Genetic duplication of tissue factor reveals subfunctionalization in venous and arterial hemostasis

Tissue factor (TF) is an evolutionarily conserved protein necessary for initiation of hemostasis. Zebrafish have two copies of the tissue factor gene (f3a and f3b) as the result of an ancestral teleost fish duplication event (so called ohnologs). In vivo physiologic studies of TF function have been difficult given early lethality of TF knockout in the mouse. We used genome editing to produce knockouts of both f3a and f3b in zebrafish. Since ohnologs arose through sub- or neofunctionalization, they can unmask unknown functions of non-teleost genes and could reveal whether mammalian TF has developmental functions distinct from coagulation. Here we show that a single copy of either f3a or f3b is necessary and sufficient for normal lifespan. Complete loss of TF results in lethal hemorrhage by 2–4 months despite normal embryonic and vascular development. Larval vascular endothelial injury reveals predominant roles for TFa in venous circulation and TFb in arterial circulation. Finally, we demonstrate that loss of TF predisposes to a stress-induced cardiac tamponade independent of its role in fibrin formation. Overall, our data suggest partial subfunctionalization of TFa and TFb. This multigenic zebrafish model has the potential to facilitate study of the role of TF in different vascular beds.


Introduction
Preventing hemorrhage in the setting of normal physiology requires a well-regulated plasma coagulation system. Under normal conditions, the potent activator tissue factor (TF, gene symbol F3) is expressed as a transmembrane glycoprotein on the surface of many cell types outside the vasculature. Disruption of the endothelium exposes TF which binds to the circulating zymogen, factor VII (FVII, 99% of circulating factor), or its active protease (FVIIa, 1%) [1]. Once bound, FVII is rapidly activated to FVIIa and the TF/FVIIa complex serves as an initiator of coagulation via the activation of factor X and factor IX [2,3]. Apart from vascular damage, infection induces proinflammatory cytokines known to trigger the expression of TF on the surface of mononuclear and endothelial cells [4]. Additionally, many neoplasms express high levels of TF which can enter circulation through the shedding of microvesicles [5].
Due to the early lethality of complete F3 knockout, a human F3 minigene was introduced into F3 -/mice, resulting in TF protein levels 1-2% of normal [17]. Although the minigene rescues survival into adulthood, low TF mice exhibit prolonged bleeding after hemostatic challenge and have increased rates of hemorrhage. Interestingly, while human TF activates coagulation in mice, mouse TF has about a six-fold lower activity in activation of coagulation in human plasma [18]. This is due to species-specific interaction between TF and FVIIa [19,20]. Therefore, an accessible model that does not rely on the complications of xenogeneic proteins is critical for understanding TF deficiency.
The zebrafish is a small aquatic vertebrate with a highly conserved hemostatic system [21,22], high fecundity, and rapid, transparent, and external development. Circulating blood is visible by 36 hours of life and assays for assessing coagulation in vivo are readily performed between 72 and 120 hours post fertilization (hpf) [23,24]. Importantly, zebrafish are able to survive severe hemostatic defects that are usually lethal in mammals, and develop into adulthood [25][26][27][28][29]. Zebrafish have two copies of f3 (f3a and f3b) due to an ancestral whole genome duplication event (referred to as "ohnologs") [30]. Previously performed antisense knockdowns of f3a and f3b in zebrafish suggest a role in early vascular development and hemostasis [31,32], although the morpholino-based knockdown technology is known for off-target effects, especially in the vasculature [33]. To our knowledge, no studies have reported the consequences of a total loss of TF activity in zebrafish, nor have studies previously genetically knocked out the ohnologs individually and in combination.
Although the first genetic heterozygous deficiency of F3 in humans has recently been described, [34] complete loss has never been seen in humans and is difficult to study in mice. Here we show the use of genome editing with CRISPR/Cas9 to generate loss-of-function alleles in both copies of zebrafish f3. Complete loss of TF is compatible with embryonic through juvenile development but leads to early adult lethality, although a single allele of either ohnolog is sufficient to enable survival into adulthood. Furthermore, we show that TFa has higher procoagulant activity than TFb in vitro and is sufficient for venous hemostasis, while TFb is sufficient for arterial coagulation. Understanding the spatial and temporal control of f3a and f3b, as well as their functional differences, could prove valuable for understanding regulation of the coagulation cascade, as well as identifying potential distinct or novel activities of TF in mammalian hemostasis.

Characterization of the f3a and f3b genes
Analysis of the Refseq database [35] revealed two orthologs of F3, annotated as f3a and f3b, with each consisting of 6 exons. f3a (NM_001245967.1) is a 13 kilobase (kb) locus on chromosome 24 and f3b (NM_001017728.2) is a 7 kb locus on chromosome 2. The coding regions of each gene were isolated from embryonic mRNA and confirmed by Sanger sequencing. The predicted amino acid sequences of TFa (294 aa) and TFb (288 aa) were used for homology analysis (S2 Table). The mature (leader peptide removed) TFa and TFb proteins share 45.2% sequence identity with each other and 34.5% and 32.9% sequence identity with human TF (NP_001984.1), respectively. Protein sequence alignment demonstrates several highly conserved regions (S1 Fig).
Humans have a well-known alternative splice form of F3, [36] but no evidence for alternative splicing was found for either TF ortholog in available sequence databases. Human TF is glycosylated at asparagines 11, 124 and 137. [2] Both TFa and TFb contain a homologous asparagine to human Asn11. However, only TFb has a predicted Asn near 137 and neither has a predicted glycosylation site near 124. [37] Additionally, TFa and TFb have 2 predicted conserved disulfide bonds in the extracellular domain and a conserved cysteine in the cytoplasmic tail that is palmitoylated in humans (S1 Fig). [38] Only TFb has potential sites of phosphorylation in the cytoplasmic tail according to PhosNet prediction. [39] Genome editing to generate f3a and f3b null alleles Large stably transmitting deletions were genetically engineered in f3a and f3b using CRISPR/ Cas9 (Fig 1A and 1B). For simplicity, when discussing both f3 genotypes, wild-type alleles will be referred to as "A" and "B" while null-alleles will be referred to as "a" and "b," e.g., double heterozygotes are Aa/Bb (A � will represent AA or Aa, B � will represent BB or Bb). qRT-PCR at 4 dpf (days post fertilization) demonstrated that homozygosity of the f3a deletion resulted in reduction of mRNA levels consistent with nonsense-mediated decay (Fig 1C). The f3b allele developed a large insertion at the site of editing. Conversely, qRT-PCR of homozygous f3b larvae revealed levels of residual transcript consistent with wild-type and heterozygous genotypes and compensatory induction of f3a (Fig 1C). When the start codon to the 3' UTR of the f3b transcript was amplified using a single primer pair and analyzed via agarose gel electrophoresis, multiple small transcripts were identified in heterozygous and homozygous f3b mutants ( Fig 1D). When this mixed pool of products was analyzed using Sanger sequencing, the most prevalent was a splice variant from exon 2 to exon 6 encoding a 93 amino acid peptide which includes the first 65 N-terminal residues of the preprotein and 28 out of frame residues from the last exon. The engineered alleles for both f3a and f3b result in significant deletions and frameshift mutations that likely generate complete loss of functional TF protein.

A single f3 allele is both necessary and sufficient for long term survival
A large group of offspring produced from incrossing Aa/Bb parents was genotyped at 8 weeks revealing no deviation from the expected Mendelian ratios. However, all 9 double homozygous mutant fish (aa/bb) died over the next week (Fig 2A). The remaining genotypes, including Aa/ bb and aa/Bb, demonstrated equivalent survival over the following year (p > 0.05). A second group was raised to 6 months of age before genotyping to eliminate the impact of genotyping stress on survival (Fig 2B). No double homozygous mutant fish were identified in this cohort, consistent with complete TF deficiency being incompatible with long-term survival. A final cohort of 7 aa/bb fish was identified at <1 month of age and followed. Six succumbed to hemorrhage by 2 months of age while the last reached sexual maturity before dying at 120 days ( Fig  2C). The cause of death across all groups appeared to be overt hemorrhage in the head or pericardial space (Fig 2D). When groups of offspring were generated from Aa/bb males crossed to aa/Bb females and Aa/bb females crossed to aa/Bb males, all possible genotypes were recovered at 1 month of age, indicating that maternal deposition of f3a/f3b transcripts does not appear to influence early survival.

Loss of TF does not alter vascular development
Hypothesizing a role of TF in angiogenesis, the gata1a:DsRed [40] (labeling erythrocytes), and kdrl:eGFP [41] (labeling endothelial cells) transgenic lines were used to investigate endothelial development and vascular perfusion. At 30 hpf, the number of vascular sprouts and their growth progress were scored prior to genotyping, and no significant difference was observed across genotypes (Fig 3A-3C). At 54 hpf, there were no obvious defects in endothelial development or erythrocyte perfusion in aa/bb larvae ( Fig 3D). No obvious morphological defects were seen in the first week of life under bright field examination. Overall, these data indicate that TF is not necessary for the differentiation, proliferation, or function of the vascular endothelium in zebrafish.

f3a and b demonstrate tissue-specific roles in venous and arterial hemostasis
Laser-induced vascular endothelial injury is a methodology that injures the endothelium leading to exposure of TF, and thereby examines the ability of the coagulation cascade to generate a proper hemostatic response. This was used to trigger thrombus formation in the venous system at 3 dpf in f3 mutants ( Fig 4A). Loss of TFb alone did not lead to any alterations in the time to occlusion (TTO). Loss of TFa resulted in a mild delay in median TTO from <20 to 32 seconds. In the absence of TFa, heterozygosity for f3b (aa/Bb) exacerbated the TTO delay to 43 seconds while complete deficiency of TF (aa/bb) resulted in the almost complete inability to form occlusive thrombi ( Fig 4A, open squares). Thus, a single copy of either f3a or f3b is necessary for venous occlusion, but only f3a is sufficient for normal TTO.
Laser-mediated endothelial injury was performed on the dorsal aorta at 5 dpf to determine the role of the f3 ohnologs in arterial hemostasis. Loss of f3b but not f3a led to an increased median TTO. Notably, complete loss of f3b led to a statistically significant increase in failure to occlude with phenotypic penetrance of 60%, 46% and 86% in AA/bb, Aa/bb, and aa/bb offspring; respectively ( Fig 4B). When subset for f3b-null larvae that occluded, the median TTO was consistent across genotypes, suggesting a binary phenotype. Endothelial injury was then performed at 2 dpf, prior to the robust expansion of thrombocytes in circulation [42,43]. Greater than 90% of Aa/Bb and aa/Bb larvae formed occlusive clots following arterial damage. In contrast, significantly fewer Aa/bb (11%) and aa/bb (0%) larvae formed occlusive thrombi ( Fig 4C). Therefore, in contrast to the stochastic role of f3a in venous thrombus formation, one allele of f3b appears to be necessary and sufficient for arterial thrombus formation.

TFa demonstrates greater in vitro plasma procoagulant activity than TFb
The differential responses in the venous and arterial vasculature suggest a possible functional difference between the two ohnologs. One method for testing this hypothesis is through examination of fibrin formation in platelet-poor plasma ex vivo, a common technique used in Complete loss of TF activity results in early lethality due to hemorrhage. (A) Fish from a double heterozygous (Aa/Bb) f3 incross were genotyped at 8 weeks of age and found to have normal Mendelian ratios (n = 135, AA/BB = 10, AA/Bb = 17, AA/bb = 7, Aa/BB = 15, Aa/Bb = 36, Aa/bb = 20, aa/BB = 10, aa/Bb = 13, aa/ bb = 9). Survival curves show that by 9 weeks, there was statistically significant loss of aa/bb offspring. By 400 days, there was a relatively even reduction of the remaining genotypes which is common in wild-type fish, and confirms that a single copy of f3a or f3b is sufficient for survival. (B) A repeat cohort was left undisturbed until genotyping at 6 months of age. The statistically significant loss of aa/bb confirms the early lethality of TF loss independent of genotyping stress. (C) A group of 7 double homozygous mutants were identified at 1 month of age and observed over the following 3 months. 6/7 were lost by 2 months with the single survivor passing at 112 days. (D) Early lethality was due to gross hemorrhage. Examples include the pericardial space (left) and head (right) (red arrows).
https://doi.org/10.1371/journal.pgen.1010534.g002 mammalian studies of hemostasis. This could determine if there are functional differences between TFa and TFb proteins. We have found that as in mammals, zebrafish venous hemostasis is more dependent on fibrin, rather than thrombocytes (platelets in mammals). We could not consistently isolate enough plasma from zebrafish, so we obtained thrombocyte-poor trout plasma and performed a manual tilt-test after activation by liposomes containing recombinant TFa or TFb proteins (Fig 4D). Plasma activated by TFa-liposomes clotted in roughly 120 seconds. Samples initiated with TFb-containing liposomes were not observed to form clots until a final 15-minute examination point. A titration of liposomes at a lower sodium concentration demonstrated roughly 10 to 100-fold higher procoagulant activity of TFa-liposomes relative to TFb-liposomes (Fig 4E), demonstrating a functional difference between the two proteins. Laser-mediated venous endothelial injury at 3 dpf shows that the time to occlusion (TTO) was not altered by loss of TFb (AA/bb, Aa/bb) but was delayed in the complete absence of TFa (aa/BB, aa/Bb). Thrombus formation was absent after complete TF loss (aa/bb). (B) Lasermediated arterial endothelial injury at 5 dpf shows that the ability to form occlusive thrombi was significantly reduced in fish lacking TFb (AA/bb, Aa/bb, aa/bb, indicated by grey bars) compared to BB and Bb groups (green and purple bars; respectively). (C) Endothelial injury at 2 dpf confirms the dependence on TFb in the arterial system prior to the presence of thrombocytes. Occlusive thrombi were nearly absent without TFb (Aa/bb, aa/bb). (D) Relipidated recombinant TFa or TFb proteins (173 nM) were mixed with citrated trout plasma, and recalcified. TFa-liposomes induced clot formation during a manual tilt test significantly faster than TFb-liposomes or empty liposome controls. The reaction was observed until 300 seconds, and then again at 15 minutes to assess for delayed clot formation (the latter observation is indicated by an open or closed circle at 300 seconds). (E) The assay was repeated with relipidated TF diluted in 0.5x HBS (20 mM HEPES pH 7.5, 50 mM NaCl) and plotted on log-log axes demonstrating 10 to 100-fold higher procoagulant activity of TFa over TFb. Statistical significance by Mann-Whitney U testing: �� < 0.01, ��� < 0.001. All phenotypic data were collected by an observer blinded to genotype. Open squares in A and B indicate individual larvae that did not occlude within 120 seconds. https://doi.org/10.1371/journal.pgen.1010534.g004

Loss of factor IX (FIX) modifies clot formation in f3a-null offspring
Zebrafish lack factor XI (FXI), raising the possibility that FIX (gene symbol f9b) activation is completely dependent on TF activity and that the previous phenotypic variabilities could represent differential substrate preference for TF:FVIIa complexes containing TFa versus TFb. Offspring from Aa/Bb; f9b +/incrosses were subjected to endothelial injury at various time points and sub-analysis of relevant genotypes explored (full dataset available in S2 Fig).
At 3 dpf, f9b genotype did not influence the ability to form occlusive venous clots in f3bnull offspring. In contrast, complete loss of f9b in f3a-null larvae resulted in a significant reduction in clot frequency to 67% compared to 91% in f9b +/ � offspring (p<0.01; Fig 5A). These data suggest that venous clot formation is primarily mediated from canonical activation of FX by TF/FVII. The requirement for FIX in the absence of TFa, and not the absence of TFb, supports the hypothesis that TFb has lower efficiency in the venous system, thus necessitating FIX-mediated amplification. At 5 dpf, f3b-null offspring formed occlusive arterial clots only 45% of the time when carrying at least 1 copy of f9b. This went up to 75% when both copies of f9b were lost, although not significant (p = 0.06). In contrast, f3a-null offspring saw a significant decrease in clot formation from 76% of f9b +/ � larvae to 33% of f9b -/offspring (p < 0.01; Fig 5B). Overall, FIX significantly enhances TFb-mediated clot formation with a larger effect in the arterial circulation ( Fig 5C).

Loss of TFb partially protects against inherited thrombophilia
Our previous studies demonstrate that the loss of antithrombin (AT3) in zebrafish results in spontaneous venous thrombosis at 3 dpf with subsequent disruption of induced clot formation due to hypofibrinogenemia. This is followed by early adult lethality due to intracardiac thrombosis [29].
A cohort of at3 -/-;Aa/Bb fish carrying a GFP-tagged fibrinogen beta transgene were incrossed and their offspring scored for spontaneous venous thrombosis at 5 dpf, followed by f3 genotyping. 39% of larvae containing at least one copy of each f3 (A � /B � ) developed thrombosis. This was not statistically different from the 36% frequency of f3a -/larvae with at least one copy of f3b (aa/B � ). However, f3b -/larvae with at least one copy of f3a (A � /bb) had a significantly lower rate of thrombosis at 12%. aa/bb larvae did not have evidence of thrombosis ( Fig 6A). These data indicate spontaneous venous thrombosis in the context of AT3 deficiency is predominantly the result of f3b activity.
Additional offspring from the at3 -/-;Aa/Bb incross were followed for survival and genotyped at 2-3 months of age. There were no clear differences in survival by genotype through the first 400 days aside from the early expected loss of aa/bb (Fig 6B). This shows that TFb-mediated modification of early acute spontaneous thrombosis in at3 -/larvae does not translate into protection against long-term adult lethality.

Loss of TF predisposes to cardiac tamponade independent of fibrin formation
The laboratory aquatic milieu, relatively low blood pressures, and lack of birthing trauma create a low stress early developmental environment for zebrafish compared to mammals. To understand if these factors explain the ability of zebrafish to survive to 2 months of age without TF, a chemical stress challenge was initiated at 3 dpf. After 16 hours of treatment with cortisol and epinephrine, the hearts of treated fish were scored by a blinded observer (Fig 7A and S1 and S2 Video). Zebrafish with overall TF deficiency developed cardiomegaly and a concomitant significant increase in erythrocytes in the pericardial space, consistent with cardiac tamponade (Fig 7B, left). Prothrombin-deficient fish exhibited a similar frequency of cardiac tamponade (Fig 7B, center). However, fibrinogen-deficient larvae were unchanged from wild-type (Fig 7B, right). These data suggest that the risk of tamponade is related to the roles of TF and prothrombin in thrombocyte activation and/or non-hemostatic signaling.

Discussion
Study of the evolution of the coagulation cascade has been enhanced with increased access to novel model organisms and their genomes [44][45][46]. The function of TF is essential for vertebrate life despite its low sequence level conservation (~60% between human and mouse and 40% between human and fish), and it is conserved even in jawless fish. TF in lamprey shares the same broad structure with humans, mice, and zebrafish, including an extracellular domain comprising two fibronectin type III modules, a transmembrane domain, and a cytoplasmic tail. Simulations suggest that despite the low sequence homology between humans and lamprey, the dynamics of TF activation of FVII are conserved [47]. It's safe to reason that these same dynamics likely apply to the two copies of TF in zebrafish, especially given demonstrated sequence level conservation of disulfide bridges and glycosylation sites. The low degree of sequence conservation has been a challenge for the study of TF, necessitating development of intricate in vitro systems and specialized reagents. While loss of TF in mice has been linked to abnormalities in development, angiogenesis, inflammation, and regeneration, it has been difficult to validate these phenotypes and eliminate artifacts in whole-organism models [4,13,48,49]. Zebrafish offer a model to address many of these concerns due to their easily visualized development, conserved coagulation cascade, resistance to hemostatic imbalance, and uniquely due to the presence of 2 distinct genomic copies of f3. Importantly, while this study focuses on phenotypes attributed to the extracellular domain of TF, there are changes in the intracellular domain that are not addressed. For example, phosphorylation of the cytoplasmic tail may contribute to TF influenced intracellular signaling. Only TFa demonstrates conserved sequence suggesting that this phosphorylation could occur.
Prior studies of TF in zebrafish have focused on the use of morpholino antisense technology to transiently knock down translation in larvae. Knockdown of either f3a or f3b using morpholinos appears to cause angiogenic defects, including vascular leak of small molecules and delay of venous angiogenesis not observed in this study. [31,32,50] In contrast to those studies, we have employed genetic mutants to study simultaneous, complete, and permanent deficiency of TFa and TFb. We found that zebrafish lacking TF typically survive to 8 weeks of age without gross defects in vascular morphology or endothelial development. We believe our large engineered genomic deletions reduce the likelihood of nucleic acid based genetic compensation [51] as an explanation and increases the likelihood that the observed compensation is driven by protein level homeostasis. The gross morphologic phenotypic differences observed in the studies using morpholinos to knock down TF may be due to a combination of developmental delays and toxicity due to off target effects inherent in morpholino experiments. However, subtle vascular-specific phenotypes could have been missed in our analyses.
We found that complete TF deficiency typically causes lethality by 2 months (early adulthood), but a single allele of either f3a or f3b is sufficient for normal survival and sexual reproduction. These data confirm that although TFa and TFb share less than 50% sequence identity, they both maintain the ability to activate the coagulation cascade. This also demonstrates the strength of the zebrafish for studying in vivo biological functions of TF that are difficult to test due to mammalian lethality. With these data, we were also able to identify evidence of TF subfunctionalization. TFa is necessary and sufficient for normal venous clot formation in larvae, but only plays a minor role in the arterial system. Conversely, TFb is necessary and sufficient for an intact arterial response to injury while occupying a lesser function in the venous system. The 10 to 100-fold greater procoagulant activity of TFa over TFb in the ex vivo trout plasma assay likely explains that the dominant venous function is due to a functional difference between TFa and TFb, but also suggests a distinct mechanism for the enhanced arterial activity of TFb. Conversely, spontaneous thrombosis in our at3-null model was more dependent on TFb rather than TFa. This raises the possibility of differential behaviors for the two types of TF activity in hemostatic versus thrombotic conditions. For example, graft failure due to thrombosis is a significant clinical problem and several studies implicate ectopic TF expression in response to hemodynamic changes as a driver of clot formation. [52][53][54] Our model suggests differential behaviors for the two TFs might inform specific and novel methods for modifying spontaneous thrombosis in the venous and arterial circulation.
Unfortunately, spatiotemporal studies for f3 expression have been unsuccessful presumably due to the low absolute levels of f3 mRNAs and the lack of available antibodies. [55] Two large single-cell transcriptome datasets find that f3a has higher expression in the ventral aspect of the embryo near the venous circulation while f3b is more broadly expressed. [56,57] Interestingly, evidence for f3b expression in hemangiogenic progenitors points to a potential function in cell differentiation that may partially explain the observed arterial phenotype. [56] Future work to understand the precise cell types where the f3 ohnologs are expressed will help sort out their disparate roles.
Although FXIa-mediated activation of FIX is well known [58,59], it is also established that FIX is an important substrate of the TF:FVIIa complex. [60,61] In fact, a mutation in FIX that interferes with its interaction with TF has been shown to cause mild hemophilia. [62] Zebrafish do not have known orthologs for FXI or FXII, suggesting that TF could be the primary activator of zebrafish FIX. [63] We hypothesize that TFa:FVII and TFb:FVII complexes may have different affinities for FIX and FX, with distinct roles in the extrinsic or intrinsic pathways. Although zebrafish have 2 copies of FIX (f9a and f9b), transcriptomic profiles of early development show that f9b is the predominant form prior to 5 dpf. [55] Therefore, we chose to focus on f9b for this study. Multigenic f3/f9b knockouts demonstrate similar patterns in both induced venous and arterial thrombus formation, more pronounced in the latter. The results suggest that TFb serves as the primary activator of intrinsic tenase, while TFa fills the canonical role of FX activation. The mild increase in TFa-mediated clot formation in the absence of FIX implies that FIX interacts with TFa and lowers the overall extrinsic tenase activity. Recently, it has been suggested that f9a plays a significant role ex vivo and at later ages. This may explain the incomplete phenotypic penetrance in this study. [64] Regardless, our work points to a role for the intrinsic pathway in mediation and a target for management of acute spontaneous thrombosis.
The ability of zebrafish to survive complete TF deficiency into early adulthood is compatible with previous studies of common pathway knockouts, [25][26][27][28], with all showing spontaneous intracranial and pericardial hemorrhage, a pattern similar to the adult mouse deletion of prothrombin. [65] The resistance to developmental lethality may be due to the low stress aquatic environment, the lack of birthing trauma, low-pressure vasculature, or species-specific differences in hemostatic factors. By chemically inducing stress, we were able to elicit a phenotype in TF-deficient larvae consistent with cardiac tamponade, which might be indicative of an inducible vascular defect. This same phenotype was observed in the f2 knockout, but not with loss of fibrinogen. The implication is the presence of secondary functions for TF and thrombin that are distinct from fibrin polymerization, e.g. thrombocyte regulation, PAR signaling, and/ or integrin binding [6][7][8]. These data suggest a possible inducible vascular defect that leads to hemorrhage. Notably, cardiac-specific f3 knockout mice demonstrate a mild hemostatic defect but significant fibrosis and hemosiderosis when chemically challenged with isoproterenol, and humanized low-TF mice similarly show significant cardiac fibrosis and left ventricular dysfunction in adulthood. [9,17] Alternative explanations include subtle cardiac-specific structural abnormalities and/or changes in vascular development such as vascular remodeling that were beyond the detection of our vascular survey.
Overall, the duplication of f3 in zebrafish and the evidence for subfunctionalization offers multiple opportunities to explore the subtleties of TF function. The work presented is consistent with the expectations from previous genetic knockouts of procoagulant factors in zebrafish. However, the novel differentiation of arterial and venous functions and potential substrate specialization facilitates asking narrower questions than previously addressable in mammals.

Ethics statement
Zebrafish maintenance and use was according to protocols approved by the University of Michigan Institutional Animal Care and Use Committee, identification number PRO00010679.

Animal care
Wild-type fish were a hybrid line generated by crossing AB and TL zebrafish acquired from the Zebrafish International Resource Center. Genetic knockouts of prothrombin (f2) and fibrinogen alpha (fga) were generated and described previously [26,28]. Tricaine (Western Chemical) was used for anesthesia and rapid chilling in an ice water bath for euthanasia.

Targeted mutagenesis using genome editing
Throughout the text, we use "TF" and "FIX" to refer to the tissue factor and factor IX proteins, and "f3" and "f9b" to refer to the genes, respectively. The ChopChop2 server was used to identify predicted high efficiency target sites for CRISPR/Cas9 engineering. [66] Three guides each were selected for f3a, f3b and f9b. DNA oligonucleotides for the target sequence were synthesized with homology to a second common backbone oligo (S1 Table). The pieces were assembled via PCR and purified using phenol:chloroform. Single guide RNAs (sgRNAs) were synthesized using the Maxiscript T7 kit and purified via ethanol/ammonium acetate precipitation. 600 ng of each sgRNA was mixed with 0.5 uL EnGen Cas9 protein (New England Bio-Labs) at a 1.5:1 molar ratio in 300 mM KCl. The solution was incubated at 37˚C for 10 minutes and 2 nL injected into the yolk of single cell embryos. Test embryos were lysed at 24 hpf and flanking primers used to amplify the target site. High resolution gel electrophoresis was performed to validate efficient cleavage. Larvae injected with 2 known successful guides were raised, bred, and their offspring screened for large deletions. Lines were established from single heterozygous founders. The f9b line is a large deletion from exon 2 to 8 (total 8 exons), so there is essentially no remaining coding sequence (Lavik and Shavit et al, in preparation).

Genotyping of mutant offspring
Whole embryos or adult fin biopsies were lysed in buffer (10 mM Tris-HCl, pH 8.0, 2 mM EDTA, 2% Triton X-100, 100 μg/mL proteinase K) for 2 hours at 55˚C followed by incubation at 95˚C for 5 minutes. One microliter was used as template for gene specific polymerase chain reactions (S1 Table) and analyzed by gel electrophoresis.

Real time quantitative polymerase chain reaction (RT-qPCR)
RT-qPCR was performed as described previously. [26] Briefly, total RNA was extracted from 4 days post fertilization (dpf) larvae using the RNeasy Mini Plus kit (Qiagen) and transcribed with oligo (dT) [12][13][14][15][16][17][18] primer and Superscript III (Invitrogen). Three equal pools of whole larvae were tested per genotype. The resulting cDNA was template for qPCR (StepOnePlus, Applied Biosystems) with gene specific PrimeTime probe-based qPCR Assays (Integrated DNA Technologies) (S1 Table). The expression levels of f3a and f3b were normalized to the actb2 gene and significance analyzed using the double delta Ct method as described. [67] Laser-induced vascular endothelial injury Vascular endothelial injury was performed as previously described. [24,29] Briefly, larvae were mounted in 0.8% low melt agarose and visualized using an Olympus IX73 microscope with attached Micropoint focusing system and pulsed-dye laser (Andor). 99 pulses were administered to the ventral surface of the posterior cardinal vein 5 somites caudal to the anal pore at 2 or 3 dpf or 3-5 somites caudal to the anal pore on the dorsal surface of the dorsal aorta at 2, 3, or 5 dpf. At 2 minutes, larvae were examined for formation of an occlusive thrombus and scored as either occluded or non-occluded by a blinded observer prior to genotyping. Venous injury was performed at 3 dpf, while arterial studies were done at 5 dpf since that is when thrombocytes first appear in circulation in substantial numbers. [43] Spontaneous thrombosis assay Tg(fabp10a:fgb-eGFP) larvae were raised to 5 dpf and screened for transgene status by presence of signal in the liver. Larvae were scored via direct visualization of the venous circulation under high-powered fluorescent microscopy. The degree of thrombosis was scored by a single blinded observer on a scale of 0-3 with 0 being no signal and 3 being diffuse, intense signal, similar to previous work. [29] During analysis, scores �1 were considered negative and scores �2 were labeled positive.

Membrane bound recombinant TF production
The cDNA sequences for zebrafish f3a and f3b were identified from genomic reference sequence, [68] then cloned and confirmed by Sanger sequencing. Nucleotides coding for the predicted leader peptide and cytoplasmic tail were removed. A pelB leader peptide (to direct expression to the E. coli inner membrane) and an HPC4 epitope tag were added (S2 Table). The resulting nucleic acid sequence was codon optimized for E. coli expression and synthesized as a gBlock (IDT, Coralville, Iowa), subcloned into the pET-26b (+) expression vector, and the validated plasmids transformed into BL21 (DE3) competent cells. For expression, transformed E. coli were grown overnight at 37˚C in minimal MDG medium [69] plus 100 μg/ mL kanamycin. The next day, 50 μL culture were diluted to 1 mL with water and the A 600 measured (result multiplied by 20 to calculate A 600 of undiluted culture). E. coli cells were collected by centrifugation (15 minutes at 6,000g), and then resuspended in BYE medium (50 g/L Bacto yeast extract, 12.75 mM KH 2 PO 4 , 54 mM K 2 PO 4 , 0.5% w/v glycerol, 100 μg/mL kanamycin) to achieve an estimated A 600 of 4.5, and shaken at 25˚C for 2 hours. IPTG was then added to 100 μM and cultures allowed to shake for another 4 hours, after which cells were collected by centrifugation and stored frozen at −80˚C.

Ex vivo clotting assays
Lake trout blood (courtesy of Kevin Keeler at the United States Geological Survey, Ann Arbor, MI) was collected by venous puncture and anticoagulated using 9:1 plasma:3.2% sodium citrate and centrifuged at 1500 x g for 15 minutes to isolate thrombocyte-poor plasma. Plasma, TF liposomes, and 25 mM CaCl 2 were mixed at a 1:1:1 ratio and placed on a rocker at 25˚C. The tubes were continually observed through repeated tilting until the solution became solid and a clot was apparent. Clots were scored by a blinded observer continuously for 5 minutes and then checked again at 15 minutes.

Chemical treatment
Larvae were placed in a light protected plate and stock solutions of epinephrine and hydrocortisone in DMSO were added to a final concentration of 25 μM epinephrine, 0.01% hydrocortisone, 0.1% DMSO, and incubated in the dark at 28.5˚C.

Computational resources and statistical analysis
Protein structure, conservation and modifications were predicted using Clustal Omega, [72] NetNGlyc, [73] NetPhos 3.1 [39] and TMHMM. [74] Occlusion data and ex vivo clotting assays were analyzed using Mann-Whitney U or binomial proportion tests. Significance testing, graphs, and survival curves were produced with Prism (Graphpad Software, California).