Skip to main page content
Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
, 550 (7674), 114-118

A Reversible Haploid Mouse Embryonic Stem Cell Biobank Resource for Functional Genomics


A Reversible Haploid Mouse Embryonic Stem Cell Biobank Resource for Functional Genomics

Ulrich Elling et al. Nature.


The ability to directly uncover the contributions of genes to a given phenotype is fundamental for biology research. However, ostensibly homogeneous cell populations exhibit large clonal variance that can confound analyses and undermine reproducibility. Here we used genome-saturated mutagenesis to create a biobank of over 100,000 individual haploid mouse embryonic stem (mES) cell lines targeting 16,970 genes with genetically barcoded, conditional and reversible mutations. This Haplobank is, to our knowledge, the largest resource of hemi/homozygous mutant mES cells to date and is available to all researchers. Reversible mutagenesis overcomes clonal variance by permitting functional annotation of the genome directly in sister cells. We use the Haplobank in reverse genetic screens to investigate the temporal resolution of essential genes in mES cells, and to identify novel genes that control sprouting angiogenesis and lineage specification of blood vessels. Furthermore, a genome-wide forward screen with Haplobank identified PLA2G16 as a host factor that is required for cytotoxicity by rhinoviruses, which cause the common cold. Therefore, clones from the Haplobank combined with the use of reversible technologies enable high-throughput, reproducible, functional annotation of the genome.

Conflict of interest statement

The authors declare no competing financial interests.


Extended Data Figure 1
Extended Data Figure 1. Properties of the haploid subclone AN3-12.
e-f, Various parthenogenic cell lines derived from independent embryos from an outcross of 129/Sv and C57/B6 and thus containing different genomic background on different chromosomes were subjected to embryoid body formation by placing 1000 cells/hanging drop. We observed downregulation of pluripotency marker genes (a) and upregulation of markers from all three germ layers (b) in all cell lines assayed on day 0 (d0), day 5 (d5) and day 12 (d12). The HMSc2 subclone AN3-12 was chosen for further study based on its growth properties in serum/LIF and absence of feeders. Data is shown as individual data points of n=2 technical replicates together the mean ± SD of one representative experiment. c, Growth curve of AN3-12 in the presence and absence of LIF. Data are shown as individual data points and mean values of 3 biological replicates ± SD. d, FACS analysis of chromosome content of AN3-12 cells (in LIF, same experiment as shown in panel c) shows the decline of haploid (1n) from 35.5% to 24.9% during the 7day culture period. e, AN3-12 cells, cultured as in panel c, maintain a robust haploid population when analyzed on day 17 in ES cell medium despite rapid proliferation. f, Differentiation of AN3-12 cells into keratinocytes resulted in a near complete loss of haploid cells among the Keratin 14 (K14) positive population; ES cells stained with anti-K14 are shown as a negative specificity control (grey curve in the K14 histogram).
Extended Data Figure 2
Extended Data Figure 2. Analysis of genome integrity.
a, M-FISH karyotypic analysis was performed on parental mouse haploid cells (AN3-12) to evaluate genomic stability. Randomly selected metaphases were karyotyped and examined by M-FISH and DAPI banding. Approximately, 200 metaphases from AN3-12 were counted for the diploid versus haploid frequency and 10 well-spread metaphases were fully karyotyped by M-FISH and DAPI-banding pattern. Images of a normal female diploid and haploid karyotype [19, X] are shown. Images were captured on a Zeiss AxioImager D1 fluorescent microscope equipped with narrow band-pass filters for DAPI, DEAC, FITC, CY3, TEXAS RED, and CY5. b, CNV (copy number variation) analysis of haploid AN3-13 cells by genome sequencing using IlluminaHiSeq2500. Mapped reads were analyzed relative to male genomes of parental mouse strains C57BL/6J and 129/Sv respectively, quotient to closer parental strain is shown. As expected, the X chromosome is overrepresented while the Y chromosome is absent. Regions of detected variation are highlighted with red boxes and shown below. Chromosome numbers are indicated. c, In AN3-12 haploid ES cells three very small deletions (on chromosomes 2, 10, and 12) and 1 duplication (on chromosome 13) were detectable as highlighted. d, Chromosomal distribution of SNP densities for in-house 129/SV and AN3-12 ES cells relative to in-house C57BL/6 are shown. Numbers of SNPs were calculated for all non-overlapping 100 kb windows across the mm10 C57BL/6J mouse reference genome. SNP density in AN3-12 shows regions of high and low number of SNPs relative to the C57BL/6J genome, as expected for a haploid cell line derived from an F1 female between 129/SV and C57BL/6.
Extended Data Figure 3
Extended Data Figure 3. Differentiation potential of AN3-12 cells.
a, Immunostaining of AN3-12 cells cultured in ES cell medium as well as time course of removal of LIF with addition of 500nM retinoic acid (analysed on the indicated days) shows downregulation of pluripotency markers Oct4, Nanog, as well as Sox2. DAPI is shown as a nuclear counterstain. Bar graphs 50μm b, Histological examination of Teratomas analysed 25 days after injection of 106 cells subcutaneously. All three germ layers were present in 6 analyzed teratomas, representative H&E images are shown. Magnifications are indicated in each panel.
Extended Data Figure 4
Extended Data Figure 4. Molecular characterization of mutagenesis vectors.
a, Schematic illustration of the universal NGS sequencing strategy. Optimized primer binding sites compatible with Illumina sequencing and 2 restriction enzymes with 4 base pair recognition sites were placed adjacent to the terminal elements (LTR, TR). An internal barcode of 32 bases with alternating weak and strong bases was inserted in a parallel cloning step. b, For mapping of integration sites, genomic DNA was amplified by inverse PCR to introduce adaptor sequences and the experimental index for NGS. Paired-end sequencing maps the genomic integration in the first read using a custom primer, the experimental index as well as the internal barcode using standard Illumina primers binding to the integrated complementary sequence. Barcode (BC) PCR was performed on genomic DNA. c, Meta-analysis of mutagen integrations around transcriptional start sites (TSS) (excluding the precise TSS site). In particular Tol2 and Retrovirus show a preference to integrate in proximity to the TSS. Retroviruses also frequently integrate into the promotor regions, while lentiviral integrations are typically located within the entire gene body. IPKM= Insertions per kilobase per million. The vectors used are described in Fig. 1 legend. d, Distribution of integration sites. Binning the number of integrations in genic and 2kb upstream regions per 10kb windows illustrates pronounced cold spots of mutagenesis using retroviral mutagenesis, where one can observe bins devoid of integrations. e, Genomic region surrounding the Gapdh locus exemplifying the distributions of integrations. While retroviral integrations strongly cluster, Tol2 displays a more uniform distribution of integration sites. Tracks are + strand (top) and – strand (bottom) integration sites. Bar lengths indicate NGS read numbers, subsequent to iPCR. f, Heat map illustrating overlap of epigenetic histone marks with integrations of the indicated mutagens, normalized to peak size. Only retrovirus and Tol2 integrations strongly correlate with DNA accessibility determined by ATACseq and active marks such as H3K4me3 and H3K27ac. In silico mutagenesis is shown as a control.
Extended Data Figure 5
Extended Data Figure 5. Insertional preferences.
a, Correlation of integration probabilities (IPKM=insertions per kilobase per million) to expression level (mean log2 of FPKM). Strongest correlation is seen for lentiviral constructs as well as Retro-GT without osteopontin enhancer elements. All mutagenesis vectors are described in Fig. 1 and methods. b, 5’RACE on a set of pooled clones with confirmed antisense integration sites revealed multiple spurious transcription initiation sites in the intronic part of the gene trap vector around the lox site, but we failed to detect spliced transcripts. Transcriptional initiation within the lox5171 site is highlighted. Red labelled sequence is marking polyGs used for 5’ tailing. c, Intersection of integration sites of the indicated mutagenesis vectors (see Fig. 1) with genomic features. Coding sequences (CDS), 5’ and 3’ untranslated regions (5’UTR and 3’UTR), 1st Intron, all other introns excluding the first intron (Intron), non-coding exons (ncExon), upstream regions (defined as 2kb upstream of transcriptional start site), and intergenic regions are indicated. Mutagenesis by Piggyback transposons as well as in silico random mutagenesis and ATACseq results are shown for comparison.
Extended Data Figure 6
Extended Data Figure 6. Generation of the mutant ES cell library.
a, Schematic work flow for generation of the mutant haploid ES cell library. Single cell derived clones were manually picked 10 or 11 days after seeding, expanded in 96 well plates, and either frozen in quadruplicates or further processed for mapping of the integration sites. b, Schematic illustration of the first step of 4D pooling. Each plate was pooled into the respective slice tray as well as a master-plate, uniting identical well coordinates of all plates. c, Schematic illustration of the second step of 4D pooling. Each master-plate was pooled into a master tower pool, a plate with lamella uniting columns, and a plate with lamella uniting columns, thereby generating pools for rows and columns over all samples. d, 4-Dimensional pooling of 9600 clones in 8 rows, 12 columns, 10 slices, and 10 towers resulting in 40 pools. Subsequent to iPCR to introduce experimental indices, pools were combined and deep sequenced. Amplification of internal barcodes confirmed clonal identity and mapping in 4 dimensions. All mapped clones were deposited to Haplobank (
Extended Data Figure 7
Extended Data Figure 7. Numbers of independent gene trap clones.
a, Numbers of independent available cell lines, carrying a single integration per cell, per gene. For about 37% (RefSeq) to 38% (Ensembl) of genes targeted, there is one gene trap clone available (5’UTR, Intron, or coding sequence), whereas about 18% of genes are targeted in two independent clones, and for ~ 43% of genes 3 or more independent clones are available. b, 24.8% (RefSeq) to 26.8% (Ensembl) of genes are represented by a single cell line if one takes all clones into account and about 40% of genes are hit in more 3 or more clones. c, Separation of all gene traps combined into biotypes in single integration Haplobank clones. Antisense and intergenic insertions are observed in all systems, in particular for enhanced gene trap vectors.
Extended Data Figure 8
Extended Data Figure 8. Distribution of integration sites relative to coding sequences.
To map the integration sites of our Haplobank clones to the ORFs (open reading frames) of the respective genes dissected ORFs into 5% intervals and annotated integration sites in introns and exons relative to the position within the ORF. All mutagenesis systems (see Fig. 1) show a strong bias towards transcript truncation proximal to the 5’end of the ORFs and are thus predicted to result in loss of function alleles. We defined integrations in the anterior 50% of coding sequence (green bars) as optimal for a gene trap allele; these clones are highlighted by a yellow star on the Haplobank homepage.
Extended Data Figure 9
Extended Data Figure 9. Interaction of Pla2g16 with Cox inhibitors in mouse ES cells.
a, Titration series of the indicated Cox inhibitors in the presence and absence of rhinovirus (RV-A1a) in mouse ES cells. No protective effect of inhibition of prostaglandin biosynthesis was detected at non-toxic concentrations. Since ES cells do not generate infectious RV-A1a efficiently, conditioned supernatant containing RV-A1a was added daily. Data is shown as individual data points and mean values. b, Haplobank ES cell clones harboring mutations in Ldlr and Pla2g16, respectively, were mixed as sister cells in sense (red) and antisense (green) orientation labelled by GFP and mCherry. Subsequently, cells were cultured in the presence and absence of rhinovirus RV-A1a for 4 days and ratios were then quantified using FACS. Selection pressure for loss of Ldlr and Pla2g16 was not affected by inhibition of Cox. Data is shown as mean of 3 biological replicates, error bars represent SD.
Extended Data Figure 10
Extended Data Figure 10. Interactions of PLA2G16 with Cox inhibitors in human HEK293T cells.
a, RV-A1a exposure causes cell death in HEK293T cells in a dose dependent manner. Cell viability was quantified 3 days after infection by Alamar blue cell viability measurement. Data is shown as individual data points of 5 biological replicates and mean values. b, Titration series for ibuprofen and indomethacin treatment in the presence and absence of rhinovirus (RV-A1a) in human embryonic kidney HEK293T cells. Protective effects of ibuprofen and indomethacin were detected at high concentration. Cell viability was quantified 2.5 days after infection by Alamar blue cell viability measurement. Data is shown as individual data points of 4 biological replicates and mean values. c, Competitive growth assays in HEK293T cells. Cells containing sgRNAs targeting PLA2G16 did not show a growth difference in the absence of RV-A1a or when treated with indomethacin at 100μM, but were significantly enriched when challenged with RV-A1a, indicting preferential survival. By contrast, control guide treated cells did not show growth advantages at any experimental condition. Data are shown as individual data points and mean values +/- SD of biological triplicates analysed on days 0, 3, 6, and 10 after RV-A1a exposure.
Extended Data Figure 11
Extended Data Figure 11. Pla2g16 domain mapping in rhinovirus infections.
a, Scheme of Pla2g16 domains. The enzymatic center of Pla2g16 is located in the cytoplasm (green); an alpha-helix in the transmembrane domain (yellow) connects it to a short vesicular domain (blue), located in endosomes14. b, Design of CRISPR sgRNAs targeting the mRNA regions encoding the vesicular domain of 7 amino acids (sgRNA1) and the 3’UTR (sgRNA2) in haploid ES cells to test essentiality of these domains in RV-A1a infections. c, Cells carrying sgRNA2 showed editing in only 1/12 cases, but upon selection with RV-A1a were enriched for deletions within the vesicular domain. For sgRNA2, all mapped deletions in control cells only affected the 3’UTR, where the expected Cas9 cuts occur; upon RV-A1a exposure, the majority of observed deletions affected the transmembrane domain, the vesicular domain, and in some cases even extended into the cytoplasmatic region. Color codes: Grey: deletion; Red: alternative reading frame and insertions.
Extended Data Figure 12
Extended Data Figure 12. Increased blood vessel sprouting in Notch1 mutant ES cells.
a, Assessment of 4 independent Notch1 targeted clones from Haplobank. The locations of the integrations are shown: 2 anti-sense (as) clones marked by green triangles, 1 sense (s) clone marked by the red triangle, and one clone with an upstream (as-up) integration (blue triangle). Flipping of the gene traps upon Cre infection is shown by PCR in the middle panel. Loss of Notch1 protein (intracellular domain, ICD) expression (clones A4, H7), and re-expression (clone D5) upon Cre recombination are shown by Western blot (lower panel). β-actin is shown as a loading control. WT, parental clone without any gene trap integration. b, Notch1 inactivation leads to a hyper-sprouting phenotype. Note the advanced progression and increased density of the vascular networks upon Notch1 deletion (sense clone) compared to anti-sense sister cells (upper panels – bright field images, lower panels IB4 immunostaining to mark endothelial cells). Scale bars 500μm. c, Angiogenic sprouting is not affected when the gene trap is located 1500bp upstream of the Notch1 gene (A2 clone). For molecular characterization of the A2 clone see Extended Data Fig. 12. GFP+ and Cre-reverted mCherry+ sister cells were analyzed in 3D blood vessel organoid cultures. Bright field images are shown. Scale bars 500 μm.
Extended Data Figure 13
Extended Data Figure 13. Candidate tip cell genes and clonal variability.
a, Differentially expressed genes in endothelial tip cells versus stalk cells from two published datasets15,16 in the murine retina were filtered for genes that have also been associated (Ingenuity pathway analysis) with candidate genes/pathways for vascular diseases in humans. Scatterplot showing the frequency of independent associations of “tip cell genes” with various human vascular diseases. Genes available at Haplobank at the beginning of the project were chosen for functional analysis in the 3D organoids. For most of the listed candidate genes, there were no functional vascular data available. b, Quantification of IB4 positive vascular structures from the indicated sister clones carrying sense and repaired antisense integrations. Clones were classified according to their sprouting capacity from low (hypo-sprouting) to high (hyper-sprouting).Data are shown as individual data points from a minimum of n=3 independent experiments for each sense/antisense sister clone combination together with the mean values ± SEM. * P < 0.05; ** P < 0.01; *** P < 0.001 (Two tailed students t-test). c, Different clones with independent integrations in the same gene showed reproducible phenotypes in sprouting angiogenesis. Vascular outgrowths were stained for endothelial specific IB4 expression, number of vessels counted and normalized to the respective anti-sense sister clones. Data are shown as individual data points from a minimum of n=3 independent experiments for each sense/antisense sister clone combination together with the mean values ± SEM. * P < 0.05; ** P < 0.01; *** P < 0.001 (Two tailed students t-test).
Extended Data Figure 14
Extended Data Figure 14. Sprouting angiogenesis in revertible sister clones.
Representative images of the indicated sense (S) and anti-sense (AS) sister clones. IB4 was used to mark endothelial cells. GFP or mCherry expression indicates the respective flipped gene traps. Note that some sense clones are GFP+ while others are mCherry+; this is due to the original orientation of the integration in sense or anti-sense, which was then reverted by the mCherry-Cre expressing virus. Scale bars for all images is 500μm. For quantification of data see Fig. 3e.
Extended Data Figure 15
Extended Data Figure 15. Generation of a chimeric vasculature in vivo.
a, Representative fluorescence image of a haploid ES cell-derived teratoma stained for endothelial specific IB4. Endothelial cells arising from haploid ES cells are positive for mCherry and IB4 (yellow), whereas host endothelial cells are only positive for IB4 and appear green. Scale bars 50 μm. b, Representative FACS analysis of teratomas following injection of chimaeric EBs into immunocompromised mice. Myst3 anti-sense (mCherry+) and sense (GFP+) sister clones were mixed at a 1:1 ratio. VE-Cadherin-negative non-endothelial cells were also determined within the teratomas. c, Parental haploid ES cells stably expressing GFP or mCherry-Cre were assessed for their ability to generate IB4+ vascular structures in the presence of VEGF-A. The number and ratios of IB4+ vessels per organoid were not apparently different between GFP and mCherry-Cre expressing cells. Scale bars 500μm. Data are shown as individual data points from n=4 independent experiments mean values ± SEM. P=0.207 (Two tailed students t-test). d, GFP and mCherry-Cre expressing parental haploid ES cells contribute equally to tip cells (49.2 % GFP+; 50.8 % mCherry-Cre+) in 1:1 mixed mosaic cultures. Data are shown as individual data points from of n=4 independent experiments together with the mean values ± SEM. P=0.823 (Two tailed students t-test).
Extended Data Figure 16
Extended Data Figure 16. Gja1/Connexin43 localizes to tip cells in the developing mouse retina and in ES cell-derived 3D blood vessels.
a, Localization of Gja1/Connexin43 protein in the murine retina at postnatal day 6 (P6). Endothelial cells are marked by IB4 staining. At the angiogenic front, Gja1/Connexin43 expression is found in endothelial cells, primarily localized at tip cells (arrows). Scale bars 50μm. b, Retinas were stained for Gja1/Connexin43 protein expression and the endothelial marker IB4 to visualize the vascular networks on postnatal day 6 (P6). Note punctate pattern of Gja1/Connexin43 adjacent to the IB4+ vessels, suggestive of Gja1/Connexin43 expression in perivascular cells. Scale bars 50 μm. c, Gja1/Connexin43 protein predominantly localizes to the tip cells (arrows) in our 3D blood vessels. Vessels are marked by CD31 immunostaining and counterstained by DAPI. Bar graph indicates percentages of vessels with highest Gja1/Connexin43 expression in the tip cell; Data is shown as individual data points of 8 independent EBs together with the mean values ± SD of vessels. Scale bars upper panels 20μm, lower panels 10μm.
Figure 1
Figure 1. A repairable mutant mES cell library.
a) Schematic representation of insertional mutagenesis vectors: Splice acceptor sites (SA) are revertible using non-compatible loxP/lox5171 and FRT/F3 sites (triangles). G418 resistance is conferred by beta-Geo (bgeo) transcribed from the revertible cassette (gene trap vectors, GT) or Neo independently from a PGK promoter (polyA trap, pA), stabilized by a splice donor (SD). Six osteopontin enhancer (OPE) elements (enhanced gene trap; Lenti-ETG, Retro-EGT, and Tol2-EGT vectors) enhance expression of beta-Geo via Oct4/Pouf51 binding. RetroRS carries a spacer sequence between loxP sites and lacks OPEs. Purple diamonds indicate internal barcodes (BC). LTR, long terminal repeats; L200/R175 and LITR/RITR, terminal repeats of Tol2 and SB. b) Heat map representing numbers of integrations per gene per 1 million integrations. Gene expression levels are shown (blue=highly expressed, white=not expressed). Color code shows numbers of integrations. c) Saturation of mutagenesis systems compared to random in silico mutagenesis. Y-axis, total numbers of insertions versus % of genes with integrations. d) Schematic representation of splice acceptor inversions. e) Loss of mESC adhesion in clones with integrations in intron 1 of Ctnna1. Inversion of the gene trap restores cell adhesion, subsequent reversion again disruptes adhesion. Phalloidin images polymerized actin; DAPI visualizes nuclei. Size bars, 10 μm. One representative experiment out of 2 biological replicates is shown.
Figure 2
Figure 2. Essential genes for mESC and common cold virus infections.
a,b) Functional annotation of essential mESC genes. a) Competitive growth assays of anti-sense (GFP+) and Cre-reverted sense (mCherry+) sister cells harboring integrations in the indicated genes. Cell populations were analyzed at the indicated days after Cre addition using flow cytometry. Means +/- SD of biological triplicates. b) FACS plots for the essential gene Psmd1 illustrating depletion of Cherry+ cells. c) Integration sites of top scoring genes in our haploid mESC survival screen of human rhinovirus RV-A1a infections. Loss of function score for integrations into the Ldlr locus p=2.9x1012 and p=1.4x1011 for Pla2g16. Sense integrations, red triangles; anti-sense integrations, green; exons, blue boxes. Transcriptional start sites are marked. d) Growth advantage of sense versus respective anti-sense sister mESC harboring integrations in Pla2g16 or Ldlr upon infection with RV-A1a. In un-infected cells, mutation of these genes did not confer growth advantages; arbitrarily set to 1. e) Human embryonic kidney HEK293T cells were transduced with 4 different sgRNAs against PLA2G16 and LDLR in biological triplicates, mixed with control GFP+ HEK293T cells at a ratio of 1:3. Ratios of control to mutated HEK293T cells were evaluated on day 13 after infection using FACS. Data in d and e are means +/- SD, normalized to uninfected cells. Individual data points and error bar (STDEV); one tailed students t-test, *< 0.1, **< 0.01, ***< 0.001. f) Targeting of the C-terminal Pla2g16 domain using CRISPR/Cas9. Upon selection of haploid cells to ensure hemizygous editing, cells were split and maintained in the presence and absence of RV-A1a.
Figure 3
Figure 3. Novel regulators of angiogenesis.
a, Generation of sprouting vasculature from haploid mESCs, differentiated into Embryoid Bodies (EBs) and cultured with VEGF-A (30ng/mL). b, CD31+ (green) endothelial cells and filopodia, indicative of tip cells in blood vessel sprouts. Luminal structures and Collagen IV+ basement membranes are shown in right panels. Scale bars, 200μm or as indicated. c, Schematic outline for functional validation of candidate genes in sprouting angiogenesis. Haplobank clones were infected with GFP or mCherry-Cre viruses to generate disruptive sense (S) and anti-sense (AS) sister cells. d, Representative images of hypo- and hyper-sprouting sense (S) and anti-sense (AS) sister clones. IB4 marks endothelial cells. Scales, 500μm. e, Quantification of IB4+ blood vessel sprouts. Data were normalized to the respective anti-sense sister clones. Means ± SEM are shown from a minimum of n=3 independent experiments. *<0.05; **< 0.01; ***< 0.001 (Two tailed students t-test).
Figure 4
Figure 4. In vivo angiogenesis and cell specifications.
a, Quantification of the indicated sense and anti-sense targeted clones to form blood vessels in teratomas. Means ± SEM are shown (n=3 independent teratomas; n=5 for the AN3-12 control). *< 0.05 (Two tailed students t-test). b, Gja1-sense (GFP+) and Gja1-antisense (mCherry+) sister mES cells were mixed 1:1 to form chimeric EBs and subsequently injected into immunocompromised nu/nu mice. Representative sections to identify IB4, GFP, and mCherry expressing cells are shown. Scale bar 50μm. c, Notch1 anti-sense and mCherry-Cre+ Notch1-sense sister mES cells were mixed (1:1 ratio) to generate mosaic blood vessels and analyzed for red or green cells at the tip position. Scale bar 200μm. d, Relative tip cell position of sister cells with sense and anti-sense integrations in the indicated genes, determined in chimeric 3D sprouts. Means ± SEM of a minimum of n=3 independent experiments. *< 0.05; **< 0.01; ***< 0.001 (Two tailed students t-test). e, Representative image of mosaic blood vessel sprouts from Gja1 sense and anti-sense sister clones. Scale bars, 500 μm and 100 μm (insert). f, Intravenous injection of a Gja1 inhibitory peptide (GAP26) into neonatal mice abrogated retinal angiogenesis. At day 5 after birth, retinas were isolated and stained for IB4+ blood vessels. Scale bars: 500μm upper, 100μm lower panels.

Similar articles

See all similar articles

Cited by 21 PubMed Central articles

See all "Cited by" articles

Publication types

MeSH terms