SorLA restricts TNFα release from microglia to shape a glioma-supportive brain microenvironment

The activity of SorLA has been especially well documented in neurons, yet recent data suggest that it might also be expressed in other brain cell types in a specific pathological context. In particular, Abdelfattah et al reported high expression levels of the SORL1 gene in GAMs in human samples (Abdelfattah et al, 2022). In line with this notion, we could indeed detect SorLA in Iba1+ cells in human glioma sections, although not in all studied patient samples (Fig. 1A; Appendix Table S1). Intrigued by the fact that not all Iba1+ cells in patients specimens stained positive for SorLA, we analyzed the proportions of AIF1 (encoding for Iba1) expressing cells that also express SORL1 in several scRNA-seq datasets from GBM samples (Data Ref.: Sankowski et al, 2019; Abdelfattah et al, 2022; Chen et al, 2021; Neftel et al, 2019; Pombo Antunes et al, 2021; Wang et al, 2022b). In all 6 datasets, SORL1 was expressed in AIF1+ cells. The percentage of SORL1+ cells among AIF1+ cells was variable and ranged from 17.2% to 97.0% depending on the dataset (Appendix Table S2). Induction of Sorl1 expression in GAMs was recapitulated in a murine model of glioma (Szulzewsky et al, 2015). Twenty days after GL261 glioma cells implantation, Sorl1 expression in GAMs (CD11b+ cells) isolated from brain tumors increased more than three times compared to the control cells. At the same time, SorCS2, another VPS10P receptor detected in this dataset, did not show such an induction (Appendix Table S3). Taken together, these data point to the hypothesis that SorLA levels in microglia/macrophages might be upregulated during activation of these cells toward a tumor-supporting phenotype.

To further investigate SORL1 expression patterns in GAMs, we re-analyzed scRNA-seq data from human glioma samples (Data Ref.: Abdelfattah et al, 2022), focusing on newly diagnosed GBM samples (ndGBM). From this dataset, we selected 7 clusters that we classified as GAMs based on the expression of marker genes (AIF1, CD68, ITGAM, P2RY12, TMEM119, CX3CR1; Fig. EV1A–C; Appendix Table S4; Dataset EV1). As reported by the authors of the study before, SORL1 expression was limited to this GAMs population (Fig. EV1B; Appendix Fig. S1A). The only other cluster with noticeable expression of SORL1 was characterized by expression of interferon-inducible genes (IFI6, IFI27, IFITM1, IFITM2, IFITM3) as well as endothelial markers (EDN1, ABCG2, FLT1, PLVAP, PECAM1) (Appendix Fig. S1A; Dataset EV1).

In all subsequent steps, we focused on the population of selected glioma-associated microglia/macrophages (GAMs). These cells were pooled for further analysis to yield 23 GAMs clusters in which SORL1 expression levels were evaluated (Fig. 1B). To shed light on the potential links between SorLA levels and the functional properties of GAMs, we next investigated the marker genes of five clusters with the highest and five clusters with the lowest SORL1 levels (Fig. 1C; Dataset EV2). Among “high-SORL1” clusters, cluster 7 was characterized by the expression of immediate early genes, SPP1, a gene associated with tumor-promoting GAMs (Szulzewsky et al, 2015), and ID2 involved in pro-tumorigenic polarization of myeloid cells (Huang et al, 2017). Cluster 22 showed high expression levels of several microglia/macrophages genes including TREM2, which was linked before to the pro-tumorigenic properties of tumor-associated macrophages in various cancers (Khantakova et al, 2022). At the same time, among “low-SORL1” clusters, cluster 9 was characterized by expression of TLR genes as well as pro-inflammatory cytokines IL1A and TNF. Clusters 11 and 17 showed relatively high expression of a pro-inflammatory factor MIF and several glycolysis-related genes (PGK1, ENO1, GAPDH, LDHA, PKM). Importantly, a metabolic switch towards glycolysis is a hallmark of pro-inflammatory activation of microglia/macrophages (Lauro and Limatola, 2020).

Furthermore, to provide a more global characteristics of GAMs, we implemented Natural Language Processing (NLP) approach, which allows to identify keywords describing functionally related groups of genes. In doing so, we analyzed marker genes of all GAMs clusters (Datasets EV2 and EV3). NLP revealed that genes enriched in “high-SORL1” clusters were associated with terms as “cancer” (clusters 2, 7, 8, 22) and “angiogenesis” (cluster 12). Concurrently, genes expressed in “low-SORL1” clusters were described with words like “glycolysis” (clusters 11 and 17), “toll-like receptor” (clusters 9 and 18) or “phagocytosis” (cluster 3). Taken together, based on the results obtained thus far we hypothesized that high and low SorLA levels might be associated with pro-tumorigenic and pro-inflammatory phenotypes of microglia/macrophages, respectively.

To gain deeper insights into the functional relevance of SorLA’s presence in GAMs, we performed Monte-Carlo Feature Selection (MCFS-ID) analysis (Dramiński et al, 2008; Dramiński and Koronacki, 2018) on a single-cell level in the same dataset. MCFS-ID allows to determine features (here expression levels of a given gene) predicting the behavior of another feature (here, SORL1 expression levels categorized to low, medium, or high). Among 25 top genes from the MCFS-ID, differential levels of gene expression in the context of discretized values of SORL1 were assessed (Fig. 1D; Dataset EV4). This analysis highlighted microglia signature genes (CX3CR1, A2M, C3) and TREM2 among the best predictors of high-SORL1 expression levels in GAMs. Moreover, high expression of a gene coding for transcription factor MEF2C, known for its role in restraining microglial inflammatory response (Deczkowska et al, 2017), was also linked to high-SORL1 transcript levels. Interestingly, we also noted a similar positive association between SORL1 and GPR34, coding for a microglial receptor required to maintain microglia in the homeostatic phenotype (Schöneberg et al, 2018). Finally, we identified LGALS1, S100A10 and MIF as predictors of low SORL1 expression in GAMs. While MIF is a well-established marker of pro-inflammatory activation of microglia/macrophages, the roles of LGALS1 and S100A10 in these cells are not entirely clear. LGALS1 expression in tumor cells has been linked to immunosuppression (Chen et al, 2019), but its induction in microglia might be relevant for inflammatory responses. Thus, increased LGALS1 expression was observed in microglia after stimulation with LPS (Kiss et al, 2023) and in a subpopulation of disease-associated microglia in multiple sclerosis (Masuda et al, 2019). S100A10 has been mostly studied in other cell types; in macrophages, it seems to be involved in their recruitment upon inflammation (O’Connell et al, 2010).

To further characterize the potential functional interactions between various cell types present in the tumor microenvironment in the context of SORL1 levels in GAMs, we implemented the CellChat analysis. First, we defined several cell types in the ndGBMs (Data Ref.: Abdelfattah et al, 2022) based on the expression of marker genes (Dataset EV1 and Appendix Tables S4–7). As a result, in addition to GAMs population subdivided into low-, medium- and high-SORL1 expressing cells, we defined the populations of tumor cells (clusters 0, 7, 8, 9, 15, expressing CDK4, MT1X, ATRX, CCND2, MDM2, SOX4, CD9, CDK6, S100B), lymphocytes (clusters 5, 17, 19; GZMK, CD3E, PTPRC, CCL5, IL32, CD69, CD52), smooth muscle cells (cluster 11; ACTA2, TAGLN), endothelial cells (cluster 14; EDN1 PECAM1 ANGPT2), oligodendrocytes (cluster 12, MBP, CNP) and other cells whose classification was dubious (clusters 4, 18, 20). Cell–cell interaction analysis revealed distinct communication patterns specific for each of these populations (Figs. 2A,B and EV2A). Tumor cells appeared to be the major contributors to outgoing communication, which at the same time was almost missing for the lymphocytes that mostly exhibited incoming patterns. Of note, GAMs tended to show different communication patterns depending on their SORL1 expression levels (Figs. 2A,B and EV2A).

More detailed analysis of the ligand-receptor pairs contributing to these communication patterns revealed several interactions specific for the GAMs with high-SORL1 expression. This population received CCL5-CCR1 signals from the lymphocytes and ICAM2-(ITGAM + ITGB2) from the endothelial cells, as well as HLA-F–LILRB1 from the lymphocytes (Fig. 2C). These interactions may be associated with enhanced tissue infiltration (Schenkel et al, 2004; Pham et al, 2012) and induction of immunosuppressive GAMs phenotype (Zeller et al, 2023), respectively. In turn, high-SORL1 GAMs sent pro-tumorigenic signals to the tumor cells (Fig. 2C). In particular, noncanonical Wnt signaling (WNT5A-FZD3) might promote glioma invasion (Pukrop et al, 2010, 2006) and glutamatergic signaling ((SLC1A3 + GLS)–GRIA2) is known to stimulate tumor cells growth, proliferation, and survival (Prickett and Samuels, 2012). Other ligand-receptor pairs revealed by this analysis included TNF-TNFRSF1A sent by medium-SORL1 expressing GAMs, TNFSF12-TNFRSF12A (high-SORL1 GAMs to tumor cells), PDGFB-PDGFRA (medium- and high-SORL1 GAMs to several populations), and THY1-(ITGAM + ITGB2) received by high-SORL1 GAMs (Fig. EV2B). In conclusion, CellChat analysis further substantiated our notion that high expression of SORL1 is one of the features characterizing a distinct GAMs population whose transcriptional profiles point to their pro-tumorigenic potential.

Finally, we repeated key analyzes on an independent dataset published by Neftel et al, (Data Ref.: Neftel et al, 2019). Here, SORL1 expression was also seen predominantly in GAMs that we selected based on the expression of the marker genes as before for the Abdelfattah et al data (Dataset EV5; Fig. EV1D,F; Appendix Fig. S1B; Appendix Table S8), but it appeared more uniform and shifted towards higher levels compared to the dataset from Abdelfattah et al (Appendix Fig. S1C,D). In the next step, GAMs were further grouped into 5 clusters, for which SORL1 expression levels and the marker genes were analyzed (Appendix Fig. S1E–H; Dataset EV6). Marker genes of the two GAMs clusters with the highest SORL1 expression included CX3CR1, P2RY12, AIF1, ITGAM, TMEM119, TREM2, CCL3, and CCL4, while GAMs cluster with the lowest SORL1 levels was characterized by LGALS1, GAPDH, PGK1, ENO1, LDHA, MIF, FTH1, HMOX1, and TLR4 marker genes (Appendix Fig. S1H). Moreover, the genes that we previously linked to SORL1 expression on a single-cell level (Abdelfattah et al data, Fig. 1D) showed similar expression patterns associated with SORL1 in this additional scRNA-seq dataset (Appendix Fig. S1I). In conclusion, our key findings on SORL1 expression patterns in GAMs were recapitulated in this independent analysis.

Taken together, we propose a functional link between the activation status of microglia/macrophages and SORL1 expression levels. In particular, our results point to the scenario where high SORL1 expression occurs in tumor-supportive GAMs, while low SORL1 expression is associated with pro-inflammatory phenotypes of microglia/macrophages.

Since SORL1 expression appeared related to the functional properties of GAMs, we tested whether SorLA levels might be specifically regulated by the cues triggering diverse microglial phenotypes. To study this phenomenon, we used primary murine microglia treated with LPS or co-cultured with GL261 glioma cells, an in vitro model to mimic pro-inflammatory stimulation and the impact of glioma-secreted factors, respectively. In line with our hypothesis, Sorl1 expression increased in the presence of glioma cells, while it dramatically decreased upon LPS treatment (Fig. 3A; Appendix Fig. S2A,B).

Distinct changes in microglial Sorl1 expression seen upon activation by LPS and in the presence of glioma cells suggested that SorLA might be an active player in shaping functional properties of microglia. SorLA controls the intracellular sorting of target proteins defining plasma membrane transport and secretion properties (Schmidt et al, 2017). As cytokines release from activated microglia is crucial for their activity and response to disease (Colonna and Butovsky, 2017), we profiled cytokines released by WT and SorLA-deficient (SorLA-KO, SLKO) murine microglia upon PMA stimulation to uncover potential factors secreted in a SorLA-dependent manner. We did not observe any global alterations in cytokines secretion from SorLA-KO microglia as compared to WT cells. Several cytokines were released in similar amounts in both genotypes, including IL-9, MCP1, MIP1α, MIP2, and MIP3 (Fig. EV3). Secretion of MIP1β tended to be decreased in SorLA-KO cells. However, the most remarkable difference was seen in the secretion of the pro-inflammatory cytokine TNFα, which was released in higher amounts from SorLA-deficient microglia (Fig. 3B). This enhanced secretion of TNFα was not due to its increased expression, as mRNA levels were not changed in SorLA-KO cells (Fig. 3C; Appendix Fig. S2C). These results indicated that the alterations in TNFα release occur post-transcriptionally and might result directly from SorLA-dependent sorting mechanisms present in WTs, but absent in SorLA-KO microglia.

To further establish the relevance of our findings for human cells, we used microglia derived from induced pluripotent stem (iPS) cells (Fig. 3D,E) either wild-type or genetically deficient for SORL1 (SLKO). As expected, expression of pluripotency markers (SOX2, NANOG, OCT4) dropped during differentiation, while microglia markers (P2RY12, TREM2, AIF1, CX3CR1, ITGAM) levels increased (Fig. 3F; Appendix Figs. S2D and S3A). Immunodetection of Iba1 and P2RY12 additionally confirmed the microglial identity of generated cells (Fig. 3G). Differentiation was unaltered by SorLA deficiency, as expression levels of marker genes AIF1, ITGAM and P2RY12 were comparable for WT and SorLA-deficient human-induced microglia (iMG, Appendix Figs. S2E and S3B). As anticipated, SorLA protein was completely lost from SLKO iMG (Appendix Fig. S3C). Using this model, we confirmed that LPS stimulation drives a remarkable decrease in SORL1 expression in WT iMG (Fig. 3H; Appendix Fig. 2F). Moreover, loss of SorLA activity led to an increased TNFα release from iMG (Fig. 3I), indicating that SorLA-dependent control of TNFα secretion is conserved between the species and highlighting its potential relevance for disease pathogenesis.

Typically, SorLA exerts its functions by binding target proteins and directing their intracellular trafficking. For example, this sorting activity of SorLA was documented for protein sorting between the TGN, endosomes and lysosomes, as well as in the recycling route via the Rab11+ compartment (Schmidt et al, 2016, 2007; Caglayan et al, 2014). To corroborate the potential role of SorLA in TNFα sorting, we first tested the colocalization of the two proteins in microglial cells. Indeed, immunostaining of PMA-stimulated BV2 cells revealed partial overlap of SorLA and TNFα signals (Fig. 4A,B). Next, we examined the interaction of SorLA with TNFα in co-immunoprecipitation (co-IP) assays. Towards this end, we overexpressed SorLA and GFP-tagged TNFα (or GFP alone) in HEK293 cells and pulled down the GFP tag. SorLA was present in the immunoprecipitate containing TNFα-GFP, while it was not visible in the control GFP-IP (Fig. 4C), supporting our notion that TNFα is a SorLA ligand.

As the structure of SorLA entails several domains capable of cargo binding (Fig. 4D), we sought to identify the domain responsible for the interaction with TNFα. Using deletion mutants lacking particular SorLA domains (Fig. EV4A) in our co-IP experiments, we observed that removing EGF-type repeat and the β-propeller (ΔEGF/βP mutant) tended to weaken SorLA binding to TNFα, although these results did not reach statistical significance (Fig. EV4B,C). Of note, these experiments did not rule out additional binding sites outside the EGF-type repeat and the β-propeller for TNFα in SorLA. Using more stringent co-IP conditions (300 mM NaCl) in order to increase the specificity of our results led to loss of binding of TNFα by full-length SorLA (Appendix Fig. S4). As an alternative approach, we used myc-tagged mini-receptors composed exclusively of particular SorLA domains (Fig. EV4A; Appendix Table S9). In line with our prior observations, the most efficient co-IP with TNFα was noted for the mini-receptor encompassing the EGF-type repeat and β-propeller (EGF/βP, Fig. 4E,F). In summary, it is plausible that SorLA binds TNFα predominantly via an extracellular motif containing EGF-type repeat and β-propeller.

Finally, we elucidated the impact of SorLA on the intracellular trafficking of TNFα. We analyzed its colocalization with the markers of subcellular compartments in WT and SorLA-deficient primary microglia stimulated with PMA. We did not observe remarkable TNFα presence in lysosomes (stained with anti-Lamp1) in any of the genotypes (Appendix Fig. S5). Rather, TNFα was colocalizing with the Golgi (GM130-positive structures), Rab7+ late-endosomes, Rab11+ recycling endosomes, as well as with Vti1b, known for its important role in the TNFα secretory route (Murray et al, 2005). SorLA deficiency did not affect the presence of TNFα in the GM130 + , Lamp1+ and Rab7+ compartments (Appendix Fig. S5), but it increased the colocalization of TNFα with Vti1b and caused a concurrent loss of TNFα from Rab11+ endosomes (Fig. 4G). These results indicated that loss of SorLA shifts the trafficking of TNFα toward the secretory pathway, which could explain the increased TNFα release from SorLA-KO microglia.

SorLA emerged as a critical factor controlling the functional properties of microglia, restricting their pro-inflammatory activities. Moreover, in glioma patients, SORL1 expression levels in GAMs were related to their transcription profiles. We further speculated that the presence of SorLA might have an impact on the functional properties of GAMs and, consequently, on tumor microenvironment and glioma progression.

Thus, we asked whether SorLA presence in the host cells has an impact on glioma progression in a murine model. Glioma GL261 cells carrying luciferase and tdTomato transgenes were implanted to the striata of WT and SorLA-KO mice, and the tumor growth was followed for 21 days. Using this model, we showed that glioma growth is limited in SorLA-deficient mice (Fig. 5A,B), supporting our notion that the presence of this sorting receptor in host cells is critical for establishing a tumor-promoting microenvironment.

To evaluate the combined impact of the developing glioma and SorLA loss from host cells on the tumor microenvironment, we first assayed the levels of selected cytokines (TNFα, MIP2, CCL5, CXCL1, IL1β, IL2, IL6, and IL10) in the tissue lysates derived from tumor-bearing and tumor-free hemispheres. At 14 days post-implantation, we did not document any major induction of the cytokines (Appendix Fig. S6A). Twenty-one days after implantation, induction of all cytokines in glioma-bearing hemispheres was visible and comparable between genotypes, except for MIP2, which was decreased in the SLKO tumor samples as compared to WTs (Appendix Fig. S6B).

Next, we evaluated the properties of microglia in the glioma model in WT and SLKO mice. The microglia activation status is reflected by the changes in their morphology (Morrison et al, 2017; Franco-Bocanegra et al, 2021; Kvisten et al, 2019). In essence, the tumor-supportive phenotype is characterized by ramified morphology with longer and more branched processes. By contrast, pro-inflammatory microglia present compact morphology and shorter extensions. These features can be evaluated by Sholl analysis, which quantifies the number of processes crossing the spheres of increasing radius, centered at the cell soma. This analysis performed on the tumor-surrounding Tmem119+ cells revealed remarkable differences between WT and SorLA-KO microglia (Fig. 5C,D). In detail, SorLA-KO microglia showed a compact morphology with shorter processes, which can be attributed to its more pro-inflammatory phenotype, while a branched morphology of microglia observed in WTs can correspond to a homeostatic or tumor-supporting state. These differences between the genotypes were also reflected by an apparent global decrease in Tmem119 signal in the tumor-surrounding tissue in SorLA-KO mice (Appendix Fig. S7A,B). Of note, in the contralateral glioma-free hemisphere, we did not observe any genotype-dependent alterations in microglia morphology (Appendix Fig. S7C,D).

To further verify the hypothesis that the pro-tumorigenic activities are blunted in the SorLA-KO animals, we focused on the phosphorylation status of STAT3 in the glioma-bearing brains. STAT3 is a transcription factor, which, when phosphorylated, activates the expression of multiple genes related to pro-tumorigenic properties of GAMs (De Boeck et al, 2020; Dumas et al, 2020). It was demonstrated that inhibition of STAT3 enhances pro-inflammatory potential of these cells, which results in suppression of tumor growth in a murine model of glioma (Zhang et al, 2009). In line with our hypothesis, the levels of p-STAT3 were remarkably reduced in glioma-bearing hemispheres of SorLA-deficient mice as compared to WTs (Fig. 5E,F).

An important consequence of the inflammatory response is the influx of peripheral immune cells into the affected tissue. Locally released factors attract circulating leukocytes and promote their migration, eventually driving their infiltration into the inflamed area. In GL261 gliomas, the infiltration of galectin-3+ macrophages and CD8 + T lymphocytes into the tumor mass was similar for both WT and SorLA-KO mice (Fig. 6A–C). Also the staining for the common GAMs marker Iba1 did not reveal any genotype-dependent differences in terms of glioma infiltration (Appendix Fig. S8). However, we noted a striking genotype-dependent difference in the neutrophil influx into the glioma. Thus, infiltration of the MPO+ neutrophils was evident in the gliomas in SorLA-KO brains, while it did not occur in the WTs and in the contralateral hemispheres (Fig. 6D,E). This was not due to the overall increase in neutrophil amounts in the circulating blood of SorLA-KO mice, as the numbers of circulating neutrophils, as well as of erythrocytes, monocytes, lymphocytes, basophils, and eosinophils, were comparable in glioma-bearing WT and SorLA-KO mice (Appendix Fig. S9). We propose that this massive neutrophil infiltration is a direct consequence of a pro-inflammatory milieu promoting their migration into the brain parenchyma in SorLA-deficient mice. One of the key mechanisms facilitating neutrophils influx into the tissue involves TNFα-driven induction of ICAM-1, an adhesion molecule critical for the transendothelial migration of these cells (Peterson et al, 2006; Yang et al, 2005). Increased soluble ICAM-1 (sICAM1) levels are also a well-established hallmark of inflammation (Bui et al, 2020). As anticipated from our data, sICAM1 was elevated in the tumor-containing hemispheres derived from SorLA-KO mice as compared to the WTs (Fig. 6F). These results strongly supported our hypothesis that loss of SorLA shifts the properties of the glioma microenvironment toward pro-inflammatory.

Finally, to further elucidate the mechanisms limiting tumor growth in SorLA-KO mice, we focused on cell death mechanisms that might be activated by TNFα itself, or by the infiltrating neutrophils. TNFα can trigger apoptosis or necroptosis via its receptor TNFR1 (Webster and Vucic, 2020), while neutrophils elicit ferroptosis (Yee et al, 2020). We thus checked which of these pathway(s) are activated in glioma specifically in SorLA-KO mice. We did not observe induction of apoptosis, as the cleavage of PARP and caspase-3 was negligible and similar for both WT and SorLA-deficient mice (Fig. 7A,B). Induction of ferroptosis was not visible either (Fig. 7C). At the same time, we noted increased levels of necroptosis markers p-RIP1 and p-RIP3 in the glioma-bearing hemispheres from SorLA-KO mice, as compared to the WTs (Fig. 6D,E). In line with these observations, we further documented a time-dependent increase of p-RIP1 in cultured glioma cells treated with TNFα (Fig. 7F). These results suggested that necroptosis might contribute to the elimination of glioma cells and, in consequence, to limiting tumor growth in SorLA-KO brains.

Mice with a targeted disruption of murine SorLA (SorLA-KO) have been described (Andersen et al, 2005). We used SorLA-KO mice on an inbred C57BL/6J background. For primary cultures, SorLA-KO and C57BL/6J wild-type newborns of both sexes were used. For in vivo studies, SorLA-KO and C57BL/6J wild-type males at 10–16 weeks of age were used. Animals were kept under a 12 h/12 h light/dark cycle with free access to food and water. Animal experimentation was performed following approval by the First Ethical Committee in Warsaw (approvals no 1102/2020, 1274/2022) and in agreement with the ARRIVE guidelines.

Primary microglial cultures were prepared using whole brains of P0-P1 C57BL/6J wild-type or SorLA-KO newborns. Brains were stripped of meninges and washed three times with HBSS. Afterward, brains were enzymatically dissociated with 1% Trypsin and 0.05% DNAse. Cell culture medium composed of DMEM w/Glutamax, 10% FBS (PanBiotech), and penicillin/streptomycin was used to inhibit enzymatic digestion. Next, tissues were centrifuged (130 × g, 4 °C, 10 min), and obtained pellets were resuspended in fresh cell culture medium. Cells were plated onto flasks coated with 0.1 mg/mL PLL. After 48 h, cells were washed three times with PBS to remove debris and maintained for the next 8 days. Then, microglial cells were separated from other glial cells by 1 h shaking at 80 RPM at 37 °C, collected by centrifugation (130 × g, 4 °C, 10 min), and seeded onto six-well (6 × 10⁵ cells) or 24-well (1 × 10⁵ cells) plates, on coverslips. Forty-eight hours after plating microglia were used for PMA and LPS simulations or co-cultured with GL261 cells.

All cell lines were routinely tested for mycoplasma contamination. HEK293 and GL261 cells were cultured in DMEM with 10% FBS (Gibco), penicillin/streptomycin and with the addition of 100 μg/mL G418, a selection antibiotic, in case of GL261-tdTomato⁺Luc⁺. BV2 cells were cultured in DMEM w/Glutamax with 2% FBS (Gibco) and penicillin/streptomycin. HEK293 transfections were performed with Lipofectamine 2000 reagent, according to the manufacturer’s protocol. The cells co-expressing SorLA mutants and TNFα-GFP (or GFP) were collected 24 h later.

Human brain tissue samples used in this study were obtained from the archives of the Department of Neuropathology of the Amsterdam UMC (University of Amsterdam, the Netherlands). Appendix Table S1 summarizes the clinical characteristics of patients. All human specimens were obtained and used in accordance with the Declaration of Helsinki and the Amsterdam UMC Research Code provided by the Medical Ethics Committee of the AMC.

The human-induced pluripotent stem cell (iPSC) line HMGU001-A/BIHi043-A (WT) and the isogenic SORL1-deficient (SLKO) line BIHi268-A18 line were used in these studies. The SLKO iPSC line was generated by CRISPR/Cas9-mediated genome editing using the gRNA targeting exon 1 in the SORL1 gene (5´-CAGTAGCGTTCGCCCGAACA-3´). SorLA deficiency in a clone containing an 8 bp frameshift deletion was confirmed by western blotting. iPSC lines were cultured on Matrigel (#356324) coated six-well plates in Essential 8™ Flex Medium (Gibco #A2858501). The culture medium was changed daily, and cells were passaged in clusters every 3–4 days at a density of 80% using 0.5 mM EDTA/PBS. iPSCs were differentiated to microglia using a previously described protocol (McQuade et al, 2018). First, iPSCs were differentiated into hematopoietic progenitors (HPs) using STEMdiff™ Hematopoietic Kit. On day 1, iPSCs with a density of 70–80% were passaged with ReLeSR™ into E8 flex containing Matrigel-coated six-well plates. Clusters at the size of 100 cells were seeded at a density of 50–100 per well. At day 0, in wells with a total of 40–80 clusters, E8 flex medium was replaced with 2 ml of medium A. On day 2, 1 ml medium A was added to the well. On day 3, the medium was removed and replaced with 2 ml of medium B. At days 5, 7, and 9, 1 ml medium B was added to the well. At day 11, HPs were present in the media as well as attached to the bottom. To increase the yield of HPs in the collected media, the adherent cells were gently washed off using a 5-ml serological pipette. The cells were centrifuged at 300×g for 5 min, resuspended in microglia differentiation medium (DMEM/F12, 2× Insulin-transferrin-selenite, 2x B27, 0.5× N2, 1× Glutamax, 1x MEM Non-Essential Amino Acids, 400 μM monothioglycerol, 5 μg/ml Insulin, 100 ng/ml IL-34, 50 ng/ml TGFβ1, and 25 ng/ml M-CSF and seeded at a density of 2 × 10⁵ cells in Matrigel (#354277) coated six-well plates. At day 13, 15, 17, 19, 21 1 ml of microglia differentiation medium was added to the well. At day 23, the medium, except 1 ml, was transferred to a 15-ml falcon tube to spin down the floating cells at 300×g for 5 min. The cells were resuspended in 1 ml medium microglia differentiation medium and returned to the well. This was repeated for days 25–35. At day 35, the cells were resuspended in microglia differentiation medium plus 100 ng/ml CD200 and 100 ng/ml CX3CL1 to further mature the microglia. On day 37, 1 ml of the maturation medium was added to the well. The microglia cells were seeded for functional studies in the period between days 38–42.

For RNA analysis, primary murine microglia were plated onto glass coverslips on a six-well plate (6 × 10⁵/well). After 48 h, cells were stimulated with 100 nM PMA or 100 ng/ml LPS (Sigma #L7770) in DMEM w/Glutamax for 24 h. Then cells were washed twice with cold PBS and collected according to the RNA isolation kit manufacturer’s protocol. For cytokine levels analysis, cell medium was collected, centrifuged to remove cell debris, and frozen at −80 °C until use.

For immunostaining, BV2 cells and primary murine microglia were plated onto glass coverslips (1 × 10⁵) on a 24-well plate, and 100 nM PMA was applied 24 h or 48 h after plating, respectively. Twenty-four hours later, cells were fixed with 4% PFA in PBS.

For co-culture experiments, GL261 cells were seeded into the cell culture inserts with 0.4-µm pores (2.5 × 10⁵ cells/insert). Microglia cells were plated onto glass coverslips on a six-well plate (6 × 10⁵/well). After 24 h, GL261 cells were cultured in microglial medium, and after 24 h of medium conditioning over GL261 cells, medium and inserts were transferred to microglia-containing wells. After 24 h of co-culture, microglial RNA was isolated.

iPSC-derived microglia were seeded at a density of 5 × 10⁴ cells per well in microglia differentiation medium in 96-well plates and stimulated with 1.0 µg/ml LPS (Sigma, #L4391) in microglia differentiation medium. After 24 h, the medium was collected for further analyses and the cells were harvested for RNA isolation.

For TNFα treatment of cultured glioma cells, GL261 cells were plated on a six-well plate (2.5 × 10⁵ cells/well) one day prior to the stimulation. Cells were washed once with PBS, and next they were cultured for 0.5, 1, 6, or 24 h in a serum-free medium supplemented with 20 ng/ml TNFα. Cells cultured in a serum-free medium served as a control. Cells were collected at the indicated timepoints and subjected to lysis.

Mouse brains were dissected from tumor-bearing mice after transcardial perfusion with PBS and 4% paraformaldehyde PFA in PBS. After post-fixation (PFA, overnight) and cryopreservation in 30% sucrose/PBS, brains were cut in 40-μm free-floating coronal sections using a cryostat. Alternatively, the brains were dissected from non-perfused animals and slowly frozen on isopentane/dry ice. Next, 12-μm coronal sections were cut using a cryostat, immediately mounted on glass slides, and kept at –20 °C until further use. Prior to staining, these sections were thawed at room temperature for 30 min and fixed in 4% PFA/PBS for 30 min. The sections were blocked in 1% horse serum in PBS and incubated with primary followed by secondary antibodies diluted in PBS, supplemented with 1% BSA, 1% NDS, and 0.5% Triton-X. Specimens were washed with PBS, counterstained with DAPI and mounted with DAKO fluorescence mounting medium.

Cells grown on glass coverslips were fixed with 4% PFA/PBS. Next, the cells were washed with PBS, blocked for 1 h in PBS supplemented with 5% NDS and 0.3% Triton-X, and incubated with primary antibodies followed by secondary antibodies diluted in PBS with 1% BSA and 0.3% Triton-X. After washing with PBS and counterstaining with DAPI, the coverslips were mounted with DAKO fluorescence mounting medium.

Human brain tissue was fixed in 10% buffered formalin, embedded in paraffin, sectioned at 5 µm, and mounted on pre-coated glass slides. Sections were deparaffinized in xylene, and ethanol (100%, 95%, 70%). Antigen retrieval was performed using a pressure cooker in 0.01 M sodium citrate buffer (pH 6.0) at 121 °C for 10 min. Slides were washed with PBS (0.1 M, pH 7.4) and incubated overnight with primary antibodies against SorLA (EMD Millipore, 1:150) and Iba1 (1:200) in antibody diluent at 4 °C. The next day, sections were washed with PBS and incubated with Alexa Fluor 568 goat anti-rabbit (1:200) or Alexa Fluor 488 donkey anti-mouse (1:200) antibody plus Hoechst 33258 (1:1000) in antibody diluent for 2 h at room temperature, washed with PBS and mounted with Vectashield.

Total RNA was extracted from cell lysates and purified with RNeasy Mini Kit or Direct-zolTM RNA MiniPrep Kit, according to the manufacturer’s protocol. Reversely transcribed cDNA was synthesized using High-Capacity RNA-to-cDNA™ Kit. Gene expression was evaluated by quantitative real time-PCR using TaqMan™ Fast Universal PCR Master Mix and the following Taqman Gene Expression Assays: Sorl1, TNFα, β2M, Hprt1, SOX2, P2RY12, TREM2, AIF1, GAPDH, TBP, NANOG, OCT4, CX3CR1, ITGAM. Fold change in gene expression was calculated using the cycle threshold (CT) comparative method (2ddCT) normalizing to CT values for housekeeping genes (in case of primary murine microglia: β2M for LPS treatment, Hprt1 for co-culture with glioma cells and PMA stimulations; in case of human iMG: GAPDH for differentiation experiments and β2M for LPS treatment; see Appendix Fig. S2).

The following enzyme-linked immunosorbent assay (ELISA) kits were used to determine cytokine levels in microglial media samples and brain tissue lysates of WT and SorLA-KO animals 14 and 21 days after glioma implantation: murine U-Plex Biomarker Group 1 assay, human TNFα U-PLEX assay and ICAM-1/CD54 Quantikine ELISA Kit. All assays were performed according to the manufacturer’s protocol.

pEGFPC2-BIO plasmid used for GFP overexpression was a kind gift from Jacek Jaworski (Swiech et al, 2011). GFP-TNFα was a gift from Jennifer Stow (Manderson et al, 2007). The SorLA-ΔCR construct (deletion of Cys1078 to Glu1552) has previously been described (Mehmedbasic et al, 2015). In this study the following constructs were generated: SorLA-ΔVPS10P construct (deletion of Met1 to Pro753) in the pSecTag2A expression vector; SorLA-ΔEGF/β-propeller construct (deletion of Leu754 to Glu1074) in the pcDNA3.1 expression vector; SorLA-ΔFN3 construct (deletion of Val1555-Ala2123) in the pcDNA3.1 expression vector. Human SorLA mini plasmids were generated by amplifying the specific SorLA domain with overlapping primers using PCR technology. The fragments were then ligated into pcDNA3.1 zeo+ vector (Invitrogen). The SORL1 stop codon was replaced with a myc-coding sequence. All mini-receptors had the original SorLA signal and propeptide sequence and furin cleavage site at the N-terminus, and the original SorLA transmembrane and cytoplasmic domain at the C-terminus. The information on SorLA domains covered by the constructs is provided in Appendix Table S9.

HEK293 cells overexpressing SorLA mutants (depleted from one of the domain or SorLA mini plasmids) and TNFα-GFP (or GFP) were scraped in lysis buffer containing 10 mM Tris pH 7.5, 150 mM NaCl, 1 mM CaCl₂, 1 mM MgCl₂, and 0.5% NP40 substitute with proteases and phosphatases inhibitors. The lysate was kept on ice for 30 min and centrifuged (16,000 × g, 4 °C, 15 min). Lysis buffer w/o NP40 substitute was added to the cleared lysate to decrease concentration of the detergent in samples to 0.2%. Protein level was measured by Pierce™ BCA Protein Assay Kit. Part of the resultant cleared lysate was kept for analyses, while the rest (500–600 µg) was used for 30 min incubation with GFP-Trap Magnetic Particles® (with rotation, in 4 °C) to pull down the GFP tag. Lysis buffer with reduced NP40 substitute content (0.2%) was used for three subsequent washing of beads. Proteins bound to the beads were released by boiling in 2× Laemmli Buffer for 10 min and further examined by western blot.

Implantations were performed as described (Kaminska et al, 2024). In brief, mice were kept under constant anesthesia during the whole procedure using 2% Isoflurane in oxygen. After subcutaneous administration of butorphanol (2 mg/kg body weight) and meloxicam (2 mg/kg body weight), as well as bupivacaine local application (8 mg/kg body weight), skin on the head was incised. The hole in the skull was drilled at the following coordinates: −1 mm anterior-posterior (AP) and 2 mm medial–lateral (ML) from bregma. In total, 8 × 10⁵ of GL261-tdTomato+Luc+ cells in 1 μl of DMEM were stereotactically injected into the right striatum at the rate of 0.25 μl/min, to a depth of 3 mm according to the brain surface. Withdrawing of the syringe was performed at the rate 1 mm/min to avoid backward outflow of the cell suspension. The skin incision was closed using sutures. After surgery, animals were carefully monitored until they fully recovered from anesthesia. For the following 3 days, mice received meloxicam (2 mg/kg body weight) once a day. Mice weight and well-being were controlled every 2–3 days until the end of the experiment.

Mice implanted with GL261-tdTomato+Luc+ were injected intraperitoneally with d-Luciferin sodium salt (150 mg/kg body weight) in PBS. After 8 min, animals were anesthetized with 2% Isoflurane in oxygen and placed immediately in X-treme Imaging System under 2% isoflurane/oxygen supply. Ten minutes after d-luciferin administration, bioluminescence emission was determined for the total time of 2 min. X-ray images were acquired following bioluminescence imaging. Tumors were visualized 7, 14, and 21 days post-implantation. Bruker Molecular Imaging (MI) Software was used for signal quantification.

Prior to perfusions, samples of venous blood were collected immediately after incision of the right atrium. EDTA tubes were used for material collection. Hematological analyses were performed by Vetlab^® company, Warsaw, Poland.

In all, 14 or 21 days after glioma implantation brains were isolated, hemispheres were dissected and homogenized in 20 mM Tris pH 7.5 150 mM NaCl 2 mM EDTA 2 mM MgCl₂, and 10% glycerol, supplemented with protease and phosphatase inhibitors. Lysates were incubated on ice for 20 min and centrifuged (1000 × g, 4 °C, 10 min). Triton-X and NP40 substitute were added to the cleared lysate at final concentration of 0.5% and the samples were incubated for 1 h with rotation in 4 °C. Lysates were centrifuged (16,000 × g, 4 °C, 15 min) and protein concentration in the supernatants was measured using Pierce™ BCA Protein Assay Kit.

Twenty-one days post glioma implantation brains were isolated, tumor-bearing hemispheres were dissected and homogenized in 20 mM Tris-HCl pH 7.5, 2 mM MgCl₂, and 0.25 M sucrose, supplemented with protease and phosphatase inhibitors. Lysates were incubated on ice for 20 min and centrifuged (1000 × g, 4 °C, 10 min). Part of the resultant cleared lysate was kept for analyses, while the rest was further centrifuged (100,000 × g, 4 °C, 1 h) to obtain the soluble protein fraction. Protein concentration was measured using Pierce™ BCA Protein Assay Kit and samples were subjected to sICAM1 ELISA.

Cell and tissue lysates were analyzed by western blot using the following antibodies: anti-myc-tag (1:1000), anti-SorLA (BD Transduction Laboratories, 1:1000 and C-term homemade 1:1000), anti-GFP (1:250), anti-p-RIP1 (1:1000), anti-RIP1 (1:1000), anti-p-RIP3 (1:1000), anti-RIP3 (1:1000), anti-caspase-3 (1:1000), anti-PARP (1:1000), anti-TRFR (1:1000), anti-GPX4 (1:1000), anti-p-STAT3 (1:2000), anti-STAT3 (1:2000), anti-GAPDH (1:25,000). After incubation with secondary antibodies conjugated to HRP (1:5000), the chemiluminescent signal was registered with the ChemiDoc Imaging System (Bio-Rad).

Following antibodies were used: anti-SorLA (1:150, EMD Millipore and 1:100 homemade antibody produced in goat), anti-Iba1 (1:200, WAKO and 1:500, Abcam), anti-P2RY12 (1:100), anti-TNFα (1:200), anti-Vti1b (1:100), anti-Rab11 (1:200), anti-GM130 (1:1000), anti-Lamp1 (1:500), anti-Rab7 (1:100), anti-Tmem119 (1:500), anti-MPO (1:40), anti-galectin-3 (1:500), anti-CD8 alpha (1:500).

For the analysis of microglial cell morphology, Z-stack images of WT and SorLA-KO murine brains after glioma implantation stained for Tmem119 were used. Each cell was selected manually using ImageJ software, and all stacks comprising it were duplicated and saved as a new file for further analyses. 3D reconstruction of microglial branches was performed in Imaris 9.1.2. Software. Imaris Filament Tracer with spot detection mode was used to determine starting and ending points. Automatically added ending points, which did not cover the cell surface were removed manually. Reconstructed cell branches were subjected to Sholl analysis.

Western blot signals were quantified using the Image Studio Lite software or Image Lab Software 6.1 (Bio-Rad). Microscopy images were quantified using Fiji software. Signal intensities were measured on single-channel images as a mean gray value for the field of view. Colocalization analysis was performed with the Colocalization Threshold plugin with manual selection of the region of interest, containing the cell cytoplasm. Single z-plane images were used for calculating thresholded Mander’s coefficient (tM).

For in vivo experiments, an indicated number n is the number of mice per group used in an experiment. For primary cell culture experiments, an indicated number n is the number of independent glial preparations (biological replicates) used for ELISA or qRT-PCR analysis. In case of colocalization studies in primary microglia, n is the number of individual cells quantified in a given experiment, and every experiment was replicated at least three times on independent microglial cultures. Statistical analyses were performed using GraphPad Prism software. Where applicable, outlier analysis was performed using ROUT test. Data variance was tested with GraphPad Prism and was similar for the compared groups. The details of statistical analysis are specified in the figure legends. No blinding was performed.

Initially, scRNA-seq pipeline was established to run analysis on Abdelfattach et al, dataset (Abdelfattah et al, 2022), but to verify obtained results, the pipeline was then adjusted to other scRNA/snRNA-seq datasets (see below). From Abdelfattah et al, only newly diagnosed glioblastoma (ndGBM) samples were selected, consisting of 11 patients (6 males and 5 females), some of whom had multiple samples taken (Abdelfattah et al, 2022). Downstream analysis of the raw count matrices obtained from the GEO series GSE182109 was performed using the R package scTools (1.0) (https://github.com/MateuszJakiel/scTools) with R version 4.2.0. A list of samples was created with scTools function scCreateRawSamples(), from which cells expressing at least 500 genes and genes expressed in at least 5 cells were selected. Then, using the scIntroQC() function the mitochondrial, ribosomal and red blood cell gene percentages were calculated. Only cells with mitochondrial read percentage <15%, ribosomal read percentage > 5% and red blood cell read percentage <0.5% were kept. Cells with gene counts that were less than 3000 were kept. These filtered samples were integrated using the scIntegrate() function. First, the SCTransform (Hafemeister and Satija, 2019) algorithm was applied on all samples. The most informative 3000 genes were selected to perform the Principal Component Analysis (PCA) and the first 30 Principal Components were used for cell clustering and Uniform Manifold Approximation and Projection algorithm (UMAP) applied (McInnes et al, 2020). Markers of obtained clusters were found with scMarkers() that utilizes Seurat’s (Hao et al, 2021) FindAllMarkers() function, its criteria were: gene presence in at least 50% of cells, only positive fold change (FC) and the logFC threshold equal to 0.25. Expert biological analysis was used to manually annotate glioma-associated microglia/macrophages cells (GAMs) that were used for further analysis. The integrated counts were extracted from the Seurat Object and were used to perform further calculations. To analyze SORL1 expression patterns in two independent datasets (ndGBMs from Abdelfattah et al, primary GBM form Neftel et al,) as well as to prepare normalized input matrix to CellChat tool, we used SCTransform method for gene expression normalization and variance stabilization (Hafemeister and Satija, 2019).

Natural Language Processing (NLP) methods were applied to cluster GAMs marker genes and retrieve sets of keywords to describe each cluster. Marker genes descriptions, gathered from various molecular databases provided by BiomaRt (Durinck et al, 2009) such as: NCBI, Gene Ontology, KEGG, Reactome, WikiPathways and Biocarta, were used. To perform this operation each gene was represented as the text document (combined from all available descriptions), and later, to calculate TF-IDF (term frequency–inverse document frequency), as a bag of words/terms. When each gene is represented by a TF-IDF vector it is possible to calculate cosine similarity between the documents and a hierarchical clustering model can be built based on these similarities. Finally, discovering a set of unique keywords that characterize each cluster provides a good functional, biological overview of input genes and groups that should be studied more closely.

The Spearman correlation of SORL1 expression with other selected genes was calculated using cor.test(); P values were adjusted using the FDR method with the threshold equal to 0.05. To detect more complex nonlinear relationships between SORL1 expression and other genes, the MCFS-ID algorithm from the rmcfs R package (Dramiński and Koronacki, 2018; Dramiński et al, 2008) was applied. SORL1 gene expression values were discretized by dividing the values of expression into three ranges: {(−3.1374856; −0.3428585], (−0.3428585; 0.8406330], (0.8406330, +inf)} i.e., “low”, “medium”, and “high” levels. This discretized SORL1 variable was used as the response in the decision table and all other gene counts were used as explanatory features resulting in 38,086 cells and 3000 genes. Next, 75% of cells from the input decision table were sampled to establish a training set, leaving the remaining cells for the validation set. The significant features set returned by MCFS-ID was obtained using the permutation method and verified its predictive quality on the validation set using the following set of classifiers: decision tree, logistic regression, random forest, Naive Bayes and Support Vector Machines. Finally, to show differential levels of gene expression, in the context of discretized values of SORL1, selected top genes from the MCFS-ID returned ranking were visualized on the heatmap using pheatmap R package (Kolde, 2012).

To perform inference, analysis, and visualization of cell–cell communication from single-cell and spatially resolved transcriptomics the R package CellChat (Jin et al, 2023) was used on 21 clusters from ndGBMs (Abdelfattah et al, 2022). These clusters were grouped into seven groups: GAMs [1, 2, 3, 6, 10, 13, 16], lymphocytes [5, 17, 19], tumor cells [0, 7, 8, 9, 15], smooth muscle cells [11], endothelial cells [14], oligodendrocytes [12], and other cells [4, 18, 20]. Moreover, the GAMs group was split into three subgroups based on the corresponding SORL1 gene expression values: GAMs_low [0; 0.5], GAMs_med (0.5; 1.5], GAMs_hi (1.5, 3] and these bins were based on the result of the SORL1 histogram (Appendix Fig. S1C).

Other datasets analyzed in this study were GSE163120 (Pombo Antunes et al, 2021), GSE141383 (Chen et al, 2021), GSE131928 (Neftel et al, 2019), GSE135437 (Sankowski et al, 2019), and GSE174554 (Wang et al, 2022b). Data processing and analysis were performed according to the pipeline described above, where relevant.