Homeostasis is essential for hematopoiesis, and its dysregulation can lead to severe pathological conditions. Retinoic acid (RA) is a key regulator that exerts concentration-dependent effects on both embryonic and adult hematopoiesis. However, the mechanisms that modulate RA signaling in hematopoietic processes remain poorly understood. Using zebrafish as a model, we identified angiopoietin-like protein 5 (Angptl5) as a critical regulator of hematopoietic homeostasis. Loss of Angptl5 function resulted in myeloid hyperplasia in the anterior lateral plate mesoderm (ALPM) and anterior expansion of erythroid progenitors in the posterior lateral plate mesoderm (PLPM)—phenotypes consistent with attenuated RA signaling. Molecular analyses confirmed impaired RA signaling in angptl5^Δ10/Δ10 mutants, and exogenous RA supplementation fully rescued the hematopoietic defects. Mechanistically, we found that Angptl5 transcriptionally activates retinol dehydrogenase dhrs9 through its interaction with Integrin α6lβ5. Our findings establish Angptl5 as a novel and essential regulator of embryonic hematopoiesis and reveal a previously unrecognized mechanism controlling hematopoietic homeostasis. These insights position Angptl5 as a potential therapeutic target for hematological disorders.

angptl5 is expressed in the primitive hematopoietic tissue during zebrafish embryonic development

Angiopoietin-like proteins are secreted glycoproteins conserved across vertebrates (S1A and S1B Fig). Members of this family feature two highly conserved domains: an N-terminal coiled-coil domain (CCD) and a C-terminal fibrinogen-like domain (FLD) (S1C Fig). To investigate their potential roles in development, we first analyzed the spatial and temporal expression patterns of angiopoietin-like genes during zebrafish embryogenesis using whole-mount in situ hybridization (WISH). Interestingly, different angiopoietin-like genes exhibited distinct tissue specificity. For example, angptl2b was expressed in the posterior spinal cord, angptl6 in the notochord, and angptl7 in the somites (S2A Fig). Among these, angptl5 was expressed in the mesendoderm during gastrulation and specifically enriched in hematopoietic tissues at 24 hours post-fertilization (hpf), including the rostral blood island (RBI) and posterior intermediate cell mass (ICM) (Figs 1A and S2B and S2C). To further confirm its expression, we generated a transgenic line Tg(angptl5:EGFP) by cloning the upstream 867 bp region of angptl5 gene and EGFP coding sequece, which revealed expression of angptl5 in the ALPM and partial co-localization with spi1b-positive cells (Figs 1B and S2D). Additionally, analysis of a single-cell RNA sequencing (scRNA-seq) dataset showed that angptl5 is expressed in a subset of spi1b-positive cells at both 18 hpf and 24 hpf (Fig 1C–1F) [20]. Together, these findings suggest that angptl5 may play a role in primitive hematopoiesis in zebrafish.

angptl5 mutation leads to hyperplasia of myeloid and erythroid progenitors

To investigate the function of Angptl5 in hematopoiesis, we generated two angptl5 mutant lines using CRISPR/Cas9: angptl5Δ10 (10 bp deletion) and angptl5Δ5 (5 bp deletion) (S3A Fig). Both mutations induced frameshifts producing premature stop codons, resulting in truncated proteins that either lacked the FLD(Angptl5Δ10) or both the CCD and FLD (Angptl5Δ5).

While expression of hemangioblasts marker tal1 appeared normal, angptl5Δ10 mutants displayed significant expansion of spi1b+ myeloid progenitors in the ALPM (Fig 2A), along with increased neutrophils (lyz+, mpx+) and macrophages (mpeg1.1+) (S3C Fig). This phenotype was consistently observed in angptl5Δ5 mutants (S3D and S3E Fig), suggests that loss of angptl5 function induces defects in primitive myelopoiesis. Based on these findings, the angptl5Δ10 mutant was selected for further functional analysis. Markedly, ectopic angptl5 mRNA expression effectively suppressed the aberrant expansion of spi1b+ myeloid progenitor in angptl5^Δ10/Δ10 embryos (Fig 2B). We then checked the expression pattern of gata1a+ erythroid progenitors. Interestingly, HCR co-staining with the somite marker myod1 revealed anterior expansion of gata1a (Fig 2C), a phenotype resembling attenuated RA signaling [10]. And quantitative real-time PCR (qPCR) also confirmed upregulation of spi1b and gata1a in angptl5^Δ10/Δ10 embryos, while tal1 and the angioblast marker etsrp remained unchanged (S3F Fig). Collectively, these data define a specific primitive hematopoietic defect in angptl5 mutants. Importantly, this defect was not associated with significant embryonic lethality under standard conditions, as determined by systematic survival analysis (S3G Fig).

To further characterize the hematopoietic phenotype caused by Angptl5 deficiency, we performed scRNA-seq on 16 hpf angptl5^Δ10/Δ10 mutant embryos and integrated these data with published WT scRNA-seq datasets [21]. While UMAP visualization revealed similar overall cell type distributions between mutant and WT embryos (Figs 2D, S4A and S4B), differential abundance analysis identified significant expansion of hematopoietic populations in mutants, including erythroid progenitors and myeloid progenitors (Fig 2E and 2F).

Angiopoietin-like proteins have been previously reported to play roles in angiogenesis [22,23]. To assess whether angptl5 affects vascular development, we first examined endothelial precursors during the critical stages of primitive hematopoiesis. The expression of key markers, including the vascular progenitor marker etsrp [24] and fli1 [25], in the ALPM and PLPM was comparable between mutant and WT embryos (S5A and S5B Fig). Then, to evaluate potential defects at later stage, we analyzed the endothelial marker kdrl [26] as well as etsrp again at 3 dpf and 3.5 dpf using WISH and transgenic reporter lines. No obvious vascular alterations were observed in angptl5^Δ10/Δ10 embryos at these later stages (S5C and S5D Fig).

Taken together, the main phenotypes observed in angptl5 mutants are myeloid hyperplasia in the ALPM and anterior expansion of erythroid progenitors in the PLPM, while the vasculogenesis remains largely unchanged.

angptl5 mutation leads to attenuated RA signaling

In addition to the hematopoietic defects in angptl5 mutants, characterized by elevated spi1b and anterior expansion of gata1a that resemble the phenotype observed upon RA signaling inhibition in zebrafish [10] and mirror vitamin A deficiency-induced promotion of hematopoietic stem cell differentiation in mice [27], we also observed mesoderm involution defects during gastrulation, as indicated by aberrant fn1a [28] expression detected via WISH (S6A Fig), reflecting mesoderm involution defects reminiscent of those observed under RA-deficient conditions [29]. Therefore, to determine whether RA signaling is affected in angptl5 mutants, we first examined the expression of RA direct target cyp26a1 [30], as well as two key enzymes required for RA biosynthesis: retinol dehydrogenase dhrs9 and aldehyde dehydrogenase aldh1a2 (S6B Fig). Interestingly, WISH and qPCR analyses revealed significantly reduced expression of cyp26a1 at 8 hpf and 18 hpf, aldh1a2 at 8 hpf, and dhrs9 at 8 hpf and 18 hpf in angptl5^Δ10/Δ10 embryos (Fig 3A–3C). To further validate these findings, we utilized the Tg(RARE:EGFP) transgenic zebrafish line, in which EGFP expression is driven by RA response elements, and thus serves as a direct readout of RA signaling activation [31]. We observed angptl5^Δ10/Δ10 embryos exhibited remarkably reduced EGFP fluorescence compared to WT embryos, providing further evidence of attenuated RA signaling in angptl5 mutants (Figs 3D and S6C).

To investigate the hierarchy relation between Angptl5 and RA signaling in hematopoiesis regulation, we treated WT and angptl5^Δ10/Δ10 embryos with RA inhibitor (DEAB, inhibitor of aldh1a2) or RA at varying concentrations starting from shield stage. As shown by the results, RA exhibited dose-dependent teratogenic effects, with concentrations ≥100 nM causing severe morphological abnormalities in WT embryos (S6D Fig) at 18 hpf. Pharmacological inhibition of RA markedly suppressed cyp26a1 expression, whereas exogenous RA administration robustly upregulated this RA target gene in a dose-dependent manner (S6E Fig). While RA treatment suppressed spi1b expression in all embryos, angptl5 mutants demonstrated significantly higher resistance to RA treatment, as evidenced by persistent spi1b and gata1a expression compared to WT controls (Figs 3E, 3F and S7A). Consistently, treatment with the RA receptor antagonist AGN 193109 (AGN) or DEAB led to a significant expansion of spi1b⁺ and gata1a⁺ cells in both WT and angptl5^Δ10/Δ10 embryos (S7B and S7C Fig).

The above results suggest that RA activation could compensate for Angptl5 deficiency. To further confirm the effect of restoring RA signaling on primitive hematopoiesis in angptl5 mutants, we performed rescue experiments by injecting dhrs9 or aldh1a2 mRNA at the 1-cell stage in both WT and angptl5^Δ10/Δ10 embryos. Overexpression of both dhrs9 and aldh1a2 enhances the cyp26a1 expression, indicating the elevated RA signaling (S7D Fig). Consistently, the expanded expression of spi1b⁺ myeloid and gata1a⁺ erythroid progenitors in angptl5 mutants was rescued (Fig 3G and 3H). Together, these results indicate that Angptl5 function as an upstream regulator of RA-mediated hematopoietic patterning.

Angptl5 physically interacts with integrin α6lβ5

As a secreted protein, Angptl5 requires receptor binding to exert its biological functions. To comprehensively identify receptors through which Angptl5 regulates RA signaling, we employed TurboID-mediated proximity labeling combined with immunoprecipitation-mass spectrometry (IP-MS) [32] in zebrafish. We engineered a construct expressing angptl5 mRNA fused with a C-terminal Flag-tagged TurboID. Secretion of engineered proteins was detected in HEK293T conditioned medium and the expression of tagged angptl5 and biotinylated proteins was detected in zebrafish embryos (S8A and S8B Fig). The engineered angptl5 mRNA exhibited similar rescue efficiencies to untagged controls, indicating that the fusion tags do not compromise Angptl5 protein functionality (S8C Fig). We then injected engineered WT and mutated angptl5 mRNA into 1-cell stage embryos, and incubated those embryos with biotin for 8 hours before collecting them for IP-MS, respectively (Fig 4A).

Comparative proteomic profiling of WT versus mutant samples, we identified integrin α6-like (Itga6l) as a potential candidate, based on its specific interaction with WT Angptl5 but not the Angptl5Δ10 mutant (Fig 4B and S1 Dataset). This aligns with established literatures demonstrating that ANGPTL family members engage various integrins-conserved transmembrane proteins that regulate biochemical signaling and mechanotransduction via force-dependent conformational changes [13,33]. Structural modeling of the Angtpl5-Itga6l complex using AlphaFold3 also revealed a considerable interface area for the WT Angptl5 (Fig 4C, up) and co-immunoprecipitation (Co-IP) assay confirmed that Angptl5 can physically associate with Itga6l (Fig 4D). Notably, overexpression of angptl5 mRNA in angptl5^Δ10/Δ10 embryos effectively suppressed the pathological expansion of spi1b⁺ myeloid progenitors, but this rescue phenotype was abrogated by co-injection with itga6l morpholino (MO), indicating that Angptl5-mediated regulation of primitive hematopoiesis is dependent on Itga6l function (S8D Fig). Conversely, the Angptl5Δ10 exhibited a reduction interface area, indicating a reduced binding affinity for the truncated Angptl5 (Fig 4C, down). However, to our surprise, the Co-IP assay showed that both Angptl5 and Angptl5Δ10 can physically associate with Itga6l (Fig 4D). These findings suggest that the loss-of-function phenotype cannot be attributed solely to disrupted binding of Angptl5 to Itga6l.

Functional integrin signaling requires the formation of α/β heterodimeric complexes at the cell membrane [34]. We thus speculate that the mutation of Angptl5 may have disrupted the formation of this potential complex. To identify the β subunit partner for Itga6l in this context, we systematically evaluated potential candidates. Bioinformatic screening of existing transcriptome datasets revealed differential expression profiles across 12 integrin β subunit genes in zebrafish [35]. Among these, five subunits (itgb1a, itgb1b, itgb4, itgb5, and itgb6) showed detectable expression during primitive hematopoiesis (S9A Fig). Thus, we performed structural modeling of the complexes formed between Itga6l and each individual β subunit. The Itga6l-Itgb4 and Itga6l-Itgb5 complexes exhibited the two largest interfacial contact areas among all Itga6l-β subunit complexes (Figs 4E and S9B–S9F). However, the Angptl5-Itga6l-Itgb4 ternary complex failed to maintain structural stability due to incomplete pairwise interaction interfaces (S9G Fig). Instead, structural modeling demonstrated that Itgb5 engages in an extensive interaction with Itga6l, burying a surface area of 2695.6 Ų at their interface (Fig 4E), and this is further confirmed biochemically by Co-IP (Fig 4F). WISH analysis showed that both itga6l and itgb5 are expressed during early development (S9H and S9I Fig).

We therefore characterized the putative Angptl5-Itga6l-Itgb5 ternary complex. Structural modeling revealed that WT Angptl5 formed stable ternary interactions through extensive interfacial contact surfaces, whereas the Angptl5Δ10 mutant exhibited disrupted binding interfaces with insufficient pairwise interaction networks (Fig 4G and 4H). Intriguingly, Co-IP assays demonstrated that Angptl5 cannot bind Itgb5 directly in the absence of Itga6l (Fig 4I). However, when Itga6l is present, WT Angptl5, but not the Δ10 mutant, mediates formation of a stable complex with Itgb5 (Fig 4I). These findings demonstrate that Angptl5 facilitates assembly of Itga6l-Itgb5 heterodimers, and that this molecular activity is abolished in Angptl5Δ10 mutant.

To elucidate the functional role of the Angptl5/Integrin α6lβ5 complex in zebrafish primitive hematopoiesis, we performed loss-of-function and gain-of-function analyses. In WT embryos, individual knockdown of itga6l or itgb5, or pharmacological inhibition of integrin function, was sufficient to increase the expression of the primitive myeloid marker spi1b and the erythroid marker gata1a, phenocopying the angptl5 mutant (S10A Fig). Conversely, either angptl5 mRNA injection or co-injection of itga6l and itgb5 mRNAs effectively suppressed spi1b expression, while angptl5Δ10 mRNA had no significant effect in WT embryos (S10B Fig). Notably, simultaneous overexpression of all three components (angptl5, itga6l, and itgb5) led to a further reduction in both spi1b and gata1 expression (S10C and S10D Fig). Most importantly, in angptl5^Δ10/Δ10 mutants, this triple overexpression (angptl5 + itga6l + itgb5) significantly reduced both the myeloid and erythroid hyperplasia (Fig 4J and 4K).

Collectively, these results suggest that integrin α6lβ5 acts as the cognate receptor for Angptl5 in hematopoietic regulation, and that the Angptl5-mediated suppression of primitive hematopoietic progenitor cells is mechanistically dependent on integrin α6lβ5 signaling.

ERK signaling is required for Angptl5-mediated regulation of primitive hematopoiesis

Integrin engagement by extracellular ANGPTL proteins typically activates intracellular signaling cascades, including FAK, NF-κB, and MAPK/ERK pathways, which regulate diverse cellular processes such as motility, proliferation, and transcriptional regulation [36–38]. To delineate the downstream effectors of Angptl5/α6lβ5 signaling in hematopoietic development, we systematically inhibited key pathways using small-molecule inhibitors from the shield stage onward, including: Defactinib (FAK), BAY 11–7082 (NF-κB), Adezmapimod (p38 MAPK), and Mirdametinib (ERK). Notably, FAK inhibition showed no significant effect on spi1b expression in the RBI, suggesting minimal involvement in primitive myeloid regulation (S11A Fig). And NF-κB blockade downregulated spi1b expression, revealing its role in promoting myeloid expansion (S11B Fig). Interestingly, MAPK or ERK inhibition significantly upregulated spi1b expression, and reduced cyp26a1 expression, phenocopying angptl5 loss-of-function mutants (Figs 5A, 5B, and S11C). Most notably, pharmacological activation of ERK signaling (C16-PAF) effectively rescued the spi1b+ myeloid hyperplasia in angptl5^Δ10/Δ10 mutants (Fig 5C).

The above results suggest that angptl5 may regulate primitive hematopoiesis through the ERK signaling pathway. To further validate this hypothesis, we generated WT ERK2 and a constitutively activated ERK2 mutant (caERK2, L84P/S162D/D330N) [39] (S11D Fig), and injected mRNA of these two versions to rescue the phenotype resulting from angptl5 loss-of-function, respectively. While WT ERK2 had no obvious effect on spi1b expression, 10 pg of caERK2 rescued spi1b expression in angptl5^Δ10/Δ10 embryos to levels comparable to the WT, and higher doses of caERK2 abolished spi1b expression in most angptl5^Δ10/Δ10 embryos and in all WT embryos (Figs 5D and S11E). Mechanistically, we found that ERK signaling activity was significantly decreased in angptl5^Δ10/Δ10 embryos compared to WT, as indicated by western blot and immunofluorescence (IF) of phosphorylated ERK1/2 (pERK1/2) (Figs 5E and S11H). Further spatial analysis of pERK distribution in hematopoietic regions revealed partially co-localization of pERK signal with spi1b+ myeloid progenitors in the ALPM (S11F Fig). In addition, pERK staining was detected within the ICM, a posterior blood-forming tissue derived from the PLPM, in WT embryos (S11G Fig). These data indicated that active ERK signaling was observed in both major primitive hematopoietic progenitor domains.

Next, we investigated whether the Angptl5/Integrin α6lβ5 complex is sufficient to active ERK signaling. We found that overexpression of angptl5 alone, co-overexpression of itga6l and itgb5, or combined overexpression of angptl5 with itga6l and itgb5 enhanced ERK signaling (S11I Fig). To further validate this, we co-injected angptl5, itga6l, and itgb5 mRNA into one blastomere of the animal pole at the 128-cell stage, and used fgf8a, a known canonical ERK activator [40] as a positive control (Fig 5F). Strikingly, combinatorial overexpression of angptl5, itga6l, and itgb5 triggered ectopic ERK activation, similar to the activation pattern observed with overexpression of fgf8a (Fig 5G and 5H).

In summary, these findings delineate a pathway in which Angptl5 engages integrin α6lβ5 to activate ERK signaling, a process required for the regulation of primitive hematopoiesis by Angptl5.

Angptl5/Integrin α6lβ5 potentiates RA activity via ERK-dependent dhrs9 transcription

Building on our previous characterization of RA signaling in Angptl5-dependent hematopoiesis, we investigated whether the Angptl5/Integrin α6lβ5-ERK axis can directly modulates RA activity. Using cyp26a1 as a readout of RA pathway activation, we found that ERK inhibition significantly reduced the cyp26a1 expression in the WT embryos, while ERK activation drastically enhanced cyp26a1 expression. Interestingly, triple overexpression of angptl5, itga6l, and itgb5 at one cell stage also dramatically increase the cyp26a1 expression. Conversely, CRISPR-Cas9-mediated knockdown of itga6l and itgb5 significantly reduced cyp26a1 expression (S13A Fig). Importantly, this Angptl5/Integrin-mediated cyp26a1 induction was completely abolished by ERK inhibition, demonstrating that the Angptl5/integrin complex enhances RA signaling through ERK-dependent mechanisms (Fig 6A).

To further explore the mechanism by which the Angptl5/Integrin α6lβ5-ERK axis regulates RA activity, we injected caERK2 mRNA or angptl5, itga6l, and itgb5 mRNA together into one blastomere at the 128-cell stage, followed by analysis of the expression of RA synthases aldh1a2 and dhrs9 (Fig 6B). Interestingly, both treatments induced dhrs9 expression at the animal pole, but not aldh1a2 (Fig 6C and 6D). To systematically dissect the activating effects of these components on dhrs9 transcription, we performed a two-stage microinjection in zebrafish embryos (Fig 6E). Our results showed neither angptl5 nor itga6l was sufficient to induce dhrs9 expression. Intriguingly, itgb5 alone, or together with angptl5, caused a slight induction of dhrs9 expression. Co-injection of itga6l and itgb5 promoted a moderate transcriptional activation of dhrs9, and this activation was dramatically boosted by the additional overexpression of angptl5. Critically, this synergistic activation was completely abolished upon pharmacological ERK inhibition (Fig 6F and 6G), suggesting its dependence on ERK signaling.

To determine the specific upstream kinase responsible of ERK activation [41], we performed a targeted inhibitor screen. Our results revealed that ERK activation downstream of the Angptl5-Integrin α6lβ5 complex is dependent on Src family kinase activity, as it was abolished by Src inhibition (S12F Fig). In parallel, we also tested two alternative mechanisms: first, whether the complex signals through the canonical integrin effector FAK; and second, whether it facilitates ERK activation indirectly by enhancing EGFR signaling. Inhibition of either FAK or EGFR had no effect on Angptl5/Integrin α6lβ5 mediated ERK activation (S12B and S12E Fig), indicating that neither protein was required for the regulatory axis.

Finally, to investigate how ERK signaling activates dhrs9 transcription, we performed a bioinformatic analysis of the dhrs9 promoter region, which revealed three conserved, high-probability binding sites for the transcription factor Junb—a canonical nuclear target of the ERK/MAPK pathway [42]—predicted based on UCSC Genome Browser and JASPAR database analyses. Knockdown of junba/junbb completely abolished the ectopic induction of dhrs9 by caERK2 (Fig 6H). Chromatin immunoprecipitation (ChIP) assays further confirmed the specific enrichment of Junb at these predicted sites in vivo (Fig 6I).

In summary, these findings indicate a complete signaling transduction cascade: Angptl5, together with Itga6l/Itgb5, activates a Src-dependent ERK pathway, resulting in Junb-dependent transcriptional activation of dhrs9, thereby modulating RA activity. This model was further supported by the observation that the RA signaling potentiation caused by co-overexpression of angptl5, itga6l, and itgb5 in angptl5^Δ10/Δ10 embryos was remarkedly attenuated by ERK inhibition (S14A Fig).

Next, we validated the regulatory role of the Angptl5/Integrin α6lβ5-ERK-Junb-RA signaling axis in primitive hematopoiesis. Co-injection of itga6l and itgb5 mRNA into angptl5^Δ10/Δ10 embryos partially attenuated myeloid hyperplasia, but this suppression was reversed upon ERK inhibitor treatment, revealing that the regulatory function of integrin α6lβ5 in primitive hematopoiesis is mechanistically reliant on ERK signaling activation (Fig 6J). In contrast, caERK2 overexpression completely abolished spi1b expression in angptl5^Δ10/Δ10 embryos, but this inhibitory effect was counteracted by RA inhibitor treatment, suggesting that ERK signaling-mediated regulation of primitive hematopoiesis requires RA activity (Fig 6K).

Taken together, these findings establish Angptl5 as a critical regulator of zebrafish primitive hematopoiesis, operating through the Angptl5/Integrin α6lβ5-ERK-Junb-RA signaling axis to suppress excessive expansion of hematopoietic progenitor cells.

Fig 7. Working model of Angptl5-mediated regulation of primitive hematopoiesis in zebrafish.

During primitive hematopoiesis, Angptl5 (but not the truncated Angptl5Δ10) specifically promotes the assembly of Itga6l and Itgb5 subunits into functional integrin α6lβ5 heterodimers, thereby activating downstream ERK signaling pathways. Activated ERK2 then induces dhrs9 expression via the transcription factor Junb, which binds directly to the dhrs9 promoter. This upregulation of dhrs9 potentiates RA signaling, ultimately suppressing excessive expansion of myeloid and erythroid progenitors in primary hematopoiesis.

https://doi.org/10.1371/journal.pbio.3003858.g007

Ethics statement

All fish maintenance and experimental procedures were performed in strict accordance with the Regulations for the Administration of Affairs Concerning Experimental Animals and the National Standard of the People’s Republic of China for the Care and Use of Laboratory Animals (GB/T 35892-2018). The study protocol was approved by the Animal Care and Use Committee of Zhejiang University (Protocol No. ZJU20220375).

Zebrafish husbandry and genetic strains

Zebrafish were raised on a 14- hours-light/10-hours-dark cycle at 28 °C [54] and staged as described [55]. The following lines were used in this study: wild-type (WT) AB strains, Tg(angptl5:EGFP), Tg(RARE:EGFP), Tg(kdrl:GFP), angptl5^Δ10/Δ10, and angptl5^Δ5/Δ5 mutants. All fish maintenance was performed per the requirements of the “Regulation for the Use of Experimental Animals in Zhejiang Province”, with the approval by the Zhejiang University Animal Care and Use Committee. The transgenic line Tg(spi1b:EGFP) [56] was obtained from Shanghai Institute of Hematology at Ruijin Hospital, SJTU.

WISH and histological sections

WISH of zebrafish was performed essentially as described previously [57]. Probe templates including tal1, spi1b, gata1a, mpx, lyz, mpeg1.1, myod1, aldh1a2, dhrs9, cyp26a1, angptl1–7, etsrp, fli1, and kdrl have been generated from zebrafish cDNA and clones inserted into the pEASY-blunt-zero vector (Transgen). Antisense RNA probes were labeled with DIG RNA labeling Mix (Roche) by transcription using T7 or T3 polymerase (Promega) in vitro.

For histological sections, embryos were washed with PBS following WISH and embedded in OCT compound (optimal cutting temperature compound, Sakura). Cryosections of 20 μm thickness were prepared along the anterior–posterior axis using an NX50 Cryostat Microtome (Thermo Scientific), and images were captured with an ECLIPSE Ni microscope.

Generation of Tg(angptl5:EGFP) and Tg(RARE:EGFP) transgenic lines

The angptl5:EGFP plasmid was generated by cloning the angptl5 promoter by cloning the upstream 867 bp region of angptl5 gene and EGFP fragment into the Tol2 backbone using a seamless cloning kit (Vazyme). For the RARE:EGFP plasmid, the EF1α enhancer-promoter in the PT2AL200R150G transposon vector [58] was replaced with a trimerized RARE sequence, which was amplified from the pGL3-RARE-luciferase plasmid (Addgene). Both plasmids were co-injected with Tol2 mRNA into zebrafish embryos at the 1-cell stage to generate the Tg(angptl5:EGFP) and Tg(RARE:EGFP) transgenic lines, respectively.

In situ hybridization chain reaction (HCR)

HCR staining was performed as described previously [59]. Briefly, zebrafish embryos were fixed with 4% (w/v) paraformaldehyde in DEPC-treated PBS at 4 °C overnight. Samples underwent proteinase K (MCE) treatment before incubation of probe. Samples were incubated with 2 pmol of each HCR probe sets in 100 μL probe hybridization buffer at 37 °C overnight. Hybridization was terminated by washing samples repeatedly with probe wash buffer at room temperature. For amplification, the probes were conjugated with fluorescent HCR amplifiers in amplification buffer for 4 hours at room temperature in the dark. The reaction was stopped by washing several times with 5× SSCT (DAPI co-staining was achieved by adding 0.1 μg/mL DAPI). Fluorescent hairpins, buffers, and DNA probes were purchased from Molecular Technologies. Fluorescent signals were detected using an OLYMPUS CSU-W1 confocal microscope.

Generation of angptl5^Δ10/Δ10 and angptl5^Δ5/Δ5 mutants using CRISPR/Cas9

To generate the angptl5^Δ10/Δ10 and angptl5^Δ5/Δ5 mutants, we synthesized gRNAs targeting the fourth exon and third exon of the zebrafish angptl5 gene, respectively. These procedures were performed as previously described [60]. The guide RNAs (S1 Table), targeting zebrafish angptl5 gene, were designed using CHOPCHOP (http://chopchop.cbu.uib.no) and synthesized using the T7 High Yield RNA Transcription Kit (Vazyme). The SpyCas9 protein (NEB) and angptl5-targeting gRNA were co-injected into WT embryos at the 1-cell stage. The angptl5 mutant lines were identified in the F1 generation by analyzing the PCR products using the primer pair listed in S1 Table.

Immunofluorescence

Whole-mount immunofluorescence staining was performed as described previously [61] using the following primary antibody: Phospho-Erk1 (Thr202/Tyr204)/Erk2 (Thr185/Tyr187) rabbit monoclonal antibody (Beyotime, 1:100), and Alexa Fluor 488-conjugated donkey anti-rabbit antibody (Invitrogen, 1:500 dilution) used as the secondary antibody. Embryos were photographed using an OLYMPUS CSU-W1 confocal microscope.

Preparing samples for sequencing and pro-processing single cell RNA-seq data

The angptl5^Δ10/Δ10 embryos were raised in 0.3× Danieau Buffer and harvested at 16 hpf. Collected samples were digested with trypsin into single cell suspension and set to Novogene for single cell sequencing.

Pro-processing single cell RNA-seq data

The raw Illumina sequencing reads were aligned to the zebrafish reference genome (GRCz11) through 10×  Genomics CellRanger pipeline v3.0.2 with default parameters. The angptl5^Δ10/Δ10 embryos yielded 7,350 cells, and the generated expression matrix was further processed by Seurat v5.1.0 [62]. Low-quality cells with fewer than 200 detected genes, more than 7,000 detected molecules, or greater than 15% mitochondrial expression were filtered out. The single cell RNA-seq dataset of 16 hpf WT embryos was obtained from the published work [21]. The single cell RNA-seq datasets of WT and angptl5^Δ10/Δ10 mutant embryos were merged for further analyses. The NormalizeData function with default settings was used to normalized the raw counts. The FindVariableFeatures function was used to identify the highly variable genes. The normalized data was then scaled with the through ScaleData function. Principle component analysis (PCA) and UMAP were employed to reduce the dimensionality of the data.

Integrative single cell RNA-seq analysis

The merged datasets were integrated using IntegrateLayers function with the canonical correlation analysis (CCA) integration method, utilizing shared principal components across two conditions. The layers of two conditions were re-joined after integration through the JoinLayers function. Spot clusters were obtained using the FindNeighbors and FindClusters function with the first 30 dimensions and the resolution set to 0.8. Nonlinear reduction was run using RunUMAP function with the first 30 dimensions. Marker genes for each cluster were identified using the FindAllMarkers function, and each cluster was then annotated based on the expression of these marker genes. The WT embryos had a total of 11,460 cells, with 170 erythroid progenitor cells (1.483%) and 19 myeloid progenitor cells (0.165%); while the angptl5^Δ10/Δ10 mutant embryos contained 7,241 total cells, with 192 erythroid progenitor cells (2.652%) and 32 myeloid progenitor cells (0.442%).

Analysis of differential abundance in cell types

Milo [63] was employed to analyze the differential abundance in cell types between WT and angptl5^Δ10/Δ10 embryos. Briefly, a k-nearest neighbor (KNN) graph was firstly constructed, and cells were assigned to the defined representative neighborhoods. The cells belonging to WT and angptl5^Δ10/Δ10 embryos in each neighborhood were counted. Differential abundance analysis was conducted within neighborhoods using the negative Binomial generalized linear model (GLM), with p-values corrected using the Spatial FDR method. A cell type label was assigned to each neighborhood by identifying the most abundant cell type among the cells within that neighborhood. A beeswarm plot was used to visualize the distribution of differential abundance fold changes among different cell types.

Chemical treatment

In general, embryos were treated with drugs from the shield stage until collection at the indicated stages. When injected at the 128-cell stage, embryos received drug treatment immediately post-injection with continuous exposure until specimen collection at the shield stage.

The drug was initially dissolved in DMSO to prepare a high-concentration stock solution, which was subsequently diluted with 0.3×Danieau buffer to the working concentration. Detailed pharmacological parameters and experimental concentrations are tabulated below: RA (sigma), Aldehyde dehydrogenase inhibitors 4-diethylaminobenzaldehyde (DEAB) (MCE, 10 μM), RA receptor (RARs) antagonists AGN 193109 (MCE, 20 μM), FAK inhibitor Defactinib (MCE, 2 μM/8 μM), MAPK inhibitor Adezmapimod (MCE, 10 μM), ERK inhibitor Mirdametinib (MCE, 10 μM), ERK activator C16-PAF (MCE, 1 μM/5 μM), NF-κB inhibitor BAY 11-7082 (MCE, 0.1 μg/mL, 0.4 μg/mL).

Quantitative Real-time PCR

Total RNA was isolated from zebrafish embryos (50 embryos per sample) using RNA isolater Total RNA Extraction Reagent (Vazyme), followed by cDNA synthesis with HiScript III 1st Strand cDNA Synthesis Kit (+gDNA wiper) (Vazyme). qPCR was conducted on a LightCycler 480 Instrument II system (Roche Diagnostics) with 2× Universal SYBR Green Fast qPCR Mix (ABclonal). The ΔΔCT method employed for data analysis, with normalization to 18S ribosomal RNA (18s rRNA) levels. Primer sequences with annealing temperatures are detailed in S1 Table.

Gene knockdown by morpholino and CRISPR/Cas9

For transient gene knockdown, two approaches were employed. Morpholino antisense oligonucleotides (Gene Tools LLC) were used to inhibit itga6l (5′- GCTCTCCTTTCTTCATCAGGTTCAT-3′) and shc1 (5′- ATAAAGAATTGGAAACCTTTCTCCT-3′). Embryos at the one-cell stage were microinjected with 60 nM (itga6l) or 3 nM (shc1) morpholino solution using standard procedures.

Alternatively, transient CRISPR/Cas9-mediated knockdown was performed for itga6l and itgb5. Gene-specific guide RNAs (gRNAs, sequences listed in S1 Table) were synthesized in vitro using the T7 High Yield RNA Transcription Kit (Vazyme). A mixture of each gRNA and SpyCas9 protein (NEB) was co-injected into WT embryos at the one-cell stage. This method achieves efficient somatic mutation and gene disruption for phenotypic analysis within the injected generation (F0).

Plasmid constructs

The full-length angptl5 coding sequence was amplified and cloned into a pCS2(+) vector with the addition of Flag or Flag-TurboID tag in frame at the C terminus. Based on Flag or Flag-TurboID tagged Angptl5 plasmid, Angptl5Δ10 was constructed by deleting 10 bp (base pair 458–467). Angptl5 and Angptl5Δ10 tagged with Flag or V5 were used for Immunoblotting (IB) and co-immunoprecipitation (Co-IP) in zebrafish embryos. Angptl5 and Angptl5Δ10 tagged with Flag-TurboID were used for IP-MS in zebrafish embryos. For Co-IP analysis in zebrafish embryos, the itga6l and itgb5 coding sequences were amplified and cloned into a pCS2(+) vector, with addition of HA, V5, or Flag tag in frame at the C terminus.

Co-IP and mass spectrometry

Co-IP analysis in zebrafish embryos was performed as described previously [64]. Briefly, embryos were co-injected with the indicated mRNA as described in the main text at the 1-cell stage, and collected at 5–6 hpf. Then, embryos were lysed with ice-cold lysis buffer [50 mM Tris at pH 7.5, 150 mM NaCl, 10% glycerol, 1% Triton X-100, and complete protease inhibitor (Roche)]. Lysates were centrifuged and the supernatants were then transferred to a Spin Column (Pierce). Samples were incubated with HA-Nanoab-Agarose (LabLEAD), Anti-Flag Affinity Gel (Bimake), or DYKDDDDK-Nanoab-Agarose (LabLEAD) at 4 °C overnight, followed by four washes with ice-cold wash buffer (50 mM Tris at pH 7.5, 150 mM NaCl, 1%Triton X-100 and complete protease inhibitor) four times before adding IgG Elution Buffer (Pierce).

IP-MS was performed as described [32]. Briefly, 800 embryos were lysed and subjected to immunoprecipitation with Streptavidin Agarose Resins (Pierce). Samples were separated by SDS-PAGE. The excised gels were analyzed by liquid chromatography-tandem mass spectrometry (LC-MS/MS) at Shanghai Applied Protein Technology Co., Ltd (APTBIO, China).

Chromatin immunoprecipitation and qPCR (ChIP-qPCR)

To assess the binding of Junb to the upstream region of the dhrs9 gene, ~300 zebrafish embryos were injected at the one-cell stage with HA-Junba mRNA and collected at 6 hpf. Following fixation with 1.45% formaldehyde, embryos were deyolked and lysed, and the lysate was treated with RNase A prior to chromatin shearing by sonication. Prior to immunoprecipitation, 10% of the sample was reserved as input. Chromatin was incubated with HA-agarose or control IgG-agarose beads overnight at 4 °C. Following sequential washes, bound complexes were eluted, reverse-crosslinked, and the DNA was purified. Quantitative PCR was performed to measure enrichment at three putative Junb binding sites upstream of dhrs9, predicted by analyses using the UCSC Genome Browser and the JASPAR database. Data were normalized to the input control.

Western blot analysis

Western blot analysis was performed as described previously [65]. Briefly, embryos or HEK293T cells were lysed with ice-cold RIPA Complete Lysis Buffer (Beyotime). For HEK293T conditioned medium (CM) preparation, the plasmids were transfected using polyethylenimine (PEI, Polysciences) at 1:2.5 DNA ratio. At 24 hours post-transfection, medium was replaced with serum-free DMEM. After 24 hours of incubation, CM samples were collected, centrifuged, and concentrated 5–10-fold using centrifugal filters (Millipore). Proteins were subjected and separated on SDS‐PAGE gel and transferred to PVDF membranes (Millipore). After blocking with nonfat milk (CST), membranes were incubated overnight with the following specific primary antibodies: ERK1/2 (Beyotime), Phospho-Erk1 (Thr202/Tyr204)/Erk2 (Thr185/Tyr187) (Beyotime), Beta Tubulin (Proteintech), Flag-Tag (Beyotime), HA-Tag (CST), and V5-Tag (Invitrogen). After four TBST washes, membranes were incubated with appropriate HRP‐conjugated secondary antibodies. After incubation, membranes were subjected to five TBST washes, then visualized using BeyoECL Moon Kit (Beyotime) or BeyoECL Star Kit (Beyotime) and detected with gel image analysis system (Tanon).

Statistical analysis

Raw image data were processed using ImageJ software. Normally distributed data were presented as mean ± SD. Comparisons between two groups were performed using the unpaired two-tailed Student t test, while comparisons among three or more groups were performed using one-way ANOVA followed by Tukey’s post hoc test. A p-value of less than 0.05 was considered statistically significant. The statistical parameters were shown in the figures and described in the figure legends or in the main text. All statistical analyses were performed using GraphPad Prism 9.5 (GraphPad Software, USA).

S1 Fig. Cross-species evolutionary comparison of ANGPTL5.

(A) Phylogenetic tree of ANGPTL5 nucleotide sequence in human, rabbit, chick, Xenopus and zebrafish using the neighbor joining method. (B) ANGPTL5 protein sequence homology among different species. (C) Structural models of the zebrafish and human ANGPTL5 proteins were modeled using AlphaFold3. CCD, coiled-coil domain; FLD, fibrinogen-like domain.

https://doi.org/10.1371/journal.pbio.3003858.s001

(TIF)

S2 Fig. Expression patterns of angptl family members during zebrafish embryonic development.

(A) WISH of angptl family members in zebrafish embryos. (B) WISH of angptl5 during zebrafish embryonic development at the indicated stages. (C) Cryosection of the trunk region for 24 hpf zebrafish embryo using O.C.T. after WISH of angptl5. (D) Time-lapse confocal micrographs of the angptl5 positive cell in Tg(angptl5:EGFP) embryos. Each imaging was performed for at least three independent replicates. LV, lateral view; DV, dorsal view; AV, anterior view. NT, Neural tube; NC, notochord; ICM, intermediate cell mass.

https://doi.org/10.1371/journal.pbio.3003858.s002

(TIF)

S3 Fig. Comparison of primitive hematopoiesis between WT and angptl5^Δ10/Δ10 embryos.

(A) Schematic diagram showing the positions of gRNA target site; the deletions of angptl5 mutant lines generated using CRISPR/Cas9; and the predicted truncated Angptl5Δ10 and Angptl5Δ5 proteins (predicted domains taken from the Uniprot database). (B) WISH of tal1 in WT and angptl5^Δ10/Δ10 embryos. (C) WISH of neutrophil markers lyz, mpx and macrophage marker mpeg1.1 in WT and angpt15^Δ10/Δ10 embryos. (D and E) WISH of spi1b (D) and lyz (E) in WT and angptl5^Δ5/Δ5 embryos. (F) qPCR shows the tal1, spi1b, gata1a and etsrp expression in WT and angptl5^Δ10/Δ10 embryos. Data presented as the mean ± SD, n = 30. Three independent biological replicates were used. Statistical significance: ***P < 0.001 (Unpaired t test). (G) Comparison of survival rates among WT, angptl5^Δ10/Δ10, and angptl5^Δ5/Δ5 zebrafish. (Kaplan–Meier survival curves of WT (blue solid line), angptl5^Δ10/Δ10 (red solid line), and angptl5^Δ5/Δ5 (green solid line) zebrafish under normal rearing conditions up to 20 days post-fertilization (dpf). The survival curves were compared using the Log-rank (Mantel–Cox) test (P = 0.7982, not significant (ns). Each genotype group consisted of n = 100). LV, lateral view; AV, anterior view. The data for this figure can be found in S1 Data.

https://doi.org/10.1371/journal.pbio.3003858.s003

(TIF)

S4 Fig. Integrative single cell RNA-seq analysis of 16 hpf WT and angptl5^Δ10/Δ10 embryos.

(A) Integrative UMAP analysis of 16 hpf zebrafish embryos, with the UMAP results separated by the conditions of WT and angptl5^Δ10/Δ10 embryos. The WT dataset is from the published work [21]. Each cell is coloured according to cell type annotations. (B) Dot plot showing the expression of two representative markers for each cluster. Color represents the gene expression level, while dot size indicates the percentage of cells within the cluster expression the gene.

https://doi.org/10.1371/journal.pbio.3003858.s004

(TIF)

S5 Fig. Comparison of primitive hematopoiesis and angiogenesis between WT and angptl5^Δ10/Δ10 embryos.

(A–C) WISH of etsrp (A), fli1 (B) and kdrl (C) in WT and angptl5^Δ10/Δ10 embryos. (D) Intersegmental blood vessels in WT and angptl5^Δ10/Δ10 embryos shown by fli1:GFP reporter. Each imaging was performed for at least three independent replicates. AV, anterior view; LV, lateral view; DV, dorsal view.

https://doi.org/10.1371/journal.pbio.3003858.s005

(TIF)

S6 Fig. Retinoic acid signaling in early zebrafish development.

(A) WISH of fn1a in WT and angptl5^Δ10/Δ10 embryos. (B) Schematic diagram of RA signaling pathway. (C) RA signaling in Tg(RARE:EGFP) embryos treated with RA from the shield stage to 24 hpf. Untreated embryos were used as control. (D) Morphology of WT embryos treated with RA inhibitor DEAB or different RA concentrations from the shield stage to 18-somite stage. Untreated embryos were used as control. (E) WISH of cyp26a1 in WT embryos treated with DEAB or RA from the shield stage to 8 hpf. Untreated embryos were used as control. LV, lateral view.

https://doi.org/10.1371/journal.pbio.3003858.s006

(TIF)

S7 Fig. Regulation of spi1b and gata1a expression by Retinoic Acid Signaling.

(A and B) WISH of spi1b in WT embryos treated with RA (A) or RA inhibitor AGN (B). Untreated embryos were used as control. (C) WISH of gata1a and myod1 in flat-mounted WT and angptl5^Δ10/Δ10 embryos treated with RA inhibitor DEAB or AGN. (D) WISH of cyp26a1 in WT and angptl5^Δ10/Δ10 embryos injected with dhrs9 or aldh1a2 mRNA at the 1-cell stage. Uninjected embryos were used as control. Statistics are shown on the right of the representative photos. AV, anterior view; LV, lateral view. The data for this figure can be found in S1 Data.

https://doi.org/10.1371/journal.pbio.3003858.s007

(TIF)

S8 Fig. Secretion and functional analysis of tagged Angptl5.

(A) Synthesis and secretion analysis of Angptl5-Flag-TurboID and Angptl5Δ10-Flag-TurboID in HEK293T cells after transfection with Angptl5-Flag-TurboID and Angptl5Δ10-Flag-TurboID plasmid for 48 hours. Conditioned medium (CM) were collected for western blot. (B) Expression of Angptl5-Flag-TurboID, Angptl5Δ10-Flag-TurboID and biotinylated proteins in zebrafish embryos. Embryos were injected with Angptl5-Flag-TurboID and Angptl5Δ10-Flag-TurboID mRNA at the 1-cell stage, and then embryos were collected for western blot at the 75% epiboly. (C) WISH of spi1b in angptl5^Δ10/Δ10 embryos injected with angptl5, angptl5-flag, or angptl5-TurboID mRNA at the 1-cell stage. Uninjected embryos were used as control. (D) WISH of spi1b in angptl5^Δ10/Δ10 embryos. Embryos were injected with angptl5 mRNA ± itga6l MO at the 1-cell stage. Uninjected embryos were used as control. AV, anterior view.

https://doi.org/10.1371/journal.pbio.3003858.s008

(TIF)

S9 Fig. Interaction between Itga6l and integrin β-subunits.

(A) Expression level of integrin β-subunits in zebrafish embryos from published work [35]. (B–F) Structural models of the extracellular domain complex between Itga6l and: Itgb1a (B); Itgb1b (C); Itgb1b.1 (D); Itgb6 (E); Itgb4 (F). (G) Structural models of the Angptl5_78–397-Itga6l_14-736-Itgb4_25–710 complexes. (H and I), WISH of itga6l (H) and itgb5 (I) in WT embryos. The three-dimensional structures of all protein complexes were predicted using AlphaFold3 via the AlphaFold Server (https://alphafoldserver.com). The top-ranked model based on the predicted template modeling score was selected for further analysis. All structural figures were generated using the open-source version of PyMOL. DV, dorsal view; LV, lateral view; AV, anterior view.

https://doi.org/10.1371/journal.pbio.3003858.s009

(TIF)

S10 Fig. Analysis of integrin α6lβ5 in regulating primitive hematopoiesis.

(A) WISH of spi1b, gata1a, and myod1 in WT embryos treated at the 1-cell stage as follows: injection of Cas9/gRNA or treatment with Integrin inhibitor (Cilengitide) from 6 hpf. KD: knockdown; (B) WISH of spi1b in WT embryos injection of respective mRNAs. Uninjected embryos were used as control. (C and D) WISH of spi1b (C), gata1a and myod1 (D) in WT embryos with angptl5 + itga6l + itgb5 mRNA at the 1-cell stage at the 1-cell stage. AV, anterior view.

https://doi.org/10.1371/journal.pbio.3003858.s010

(TIF)

S11 Fig. ERK signaling is reduced in angptl5^Δ10/Δ10 embryos.

(A–C) WISH of spi1b or cyp26a1 in WT embryos treated with FAK inhibitor (A), NF-κB inhibitor (B), MAPK inhibitor and ERK inhibitor (C) from shield stage to 18-somite stage. Untreated embryos were used as control. (D) Schematic diagram of ERK2 and constitutively activated ERK2 (caERK2). (E) WISH of spi1b in WT embryos injected with ERK2 or caERK2 mRNA at the 1-cell stage. Uninjected embryos were used as control. (F) Projection of images from confocal stacks to show co-localization of spi1b and p-ERK in Tg(spi1b:EGFP). Each imaging was performed for at least three independent replicates. (G) Cryosection of the trunk region for 24 hpf WT zebrafish embryo using O.C.T. after IF of p-ERK. NT, Neural tube; ICM, intermediate cell mass. (H) Projection of trunk region images from confocal stacks to show p-ERK1/2 in WT and angptl5^Δ10/Δ10 embryos by IF. (I) Expression of p-ERK in WT embryos injected with angptl5 + itga6l + itgb5 mRNA at the 1-cell stage. Statistics are shown on the right. AV, anterior view; LV, lateral view. The data for this figure can be found in S1 Data.

https://doi.org/10.1371/journal.pbio.3003858.s011

(TIF)

S12 Fig. Kinases and adaptor proteins mediating Angptl5-Itga6lβ5-induced ERK signaling identified by immunofluorescence.

(A–F) IF of p-ERK1/2 in WT embryos. Embryos were injected in one animal pole blastomere at the 128-cell stage with the indicated mRNAs or MO, and then treated as follows: with the FAK inhibitor Defactinib (B); with the EGFR inhibitor Gefitinib (D and E); with the Src inhibitor Saracatinib (F); or with a shc1 MO (G and H). All embryos were imaged at 6 hpf. AV, animal view.

https://doi.org/10.1371/journal.pbio.3003858.s012

(TIF)

S13 Fig. Integrin α6lβ5 regulates RA signaling.

(A) WISH of cyp26a1 in WT embryos. Embryos were co-injected with Cas9 protein and gRNAs target against itga6l and itgb5 at the 1-cell stage. Untreated embryos were used as control. KD: knockdown; LV, lateral view.

https://doi.org/10.1371/journal.pbio.3003858.s013

(TIF)

S14 Fig. RA signaling is regulated by Angptl5-Integrin α6lβ5-ERK signaling cascade.

(A) RA signaling in angptl5^Δ10/Δ10 embryos injected with angptl5 + itga6l + itgb5 mRNA at the 1-cell stage and treated with or without ERK inhibitor from the shield stage to 18-somite stage. Data presented as the mean ± SD, n(angptl5^Δ10/Δ10) = 19, n(angptl5^Δ10/Δ10 + angptl5 + itga6l + itgb5) = 20, n(angptl5^Δ10/Δ10 + angptl5 + itga6l + itgb5 + ERKi) = 16. Statistical significance: ***P < 0.001 (One-way ANOVA). The data for this figure can be found in S1 Data.

https://doi.org/10.1371/journal.pbio.3003858.s014

(TIF)