and zygotic dicer activity.

Generation of Maternal-Zygotic dicer Mutants

To eliminate all maternal contribution in dicer mutants, we took advantage of the germ line replacement technique (22). Wild-type zebrafish embryos depleted of their germ cells served as hosts for germ cells from homozygous dicer mutant donor embryos (fig. S1). The resulting fish were fertile even though they had a germ line that was exclusively constituted by mutant donor cells. As donors, we used dicerhu715/hu715 mutants, an allele that codes for a truncated Dicer protein that disrupts the RNaseIII and double-stranded RNA (dsRNA)-binding domains (21). Intercrossing of fish that had a dicer mutant germ line generated embryos that were maternal-zygotic mutant for dicer (MZdicer). In marked contrast to Zdicer mutants, MZdicer embryos did not generate mature miRNAs and displayed severe morphogenesis defects.

Loss of pre-miRNA Processing in MZdicer Mutants

Similar to other model systems, wild-type zebrafish embryos generate mature miRNAs from endogenous (21, 23) or exogenously provided pri-miRNAs, resulting in the post-transcriptional repression of reporter genes (fig. S2). miRNAs induce the cleavage of reporter RNAs with perfectly complementary target sites (PT) in the 3′ untranslated region (3′UTR), whereas imperfectly complementary sites (IPT) result in the noneffective translation of reporter mRNAs (24) (fig. S3). Previous biochemical and genetic studies have shown that Dicer is required for the generation of mature miRNAs (4, 5). To determine whether MZdicer embryos lack mature miRNAs, we first hybridized total RNA from 1-day-old zebrafish embryos to a microarray of probes for 120 different zebrafish mature miRNAs (10). Although such arrays are susceptible to cross-hybridization artifacts, we observed a marked reduction of signals in MZdicer mutants compared with wild-type embryos and zygotic dicer mutants (Fig. 1A). Of the 120 miRNA probes, 59, 35, and 9 gave a detectable signal in wild-type embryos, Zdicer mutants, and MZdicer mutants, respectively. To test for the presence of mature miRNAs more specifically, we performed Northern blot analyses. We found that of eight miRNAs present in wild-type embryos, none was detected in MZdicer mutants (Fig. 1B) (25). In most of these cases, the lack of processing resulted in an accumulation of the pre-miRNA (Fig. 1B). Northern analyses also suggested that the nine positive signals on the microarray probed with MZdicer RNA are unlikely to be due to mature miRNAs. First, we found that one miRNA (miR-206) is processed in wild-type but not mutant embryos (Fig. 1B). Second, two miRNAs (miR-223 and miR-122a) were detectable neither in wild-type embryos nor MZdicer mutants (Fig. 1B) (25). These results suggested that mature miRNAs were not generated in MZdicer mutants.

Fig. 1.

MZdicer mutants lack mature miRNAs. (A) miRNA array expression data from MZdicer, Zdicer, and wild-type (wt) embryos at 32, 28, and 28 hpf, respectively. The range of signal was from -100-fold to 0 + to 100-fold. Yellow denotes high signal and blue denotes low signal. The asterisks highlight miRNAs whose expression was also analyzed by Northern blot. (B) Northern blot analysis of different miRNAs in MZdicer mutants (32 hpf) and wild-type embryos (28 hpf).

As an additional assay for miRNA maturation in MZdicer mutants, we examined the response of a green fluorescent protein (GFP) reporter (3xPT-miR-430b) containing target sites for members of the miR-430 family of miRNAs (Fig. 2A). These miRNAs are transcribed at high levels in the embryo and silence reporter expression in wild-type embryos (Fig. 2, B and C). In contrast, the reporter is not silenced in MZdicer mutants, consistent with the lack of mature miRNAs (Fig. 2, B and C). We also injected a pri-miRNA (pri-miR-1) and followed its processing. Whereas pri-miR-1 mRNA was processed into mature miR-1 in wild-type embryos, no mature miR-1 was detected in MZdicer embryos (fig. S4). Moreover, reporter genes containing perfect or imperfect targets for miR-1 were no longer silenced in MZdicer mutants by pri-miR-1 injection (fig. S4). Taken together, these results indicate that MZdicer mutants do not process endogenous and exogenously provided pre-miRNAs and thus are devoid of mature miRNAs.

Fig. 2.

Silencing activity of miRNA duplexes but not pri-miRNAs in MZdicer. (A and B) Coinjection of miRNA duplexes (miR-204, miR-430a, or miR-430b) with GFP sensors that contain the coding sequence of GFP and three (3x) perfect target (PT) sites for the different miRNAs. See figs. S2 and S3 for details. (A) Schematic representation of the experimental set up. (B) Coinjection of GFP sensors with buffer (-) or miR duplexes into wild-type embryos and MZdicer mutants. Fluorescent microscopy shows GFP target expression (green) at 24 to 30 hpf. Bright-field image of embryos is shown below. The specific silencing of the targets can be identified by their corresponding miRNA duplexes in wild-type and MZdicer embryos. Endogenous miR-430 repressed the expression of its GFP sensor in wild-type embryos but not in MZdicer mutants. (C) Northern blot analysis of endogenous miR-430b in wild-type and MZdicer embryos, showing the accumulation of the miR-430b precursor and absence of the mature form of this miRNA in MZdicer embryos.

miRNA Duplexes Are Active in MZdicer Mutants

Injection of synthetic miRNA duplexes into wild-type zebrafish embryos initiates the effector step of RNA silencing and leads to the down-regulation of GFP reporters that contain complementary target sequences (24). Because recent studies have implicated Dicer in the assembly and function of the silencing complex (26-30), we analyzed the ability of exogenously provided miRNA duplexes to repress target expression in MZdicer mutants. Wild-type and MZdicer mutant embryos were injected with GFP sensors for three different miRNAs and with either complementary or unrelated miRNA duplexes (Fig. 2, A and B, and fig. S4). We found that miRNA duplexes specifically repressed expression of their cognate targets in both wild-type and MZdicer mutant embryos (Fig. 2B and fig. S4). These results show that miRNAs can be incorporated into active silencing complexes in MZdicer mutants, indicating that the RNaseIII and dsRNA-binding domains of Dicer are not required for loading into the complex and subsequent silencing activity.

Dicer Is Essential for Embryonic Morphogenesis but Not Axis Formation

The absence of mature miRNAs in MZdicer mutants allowed us to determine their global requirement during early zebrafish development. The MZdicer phenotype notably differs from that of Zdicer mutants, which are indistinguishable from wild-type embryos during these stages (21) (Fig. 3, A to C). Morphological analysis during the first 5 days of development revealed that axis formation and the regionalization of MZdicer mutants were intact (Fig. 3C). Major subregions and cell types were present, ranging from forebrain, eye, midbrain, hindbrain, ear, pigment cells, and spinal cord to hatching gland, heart, notochord, somites, and blood (Fig. 3C and figs. S5 and S9). In contrast, morphogenetic processes during gastrulation, somitogenesis, and heart and brain development were severely affected (Fig. 3, C and E to G, and figs. S5 and S10). MZdicer mutants also developed more slowly than wild-type embryos, with 3 to 4 hours of delay within the first 24 hours of development (fig. S5) (31).

Fig. 3.

Morphogenesis defects in MZdicer mutants. Compared with wild-type embryos (A), Zdicer mutants (B) have no morphological defects at 36 and 90 hpf. (C) MZdicer mutants display morphogenesis defects. This phenotype is fully penetrant and expressive. (D) MZdicer injected at the one cell stage with miR-430a duplex. Brain morphogenesis (white bracket), and the midbrain-hindbrain boundary (*) are rescued, and trunk morphology and somite boundary formation are partially rescued. Phenotypes that are not rescued include the lack of blood circulation (black bracket), heart edema (arrow), and defective ear development (arrowhead). (E to G) (Left) Scheme representing the cell movements (arrows) during gastrulation in zebrafish: (1) epiboly, (2) internalization, (3) convergence, and (4) extension; (middle) wild-type embryo; (right) MZdicer mutants. (E) Embryos at 80% epiboly stage; arrow shows the similar extent of prechordal plate extension in wild-type embryos and MZdicer mutants; bracket shows a reduced extent in epiboly in MZdicer mutants compared with wild-type embryos. (F) Tail bud stage. Note the accumulation of cells in the region of the anterior axial mesendoderm (MZdicer, arrow and bracket). (G) Nine-somite stage. Note the abnormal development of the optic primordium (MZdicer, arrowhead) and reduction in axis extension that results in a shorter embryo (MZdicer, double arrow).

Gastrulation. During zebrafish gastrulation, four concomitant cell rearrangements take place: (i) epiboly (spreading of the embryo over the yolk, (ii) internalization (formation of mesodermal and endodermal germ layers, (iii) convergence (movement of cells toward the dorsal side), and (iv) extension (lengthening of the embryo) (Fig. 3, E to G). MZdicer mutants failed to coordinate epiboly and internalization (Fig. 3E). This resulted in mutant embryos that had undergone prechordal plate migration corresponding to 80% epiboly in wild-type embryos, yet epiboly movements were delayed to a stage equivalent to 50 to 60% epiboly (Fig. 3E). MZdicer embryos also displayed a reduced extension of the axis, resulting in a shortening of the embryo and an accumulation of cells in the head region (Fig. 3, F and G). Later during development, MZdicer mutants had a reduced posterior yolk extension (fig. S5).

Neural development. Neurulation was severely affected in MZdicer embryos. The mutant neural plate gave rise to the neural rod, but the subsequent formation of the neurocoel and neural tube was notably impaired (Fig. 4, A, B, E, and F, and fig. S5). The formation of the brain ventricles was severely reduced. In wild-type embryos, several constrictions subdivide the brain into distinct regions. These constrictions did not form in MZdicer mutants. For example, the midbrain-hindbrain boundary that is very prominent in wild-type embryos did not form in MZdicer mutants (Fig. 4, B and arrowhead in F, and fig. S5). In addition, retinal development was affected (Fig. 4, A and B). Defects in the spinal cord were manifested by a rudimentary neurocoel and a reduction of the floor plate in the trunk (fig. S5).

Fig. 4.

miR-430 miRNAs rescue brain morphogenesis in MZdicer embryos. (A to D) differential interference contrast (DIC) images of wild-type embryos (A), MZdicer mutants (B), MZdicer mutants injected with miR-430 duplex (MZdicer+miR-430) (C), and MZdicer mutants injected with miR-430b-mis duplex (MZdicer+miR-430b-mis) (D) that contains two mismatches in the seed of miR-430b. (E to H) Confocal dorsal view of embryos with the same genotype as in (A) to (D). Cell membranes were labeled in green (BODIPY) and the brain ventricles were labeled in red by injection of Texas-Red dextran into the brain. Wild-type embryos displayed the characteristic fold of the midbrain-hindbrain boundary (MHB) (arrowhead) and have brain ventricles (red) (E). MZdicer mutants do not form a midbrain-hindbrain boundary, lack normal brain ventricles, and display defects during eye development. [(C) and (G)] Injection of MZdicer mutants with the miR-430 duplex rescued brain development, including the midbrain-hindbrain boundary (arrowhead) and ventricle formation. Most (84%) of the embryos were rescued (n = 104). Eye development was also partially rescued. [(D) and (H)] Injection of the miR-430b-mis duplex, which contains two mismatches in the 5′ seed, did not rescue these defects (0% rescued, n = 50).

Despite the gross morphological malformations of the nervous system, gene expression analysis suggested that anterior-posterior and dorsal-ventral patterning were not severely disrupted (fig. S6). Analysis of anterior-posterior and dorsal-ventral markers revealed normal specification of the optic stalk, forebrain, midbrain-hindbrain boundary, otic vesicles, hindbrain rhombomeres, and the dorsal and ventral neural tube.

Analysis of neuronal differentiation and axonal markers, with the use of HuC and HNK antibodies, revealed mispositioned trigeminal sensory neurons adjacent to the eye (fig. S7). In addition, we observed defasciculation of the postoptic commissure in MZdicer embryos (fig. S7). In the hindbrain, multiple neurons project longitudinal axons anteriorly and posteriorly and form a ladder-like structure on each side of the midline. This scaffold was disrupted and defasciculated in MZdicer mutants, but longitudinal axonal projections were established (fig. S7). In addition, touch-induced escape behavior was severely diminished in MZdicer mutants (fig. S8). Taken together, these results indicate that early patterning and fate specification in the embryonic nervous system are largely unaffected by lack of miRNAs. In contrast, normal brain morphogenesis and neural differentiation and function require Dicer activity.

Nonneural development. During somitogenesis, the paraxial mesoderm becomes segmented. MZdicer embryos formed normally spaced somites and expressed the muscle marker myoD similar to wild-type embryos (fig. S9). Later in development, the somites acquired a chevron shape in wild-type embryos but formed irregular boundaries in MZdicer mutants (fig. S5). Endothelial and hematopoietic precursor cells were present as judged from the expression of the markers fli-1 and scl, respectively, but endocardial fli-1 expression was reduced and blood circulation disrupted in MZdicer mutants (fig. S9). Analysis of the markers pax2a, GFP-nanos-3UTR, fkd1, cmlc2, and fkd2 revealed that pronephros, germ cells, endoderm, cardiomyocytes, and liver cells, respectively, were specified (fig. S6) (25). MZdicer mutants had contractile cardiomyocytes but the two chambers characteristic of the wild-type heart did not form; instead, a tubular heart and pericardial edema developed (fig. S10).

Taken together, these results indicated that MZdicer mutant embryos were patterned correctly and had multiple specified cell types but underwent abnormal morphogenesis, in particular during neural development and organogenesis.

The miR-430 miRNA Family

To identify miRNAs that might play important roles during early zebrafish development, we cloned small RNAs (∼18 to 28 nt) from eight developmental stages between fertilization and 48 hours of development (32). These experiments identified miR-430a, miR-430b, and miR-430c as three highly expressed miRNAs, as well as several related species, miR-430d to miR-430h, which were expressed at lower levels (Fig. 5, A and B). The miR-430 family members each had the same sequence at nucleotides 2 to 8, which is known as the “seed” and has been shown to be the miRNA segment most important for target recognition (8, 13, 24, 33). The family members also have strong homology in their 3′ region, but differ in their central and terminal nucleotides. Mapping of the miR-430 family to the zebrafish genome revealed a locus composed of multiple copies of the miR-430a,c,b triplet, with more than 90 copies of the miRNAs within 120 kb (Fig. 5C). miRNA genes are sometimes observed in clusters of about two to seven, which are frequently transcribed as a single polycistronic transcript (34, 35), but the zebrafish miR-430 cluster has many more miRNAs than reported in other clusters. The miR-430 miRNAs are conserved and clustered in other fish genomes, including Fugu rubripes and Tetraodon nigroviridis (Fig. 5C). The miR-430 miRNAs belong to a superfamily that includes the vertebrate miR-17-miR-20 family, which are found in much smaller clusters in mammalian genomes (Fig. 5B). Despite the sequence similarities of the two families, members of the miR-17-miR-20 family derive from the opposite arm of their precursors, which suggest convergent rather than divergent origins of the two families. The miR-430 RNAs might share evolutionary origins with some of the miRNAs expressed specifically in mammalian embryonic stem cells (12), including miR-302 and miR-372, which have the same seed nucleotides and derive from the same arm of the hairpin.

Fig. 5.

Identification of a highly expressed miRNA family. (A) Predicted hairpins of three miR-430 miRNAs together with the corresponding duplexes used for injection; mature miRNAs shown in red. miR-430b-mis contains two mismatches (black) in the 5′ seed. (B) miR-430 miRNAs cloned from zebrafish (dre) and predicted in Fugu (fru) aligned with the miR-17-miR-20 family of human (hsa) miRNAs. (C) Color-coded representation of a miR-430 genomic cluster in the zebrafish and Fugu genomes. Each bar represents a predicted miRNA hairpin. (D) Northern blot analysis of the expression profile of miR-430a in wild-type embryos at different developmental stages.

The miR-430 miRNAs are initially expressed at about 50% epiboly [5 hours postfertilization (hpf)], continue to be expressed during gastrulation and somitogenesis, and then decline at about 48 hpf (Fig. 5D) (25). Analysis of GFP sensors with perfect target sites for miR-430a or miR-430b suggested that the miR-430 miRNAs are ubiquitously expressed and active during early development (Fig. 2B) (25).

miR-430 Rescues Brain Morphogenesis in MZdicer Mutants

As described above, miRNA duplexes are still active in MZdicer mutants. This allowed us to determine if aspects of the MZdicer mutant phenotype could be suppressed by providing specific miRNAs that are normally expressed during early zebrafish development (miR-1, miR-204, miR-96, miR-203, miR-430a, miR-430b, or miR-430c). We also reasoned that such rescue would unequivocally demonstrate that a particular phenotype is caused by the loss of a specific mature miRNA and not by the lack of small interfering RNAs (siRNAs) or the abnormal accumulation of pre-miRNAs in MZdicer mutants (5, 7, 20). We found that injection of miR-430 duplexes (miR-430a, miR-430b, or miR-430c) rescued the brain morphogenesis defects in MZdicer mutants (Figs. 3D and 4, C and G). This rescue was specific, as indicated by two control experiments. First, injection of unrelated miRNA duplexes did not cause any rescue (fig. S11) (25). Second, injection of a miRNA duplex with two point substitutions in the 5′ seed did not rescue the MZdicer phenotype (miR-430b-mis; Figs. 4, D and H, and 5A; fig. S11). Rescue of MZdicer mutant embryos by miR-430 (MZdicer+miR-430) resulted in normal brain ventricles and brain constrictions (Fig. 4, D and G, and fig. S11). For example, the midbrain-hindbrain boundary formed in MZdicer+miR-430 as in wild-type embryos (Fig. 4G and fig. S11). Injection of miR-430 also induced a substantial rescue of the neuronal defects observed in MZdicer mutants (fig. S7). MZdicer+miR-430 also displayed partially rescued gastrulation, retinal development, somite formation, and touch response (Figs. 3D and 4C and fig. S8). In contrast, the defects in the development of the ear and heart and the lack of circulation were not rescued (Fig. 3D and fig. S10). Later during development (90 hpf), MZdicer+miR-430 embryos were developmentally delayed and displayed reduced growth similar to MZdicer. These results indicate that loss of miR-430 miRNAs accounts for some but not all of the defects observed in MZdicer embryos.

Our study of zebrafish that lack Dicer RNaseIII activity and mature miRNAs provides three major insights into the roles of miRNAs during embryogenesis. First, our results suggest that mature miRNAs do not have widespread essential roles in fate specification or signaling during early zebrafish development. Phenotypic comparison between MZdicer mutants and embryos with aberrant signaling pathways (Nodal, Hedgehog, Wnt, Notch, CXCR4, FGF, BMP, retinoic acid, or STAT3) suggests that none of these pathways is markedly affected by the absence of miRNAs (36). For example, MZdicer mutants do not display the phenotypes seen upon an increase or decrease in Nodal or BMP signaling. This suggests that miRNAs might have modulating or tissue-specific rather than obligatory roles in various signaling pathways. Similarly, our study reveals that MZdicer mutants can differentiate multiple cell types during development. This suggests that mature miRNAs are not required to specify the major embryonic cell lineages in zebrafish. Our results do not exclude more specific roles in fate specification, such as modulating the choice between highly related cell fates. For example, lsy-6 in Caenorhabditis elegans controls the distinction between two closely related neurons, and mouse miR-181 seems to regulate the ratio of cell types within the lymphocyte lineage (18, 37). miRNAs might also function at later stages to stabilize and maintain a particular fate. For instance, miRNAs might repress large numbers of target mRNAs to maintain tissue homeostasis by dampening fluctuations in gene expression (38, 39). However, our transplantation results argue against an absolute requirement for miRNAs in every cell type. In particular, we generated fertile adults from MZdicer mutant donors by germ cell transplantation (fig. S1). This indicates that primordial germ cells, the ultimate stem cells, proliferate and remain pluripotent to form the adult germ line in the absence of miRNAs. Multigeneration transplantation studies are required to determine if the lack of miRNAs has effects on germ cell maintenance (40, 41). More exhaustive analysis of different cell types and signaling pathways is needed to test for more subtle or later roles of miRNAs in zebrafish, but our current study excludes a general role in signaling, embryonic fate specification, or germ line stem cell development.

Second, our results suggest important roles for miRNAs during embryonic morphogenesis and differentiation, ranging from epiboly and somitogenesis to heart, ear, and neural development. For example, loss of Dicer leads to defects in the positioning of neurons, the defasciculation of axons, and impaired touch-induced behaviors. Most notably, mutants form a neural rod but fail to generate normal brain ventricles. In addition, the morphological constrictions that subdivide the anterior-posterior axis do not form in the absence of Dicer, despite the regionalization observed by marker analysis. These results reveal essential roles of miRNAs during zebrafish morphogenesis.

Third, our study identifies a previously unknown miRNA family, the absence of which is likely to account for the brain morphogenesis defects in MZdicer mutants. The miR-430 family has more genes than any miRNA family described to date, is conserved in fish, and is part of a superfamily found in other vertebrates. Injection of miR-430 duplexes suppresses the brain morphogenesis defects in MZdicer mutants. This complementation approach can now be applied to determine which miRNAs (or siRNAs) account for the MZdicer phenotypes that cannot be rescued by miR-430. The miR-430 family might inhibit mRNAs that are provided maternally or expressed during early embryogenesis but are detrimental to later steps in morphogenesis. Cell shape changes, cell rearrangements, and fluid dynamics are thought to generate both extrinsic and intrinsic forces that contribute to neural tube and ventricle formation, but the underlying molecular mechanisms are poorly understood (42). The study of the miR-430 family and its targets therefore provides a genetic entry point to dissect the molecular basis of brain morphogenesis.

Supporting Online Material

Materials and Methods

Figs. S1 to S11


References and Notes

View Abstract

Stay Connected to Science

Navigate This Article