Elevated expression of the adhesion GPCR ADGRL4/ELTD1 promotes endothelial sprouting angiogenesis without activating canonical GPCR signalling

ADGRL4/ELTD1 is an orphan adhesion GPCR (aGPCR) expressed in endothelial cells that regulates tumour angiogenesis. The majority of aGPCRs are orphan receptors. The Stachel Hypothesis proposes a mechanism for aGPCR activation, in which aGPCRs contain a tethered agonist (termed Stachel) C-terminal to the GPCR-proteolytic site (GPS) cleavage point which, when exposed, initiates canonical GPCR signalling. This has been shown in a growing number of aGPCRs. We tested this hypothesis on ADGRL4/ELTD1 by designing full length (FL) and C-terminal fragment (CTF) ADGRL4/ELTD1 constructs, and a range of potential Stachel peptides. Constructs were transfected into HEK293T cells and HTRF FRET, luciferase-reporter and Alphascreen GPCR signalling assays were performed. A stable ADGRL4/ELTD1 overexpressing HUVEC line was additionally generated and angiogenesis assays, signalling assays and transcriptional profiling were performed. ADGRL4/ELTD1 has the lowest GC content in the aGPCR family and codon optimisation significantly increased its expression. FL and CTF ADGRL4/ELTD1 constructs, as well as Stachel peptides, did not activate canonical GPCR signalling. Furthermore, stable overexpression of ADGRL4/ELTD1 in HUVECs induced sprouting angiogenesis, lowered in vitro anastomoses, and decreased proliferation, without activating canonical GPCR signalling or MAPK/ERK, PI3K/AKT, JNK, JAK/HIF-1α, beta catenin or STAT3 pathways. Overexpression upregulated ANTXR1, SLC39A6, HBB, CHRNA, ELMOD1, JAG1 and downregulated DLL4, KIT, CCL15, CYP26B1. ADGRL4/ELTD1 specifically regulates the endothelial tip-cell phenotype through yet undefined signalling pathways.


Results
Codon optimisation increased ADGRL4/ELTD1 expression. After experiencing difficulties overexpressing the ADGRL4/ELTD1 protein (Fig. 1A) we observed that ADGRL4/ELTD1's nucleotide sequence has a GC content of 34% (codon adaptation index 0.18), the lowest within the adhesion and secretin GPCR families ( Figure S1A). The average GC content for coding sequences in Homo sapiens is 52% 21 . Wild type (WT) full length (FL) ADGRL4/ELTD1 was codon optimised (co) retaining its amino acid coding sequence and increasing its GC content to 58% (codon adaption index 0.76), a level higher than the majority of compared receptors ( Figure S1A) with the resulting co FL ADGRL4/ELTD1 producing a much higher level of protein expression compared to the original sequence when transfected into HEK293T cells. (Fig. 1A). Codon optimised versions of ADGRL4/ELTD1 were then used for all subsequent experiments.

ADGRL4/ELTD1 construct design and expression. Two recombinant forms of co ADGRL4/ELTD1
were created to test the tethered agonist hypothesis 14 : (1) a FL form-showing more or less basal activity; and (2) a C-terminal fragment (CTF) mutant, where everything N-terminal to the GPS consensus cleavage site (amino acids HL//T 22,23 ) was removed-which was shown to induce constitutive activity in aGPCRs (Fig. 1B). To minimise the risk of tags interfering with signalling, two series of constructs were designed with different tagging strategies: one contained both N-terminal HA and C-terminal FLAG tags, as used in the original studies 14,24,25 , whilst a second contained only a C-terminal GFP tag (Fig. 1B). Wild type ADGRL4/ELTD1's signal peptide was included in all recombinant versions created (both FL and CTF forms).
Each was transiently transfected into HEK293T cells and assessed for expression of the appropriate protein tag at 48 h using FACS. The N-terminal HA tagged FL and CTF constructs were expressed on the cell surface with the FL construct having a higher surface expression than the CTF construct (55% vs. 30% positivity) (Figure S1B). FACS results for the whole cell C-terminally GFP tagged FL and CTF constructs showed equal overall fluorescence for both constructs ( Figure S1B) that was greater than the WT equivalent ( Figure S1C). Similar results for cell surface expression were obtained using an 'in-house' anti-ADGRL4/ELTD1 monoclonal antibody raised against the NTF 4 ( Figure S1C).
Confocal microscopy also showed surface and cytoplasmic GFP expression for both constructs ( Figure S1D). Surface expression of ADGRL4/ELTD1's N-terminal HA tagged constructs was comparable to that of HA tagged FL and CTF ADGRG6/GPR126 14 as well as HA tagged P2Y12 (a Rhodopsin-like GPCR with very high surface expression) ( Figure S1E).
Neither full length nor CTF ADGRL4/ELTD1 or ADGRL4/ELTD1's Stachel peptides activate canonical GPCR signalling. GPCR signalling assays were performed using three experimental approaches (see methods), assessing second messenger products or transcription factors downstream of specific G-protein pathway coupling. As all of these assays are accumulation assays, they were performed at 48 h following transfection to allow sufficient time for high levels of ADGRL4/ELTD1 surface expression.
HTRF FRET assays in transiently transfected HEK293T cells showed no Gαq or Gαs signalling activity from either FL or CTF ADGRL4/ELTD1 compared to an empty vector control (Fig. 1C), despite adequate cell surface expression and positive controls. Luciferase reporter assays (downstream of a wider range of G proteins) in similarly transfected HEK29T cells showed no sign of canonical GPCR signalling activity, despite proper positive controls (Fig. 1D), as did the Alphascreen system ( Figure S1F). In summary neither FL nor CTF ADGRL4/ ELTD1 constructs (in differently tagged configurations) elicited any canonical GPCR signalling, despite adequate and appropriate localisation of expression, when compared to controls under conditions where signalling was observed for other aGPCRs.
We also investigated whether a peptide analogous to ADGRL4/ELTD1's putative tethered agonist region could elicit canonical GPCR signalling. 21 peptides of varying length corresponding to the amino acid sequence beginning at ADGRL4/ELTD1's consensus GPS cleavage site and extending C-terminally to the predicted start of the first 7TM loop ( Fig. 2A) Fig. 2A) as most published Stachel peptides had been of a length within this range. The 11 selected ADGRL4/ELTD1 Stachel peptides were then assayed for activation of co FL ADGRL4/ELTD1 transiently transfected in HEK293T cells using luciferase assays corresponding to all canonical GPCR signalling pathways. With peptide insolubility being of concern we wanted to include some additional peptide controls. Thus, phylogeny comparisons were conducted on the predicted GPS sequences of all human aGPCR family members to find the relative with the highest GPS similarity to ADGRL4/ELTD1's GPS (ADGRA1/GPR123 does not contain a GPS domain and was not included) as it has been shown that aGPCRs that cannot be activated through their own tethered agonist can be induced through closely related Stachel peptides 26 . ADGRD1/GPR133 had the most similar GPS (Fig. 2B) and ADGRG5/ GPR114's peptide (pGPR114) (Wilde et al., 2016) was chosen as an additional control peptide. Neither of these peptides showed Gαq, Gαs, Gαi, Gα12/13, or Gβγ pathway signalling activity with ADGRL4/ELTD1 (Fig. 2C). Results from HTFR FRET assays performed in similarly transfected HEK293T cells as well as on non-transfected HUVECS (which constitutively express WT FL ADGRL4/ELTD1) using all 21 Stachel peptides showed no Gαs or Gαi signalling ( Figure S2A-B). In summary, ADGRL4/ELTD1 did not appear to activate any form of canonical GPCR signalling (Fig. 2D). www.nature.com/scientificreports/ ADGRL4/ELTD1 overexpression induces sprouting angiogenesis and regulates the endothelial tip-cell phenotype. Although ADGRL4/ELTD1 knockdown has previously been shown to inhibit sprouting angiogenesis in HUVECs 4 , the effect of overexpression remains unknown. This is biologically relevant to disease pathology as ADGRL4/ELTD1 is upregulated in endothelial cells found in the tumour microenvironment 4 . Functionality of the overexpressed co ADGRL4/ELTD1 protein was investigated in different batches of stably transduced low passage primary endothelial cells (Fig. 3A,B, Figure S3A,B) which constitutively express WT ADGRL4/ELTD1 at a low level. Wild type ADGLR4/ELTD1 mRNA expression (distinguishable from co ADGLR4/ELTD1 at the transcript level) was similar for both control and co ADGRL4/ELTD1 transduced samples, showing that ADGRL4/ELTD1 protein overexpression did not exert negative feedback onto wild type ADGLR4/ELTD1 expression ( Figure S3C). ADGRL4/ELTD1 overexpression in HUVECs induced sprouting angiogenesis (increased total spheroid surface area [24 h p = 0.002; 48 h p = 0.0008], increased the number of sprouts [24 h p = 0.002; 48 h p = 0.03] (Fig. 3C) and increased endothelial capillary tube formation (p = 0.03) (Fig. 3D). The fibrin gel bead assay showed that ADGRL4/ELTD1 overexpression inhibited vessel branching reducing vascular anastomoses/branches between adjacent vessels at the 3-week conclusion of the assay (p < 0.00001) (Fig. 3E). In both groups, lumen formation (C) Gαq, Gαs, Gαi, Gα12/13, Gβγ Stachel peptide luciferase assays in HEK293T cells transfected with co FL ADGRL4/ELTD1. Positive controls used were the CHRM3/M3 GPCR receptor exposed to 10 µM carbachol for Gαq, Gαs, Gαi, Gα12/13, Gβγ (p < 0.0001) and GPR133 exposed to its Stachel peptide (p133) for Gαs (p < 0.0001). (D) Graphical summary of canonical GPCR signalling pathways investigated in this study. www.nature.com/scientificreports/ appeared to be similar (Fig. 3E). HUVEC proliferation was decreased with 40% fewer cells at 24 h (p = 0.01) and 28% fewer cells at 48 h (p = 0.01) (Fig. 3F). There was no effect on endothelial cell migration (start of assay p NS; end of assay at 12 h p NS) ( Figure S3D), or adhesion to various angiogenesis-associated extracellular matrix components (collagen I, collagen IV, laminin, vitronectin, fibronectin p NS for all) ( Figure S3E) and matrigel (p NS) ( Figure S3F). Gαq, Gαs, Gαi HTRF FRET assays were performed in both ADGRL4/ELTD1 overexpression and control HUVECs and showed no significant induction of any Gαq, Gαs, or Gαi signalling (p NS for all) (Fig. 4A). Western blotting revealed no activation of the MAPK/ERK, PI3K/AKT JNK, JAK/HIF-1α, beta catenin, NOTCH and STAT3 pathways in ADGRL4/ELTD1 overexpressing HUVECs (Fig. 4B).

Discussion
We investigated ADGRL4/ELTD1 receptor signalling and found that ADGRL4/ELTD1 had the lowest GC content in the aGPCR family, did not show canonical GPCR signalling, and induced an endothelial tip-cell phenotype when overexpressed (as occurs in tumour-associated endothelium 28 ). The finding that wildtype ADGRL4/ ELTD1's GC content was the lowest in the aGPCR family is interesting from a functional and evolutionary perspective. Codon usage regulates mRNA stability and translational efficiency with optimal codons associated with abundant tRNAs for the cognate codon, and higher levels of protein expression, whilst those with low GC content and non-optimal codons are associated with low level expression [29][30][31] . ADGLR4/ELTD1's non-optimal codon usage suggests that it has been evolutionary beneficial for endothelial and smooth muscle cells to tightly www.nature.com/scientificreports/ control ADGRL4/ELTD1 expression, despite natural selection generally opting for more optimal codon usage across the genome 32 . An example of this are the genes of the cell cycle, which have a low GC and non-optimal codon usage as a mechanism to tightly control expression 33 .
A recent publication suggested that ADGRL4/ELTD1 signals via the JAK/STAT3/HIF-1α pathway in glioblastoma to regulate proliferation and migration 34 although this is not a known direct cascade for GPCRs. We did not observe any downstream signalling events in HEK293T, COS-7 or HUVEC cells, indicating that the interaction between ADGRL4/ELTD1 and the JAK/STAT3/HIF-1α pathway is specific for glioblastoma or other cancer cell types and does not apply within endothelial cells. Currently, the direct intracellular consequences of ADGRL4/ELTD1 activation remain unknown. This lack of signalling could potentially be due to: (1) lack of the endogenous Stachel sequence or insufficient stimulus for its exposure, (2) G protein-independent signalling (3) function via cell surface interactions.
ADGRB1/BAI1 and ADGRG1/GPR56 are two aGPCR S which induce G protein-coupled signalling through both the tethered agonist as well as through a non-tethered agonist mechanism 35 . For both receptors, both their CTF forms as well as a truncated form of the 7TM (without a tethered agonist sequence) signal constitutively but also differentially. For ADGRG1/GPR56, its CTF induced greater Gα 12/13 and Gα q signalling than its 7TM form, whilst for ADGRB1/BAI1, its CTF induced greater Gα 12/13 signalling whilst its 7TM form induced greater Gα q signalling 35 . However, even when Stachel-independent signalling was detected it did not reveal a signal that was absent in full length or Stachel-containing constructs. Thus, we believe it rather unlikely that 7TM constructs would reveal any signal in the investigated assays.
Although ADGRL4/ELTD1 lacks a PDZ motif, interestingly its overexpression induced the expression of another ELMO protein family member, ELMOD1. ELMOD1 is a GTPase-activating protein, activates the GTPbinding protein ARF6 to regulate actin assembly and membrane trafficking 40 . In endothelial cells, ARF6 is known to regulate angiogenesis, and does so by modulating the trafficking and recycling of VEGFR2 from the cell surface, ensuring adequate VEGFR2 expression 41,42 . Other known G protein-independent aGPCR signalling components include GABPγ associating with ADGRB2/BAI2 and leading to VEGF induction 43 , ADGRC1/ Celsr1 forming a complex with Dishevelled, DAAM1 and PDZ-RhoGEF and ADGRA3/GPR125 influencing dishevelled and Wnt signalling 44 .
Another possible explanation for the lack of detectable signalling pathways is that other means of activation need to be employed to activate the receptor. Among them mechanical forces or incubation of the full length receptor with a ligand that has not yet been identified. Further, small molecule compounds have been shown to activate aGPCRs. Examples include beclomethasone dipropionate on GPR97 45 or Gedunin-and Khivorinderivatives on GPR56 and GPR114 46 . It is possible that endogenous small molecules could activate ADGRL4/ ELTD1 irrespective of the existence of a functional tethered agonist. Also, ADGRL4/ELTD1's function may act in trans as a ligand for other molecules as has been suggested for ADGRL1/LPHN1 47 or ADGRB1/BAI1 48 . Future experiments are planned to determine whether ELTD1/ADGRL4's NTF or 7TM regions are responsible for its angiogenesis phenotype seen when FL ELTD1/ADGRL4 is overexpressed in endothelial cells. This will be performed using two approaches: CTF form overexpression and 7TM-mutant form overexpression. Should CTF overexpression not induce an angiogenesis phenotype, this suggests that ADGRL4/ELTD1's NTF is central to promoting its angiogenic effect with activity possibly influenced by NTF adhesion. Conversely, if 7TM-mutant overexpression (FL ELTD1 truncated prematurely after the first or second 7TM loop) does not induce an angiogenesis phenotype, this would suggest that ADGRL4/ELTD1's 7TM is central to promoting its angiogenic phenotype.
Limitations of our signalling analysis include the use of transient transfection and HEK293T cells which may not contain the cellular machinery necessary for ADGRL4/ELTD1 signalling. Despite this, the majority of published aGPCR Stachel papers have used these cells for GPCR signalling assays although they do not constitutively express the receptor. ADGRL4/ELTD1's Stachel peptides were all very water insoluble and difficult to dissolve. Various modifications were added to the N-and C-peptide termini to try and improve solubility with no effect. In contrast, some aGPCRs have more water-soluble Stachel peptides, for example GPR116's p116 19,26 . However, Stachel peptides of GPR114 or GPR133 are similarly insoluble, yet they yield significant activation on full length and mutant aGPCRs 14,16 .
Our finding that ADGRL4/ELTD1 induced a tip phenotype in endothelial cells is opposite and complementary to the effect seen on ADGRL4/ELTD1 silencing 4 , as is the effect on the core angiogenic Notch genes DLL4 and JAG1, and KIT. The finding that ADGRL4/ELTD1 overexpression decreases anastomoses suggests that there may be a failure to switch off the tip cell phenotype as vascular anastomoses occur when two sprouts led by endothelial tip cells meet and merge, with the resultant loss of the tip cell phenotype and induction of the stalk phenotype.
ADGRL4/ELTD1 overexpression also altered patterns of gene expression in HUVECs by inducing ANTXR1, SLC39A6, HBB and CHRNA1 and repressing CCL15. It is possible that some of the products of these genes may contribute to ADGRL4/ELTD1's ability to promote angiogenesis. ANTXR1 is a transmembrane receptor that similarly to ADGRL4/ELTD1 is upregulated in tumour endothelial cells 49 and which promotes tumour angiogenesis and endothelial proliferation 50 . Similarly, to ADGRL4/ELTD1, ANTXR1 silencing in xenografts of diverse tumour origin reduced tumour size 51 . SLC39A6 is a zinc transporter associated with the metastasis of pancreatic, oesophageal, prostate and lung cancers 52-54 however its link to endothelial biology remains unexplored. Interestingly, ADGRL4/ELTD1 silencing was shown to inhibit the metastasis of ovarian cancer xenografts 4 .
HBB is the β-subunit of haemoglobin expressed by red blood cells. This is an interesting association considering the link between progenitor endothelial cells which develop into haematopoietic stem cells (HSCs) which www.nature.com/scientificreports/ produce erythrocytes 55 . The association with the haemogenic endothelium is further strengthened through ADGRL4/ELTD1's antagonistic relationship with KIT 27 , a positive regulator of haematopoietic stem cells 56 and the endothelial niche 57 . Furthermore, ADGRL4/ELTD1 (along with ADGRG1/GPR56, another aGPCR) forms part of the transcriptional programme behind the transdifferentiation of haemogenic endothelial cells (HECs) to HSCs, which then produce the cells of the haematopoietic system 58 . This presents an exciting new direction of ADGRL4/ELTD1 research. ADGRL4/ELTD1 upregulation in the tumour vasculature has been identified as a common finding in several different cancer types, where it has been hypothesised to promote tumour angiogenesis 4 . Here we show that ADGRL4/ELTD1 overexpression in normal primary endothelial cells does indeed have a pro-angiogenic effect. While we could not identify the signalling cascades utilised by ADGRL4/ELTD1, we identified several genes whose expression is altered on ADGRL4/ELTD1 overexpression providing possible mechanistic insights into how it may drive tumour angiogenesis. The further study of such markers related to ADGRL4/ELTD1 function will also provide valuable information for monitoring the efficacy of future therapies aiming to inhibit the expression and/or function of ADGRL4/ELTD1 in the tumour microenvironment.

Methods
Codon optimisation. GC content, codon usage analysis and optimisation was performed using JCat 59 and Biomart 60 . We transiently transfected both codon optimised (co) and wild type (WT) forms of full length (FL) ADGRL4/ELTD1 into HEK293T cells (which do not constitutively express ADGRL4/ELTD1) and analysed total cell ADGRL4/ELTD1 protein expression after 48 h. HEK293T cells were selected as the majority of aGPCR signalling experiments to date have been performed in these cells. RNA analysis and qPCR. RNA was extracted using RNeasy Mini Kit (Qiagen) and reverse transcription was performed using the High Capacity cDNA Reverse Transcription kit (Applied Biosystems) according to the manufacturer's guidelines. qPCR was performed using the SensiFAST SYBR No-ROX kit (Bioline). qPCR primers were designed using Primer-BLAST 64 and were synthesised by Invitrogen. Samples were run using the 7900HT Fast Real-Time PCR System (ThermoFisher) in a 384 well format. Primers are listed in Supplementary Methods Table 2.

ADGRL4/ELTD1 construct design.
Flow cytometry. Cells were prepared for FACS using standardised FACS methods and measured using a BD FACSCalibur flow cytometer (BD Biosciences). Primary antibodies are listed in Supplementary Methods Table 3. Data was analysed using FlowJo 10.0.7 (FlowJo LLC). Signalling assays. In brief, constructs were transiently transfected into HEK293T cells (co-transfection was performed when using the luciferase reported assays) and assayed at 48 h followed by FACS to determine ADGRL4/ELTD1 surface expression. For the Stachel peptide experiments in HEK293T cells, cells were transiently transfected with co FL ADGRL4/ELTD1 for 48 h before being exposed to the ADGRL4/ELTD1 Stachel peptides for 30 min and commencing the signalling readout assays. AlphaScreen assay. PerkinElmer's AlphaScreen cAMP assay (PerkinElmer) was performed as per the manufacturer's protocols. In brief, cells were transfected as above. At 48 h, cells were incubated in medium containing IBMX and then lysed in LI buffer (PerkinElmer). Plates were then frozen for 24 h at − 20 °C. The Alpha Screen cAMP assay kit was used to measure cAMP concentration as per the manufacturer's guidelines. Accumulation was measured in opaque-walled 96 well microplates (PerkinElmer) and read using the PerkinElmer EnVision plate reader (PerkinElmer).

HTRF FRET assays. Homogeneous
Stachel peptide dilution and application. ADGLR4/ELTD1's Stachel peptides were first diluted with DMSO and then with 50 mM tris(hydroxymethyl)aminomethane (TRIS) (pH8) to yield a final peptide concentration of 1 mM peptide in 1% DMSO (DMSO alone used as control). Peptides were applied to HUVECS and transiently transfected HEK293T cells for 30 min before assessing canonical GPCR signalling activity as described above.
Cloning and lentiviral transduction. Cloning  Fibrin gel bead assay. This assay was performed as originally described 66 using ADGRL4/ELTD1 overexpressing HUVECS coated onto cytodex beads which were then embedded in a fibrin gel overlaid with fibroblasts. Medium was changed every 2 days with EGM2 (Lonza) and images were taken every second day. See supplementary methods for further details.
Tube formation assay on matrigel. Control and ELTD1 overexpressing HUVECS were plated at 5 × 10 4 cells/ well on top of 200 µl of matrigel in a 24-well plate (BD Biosciences) The plate was then placed within an Incucyte Imaging Incubator (5% carbon dioxide at 37 °C) (Essen BioScience) and 10 photos were taken per well every hour for 24 h. Tube formation was analysed by counting the number of complete polygons (comprising endothelial tubes) in each image field. Results were analysed using FIJI.
Endothelial proliferation assay. In this assay, control and ELTD1 overexpressing HUVECS were plated at 1 × 10 5 cells per well (low confluence) into multiple 12-well plates. Every day, one plate was selected for photographing, harvesting and cell counting. This was repeated until both control and ELTD1 groups achieved confluence. Photographs were performed using an Incucyte Imaging Incubator (5% CO 2 at 37 °C). Sixteen photographs were taken of each well, then cells were harvested and counted as described in the cell lines and maintenance methods section. Results were analysed using FIJI.
Endothelial migration assay. Control and ELTD1 overexpressing HUVECS were seeded at 5 × 10 5 cells per well into a 24-well ImageLock plate (Essen BioScience) and grown to confluence. Once confluent, a scratch defect was introduced into each well using a 4 pin Essen Wound Maker (Essen BioScience) and scratches were imaged every 30 min, using an Incucyte Imaging Incubator (5% CO 2 at 37 °C), until the defect had been repaired. The cell-free scratched images of each group at individual time points were then manually traced using the self-made stylus described above with the area quantified using FIJI.
MilliCoat extracellular matrix (ECM) adhesion assay. This assay was performed using the MilliCoat ECM Screening Kit (Merck: ECM105) according to the manufacturer's instructions, seeding 1 × 10 4 HUVECS into wells coated individually with the various ECM adhesion components (96-well format). Absorbance was read at 560 nm using a FLUOstar optima microplate reader (BMG Labtech). microplate reader. Results were analysed using FIJI.
Matrigel adhesion assay. Following matrigel solidification, control and ELTD1 overexpressing HUVECS were seeded at 1 × 10 4 cells into each well of a 96-well plate. After 1 h cells were washed 3 × with warmed PBS and the centre of each well was photographed using an AMG Evos XL Core digital microscope (Fisher Scientific). Images were analysed using FIJI.

Microarray analysis
Gene expression profiling analysis of ADGRL4/ELTD1 overexpressing HUVECs was performed by the Wellcome Trust Centre for Human Genetics (Oxford). Microarrays were performed by reverse transcribing extracted mRNA into cDNA, labelling the cDNA, and then hybridising the labelled target to the microarray chip. Raw data outputted from the Illumina microarray scanner (Illumina iScan) using a HumanHT-12 v4 Expression BeadChip Kit (Illumina) was analysed using using R (version 3.3.2) and Limma 67 . Differentially expressed genes were deemed statistically significant if they fulfilled standard microarray criteria: (1) an adjusted p value < 0.05; and (2) a fold change of > 2 < − 2. Visualisation was performed using "GGplot2" 68 . All microarray experiments were MIAME compliant 69 .