Introduction
The concept of tumor angiogenesis refers to the ability of a nascent tumor mass to promote vascularization in order to sustain its growth and survival [
1]. A fundamental outcome of this proposition is that by inhibiting the release of these factors, tumor development and hematogenous dissemination of tumor cells could be blocked, in practice setting the basis for anti-angiogenic therapy. Indeed, the presence of a vascular network to support the metabolism of cancer cells and to allow their spread to distant organs is a hallmark of solid malignancies [
2]. In 2004, the American Food and Drug Administration (FDA) approved the clinical use of bevacizumab, a humanized monoclonal antibody against vascular endothelial growth factor (VEGF)-A, in combination with standard chemotherapy in patients with metastatic colorectal cancer. Unfortunately, the use of this agent showed limited efficacy in breast cancer when administered together with the chemotherapeutic agent paclitaxel: despite almost doubling the progression free survival compared to paclitaxel alone (11.8 vs. 5.9 months), addition of bevacizumab did not extend the overall survival in patients (26.7 vs. 25.2 months) [
3]. Experimental studies later uncovered the shortcomings of such type of treatment, i.e., a response phase followed by adaptation to the therapy and bypass of the inhibition [
4,
5]. In fact, different modalities to overcome the blockade of angiogenesis are now recognized, from upregulation of pro-angiogenic factors to sprouting angiogenesis, vasculogenesis, intussusception, vessel cooption, vascular mimicry, and cancer stem cell-to-endothelial cell differentiation [
6].
In parallel to the increasing realization of inadequate efficacy of VEGF-targeted agents, the search for alternative pathways that regulates neo-angiogenesis ensued. In this context, considerable attention has been given to ALK1. ALK1 is a type I receptor of the TGF-β superfamily and mediates bone morphogenetic protein (BMP)9- and BMP10-induced signaling in the endothelium via the downstream mediators SMAD1/5/8 to orchestrate the development of blood vessels [
7]. Despite that ALK1 inhibitors exhibited promising results in a range of different mouse models of cancer [
8‐
11], clinical trials with the receptor decoy dalantercept (Acceleron Pharma) failed to show substantial benefit in different cancer types [
12,
13]. One of the major limitations of the drug development has been the absence of validated predictive biomarkers for ALK1 activity in cancer. Also, despite the increasing knowledge about the role of ALK1 in endothelial cell biology, the functional gene network acting downstream of ALK1 remains largely elusive, precluding informed predictions about suitable partners in combinatorial treatment regimens involving ALK1 blockade.
Here, we provide insights to the broader regulatory network associated with ACVRL1 expression in different human cancers. By interrogating publicly available data on gene expression, we reveal a previously unidentified association between ACVRL1 and genes controlling immune cell function. Moreover, analysis of the conserved set of ACVRL1-correlated genes in 14 different tumor types highlighted an 8-gene signature indicative of ALK1 activity. The gene with the highest median co-expression coefficient across all cancers is CLEC14A, which we infer to be a potential direct transcriptional target of ALK1 signaling through SMAD1/5. Taken together, our work prompts further validation of the use of CLEC14A as a surrogate marker for ALK1 activity to guide precision anti-angiogenic therapy in patients, possibly in combination with immunotherapy.
Materials and methods
Cell culture, in vitro stimulation, RNA extraction, and qPCR
Mouse endothelial MS1 cells were maintained in culture in DMEM (Invitrogen) supplemented with 10% FCS, penicillin, and streptomycin, in a humidified incubator at 37 °C and 5% CO2. Cells were seeded in 6-well plates at a density of 3 × 105 cells/well and cultured overnight. Next, cells were starved in serum-free medium for 5 h, and further cultured as non-treated, BMP-9-treated, or TGFβ-treated (50 ng/ml and 10 ng/ml, respectively; R&D Systems) in serum-free conditions for 24 h. All experiments were performed in triplicate wells for each condition. Subsequently, cells were washed with PBS, trypsinized, and collected as pellets, which were lysed in RLT buffer. RNA isolation was performed with the RNeasy Mini Kit (Qiagen) according to the manufacturer’s protocol. 0.5 µg total RNA was subsequently reverse-transcribed to cDNA using the iScript cDNA Synthesis Kit (Bio-Rad). 1 µl of the template was used for qPCR (Quant Studio 7 Flex Thermo Fisher Scientific). Expression levels were calculated relative to expression of the reference ribosomal gene RPL19, as calculated by the formula 100*2− ΔCt. Primer sequences (forward and reverse, respectively, Invitrogen) for the specific targets were as follows:
-
Rpl19 (GGTGACCTGGATGAGAAGGA, TTCAGCTTGTGGATGTGCTC);
-
Id1 (GAGTCTGAAGTCGGGACCAC, TTTTCCTCTTGCCTCCTGAA);
-
Id3 (ACTCAGCTTAGCCAGGTGGA, GTCAGTGGCAAAAGCTCCTC);
-
Pai1 (TGCATCGCCTGCCATT, CTTGAGATAGGACAGTGCTTTTTCC);
-
Pdgfb (CCTCGGCCTGTGACTAGAAG, CCTTGTCATGGGTGTGCTTA);
-
Clec14a (TGGCCAGGTCAGGTCTATGA, CAGGGGGCGAAGATGTGTAG).
Patient consent, RNAscope, and imaging
Tissue samples were provided by the Sweden Cancerome Analysis Network - Breast: Genomic Profiling of Breast Cancer (SCAN-B) consortium (Permit DNR 2009/658 approved by the national Ethical review board). Patients were enrolled in the clinical trial with the Identifier NCT02306096. Clinical and/or personal data connected to the tissue sample were not disclosed. Informed consent was obtained from all individual participants included in the study. This article does not contain any studies with animals performed by any of the authors.
Tumor pieces from breast cancer patients were directly obtained from surgery and were fresh-frozen in optimum cutting temperature (OCT) cryomount medium (Histolab). 5-µm-thick sections were used for RNAscope detection, following an optimized version of the RNAscope Fluorescent Multiplex Assay protocol (Advanced Cell Diagnostics, ACD). Briefly, sections were fixed in ice-cold 4% fresh paraformaldehyde, for 30 min on ice, washed with PBS, and dehydrated to 100% ethanol. Samples were pre-treated with Protease III, for 30 min at room temperature, followed by probe hybridization (Hs-ACVRL1, #55922; custom-made Hs-CLEC14A-C2, based on #510761) and four canonical steps of amplification at 40 °C, to allow for the appropriate detection of fluorescent signals. Sections were washed and mounted with ProLong Gold anti-fade mounting medium with DAPI (Thermo). The ACD 3-plex negative control probe for channels 1, 2, and 3 was used to determine the specificity and background of the signal. Images were acquired with a LSM 710 laser scanning microscope (Zeiss). At least 4 fields of 3 individual human samples were used for the quantification.
Gene set enrichment analysis, mutation profiling, and conserved gene signature
The lists of genes co-expressed with
ACVRL1 were obtained by enquiring the “cBioPortal for cancer genomics” [
14] in selected provisional studies of The Cancer Genome Atlas (TCGA) repository and the glioblastoma cohort reported in 2013 [
15]. Gene ranking was based on Pearson’s R coefficient. RNK files were generated from co-expression data from the cBioPortal and used as ranked list inputs for gene set enrichment analysis (GSEA) preranked analysis (“Hallamarks” gene matrix database, 1000 permutations). In order to obtain a signature of ACVRL1-coexpressed genes conserved across tumor types, the intersection of the different ranked gene lists was calculated with the online tool found at
http://bioinformatics.psb.ugent.be/webtools/Venn/.
TCGA data acquisition and analysis
TCGA RNA-Seq upper quantile normalized FPKM gene expression data and masked Affymetrix SNP 6.0 segmented copy number profiles were downloaded from the Genomic Data Commons (GDC) Data Portal by November 2016. Only primary tumor and normal tissue sample data were used in downstream analyses. In total, 10397 samples from 32 different TCGA projects were included, 21 of which had matching normal tissue samples. For gene expression data, log2 RNA-Seq upper quantile normalized FPKM expression values were calculated after adding an offset of 105. Matched RNA-Seq and copy number profiles were available for 492 primary prostate tumors. Copy number segments less than 10 probes were removed and neighboring segments with log2 fold differences < 0.075 were merged into continuous segments. Plots were produced in R using the “ggplot2” package.
For gene expression data of the BLCA cohort, log2 values of the normalized RNA-seq by expectation maximization (RSEM) counts were downloaded from UCSC Xena hub. The dataset (n = 407 samples) was median re-centered. Samples were classified according to Lund taxonomy classification [
16]. The mutation and copy number data for the BLCA cohort were downloaded from Broad Institute of MIT and Harvard. Gene expression-based quantification of immune and stromal cell abundance was carried out using the Microenvironment Cell Populations-counter package [
17] in R.
ChIP-Seq datasets analysis
Feature tracks from previously published ChIP-seq data [
18] were visualized with the Integrative genomics viewer (IGV).
To identify transcription factors binding to the CLEC14A DNA region, data from 7353 transcription factor (TF) ChIP-Seq experiments in 31,081 different cell and tissue types were obtained from the ChIP-Atlas database. Thresholds for TF binding were set to ± 5 kb relative to the transcription start site of CLEC14A.
Statistical analysis
All measurements are depicted as mean ± standard deviation (SD), and statistical analyses were performed using an unpaired two-tailed Student’s t test, either with R software or with GraphPad Prism 7. Statistical significance was considered using α = 0.05.
Discussion
Collectively, our study has revealed a broader regulatory network associated with ALK1 activation in cancer. The use of computational analysis enabled the generation of cancer-specific sets of genes associated with ACVRL1 expression that could be further refined to obtain a single list of common factors conserved across different tumor types. Ultimately, the validation of CLEC14A as a transcriptional target of ACVRL1 for biomarker use, and the regulation of immune response as a process correlated with ACVRL1 expression may hold utility for re-evaluating the clinical development of already existing ALK1-blocking agents.
Our cross-cancer analysis highlights a variable, but consistent, expression of
ACVRL1 in all tumor types. In particular, the reduced levels of
ACVRL1 compared to the corresponding normal tissues suggest the inability of the proliferating malignant mass to develop a vascular tree to adequately sustain the metabolic needs of the tumor cells. Our data are compatible with findings indicating a specific role for ALK1 in mediating the maturation phase of angiogenesis [
32] and are in agreement with the known aberrant nature of tumor-associated vessels. The higher expression of
ACVRL1 in KIRC and GBM tumors might reflect the architecture of the organs in which these cancers arise and develop, including a naturally strict dependency on the vasculature of these tumor types.
In light of the GSEA analysis, the association of
ACVRL1 to processes related to immune cell regulation might constitute a rationale for a combined treatment regimen based on ALK1 inhibition and immunotherapy agents. In support of this hypothesis, ALK1-co-expressed genes were highly enriched in IL2/STAT5 and IL6/JAK/STAT3 pathways; the former has been implicated in the proliferation and development of peripheral T cells and regulatory T cells [
33,
34], whereas the latter is a strong and recognized tumor immunosuppressive signaling cascade [
35,
36]. In this context, characterization of the immune infiltration will be of paramount importance to determine whether specific tumor types might benefit from combined therapy. Recent work proposes a triggering of the intra-tumoral immune response following vessel normalization induced by anti-VEGF therapy [
37,
38]. Intriguingly, an analogous vascular phenotype was reported in different studies that characterized the
in vivo activity of ALK1-Fc [
8,
9,
39,
40].
As already mentioned, the clinical benefit of targeted therapy has been limited by tumor evolution and adaptation to anti-cancer agents. The analysis of the mutational landscape of
ACVRL1 indicated that this locus is affected by somatic alterations with a relatively low frequency. In line with these observations, amplification of
ACVRL1 did not confer tumors the biological advantage typical of a putative oncogenic driver. Interestingly, some of the loss of function mutations observed in tumors are reported to affect the receptor kinase domain (e.g., the missense R411Q [
41], and the truncating W406*, S462*, and E470* mutations [
42‐
44]) and have already been described in human hereditary telangiectasia (HHT)2, an autosomal dominant genetic vascular disorder caused by mutations in
ACVRL1. Again, these events were randomly distributed in the different datasets and did not confer any overt advantage to the tumors.
Our effort to identify a set of common
ACVRL1-related genes whose expression is preserved across different cancer types confirms the primary role of ALK1 as a mediator of endothelial cell fate. Among the 8 conserved genes across different tumor types,
CLEC14A was the one with the highest mean co-expression coefficient. CLEC14A was initially described as a fundamental component of the cell-to-cell adhesion machinery [
29] and just a year later, a function as a tumor endothelial-specific marker was proposed [
25]. The exact role of CLEC14A in angiogenesis is still debated, as two independent studies (based on rather different investigational endpoints) reported opposite effects when knocking out
Clec14a in a mouse model of lung carcinogenesis [
30,
45]. Nonetheless, based on much more similar experimental setups, the reduced sprouting of VEGF-stimulated HUVEC
in vitro, as well as the reduced tumor volume and associated vascular density
in vivo reported in
Clec14a knock-out mice [
45], phenocopies the effects of ALK1 inhibition observed in different studies [
8,
10,
11]. Similarly, analysis of
Clec14a expression in a transgenic mouse model of pancreatic neuroendocrine tumorigenesis demonstrated an increased expression only in full-blown tumors with a more mature vessel phenotype, but not in islets that have undergone an angiogenic switch to fuel their proliferation [
46], supporting the reports of ALK1 expression in the resolution phase of angiogenesis [
7].
As ascertained by ChIP-seq data of human endothelial cells, stimulation with the high-affinity ligand BMP9 produced a strong binding peak of SMAD1 in the promoter region of CLEC14A. To confirm this type of regulation, dual RNAscope-ISH on human breast cancer samples unveiled that expression of ACVRL1 is required for the concurrent detection of CLEC14A in the same cell. In conclusion, we propose that CLEC14A is under the transcriptional control of ACVRL1. The lack of reliable reagents for the detection of ALK1, paired with the paucity of predictive biomarkers for ALK1 blockade, has hampered the translation of ALK1 inhibitors to clinical care. This highlights the need for activity-based assessment of the ALK1 pathway as a predictive biomarker for patient selection. In this context, our 8-gene profile, as well as CLEC14A, might represent a starting point for the development of a companion tool for precision targeting of ALK1-driven tumor angiogenesis.
Furthermore, our results suggest other modalities of
CLEC14A regulation exerted by
ACVRL1, bringing together some of the aspects we have already discussed, e.g., the relationship between endothelial ALK1 expression and the modulation of the properties and the composition of the tumor microenvironment. The unbiased assessment of transcriptional regulators bound to the promoter of
CLEC14A revealed a significant occupancy of the coactivator EP300, an acetyltransferase that orchestrates transcription via chromatin remodeling [
47]. Although EP300 is a common cofactor with very broad functions in cell growth and division, the presence of this enzyme is relevant given its ability to cooperate with another enriched factor bound to the promoter region of
CLEC14A, namely RELA/p65 (encoding for the p65 subunit of NF-κB). Indeed, RELA and EP300 can promote the activation of E-selectin and vascular cell adhesion molecule (VCAM)-1, fundamental mediators of leukocyte adhesion to endothelial cells [
48], allowing the extravasation and tissue infiltration steps of the inflammation cascade to further coordinate the immune response. In line with these observations, our results show that genes associated to
ACVRL1 expression are significantly enriched in “TNF-α via NF-κB signaling.” Lastly, the bromodomain containing protein (BRD)4 was the most significantly enriched element bound to
CLEC14A. Of note, BRD4 and RELA/p65 jointly drive the inflammatory transcriptional response [
49], whereas more recently a study focused on the direct interaction between these two proteins following TNF-α stimulation of endothelial cells [
50].
In conclusion, our results shed light on previously unknown functional associations elicited by the downstream effectors of ALK1 in endothelial cells. The future validation of our 8-gene signature and CLEC14A as biomarkers to follow the activation status of ALK1, paired with the potential combination of ALK1 inhibitors with immunomodulatory compounds, may motivate reconsideration of the halted clinical development of already existing ALK1-blocking agents.