Niche-derived soluble DLK1 promotes glioma growth

Highlights • Astrocytes secrete DLK1 after exposure to hypoxia or irradiation.• Soluble DLK1 promotes stemness in glioma, in part by increasing HIF-2alpha stabilization.• High levels of soluble DLK1 are associated with tumor aggressiveness and lethality.


Introduction
Glioblastoma recurrence following standard-of-care treatment with radiotherapy, surgery, and chemotherapy invariably gives rise to incurable lesions, and the median survival following diagnosis remains at 12 to 20 months despite recent advances in our understanding of glioblastoma at a molecular level [1] . Evidence from human patient samples and murine models of brain tumors suggest that inherent therapeutic resistance within a subset of tumor cells with stem cell characteristics may be the primary source of recurrent tumors [2 , 3] . While the origin and fate of such cells remain controversial, glioblastoma cell phenotypes are highly plastic, and non −stem-like cells can acquire characteristics of stem cells as a result of microenvironmental interactions with the extracellular matrix, with growth factors, or as a result of altered oxygenation or pH [3][4][5][6] . Indeed, tumor cells with stem cell properties appear to be spatially restricted to specific microenvironments such as the perinecrotic and perivascular niches, suggesting that these niches may control residing tumor cell phenotypes [4 , 5 , 7] .
Delta Like noncanonical Notch ligand 1 (DLK1) is a transmembrane protein in the Notch family of ligands, that is capable of signaling in a Notch-dependent and -independent manner depending on cellular context Statistical significance was determined with t test with Welch's correction for unequal variances applied to Pearson's coefficients (F-H). In the whole figure significance is represented as * P < 0.05 and * * P < 0.01 vs bulk tumor. DLK1, delta-like noncanonical Notch ligand 1; GFAP, glial fibrillary acidic protein. [8][9][10][11][12][13] . Expression of DLK1 is increased with tumor grade in glioma, and its signaling has been associated with various properties of aggressive tumor cells [14 , 15] . The mechanisms underlying these effects on tumor cell behavior remain poorly understood, but likely include signaling from the extracellular, soluble domain of DLK1 [16] . Indeed, soluble DLK1 secreted from astrocytes was recently shown to be a critical component of the subventricular zone neural stem cell niche [11] . Astrocytes represent a prominent cell type in the brain tumor microenvironment [17] , and recent studies revealed that DLK1 is one of the top upregulated genes in tumor-associated astrocytes of highgrade vs low-grade gliomas [18] . The regulation of DLK1 expression is poorly and t test (C-F), in E and F Welch's correction for unequal variances was applied. In the whole figure significance is represented as * P < 0.05, * * P < 0.01, and * * * P < 0.001 vs untreated controls. DLK1, delta-like noncanonical Notch ligand 1.
understood, but some elements of the brain tumor microenvironment such as hypoxia have been shown to drive DLK1 expression [19 , 20] .
Here, we sought to investigate the role of soluble DLK1 in the highgrade glioma tumor microenvironment. We found increased secretion of DLK1 from tumor-associated astrocytes subjected to stresses of the tumor microenvironment, such as hypoxia and ionizing radiation. Soluble DLK1 increased proliferation and stem cell characteristics of glioma cells, and promoted tumor growth in a genetically engineered mouse model of glioma. Together, our findings suggest that soluble DLK1 is a niche-derived mediator of aggressive tumor growth in brain tumors.
Soluble DLK-S was cloned into the RCAS vector and mice were co-injected with a 1:1 mix of DF1 cells expressing RCAS-PDGFB and RCAS-DLK-S or empty RCAS as indicated. Each litter was allocated to 1 experimental group. Mice were monitored daily and sacrificed upon displaying brain tumor symptoms. All procedures were approved by the Swedish Board of Agriculture through the Malmö-Lund Regional Committee (permit M186-14). The sample number was determined based on the law of diminishing returns with the resource equation method (total number of animals -total number of groups > 10). A total of 6 pups were excluded due to nontumor symptoms during week 0 to 3, the final numbers were n = 26 for PDGFB and n = 24 for DLK-S.

Immunofluorescence
Whole brains were embedded in optimal cutting temperature compound (OCT) (ThermoFisher) and frozen in precooled isopentane. Five micrometers thick cryosections were air-dried for 30 minutes, then fixed in ice-cold acetone or 4% paraformaldehyde and permeabilized using 0.3% Triton X-100 in phosphate buffered saline (PBS) (Sigma). Blocking was performed using serum-free protein block (DAKO), then sections were incubated overnight with primary antibodies at 4 °C with background reducing components (DAKO). Alexa Fluor secondary antibodies (Abcam) were used, and Vectashield Mounting medium with DAPI (Vector Laboratories) was used for mounting. Images were acquired using an Olympus BX63 microscope and DP80 camera and CellSens software (Olympus CellSens Software, RRID:SCR_016238).
For DLK1 and GFAP localization images ( Figure 1 ), minimal postproduction consisting of background subtraction and automated level optimization was equally applied with ImageJ (Fiji, RRID:SCR_002285).
Colocalization analysis was performed with ImageJ (Fiji, RRID:SCR_002285) Coloc2 plugin on the selected Regions of Interest of at least 3 independent experiments.
Ki67 quantification was performed with CellProfiler (CellProfiler Image Analysis Software, RRID:SCR_007358), at least 3 fields were analyzed for each tumor, for a total of 101,717 nuclei analyzed, with n = 45,466 for PDGFB tumors and n = 56,251 for DLK-S tumors.
Areas of necrosis were identified by 2 independent researchers based on cellularity and the presence of pseudopalisading nuclei, based on the DAPI stain. Vessels were identified by morphological inspection of DAPI and GFAP stains. replicates are as follow, (B) n = 5 for U3082MG, U3084MG, and U3065MG primary spheres and n = 3 PIGPC cells, (C) n = 8 except PIGPC where n = 3, (D) n = 4. Statistical significance was determined by t test, in C and D Welch's correction for unequal variances was applied. In the whole figure significance is represented as * P < 0.05, * * P < 0.01, and * * * P < 0.001 vs untreated controls. DLK1, delta-like noncanonical Notch ligand 1; PIGPC, primary murine glioma cell.

Proliferation assay
Thousand cells/well (PIGPC) or 2500 cells/well (U3082MG, U3084MG and U3065MG) were seeded in 96 well plates. For astrocyte conditioned media (ACM) transfer experiments, 24 hours after seeding, media from Astrocyte and transfected cells was filter sterilized and used to replace culture media. For astrocyte experiments, media was replaced every 2 to 3 days and proliferation was assessed after 9 days. For transfected cells, proliferation was assessed after 72 hours.
For recombinant protein experiments, 24 hours after seeding cells were treated with serial dilutions of recombinant DLK1 (0-200 ng/mL range) and grown for 72 hours. At the moment of the assay, 10 μL of WST-1 solution (Roche) were added to each well and after 2 hours of incubation at 37 °C and 5% CO2 450 nm absorbance was read on a Synergy 2 plate reader (BioTek).

Western blot
Cells were lysed in radioimmunoprecipitation assay buffer supplemented with Complete Phosphatase and Complete Protease inhibitor cocktails (Roche). After dilution in Laemmli buffer with DTT and boiled for 5 minutes, samples were loaded on 4% to 20% Mini-PROTEAN TGX Precast Protein Gels (Biorad). Proteins were transferred on polyvinylidene fluoride (PVDF) membranes using a Transblot Turbo System (Biorad), blocked in 5% nonfat dry milk/PBS, and incubated overnight at 4 °C with primary antibodies. After washing, membranes were incubated for 1 hour with secondary antibodies (Abcam). Images were acquired using a Fujifilm LAS 3000 Imager. Densitometric analysis were performed with ImageJ software (Fiji, RRID:SCR_002285). Band signal intensity was normalized for the respective loading control values (actin or SDHA).

Colony and sphere formation assays
Mechanical dissociation with Accutase (ThermoFisher) was used to prepare single cell suspensions. Cells were counted using a hemocytometer. For colony formation assay, 350 cells were seeded in 5 cm dishes coated with polyornithine (Sigma) and laminin (Biolamina). U3082MG, U3084MG and U3065MG cells were cultured for 14 days while PIGPCs for 8 days, under the indicated conditions, then washed in PBS and fixed using 4% paraformaldehyde. Cells were stained using 0,01% crystal violet/H2O. Wells were washed gently in water, then airdried for 24 hours. Images were acquired with a Fujifilm LAS 3000 Imager.
Sphere formation assay was performed with the hanging-drop method. 10 cells in 35 μL drops were seeded on the lid of a 48 well plate and grown under the indicated conditions for 2 weeks. For secondary sphere assay, primary spheres were pooled, pelleted, dissociated with Accutase and reseeded at the indicated conditions. Wells with spheres were manually counted and images were acquired with a Zeiss AX10 inverted microscope.

Real-time quantitative PCR
The RNeasy Mini Kit was used with Qiashredder (QIAGEN) according to the manufacturer's instructions for RNA isolation, and cDNA was synthesized using random primers and Multiscribe reverse transcriptase (Applied Biosystems). A QuantStudio 7 real-time PCR system (Applied Biosystems) with SYBR Green Master Mix (Applied Biosystems) was used for amplification. Gene expression levels were normalized to the expression of 3 housekeeping genes (UBC, SDHA, and YWHAZ) using the comparative CT method. Transfections and luciferase reporter assay DLK-S expression plasmid was kindly provided by prof. Anne Ferguson-Smith [11] , cloned into RCAS vector by classic restriction enzyme technique and transfected into DF-1 cells. For luciferase reporter assay, cells were cotransfected with hypoxia-responsive element (HRE)-luc (Addgene) [26] or 8xCSL-luc (gift from Håkan Axelson) and pCMV-renilla (Promega) and analyzed using the Dual-Luciferase Reporter Assay System (Promega) on a Synergy 2 platereader (BioTek). Xtreme gene 9 (Roche) reagent was used according to manufacturer's recommendations for transient transfections.
For proliferation experiments, the EC50 and span were estimated with a nonlinear regression curve, using a log. agonist vs normalized response (variable slope) equation fitted in Graphpad Prism 5 (GraphPad Prism, RRID:SCR_002798). For immunofluorescence experiments, colocalization was measure by Pearson's R coefficient in 3 independent experiments with ImageJ Coloc2 plugin. For Ki67 quantification, at least 3 fields were analyzed for each tumor, for a total of 101,717 nuclei analyzed, with n = 45,466 for PDGFB tumors and n = 56,251 for DLK-S tumors.
After normal distribution and variance similarity evaluation, 2-sided unpaired t test (eventual Welch's correction for groups with different variances), Mann-Whitney for nonparametric data, 1-way ANOVA with Bonferroni post hoc test and 2-way ANOVA (timelines only) tests were used to determine statistical significance, as indicated in respective figure legends. For survival evaluation, the Kaplan-Meier method was used to investigate variables and overall survival correlation, while a log-rank test was employed to compare survival curves. In all figures data are shown as mean ±SEM, analyzed using GraphPad Prism 5 software and significance expressed as P values ( * P < 0.05, * * P < 0.01, * * * P < 0.001).

Data availability
The data that support the findings of this study are available from the corresponding author upon reasonable request.

DLK1 is expressed and secreted by tumor-associated astrocytes in the glioma microenvironment
To test whether tumor-associated astrocytes could be a source of DLK1 in the glioma tumor microenvironment, we generated PDGFB/shp53-induced murine gliomas using the RCAS/tv-a system as previously described [14] , then co-stained tumors for the astrocyte marker glial fibrillary acidic protein (GFAP) and DLK1. While the bulk of the tumor cells appeared negative for DLK1 expression, DLK1 signal was detected in areas of GFAP staining both in perinecrotic and perivascular tumor areas, and both in GFAP positive and negative cells, suggesting DLK1 expression in astrocytes, tumor cells, and potentially other cell types present in these areas ( Figure  DLK1 gene expression was previously reported to be upregulated in isolated GFAP + tumor-associated astrocytes of high-grade glioma compared to those of lower-grade tumors [18] , further confirming DLK1 expression in astrocytes in this model system. We speculated that this DLK1 induction could be mediated by microenvironmental factors. As hypoxia is 1 major microenvironmental reality of high-grade glioma compared to low-grade glioma [5 , 30] , we cultured human fetal astrocytes under normoxic or hypoxic conditions for up to 10 days. We also subjected astrocytes to a single dose of irradiation to mimic another physiological response to a therapeutic intervention relevant to high-grade glioma. Both astrocytes subjected to growth in hypoxia and those subjected to irradiation displayed increased DLK1 levels secreted into the culture media, as measured by an ELISA assay ( Figure 2 A). The higher levels of DLK1 in the media were sustained for the entire 10 day period in the case of astrocytes cultured in hypoxia, whereas irradiated astrocytes displayed a peak secretion at 2 days post-treatment with levels returning to baseline after 9 days ( Figure 2 A). Together, these data support that astrocytes may secrete DLK1 into the tumor microenvironment in glioma.

Soluble DLK1 promotes glioma cell proliferation, survival and self-renewal
We next examined what effect soluble DLK1 may have on glioma cells. We first performed a media transfer experiment by treating human glioblastoma cell lines maintained in serum-free, stem cell-promoting conditions with media from astrocytes cultured in normoxic (ACM CTRL) and hypoxic (ACM 1% O 2 ) conditions for up to 9 days. In 2 out of 3 cell lines, media from hypoxic astrocytes induced a significant increase in the proliferation rate ( Figure 2 B). Since ACM contains many other different factors that may influence cancer cell growth, we then directly investigated the effects of soluble DLK1 with 2 different approaches. First, we transiently transfected human glioblastoma cell lines with a plasmid containing the Nterminal soluble part of DLK1 (DLK-S, His-tagged). All the 3 transfected cell lines overexpressed and secreted similar levels of DLK-S, as verified by western blot and ELISA experiments ( Figure 2 C −D). A media transfer experiment showed that 2 out of 3 cell lines significantly increased their proliferation ( Figure 2 E) when grown in DLK-S conditioned media. We then treated the human glioblastoma cell lines with a recombinant protein corresponding to the DLK1 secreted part, and once again, the 2 DLK-responding cell lines showed significant increase in their proliferation ( Figure 2 F). Taken together, these data demonstrate that soluble DLK1 is able to induce glioma cell proliferation, irrespectively of its origin.
As the use of the recombinant protein allows for better control of DLK1 concentrations, we then moved forward with this approach. We first generated a dose-response curve in all 3 human glioblastoma cell lines and in PDGFB-induced glioma primary cultures (PIGPCs) derived from the glioma mouse model. All the cell lines, with the exception of the nonresponding U3065MG, showed a dose dependent increase in cell proliferation, with a plateau obtained at 200 ng/mL DLK1 and EC50s in between 25 and 35 ng/mL ( Fig 3 A).
In line with these findings, all cell lines that responded to DLK1 in the proliferation assay also increased their colony formation ability in a dose-dependent manner when exposed to sub-maximal soluble DLK1 concentrations similar to those obtained in hypoxic astrocytes ( Figure 3 B −C).
Furthermore, soluble DLK1 strongly enhanced the self-renewal ability of responsive glioma cell lines, as measured by the serial sphere-formation assay ( Figure 4 A −B) performed at clonal density (Supplementary Figure 3), and induced a significant increase in the stem cell markers OCT4, NANOG, and SOX2 ( Figure 4 C). Since DLK1 has been reported to influence the Notch pathway [8 , 12] , we tested if these effects were Notch-dependent. Luciferase experiments performed at different time points revealed no significant alterations in Notch activity in any tested cell lines ( Figure 4 D).

DLK1-effects are mediated in part by HIF-2a
Because DLK1 secretion was increased by astrocytes under hypoxic conditions, and because of known previous links between DLK1 expression and function to hypoxia [20] , we asked whether soluble DLK1 could influence the hypoxic response of glioma cells. We cultured U3082MG and U3084MG human glioblastoma cells for 24 or 72 hours in 1% O 2 , stimulated or not with soluble DLK1. While there was no difference in HIF-1a stabilization with DLK1 treatment, Western blots showed significantly increased HIF-2a protein levels at 72 hours in cells cultured with soluble DLK1 ( Figure 5 A −B). This increased HIF-2a expression was reflected in a stronger hypoxic response, as 2/3 human glioblastoma lines and PIGPCs displayed increased activation of HREs at 72 hours of culture in hypoxia with DLK1 stimulation, as measured in an HRE-luciferase assay ( Figure 5 C). Moreover, analysis of PDGFB/shp53-induced murine gliomas revealed that both DLK1 and HIF-2a were strongly expressed and showed a significant co-localization only in the perivascular and perinecrotic niches ( Fig. 5 D).
As HIF-2a is a known driver of stem cell characteristics in glioma and other tumor forms [31 , 23 , 32-34 ], we next tested whether effects of DLK1 on glioma cell behavior were mediated by HIF-2a. The treatment with the specific HIF-2a inhibitor PT2385 at concentrations with significant effects on HIF-2a protein [35 , 36] (Supplementary Fig. 4) was able to revert the DLK1-induced increase in hypoxia response ( Figure 6 A). Similarly, while stimulation of glioma cells with soluble DLK1 boosted the increase in the expression of the stem cell marker genes NANOG, OCT4, and SOX2 in hypoxic cells, addition of PT2385 blocked this specific DLK1 effect in all cell lines tested ( Figure 6 B). Moreover, PT2385 decreased the colony formation ability of glioma cells exposed to soluble DLK1 in hypoxia ( Figure 6 C). Notably however, PT2385 treatment did not significantly affect DLK1induced gene expression in normoxia, and showed surprisingly modest effects in DLK1-untreated cells at hypoxia. Together, these data suggest that DLK1 promotes the glioma stem cell character in part via HIF-2a stabilization.

DLK1 promotes aggressive glioma growth in vivo
To test the effects of soluble DLK1 on glioma growth in vivo, we generated a mouse model for the overexpression of soluble DLK1 together with PDGFB using the RCAS/tv-a system ( Figure 7 A). Co-injection of RCAS-PDGFB with RCAS-DLK-S (soluble) resulted in more aggressive tumors as compared to RCAS-PDGFB with empty vector control, as measured by survival time following injections ( Figure 7 B). Evaluation of Ki67 expression revealed a significant increase in cell proliferation in murine DLK-S tumors as compared to controls ( Figure 7 C, D), thus confirming the in vitro data ( Figure 2 E −F, 3A). In agreement with our in vivo data using a mouse model that gives rise to a range of low-to-high-grade gliomas, analysis of the human TCGA LGGGBM data set [27] revealed that tumors expressing high levels of DLK1 were significantly more aggressive than those with low levels of DLK1 ( Figure 7 E), presumably as a result of the higher DLK1 levels reported in high-grade glioma. These findings were replicated in an independent data set (Supplementary Figure 5 A −B). Notably, in data sets comprising GBM only, DLK1 expression was either associated with slightly longer survival, or not associated with any significant survival difference at all, suggesting that the link between DLK1 and shorter survival in the LGGGBM data set is related to increased expression in higher-grade tumors (Supplementary Figure 5 A −B).

Discussion
An increasing focus on cancer stem cell characteristics has revealed parallels between normal neural stem cell regulation and cancer stem cell characteristics in brain tumors [37 , 38] . Control over tumor cell phenotypes by specific, local microenvironments within a tumor, for example, is reminiscent of the way that normal tissue stem cells reside within and rely on their niche to maintain the stem cell character [3 , 4] . It is likely that some of the same mechanisms involved in neural stem cell maintenance in the vascular niche of the subventricular zone may also be involved in maintaining the cancer stem cell character of brain tumor cells located in a perivascular or periarteriolar [39] niche. We describe one such example here: soluble DLK1 secreted from astrocytes appears to be involved in stem cell maintenance both of normal neural stem cells and glioma cells, as shown here. An association between DLK1 expression and aggressive tumor growth in glioblastoma has previously been established [20] . By generating a mouse model for testing effects of soluble DLK1 overexpression specifically in the context of glioblastoma, we show that the previously reported association between DLK1 expression and tumor grade in glioma [14 , 15] may at least in part be caused by DLK1 itself, as soluble DLK1-overexpressing tumors had a higher proliferation rate and significantly decreased mice survival as compared to controls. It is important to note that DLK1 expression is not limited to tumor-associated astrocytes, and soluble DLK1 may be derived both from other stromal cell types and tumor cells themselves. Our experiments indicate that soluble DLK1 affects tumor cell proliferation similarly regardless of whether it was produced by astrocytes or tumor cells. In a previous study, we described release of and signaling from the intracellular domain of DLK1 in glioma cells [14] . Data presented here do not link signaling from the soluble DLK1 to that of the intracellular fragment, however, both appear to be regulated by the hypoxic tumor microenvironment. It is yet unclear whether or not DLK1 expression is required in tumor cells to be affected by soluble DLK1 secreted into the niche [11] .
As with astrocyte-derived DLK1 in regulation of normal neural stem cells, the exact mechanism(s) by which soluble DLK1 signals to glioma cells remains to be investigated. We show here that soluble DLK1 can contribute to a stronger and more prolonged response to hypoxia, as mediated by increased HIF-2a stabilization in DLK1 treated cells. This effect on HIF-2a stabilization indeed seemed important for the tumor-promoting effects of DLK1 signaling as inhibition of HIF-2a transcriptional activity by use of the specific HIF-2a inhibitor PT2385 abolished all effects of DLK1 on stem cell marker gene expression and colony formation under hypoxic conditions. Interestingly, DLK1 expression itself has been shown to be regulated by hypoxia in other cell systems [19 , 20] , suggesting that there may be a DLK1-HIF feedback loop in hypoxic tumor cells. In the present investigation, effects of DLK1 treatment were enhanced by hypoxic culture conditions. Importantly, however, HIF-2a inhibition did not significantly affect DLK1mediated stem cell marker expression under normoxic conditions, suggesting that there are other mediators downstream of DLK1 that can contribute to DLK1 signaling in glioma.
Several questions remain regarding signaling mediated by soluble DLK1, including that of potential receptors for DLK1. Among the human glioma cell lines and genetically engineered glioma mouse model tested in this study, all but one cell line (U3065MG) responded to soluble DLK1. Based on the experiments presented here, it is difficult to determine the reason for the lack of response in this cell line. It is possible that U3065MG cells lack expression of necessary components downstream of soluble DLK1. Further investigation into potential receptors and downstream mediators of DLK1 signaling is warranted, to better infer the applicability of the DLK1-mediated effects described in this study. Furthermore, it is somewhat counterintuitive that soluble DLK1 appears to simultaneously promote glioma cell proliferation and stem cell characteristics, as it is widely assumed that stem-like cells display lower proliferation rates than more differentiated, non −stem-like cells.

Conclusions
Taken together, our data support a role for soluble DLK1 as a tumorpromoting stem cell niche factor in glioma. Further research is warranted to investigate whether or not signaling by DLK1 can be therapeutically targeted, either via HIF2-a inhibition or by targeting upstream signaling.