doi:10.1093/braincomms/fcz043 BRAIN COMMUNICATIONS 2020: Page 1 of 20 | 1 BRAIN AIN COMMUNICATIONS Phagocytic glioblastoma-associated microglia and macrophages populate invading pseudopalisades Elena Saavedra-López,1 Meritxell Roig-Martı́nez,1 George P. Cribaro,1 Paola V. Casanova,1 José M. Gallego,2 Ana Pérez-Vallés3 and Carlos Barcia1 Hypoxic pseudopalisades are a pathological hallmark of human glioblastoma, which is linked to tumour malignancy and aggres- siveness. Yet, their function and role in the tumour development have scarcely been explored. It is thought that pseudopalisades are formed by malignant cells escaping from the hypoxic environment, although evidence of the immune component of pseudopali- sades has been elusive. In the present work, we analyse the immunological constituent of hypoxic pseudopalisades using high-reso- lution three-dimensional confocal imaging in tissue blocks from excised tumours of glioblastoma patients and mimic the hypoxic gradient in microfluidic platforms in vitro to understand the cellular motility. We visualize that glioblastoma-associated microglia and macrophages abundantly populate pseudopalisades, displaying an elongated kinetic morphology across the pseudopalisades, and are oriented towards the necrotic focus. In vitro experiments demonstrate that under hypoxic gradient, microglia show a par- ticular motile behaviour characterized by the increase of cellular persistence in contrast with glioma cells. Importantly, we show that glioblastoma-associated microglia and macrophages utilize fibres of glioma cells as a haptotactic cue to navigate along the an- isotropic structure of the pseudopalisades and display a high phagocytic activity at the necrotic border of the pseudopalisades. In this study, we demonstrate that glioblastoma-associated microglia and macrophages are the main immune cells of pseudopalisades in glioblastoma, travelling to necrotic areas to clear the resulting components of the prothrombotic milieu, suggesting that the scav- enging features of glioblastoma-associated microglia and macrophages at the pseudopalisades serve as an essential counterpart for glioma cell invasion. 1 Neuroimmunity Research Group, Department of Biochemistry and Molecular Biology, School of Medicine, Institut de Neurociències, Universitat Autònoma de Barcelona, Bellaterra, Cerdanyola del Vallès, Barcelona 08193, Spain 2 Department of Neurosurgery, Valencia General Hospital, Valencia 46014, Spain 3 Department of Pathology, Valencia General Hospital, Valencia 46014, Spain Correspondence to: Carlos Barcia, PhD, Neuroimmunity Research Group, Department of Biochemistry and Molecular Biology, School of Medicine, Institut de Neurociències, Lab M2-107, Universitat Autònoma de Barcelona, Bellaterra, Cerdanyola del Vallès, Barcelona 08193, Spain E-mail: carlos.barcia@uab.es Keywords: glioblastoma; pseudopalisades; hypoxia; microglia; macrophages Abbreviations: 3D ¼ three-dimensional; BV ¼ blood vessel; CD ¼ cluster of differentiation; COL-IV ¼ collagen IV; D ¼ distal; DAPI ¼ 40 ,6-diamino-2-phenylindole; GAMMs ¼ glioblastoma-associated microglia and macrophages; GBM ¼ glioblastoma; GFAP ¼ glial fibrillary acidic protein; HA ¼ hyper-cellular area; HLA ¼ human leucocyte antigen; HS ¼ horse serum; I ¼ inter- mediate; Iba-1 ¼ ionized calcium-binding adapter molecule 1; Ig ¼ immunoglobulin; MHC ¼ major histocompatibility complex; MTOC ¼ microtubule-organizing centre; NF ¼ necrotic focus/foci; P ¼ proximal; PBS ¼ phosphate-buffered saline; PPs ¼ pseudo- palisades; TP ¼ tumour parenchyma Received April 2, 2019. Revised September 2, 2019. Accepted October 9, 2019. Advance Access publication December 4, 2019 C The Author(s) (2019). Published by Oxford University Press on behalf of the Guarantors of Brain. V This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/4.0/), which per- mits non-commercial re-use, distribution, and reproduction in any medium, provided the original work is properly cited. For commercial re-use, please contact journals.permissions@ oup.com 2 | BRAIN COMMUNICATIONS 2020: Page 2 of 20 E. Saavedra-López et al. Graphical Abstract Introduction In the present work, we revisited and studied the neu- roinflammatory compound of PPs in human GBM speci- Pseudopalisades (PP)-s are one of the critical pathological mens, and we attempted to understand its function in features of glioblastoma (GBM; Brat and Van Meir, this tumour microenvironment. We particularly focused 2004; Rong et al., 2006). They are densely populated on studying GBM-associated microglia and macrophages layers of cells, surrounding certain necrotic areas of the (GAMMs) because they are a predominant immune-cell tumour (Wippold et al., 2006). Their origin and function population in this type of brain tumour (Poon et al., are still debated but they are considered as a canonical 2017), and most importantly, because it is established GBM diagnostic feature for neuropathologists, as well as that GAMMs predominantly have a pro-tumoural influ- an indicator of high malignancy and invasive capacity. ence (Hambardzumyan et al., 2016). However, the ma- Previous studies have described PPs as a population of chinery by which they contribute to the invasive features tumour cells that escape from the necrotic and hypoxic of GBM is still poorly understood. foci to form rather organized structures (Brat and Van We observed that a large population of GAMMs coex- Meir, 2004; Brat et al., 2004). However, despite the ist as a big part of the cellular structure of GBMs’ PPs. advances in GBM research, the cause and implications of Experiments with cell culture microfluidic platforms show PPs in brain tumours still remain mysterious. that microglial cells are highly motile compared with gli- Importantly, these strata of gathered cells are orderly dis- oma cells and show particular dynamics when exposed to posed around necrotic foci (NF), geographically associ- a hypoxic gradient, consistent with the migrating features ated with a central blood vessel (BV), which appears to observed in PPs. Detailed study of kinetic microglia under be occluded upstream by factors derived from the tumour hypoxia revealed the preferential orientation of the micro- cells and probably by the tumour mass development itself tubule-organizing centre (MTOC) towards the cells’ lead- (Rong et al., 2006). It is thought that the lack of oxygen ing edge. This feature was then analysed in GAMMs in and probably low glucose levels cause massive cell death the PPs of human biopsies, demonstrating the centripetal in these areas. A circumstance that creates a hypoxic mi- and anisotropic tendency of these cells towards the NF. lieu where tumour malignant cells escape from, in a cen- Finally, three-dimensional (3D) reconstructions and trans- trifugal way, forming these aligned structures. The cells parencies of tissue blocks revealed the haptotactic inter- building up PPs are placed in very close apposition to action of GAMMs with fibrous tumour cells and the each other, delimiting necrotic tissue and probably con- increased phagocytic capacity of migrating GAMMs in taining this adverse environment. Considering the active the internal limit of the PPs. With this study, we demon- cell death process within these areas, and the high quan- strate that a large proportion of the cellular component tity of cellular debris, it would be expected to be associ- of PPs is microglia and macrophages, having a significant ated with a robust inflammatory response. However, the implication for the dynamics of glioma. most cited and comprehensive study of PPs to date, These findings, unascertained until now, challenge pre- reported very limited, if any, amount of inflammatory vious knowledge describing a barely existent immune cells in these arranged structures (Brat et al., 2004). component in this GBM hallmark and indicate that GAMMs contribute to GBM invading niche BRAIN COMMUNICATIONS 2020: Page 3 of 20 | 3 GAMMs migration and debris clearance in this tumour Immunohistochemistry location is essential for glioma cell invasion. Free-floating immunohistochemistry by diaminobenzidine detection was performed on tumour sections to visualize immune-cell markers and glial-tumour-cell markers. In Material and methods order to identify GAMMs, antibodies against antigens commonly expressed by macrophages were used. Anti-in- Human tissue studies flammatory macrophages were identified by positive immunostaining of the cell-surface marker Cluster of Patients and samples Differentiation 163 (CD163), whereas the panmacrophage/ For this research, we selected biopsies from six patients microglia marker used was Ionized calcium-binding adapt- from a cohort of samples of the General Hospital of er molecule 1 (Iba-1; also named AIF-1). Human leucocyte Valencia diagnosed with GBM containing evident necrotic antigen (HLA)-DP, -DQ, -DR (major histocompatibility PPs. Before surgery, the patients presented the classical complex II or MHCII) was also used to see the myeloid GBM physiopathology and symptomatology, with the cell population within the pro-inflammatory spectrum. expected clinical evolution and the confirmation of the Specifically, primary antibodies used were those to recog- tumour presence by neuroimaging. Tumour excision was nize: Iba-1 (1:500, polyclonal rabbit immunoglobulin (Ig) performed at the surgery room and biopsies were handled G; Wako Pure Chemical Industries, Ltd.; Osaka, Japan); with minimal manipulation. Part of each sample was CD163 (1:150, monoclonal mouse IgG; Abcam; used for routine neuro-pathological examinations and Cambridge, UK); and MHCII (1:100, mouse IgG1; Dako diagnosis, while the samples used from this study came Cytomation; Glostrup, Denmark). Samples were first from 2.5 ml of the tissue, which were submerged during treated with citrate (10 mM, pH 6, 60 C, 20 min) to 1 week in 4% paraformaldehyde before being washed ameliorate antibody binding by antigen retrieval. At that with phosphate-buffered saline (PBS) 0.1 M/l, pH 7.4 and point, 0.3% hydrogen peroxide was used to block en- stored in PBS with 0.1% sodium azide until being sec- dogenous peroxidase, and sections were blocked for 1 h tioned by vibratome (Leica Microsystems, Germany) into with 0.5% Triton X-100 with 10% horse serum (HS10%; 60 lm slices. The study was performed with the approval Sigma-Aldrich; St. Louis, MO, USA) before incubating of both institutions involved and in compliance with the overnight with primary antibody diluted in Trizma Base 1964 Declaration of Helsinki, including the informed Saline (TBS)-0.5% Triton X-100, 1% horse serum and consent of the patients. The excised tumour had the typ- 0.1% sodium azide. The primary antibodies used are ical GBM pathological features, such as high glial fibril- described above. Secondary antibodies were biotinylated lary acidic protein (GFAP) and Vimentin reactivity, as goat anti-mouse or biotinylated goat anti-rabbit (1:1000, well as Ki-67 expression, and showed different layers of goat; Dako; Denmark), diluted in 0.5% Triton X-100 the tumour, from the peripheral healthy-like tissue to the with 1% horse serum and 0.1% sodium azide, accordingly necrotic core. Besides the clear presence of PPs, all to the host of the primary antibodies. Secondary antibod- tumours showed hyperchromatism and pleomorphism, ies were detected by the Vectastain Elite ABC horseradish with marked cellularity, glomeruloid vessels, gemistocytic peroxidase method (Vector Laboratories; CA, USA). formations and aberrant mitoses. Sections were mounted on gelatinized glass slides and were dehydrated through graded ethanol solutions (70%, 80%, Histochemistry 90% and 100%) and submerged in Xylene before cover- Thionine staining was performed to analyse the anatom- slipping. The sections were then analysed at the micro- ical structure of the tissue, detect the nuclei of the cells scope (Eclipse 80i microscope, Nikon), taking the required and visualize the cell-division status of each tumour. For pictures with a digital camera connected to the microscope each sample, we were able to see the classical GBM (DXM 1200F Digital Camera, Nikon). histopathological features, including the presence of PPs (Fig. 1). For this, thionine (acetate; Certistain; Merck Immunohistofluorescence KGaA; Darmstadt, Germany) solution (25 mg/100 ml acet- Free-floating immunofluorescence was performed on tu- ate buffer pH: 4.6) in acetate buffer solution (Stem solu- mour sections for multiple stainings to visualize Iba-1þ tion A, sodium acetate 0.1 M; Stem solution B, Acetic cells with their MTOC or combined to also see GFAPþ acid 0.1 M) was used. All the samples were first mounted cells and the cells’ relation to BV. Antigen retrieval and on gelatinized slides, dried and then rinsed in distilled blocking were done as previously mentioned before incu- water, stained in thionine solution for 5 min, washed in bating during 48 h with the primary antibodies diluted in distilled water, dehydrated in increasing concentrations of 1% HS. The primary antibodies used were anti-human ethanol and Xylene. Then, sections were cover-slipped to Iba-1 as described for the immunohistochemistry (1:500, be analysed at the conventional light microscope (Eclipse rabbit polyclonal; Dako Cytomation; Glostrup, 80i microscope, Nikon), taking the required pictures with Denmark); anti-MHCII (1:100, mouse IgG1; Dako a digital camera connected to the microscope (DXM Cytomation; Glostrup, Denmark); anti-GFAP (1:500, 1200F Digital Camera, Nikon). chicken IgY; Abcam, Cambridge, UK); anti-c-tubulin (to 4 | BRAIN COMMUNICATIONS 2020: Page 4 of 20 E. Saavedra-López et al. Figure 1 GAMMs constitutively form PPs in GBM. (A) Confocal image and diagram of a classic PPs structure, containing HA, central BV and NF. [Diagram based on confocal image with characteristic architecture of a PP. BV identified by COL-IV (green) is located at the centre of the NF. Malignant tumour cells and remaining astrocytes, identified by GFAP (red) appear in a centrifuging manner, escaping from the necrotic areas (as previously described by Brat et al., 2004). A clear accumulation of DAPIþ nuclei (blue) can be appreciated forming the palisade arrangement]. (B) Thionine-based Nissl staining evidences necrotic PPs in GBM samples. (C) Adjacent histological sections of the HA show a dense population of elongated Iba-1þ GAMMs, with a scarce frequency of the pro-inflammatory spectrum (MHCII), and expression of the anti-inflammatory spectrum (CD163þ) delineating the HA (asterisk indicates the central BV). (D) Confocal transparency shows that the area of high density of DAPIþ nuclei (blue) (HA) is densely populated by Iba-1þ GAMMs. (E) 3D reconstruction of the PP (shown in D) displays elongated Iba-1 cells at the HA. (F) Three-dimensional reconstruction from E with a central clipping plane to visualize elongated cells at the HA. (G) Confocal transparencies show the distribution of Iba-1 (top panel) and MHCII (bottom panel) within the HA. Measurements of the relative fluorescence are shown in the plots obtained from the region of interest (yellow broken line) at the HA. MHCII is notably increased at the NF whereas Iba-1 expression is steady along the HA. detect the MTOC; 1:500, mouse IgG1; Sigma-Aldrich; Carlsbad, CA, USA; likewise for CD31); and on the Saint Louis, MO, USA); anti-CD31 (1:200, mouse mono- other hand, 488 Goat anti-rabbit 1:1000 (Life clonal IgG1; Abcam, Cambridge, UK); anti-collagen IV Technologies; Carlsbad, CA, USA) and AlexaFluor 555 (COL-IV; 1:200, rabbit polyclonal; Abcam, Cambridge, Goat anti-Chicken 1:500 (Life Technologies; Carlsbad, UK); and anti-CD3 (1:100; rabbit polyclonal; Dako CA, USA) to see the cells’ interactions. When staining Cytomation; Glostrup, Denmark). Secondary fluorescent CD3, the secondary antibody used was AlexaFluor 647 antibodies diluted in 1% HS were used accordingly with Goat anti-rabbit 1:1000 (Life Technologies; Carlsbad, the host of the primary antibodies. For the study of the CA, USA). Then 40 ,6-diamino-2-phenylindole (DAPI; cells’ directionality, we used AlexaFluor 488 Goat anti- 1:1000; Life Technologies; Carlsbad, CA, USA) was used rabbit 1:1000 (Life Technologies; Carlsbad, CA, USA; to stain the nuclei of the cells. Sections were mounted on also used to detect anti-COL-IV) and AlexaFluor 555 glass slides and cover-slipped using Prolong antifade re- Goat anti-mouse IgG1 1:1000 (Life Technologies; agent (Invitrogen; Carlsbad, CA, USA). GAMMs contribute to GBM invading niche BRAIN COMMUNICATIONS 2020: Page 5 of 20 | 5 Confocal microscopy, 3D analysis and quantification In the case of the Iba-1/GFAP staining, a grid was Human biopsies stained by immunofluorescence were drawn on the image with Image J software on 23 palis- imaged with a confocal laser-scanning microscope (LSM ade images from five tumours. This grid was made of 700, Carl Zeiss, Oberkochen, Germany) utilizing a Plan- nine centred 1600 mm2 squares, which determined three Apochromat 40/1, 3 Oil DIC M27 objective lens (Carl areas, which we again called P, I and D, corresponding Zeiss) and processed with the ZEN 2010 software (Carl to the relative distance of the zones to the NF. Of the Zeiss). Three-dimensional reconstructions were generated nine squares, only the central column was taken into ac- using the a blending softwares ZEN 2010, Imaris count, and only one central slice of each palisade was (BitPlane, South Windsor, USA) and iLlucida FX (iLlucida analysed to obtain the values of the number of nuclei, LLC, Los Angeles, USA) as we recently described (Diaz the different types of phagocytic events and the kind of et al., 2018). Z-stacks in the regions of interest were intercellular contacts (longitudinal or perpendicular), acquired to cover the depth of the tissue (25–30 images/ which were then extrapolated in order to obtain values stack) with a 0.5–lm-step size, and the images for quanti- per mm2. The percentage of Iba1þ cells was then com- fication were always taken with the NF on the upper part pared in the HA and in the tumour parenchyma (TP). of the image, considering that the scanning field could be The TP images were taken in the same samples and with rotated and oriented. Raw images were imported into the same conditions, and equally a grid of 1600 lm2 both Imaris (Bitplane Zurich, Switzerland) and Image J squares was established for stereological purposes. To see (version 1.4.3.67, NIH, USA) softwares, where they were the number of contacts in general, by means of the further processed for morphometric analyses and cell ImageJ software, a cross grid with an area of 800 mm2 counts. For the cells’ directionality, in order to better visu- per point was drawn. An intermediate optical slice of alize the MTOC, zoom 1.7 was used, while the Iba-1/ each Z-scan per palisade was chosen, and Iba-1þ cells GFAP staining was visualized with zoom at 1.0. On the hitting the crosses were analysed to see the number of former, the Iba-1 channel composing the Z-stack was contacts with GFAPþ cells in the depth of the z-stack. iso-surfaced using Imaris Surpass, in order to view a 3D This way we were able to determine the contacts between representation of the Iba-1þ cells. To detect the MTOC Iba-1þ cells and GFAPþ material in the different areas of within the cells, the module Spots was used. Within the the palisade, as two rows of crosses corresponded to cell module, the nuclei were detected and then the each of the ‘P’, ‘I’ or ‘D’ zones, as we resolved before. MTOCs were imported with the Vesicles tool. The Iba-1 surfaces were imported as the cells’ cytoplasms. This way a blending software Imaris, (Bitplane, Zurich, Switzerland) Cell culture studies and iLlucida FX (iLlucida LLC, Los Angeles) allowed the Cell culture and treatment free rotation of the 3D reconstructions of 23 images from BV2 (murine microglia), HAPI (rat microglia; both all tumours, which were analysed to see Iba-1þ cells’ dir- obtained from the Cell Cultures unit at the Institut de ectionality by the observation of the distribution of the Neurociències) and GL261 (murine glioma; purchased MTOC of interest within the cells. The same procedure from the National Cancer Institute at Frederick) were was performed when analysing the cells’ diameter and per- maintained in RPMI-1640 medium (Sigma-Aldrich, Saint imeter at the different positions within the PPs. This time, Louis, MO) supplemented with 10% heat-inactivated foe- seven PPs were divided into three regions, which we called tal bovine serum, 0.1% penicillin/streptomycin at 37 C “proximal” (P), “intermediate” (I) or “distal” (D), corre- and 5% CO2 in a humidified incubator. C6 (rat glioma; sponding to the relative distance of the zones to the NF as Obtained from Culture Collections, Public Health illustrated in descriptive diagrams (Fig. 2). The criteria for England, Salisbury, UK) was maintained in F12 medium this regionalization were established by the following (Sigma-Aldrich, Saint Louis, MO) supplemented with 5% defined steps: once a PP was identified, the internal and foetal bovine serum, 0.1% penicillin/streptomycin at external limits of the hyper-cellular area (HA) were set 37 C and 5% CO2 likewise in a humidified incubator. In according to the position of the NF. Then, an equidistant order to do immunostainings, cells were seeded on cover- distribution of the three regions of interest (P, I and D) slips pre-treated with poly-L-lysine at a concentration of was established according to these limits. As indicated 25 lg/ml and washed twice with PBS. above, the confocal sampling was made allowing the rota- In regard to seeing how glucose deprivation affected tion of the scanning head, always positioning the NF on cells, murine microglia and glioma cell lines were seeded the upper part of the image (Fig. 2). This rotation allowed into 24-well culture plates 1 day before treatment at a the analysis of the orientation of GAMMs in 3D recon- density of 50 000 cells/ml, and the day of the experiment structions. Further morphometric studies were performed the medium was changed for fresh medium, either glu- on single stereological dissectors of confocal images of cose-free RPMI-1640 supplemented with 0.1% penicillin/ eight PPs, either in the HA or the NF, and by means of streptomycin, or the same medium but also supplemented image analysis software (Image J), we analysed the Area, with L-glucose at a final concentration of 2 g/l. After Aspect Ratio, Roundness and Circularity of cells in these 24 h, cells were fixed with 4% paraformaldehyde before two areas. immunostaining. 6 | BRAIN COMMUNICATIONS 2020: Page 6 of 20 E. Saavedra-López et al. Figure 2 Differential distribution of GAMM and tumour cells within PPs. (A) Diagram illustrating the criteria utilized for the classification of the HA in proximal (P), intermediate (I) and distal (D) regions. (B) The 3D reconstructions of a representative human tissue block at a PP containing the NF and HA, which was divided in P, I and D. High-resolution scanning frames of the HA were captured with rotated orientation comprising P, I and D for quantification considering x-, y- and z-axis. (C) Representative zenithal view of the areas of quantification. (D) Estimation of nuclei density in the HA of PPs. (E) Estimation of GFAPþ nuclei (left) and percentage from the total nuclei (right). (F) Quantification of the percentage of Iba-1þ cells in the HA compared with the TP. (G) Estimation of Iba-1þ nuclei (left) and percentage from the total nuclei (right). GAMMs contribute to GBM invading niche BRAIN COMMUNICATIONS 2020: Page 7 of 20 | 7 For the hypoxia experiment, both murine cell lines inhibitor (Gibco) and 0.1 ml MgSO4 at 3.8%. Solution 4 were seeded into 24-well culture plates 1 day before consisted of 8.4 ml Solution 1 and 1.6 ml Solution 3. treatment as previously described, and the day of the ex- Finally, Solution 5 was a mix of 5 ml Solution 1, 40 ll periment the medium was replaced with fresh supple- MgSO4 3.8% and 6 ll CaCl2 1.2%. After the whole mented RPMI-1640 medium and cells were incubated in brain was extracted, the meninges were discarded and the either the humidified incubator with normoxic (21% oxy- required tissue was separated from the rest of the brain. gen) conditions, in the same conditions with CoCl2 It was cut in small sections and re-suspended in 15 ml (100 lM) as a positive control for hypoxia; or in the Solution 1. This was followed by centrifugation (30 s, hypoxic (Baker Ruskinn InvivO2 200) workstation, al- 1500 rpm) and the pellet containing the cells was re-sus- ways at 37 C and 5% CO2. After 24 h, cells were fixed pended in the Solution 2 and incubated 10 min at 37 C, with 4% paraformaldehyde before immunostaining. with gentle mixing every 2–3 min to better digest the tis- Microfluidic plates (CellASICTM ONIX M04G-02-5PK, sue. This enzymatic digestion was stopped when adding EMD Millipore Corporation; Hayward, CA, USA) were Solution 4 and everything was centrifuged again at used to record the natural movement of BV2 and GL261 1500 rpm. The pellet was re-suspended in 3 ml Solution 3 cells, thus under normoxic and normoglycemic condi- and mechanical disaggregation was performed by gently tions. For this, after treating the plate with poly-L-lisine pipetting up and down with a Pasteur pipette 10 times. (25 lg/ml) and washing it, cells were plated at a low The cells were isolated into a uniform single-cell suspen- density to avoid confluence during the experiment. After sion by the use of a cell strainer and gently pipetting 24 h in the humidified incubator, flows of normally sup- again up and down 10 times. All this cell suspension was plemented RPMI-1640 were created, what permitted us added to the tube with Solution 3 before centrifuging to record the cells in an always-fresh medium. 5 min at 1000 rpm. Again, the supernatant was discarded The same microfluidic plates (CellASICTM ONIX and the pellet was re-suspended in 10 ml DMEM supple- M04S-02, Merck) were also used to see how BV2 and mented with 1% penicillin/streptomycin and 10% foetal GL261 cells reacted to a hypoxic gradient, simulating bovine serum, medium in which the cells were later on pseudopalisades’ dynamics, similarly to a recently pub- cultured. After re-suspension, cells were counted to be lished system (Ayuso et al., 2017). Hence, after treating cultured at the desired density, 300 000 cells/ml, in a cul- the plate with Poly-L-lisine (25 lg/ml) and washing it, ture flask or 24-well plates at 37 C and 5% CO2. The cells were plated at a low density to avoid confluence medium of these cells was partially changed (50%) for during the experiment. Creating flows of Normoxic fresh medium every week until confluence was achieved. RPMI-1640 medium normally supplemented and the Then, the flasks were shaken at 300 rpm for 2 h in order same medium but bubbled with N2 in order to displace to extract the microglia and seeded on poly-L-lysine pre- the O2 in it (hypoxic medium), we achieved the creation treated coverslips at 100 000 cells/ml. of a gradient in the chambers where the cells were In order to study the phagocytic capacity of microglial seeded. cells, a primary culture of rat cortical microglia was per- The position of the MTOC was analysed in BV-2, formed like described above. Once the culture reached GL261 and HAPI cell lines, as well as primary rat micro- confluence, the glia was agitated and the cortical micro- glia. While the cell lines were seeded as previously glia extracted was cultured at 100 000 cells/ml in 24-well described in poly-L-lysine pre-treated coverslips at a con- plates with 50% conditioned media. To activate micro- centration of 50 000 cells/ml; primary rat microglial cells glia, interferon gamma (25 ng/ml) and lipopolysaccharide were seeded at a concentration of 100 000 cells/ml. For (10 ng/ml) were added. After 24 h, astrocytes from the this to be done, newborn Sprague Dawley rats (post-natal same rats or early-passage C6 glioma cells were collected days 0–2) were sacrificed in order to obtain the glia from by tripsinization and co-cultured with the microglia at their cortex and/or mesencephalon. The animals were also 100 000 cells/ml for 2.5 h before cell culture fix- obtained from the animal facility of the Universtitat ation. To detect C6 glioma cells in the co-culture, cells Autònoma de Barcelona (UAB), where the experiments were previously centrifuged (1500 rpm, 2 min) and re-sus- were carried out, following the protocol approved by the pended in PBS to be stained with CellMaskTM Deep Red Ethics Committee on Animal and Human Research of the Plasma Membrane stain (1:1000, ThermoFisher UAB. Five solutions were used throughout the culture, Scientific), for 150 at 37 C, before dilution to 100 000 and all of them were filtered (0.2 mm filter) prior their cells/ml in their medium. Alternatively, GFAP immunocy- use: Solution 1 consisted of 50 ml Krebs buffer (120 mM tofluorescence was also performed in the co-cultures as NaCl, 4.8 mM KCl, 1.2 mM KH2PO4, 25 mM NaHCO3, described in the following section. 14.3 mM Glucose), with 0.15 g BSA (Sigma-Aldrich, St. Louis, MO, USA) and 0.4 ml MgSO4 3.8%. Solution 2 Immunocytofluorescence was created with 10 m Solution 1 and 2.5 mg trypsin The cells were stained to analyse their morphology. After (Sigma-Aldrich, St. Louis, MO, USA). To make Solution cell fixation, paraformaldehyde was washed with PBS 3, 10 ml of Solution 1 were mixed with 0.8 mg DNase and cells were treated for antigen retrieval with PBS with (Sigma-Aldrich, St. Louis, MO, USA), 5.2 mg of trypsin 0.02% saponine for 7 min. After washing this solution, a 8 | BRAIN COMMUNICATIONS 2020: Page 8 of 20 E. Saavedra-López et al. mild blocking solution was added (PBS with 0.01% sapo- done triplicates. Only cells with clear protrusions with an nine, 10 mM glycine) for 15 min before blocking with a aspect ratio >1.5 were taken into account. more concentrated solution for 1 h (PBS with 0.01% sap- In order to analyse the cells’ morphology, all replicates onine, 10 mM glycine, 5% BSA). The primary antibody were sampled homogeneously using a fluorescence (Nikon anti-c-tubulin (1:500, mouse IgG1; Dako Cytomation; Eclipse 90i) microscope attached to a DXM 1200F digital Glostrup, Denmark) was diluted in PBS with 0.01% sap- camera. Pictures were taken at a 20 objective using the onine, 1% BSA, and left in a humidity chamber over- software ACT1 version 2.70 (Nikon corporation). night at 4  C. AlexaFluor 555 Goat anti-mouse IgG1 Regarding the cells’ kinetic activity, we imaged the cells 1:1000 (Life Technologies; Carlsbad, CA, USA) was used in the microfluidic plates with an Eclipse TE2000-E as a secondary antibody for 45 min, and after washing, inverted microscope, taking pictures at 10 every 15 min Alexa Fluor 647 Phalloidin (Life Technologies; Carlsbad, for 24 h. Three time-lapse videos per condition were ana- CA, USA) was used for the analysis of the protrusions lysed to track the cells in hypoxic gradient, using the and Alexa Fluor 488 Phalloidin (Life Technologies; Chemotaxis Tool Plugin in ImageJ (version 1.4.3.67, NIH, Carlsbad, CA, USA) was used for the morphometric anal- USA). The analysis of the motility provided the following yses, both at 1:40 for another 45 min. In the case of the parameters: directionality, centre of mass, accumulated dis- co-cultures, primary rat microglia and C6 rat glioma cells tance, velocity and Euclidean distance (note: Euclidean dis- were stained to visualize the phagocytic capacity. After tance is defined as the ordinary straight-line distance cell fixation, paraformaldehyde was washed with PBS between two points. Increasing Euclidean distance indicates and cells were treated for antigen retrieval with PBS with a straighter displacement, which in cell biology is recog- 0.02% saponine for 7 min. After washing this solution, a nized as cellular persistence). mild blocking solution was added (PBS with 0.01% sapo- nine, 10 mM glycine) for 15 min before blocking with a more concentrated solution for 1 h (PBS with 0.01% Statistical analysis Saponine, 10 mM glycine, 5% BSA). The primary anti- For human tissue, the number of cells in the palisades, bodies used were: anti Iba-1 (1:500, rabbit IgG; Wako, the contacts and the phagocytic events were calculated Osaka Japan) or anti- OX42 (1:500, mouse IgG2a; from over 20 3D reconstructed multi-colour z-stacks of Abcam; Cambridge, UK) to stain the microglial cells and the PPs along the tumour samples (total 3D stacks, anti-GFAP (1:500, chicken IgY; Abcam, Cambridge, UK) n ¼ 23). The detailed analysis of the cells’ directionality, for C6 and astrocytes. They were diluted in PBS with 3D individual cell-reconstructions, iso-surface rendering, 0.01% Saponine, 1% BSA and left in a humidity cham- morphology and MTOC position within PPs was scruti- ber overnight at 4 C. AlexaFluor 488 Goat anti-rabbit nized from 59 cells (n ¼ 59). In vitro, experiments for the IgG 1:1000 (Life Technologies; Carlsbad, CA, USA) was study of the MTOC position with respect to the protru- used as a secondary antibody to detect Iba-1þ cells; sion and the cells’ morphology were performed twice, AlexaFluor 488 Goat anti-mouse IgG2a 1:1000 (Life while the time-lapsed cells tracked were 30 cells each ex- Technologies; Carlsbad, CA, USA) was used as a second- periment (n ¼ 30). Student’s T-test was used for depend- ary antibody to detect OX42þ cells; while AlexaFluor ent/independent samples and Mann–Whitney U test was 555 anti-chicken 1:500 (Life Technologies; Carlsbad, CA, used to compare data between groups. For instance, the USA) to see C6 cells and astrocytes. The secondary anti- cells’ directionality was seen manually, and T-tests were bodies were incubated for 45 min. Nuclei were counter- performed to see if there was a statistical difference be- stained with DAPI for 5 min before washing with PBS tween the number of cells going towards or away from and mounting the coverslips on glass slides using Prolong the necroses. The cells’ perimeter and diameter in the antifade reagent (Invitrogen; Carlsbad, CA, USA). tumours were given by the Imaris software when selecting individual cells, and the results were submitted to Microscopy, analysis and Kruskal–Wallis test with Dunn’s multiple comparisons test, as the data did not follow a normal distribution. quantification For the same reason, the number of nuclei, the phagocyt- Cells stained to see their MTOC and actin in 647 were ic events and the intercellular contacts were submitted to sampled with the confocal laser-scanning microscope (LSM Kruskal–Wallis tests with Dunn’s multiple comparisons 700, Carl Zeiss) and a Plan-Apochromat 40/1, 3 Oil test. The results of the percentage of Iba1þ cells in the DIC M27 objective lens (Carl Zeiss), to be processed with HA and TP, the cells’ morphology, as the kinetic differ- the ZEN 2010 software (Carl Zeiss). Three-dimensional ences between BV2 and GL261 lines, were analysed by reconstructions were generated using the a blending soft- Student’s T-test. Cells’ persistence when time-lapsed was wares ZEN 2010, Imaris (BitPlane, South Windsor, CT, determined by Rayleigh test, where P  0.05 indicated USA) and iLlucida FX (iLlucida LLC, Los Angeles, CA, directionality. In all cases, P  0.05 was considered sig- USA). Z-stacks were acquired to cover the entire mono- nificant. Graphs show the mean values obtained 6 stand- layer of the culture (20–25 images/stack) with a 0.5 lm ard error of the mean. All statistical analyses were step size. One picture was taken for each replicate, being performed using Prism 6 software (GraphPad Software), GAMMs contribute to GBM invading niche BRAIN COMMUNICATIONS 2020: Page 9 of 20 | 9 except the Rayleigh test, which was performed by ImageJ’s chemotaxis plugin. Data availability All the raw data supporting the results of the present study are available at the reader’s request. Results GAMMs densely populate PPs in human GBM We selected GBM samples from a cohort of biopsies obtained from patients that underwent surgery and based on the histological features, we evaluated the presence of Video 1 The 3D rotation of a human tissue block canonical PPs in the tissue sections. Necrotic PPs were containing a PP to visualize GAMMs at the HA. A clipping identified by a HA, well organized and distributed sur- plane is set to facilitate the identification. rounding a NF with a central BV (Fig. 1 and Supplementary Fig. 1). Nuclear staining clearly revealed the HA, accompanied by GFAP immuno-reactive cells demonstrating the presence of malignant cells and remaining reactive astrocytes. Marker for vascular base- essentially characterized by hypoxia (Brat et al., 2004), ment membrane, COL-IV, also demonstrated the central yet coexistent with hypoglycaemia, may induce this mor- position of BV along the 3D organization of PPs (Fig. 1 phological change. Therefore, we set up an in vitro ex- and Supplementary Fig. 1). Serial sections of the tumours periment with microglial and glioma cell lines to analyse with identifiable PPs were stained for several GAMM morphometric parameters of the two cell types under markers (Iba-1, CD163 and HLA-DR), as well as for a these conditions. Our experiments revealed that microglial general marker for T lymphocytes (CD3). Iba-1 and cells, in contrast with glioma cells, become patently elon- CD163 immunohistochemistry showed cells clearly outlin- gated in hypoxic environments suggesting a different mo- ing the PPs at the HA, demonstrating the big proportion tility pattern in these settings, compatible with the of the immune component of the PPs. However, HLA- palisade-like arrangement (Fig. 3E–I). On the contrary, DR (the human MHCII form) staining did not reveal an after hypoglycaemic conditions, although GL261 glioma evident palisade outline, being patent outside the HA at cells did become elongated, no changes in elongation the PPs borders (Fig. 1C and G). In addition, T cells were seen in microglial cells (Supplementary Fig. 3). were only limited to the BV surroundings but absent Thus, we focused on hypoxic conditions for studying from the HA of the PPs (Supplementary Fig. 2). microglial kinetics. In order to know whether elongated Our transparent 3D reconstructions of the tissue blocks morphology coincides with cell motility, we analysed in allowed the visualization of elongated GAMMs crossing the time-lapse the behaviour of microglia. Importantly, micro- HA areas (Fig. 1 D–F and Video 1). Furthermore, we compart- glial cells in movement adopt a particular shape when mentalized the HA of the PPs in three regions, proximal (P) to travelling, characterized by extended structure, the forma- the NF: intermediate (I); and distal (D), the latter appearing tion of a protruding leading lamella and a trailing uro- close to the malignant TP (Fig. 2A–C); and quantified the pro- pod (Fig. 4A–C and Supplementary Video 2). F-actin portion of Iba-1þ GAMMs and GFAPþ cells (Fig. 2C–G). The staining evidenced the actin-rich lamella and the kinetic percentage of Iba-1þ GAMMs in the HA had a tendency to be shape of microglial cells (Fig. 4C–E). Quantification and slightly higher than the rest of the tumour. However, we detailed analyses revealed an apparent increase of kinetic observed a differential distribution of cells within the HA, morphology of microglia under hypoxic, in contrast with being the density of GAMMs higher at the P region contrast- hypoglycaemic conditions (Fig. 4F and G). Differently, ing with GFAPþ cells, which were fewer in this area (Fig. 2 no significant changes were seen in glioma cells and Supplementary Video 1). (Supplementary Fig. 4A and B). Since the position of the MTOC is potentially oriented towards the leading edge GAMMs become elongated in in migrating cells (Nemere et al., 1985; Etienne- Manneville, 2004), we also labelled the MTOC, with c- hypoxic conditions Tubulin to analyse its positioning with respect to the Since GAMMs at the HA showed elongated morphology leading protrusion (Fig. 4). We observed that the posi- differently from an amoeboid morphology found elsewhere tioning of the MTOC in kinetic microglia (BV2, HAPI (Fig. 3A–D), we hypothesized that this environment, and primary microglia) was preferentially located adjacent 10 | BRAIN COMMUNICATIONS 2020: Page 10 of 20 E. Saavedra-López et al. Figure 3 Microglia adopt elongated morphology in hypoxic conditions. (A) The 3D reconstruction of a human GBM PP containing Iba- 1þ GAMMs within the palisading HA and NF. (B) GAMMs at the HA showed elongated morphology evidenced by a reduction of roundness. (C) Detail of morphological differences between Iba-1þ cells from NF and HA from human GBM. (D) GAMMs show reduced circularity in HA areas when compared with NF. (E) BV2 microglial cells show elongated shape under hypoxic conditions. (F) Representative confocal images of BV2 cells in normoxia and hypoxia. (G) Representative images of glioma cells in normoxia and hypoxia. (H) BV2 microglial cells, in contrast with non-myeloid GL261 cells, show an increase in cell area, perimeter and elongated morphology 24 h after hypoxic treatment. (I) No significant differences in morphometric parameters were seen in glioma cells. GAMMs contribute to GBM invading niche BRAIN COMMUNICATIONS 2020: Page 11 of 20 | 11 Figure 4 Hypoxia induces oriented kinetic morphology on microglia. (A) Time-lapse video reveals the steps that microglial cells undergo when migrating in contrast with neighbouring static cells. (B) Confocal analysis of microglial kinetic morphology evidenced by F-actin (green), c-Tubulin (red) and nuclei counterstained with DAPI (blue). (C) (1) BV2 microglial cell in kinetic movement from time-lapse video frame. (2) Maximum intensity projection of BV2 kinetic microglia shows an actin-rich lamella, trailing uropod and c-Tubulin-rich MTOC. (D) Categories of kinetic microglia (adjacent or opposite) according to MTOC position in relation to nucleus and leading lamella. (E) HAPI microglial cell line (a) and primary rat microglia (b) show equivalent distribution of the MTOC in relation to the nucleus, protruding lamella and trailing uropod when kinetic. (F) Quantification of non-kinetic and kinetic morphology of BV2 microglia in normoxia versus hypoxia. (G) Quantification of kinetic and non-kinetic morphology in medium with or without glucose in BV2 microglia. (H) Quantification of MTOC position in kinetic-shaped BV2 microglia in normoxia and hypoxia. (I) Quantification of MTOC positioning in BV2 microglia with kinetic morphology in glucose-rich medium versus hypoglycaemia. A minimum of 20 cells (n ¼ 20) was quantified for each parameter and condition. 12 | BRAIN COMMUNICATIONS 2020: Page 12 of 20 E. Saavedra-López et al. to the leading edge in contrast to opposite (back) position GAMMs interact haptotactically (Fig. 4; 95% in BV2, 70% in primary microglia and with GFAP cells 65% in HAPI) indicating that the motility of microglial migrating cells encompasses this MTOC positioning. In To further understand the strategy of microglial cells to addition, similar MTOC position was appreciated in kin- navigate the HA of the PPs, we analysed the potential etic GL261 glioma cells (94%; Supplementary Fig. 4C interactions of GAMMs with GFAPþ cells. We analysed in and D). Importantly, this location was maintained under 3D and high-resolution detail, the combined GFAP/Iba-1 hypoxic conditions suggesting that the motility and direc- immunofluorescence staining. We observed GFAPþ cells tionality of microglia are run by this centriole position displaying a fibrous anisotropic nature at the PPs (Fig. 4 and Supplementary Fig. 4). (Supplementary Fig. 5), in contrast with the isotropy found in other areas, and GAMMs frequently contacting GFAPþ material at the HA, sharing a similar anisotropic orienta- PP-located GAMMs show tion (Fig. 7 and Video 8). These intercellular contacts appeared higher at the I area (Fig. 7), implying the hapto- preferential cellular orientation tactic travelling of GAMMs through the HA. These inter- according to necrosis location cellular appositions seem to be critical since 100% of GAMMs at the I of the HA were interacting with GFAP Knowing the leading MTOC position adopted in kinetic- fibres (Fig. 7 and Videos 9 and 10). Interestingly, the elongated microglial cells under hypoxic conditions, we interactions were established with the tumourigenic GFAP analysed in 3D the location of MTOC in GAMMs within fibres both perpendicular and longitudinal to the NF. the PPs of human GBM samples to establish their prefer- Quantifications revealed that longitudinal appositions were ential orientation. We analysed with high resolution and significantly longer than the perpendicular ones demon- Z-depth the HA of the PPs separating P, I and D regions, strating the joined anisotropic orientation of GAMMs and according to the NF position (Fig. 5). We generated spe- GFAPþ fibres (Fig. 7 and Video 10). These results imply cific 3D iso-surfaces and spots in imaging software to lo- that GAMMs may utilize the elongated GFAP fibres of cate the orientation of the MTOC (Fig. 5 and Video 2). the PPs to migrate towards the NF and together with the Our 3D reconstructions show that GAMMs become differential GFAPþ density along the HA (previously larger and elongated at the HA, being more prominent at shown in Fig. 2) suggest that haptotaxis could be a crucial the I area (Fig. 5C–E) and preferentially orienting the strategy of cell motility in this microenvironment. MTOC towards the NF at I and D regions (Fig. 5F–G and Video 3), which indicates that GAMMs preferentially migrate to the NF. Interestingly, GAMMs at the P area Microglial phagocytic events do not show a preferential directionality suggesting that they may have reached their target location. increase at the P regions of the HA Since the haptotactic migration of GAMMs at the PPs is polarized towards the NF, we were wondering about the Microglia show cellular persistence physiological relevance of reaching this area. Particularly, GAMMs found at the NF of the PPs show a pro-inflam- under hypoxic gradient matory phenotype evidenced by MHCII reactivity Since the directionality of GAMMs at the HA in human (Fig. 1), which may indicate an increase of scavenging PPs is mostly leading towards the NF, we set up an properties. Thus, to ascertain this, we explored the in vitro gradient experiment to know whether or not a phagocytic capacity of microglia in vitro after classical hypoxic gradient induces changes in microglial migratory pro-inflammatory stimulus. We analysed in co-cultures behaviour. We observed that hypoxia induces a signifi- the ability of primary microglia to phagocytose C6 gli- cant escape of glioma cells towards normoxic media as oma cells after activation with interferon gamma and previously demonstrated by others (Brat et al., 2004; lipopolysaccharide. Microscopy analysis of this in vitro Ayuso et al., 2017), which corroborates the centrifugal scenario revealed abundant phagocytic events after classic characteristics of glioma invasion. In contrast, microglial pro-inflammatory activation (Supplementary Fig. 6), veri- directionality is not changed either towards normoxia or fying the phagocytic competence of microglia towards gli- hypoxia but seems to be affected by the set up gradient oma cells when activated. (Fig. 6 and Videos 4–7), as cells migrate in a different Thus, considering the scavenging and phagocytic cap- way. Microglial cells appear clearly elongated and navi- acity of GAMMs, we explored and imaged with high gate in a straighter manner, defined by the significant in- resolution the phagocytic events along the PPs. We crease of the Euclidean distance (Video 5). Thus, observed events of GAMMs phagocytosing pyknotic nu- microglial cells increase their cellular persistence when clear material or GFAPþ fragments (Fig. 8). Phagocytic exposed to a hypoxic gradient, which explains the elong- GAMMs, detected by Iba-1 immunolabelling, were identi- ation at this setting and clarifies the non-centrifugal be- fied by having a DAPI-low/Iba-1þ nucleus, and contained haviour towards the NF at the PPs. a contiguous Iba-1- space, the phagosome, containing GAMMs contribute to GBM invading niche BRAIN COMMUNICATIONS 2020: Page 13 of 20 | 13 Figure 5 Analysis of GAMMs MTOC position in relation to NF within PPs. Human samples stained for Iba-1 and c-Tubulin were analysed with 3D high-resolution confocal microscope. We defined the MTOC position towards the NF or TP in proximal (P), intermediate (I) and distal (D) areas. (A) (1) Indicative PP diagram for orientation. (2) Confocal image of a representative HA of a necrotic PP between the TP and NF limits. (3) Separation of P, I and D areas in relation to NF proximity. (4) The scanning field at HA was set up rotating the x and y axes of the system (green square) according to the NF and TP orientation (orange arrow). (5) Images were analysed with these criteria of orientation. (B) Representative confocal image of GAMMs labelled with Iba-1 (green), c-Tubulin (red) and DAPIþ nuclei (blue) within the HA of the PPs. At the higher magnification of the MERGE the MTOC can be identified as c-Tub accumulations. (C) Quantification of the area and aspect ratio of GAMMs in NF and HA. (D) Cells could be rendered in 3D to visualize the shape and MTOC position regarding the nucleus and NF location. Representative cells are indicated by 1 and 2. (E) Quantification of diameter and perimeter of Iba-1 cells at the PPs show a significant increase of the size at the intermediate area (compatible with elongated shape seen in vitro). (F) Diagram of representative cells (1 and 2) with prevalent MTOC orientation. (G) The quantification of the percentage of cell orientations shows that the majority of cells are oriented towards the NF except in the P area. 14 | BRAIN COMMUNICATIONS 2020: Page 14 of 20 E. Saavedra-López et al. anti-inflammatory spectrum phenotype at this anatomical compartment. In contrast, GAMMs within the pro-inflam- matory spectrum (MHCIIþ) emerge at the verge of the NF of PPs. These results suggest that GAMMs may switch their phenotypic spectrum, from anti- to pro-inflammatory, when arriving to the P areas. This is consistent with the increase of MHCII (HLA-DR) immunoreactivity at the necrotic border, particularly at the P region of the HA, where GAMMs significantly increase their phagocytic capacity. By contrast, the anti-inflammatory spectrum appearance is compatible with the migratory behaviour at the I area Video 2 Representative 3D processing of the images from of the HA. Since CD163 is a receptor that interacts with human PPs. Original stacks stained for Iba-1, c-Tubulin and DAPI haemoglobin/haptoglobin complex (Kristiansen et al., were treated with the following steps: step 1, Creation of DAPI iso- 2001; Schaer et al., 2007; Wang et al., 2018), CD163 ex- surface; step 2, detection of MTOC with c-Tub staining using Imaris Spots; step 3, creation of Iba-1 iso-surface; step 4, filtering of entire pression may imply that GAMMs migration towards the cells; and step 5, merging the iso-surfaces and spots to detect cells’ NF may be facilitated by the haemorrhage and throm- orientation. bosis created by the central BV clot. This could result in a haemoglobin/haptoglobin-rich environment in the nec- rotic areas, product of erythrocyte lysates, that may leak to the HA in a gradient manner. GAMMs motility in tumours, and particularly in GBM, is not fully understood. In the neoplastic context, the hypoxic environment of PPs influences GAMMs motility in a different way than glioma cells. On the one hand, our cell culture set up demonstrated, as pre- viously reported (Brat et al., 2004), that hypoxia indu- ces the escape of glioma cells towards normoxic conditions, what is congruent with their centrifugal es- cape in PPs. On the other hand, hypoxic gradient does not induce a chemotactic effect on microglia but im- portantly induces a straighter migrating behaviour com- patible with the higher elongation and their straight Video 3 Analysis of a processed image to determine the directionality found at the human PPs. These results are orientation of GAMMs regarding the MTOC position. consistent with the fact that macrophages may be reprogrammed under hypoxic conditions, having a ten- dency to present an anti-inflammatory spectrum pheno- type (Raggi et al., 2017), being hypoxia a trigger of DAPI-high/Iba-1- elements, or GFAPþ material (Fig. 8 and migration and the transition to an alternative spectrum Video 11). Careful examination of the samples and quan- (Leblond et al., 2016). tification showed a differential frequency of phagocytic Besides the effect of hypoxia, our study also reflects that events within the three areas of the HA, being significant- GAMMs may use a haptotactic strategy to navigate the ly higher at the P area (Fig. 8). These results indicate HA of the PP. This strategy is not rare in tumour microen- that GAMMs actively cross the HA of the PPs searching vironments, and specific physical changes as the fibrillary for cellular elements to be engulfed and eliminated. disposition, may directly intervene in the invasion process (Oudin and Weaver, 2016). Interestingly, mathematical models also show that haptotaxis could be the connection between cell locomotion and the resulting persistent walk Discussion behaviour on long-time scales (Dickinson, 2000). Thus, tis- The present study demonstrates that PPs in GBM are sue fibres can serve as a haptotactic cue. As tumour cells populated by a large number of GAMMs that actively enhance their efficiency of migration restricting protrusions migrate towards the NF to phagocytose cellular debris. along aligned fibres such as collagen (Riching et al., 2014), This migration appears to be taking place in a haptotac- in our scenario, the fibrillary nature of GFAPþ cells at the tic manner, being GFAPþ material the substrate and HA may also facilitate the persistent motility of GAMMs facilitated by the hypoxic gradient. within the PPs. This is consistent with the elongation and PP-integrating GAMMs appear elongated at the HA and focalization of the GAMMs’ protrusions and the variation express CD163, suggesting a skewing of GAMMs to an of GFAPþ density along the HA. GAMMs contribute to GBM invading niche BRAIN COMMUNICATIONS 2020: Page 15 of 20 | 15 Figure 6 Microglial cells do not escape from hypoxia and navigate straightly. (A) Microglia behave similarly when exposed 24 h to normoxic (representative cell trajectories in plot graph 1 and distribution of migrating angles in rose graph 2) and hypoxic conditions (analogous representation of cell tracks in 3 and rose graphs in 4). Black tracks are cells ending in the upper quadrant while red tracks are cells ending in the lower quadrant. (B) Glioma cells migrate away from hypoxic conditions. When comparing the exposure to normoxic media (1 and 2) and hypoxic gradient (3 and 4) during 24 h, glioma cells move away from the hypoxic source. Green dots in the graphs (1 and 3) represent the centre of mass. Rayleigh test for directionality revealed a significant directionality only for glioma GL261 cells [P < 0.01 (P ¼ 0.00282)] in a hypoxic gradient towards normoxia in contrast with no gradient (P > 0.05) and microglia BV2 cells (P > 0.05 with and without gradient). In all cases, microglia show higher motility, migrating with higher velocity (C) and travel longer distances (D) than glioma cells. (E) Although hypoxia– normoxia gradient does not modify the velocity and accumulated distance, microglia show higher Euclidean distance, meaning that they migrate straighter in these conditions. The tracks of 30 cells (n ¼ 30) were quantified for each parameter and condition. The advantageous effect for tumour expansion of angiogenesis (Casazza et al., 2013), highlighting the rele- GAMMs recruitment at the PPs is to be understood, but vance of the microenvironment to determine pro-tumou- it could be related with the generation of new BVs and ral activity (Rivera and Bergers, 2013). Furthermore, the debris clearance. In fact, it has been demonstrated that haemoglobin-haptoglobin receptor CD163 may indeed tumour-associated macrophages promote angiogenesis in play a critical role in this particular process. Recently, al- tumours and specifically in glioma (Chen et al., 2014). though in a different clinical scenario, it has been found Thus, our results suggest that particularly the migration that haemoglobin-haptoglobin-induced CD163 signalling of anti-inflammatory GAMMs to these thrombotic areas in macrophages promotes the production and release of may induce vasculogenesis in this environment, facilitat- angiogenic factors (Guo et al., 2018). Coherently, the ex- ing the development of the tumours. Previous studies in pression of CD163þ by microglia-macrophages populat- other neoplastic scenarios, such as experimental models ing PPs suggests their putative angiogenic properties to of breast cancer, have shown that anti-inflammatory stimulate GBM growth within this microenvironment. skewed macrophages migrate to avascular hypoxic We also show that, while tumour cells escape from the areas promoting tumour progression by stimulating hypoxic core, GAMMs travel through the HA towards 16 | BRAIN COMMUNICATIONS 2020: Page 16 of 20 E. Saavedra-López et al. Video 4 Microglial cells present a rapid random walk when Video 6 Glioma cells present a slow random walk when exposed to normoxic conditions. A 24-h time-lapse is shown exposed to normoxic conditions. A 24-h time-lapse is shown on the left panel. The track of three representative cells is displayed on the left panel. The track of three representative cells is displayed in the central panel. A merge of the two panels is shown on the in the central panel. A merge of the two panels is shown on the right. right. Video 7 Glioma cells slowly escape the hypoxic Video 5 Microglial cells present cellular persistence when environment when exposed to hypoxic gradient. A 24-h exposed to hypoxic gradient. A 24-h time-lapse is shown on time-lapse is shown on the left panel. The track of three the left panel. The track of three representative cells is displayed in representative cells is displayed in the central panel. A merge of the the central panel. A merge of the two panels is shown on the right. two panels is shown on the right. Tumour infiltrated macrophages/microglia play a sup- the NF to phagocytose, probably repairing and contain- portive role of GBM (Zhou et al., 2015); thus, it is crit- ing the adverse environment. This may also be an im- ical to understand their behaviour within the different portant process of the development of the PPs, microenvironments of the tumour and how interplay paradoxically contributing to the tumour expansion. with the growth of the tumour mass occurs. Particularly Apoptotic cells at the necrosis may be a strong attracting for PPs, since they are distinguishable pathological hall- signal for phagocytes to this site (Cullen et al., 2013; marks of the malignancy of GBMs, the role of GAMMs Elliott and Ravichandran, 2016). While apparently the in this scenario becomes relevant for the progression of phagocytic capacity of GAMMs may serve in favour of the tumour, and therefore, should be considered as an tumour clearance, the focus on phagocytosis of apoptotic important niche for therapeutic intervention. We suggest corpses may as well divert microglia macrophages from that GAMMs in PPs may be a critical factor for tumour eliminating tumour cells (Reiter et al., 1999). In PPs, the development in this microenvironment since they repair location is especially critical because the phagocytic cap- the vaso-occlusive and prothrombotic ecosystem of the acity is limited to the P areas towards the NF, which rep- NF that may contribute to accelerate tumour mass resents the opposite direction of the tumour expansion. growth and invasion. Alternatively, since the phagocytic As a result, corpse clearance and repair at the NF can capacity of GAMMs persists in GBM, new therapeutic also be essential for preparing the tissue for new neoplas- approaches should be oriented towards the re-program- tic invasion or, as mentioned above, the establishment of ing of these cells towards the phagocytosis and clearance new vasculature. of malignant cells in the TP. GAMMs contribute to GBM invading niche BRAIN COMMUNICATIONS 2020: Page 17 of 20 | 17 Figure 7 Haptotactic interaction of GAMMs with GFAP1 fibres at the human PPs. (A) Representative PP by confocal scanning the nuclei (DAPI), GAMMs (Iba-1) and glioma cells (GFAP) at the HA limited by the NF and TP. (B) The 3D reconstruction of the image shown in A. (C) Confocal visualization of GAMMs at the HA near the NF limit. Higher magnification of broken white line square is displayed to the right to show GAMMs interacting with GFAPþ fibrils, also shown in different optical planes and indicated in orthogonal views at the Z planes. (D) Diagram representing the analysis of Iba-1-GFAP interactions based on C including the orthogonal views. (E) Quantification of the number of interactions with GFAP per Iba-1þ cell at the P, I and D areas. The number of contacts is significantly increased at the I area. (F) Representative Iba-1 cells interacting with fibrous GFAP cells at the PPs. (G) Quantification of the percentage of Iba-1 cells establishing contacts with GFAP cells at the HA of the PP. (H) The 3D reconstruction of Iba-1 and GFAP interactions at the HA. Representative perpendicular (1) and longitudinal (2) interactions are indicated in white broken-line squares, which are shown in higher magnifications in 10 and 20 . Orthogonal view at the level of the longitudinal interaction is shown. Illustrations of perpendicular and longitudinal interactions are shown in 100 and 200 . (I) Quantification of the number and distance of longitudinal and perpendicular contacts at the P, I and D areas. (J) The 3D reconstruction of GAMMs within the HA of a human PP in apposition to GFAP fibrils. Zoom from white rectangle is shown. Oblique view is shown in image on the right. 18 | BRAIN COMMUNICATIONS 2020: Page 18 of 20 E. Saavedra-López et al. Figure 8 Increased phagocytic capacity of GAMMs at the NF proximity. (A) Engulfing GAMMs at the human PPs evidenced with Iba-1 and DAPI stainings. Nuclei with pyknotic characteristics could be seen in Iba-1 space within GAMM. Orthogonal views evidenced the engulfing along the z-stack (a). The 3D reconstruction at central clipping plane shows the phagosome containing pyknotic nuclei (arrowhead) surrounded by Iba-1 (b). Analysis of relative fluorescence intensity was measured along yellow broken line (c). Plot profile from c is shown in d. Nuclei (normal, 1 and pyknotic, 2) are also indicated in the profile plot (d). Analogous examples of phagocytosis are shown in B and C (in plot profiles 1 is for regular nucleus and 2 for pyknotic nuclei). (D) Phagocytosis of GFAPþ material at the human PPs. Confocal overview of the HA stained for Iba-1, GFAP and DAPI (a). Detail of a phagocytic event from the P region of the HA and orthogonal views at the indicated planes (b). The 3D visualization at central clipping plane showing GFAPþ material at the phagosome (c). Lateral view of the phagocytic event from clipping plane shown in c with white broken line (d). (E) Example of a HA of a PP indicating the P, I and D areas. Pyknotic nuclei can be detected phagocytosed (1) or isolated (2), especially at the P and I regions. (F) Quantification of phagocytosed nuclei in P, I and D. (G) Quantification of isolated pyknotic nuclei in P, I and D. (H) Estimation of the frequency of phagocytic events in P, I and D. GAMMs contribute to GBM invading niche BRAIN COMMUNICATIONS 2020: Page 19 of 20 | 19 Video 8 Rotation of a 3D tissue transparency to show the Video 10 The 3D reconstruction of a representative HA of anisotropy of the PP shared by GAMMs (Iba-1), and a human PP. Iso-surfaces of Iba-1, GFAP and DAPI revealed the malignant glioma cells and remnant astrocytes (GFAP) haptotactic interactions of GAMMs with fibrils. across the HA. Video 11 Visualization of a phagocytic event at the P Video 9 Confocal Z-scan at the HA area of a PP including region of a human PP. The 3D reconstruction of a confocal P, I and D regions. Interactions of GAMMs (Iba-1) and malignant image shows an Iba-1þ cell engulfing DAPIþ pyknotic material. cells and remaining reactive astrocytes (GFAP) can be especially seen at the I region. Phagocytic events can be observed at the P area. Barcelona, very especially to the excellent technicians at the Microscopy Core, Núria Barba and Saioa Mendizuri and the Cell Culture Unit, Cristina Gutierrez and Antonio Cambero. Supplementary material Supplementary material is available at Brain Communications online. Funding This project was supported by grants from the Spanish Ministry of Economy and Competitiveness, and the Acknowledgements European Regional Development Fund (Fondo Europeo de Desarrollo Regional, FEDER; Reference Grants: RYC- This work is dedicated to the loving memory of Carlos 2010-06729, SAF2013-45178-P and SAF2015-64123-P), Barcia Mari~ no (1941–2014), Head of the Department of Generalitat de Catalunya (Reference Grant: 2014 SGR-984), Neurosurgery, Valencia General Hospital. We would like to Spanish Ministry of Science, Innovation and Universities thank all the supporting institutions, which are listed in the (Reference Grant: PGC2018-096003-B-I00) and by the funding section. We also acknowledge the contribution of Asociación Espa~ nola Contra el Cancer (AECC). the National Cancer Institute for contributing with Research Material (GL261 Cells, NCI MTA#1-4638-14). We would also like to thank all the personnel from the Administration and Technical Laboratories of the Institut de Neurociències, Competing interests for the help provided at the Universitat Autònoma de The authors report no competing interests. 20 | BRAIN COMMUNICATIONS 2020: Page 20 of 20 E. Saavedra-López et al. Leblond MM, Gerault AN, Corroyer-Dulmont A, MacKenzie ET, Petit References E, Bernaudin M, et al. Hypoxia induces macrophage polarization and re-education toward an M2 phenotype in U87 and U251 glio- Ayuso JM, Monge R, Martinez-Gonzalez A, Virumbrales-Munoz M, blastoma models. Oncoimmunology 2016; 5: e1056442. Llamazares GA, Berganzo J, et al. Glioblastoma on a microfluidic chip: Nemere I, Kupfer A, Singer SJ. Reorientation of the Golgi apparatus generating pseudopalisades and enhancing aggressiveness through and the microtubule-organizing center inside macrophages subjected blood vessel obstruction events. Neuro-Oncology 2017; 19: 503–13. to a chemotactic gradient. Cell Motil 1985; 5: 17–29. Brat DJ, Castellano-Sanchez AA, Hunter SB, Pecot M, Cohen C, Oudin MJ, Weaver VM. Physical and chemical gradients in the tumor Hammond EH, et al. Pseudopalisades in glioblastoma are hypoxic, microenvironment regulate tumor cell invasion, migration, and me- express extracellular matrix proteases, and are formed by an actively tastasis. Cold Spring Harb Symp Quant Biol 2016; 81: 189–205. migrating cell population. Cancer Res 2004; 64: 920–7. Poon CC, Sarkar S, Yong VW, Kelly J. Glioblastoma-associated micro- Brat DJ, Van Meir EG. Vaso-occlusive and prothrombotic mechanisms glia and macrophages: targets for therapies to improve prognosis. associated with tumor hypoxia, necrosis, and accelerated growth in Brain 2017; 140: 1548–60. glioblastoma. Lab Invest 2004; 84: 397–405. Raggi F, Pelassa S, Pierobon D, Penco F, Gattorno M, Novelli F, et al. Casazza A, Laoui D, Wenes M, Rizzolio S, Bassani N, Mambretti M, Regulation of human macrophage M1-M2 polarization balance by et al. Impeding macrophage entry into hypoxic tumor areas by hypoxia and the triggering receptor expressed on myeloid cells-1. Sema3A/Nrp1 signaling blockade inhibits angiogenesis and restores Front Immunol 2017; 8: 1097. antitumor immunity. Cancer Cell 2013; 24: 695–709. Reiter I, Krammer B, Schwamberger G. Cutting edge: differential effect Chen X, Zhang L, Zhang IY, Liang J, Wang H, Ouyang M, et al. of apoptotic versus necrotic tumor cells on macrophage antitumor RAGE expression in tumor-associated macrophages promotes angio- activities. J Immunol 1999; 163: 1730–2. genesis in glioma. Cancer Res 2014; 74: 7285–97. Riching KM, Cox BL, Salick MR, Pehlke C, Riching AS, Ponik SM, Cullen SP, Henry CM, Kearney CJ, Logue SE, Feoktistova M, Tynan et al. 3D collagen alignment limits protrusions to enhance breast GA, et al. Fas/CD95-induced chemokines can serve as “find-me” sig- cancer cell persistence. Biophys J 2014; 107: 2546–58. nals for apoptotic cells. Mol Cell 2013; 49: 1034–48. Rivera LB, Bergers G. Location, location, location: macrophage posi- Diaz LR, Saavedra-Lopez E, Romarate L, Mitxitorena I, Casanova PV, tioning within tumors determines pro- or antitumor activity. Cancer Cribaro GP, et al. Imbalance of immunological synapse-kinapse Cell 2013; 24: 687–9. states reflects tumor escape to immunity in glioblastoma. JCI Insight Rong Y, Durden DL, Van Meir EG, Brat DJ. ‘Pseudopalisading’ necro- 2018; 3: e120757. sis in glioblastoma: a familiar morphologic feature that links vascu- Dickinson RB. A generalized transport model for biased cell migration lar pathology, hypoxia, and angiogenesis. J Neuropathol Exp in an anisotropic environment. J Math Biol 2000; 40: 97–135. Neurol 2006; 65: 529–39. Elliott MR, Ravichandran KS. The dynamics of apoptotic cell clear- ance. Dev Cell 2016; 38: 147–60. Schaer DJ, Alayash AI, Buehler PW. Gating the radical hemoglobin to Etienne-Manneville S. Cdc42–the centre of polarity. J Cell Sci 2004; macrophages: the anti-inflammatory role of CD163, a scavenger re- 117: 1291–300. ceptor. Antioxid Redox Signal 2007; 9: 991–9. Guo L, Akahori H, Harari E, Smith SL, Polavarapu R, Karmali V, Wang G, Wang L, Sun XG, Tang J. Haematoma scavenging in intra- et al. CD163þ macrophages promote angiogenesis and vascular per- cerebral haemorrhage: from mechanisms to the clinic. J Cell Mol meability accompanied by inflammation in atherosclerosis. J Clin Med 2018; 22: 768–77. Invest 2018; 128: 1106–24. Wippold FJ 2nd, Lammle M, Anatelli F, Lennerz J, Perry A. Hambardzumyan D, Gutmann DH, Kettenmann H. The role of micro- Neuropathology for the neuroradiologist: palisades and pseudopali- glia and macrophages in glioma maintenance and progression. Nat sades. AJNR Am J Neuroradiol 2006; 27: 2037–41. Neurosci 2016; 19: 20–7. Zhou W, Ke SQ, Huang Z, Flavahan W, Fang X, Paul J, et al. Kristiansen M, Graversen JH, Jacobsen C, Sonne O, Hoffman HJ, Periostin secreted by glioblastoma stem cells recruits M2 tumour- Law SK, et al. Identification of the haemoglobin scavenger receptor. associated macrophages and promotes malignant growth. Nat Cell Nature 2001; 409: 198–201. Biol 2015; 17: 170–82.