Age‐related macular degeneration (AMD) is a leading cause of vision loss among the elderly. AMD pathogenesis involves chronic activation of the innate immune system including complement factors and microglia/macrophage reactivity in the retina. Here, we show that lack of interferon‐β signaling in the retina accelerates mononuclear phagocyte reactivity and promotes choroidal neovascularization (CNV) in the laser model of neovascular AMD. Complete deletion of interferon‐α/β receptor (Ifnar) using Ifnar1−/− mice significantly enhanced early microglia and macrophage activation in lesion areas. This triggered subsequent vascular leakage and CNV at later stages. Similar findings were obtained in laser‐treated Cx3cr1CreER:Ifnar1fl/fl animals that allowed the tamoxifen‐induced conditional depletion of Ifnar in resident mononuclear phagocytes only. Conversely, systemic IFN‐β therapy of laser‐treated wild‐type animals effectively attenuated microgliosis and macrophage responses in the early stage of disease and significantly reduced CNV size in the late phase. Our results reveal a protective role of Ifnar signaling in retinal immune homeostasis and highlight a potential use for IFN‐β therapy in the eye to limit chronic inflammation and pathological angiogenesis in AMD.
Innate immune activation is a hallmark of age‐related macular degeneration (AMD). Here, interferon‐beta signaling was identified as an intrinsic regulator of retinal mononuclear phagocytes and therefore as potential therapeutic target for AMD.
Loss of Ifnar1 signaling triggers retinal microglia and macrophage reactivity and promotes angiogenesis in a laser‐induced model for AMD.
Specific knockdown of Ifnar1 in mononuclear phagocytes affects their immunomodulatory potential and exacerbates choroidal neovascularization.
IFN‐β therapy attenuates microgliosis and macrophage reactivity and thereby reduces choroidal neovascularization.
Age‐related macular degeneration (AMD) is the leading cause of visual impairment among the elderly, often resulting in blindness (Augood et al, 2006). It occurs in two main clinical forms, the dry form with geographic atrophy and the wet or exudative form. The dry form is characterized by accumulation of cellular debris in the subretinal space, while the wet form typically presents with neovascular processes (Noel et al, 2007). Anti‐VEGF medication with monoclonal antibodies or aptamers is presently the gold standard in the treatment of choroidal neovascularization (CNV) (Rofagha et al, 2013). However, inhibition of angiogenesis alone does not affect the cellular immunological events underlying CNV and also fails to be effective in the dry form of AMD (Jager et al, 2008). Furthermore, VEGF inhibition may cause a number of severe retinal and systemic adverse events (Sene et al, 2015). Therefore, there is an urgent need for identifying alternative approaches to treat AMD that is based on the underlying immunological pathogenesis.
Microglial cells, the resident immune cells of the CNS, play a key role in the initiation and perpetuation of chronic inflammatory events in the aging retina (Sierra et al, 2007; Damani et al, 2011). Furthermore, mononuclear phagocyte reactivity associated with subretinal migration is a common hallmark in human AMD and related mouse models (Gupta et al, 2003; Combadière et al, 2007; Levy et al, 2015). Consequently, signaling pathways and molecular targets that modulate microglia and macrophage activity represent attractive therapeutic tools for the treatment of degenerative and inflammatory diseases of the retina, including AMD (Langmann, 2007).
Type 1 interferon signaling through the interferon‐alpha/beta receptor (Ifnar) is critically important for innate immune defense (Sadler & Williams, 2008). There is also strong evidence that interferon‐β (IFN‐β) has potent immunomodulatory functions on microglia and thereby limits autoimmunity in the CNS (Axtell & Steinman, 2008; Prinz et al, 2008). Likewise, mice lacking Ifnar or the IFN‐β gene display extensive microglia activation and develop a more severe and chronic course of experimental autoimmune encephalomyelitis (EAE) (Teige et al, 2003; Prinz et al, 2008). Furthermore, IFN‐β can block the production of neurotoxic superoxide radicals in microglia in vitro (Jin et al, 2007), and induction of endogenous IFN‐β by TLR3 or MDA‐5 and RIG‐I (Dann et al, 2011) ligands protects from EAE via an immunoregulatory pathway (Touil et al, 2006).
Here, we have studied the role of IFN‐β and its receptor Ifnar in the laser‐induced choroidal neovascularization model mimicking age‐related macular degeneration in mice (Lambert et al, 2013). Previous studies have shown beneficial effects of IFN‐β treatment on laser‐induced CNV in rabbits (Kimoto et al, 2002) and monkeys (Tobe et al, 1995) by influencing the function of RPE and endothelial cells. Here, we demonstrate a pivotal effect of Ifnar/IFN‐β signaling in retinal microglia and macrophages cells that reduce the inflammatory and angiogenic events and thereby limit the development of CNV lesions.
Loss of Ifnar1 signaling promotes retinal microglia/macrophage reactivity and angiogenesis
We first determined whether Ifnar1 signaling affects the three key events in laser‐induced CNV, (i) microglia/macrophage activation, (ii) vascular leakage, and (iii) neovessel formation. These three processes therefore were characterized in a kinetics analysis 3, 7, and 14 days after laser treatment of wild‐type animals and Ifnar1−/− mice, respectively (Fig 1A). The confocal analysis of retinal flat mounts revealed that Iba1+ cells with mixed ramified and amoeboid phenotypes were present at the lesion site 3 days post‐laser damage (Fig 1B). In contrast, retinas from Ifnar1−/− mice predominantly revealed amoeboid‐shaped cells in the laser lesion, indicating a more reactive cell population (Fig 1C). The quantification of amoeboid Iba1+ cells from z‐stack images then revealed significantly higher cell numbers in the laser spots of Ifnar1−/− mice (14 ± 1.9 cells/spot) compared to control animals (4.5 ± 1.2 cells/spot) (Fig 1D). The total number of Iba1+ cells within the retina was not statistically different in both groups of animals (Appendix Fig S1A). To further define the cellular phenotype at the lesion sites, changes in their ramification and the length of processes were determined by counting the number of crossing points per individual cell using a grid image system (Chen et al, 2012). These observations also revealed a significant reduction in grid cross points in Ifnar1−/− retinas compared to controls, which clearly indicates a general shift in the morphology of microglia and macrophages (Fig 1E).
Given the strong influence of Ifnar1 deficiency on early mononuclear phagocyte activation, we next analyzed vascular leakage and CNV formation using fundus fluorescein angiography and lectin staining of RPE/choroidal flat mounts at different time points after laser lesion. We found that Ifnar1 deficiency had no major impact on CNV at day 3 and day 7 (Fig 1F, G, I and J), but resulted in a significantly higher vascular leakage at day 14 compared to controls (Fig 1H and K). Furthermore, lectin staining revealed a significant increase in the total CNV area in Ifnar1−/− mice compared to wild‐type animals at 14 days after laser damage (Fig 1M–Q). We also analyzed retinal sections at day 3 and noticed prominent co‐labeling of Iba1 and lectin in the subretinal area of Ifnar1−/− mice compared to wild‐type controls (Fig EV1A and B). Moreover, in Ifnar1−/− mice reactive Iba1+ cells persisted longer in this area as choroidal flat mounts revealed considerable Iba1‐lectin co‐staining at day 7 post‐laser treatment (Fig EV1C and D). These findings indicate that loss of Ifnar1 signaling in the retina triggers early microglia/macrophage reactivity that persists for several days and thereby negatively influences late outcome of CNV.
IFN‐β therapy attenuates mononuclear phagocyte reactivity and limits choroidal neovascularization
Having demonstrated that laser‐induced CNV was enhanced in animals lacking Ifnar1, we next investigated the potential therapeutic effect of IFN‐β. After laser coagulation, wild‐type mice were systemically treated with 10,000 units recombinant human IFN‐β every second day and CNV was analyzed after 3, 7, and 14 days, respectively (Fig 2A). IFN‐β‐treated animals displayed a significantly lower number of amoeboid Iba1+ cells in the laser spots at day 3 when compared to untreated controls (Fig 2B–D). In addition, IFN‐β therapy strongly influenced the phenotype of microglia/macrophages toward a highly ramified state characterized by more cellular processes (Fig 2E) without influencing the total number of Iba1+ cells within the retina (Appendix Fig S1B). Furthermore, IFN‐β treatment considerably improved disease outcome by reducing vascular leakage (Fig 2F–L) and CNV lesion size (Fig 2M–Q) 14 days post‐laser damage. The analysis of retinal sections at day 3 (Fig EV2A and B) and of RPE/choroidal flat mounts at day 7 (Fig EV2C and D) showed much weaker Iba1‐lectin co‐staining in IFN‐β‐treated animals, indicating less pronounced immune reactivity associated with smaller CNV.
Loss of Ifnar1 on resident microglia/macrophages affects their immunomodulatory potential and supports choroidal neovascularization
To establish whether Ifnar1 signaling in retinal microglia and potentially also long‐lived choroidal macrophages directly modulates CNV formation, we used a specific targeting strategy by crossing Ifnar1fl/fl mice with Cx3cr1CreER mice (Goldmann et al, 2013). This mouse line carries a tamoxifen‐inducible Cre recombinase under the control of the Cx3cr1 (fractalkine receptor) promoter (Yona et al, 2013). To be targeted by this genetic system, cells must express CX3CR1 at the time of tamoxifen application and be self‐maintaining over a longer time period. Thus, tamoxifen‐induced Cre recombinase expression is sustained in long‐lived microglia, whereas short‐lived myeloid cells lose Cre activity (Bruttger et al, 2015). We first tested microglia‐specific targeting of Cx3cr1CreER mice that were crossed with R26tomato Cre reporter mice and treated them with tamoxifen. The confocal analysis of retinal flat mounts in these Cx3cr1CreER:R26tomato animals revealed a prominent co‐localization of the tomato signal with Iba1 staining, indicating high recombination efficiency in microglia (Appendix Fig S2).
We next generated Cx3cr1CreER:Ifnar1fl/fl mice that did not show any apparent phenotypic abnormalities. Cx3cr1CreER mice and Ifnar1fl/fl mice carrying the floxed Ifnar alleles but lacking Cre expression were used as controls. PCR‐based genotyping and Western blotting with a specific antibody confirmed Cre‐mediated excision of Ifnar1 exon 10 (Appendix Fig S3A) and reduced Ifnar1 expression (Appendix Fig S3B) in retinal extracts of tamoxifen‐injected Cx3cr1CreER:Ifnar1fl/fl mice, respectively. Co‐staining of retinal sections with Ifnar1 and Iba1 also revealed only weak Ifnar1 signals that were mainly confined to Iba1+ monocytes located in inner retinal blood vessels (Appendix Fig S3C).
All three mouse lines were then subjected to laser coagulation and further analysis of immune cell behavior and CNV formation (Fig 3A). Ifnar‐deficient retinal Iba1+ cells displayed a strong amoeboid morphology compared to lesion‐associated cells from Cx3cr1CreER and Ifnar1fl/fl mice (Fig 3B–E). The mononuclear phagocytes in Cx3cr1CreER:Ifnar1fl/fl animals also showed significantly less ramifications in the grid cross analysis system, indicating a more reactive phenotype (Fig 3F). We then analyzed vascular leakage and CNV formation in tamoxifen‐treated Cx3cr1CreER:Ifnar1fl/fl mice and their two control counterparts. Both control mouse lines developed a typical laser‐induced vascular leakage at all three time points (Fig 3G–L). In contrast, Cx3cr1CreER:Ifnar1fl/fl animals showed a significantly increased vascular leakage after 14 days (Fig 3M–P). Quantification of lectin‐stained flat mounts then also revealed a significantly larger CNV size at 14 days after laser treatment in Cx3cr1CreER:Ifnar1fl/fl animals compared to Cx3cr1CreER and Ifnar1fl/fl controls (Fig 3Q–W). Finally, we analyzed the temporal correlation of Iba1+ cells with CNV lesions in retinal sections at day 3 and RPE/choroidal flat mounts at day 7 in all three mouse lines (Fig EV3). There was a clear overlap of amoeboid Iba1+ cells with increased lectin staining in tamoxifen‐treated Cx3cr1CreER:Ifnar1fl/fl animals compared to Cx3cr1CreER and Ifnar1fl/fl controls at both time points (Fig EV3). These data together clearly demonstrate that loss of Ifnar signaling in mononuclear phagocytes only causes sustained immune cell activation and exacerbates the development of laser‐induced CNV lesions.
Our study demonstrates a strong influence of Ifnar1 signaling on retinal microglia and macrophage activity and angiogenesis in the murine laser‐induced photocoagulation model that mimics several features of neovascular AMD (Lambert et al, 2013). Ifnar1−/− animals, lacking the α‐subunit of the type 1 interferon α/β‐receptor (Muller et al, 1994), showed a stronger and prolonged disease course after laser induction that was associated with an accumulation of predominantly amoeboid mononuclear phagocytes at the lesion sites. These results are in full agreement with a previous study reporting that interferon‐β/Ifnar1 signaling has potent immunomodulatory effects in experimental autoimmune encephalomyelitis (EAE), a model for autoimmune brain inflammation and multiple sclerosis (Prinz et al, 2008). IFN‐β was produced locally in the CNS of EAE animals, and mice lacking Ifnar1 expression in all tissues developed exacerbated clinical disease symptoms accompanied by stronger inflammation, demyelination, and lethality (Prinz et al, 2008). This negative influence of Ifnar1 deficiency on EAE was not seen upon Cre‐mediated deletion of Ifnar in B cell or T cells but was fully recapitulated by LysM‐Cre‐directed deletion of Ifnar in myeloid cells (Prinz et al, 2008). Similar findings were reported for the Ifnar1 ligand IFN‐β, where its gene disruption in mice caused augmented and sustained demyelination in EAE (Teige et al, 2003).
Microglia evolve from distinct primitive yolk sac progenitors (Kierdorf et al, 2013) and can be regarded as the primary innate immune effector cells in pathologies of the brain and the retina (Nimmerjahn et al, 2005; Kettenmann et al, 2011; Karlstetter et al, 2015; Zhao et al, 2015). To specifically address the role of Ifnar1 in retinal microglia function, we made use of a tamoxifen‐inducible Cx3cr1CreER mouse. This mouse line was established to specifically target microglia in vivo, by facilitating inducible microglia‐specific gene deletion in adult animals (Goldmann et al, 2013; Yona et al, 2013). Despite the redundant expression of Cx3cr1 on all myeloid cell populations, microglia can be distinguished from CNS infiltrating monocytes in Cx3cr1CreER mice by their unique features of self‐renewal and longevity (Wieghofer et al, 2015). However, other long‐lived resident mononuclear phagocyte populations located close to the retina such as choroidal macrophages may be also potentially targeted by this system (McMenamin, 1999). Our laser‐CNV analysis in tamoxifen‐induced Cx3cr1CreERT2:Ifnar1fl/fl animals showed a significantly enhanced retinal pathology comparable to that of complete Ifnar1 deletion. Thus, the disease‐promoting effects of Ifnar1 deletion seem to be confined to microglia and potentially other long‐lived macrophage subsets, revealing a significant contribution of these cell types to increase angiogenesis in the laser‐CNV model.
Since IFN‐β therapy is an established treatment option in relapsing remitting multiple sclerosis (Steinman et al, 2012) and inhibits EAE via different cellular and humoral mechanisms (Inoue & Shinohara, 2013), we postulated an immunomodulatory potential in the eye. Our data clearly revealed decreased microgliosis and less CNV in the laser‐damage model. This is in agreement with data that showed protective effects of IFN‐β in experimental autoimmune uveoretinitis, a model for human intraocular inflammation, by suppressing Th1 and Th17 cells (Sun et al, 2011). Moreover, systemic IFN‐β was tested in a CNV rabbit model without directly analyzing microglia/macrophage reactions (Yasukawa et al, 2002). In this model, subretinal injection of gelatin microspheres containing basic fibroblast growth factor (bFGF) triggered neovascularization for approximately 4 weeks. Continuous systemic therapy with dextran‐conjugated IFN‐β was very effective in lowering CNV in the incipient stage but did not affect CNV progression in later phases (Yasukawa et al, 2002). In contrast, our data showed significant effects of IFN‐β especially in the late phase. We hypothesize that the laser‐coagulation model mainly involves chronic inflammatory events whereas the bFGF model may act predominantly via the formation of neovascular membrane scars. In accordance with this hypothesis, IFN‐β treatment also ameliorated laser‐induced CNV in rabbits (Kimoto et al, 2002) and monkeys (Tobe et al, 1995). Of note, a patient with multiple sclerosis and punctate inner choroidopathy could significantly profit from systemic IFN‐β therapy and was subsequently free of active CNV (Cirino et al, 2006).
Taken together, our data in the laser‐coagulation model showed that Ifnar1 deficiency enhanced retinal microglia/macrophage reactivity and that IFN‐β inhibited this immune cell activation, vessel leakage, and CNV. Targeting Ifnar1/IFN‐β signaling may therefore highlight new therapeutic strategies for AMD and potentially other chronic inflammatory and degenerative diseases of the retina.
Materials and Methods
Animals and tamoxifen administration
Experiments were conducted with 6‐ to 10‐week‐old male and female Ifnar1−/− mice (Muller et al, 1994) and Cx3cr1CreER:Ifnar1fl/fl mice, which were obtained by breeding Cx3cr1CreER mice (Yona et al, 2013) and Ifnar1fl/fl animals (Kamphuis et al, 2006; Detje et al, 2009). Cx3cr1CreER mice were crossed with R26tomato reporter mice (Soriano, 1999). Ifnar1−/−, Ifnar1fl/fl, and R26tomato mice were on C57BL6/J and Cx3cr1CreER mice were on C57BL6/N background. All animals were maintained in an air‐conditioned environment at 22°C on a 12‐h light–dark schedule, had access to phytoestrogen‐free food and water ad libitum, and were health‐monitored on a regular basis. For induction of Cre recombinase, Cx3cr1CreER mice and Cx3cr1CreER:Ifnar1fl/fl mice were treated with 4 mg tamoxifen (T5648; Sigma) dissolved in 200 μl corn oil (C8267; Sigma) injected subcutaneously at two time points 48 h apart. The animals had consecutive numbers which were allocated to the genotype only after complete experimental evaluation. All experiments were approved by the governmental body responsible for animal welfare in the state of North Rhine‐Westphalia, Germany (Landesamt für Natur, Umwelt und Verbraucherschutz Nordrhein‐Westfalen, Germany), with the permission number Az 84‐02‐04‐2014‐A466.
Laser damage of the retina was performed using a slit‐lamp‐mounted diode laser system (Viridis). The mice were anesthetized by an intraperitoneal injection of 150 μl ketamine hydrochloride (100 mg/kg body weight, Ketavet; Pfizer Animal Health) and xylazine hydrochloride (5 mg/kg body weight, 2% Rompun; Bayer HealthCare) diluted in 0.9% sodium chloride. Before laser treatment, the pupils of the animals were dilated using Phenylephrin 2.5%–Tropicamid 0.5%. Three laser burns were created equally distributed around the optic nerve of both eyes (energy 125 mW, duration 100 ms, spot size 100 μm). The animals were divided into three groups and analyzed after 3, 7, or 14 days.
Animals were treated with recombinant human interferon‐beta 1a, produced in CHO cells (300‐02BC; PeproTech). A dose of 10,000 U diluted in 100 μl phosphate‐buffered saline (PBS) was injected i.p. every other day from day 0 until the end of the experiment. To minimize the effect of subjective bias, mice were treated cage‐wise and cages were allocated randomly to the experimental groups.
Preparation of flat mounts and immunohistochemistry
The eyes were enucleated and fixed in 4% paraformaldehyde (0335.2; Roth) for 4 h at room temperature. Retinal and RPE/choroidal flat mounts were dissected and retinal flat mounts were permeabilized overnight (5% Triton X‐100, 5% Tween‐20 in PBS). After blocking the unspecific antigens with BLOTTO (1% milk powder, 0.01% Triton X‐100 in PBS) for 1 h at room temperature, the retinal flat mounts were incubated in the primary antibody overnight at 4°C (Iba1, rabbit polyclonal, 019‐19741; Wako Chemicals). The RPE/choroidal flat mounts were incubated for permeabilization and blocking in 1% BSA (A2153; Sigma), 5% nonspecific goat serum (16210‐064; Gibco), and 0.3% Triton X‐100 in PBS overnight at 4°C, followed by an incubation with a 1:500 dilution of Iba1 antibody and 1:10 dilution of primary TRITC‐conjugated lectin (L5264; Sigma) to label microglia/macrophages and endothelial cells, respectively. All flat mounts were washed in PBS before incubating with a 1:1,000 dilution of goat anti‐rabbit AlexaFluor 488 nm‐conjugated secondary antibody (A11008; Life Technologies). After washing, retinal and RPE/choroidal flat mounts were mounted on a microscope slide and embedded with fluorescence mounting medium (S3023; DakoCytomation).
The stainings were evaluated with a Zeiss Imager M.2 with an ApoTome.2. The total number of Iba1+ cells and the number of amoeboid‐shaped cells were counted within a circular region of 200 μm diameter around the laser spots. Cellular morphology was analyzed using a grid system to determine the number of grid crossing points per cell (n = 40–70 cells; from at least 3 different retinas per group) (Chen et al, 2012). CNV areas in RPE/choroidal flat mounts were quantified with the spline function of the graphic tool of ZEN software (Zeiss).
Fundus fluorescein angiography (FFA)
The animals were anesthetized, the pupils were dilated, and the vascular leakage was determined with FFA using the Spectralis™ HRA device. Following SD‐OCT analysis, 100 μl of 2.5% fluorescein (Alcon) diluted in 0.9% sodium chloride was injected intraperitoneally. For FFA analysis, late‐phase pictures (10 min after fluorescein injection) were taken. To quantify the laser‐induced vascular leakage, the pixel intensity was measured in two regions of interests (ROIs) within and one ROI outside each laser spot using the image processing program ImageJ (NIH). Because three laser spots were induced per eye, we quantified the pixel intensity of six ROIs within and three ROIs outside the fluorescein leakages. After averaging the values and subtracting the background, one data point represented the mean laser‐induced leakage per eye. Eyes were excluded from the analysis in case of choroidal hemorrhages and when laser lesions completely fused with each other or the optic nerve head.
Statistical analysis was performed using GraphPad Prism 6. For analysis of two groups, unpaired Student's t‐test was performed. To compare more than two groups, one‐way ANOVA followed by Tukey's post‐test was used. All values are presented as mean ± SD. P‐values ≤ 0.05 were considered significant.
The paper explained
Age‐related macular degeneration (AMD) is a leading cause of blindness in the Western World with increasing prevalence in the aging population. AMD is often accompanied by an activation of the innate immune system involving complement factors and reactive mononuclear phagocytes. Interferon‐β is an immunomodulatory drug to treat multiple sclerosis, an autoimmune demyelinating disease of the brain. It was previously unknown whether interferon‐β signaling in retinal immune cells may be a therapy target for retinal degenerative diseases including AMD.
Using the laser‐coagulation model of neovascular AMD in complete or microglia/macrophage‐specific Ifnar1‐deficient mice, we report an essential role of interferon‐β signaling in regulating immune cell reactivity and pathological angiogenesis. Loss of Ifnar1 triggered microglia/macrophage activation, vessel leakage, and choroidal neovascularization (CNV). In contrast, IFN‐β therapy attenuated retinal immune cell response and CNV development.
Our findings indicate a key role for Ifnar1 signaling in retinal immune mechanisms. Although studied in an experimental model of laser‐induced CNV in mice, the immunomodulatory potential of IFN‐β is a promising new strategy for future therapy approaches to control chronic inflammation in AMD.
AL performed experiments, analyzed data, and wrote the manuscript. AC performed laser damage and angiography. RS performed experiments, and MP and UK contributed Ifnar1−/− and Cx3cr1CreER:Ifnar1fl/fl mice and corrected the manuscript. TL designed and supervised the study, obtained funding, and finalized the manuscript. All authors read and contributed to the final manuscript.
Conflict of interest
The authors declare that they have no conflict of interest.
Expanded View Figures PDF
The authors thank Prof. Thomas Wunderlich, Max Planck Institute for Metabolism Research, and Institute for Genetics, University of Cologne, for providing R26tomato mice. We also thank Prof. Andrew Dick, Unit of Ophthalmology, School of Clinical Sciences and School of Cellular and Molecular Medicine, University of Bristol, for helpful discussion and Khalid Rashid for critical reading of the manuscript. This work was supported by grants from the Graduate Program in Pharmacology and Experimental Therapeutics at the University of Cologne in collaboration with Bayer AG, the DFG (LA1203/6‐2, LA1203/9‐1, LA1203/10‐1, and FOR2240), the ProRetina Foundation, and the Hans and Marlies Stock Foundation.
FundingGraduate Program in Pharmacology and Experimental Therapeutics at the University of Cologne
This is an open access article under the terms of the Creative Commons Attribution 4.0 License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited.
- © 2016 The Authors. Published under the terms of the CC BY 4.0 license