The Amino-Terminal Oligomerization Domain of Angiopoietin-2 Affects Vascular Remodeling, Mammary Gland Tumor Growth, and Lung Metastasis in Mice

The angiopoietin (Angpt1, 2 and 4)/Tie receptor tyrosine kinase signaling pathway (ANGPT, human; Angpt, mouse) regulates vascular remodeling in development and during tumorigenesis (1). Angpt1 is an agonistic Tie2 ligand having a stabilizing and anti-inflammatory role (2), whereas Angpt2 is implicated in vascular remodeling and endothelial cell (EC) activation (3). Angpt4 is required for venous morphogenesis in the retina (4).

In ECs, Angpt2 is stored in and released from EC-specific storage granules, the Weibel–Palade bodies (WPB; ref. 5). In mouse development, Angpt2 is required for lymphangiogenesis, development of retinal blood vessels and regression of hyaloid vessels in the eye (6). In cancer, Angpt2 promotes vascular sprouting and destabilization, inflammation, and metastasis (1). ANGPT2 expression pattern (low in normal tissue, induced in many cancers) and cancer-promoting effects have identified ANGPT2 as a potential target molecule for anticancer treatments (1); however, treatment responses in patients have been less promising (7, 8) as expected on the basis of preclinical studies. Clearly, thorough characterization of ANGPT2 is necessary to fully evaluate its potential as a target molecule for anticancer drugs.

Alternative mRNA splicing is a common mechanism for generation of multiple isoforms from a single gene and it represents a major mechanism to increase genetic diversity. In normal development, expression of protein isoforms is carefully regulated, whereas in cancer significant alterations in the transcriptome are caused by differential or dysregulated mRNA splicing (9). Alternative splicing generates the ANGPT2₄₄₃ isoform lacking exon 2 (10) that in full-length ANGPT2 mRNA encodes for 53 amino acids of the coiled-coil domain (CCD) responsible for ligand oligomerization. In cultured human ECs, the relative abundance of ANGPT2₄₄₃ is approximately 10% of ANGPT2 (10, 11) and its expression is also documented in human macrophages (10). ANGPT2₄₄₃ is upregulated in tumor cell lines and human tumor biopsies (10, 12). Interestingly, some tumor cells have been shown to express solely ANGPT2₄₄₃ (10), and protein forms possibly representing smaller isoforms are reported in hepatitis B–infected liver (13) and in the plasma of patients with acute lung injury (14). Apart from human malignancies, ANGPT2₄₄₃ expression is also reported in canine adrenocortical tumor tissue (15). Our analysis of ANGPT2 splice variants from The Cancer Genome Atlas (TCGA) mRNA-sequencing data (16) revealed that in normal tissue, the ratio of ANGPT2⁴⁴³ to full-length ANGPT2 is less than 20% whereas in breast cancer patient tissues the ratio is increased (Fig. 1A). Collectively, these data suggest that ANGPT2₄₄₃ is expressed in low abundance in unstressed tissues and upregulated in tumorigenesis and in activated ECs. The in vivo functions of ANGPT2₄₄₃ variant and whether they differ from those of ANGPT2 are, however, unknown.

Here, we hypothesized that alternative protein forms increase diversity of the functions of Angpt2 and generated a targeted knock-in mouse allele expressing Angpt2₄₄₃. The Angpt2⁴⁴³ mice allowed us to address the physiological importance of a specific domain of Angpt2 in development, in primary tumor growth, and in metastasis by using genetic and syngeneic breast cancer and lung metastasis models. We found that alternative protein forms were less oligomerized, potent competitive inhibitors of TIE2 activation affecting venous remodeling, tumor vasculature, and metastasis. Our findings identify the in vivo importance of the N-terminal domain of ANGPT2 and provide new insights into mechanisms underlying the versatile functions of ANGPT2 in cancer.

ANGPT2 isoform expression data in human breast cancer were analyzed using TCGA cohort data from the TSVdb database (http://www.tsvdb.com).

Human umbilical vein ECs (HUVEC, C-12203, PromoCell) were cultured in Medium 200 (M-200–500, Cascade Biologics) supplemented with growth supplement (LSGS, S00310, Cascade Biologics), penicillin–streptomycin (P/S, Sigma-Aldrich) and 10% heat-inactivated FBS (HI-FBS; S1810–500, Biowest) in cell culture plates pre-coated with Attachment Factor (123–500, Cell Applications). HUVECs were transfected with human TIE2 (TIE2-WT), ANGPT2, ANGPT2-Flag, ANGPT2-GFP, ANGPT2₄₄₃, ANGPT2₄₄₃-GFP, ANGPT2₄₄₃-His or ANGPT2_DAP via retroviral gene transfer (DNA constructs are listed in Supplementary Table S1; ref. 17). An African green monkey kidney fibroblast-like cell line (COS-1, ATCC, RRID:CVCL_0223) was cultured in DMEM (31966–021, Gibco) supplemented with 10% HI-FBS and 1% P/S. The human cervical epidermoid carcinoma cancer cell line (CaSki, ATCC, RRID:CVCL_1100) was maintained in RPMI-1640 medium (30–2001, ATCC) supplemented with 10% HI-FBS and 1% P/S, and retrovirally transfected to express H2B-mCherry. C57BL-syngeneic mouse skin melanoma cell line B16F10 (ATCC, RRID:CVCL_0159) was cultured in advanced DMEM F12 (11320033, Gibco) supplemented with 2 mmol/L l-glutamine (G7513, Sigma-Aldrich), 10% HI-FBS and 1% P/S. E0771 mouse breast cancer cell line (CH3 BioSystems) was cultured in RPMI-1640 (ATCC) supplemented with 10% HI-FBS and retrovirally transfected to express ANGPT2, ANGPT2₄₄₃ or ANGPT2_DAP when indicated. Cells purchased from the ATCC were authenticated by the manufacturer. HUVECs were primary cells from pooled donors used at low passages and authenticated by the expression of EC-specific genes and monolayer morphology. E0771 cells were not authenticated after receiving from the vendor. Non-primary cells were kept in culture for maximum 2 months, and all cells were tested Mycoplasma-negative upon receival and regularly while performing the experiments with the EZ-PCR Mycoplasma test kit (Biological Industries). Details of protein production in COS-1 cells, HUVECs or E. coli, ELISA assays, ANGPT2_DAP peptide sequence analysis and in vitro experiments are described in the Supplementary Materials and Methods.

Whole-cell lysates, conditioned media or immunoprecipitated cell or tissue samples were separated under reducing or non-reducing conditions with SDS-PAGE and analyzed by Western blotting as described in ref. 4. Purified human recombinant ANGPT2 protein (R&D Systems) was included as a control when indicated. All antibodies and special reagents are listed in Supplementary Table S2. See Supplementary Materials and Methods for detailed immunoprecipitation protocols.

Experiments involving mice were performed under permission from the Finnish Animal Experiment Board following the regulations of EU and Finnish legislation. A mouse allele expressing solely Angpt2₄₄₃, named as Angpt2⁴⁴³, was generated using a conventional BAC recombineering technology. The targeting construct lacked the 156 base pair (bp) long exon 2 and intronic sequences 834 bp upstream and 1620 bp downstream from exon 2. In the targeting construct, deleted region was replaced with a neo cassette, including PGK/EM7 promoter and a polyA sequence for guiding neomycin expression as an independent transcript. Final targeting vector contained 9.5 and 1.8 kb genomic arms and a 3 kb neo cassette. The SalI-linearized targeting construct was electroporated into B6 ES cells and selected with G418 (Gibco). Genomic DNA isolated from resistant colonies was screened by PCR. Cells from correctly targeted clones were injected into C57BL blastocysts, and two separate mouse lines were generated by standard methods. For Angpt2⁴⁴³ allele expression analysis, PCR from lung cDNA was performed with primers spanning from exon 1 to 4 (Supplementary Table S3) allowing detection of Angpt2 isoforms. PCR products were separated on a 2% agarose gel and verified by sequencing. Angpt2⁴⁴³ mice were kept in C57BL/6N genetic background. Angpt4^−/− mouse line has been described before (4), and Angpt2^−/− mouse line is described in the Supplementary Materials and Methods.

Generation of polyomavirus middle T antigen mouse breast cancer model FVB/N-Tg(MMTV-PyVT)634Mul (The Jackson Laboratories, hereafter PyMT) has been described elsewhere (18). Male FVB PyMT mice were crossed with heterozygous female C57BL/6N Angpt2^+/443 mice, resulting in four different genotypes in F1 generation (Angpt2^+/+, Angpt2^+/443, PyMT; Angpt2^+/+, PyMT; Angpt2^+/443) all in the same FVB-C57BL/6N hybrid genetic background. The tumor-bearing female mice were sacrificed at 13 and 18 weeks of age and nontumor control mice were sacrificed at 19 weeks. Genotyping PCR primers for the neo gene-cassette, PyMT transgene and Angpt2 are presented in Supplementary Table S3.

All histological tissue staining methods, tissue imaging, and analysis are described in the Supplementary Materials and Methods. All antibodies are listed in Supplementary Table S2.

Fluorescent beads were administrated into carotid artery of 5-months-old mice as previously published (4) except that Alexa Fluor 594–conjugated WGA-lectin (Thermo Fisher) was injected 5 minutes before the beads to visualize blood vessels.

MMTV-PyMT mice were injected with Evans Blue (EB; Sigma-Aldrich), after 30 minutes they were euthanized, vasculature was flushed from excess EB and tumor EB content was measured. More detailed protocol is presented in the Supplementary Materials and Methods.

Quantitative RT-PCR was performed as described before (4) using GAPDH and β-actin as housekeeping genes. qPCR primers are presented in Supplementary Table S3.

B16F10 melanoma cells were trypsinized, washed twice with serum-containing medium and twice with ice-cold Hank's Balance Salt Solution (HBSS; Sigma-Aldrich), strained through cell strainer and cell number was adjusted to 1 × 10⁶ cells/mL in HBSS. Adult mice were anesthetized with isoflurane and 1.5 × 10⁵ cells (150 μL)/mouse was injected into tail vein. After 14 days, the lungs were collected, and metastases analyzed with optical projection tomography as described in the Supplementary Materials and Methods.

Lungs from one-year-old mice were collected and immediately weighed, dried at 65°C for 3 days and weighed again to establish wet/dry lung weight ratio (19).

Trypsinized E0771 cells were washed twice with serum-containing medium and twice with HBSS, and 1 × 10⁷ cells/mL were suspended in RPMI-1640 medium, and 5 × 10⁵ cells (50 μL) were injected into four mammary fat pads (right and left thoracic and inguinal)/12-weeks-old female gene-targeted or C57BL/6N mouse. Tissues were collected after 29 days.

The Mann–Whitney U test was used in Figs. 1A, 4M and N. Otherwise, comparisons between two groups were performed using two-tailed unpaired Student t test (in Figs. 1E, K, 4 and 5) and between multiple groups with one-way ANOVA, followed by Tukey (in Figs. 1H, 2, 3, 6 and 7) post-hoc tests using OriginPro software. Statistical significances are marked into figures with an asterisk: *, P < 0.05; **, P < 0.01; and ***, P < 0.001. Box plots show median as a vertical line, average as a small box, 25%–75% interquartile range as a box, 5th and 95th percentile as whiskers and outliers as dots. Dot plots show measurement points and a bar as a mean. In column charts, data are presented as mean ± SD.

Analysis of cell lysates from retrovirally transduced HUVECs using polyclonal ANGPT2 antibody showed higher and lower molecular weight (MW) protein forms for ANGPT2 and ANGPT2₄₄₃ under reduced conditions in Western blot (WB) analysis. Analysis of conditioned media (CM) from the same cells under reducing conditions indicated that HUVEC-secreted ANGPT2 is mainly a single MW protein, whereas ANGPT2₄₄₃ showed two distinct bands (Fig. 1B). Under nonreducing conditions, both isoforms run mainly as dimers in the HUVECs' CM, however, in ANGPT2₄₄₃, an increase in monomeric 50 kDa protein form was detected (Fig. 1C). In addition, when compared with ANGPT2, ANGPT2₄₄₃ was less multimerized in accordance with a previously proposed role of the N-terminus in ligand oligomerization and multimerization (20). To investigate in detail the ANGPT2₄₄₃-derived low MW protein form, His-tagged proteins were immunoprecipitated for peptide analysis. Mass spectroscopy analysis of trypsin-digested samples identified the first peptide for the uncleaved ANGPT2 and ANGPT2₄₄₃ starting from glutamine 33 (Q33). In contrast, in ANGPT2₄₄₃-derived low MW protein, the first peptide identified started from aspartic acid 68 (D68; Fig. 1D). This cleavage site (…VQR/DAP…) was confirmed by N-terminal peptide sequencing. The resulting ANGPT2₄₄₃-derived 50 kDa protein was named as ANGPT2_DAP based on the first three N-terminal amino acids. Analysis of potential protease cleavage site (http://prosperous.erc.monash.edu/) revealed consensus sequence for proprotein convertase 1 and furin-like protein convertase (AVQR/DA). In line with the in silico analysis, furin inhibitor decreased the extent of ANGPT2_DAP cleavage (Fig. 1E). ANGPT2_DAP lacked the entire super-clustering domain (Supplementary Fig. S1) needed for formation of higher order multimers, consistent with the enrichment of monomeric ANGPT2 proteins. Similarly to HUVECs, CM from the ANGPT2₄₄₃-transfected COS-1 cells showed less higher oligomerized/multimerized forms and two distinct monomeric forms whereas ANGPT2_DAP run as a non-disulfide linked monomer in SDS-PAGE (Fig. 1F and G). In solid-phase binding assay, ANGPT2₄₄₃ and ANGPT2_DAP bound to TIE2 but with lower binding affinity (Fig. 1H). This was in line with previous results, indicating that designed low-oligomerized forms of ANGPT1 bind less to TIE2 (21). In addition to TIE2, ANGPT2 can bind to integrins to enhance sprouting angiogenesis (22) and endothelial destabilization (23). Our special interest was to investigate ANGPT2 binding to α5β1 integrin, which is involved in breast cancer metastasis (24), abnormal vascular remodeling (25), and leakage (26). C-terminal ANGPT2 fragment (K275 to F469, common to all variants) bound to α5β1 at a relatively high concentration but the N-terminal (E72 to A274) fragment did not (Fig. 1I). Binding of full-length ANGPT2 showed significantly higher affinity to α5β1 than the C-terminal fragment, whereas affinity of ANGPT2₄₄₃ was somewhat lower and ANGPT2_DAP did not bind to α5β1 at 2 nmol/L concentration (Fig. 1J). To test whether oligomerization of ANGPT2 is necessary for binding to α5β1, disulfide bonds in the oligomeric full-length ANGPT2 were reduced using dithiothreitol (DTT). DTT treatment efficiently reduced α5β1 binding (Fig. 1K).

To test whether ANGPT2 forms mediate differential biological effects, ligand isoforms were first compared in TIE2 activation assays as single ligands or in the presence of ANGPT1 or ANGPT4. Similarly to ANGPT2 (27), ANGPT2₄₄₃ acted as a weak TIE2 agonist (about 20% of tyrosine phosphorylation of ANGPT1-stimulated TIE2) whereas ANGPT2_DAP did not activate TIE2 (Fig. 2A, left). ANGPT2₄₄₃ has been shown to be more potent competitive inhibitor of ANGPT1 than native ANGPT2 (10). In our experimental setting, ANGPT2₄₄₃ was approximately twice as effective as ANGPT2 to reduce ANGPT1-induced TIE2 phosphorylation, whereas ANGPT2_DAP was the most potent competitive inhibitor (Fig. 2A, right). In EC–EC junctions, ANGPT2 inhibits ANGPT1-induced TIE2 activation (17), and in this subcellular compartment ANGPT2₄₄₃ and ANGPT2_DAP were more potent to do so (Fig. 2B). Similarly to ANGPT1 blocking, low oligomeric ANGPT2 forms were effective to reduce ANGPT4-induced TIE2 activation in EC–EC junctions (Supplementary Fig. S2). WB analysis (Fig. 2A) suggested that ANGPT2 and ANGPT2₄₄₃ can activate TIE2 at low level when present as single ligands. In the retracting cell edges, a cell compartment that does not contain VE-PTP (17) downregulating the agonistic role of ANGPT2 (28), ANGPT2 and ANGPT2₄₄₃ induced some TIE2 activation but ANGPT2_DAP did not (Supplementary Fig. S3).

We next compared ANGPT2 forms in functional EC assays. In cultured HUVECs (that express ANGPT2 but no other angiopoietins at significant levels), endogenous ANGPT2 increases cell viability (29, 30). ANGPT2 mRNA silencing reduced HUVEC cell number, and viability was improved when these cells were cultured in the presence of ANGPT2 or ANGPT2₄₄₃ but not with ANGPT2_DAP (Fig. 2C). In fibrin gel model of angiogenesis, native ANGPT2 promotes tube formation (27). Similarly, in our sprouting assay in fibrin gel, ANGPT2 elongated sprouts, whereas other forms did not (Fig. 2D). In contrast, lower oligomeric ANGPT2₄₄₃ and ANGPT2_DAP increased the number of sprouts, suggesting that full-length and lower oligomeric ANGPT2 forms may have complementary functions. We have previously shown that substrate-bound native ANGPT2 weakly supports HUVEC adhesion and spreading (31). This was also true for lower oligomeric ligand forms (Fig. 2E), indicating that extracellular matrix (ECM)–bound (negative cellular effect) and soluble ANGPT2 (weak agonist) have differential cellular effects.

As indicated in Fig. 1A, ANGPT2 and ANGPT2⁴⁴³ mRNAs are simultaneously expressed in tumor tissue. To test whether the presence of N-terminally variable ANPGT2 proteins may negatively affect their assembly into oligomeric forms (dominant-negative effect), COS-1 cells were retrovirally transduced either with single ANGPT2-Flag or ANGPT2₄₄₃-His or both (Fig. 2F). Comparison of CMs collected from single- and double-transduced cells revealed that simultaneous expression of ANGPT2-Flag and ANGPT2₄₄₃-His did not affect their mobility in nonreducing SDS-PAGE analysis (Fig. 2G), indicating that these N-terminally different proteins do not form intermolecular oligomers.

To investigate the in vivo importance of Angpt2₄₄₃ variant, a gene-targeted mouse allele was generated, in which alternative exon 2 was replaced with a neo gene-cassette (Supplementary Fig. S4A), thus expressing solely Angpt2⁴⁴³. RT-PCR analysis confirmed that the sequence encoded by exon 2 was lacking in homozygous Angpt2^443/443 mice mRNA (Supplementary Fig. S4B–S4D).

Development of mouse retinal vasculature is particularly sensitive for changes in the Tie/Angiopoietin signaling pathway (32) and was used as a model to analyze possible vascular phenotypes in Angpt2⁴⁴³ mice. As Angpt2 is required for formation of the retinal capillary plexus (6, 33), postnatal and adult retinas were analyzed (Fig. 3A–C). At postnatal day (P) 6, we found no delay in sprouting angiogenesis; however, Angpt2^443/443 retinas showed impaired lumen formation in the veins, which is dependent on Angpt4 (4) and Tie2 (Fig. 3D–E; ref. 34). In normal mouse retina, two or occasionally three main veins extend to the retinal periphery and branch to form an annular structure (Supplementary Fig. S4E). Interestingly, in Angpt2^443/443 as well as in Angpt4^−/− mice, the number of veins extending retinal periphery was elevated (Fig. 3F), indicating that Angpt2⁴⁴³ inhibits Angpt4-dependent vascular development. Also, in the heterozygous Angpt2^+/443 mice vein diameter was diminished and number of veins extending to periphery elevated (Fig. 3B–C and E–F), suggesting a dose-dependent effect of Angpt2₄₄₃. To rule out the possibility that defects in Angpt2^+/443 and Angpt2^443/443 mice are caused by differences in Angpt2 protein levels or location, P4 retinas were stained with Angpt2 antibody. Similar levels of staining in retinal tip cells and neuronal layer were detected in all genotypes (Supplementary Fig. S4F). These data suggested that exon 2 deletion does not have a loss-of-function effect, which we further investigated by comparing retinal phenotype to Angpt2-deficient mice. In contrast with Angpt2^443/443 mice, Angpt2^−/− retinal vasculature lacked hierarchical organization of arterial and venous trees (Supplementary Fig. S4G). Furthermore, retinal vasculature in Angpt2^−/− mice did not extend to the periphery, resulting in avascular areas that were not observed in other genotypes (Supplementary Fig. S4H). Instead, exon 2 deletion more closely resembled the phenotype observed in Angpt4^−/− mice and may rather interfere with Tie2 signaling that promotes vessel enlargement and vein morphogenesis (4, 34, 35). This was confirmed by retina whole-mount staining of phosphorylated Tie2 (pTie2) and total Tie2 that showed decreased pTie2/Tie2 ratio in Angpt2^443/443 retinal veins at P8 (Fig. 3G and H) when vein morphogenesis occurs.

Developmental defect in Angpt4^−/− veins also includes maturation abnormality of SMCs (4). Reduced αSMA staining was indeed seen in some Angpt2^443/443 P12 retinal veins (Fig. 3I and J) but this was less severe than reported in Angpt4^−/− veins. To investigate whether venous drainage function is reduced in Angpt2^443/443 mice, fluorescent beads were injected into carotid artery and their distribution in retinal veins was investigated shortly after injection. Some veins in Angpt2^+/443 and Angpt2^443/443 mice had increased accumulation of beads (Fig. 3K); however, decrease in drainage was not as prominent than observed in Angpt4^−/− mice, in which venous narrowing is also more severe (4).

Primary tumor growth in the MMTV-PyMT model has previously been shown to be dependent on Angpt2 (36). To test whether Angpt2₄₄₃ has a differential effect, as one can expect based on its proteolytic processing, lower integrin binding and increased potency to interfere TIE2 signaling, we generated PyMT; Angpt2^+/443 mice by crossing transgenic PyMT (18) males (in pure FVB/N background) with Angpt2^443/443 females (in pure C57Bl/6N background). This allowed us to investigate tumor growth in a homogenous F1-hybrid genetic background. As both full-length ANGPT2 and ANGPT2⁴⁴³ are simultaneously expressed in human tumor tissues, heterozygous state (Angpt2^+/443) in mice should well represent the ANGPT2 isoform balance in cancer (Fig. 1A). Two cohorts of tumor-bearing mice were sacrificed at 13 and 18 weeks of age and the mammary glands and tumors were harvested for gravimetric (Fig. 4A) and carmine alum (Fig. 4B) analysis. Mammary gland weights were equal in nontumor Angpt2^+/+ and Angpt2^+/443 controls and showed no alterations in their vasculature. A trend for decreased tumor weight was observed in PyMT; Angpt2^+/443 already at 13 weeks of age. In the older cohort, the total primary tumor burden in the PyMT; Angpt2^+/443 mice was smaller than in the PyMT; Angpt2^+/+ mice (Fig. 4A).

Angpt2 affects tumor angiogenesis (36), vessel lumen size (37), pericyte (PC) coverage (37, 38) in tumor vasculature, and vascular leakage (39). Immunofluorescent microcopy revealed decreased blood vessel (CD31) count (Fig. 4C and D) and reduced number of lumen structures in the tumor vessels (Fig. 4E) in PyMT; Angpt2^+/443 mice. Those vessels where lumens were identified, however, showed no differences in the lumen size when compared with the PyMT; Angpt2^+/+ mice (Fig. 4F), suggesting that not all tumor vessels express or respond to Angpt2₄₄₃ equally. There was a trend for increased number of PC-deficient vessels in the PyMT; Angpt2^+/443 whole-tumor sections stained for CD140b and CD31 (Fig. 4G). On the basis of transmission electron microscopy (TEM), in which stromal and tumor compartments can be reliably distinguished in high resolution, PC abnormalities were primarily found in PyMT; Angpt2^+/443 stromal blood vessels, whereas PC coverage was not different in vessels that were located within the tumor cell mass without defined perivascular ECM (Fig. 4H–J). In addition, vascular permeability was increased in the PyMT; Angpt2^+/443 mammary tumors (Fig. 4K).

To rule out that slower tumor growth is not caused by lack of Angpt2₄₄₃ allele expression, we investigated mRNA expression using Angpt2⁴⁴³-specific qPCR primers spanning from exon 1 to exon 3 (forward) and exon 4 (reverse; Supplementary Table S3). Angpt2⁴⁴³ mRNA was detected in PyMT; Angpt2^+/443 but not in PyMT; Angpt2^+/+ tumors (Supplementary Fig. S5A). On the basis of qPCR primers from exons 8 and 9, common to both variants, total Angpt2 expression was at the same level (Supplementary Fig. 5B). Angpt1 and Vegfa are well-known factors that can modulate destabilizing and angiogenic effects of Angpt2 on vasculature (1). Angpt1 mRNA levels remained unchanged, whereas Vegfa levels were increased in PyMT; Angpt2^+/443 mice (Supplementary Fig. S5B). Expression levels of CaIX, Eno1, Glut1, Ldha, and Pgk1, which upregulation would indicate hypoxic conditions in mouse breast cancer tumors (40, 41), were found to be unaltered (Supplementary Fig. S5B). Number of apoptotic ECs, PCs, and tumor cells as well as necrotic areas in tumors were not changed (Supplementary Fig. S5C–S5F).

Metastases were annotated from hematoxylin and eosin–stained lung sections (Fig. 4L). PyMT; Angpt2^+/+ mice showed larger metastatic area (Fig. 4M) and a trend for increased metastasis count (Fig. 4N, P = 0.057) when compared with PyMT; Angpt2^+/443 mice. Notably, 67% of the PyMT; Angpt2^+/443 mice had less than 0.5% metastasis versus the lung area, whereas 100% of the PyMT; Angpt2^+/+ lungs had more than 0.5% metastasis (Fig. 4M).

To investigate further the observed heterogeneity in lumen-forming and PC-deficient vessels in MMTV-PyMT tumor vasculature, we analyzed Angpt2 mRNA and protein localization using Angpt2 in situ hybridization (4) and an Angpt2-specific antibody (25). Notably, only a subset of ECs showed Angpt2 expression (Fig. 5A–D) as has previously been shown also in the vasculature of human glioma (42). Interestingly, a large proportion of Angpt2 immunoreactivity was located in the perivascular matrix (Fig. 5E–H) in both PyMT; Angpt2^+/+ and PyMT; Angpt2^+/443 tumors (Fig. 5H and I).

As Angpt2 immunostaining was surprisingly more intense in the subendothelial ECM than in the ECs where it was produced, we next investigated ECM secretion of ANGPT2 further using ANGPT2-GFP HUVECs. Additional motivation to investigate ECM secretion mechanism was the observation that ECM-bound and soluble ANGPT2 showed differential effects (Fig. 2) and that how paracrine, EC-produced ANGPT2 can affect PCs that locate within the perivascular ECM. In stable adherent HUVECs, both ANGPT2-GFP and ANGPT2₄₄₃-GFP were located in matured WPBs containing also von Willebrand Factor (vWF; Supplementary Fig. S6A). To analyze ECM secretion, we used total internal fluorescence (TIRF) microcopy to visualize ANGPT2-GFP with high axial resolution in freshly plated, actively spreading HUVECs where ECM constitutes are secreted and EC–ECM contacts are formed and remodeled (Supplementary Movies S1–S2). Interestingly, ANGPT2-GFP was detected on the basal cell membrane first as vesicles that emerged to the TIRF field. These vesicles were then fast immobilized when they came in contact with the ECM (Supplementary Movie S2). In fixed and vWF-stained ANGPT2-GFP HUVECs, TIRF imaging revealed co-localization of ANGPT2-GFP and vWF in the subendothelial ECM (Supplementary Fig. S6B and S6C). A previous study suggested that ANGPT2 and vWF are secreted from ECs by exosomes (43). However, our analysis of the ultrastructure of the EC–ECM interphase revealed ANGPT2 immunostaining at the basal cell membrane and ECM interphase but not within the membrane-covered extracellular vesicles (Supplementary Fig. S6D–S6H).

Angpt2 blocking has previously been shown to decrease lung metastasis in a B16F10 metastasis model (44), so we next investigated lung colonization of syngeneic B16F10 melanoma cells in Angpt2^+/+, Angpt2^+/443 and Angpt2^443/443 mice using optical projection tomography of whole lung lobes (Fig. 6A and B). In Angpt2^+/+ mice, extensive B16F10 colonization (>3% of total lung volume) was observed in 7.1% of lung samples, whereas in Angpt2^+/443 mice, the percentage was 44.4% and in Angpt2^443/443 mice, 60%. Increased volume of melanoma cell colonies was observed in Angpt2^443/443 mice when compared with Angpt2^+/+ mice (P = 0.008 in two-tailed t test). In the Angpt2^+/443 mice, the average volume of melanoma colonies was roughly in between Angpt2^443/443 and WT mice; however, due to notably high variation in Angpt2^+/443 mice, comparisons of all three groups in ANOVA resulted in a P value of 0.052 (Fig. 6B).

In a previous study, Angpt2 overexpression increased whereas Angpt2-blocking antibodies alleviated vascular destabilization in the B16F10 model (44). To investigate whether Angpt2₄₄₃ delivers stronger destabilizing effects than native Angpt2, ultrastructure of the lung vasculature was investigated using TEM. As illustrated in Fig. 6C and D, the EC–EC adherent junction area (Fig. 6E) and the length of EC–EC contacts (Fig. 6F) were decreased in both Angpt2^+/443 and Angpt2^443/443 host vessels close to the metastasis foci. EC degenerative changes (cell organelle–free cytoplasm, pyknosis, EC–ECM detachment, and vacuolization) were most frequently observed in B16F10-injected Angpt2^443/443 mice (Fig. 6G). In nontumor cell-loaded lungs, we found no changes in pulmonary vasculature (Tie2 phosphorylation, markers for permeability, lung wet/dry weight ratio indicative of edema) in either Angpt2^+/+, Angpt2^+/443 or Angpt2^443/443 mice (Supplementary Fig. S7A–S7C). This is in line with previous data, indicating that in the normal lung ECs Angpt2 is not significantly expressed (45), but it is induced in the host vessel ECs in response to tumor cell injection (46). Accordingly, Angpt2 protein was localized in some ECs at the margin of B16F10 colonies (Supplementary Fig. S7D). In comparison with MMTV-PyMT mammary tumors, B16F10 cells were closely associated with host pulmonary vessels, induced only little intratumoral angiogenesis, did not form tumor stroma and the pulmonary ECM did not show Angpt2 immunoreactivity. Vegfa, which at low levels enhances Angpt2-induced vascular regression (47), was slightly downregulated in B16F10-colonized lungs of Angpt2^+/443 and Angpt2^443/443 mice, whereas levels of Angpt1 were not significantly changed (Supplementary Fig. S7E). As in PyMT tumors, total Angpt2 mRNA levels were equal in all genotypes (Supplementary Fig. S7E), whereas the Angpt2₄₄₃ variant-specific primer pair showed Angpt2⁴⁴³ allele expression only in the Angpt2^+/443 and Angpt2^443/443 mice but none in the WT mice (Supplementary Fig. S7F). Our observation that WT mice do not naturally express Angpt2⁴⁴³ goes along with the GenBank database (https://www.ncbi.nlm.nih.gov/genbank/) and indicates that ANGPT2⁴⁴³ is not expressed in all mammalian species, and that the splicing process generating Angpt2⁴⁴³ is not completely conserved.

To test whether vascular destabilization by ANGPT2₄₄₃ could also be observed in human cell–based assay, we performed in vitro EC transmigration assay (48) using CaSki cancer cell line with non-transduced and ANGPT2 or ANGPT2₄₄₃-expressing HUVECs. In this assay, ANGPT2₄₄₃ but not ANGPT2 increased tumor cell transmigration (Fig. 6H).

Results from MMTV-PyMT and B16F10 models were inconsistent with respect to the lung metastasis potential of Angpt2₄₄₃. To test all three genotypes in a model that allowed analysis of both mammary gland primary tumors and lung metastases, we next used orthotopic injection of E0771 breast cancer cells into mouse mammary fat pad. In the first experimental setting, we injected parental E0771 cells into Angpt2^+/+, Angpt2^+/443, and Angpt2^443/443 mice. As E0771 model is not strongly metastatic and to test the ANGPT2_DAP variant, we also generated E0771 cells stably expressing ANGPT2 protein forms and transplanted the resulting E0771^ANGPT2, E0771⁴⁴³, and E0771^DAP cells into wild-type C57BL mice. In all cohorts, we found no changes in the primary tumor weight (Fig. 7A). In addition, tumor blood vessel count and lung metastasis area were not changed in Angpt2^+/443 and Angpt2^443/443 mice (Fig. 7B–E). In contrast with the genetic MMTV-PyMT model, E0771 transplantation tumors contained only little ECM-rich stroma and Angpt2 protein was localized in CD31-positive vascular structures (Fig. 7C). Furthermore, frequency of Tie2-positive blood vessels was low in E0771 tumors when compared with MMTV-PyMT tumors (Supplementary Fig. S8A and S8B), providing likely explanation for differential tumor growth and vasculature in the presence of Angpt2_443.

E0771^ANGPT2, E0771⁴⁴³, and E0771^DAP cell lines produced ANGPT2 proteins in similar amounts based on WB analysis and ELISA from concentrated CM (Supplementary Fig. S8C and S8D) before their transplantation. In contrast with ECs, where ANGPT2 is stored in WPBs, E0771 cells immediately secreted ANGPT2 forms into CM as cell lysates were negative for ANGPT2. In E0771^ANGPT2, E0771⁴⁴³ or E0771^DAP mice, ANGPT2 was present in serum as measured by human ANGPT2 ELISA (Supplementary Fig. S8E). ANGPT2 and ANGPT2₄₄₃ serum concentrations were lower, possibly due to increased solubility of ANGPT2_DAP. Primary tumor blood vessel count and lung metastasis area were increased in E0771⁴⁴³ Angpt2^+/+ mice (Fig. 7B–E). As lung metastasis was more enhanced in mice transplanted with ANGPT2 variant-expressing E0771 cells, this cohort was investigated in more detail. As in B16F10-colonized lungs, ultrastructural analysis showed reduced cortical actin (Fig. 7F) and length of EC–EC junctions (Fig. 7G), and increased EC degeneration adjacent to lung metastasis (Fig. 7H). The junctional alterations were most prominent in E0771⁴⁴³ Angpt2^+/+ mice, and increased EC degeneration was present in E0771⁴⁴³ Angpt2^+/+ and E0771^DAP Angpt2^+/+mice (Fig. 7H).

The major obstacle for development of effective cancer treatments is biological diversity in cancer. Alternative mRNA splicing is a main mechanism to generate protein isoforms (9). Additional confusing factors are those isoforms that occur in human cancer but not in normal development or in the preclinical animal models. In this study, we characterized one such alternative isoform, ANGPT2₄₄₃, and investigated its relevance for mammary gland tumors and lung metastasis in mice. Our comparison with full-length ANGPT2 revealed that lower oligomeric ANGPT2 protein forms have different effects on TIE2 activation and integrin binding, the two key mechanisms how ANGPT2 regulates vascular remodeling. Metastasis is the major cause of cancer morbidity and mortality, currently providing one of the most important challenges in clinical oncology. Vascular normalization, which can be achieved by Tie2 activation and Angpt2 inhibition (44, 49), offers one possible strategy to generate less malignant tumor microenvironment and reduce metastases. ANGPT2 is identified as a molecular constituent of the metastatic microenvironment, and our study directly indicates the metastasis-promoting role of monomeric and lower oligomeric ANGPT2 protein forms in the lungs.

In mammary gland tumor and lung metastasis models, the extent of lung metastasis was associated with differential Tie2 expression in ECs and Angpt2 localization in the stromal ECM. In MMTV-PyMT tumor vasculature, Tie2 was expressed relatively frequently in ECs, and in this model Angpt2₄₄₃ negatively affected blood vessel count and number of lumen-forming vessels, leading to reduced primary tumor growth. Reduced tumor vasculature and lower primary tumor burden likely decreased seeding of metastatic cells into the lungs in PyMT; Angpt2^+/443 mice. In contrast with MMTV-PyMT, in E0771 tumor vasculature Tie2-positive vessels were less frequent, stromal ECM compartment was poorly developed, Angpt2 staining was not detected outside of ECs and no alterations in the tumor vasculature in E0771 Angpt2^+/443 or E0771 Angpt2^443/443 mice were found. Interestingly, when ANGPT2₄₄₃-expressing E0771 cells (E0771⁴⁴³) were transplanted into WT mice (E0771⁴⁴³ Angpt2^+/+ mice) blood vessel count and lung metastasis were increased, suggesting a complementary effect of Angpt2 and ANGPT2₄₄₃. A probable contributing factor resulting in differential phenotypes in E0771⁴⁴³ Angpt2^+/+ and E0771 Angpt2^+/443 mice is that in E0771⁴⁴³ cells ANGPT2₄₄₃ is constitutively expressed under viral promoter and on top of that homozygous Angpt2^+/+ allele expression results in double the amount of endogenous Angpt2 than in Angpt2^+/443 mice.

On the basis of published RNAseq data, Tie2 (in ECs) and Angpt1 (in stromal cells) are constitutively expressed in the matured healthy lungs whereas Angpt2 (in ECs) levels are very low (45). Without seeding of metastatic cells, Angpt2⁴⁴³ mice showed no deleterious effect in lung vasculature likely due to insignificant amount of Angpt2 expression. In the B16F10 model, Angpt2 protein was localized in host vessel ECs around the B16F10 cell colonies and ultrastructural analysis revealed EC destabilization in Angpt2^+/443 and Angpt2^443/443 mice. Ultrastructural changes suggestive for enhanced EC destabilization were also observed in the pulmonary vasculature in E0771⁴⁴³ Angpt2^+/+ lungs when metastatic cells were present. In E0771 model, ANGPT2_DAP did not affect primary tumor vasculature; as a large proportion of blood vessels were Tie2-negative and ANGPT2_DAP does not bind to ubiquitously expressed integrins, ANGPT2_DAP probably lacked a mechanism to enhance metastatic spread in primary tumors. In the pulmonary venules adjacent to lung metastasis, EC–EC junctional changes were less prominent in E0771^DAP than in E0771⁴⁴³ mice that may reflect their differential integrin binding shown to have a role in EC junctional destabilization (23). Although EC–EC junctions were less affected, E0771^DAP lungs with metastatic foci showed cellular changes suggestive for EC degeneration that may be caused by blocking of Angpt1/Tie2-induced pro-survival signaling. Collectively, these data indicated that the tumor- and lung metastasis–promoting or suppressing roles of ANGPT2₄₄₃ depend on the tumor microenvironment and that simultaneous expression of both ANGPT2 and ANGPT2₄₄₃ may have complementary effects on Tie2 inhibition and integrin binding. In addition, EC-destabilizing effect in ultrastructural studies was evident in those lungs where metastatic cell colonies were present, suggesting metastasis-promoting rather than metastasis-priming effect for Angpt2₄₄₃.

To investigate the mechanisms underlying differential effects of ANGPT2 forms, their molecular and cellular characteristics were compared in biochemical and EC assays and in mouse retina model of angiogenesis. Collectively, these data indicated that the main Tie2-dependent mechanism of how Angpt2₄₄₃ mediates its effects is competitive inhibition of Tie2 activation by Angpt1 and Angpt4. Tie2 signaling interfering function on the integrity of lung endothelium by Angpt2₄₄₃ is well in line with the effect of Angpt1 and Tie2 deletions resulting in increased colonization of melanoma cells as previously published (50). In addition, manipulations that artificially increase the oligomerization state of native ANGPT2 have been shown to increase TIE2 activation and vascular protection (51). In conditions where TIE2 expression is low (such as in tip cells in vascular sprouts and in some tumor ECs) ANGPT2 can also bind to integrins to promote sprouting and vascular destabilization (22, 23). Our data confirmed earlier results that ANGPT2 binding to α5β1 occurs via C-terminal peptide (52) and further indicated that N-terminal domain-mediated oligomerization increases ANGPT2 valency to α5β1 binding. In the mouse retina model, however, we found no sprouting phenotype suggestive for impaired ANGPT2/integrin signaling implying that the affinity of Angpt2₄₄₃ is still sufficient to support physiological angiogenesis in mouse retina.

On the basis of literature, a change in protein structure and folding may result in an increased susceptibility to proteolytic cleavage (53), which can be caused by alternative splicing (54) and involve CCDs (55, 56). We propose that lack of exon 2-encoded part of the CCD in ANGTP2₄₄₃ results in a conformational change that may unmask protease recognition sequences and/or remove the protective effect of native protein structure against proteolysis. Outstanding remaining question to completely understand the importance of ANGPT2₄₄₃ in cancer would be the investigation of native ANGPT2 protein forms in malignant tissues. Despite a vast number of different experiments, our analysis failed to reliably show Angpt2 and its fragments in WB from mouse tissue lysates. On the basis of ELISA, in our experimental settings, Angpt2 concentrations were lower than typical WB detection range. Nevertheless, it is interesting that Angpt2 fragments have been previously identified in mouse tumors (57) and in the liver of chronic hepatitis B patients (13), and that angiopoietin-like proteins are known to be cleaved by furin in vivo (58) in their N-terminuses.

Currently, trebananib (AMG386) is the most extensively evaluated ANGPT2 inhibitor in clinical cancer trials. In combination with doxorubicin, trebananib improved the objective response rate (ORR) and duration of response, but not the progression-free survival (PFS; ref. 8). In combination with paclitaxel, trebananib improved PSF and ORR (59); however, in combination with paclitaxel and carboplatin, trebananib showed no effect in a phase 3 clinical trial for ovarian cancer (7). As trebananib prevents the interaction of both ANGPT1 and ANGPT2 with TIE2 (60), it can interfere with ANGPT1-induced blood vessel stabilization and therefore not result in a best possible outcome. As binding to TIE2 occurs via ANGPT2 C-terminal domain (common to all ANGPT2 forms investigated), it is most likely that the pharmaceuticals that prevent ANGPT2 binding to TIE2 do not separate between different ANGPT2 forms. Our finding that less-oligomerized Angpt2₄₄₃ promotes lung metastasis and vascular destabilization further justifies the development of novel drugs targeting ANGPT2, such as ABTAA antibody that clusters ANGPT2 for TIE2 activation and more effectively prevents lung metastasis than Angpt2 blocking alone (49).

All in all, we observed marked heterogeneity between the tumor models. Among them, MMTV-PyMT best recapitulated the multistep processes of tumorigenesis and metastasis, relevant stromal microenvironment, Tie2-positive ECs and PC coverage. Common to all primary tumor models, expression of Angpt2₄₄₃ did not increase mammary gland tumor growth; and common to all lung metastasis models, Angpt2₄₄₃ decreased vascular integrity of EC-EC junctions adjacent to metastatic cell foci. From a translational perspective, TCGA provided the most comprehensive cancer patient data available that take into account mRNA splice variants. This included 1,081 primary tumor samples; however, number of individuals in advanced/metastatic stage was low (n = 7), thus lacking statistical power to reliably compare with our mouse data. Nevertheless, our study provides a rationale for further investigation of ANGPT2⁴⁴³ as a risk factor for the progression of metastasis in patients.

M.K. Kihlström reports a patent 20205501 pending to Finnadvance Ltd., as well as a personal grant from Academy of Finland on organ-on-chip research for the last 9 months at Finnadvance Ltd. L. Ruddock reports grants from Academy of Finland during the conduct of the study. No disclosures were reported by the other authors.

E. Kapiainen: Conceptualization, formal analysis, validation, investigation, visualization, methodology, writing-original draft, writing-review and editing. M.K. Kihlström: Conceptualization, formal analysis, validation, investigation, visualization, methodology, writing-original draft. R. Pietilä: Conceptualization, formal analysis, validation, investigation, visualization, methodology, writing-original draft. M. Kaakinen: Formal analysis, validation, investigation, methodology. V.-P. Ronkainen: Formal analysis, validation, investigation, methodology. H. Tu: Formal analysis, validation, investigation. A. Heikkinen: Investigation, visualization, methodology. R. Devarajan: Investigation. I. Miinalainen: Investigation. A. Laitakari: Investigation, methodology. M. Ansarizadeh: Investigation. Q. Zhang: Investigation. G.-H. Wei: Formal analysis, supervision. L. Ruddock: Resources, formal analysis, supervision. T. Pihlajaniemi: Resources. H. Elamaa: Formal analysis, validation, investigation, methodology. L. Eklund: Conceptualization, resources, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing-original draft, project administration, writing-review and editing.

Jaana Träskelin, Riitta Jokela, Teemu Karjalainen, Hannele Härkman, and Tiia Kylli are acknowledged for their excellent technical assistance. Biocenter Oulu Light Microscopy, Transgenic, Proteomics, Virus, Sequencing, and Electron Microscopy Core facilities and Oulu Laboratory Animal Center supported by University of Oulu and Biocenter Finland are acknowledged for technical advice, analysis of samples, and help with the interpretation of the results. Angpt2 antibody was a generous gift from Gou Young Koh. We also thank Heli Alanen for cloning and Yuko Ushida for producing ANGPT2 fragments in E. coli, Erja Tomperi and Katja Porvari for IHC, Tarja Piispanen and Biobank Borealis of Northern Finland for scanning lung sections, Peppi Karppinen for hypoxia qPCR oligos and Tie2 ab agarose beads, Florence Naillat, Abhishek Sharma, and Johanna Kekolahti-Liias for help with in situ hybridization, Peter Friedl and Cindy E. Dieteren for H2B-mCherry construct, and Proteomics Unit, Institute of Biotechnology, University of Helsinki for N-terminal sequencing. The results shown in Fig. 1A are based upon data generated by the TCGA Research Network (https://www.cancer.gov/tcga). This research was supported by the Academy of Finland grants (to L. Eklund; 251314, 136880, and 310986).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.