Abstract

   Gliomas, the most frequent type of primary tumor of the central nervous
   system in adults, results in significant morbidity and mortality.
   Despite the development of novel, complex, multidisciplinary, and
   targeted therapies, glioma therapy has not progressed much over the
   last decades. Therefore, there is an urgent need to develop novel
   patient-adjusted immunotherapies that actively stimulate antitumor T
   cells, generate long-term memory, and result in significant clinical
   benefits. This work aimed to investigate the efficacy and molecular
   mechanism of dendritic cell (DC) vaccines loaded with glioma cells
   undergoing immunogenic cell death (ICD) induced by photosens-based
   photodynamic therapy (PS-PDT) and to identify reliable prognostic gene
   signatures for predicting the overall survival of patients. Analysis of
   the transcriptional program of the ICD-based DC vaccine led to the
   identification of robust induction of Th17 signature when used as a
   vaccine. These DCs demonstrate retinoic acid receptor-related orphan
   receptor-γt dependent efficacy in an orthotopic mouse model. Moreover,
   comparative analysis of the transcriptome program of the ICD-based DC
   vaccine with transcriptome data from the TCGA-LGG dataset identified a
   four-gene signature (CFH, GALNT3, SMC4, VAV3) associated with overall
   survival of glioma patients. This model was validated on overall
   survival of CGGA-LGG, TCGA-GBM, and CGGA-GBM datasets to determine
   whether it has a similar prognostic value. To that end, the sensitivity
   and specificity of the prognostic model for predicting overall survival
   were evaluated by calculating the area under the curve of the
   time-dependent receiver operating characteristic curve. The values of
   area under the curve for TCGA-LGG, CGGA-LGG, TCGA-GBM, and CGGA-GBM for
   predicting five-year survival rates were, respectively, 0.75, 0.73,
   0.9, and 0.69. These data open attractive prospects for improving
   glioma therapy by employing ICD and PS-PDT-based DC vaccines to induce
   Th17 immunity and to use this prognostic model to predict the overall
   survival of glioma patients.

   Subject terms: CNS cancer, Preclinical research, Cancer immunotherapy,
   Cell death and immune response, Prognostic markers

Introduction

   Gliomas, the most frequent intrinsic type of primary tumors of the
   central nervous system (CNS) in adults, are associated with significant
   morbidity and mortality [[56]1]. According to the newest World Health
   Organization classification of tumors of the CNS [[57]2], gliomas,
   glioneuronal tumors, and neuronal tumors are divided into six different
   families. Among these are adult-type diffuse gliomas (i.e., most adult
   patients with primary brain tumors, e.g., glioblastoma (GBM), IDH-
   wildtype), pediatric-type diffuse low-grade gliomas (with favorable
   prognoses), and pediatric-type diffuse high-grade gliomas (with poor
   prognoses) [[58]3]. The pediatric and adult types of gliomas are
   distinctively different biologically and genetically. Of note,
   pediatric-type diffuse gliomas have been subdivided into low-grade
   gliomas (LGG) and high-grade gliomas (HGG) [[59]4]. GBM is classified
   as a grade 4 malignancy; it is the most aggressive type of cancer of
   the central nervous system and has a poorer prognosis [[60]3, [61]5,
   [62]6]. Despite the development of novel, complex, multidisciplinary,
   targeted therapies, such as focal radiotherapy and adjuvant
   chemotherapeutics in combination with surgical resection, glioblastoma
   therapy has not progressed much over the last decades [[63]7]. The
   median survival of patients diagnosed with glioblastoma is 12–15
   months, with a five-year survival rate of 5% [[64]8, [65]9]. Therefore,
   there is an urgent need to develop novel patient-adjusted anticancer
   immunotherapies that actively stimulate antitumor T cells, generate
   long-term memory, and result in significant clinical benefits.

   Several recent, novel, therapeutic approaches have emerged that rely on
   vaccination to activate the patient’s own immune system and to induce a
   potent and long-lasting immune response against cancer antigens.
   Dendritic cells (DCs) are key to initiating and directing immune
   responses [[66]10, [67]11], and one of these approaches involves the
   use of DCs loaded with antigenic material derived from or based on the
   autologous tumor. One such approach is based on the identification of
   neo-antigens, but it has low efficacy due to the high antigenic
   heterogeneity of glioma (e.g., glioblastoma multiforme) [[68]12].
   Moreover, this approach is complex, labor-intensive, and costly. In
   contrast, the preparation of cancer cell lysate from the glioma tissue
   of a patient is less complex and the lysate includes neo-antigens as
   well as non-mutated tumor antigens, which can result in a broader
   immune response. However, though the immunogenicity of the lysate
   loaded in the DCs is important [[69]13–[70]15], whole glioma cells are
   usually killed by freeze-thawing (F/T) [[71]16, [72]17], which induces
   an accidental and unregulated form of necrotic cell death of low
   immunogenicity [[73]18–[74]20]. One way to increase the immunogenicity
   of the lysate is to kill the glioma cells by a method that induces
   immunogenic cell death (ICD) [[75]21, [76]22].

   ICD has recently been shown to be a prerequisite for the activation of
   the patient’s immune system. Thus, induction of ICD provides two
   benefits, effectively killing cancer cells and activating an immune
   response specific for the cancer cells. ICD is characterized by the
   release or surface exposure of damage-associated molecular patterns
   (DAMPs), which function as adjuvants to activate strong anticancer
   immunity [[77]23–[78]25]. Lately, photodynamic therapy (PDT) has been
   added to the list of therapeutic strategies that can induce typical
   features of ICD [[79]14, [80]26, [81]27]. PDT is a two-stage procedure.
   The cancer cells are first loaded with a specific drug
   (photosensitizer, PS), which is then activated by light of a specific
   wavelength corresponding to the absorption spectrum of the
   photosensitizer. This results in the generation of singlet oxygen
   (^1O[2]) and other toxic reactive oxygen species, which are components
   of ICD-inducing signaling in cancer cells but are not the only
   components [[82]26, [83]28, [84]29]. We recently demonstrated that
   clinically approved photosensitizers (photosens [PS] and
   phthalocyanines complexed with aluminum) could be used to efficiently
   trigger ICD in several cancer cell types, including glioma cells
   [[85]20]. Nevertheless, though several methods have been developed to
   induce ICD in glioma, an effective treatment strategy has not been
   developed yet.

   Here, with the aim of increasing the efficacy of glioma therapy, we
   used several subcutaneous and orthotopic mouse models to investigate
   the potential of vaccines based on glioma cells undergoing ICD
   triggered by PS-PDT. RNA-seq analysis showed that glioma cells
   undergoing ICD after PS-PDT induced a typical Th17 signature in the DC
   vaccines, which were highly effective in protecting mice against
   gliomas. Moreover, we show that inhibition of retinoic acid
   receptor-related orphan receptor-γt (RORγt), a regulator of Th17
   responses, significantly decreased the effects of the DC vaccines and
   shortened mouse survival because it reshaped the tumor microenvironment
   by depleting IL17 in the tumor. Comparison of the transcriptome program
   of the DC vaccines loaded with GL261 cells undergoing ICD after PS-PDT
   with the transcriptome data from The Cancer Genome Atlas (TCGA-LGG)
   dataset identified a four-gene signature (CFH, GALNT3, SMC4, and VAV3)
   associated with overall survival of glioma patients. These prognostic
   four gene signatures for predicting patients’ overall survival were
   validated on different cohorts of gliomas patients, including the
   datasets of the Chinese Glioma Genome Atlas (CGGA)-LGG, TCGA-GBM, and
   CGGA-GBM. Our results demonstrate the novel role of Th17 responses in
   the protection generated by DC vaccines based on ICD induced by PS-PDT
   and open promising avenues for the use of the prognostic model to
   predict the overall survival of glioma patients.

Results

Vaccination with glioma GL261 cells undergoing ICD pulsed with PS-PDT is
protective in the subcutaneous prophylactic vaccination mouse model

   We first tested the ability of dying GL261 glioma cells to activate the
   adaptive immune system by using the gold-standard prophylactic tumor
   vaccination model in immunocompetent C57BL/6J mice [[86]22]. We adapted
   the model and subcutaneously vaccinated mice twice with 5 × 10^5 glioma
   cells with a one-week interval and then challenged them with 1 × 10^5
   viable glioma cells (Fig. [87]1A). Tumor growth and appearance were
   monitored to estimate the success of priming the adaptive immune
   system. As a positive control, we included mitoxantrone (MTX), a
   well-known ICD inducer [[88]19, [89]30] that reduces the risk of death
   in patients with recurrent GBM [[90]31, [91]32]. For the negative
   control (non-ICD), we injected mice either with PBS or with 5 × 10^5
   F/T (accidentally necrotic) mouse glioma GL261 cells. Incubation of
   glioma GL261 cells with 1.4 µM of the PS and subsequent irradiation
   with a light dose of 20 J/cm^2 induced a mixed type of regulated cell
   death with both apoptotic and ferroptotic features (Suppl. Fig. [92]1A,
   B), confirming our previously published findings [[93]20]. The mice
   immunized twice with glioma GL261 cells treated with PS-PDT
   (GL261_PS-PDT) showed signs of robust activation of the adaptive immune
   system, better survival (Fig. [94]1B), and protection against tumor
   growth (Fig. [95]1C) resembling that in mice vaccinated with glioma
   GL261 cells treated with MTX (GL261_MTX; positive control).
   Importantly, mice vaccinated with GL261_PS-PDT developed no measurable
   tumors and all mice survived, indicating that GL261_PS-PDT is strongly
   immunogenic. Most of the mice immunized with PBS showed extensive tumor
   growth at the challenge site (Fig. [96]1B, C). Notably, the mice
   vaccinated twice with the same number of F/T GL261 cells developed
   significantly larger tumors (Fig. [97]1C). These data indicate that
   glioma GL261 cells undergoing accidental necrosis after F/T have weak
   immunogenicity and that even priming and boosting the mice with such
   cells does not provide effective protection against challenge with
   viable GL261 cells. These data are in agreement with previously
   published findings using other types of F/T cancer cells for
   vaccination and indicate that accidentally necrotic cells are less
   immunogenic [[98]18–[99]20]. Importantly, tumor growth at the challenge
   site of the unvaccinated (PBS) mice and those vaccinated with F/T
   glioma GL261 cells (negative control) were significantly larger than
   the tumors on the mice vaccinated with GL261_PS-PDT (Fig. [100]1C).
   However, when the mice were subcutaneously vaccinated only once with
   5 × 10^5 glioma cells and one week later challenged at another site
   with 1 × 10^5 viable GL261 glioma cells (Suppl. Fig. [101]1C),
   GL261_PS-PDT provided better protection against challenge with viable
   glioma GL261 cells than in the PBS group (i.e., better survival) though
   the difference was not statistically significant (Suppl. Fig. [102]1D).
   Interestingly, the tumors growing at the challenge site of the
   unvaccinated (PBS) mice and those vaccinated with F/T glioma GL261
   cells (negative control) were significantly larger than the tumors
   developing on the mice that were vaccinated with GL261_PS-PDT and those
   in the positive control MTX group, indicating the induction of an
   immune response in vivo (Suppl. Fig. [103]1E). Together, these data
   demonstrate that primer and booster subcutaneous vaccination of mice
   with GL261_PS-PDT activated anti-tumor immunity.

Fig. 1. Vaccination with glioma GL261 cells pulsed with PS-PDT in the
subcutaneous prophylactic vaccination mouse model.

   [104]Fig. 1
   [105]Open in a new tab

   A Prophylactic vaccination of mice was performed by injecting them in
   the left flank on days 0 and 7 with dying/dead GL261 cells treated with
   three F/T cycles, 2.0 µM MTX, or PS-PDT, or by injecting them with PBS
   (negative control). Seven days later, the mice were challenged with
   viable GL261 cells in the right flank. B Tumor appearance (% survival)
   and growth C at the challenge site in mice subjected to vaccination and
   challenge with viable GL261 cells; n = 5–6 per group. *Statistically
   significant difference from the PBS group, (p < 0.05); ^#statistically
   significant difference from the F/T group, (p < 0.05), Wilcoxon test.

Subcutaneous vaccination with glioma GL261 cells pulsed with PS-PDT protects
in the orthotopic glioma mouse model

   We examined whether GL261 glioma cells treated with PS-PDT can induce
   anti-glioma protective immunity in a prophylactic setup in the
   orthotopic glioma mouse model. Such models are widely used to test
   different novel experimental treatment strategies and to characterize
   the immunogenicity of dying glioma cells [[106]14, [107]27, [108]33].
   Immunocompetent, syngeneic C57BL/6 mice were first subcutaneously
   vaccinated twice with 5 × 10^5 glioma cells treated with PS-PDT or MTX,
   or with GL261 cells subjected to F/T (Fig. [109]2A). Eight days later,
   the mice were intracranially (intraventricularly) challenged with
   2 × 10^4 viable glioma GL261 cells and monitored for symptoms of
   neurological deficit [[110]14, [111]27, [112]33] and for survival.
   Remarkably, the mice immunized twice with GL261_PS-PDT showed signs of
   activation of anti-tumor immunity and exhibited better survival and
   protection against intracranial challenge with viable glioma GL261
   cells in comparison to mice injected with PBS (Fig. [113]2B). In the
   same model, vaccination with GL261_F/T or GL261_MTX also provided
   considerable protection against challenge with viable GL261 cells, but
   the results were not significantly different from GL261_PS-PDT
   vaccination (Fig. [114]2B). Similarly, analysis of the glioma-induced
   neurological deficit grades revealed a considerable delay in the onset
   of clinically relevant symptoms in the mice vaccinated with
   GL261_PS-PDT as compared to the PBS group (Fig. [115]2C, D). Of note, a
   single vaccination with GL261 cells pulsed with GL261_PS-PDT was not
   protective against intracranial challenge with viable glioma GL261
   cells (Suppl. Fig. [116]2A, B). These data indicate that induction of
   ICD in glioma GL261 cells by PS-PDT followed by their subcutaneous
   injection twice induces anti-tumor immunity in the orthotopic glioma
   mouse model and protects against intra-cranial challenge with viable
   GL261 cells.

Fig. 2. Subcutaneous vaccination with glioma GL261 cells pulsed with PS-PDT
protects in the orthotopic high-grade glioma (HGG) mouse model.

   [117]Fig. 2
   [118]Open in a new tab

   A Experimental setup for the prophylactic vaccination of mice injected
   with GL261 cells treated with three F/T cycles or 2.0 µM MTX, or loaded
   with 1.4 µM photosens and exposed to PDT (PS-PDT), or injected with
   PBS. The mice were vaccinated on days 0 and 7 and challenged by
   intracranial stereotactic injection of viable GL261 cells on day 14. B
   Survival of mice vaccinated and challenged with GL261 cells as
   described in (A). *p < 0.01, Mantel-Cox logarithmic test. C, D The
   percentages of mice showing neurological alterations is shown (grade 0,
   grade 1, grade 2, grade 3, grade 4). The neurological status of the
   mice was assessed every 2–4 days for up to day 31 after intracranial
   tumor inoculation. N = 5–6 per group. *p < 0.05, a statistically
   significant difference from the PBS group; Wilcoxon test.

ICD-based DC vaccines induce significant protective immunity against glioma

   We evaluated the immunogenic potential of GL261_PS-PDT and its ability
   to trigger anti-glioma protective immunity and examined whether this DC
   immunotherapy could protect mice against intracranial primary tumor
   challenge with viable GL261 cells. Indeed, GL261_PS-PDT was efficiently
   phagocytosed by murine DCs in a ratio-dependent manner (Suppl. Fig.
   [119]2C). The rates of engulfment of GL261 glioma cells treated with
   PS-PDT and those treated with MTX were similar (Suppl. Fig. [120]2C).
   These data confirm our previously published findings [[121]20]. Next,
   immunocompetent syngeneic C57BL/6 mice were vaccinated twice
   intraperitoneally with DCs loaded ex vivo with GL261_PS-PDT
   (DC-GL261_PS-PDT) (Fig. [122]3A). As a positive control, we vaccinated
   mice with DCs loaded ex vivo with MTX-treated glioma GL261 cells, and
   for the negative control, we used PBS or DCs loaded ex vivo with F/T
   glioma GL261 cells. Thereafter, all the mice were intracranially
   (intraventricularly) inoculated with 2 × 10^4 live GL261 glioma cells
   and then monitored the development of symptoms of neurological deficit
   and survival. Interestingly, the mice vaccinated with DC-GL261_PS-PDT
   before tumor challenge demonstrated a significant increase in median
   survival compared to mice injected with PBS (24 days versus 18 days,
   p < 0.03) or with DCs loaded with F/T GL261 (24 days versus 18 days,
   p < 0.02) (Fig. [123]3B). Mice vaccinated with DC-GL261_PS-PDT and
   orthotopically challenged with live GL261 cells showed significantly
   lower tumor mass than control mice (Fig. [124]3C). Consistent with
   overall survival, monitoring of the glioma-induced neurological deficit
   grades (Fig. [125]3D, E) revealed not only diminished severity of
   clinical manifestation but also later onset of symptoms in mice
   vaccinated with DC-GL261_PS-PDT (18 days versus 8 days). Remarkably, we
   also observed earlier onset of symptoms in mice vaccinated with DC
   vaccines loaded with F/T glioma GL261 cells compared to mice vaccinated
   with DC-GL261_PS-PDT (11 days versus 18 days). All these data indicate
   that the DC-GL261_PS-PDT vaccine induces anti-tumor immunity in the
   orthotopic glioma mouse model and protects mice against intra-cranial
   challenge with viable GL261 cells.

Fig. 3. ICD-based DC vaccines provide significant protective immunity against
glioma.

   [126]Fig. 3
   [127]Open in a new tab

   A Experimental setup for the prophylactic vaccination of mice with
   DC-based vaccines loaded with GL261 cells treated with PS-PDT (PS at a
   dose of 1.4 µM). As controls, we used DC-based vaccines loaded with
   GL261 cells subjected to F/T cycles or treated with 2.0 µM MTX or mire
   were injected with PBS. The mice were injected on days 0 and 7, and
   seven days after the last vaccination they were intracranially injected
   with viable GL261 cells using stereotactic coordinates. B The curve
   represents the survival of mice in the four treatment groups of 6–7
   mice per group for up to 24 days. Statistical significance was
   determined by the Mantel-Cox logarithmic test, *p < 0.01. C
   Diffusion-weighted tomography images for determining tumor volume
   (n = 6–7 per group). Statistical significance was determined by
   unpaired Mann–Whitney U test, *p < 0.05. D, E Temporal progression of
   neurological deficits in mice treated as described in (A). The
   neurological status of the mice for up to day 22 after intracranial
   tumor inoculation. The percentage of mice after intracranial tumor
   inoculation in the DC + PS-PDT or DC + MTX group showed a significant
   difference in the degree of neurological alterations (grades 0–4);
   n = 6–7 per group; *p < 0.01, a statistically significant difference
   from the PBS group; ^#statistically significant difference from the F/T
   group; Wilcoxon test.

   Although prophylactic vaccination is valuable for the analysis of
   molecular mechanisms, it does not reflect the actual clinical situation
   when patients are therapeutically vaccinated. Therefore, we tested the
   effectiveness of DC-GL261_PS-PDT vaccines in the therapeutic orthotopic
   mouse model (Suppl. Fig. [128]3A). We found that four consecutive
   DC-GL261_PS-PDT or DC-GL261_MTX vaccine injections significantly
   increased the median survival of glioma-inoculated mice by about 38%
   (37 days versus 51 days, p < 0.02) and also resulted in about 66% more
   long-term cured survivors compared to mice injected with PBS (Suppl.
   Fig. [129]3B). This finding was confirmed by analysis of neurological
   scores, ex vivo MRI and histological analysis (Suppl. Fig. [130]3C–F).
   To further examine the adaptive immune response induced in the
   therapeutic setting by DC-GL261_PS-PDT vaccines, we performed immune
   cell phenotyping of the isolated draining lymph nodes of vaccinated
   mice. Interestingly, we found that the draining lymph nodes of mice
   therapeutically vaccinated with DC-GL261_PS-PDT contained a
   significantly increased number of CD8^+T cells compared to the control
   while the number of DCs and macrophages remain unchanged (Suppl. Fig.
   [131]3G).

Glioma cells undergoing immunogenic cell death after PS-PDT induce a Th17
signature in DCs

   To identify the pathways induced in DC vaccines after in vitro
   coculture with dying GL261_PS-PDT or GL261_MTX (positive control), we
   sequenced the RNA of bone-marrow-derived DCs. Before sequencing, we
   performed a two-step enrichment for DCs (Fig. [132]4A). Flow cytometry
   showed that this resulted in the enrichment of DCs from 11.3% to 85.3%
   of CD11c^+ cells.

Fig. 4. RNA sequencing expression analysis of bone marrow-derived dendritic
cells (DCs) co-cultured with dying/dead GL261 cells.

   [133]Fig. 4
   [134]Open in a new tab

   A DCs were co-cultured with GL261 cells treated with 2.0 µM MTX, or
   loaded with 1.4 µM photosens and exposed to PDT (PS-PDT) for 6 h. The
   DCs were depleted of dead cells and CD11-positive cells were purified
   as described in Materials and Methods and in Efimova et al. [[135]84].
   The total RNA extracted from the purified DCs was subjected to RNA-seq
   analysis. B Venn diagram showing the total number of genes that were
   differentially expressed more than twofold (|Fold Change| ≥ 2; adjusted
   p-value < 0.05, base Mean > 100) in the PS-PDT and MTX groups compared
   to controls. C Histograms with fold change in expression level of
   differentially expressed genes (adjusted p-value < 0.05, base Mean >
   100) between the experimental groups (PS-PDT, MTX) and control. D, E
   Gene-gene correlation matrices. Pearson-correlation matrix of 61 marker
   genes of DC maturation in samples under the action of PS-PDT (D) and
   samples under the action of MTX (E). Each colored square within the
   figure illustrates the correlation between two genes. Red indicates a
   very strong positive correlation, black no correlation, and green a
   very strong negative correlation. F Box plots showing the expression of
   DC genes the products of which activate a Th17 response. The Tgfb3,
   Il6, and Il23a genes were strongly expressed in the PS-PDT and MTX
   groups compared with the control group. The x-axis of the plot
   represents different groups: PS-PDT, MTX, and control. The y-axis shows
   the expression data after log[2](TPM + 1) transformation. Statistical
   significance analysis was performed using the Wald test from DESeq2.
   Adjusted p-value using Benjamini–Hochberg mode < 0.05 was considered
   statistically significant. *p < 0.05, **p < 0.01, ***p < 0.001,
   ****p < 0.0001. TPM transcripts per million.

   RNA sequencing of the DCs after 6 h of co-culture with glioma GL261
   cells treated with either PS-PDT or MTX indicated distinct
   transcriptional changes (Fig. [136]4B, C). By comparing the cell
   sequencing results of GL261_PS-PDT with those of control DCs, we
   identified 1357 differentially expressed genes (1242 up and 115 down)
   (Fig. [137]4B, C). Pathway enrichment analysis revealed that
   GL261_PS-PDT cells altered their gene expression programs linked to
   cellular processes, biological regulation, metabolic processes,
   response to stimuli, signaling, developmental processes, multicellular
   organismal processes, and immune system processes (Suppl. Fig.
   [138]4A). Our positive control group, DCs cocultured with glioma GL261
   cells treated with MTX, a well-known ICD inducer [[139]19, [140]30],
   had 4072 differentially expressed genes compared with control DCs (2354
   up and 1718 down, Fig. [141]4B, C). We also identified 1269 genes
   common between the PS-PDT and MTX groups (among them 1181
   simultaneously up, 87 simultaneously down and 1 gene with changes in
   opposite directions). In addition, we analyzed markers of DC
   maturation, including exogenous signals (cytokines, chemokines) and
   ligands on the surface of DCs, and found that 61 genes were responsible
   for the expression of these molecules (Fig. [142]4D, E). Pearson
   correlation matrices showed that though both PS-PDT and MTX triggered
   ICD, they induced quite different expression profiles of the selected
   genes (Fig. [143]4D, E).

   Next, to identify the immunogenic signature triggered in DCs by dying
   GL261_PS-PDT cells, we analyzed the marker genes responsible for T cell
   differentiation (Th1, CTL, Th17, and Treg). We checked whether these
   marker genes are among the differentially expressed genes. Remarkably,
   we found that DCs cocultured with GL261_PS-PDT cells showed high
   expression levels of genes that activate Th17 cells (Fig. [144]4F). The
   Tgfb3, Il6, and Il23a genes were strongly expressed in DCs cocultured
   with dying glioma cells pulsed with PS-PDT or with MTX (positive
   control) compared with the control group. At the same time, the marker
   genes of Th1, CTL, and Treg cells were not differentially expressed.
   These data suggest that the immunogenicity of the ICD-based DC vaccine
   is associated with a Th17 signature in DCs.

Blocking RORγt reduced the protection by DC vaccines loaded with
DC-GL261_PS-PDT cells in the orthotopic glioma mouse model

   Next, we used the orthotopic murine model to validate the role of Th17
   cells in the anti-glioma protective immunity induced by the
   DC-GL261_PS-PDT vaccine. To that end, we blocked RORγt, a transcription
   factor regulating the expression of the pro-inflammatory cytokine IL-17
   in human Th17 cells [[145]34, [146]35]. We used a potent RORγt
   inhibitor (GSK805) that can penetrate into the central nervous system
   [[147]36, [148]37]. GSK805 was intraperitoneally injected 12, 24, 48,
   and 72 h after intraperitoneal injection of the DC-GL261_PS-PDT vaccine
   (Fig. [149]5A). In mice vaccinated with the DC-GL261_PS-PDT vaccine,
   treatment with 10 mg/kg of the RORγt inhibitor (GSK805) significantly
   decreased median survival (Fig. [150]5B). In parallel with this
   decrease in overall survival, GSK805 also resulted in more noticeable
   clinical manifestations and earlier onset of symptoms (Fig. [151]5C,
   D). To support these data, we non-invasively monitored the mice by
   using MRI imaging (Fig. [152]5E, F). The group of mice treated with the
   RORγt inhibitor showed disguised glioma formation that altered
   ventricular morphology, deformed the cerebral cortex, and deepened the
   tumor lesion towards the optic nerve, resulting in exophthalmos. On the
   other hand, almost none of the mice vaccinated with the DC-GL261_PS-PDT
   vaccine showed noticeable glioma masses at the site of inoculation and
   all of them retained better brain morphology. Further, we observed by
   immunohistochemistry infiltration of IL-17^+ cells in the brain after
   vaccination with the DC-GL261_PS-PDT vaccine (Fig. [153]5G). Moreover,
   the depletion of IL-17^+ cells in the brains of mice treated with
   DC-GL261_PS-PDT and GSK805 was confirmed by immunohistochemistry (Fig.
   [154]5G). These data demonstrate that pharmacological inhibition of
   RORγt significantly reduces the immunogenic potential of the
   DC-GL261_PS-PDT vaccine. Collectively, these results suggest the
   importance of Th17 cell responses for efficient glioma DC-based
   therapy.

Fig. 5. The efficacy of ICD-based DC vaccines depends on RORγt signaling.

   [155]Fig. 5
   [156]Open in a new tab

   A Mice received DC vaccines loaded with PS-PDT on days 0 and 7. After
   each vaccination, the mice received four intraperitoneal injections of
   the RORγt inhibitor (GSK805, 10 mg/kg) or vehicle (control) 12, 24, 48,
   and 72 h postvaccination, or were injected with PBS. In addition, a
   control group received PBS instead of a vaccine. Seven days after the
   last vaccination with the DC-based vaccine, the mice were
   intracranially injected with viable GL261 cells. B The curve represents
   survival of the mice for up to 30 days after the intracranial
   inoculation of tumor cells. P < 0.02, Mantel-Cox logarithmic test
   (n = 13–14). C Analysis of the neurological status of mice for up to
   day 31 after the intracranial tumor inoculation. The percentage of mice
   after intracranial tumor inoculation showing different degrees of
   neurological alterations (grades 0–4) in the different groups is shown.
   D The temporal progression of neurological deficits in mice treated as
   described in (A) is shown for each group (n = 11–14). *p < 0.01, a
   statistically significant difference from the PBS group, ^#p < 0.01, a
   statistically significant difference from the DC + PS-PDT + GSK805
   group; Wilcoxon test. E Representative T1-tomograms of layer-by-layer
   frontal brain sections on day 16. The tumor mass is indicated by red
   arrows. F Tumor volume was obtained by analysis of diffusion-weighted
   MRI images; n ≥ 6 per group. G The sagittal brain sections of mice
   treated with DC + PS-PDT or DC + PS-PDT in the presence of GSK805 or
   treated with vehicle or PBS. The sections stained with anti-IL-17
   antibodies (green) demonstrated the presence of IL-17^+ cells in the
   brain of mice treated with DC + PS-PDT, but these cells were not
   present in the brains of mice treated with DC + PS-PDT in the presence
   of GSK805. Scale bars 20 μm.

Relationship between the Th17-associated genetic signature and patient
survival

   It has been proposed that the presence of specific T-lymphocyte subsets
   and the absence of immunosuppressive cells is associated with improved
   prognosis in cancer patients and can yield information relevant to the
   prediction of treatment response and various other pharmacodynamic
   parameters [[157]38]. Therefore, we studied the clinical prognostic
   potential of the Th17 gene signature observed in the DC-GL261_PS-PDT
   vaccine. To that end, we analyzed a Th17-based immune contexture in
   patients with low-grade glioma (LGG) by making use of the very large,
   standardized and publicly available cohort of 508 LGG patients from The
   Cancer Genome Atlas (TCGA) [[158]39]. The lymphocyte subtype-specific
   mRNA signature for Th17 cells is available from a previous study
   [[159]40]. This mRNA signature was analyzed in the TCGA-LGG dataset to
   select a cluster of genes within this signature showing strong
   collective co-expression (so-called “metagene”) [[160]14, [161]41]
   centered on the standard/specific Th17 cell marker (IL17A) [[162]42].
   To identify the LGG-specific Th17 cell-associated metagene, we
   calculated the co-expression of genes in the signature (correlation
   matrix) (Suppl. Fig. [163]4B). The correlation matrix was subjected to
   unsupervised hierarchical clustering with Euclidean distance
   measurement and average linkage clustering. Then, in the TCGA-LGG
   cohort, we calculated the prognostic impact of the expression of the
   Th17 metagene on overall patient survival, dividing patients into
   groups with low and high metagene expression by the 75th percentile. In
   addition, we calculated the percent difference in median survival
   (%ΔMS) between the high-expression and low-expression groups. Though
   the difference in overall survival was not statistically significant
   between the groups with high and low expression of the Th17 metagene
   (log-rank p-value > 0.05), a positive %∆MS of +10% indicates a possible
   trend that the strong expression of the Th17 metagene is associated
   with prolonged overall survival (Suppl. Fig. [164]4C).

   These data are in line with our in vitro RNA-seq data, where glioma
   undergoing ICD after PS-PDT induced a significantly stronger Th17
   signature in murine DCs as compared to control (Fig. [165]4F) and
   provided support for the effect of the pharmacological inhibition of
   RORγt on the immunogenic potential of the DC-GL261_PS-PDT vaccine (Fig.
   [166]5B–F) and the depletion of IL-17^+ CD4^+ T cells in the brains of
   mice treated with DC-GL261_PS-PDT and GSK805 (Fig. [167]5G).
   Collectively, these results suggest an important role for Th17 cell
   responses in the efficacy of the DC-based immunotherapy of glioma.

Identification of the four-gene signature associated with overall survival of
glioma patients

   To identify the genes associated with the survival of patients with
   glioma, we analyzed the transcriptome data from the TCGA-LGG dataset.
   The differential expression of 249 genes was statistically significant
   (|Fold Change| ≥ 2, adjusted p-value < 0.05, base Mean > 50) between
   the DEAD and ALIVE groups of patients; 236 genes were upregulated and
   13 were downregulated in the DEAD group relative to ALIVE patients.
   Among the genes found, we identified 158 genes that were differentially
   expressed (adjusted p-value < 0.05, base Mean > 50) in ALIVE TCGA-LGG
   patients compared to CONTROL patients, of which 119 genes were
   upregulated and 39 were downregulated.

   Next, 158 genes were matched with 5135 differentially expressed genes
   (adjusted p-value < 0.05, base Mean > 100) from our RNA-seq data of DCs
   cocultured with glioma GL261 cells treated with PS-PDT or MTX (Fig.
   [168]4). We identified five genes (CFH, CYP1B1, GALNT3, SMC4, and VAV3)
   the expression of which changed in the in vitro DC experiments, and
   they were changed in the same direction in the ALIVE patients. Next,
   based on log[2](TPM + 1) normalized expression data from the TCGA-LGG
   dataset, we performed univariate Cox proportional hazards regression
   analysis for each gene. This analysis allowed us to select four
   prognosis-associated genes (CFH, GALNT3, SMC4, and VAV3) that were
   statistically significantly correlated with overall survival (p-value <
   0.05). The change in the expression of these four prognostic genes in
   the DEAD/ALIVE groups of TCGA-LGG patients and in DCs co-cultured with
   glioma 261 cells killed by PS-PDT or MTX groups are compared to the
   corresponding control groups (Fig. [169]6A). This analysis indicates a
   commonality of mechanisms with a good prognosis (ALIVE patients with
   low expression of these genes) and with the use of PS-PDT or MTX.

Fig. 6. Four-gene prognostic signature and its relationship with overall
survival in the TCGA-LGG dataset.

   [170]Fig. 6
   [171]Open in a new tab

   A Plot of log[2] Fold Change expression of the genes from the four-gene
   prognostic model. For each gene, values of log[2]FC are presented for
   the DEAD/ALIVE patient groups and for the PS-PDT/MTX groups versus the
   corresponding control groups. The conditional control level is shown
   with a green dash-dotted line. B The risk score of each LGG patient.
   The median risk score for categorizing patients into low-risk (blue) or
   high-risk (red) groups is 1.92. C The time to death of dead patients
   (black) and time until last follow-up of live patients (orange). D
   Heatmap of gene expression profiles of the four prognostic genes. E
   Kaplan–Meier survival analysis of high-risk versus low-risk groups. F
   Time-dependent receiver operating characteristic curve analysis of the
   four-gene predictive model. G–I Correlation between the four-gene
   prognostic signature for TCGA-LGG and the infiltration of immune and
   cancer cell subtypes: CAFs (G), endothelial cells (H) and macrophages
   (I).

   Then, we used prognosis-related genes as variables in the final
   multivariate Cox regression model. The С-index (index of concordance)
   was considered an assessment of the predictive model. Its value was
   0.82, which indicates that the four-gene signature can successfully
   predict the prognosis of patients with glioma. In our prognostic model,
   the risk score = (0.121232 × expression value of CFH) + (0.114831 ×
   expression value of GALNT3) + (0.521462 × expression value of
   SMC4) + (0.329177 × expression value of VAV3). The coefficients of all
   four genes are positive, which means that patients with high expression
   levels of CFH, GALNT3, SMC4, and VAV3 have low survival rates. Next,
   the risk score for every patient based on our prognostic model was
   calculated and the 508 patients were separated into low-risk (n = 254)
   and high-risk (n = 254) subgroups based on the median risk score of
   1.92.

   The distribution of the TCGA-LGG patients’ risk scores (with
   color-coded risk level), time to death or until the last follow-up, and
   RNA-seq expression of the patients are shown in Fig. [172]6B–D. The
   survival (or censoring) time of patients with high-risk scores was
   lower than in those with low-risk scores (Fig. [173]6B, C). The
   expression of prognostic genes was higher in high-risk patients (Fig.
   [174]6B, D). A heatmap was generated using the Z-score on
   log[2](TPM + 1) normalized expression value to illustrate the relative
   expression levels of the genes (Fig. [175]6D). Z-score normalization
   was used in addition to centering and variance stabilization. The
   Kaplan–Meier overall survival curves of the two groups based on the
   four prognostic genes were significantly different (log-rank p-value =
   7.02e−10 < 0.05) (Fig. [176]6E). We also used receiver operating
   characteristic curve analysis to estimate the accuracy of the risk
   score’s prediction of the clinical outcomes of TCGA-LGG patients (Fig.
   [177]6F). The calculated risk score was most accurate in assessing the
   one-year prognosis of TCGA-LGG patients, with an area under the curve
   of 0.88. Generally, the accuracy of the prognosis signature exceeded
   0.7 in the clinical outcome prediction from one to five years (area
   under curve of 0.85 and 0.75 after 3 and 5 years, respectively).

   Next, we analyzed the correlation between the prognostic signature and
   the infiltration of immune cells and cancer cells in TCGA-LGG. The
   fraction of B cells, CD4^+ T cells, CD8^+ T cells, macrophages, NK
   cells, cancer-associated fibroblasts, and endothelial cells was
   predicted using the EPIC deconvolution method. Cancer-associated
   fibroblasts, endothelial cells, and macrophages were significantly
   correlated (p-value < 0.05) with the risk score (Fig. [178]6G–I), and
   Pearson’s correlation coefficients were 0.1761, 0.3074 and 0.1552,
   respectively. This indicates that the infiltration of cancer-associated
   fibroblasts, endothelial cells, and macrophages is positively
   correlated with the poor prognosis of TCGA-LGG, and an increase in the
   proportions of these cells is associated with an increase in the risk
   score. There was also a trend towards a decrease in the proportion of
   CD4^+ and CD8^+ T cells in patients with high-risk scores, but this
   correlation was not statistically significant (data not shown).

Validation of the prognostic four-gene signature on overall survival in the
CGGA-LGG, TCGA-GBM, and CGGA-GBM datasets

   To determine whether the four-gene prognostic signature had similar
   prognostic value in different cohorts, we first used the CGGA-LGG
   dataset for validation. The risk score of each patient was calculated
   according to the risk score formula derived from the training TCGA-LGG
   dataset. The 408 patients in the validation CGGA-LGG set were separated
   into low-risk (n = 186) and high-risk (n = 222) subgroups based on the
   training set cutoff value of 1.92. Next, the distribution of risk
   score, survival status, and the heatmap of prognostic gene expression
   in the CGGA-LGG dataset were analyzed (Fig. [179]7A–C). Consistent with
   the results of the TCGA-LGG dataset, the Kaplan–Meier survival curves
   and log-rank test (p-value = 9.28e−07) revealed a significant
   difference in overall survival between the low- and high-risk groups in
   the CGGA-LGG dataset (Fig. [180]7D). Receiver operating characteristic
   curve analysis of the four-gene model was conducted: the values of area
   under curve were 0.78, 0.78, and 0.73, respectively, for predicting 1-,
   3-, and 5-year survival rates (Fig. [181]7E).

Fig. 7. Validation of the prognostic four-gene signature for overall survival
in the CGGA-LGG, TCGA-GBM and CGGA-GBM datasets.

   [182]Fig. 7
   [183]Open in a new tab

   A–E Characteristics of the four-gene prognostic signature in the
   validation CGGA-LGG dataset. A The low- and high-risk scores for each
   CGGA-LGG patient. B The time to death of patients (black, DEAD) and
   time until last follow-up of live patients (orange, ALIVE) from the
   CGGA-LGG dataset. C Heatmap of gene expression profiles of the four
   prognostic genes in the CGGA-LGG dataset. D Kaplan–Meier survival
   analysis of high-risk versus low-risk groups in the CGGA-LGG dataset. E
   Time-dependent receiver operating characteristic curve analysis of the
   four-gene predictive model in the CGGA-LGG dataset. F–J Characteristics
   of the four-gene prognostic signature in the validation CGGA-GBM
   dataset. F The low- and high-risk scores of each CGGA-GBM patient. G
   The time to death of dead patients (black, DEAD) and time until last
   follow-up of live patients (orange, ALIVE) from the CGGA-GBM dataset. H
   Heatmap of gene expression profiles of the four prognostic genes in the
   CGGA-GBM dataset. I Kaplan–Meier survival analysis of high-risk versus
   low-risk groups in the CGGA-GBM dataset. J Time-dependent receiver
   operating characteristic curve analysis of the four-gene predictive
   model in the CGGA-GBM dataset.

   In addition, the performance of the prognostic model built for patients
   with LGG was tested on patients with glioblastoma multiforme (GBM). To
   do this, we repeated the validation procedure for the TCGA-GBM and
   CGGA-GBM datasets. When dividing TCGA-GBM patients into low- and
   high-risk groups according to the cutoff value, 149 of the 151 patients
   were in the high-risk group. Receiver operating characteristic curve
   analysis of the four-gene model revealed that the values for the
   TCGA-GBM dataset of area under curve were 0.9 for predicting 5-year
   survival rate. A similar trend, but not as extreme, was observed in the
   CGGA-GBM dataset: there were more patients in the high-risk group
   (n = 195) than in the low-risk group (n = 23) (Fig. [184]7F–H). These
   results were expected, since glioblastoma is a more aggressive tumor
   than glioma and the number of dead patients in the GBM datasets is
   greater than the number of those alive. Consistent with the results of
   the low-grade glioma datasets, patients in the high-risk group had a
   poorer prognosis in the CGGA-GBM dataset, with log-rank p-value = 0.03
   (Fig. [185]7I). The 1-, 3- and 5-year area under curve were 0.58, 0.63
   and 0.69, respectively (Fig. [186]7J).

Discussion

   In this study, we found that DC vaccines primed with glioma cells
   undergoing ICD after PS-PDT protect mice not only against challenge
   with viable glioma GL261 cells in several orthotopic glioma models in
   the prophylactic mode but are also effective in the curative setting,
   which most closely resembles the clinical setting. Moreover, by using
   RNA-seq analysis, we found that while glioma cells are undergoing ICD
   after PS-PDT, they induce a typical Th17 signature in the DC vaccines
   and that these vaccines were highly effective in protecting mice
   against gliomas (Fig. [187]8). Furthermore, the effect of these DC
   vaccines was significantly reduced when a specific RORγt inhibitor was
   used, confirming that a Th17-induced anti-tumor immune response is
   required for the efficacy of these DC vaccines in the murine orthotopic
   glioma model. Furthermore, by comparing the transcriptome program of
   the ICD-based DC vaccine with the transcriptome data from the TCGA-LGG
   patients, we identified a four-gene signature (CFH, GALNT3, SMC4, and
   VAV3) that is strongly associated with overall survival of glioma
   patients, and we validated it on the CGGA-LGG, TCGA-GBM, and CGGA-GBM
   datasets (Fig. [188]8).

Fig. 8. DC vaccines loaded with glioma cells killed by PS-PDT induce Th17
anti-tumor immunity and provide a four-gene signature for glioma prognosis
(graphical abstract).

   [189]Fig. 8
   [190]Open in a new tab

   We developed ICD-based DC vaccines and demonstrated that they induce
   significant protective immunity against glioma in the orthotopic model.
   To investigate the molecular mechanism of DC vaccines loaded with
   glioma cells undergoing ICD, we performed a comparative analysis of
   their transcriptional levels in two ICD-inducing modalities (PS-PDT and
   MTX) and we analyzed the marker genes responsible for T cells
   differentiation (Th1, CTL, Th17, and Treg). This study revealed that
   the expression of DC genes the products of which (Tgfb3, Il6, and
   Il23a) activate Th17 cells were expressed at high levels in DCs
   cocultured with dying glioma cells treated with PS-PDT or MTX compared
   to control. Notably, blocking RORγt reduced the protection of DC
   vaccines loaded with dying glioma cells treated with PS-PDT in the
   orthotopic glioma model. Finally, by matching differentially expressed
   genes (DEG, between “ALIVE” and “CONTROL” patients in TCGA-LGG
   datasets) from our RNA-seq data of DCs cocultured with glioma GL261
   cells treated with PS-PDT or MTX, we established a predictive model
   based on the four-gene signature (CFH, GALNT3, SMC4, and VAV3).
   Application of this signature to the TCGA-LGG dataset predicted the
   patients’ overall survival. When it was validated on the overall
   survival of the CGGA-LGG, TCGA-GBM, and CGGA-GBM datasets, it
   accurately predicted the five-year survival rates. In conclusion, in
   this study we have shown that DC vaccines loaded with glioma cells
   killed by photodynamic therapy induce Th17 anti-tumor immunity and
   provide a four-gene signature for glioma prognosis. These findings open
   attractive prospects for improving glioma therapy by employing ICD and
   PS-PDT-based DC vaccines to induce Th17 immunity and to using the
   prognostic model to predict the overall survival of glioma patients.
   ^1Of note, in the previous study we demonstrated that dying/dead GL261
   cells treated with PS-PDT undergo typical hallmarks of ICD such as
   exposure on their surface calreticulin and release DAMPs such as ATP
   and HMGB1 [[191]20]. Importantly dying/dead GL261 treated with PS-PDT
   induce efficient activation and maturation of DCs in vitro [[192]20]
   justifying their use as a vaccine in the current study.

   Immunotherapy has emerged as a standard of care and the first-line
   treatment for several cancer types, including glioma, mainly due to the
   discovery of immune checkpoint inhibitors and their significant
   clinical impact [[193]43]. One of the promising concepts of cancer
   immunotherapy relies on the induction of ICD, which is on the one hand
   characterized by adjuvanticity and emission of DAMPs and cytokines,
   leading to activation of anti-tumor immunity, and on the other hand
   dictated by the antigenicity of dying cancer cells, which is defined by
   the level of tumor (neo)antigens [[194]23, [195]44, [196]45].
   Adjuvanticity and antigenicity of dying cancer cells are required for
   generation of anti-tumor immunity and long-lasting immune memory, which
   is required for lifelong protection of patients. In recent years,
   several promising ICD-inducing therapeutic modalities have been
   introduced, including therapeutic strategies based on PDT, which is
   effective for certain types of cancer. PDT involves the use of a
   photosensitizing agent and photoexciting light, which, in the presence
   of molecular oxygen, generate singlet oxygen and other cytotoxic
   oxidants that trigger ICD [[197]26, [198]28, [199]46]. Importantly, the
   potential of ICD in cancer therapy has been well-established
   [[200]23–[201]25, [202]44, [203]45], and most studies have focused on
   the induction of ICD in murine heterotopic cancer models [[204]14,
   [205]19, [206]27, [207]29, [208]47]. However, in this study, we
   developed DC vaccines primed with glioma cells undergoing ICD after
   PS-PDT for glioma therapy. It is noteworthy that several different
   methods are available for designing patient-derived cancer cell
   vaccines. Two widely used methods are the use of dying cancer cells
   [[209]14, [210]19, [211]27, [212]30, [213]47] and DCs primed with tumor
   cell lysates [[214]14, [215]48]. In our work, we subjected glioma cells
   to PS-PDT, leading to the induction of the typical features of ICD,
   followed by preparing tumor cell lysates and priming DCs with them. We
   had already characterized PS-PDT in our previous work [[216]20] and
   showed that glioma GL261 cells subjected to PS-PDT undergo ICD with the
   emission of the three major DAMPs (CRT, ATP, and HMGB1), thereby
   inducing activation and maturation of DCs in vitro [[217]20]. In our
   current study, we developed in orthotopic glioma models a novel
   immunotherapy based on a DC vaccine pulsed with GL261 glioma cells
   treated with PS-PDT.

   Importantly, whole glioma tumor cells are often lysed by several F/T
   cycles, which leads to unregulated cell death known as accidental
   necrosis, which has limited immunogenic potential [[218]19, [219]20,
   [220]27]. In our current study, we used two different orthotopic glioma
   prophylactic mouse models: subcutaneous vaccination with glioma GL261
   cells undergoing ICD (Fig. [221]2) and intraperitoneal vaccination with
   DC vaccines pulsed in vitro with glioma GL261 cells killed by PS-PDT or
   F/T, or treated with MTX as a positive control (Fig. [222]3). The
   results demonstrate that the ICD induced in glioma GL261 cells killed
   by PS-PDT provided considerable survival benefits against challenge
   with viable GL261 cells in both mouse models as compared to vaccination
   with glioma cells after F/T. Notably, the DC-GL261_PS-PDT vaccine
   increased the median survival time by more than 12% as compared to the
   F/T group. These data confirm the previously reported observation of
   the non-immunogenicity of cancer cells undergoing accidental necrosis
   [[223]19, [224]20, [225]27, [226]29]. Interestingly, DC-GL261_PS-PDT
   was also effective for treating existing glioma in mice in the curative
   mode as well (Suppl. Fig. [227]3). Moreover, we found that PS-PDT
   induced a mixed type of regulated cell death in GL261 cells with
   features of both apoptosis and ferroptosis. Induction of cell death by
   PDT can also be beneficial because PDT can work synergistically with
   ferroptosis to provide a source of reactive oxygen species for the
   Fenton reaction [[228]28]. The dose of photosensitizer used to trigger
   cell death must also be considered because there is a non-linear
   relationship between photosensitizer concentration and the PDT-induced
   antitumor immune response [[229]49].

   To investigate the efficacy of DC vaccines further and to unravel the
   immunogenic signature triggered in the DCs by dying GL261_PS-PDT cells,
   we performed RNA-seq analysis of DCs co-incubated with dying glioma
   GL261 cells triggered by PS-PDT or MTX treatment. Importantly, we found
   significantly different expression levels of the Tgfb-3, IL-6, and
   IL-23a genes in the DCs pulsed with dying cells undergoing ICD after
   PS-PDT or MTX treatment as compared to control untreated DCs. It is
   known that the Th17 genetic signature is specific to a Th17 immune cell
   response and contributes to the steering of this response in mice and
   in humans [[230]50]. Th17 cells have been identified as an independent
   subtype of inflammatory T cells with an IL-17 and transcription factor
   RORγt profile [[231]51]. Of note, naive CD4^+ T cells can be induced to
   differentiate, depending on the local cytokine milieu, towards a T
   helper-1 (Th1), Th2, Th17, or regulatory T cell phenotype with unique
   signaling pathways and expression of specific transcription factors
   [[232]52]. Thus, our RNA-seq data clearly point to the activation of
   the Th17 signature in the DC vaccines after their coculture with
   GL261_PS-PDT cells. We previously reported that glioma GL261 cells
   subjected to PS-PDT and co-cultured with DCs induced the production of
   IL-6 [[233]20], indirectly supporting our current RNA-seq data (Fig.
   [234]4). It is also known that dying cancer cells undergoing ICD induce
   a typical Th1 signature in vitro and in vivo [[235]14, [236]19]. But
   our data provide more specific direct experimental evidence pointing to
   the ability of glioma undergoing ICD to induce a Th17 molecular
   signature in antigen-presenting cells (i.e., DCs). Furthermore, in the
   prophylactic orthotopic glioma model, co-injection of a specific
   inhibitor (GSK805) of RORγt, the transcription factor of the Th17
   response, significantly reduced the protective effects of the DC
   vaccine based on ICD and PS-PDT. Moreover, we found that TCGA-LGG
   patients have +10% %∆MS, which suggests that the Th17 metagene might be
   associated with prolonged overall survival (Suppl. Fig. [237]4B, C),
   confirming a previous report on the relationship between high Th17
   metagene expression and longer survival of patients with GBM [[238]14].
   Interestingly, when patients with mutated isocitrate dehydrogenase 1
   (IDH1), a molecularly distinct subtype of diffuse glioma, were
   vaccinated with IDH1(R132H)-specific peptide vaccine, they showed
   production of tumor necrosis factor (TNF), interferon-γ (IFNγ), and
   IL-17 upon in vitro re-stimulation of peripheral IDH1-vaccine-induced T
   cells with IDH1(R132H), which indicates the involvement of Th1 and Th17
   subtypes of Th cells [[239]53]. These results also indirectly support
   our current data obtained in a prophylactic orthotopic glioma model in
   which a specific inhibitor of RORγt was also injected. The results
   point to a promising strategy for glioma therapy by employing ICD and
   PS-PDT-based DC vaccines to induce Th17 immunity. However, it is
   important to stress that even though multiple types of IL-17-producing
   cells are found in the tumor bed in mouse models and in humans, their
   role in tumor progression remains controversial [[240]54]. It is
   noteworthy that our findings are in striking contrast to the previously
   reported role of Th17 and IL-17 in glioma promotion, where a direct
   correlation between two-year progression-free survival and low
   incidence of IL-17 producing cells was reported [[241]55]. However, the
   molecular mechanism of IL-17-mediated glioma progression was not shown
   in that study. In another recent study, the CD8^+ and CD4^+
   tumor-infiltrating lymphocyte compartment was characterized in depth.
   The results pointed to a pronounced Th17 commitment of CD4^+
   tumor-infiltrating lymphocytes in untreated GBM patients [[242]56].
   Although the authors proposed that exaggerated Th17 responses in the
   GBM bed may create a dominant-negative environment for productive Th1
   and CTL responses blocking adaptive antitumor immunity, a direct link
   between exaggerated Th17 responses and survival of GBM patients has
   been not identified. Indeed, the brain tumor microenvironment varies
   greatly during its progression from early to late disease, among
   different tumor types, among individuals with the same diagnosis, and
   in non-neoplastic cell types and cell states [[243]57]. In this regard,
   it has been shown that myeloid cells in GBM tissue are the dominant
   immune cell type [[244]57–[245]59] and that GBM has much fewer
   lymphocytes, and in particular T cells such as CD4^+ and CD8^+ T cells,
   which represent about 10 and 7%, respectively, of the immune cell
   infiltrate in GBM [[246]56]. Therefore, the underlying molecular and
   cellular mechanisms of the role of IL17 in cancer are still not
   completely understood. Many interesting and challenging findings are
   expected.

   Our study allowed us to identify the four-gene signature (CFH, GALNT3,
   SMC4, and VAV3) associated with the overall survival of glioma TCGA-LGG
   patients. Currently, there are various prognostic models to predict the
   survival of LGG patients. In our study, we looked at genes that show
   commensurate expression changes in a group of live LGG patients and in
   the PS-PDT and MTX groups in our in vitro experiments. Although PS-PDT
   and MTX have different effects on the activation of immune system
   genes, both have pronounced positive effects in modeling the
   oncological process. Using TCGA-LGG data, we show that despite the
   differences in mechanisms, some genes could have significant prognostic
   value in distinguishing good from poor prognosis. Based on the immune
   cell deconvolution method, we found a significant correlation between
   the four-gene prognostic signature and the infiltration of immune and
   cancer cell subtypes. The results demonstrate that patients with higher
   infiltration levels of cancer-associated fibroblasts, endothelial cells
   and macrophages have a poor prognosis. Of note, these four genes that
   constitute the prognostic model have been shown to have the following
   associations: (i) high expression level of Complement Factor H (CFH)
   has been associated with the progression of cutaneous squamous cell
   carcinoma [[247]60]; (ii) GALNT3 (polypeptide
   N-acetylgalactosaminyltransferase 3) has been linked to neuroblastoma
   [[248]61]; (iii) strong expression of structural maintenance of
   chromosomes 4 (SMC4) promotes an aggressive phenotype in glioma cells
   [[249]62]; (iv) depletion of Vav3 (guanine nucleotide exchange factor)
   by siRNA oligonucleotides suppresses GBM cell migration and invasion
   [[250]63]. These findings reinforce the functional relevance of this
   four-gene prognostic signature.

   Altogether, our findings point to the importance of the Th17 signature
   as a prognostic marker and to its positive therapeutic impact in glioma
   therapy based on DC vaccines pulsed with dying cancer cells undergoing
   ICD triggered by PS-PDT (Fig. [251]8). Considering the wide GBM
   heterogeneity and plasticity, which result in the lack of “quality”
   neoantigens, DC vaccines pulsed with dying cancer cells undergoing ICD
   triggered by PS-PDT may represent an attractive approach for producing
   whole-tumor derived immunogenic neoantigens for effective glioma
   therapy. Moreover, a key interpretation of our new four-gene signature
   model, which demonstrated a strong predictive power in both the
   training and validation cohorts, may help to develop novel and testable
   prognostic and therapeutic opportunities for glioma patients.

Materials and methods

Cell lines

   Murine glioma GL261 cells were cultured at 37 °C under 5% CO[2] in
   Dulbecco’s Modified Eagle’s Medium (DMEM) containing 4.5 g/L glucose
   and supplemented with 2 mM glutamine, 100 µM sodium pyruvate, 100
   units/ml penicillin, 100 µg/L streptomycin and 10% fetal bovine serum,
   all purchased from Thermo Fisher Scientific. GL261 cells were kindly
   provided by Prof. P. Agostinis (Laboratory of Cell Death Research &
   Therapy, Department of Cellular and Molecular Medicine, KU Leuven,
   Leuven, Belgium).

Quantification of cell death induction by PS-PDT and MTX and analysis of cell
death inhibitors

   Cell death was induced by photosens (PS)-based PDT or mitoxantrone
   (MTX, Sigma Aldrich). For PS-PDT, GL261 cells were first incubated with
   1.4 µM PS in serum-free DMEM for 4 h and then irradiated with a light
   dose of 20 J/cm^2 in photosensitizer-free media. After PDT, the cells
   were cultured in complete DMEM for 24 h. For MTX induction, the cells
   were cultured in full medium with 2.0 µM MTX for 24 h. Cells loaded
   with PS or MTX were handled in the dark or in subdued light. Control
   cells were cultured in the same conditions but without agents or PDT.
   Accidental necrosis in glioma GL261 cells was induced by three cycles
   of freezing (–80 °C) and thawing (37 °C).

   The following cell death inhibitors were used: the pan-caspase
   inhibitor
   carbobenzoxy-valyl-alanyl-aspartyl-[O-methyl]-fluoromethylketone
   (zVAD-fmk, 25 μM, Sigma-Aldrich), the RIPK1 inhibitor necrostatin-1s
   (Nec-1s, 20 μM, Abcam), the inhibitor of reactive oxygen species and
   lipid peroxidation ferropstatin-1 (Fer-1, 1 μM, Sigma-Aldrich) and the
   iron chelator, deferoxamine (DFO, 10 μM, Sigma-Aldrich). The cell death
   inhibitors were added together with the corresponding reagent and the
   cells were incubated for 4 h in serum-free DMEM with PS and 24 h in
   complete DMEM with MTX. Before PDT, the medium was replaced with
   complete DMEM containing the respective cell death inhibitor, the cells
   were irradiated with light at 20 J/cm^2, and then they were incubated
   for 24 h.

   MTT assay (AlfaAesar) was performed using
   3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyl-2H-tetrazolium bromide
   according to the manufacturer’s instructions and the optical density
   was measured at 570 nm.

Generation of bone-marrow-derived dendritic cells (DCs)

   Bone marrow was isolated from tibias and femurs in RPMI medium (GIBCO)
   supplemented with 5% heat-inactivated fetal calf serum, 20 ng/ml murine
   GM-CSF (UGent-IRC-VIB Protein Core Facility), 1% L-glutamine, and 50 μM
   2-mercapthoethanol. The bone marrow was suctioned with a 25 G needle
   (0.5 × 25 mm), resuspended, and coarse debris was filtered through a
   Cell Strainer 70 μm (Falcon). The suspension was cleared of
   erythrocytes with a lysing solution. The cells were grown for up to 10
   days. Fresh culture medium was added on day 3, and on days 6 and 9 the
   medium was fully refreshed.

Phagocytosis assay

   Target glioma GL261 cells were labeled with 1 μM CellTracker Green
   CMFDA (Molecular Probes) in serum-free DMEM for 30 min and then induced
   by MTX as described above. For the PDT group, GL261 cells were loaded
   with 1.4 µM PS in serum-free DMEM for 4 h and then either left
   untreated or cell death was induced by PDT as described above. The
   cells were collected, washed, and co-cultured with bone marrow-derived
   dendritic cells (DCs) in ratios of 1:2 or 1:5 for 2 h. Next, the
   co-cultured cells were harvested, incubated with a mouse Fc block
   CD16/CD32 (ThermoFisherScientific), stained with PE-Cy7-anti-CD11c (BD
   PharMingen, 561022), and finally analyzed by flow cytometry on a
   CytoFlex (Beckman Coulter). Analysis was performed by using CytExpert
   software. True uptake of dead cells labeled with
   5-chloromethylfluorescein diacetate (CMFDA) by DCs was determined by
   using a gating strategy that allows analysis only of singlets that are
   identified as CD11c^+ CMFDA double-positive cells.

DCs enrichment after coculture with glioma GL261 cells for RNA isolation

   DCs isolated from four C57BL/6J mice were used for each experimental
   group. DCs were cocultured for 6 h with dying/dead glioma GL261 cells
   in a ratio of 1:5. There were three experimental groups: DCs alone
   (negative control) and DCs cocultured with glioma GL261 cells pulsed
   with PS-PDT (PS at a dose of 1.4 µM, 24 h) or with MTX (positive
   control, 2 µM, 24 h). For the cocultures of DCs with treated GL261
   cells, additional technical replicates were included. After 6 h of
   coculture, the cells were harvested and washed once with DPBS.
   Enrichment of DCs from the coculture proceeded as follows. First, dead
   cells were removed with a Dead Cell Removal Kit (MACS Miltenyi Biotec).
   Then, the final eluate was loaded onto MS columns (MACS Miltenyi
   Biotec) and treated with CD11c MicroBeads UltraPure (MACS Miltenyi
   Biotec) to enrich for CD11c^+ DCs. The purity of CD11c^+ DCs was
   analyzed by flow cytometry on a BD FACS Canto II flow cytometer after
   each step. Enriched CD11c^+ cells were snap frozen in liquid nitrogen,
   and total RNA was isolated using RNeasy Mini Kit (QIAGEN). The quantity
   and integrity of the RNA were determined with a NanoDrop 8000
   spectrophotometer (Thermo Fisher) and a Fragment Analyzer 5200 system
   (Agilent), respectively.

RNA-seq analysis, RNA-seq pipeline, and data quantification

   The TruSeq Stranded mRNA kit (Illumina) was used to prepare a RNA-seq
   library according to the manufacturer’s protocol, followed by PE100
   cycle sequencing on one lane of a NovaSeq 6000 S1 (Illumina). Quality
   assessment of the raw FASTQ files obtained after sequencing was
   controlled with FastQC tool v0.11.5 [[252]64]. All the files passed
   quality control. Next, the BWA-MEM tool v.0.7.17 [[253]65] was used to
   map raw RNA-seq reads against Ensembl Mus musculus GRCm39 and
   FeatureCounts v2.0.1 [[254]66] to get quantification estimates at the
   transcript level.

Mice experiments

   Female C57BL/6J mice (7–8 weeks old) were housed in specific
   pathogen-free conditions. The mouse experiments were performed
   according to the guidelines of the local Ethics Committees of the
   National Research Lobachevsky State University of Nizhny Novgorod
   (Russia) and the Faculty of Medicine and Health Sciences of Ghent
   University (Belgium; ECD 22-12). The G*Power 3.1.5 software was used to
   determine the sample size for in vivo experiments. In some experiments,
   5–14 mice per group were used without the calculation of the power.

In vivo prophylactic tumor sub-cutaneous vaccination mouse model

   Cell death was induced in GL261 cells in vitro by PS-PDT or MTX as
   described above. Next, the GL261 cells were collected, washed once in
   PBS, and re-suspended at the desired cell density in PBS. Mice were
   inoculated subcutaneously with 5 × 10^5 dying GL261 cells or with PBS
   in the left flank. The mice were immunized once or twice with an
   interval of seven days. Eight days after the last vaccination, the mice
   were challenged subcutaneously on the opposite flank with 1 × 10^5 live
   GL261 cells. Tumor growth at the challenge site was monitored using a
   caliper for up to four weeks after the challenge. The mice were
   sacrificed when the tumors became necrotic or exceeded 2000 mm^3.

In vivo prophylactic tumor sub-cutaneous vaccination followed by orthotopic
intracranial challenge

   C57BL/6 J mice were immunized subcutaneously once or twice with an
   interval of 7 days with GL261 cells stimulated in vitro with PS-PDT or
   with MTX or subjected to F/T cycles as described above. Seven days
   after the last vaccination, the mice were anesthetized with isoflurane
   (5% induction, 1.5–2% maintenance), an incision was made in the scalp,
   and the skull was exposed. GL261 glioma cells (20,000 cells/3 μl
   saline) were injected into the brains at the following coordinates: AP:
   –2.0 mm, ML: –2.0 mm, DV: –3.3 mm (relative to bregma). The cells were
   injected at a rate of 0.3 µl/min via a Hamilton syringe mounted in a
   motorized stereotactic injector (World Precision Instruments). After
   injection, the needle was left in situ for 5 min and then removed
   slowly. The scalp was then sewn shut and analgesia was administered
   (Xylanite 0.02 mg/kg) (NITA-PHARM, Russia).

DC vaccination in an orthotopic glioma mouse model and pharmacological
inhibition of RORγt

DC vaccines

   DCs were isolated according to the protocol described above. Between
   days 8 and 10 of cultivation, cells were collected for co-culturing.
   GL261 tumor cells were stimulated as described above and incubated for
   24 h. The cells were then subjected to six cycles of freezing (–80 °C)
   and thawing (+55 °C). Total protein in the cell lysate was measured
   with a commercial BCA Protein Assay Kit (Sigma-Aldrich) and a Synergy
   MX spectrophotometer (BioTek Instruments Inc., USA). Two mg of protein
   was added to a suspension of 10 × 10^6 DCs for 90 min. To activate the
   DCs, they were treated with lipopolysaccharide (0.5 μg/ml) for 24 h. In
   some experiments, PBS or DCs co-cultured with GL261 glioma cell lysates
   subjected to several freeze/thaw cycles to induce accidental necrosis
   (without photoinduction) were used as controls.

Prophylactic protocol

   Female C57BL/6j mice (6–8 weeks old) were injected intraperitoneally
   twice seven days apart with a suspension containing 1 × 10^6 prepared
   DCs. Seven days after the last injection, 2 × 10^4 viable GL261 glioma
   cells were injected intracranially. All animals were anesthetized with
   a mixture of medical oxygen and isoflurane (induction: 5%; maintenance:
   2%) and immobilized in a stereotaxic frame. The injection was performed
   using a stereotactic device 2 mm lateral and 2 mm posterior to the
   bregma and 3 mm below the dura mater according to a previously
   described protocol [[255]14]. The skin was sutured, and meloxicam was
   administered subcutaneously (1 mg/kg, 2 mg/mL) to manage post-operative
   pain.

   To inhibit a Th17 cell response, GSK805 (10 mg/kg; InvivoChem)
   dissolved in 10% DMSO and 90% corn oil (Sigma Aldrich) or vehicle (10%
   DMSO and 90% corn oil) were administered intraperitoneally to mice 12,
   24, 48, and 72 h after injection of the DC vaccine.

Therapeutic protocol

   Female C57BL/6j mice (6–8 weeks old) were anesthetized with a mixture
   of medical oxygen and isoflurane (induction: 5%; maintenance: 2%) and
   intracranially injected with 2 × 10^4 viable GL261 glioma cells. The
   injection was performed using a stereotactic device 2 mm lateral and
   2 mm posterior to the bregma and 3 mm below the dura mater according to
   a previously described protocol [[256]14]. The skin was sutured, and
   meloxicam was administered subcutaneously (1 mg/kg, 2 mg/mL) to manage
   post-operative pain. To prepare the DC vaccine, cell death was induced
   in GL261 cells in vitro by PS-PDT or MTX as described above. The mice
   were injected intraperitoneally with a suspension containing 1 × 10^6
   of the prepared DCs (as described above) on days 2, 6, 10, and 17 after
   intracranial injection of viable GL261 cells. Local (inguinal and
   axillary) draining lymph nodes were collected on day 37 after
   intracranial injection with GL261 cells. The immune cells in the
   draining lymph nodes were stained by anti-CD8a (eBioscience,
   12-0081-81), anti-CD45 (Biolegend, 103125), mouse Fc block
   (eBioscience, 16-0161-85), anti-CD11b (Invitrogen, 12-0112-83) and
   anti-CD11c (BD Pharmingen, 561022) and analyzed on a BD FACS Canto II
   flow cytometer.

Neurological status assessment

   After intra-cranial inoculation with glioma Gl261 cells and/or DC
   vaccines, the mice were monitored three times per week and clinical
   symptoms were scored with a neurological deficit grading scale
   [[257]14]. The dynamics of the functional state of the central nervous
   system was evaluated on a scale to assess the severity of neurological
   deficit, with modifications for mice. The scale includes several tests
   of motor activity, coordination, reflexes, muscle tone, ptosis, and
   exophthalmos. Each test was scored 2 points for no reaction, 0 for
   good/normal reaction, and –1 for some disturbances. The values were
   summed up and interpreted as severe central nervous system damage
   (10─20 points), moderate damage (6─9 points), or light damage (1─5
   points). The neurological score was evaluated by a blinded
   investigator.

Magnetic resonance imaging

   To assess the dynamics of intracranial tumor growth in the prophylactic
   model, magnetic resonance imaging (MRI) was applied using a high-field
   magnetic resonance tomograph, Agilent Technologies DD2-400 9.4 T
   (400 MHz) with a volume coil M2M (Н1). The animals were kept under
   general anesthesia (0.2 mg Zoletil and 0.5 mg Xylanit, intramuscular)
   in a fixed position inside the magnet tunnel for 40 min. The VnmrJ
   program was used to obtain and process data. T1-tomograms of
   layer-by-layer frontal brain sections weighted by proton density were
   obtained using the multi gradient-echo multi slice (MGEMS) pulse
   sequence with the following parameters: TR = 1000 ms, TE = 1.49 ms, 6
   echoes, FOV 20 × 20 mm, matrix 128 × 128 and after −256 × 256, slice
   thickness 1 mm, 15 slices, 17 min and 4 s scanning time.

   To assess the dynamics of intracranial tumor growth in the therapeutic
   model, ex vivo MRI was performed on 7-T micro-MRI (PharmaScan 70/16,
   Bruker BioSpin, Ettlingen, Germany) as previous described [[258]67].

Immunohistochemical analysis in an orthotopic glioma mouse model

   In the prophylactic model, the mice were terminally euthanized and
   perfused with sodium chloride followed by 4% formalin. The brains were
   dissected and embedded in paraffin, and 10-µm sections were cut.
   Following antigen retrieval in citric acid buffer the sections were
   stained overnight at +4 °C with rabbit polyclonal anti-IL-17A antibody
   (ab 79056, Abcam, Cambridge, UK). As secondary antibodies, goat
   anti-rabbit IgG conjugated to AlexaFluor488 (A11034), Invitrogen were
   used.

   In the therapeutic model brains were rinsed three times in PBS. The
   brains were dissected and embedded in paraffin, and 10-µm sections were
   cut and stained with hematoxylin/eosin. The images were taken on a
   spinning disk confocal Nikon Ti2 fluorescence microscope (Nikon,
   Japan).

Public datasets

   The brain lower grade glioma (LGG) project dataset was downloaded from
   The Cancer Genome Atlas (TCGA) [[259]39]
   ([260]https://portal.gdc.cancer.gov/) in August 2022. Patients with no
   reported vital status, with recurrent tumor, or with an unknown
   survival time were excluded. There were 508 patients with the primary
   tumor, of whom 125 had died and the remaining 383 were alive. We used
   STAR-counts files containing the number of mapped reads for each gene.

   We also constructed a control group of 42 healthy postmortem brain
   transcriptome samples. To do this, we took the gene’s count numbers in
   healthy samples from publicly available datasets [261]GSE80336
   [[262]68] and [263]GSE78936 [[264]69] from the Gene Expression Omnibus
   (GEO) [[265]70] repository ([266]https://www.ncbi.nlm.nih.gov/geo/).

   As for the validation dataset for the prognostic model, we used samples
   with LGG from the Chinese Glioma Genome Atlas (CGGA) [[267]71]
   ([268]http://www.cgga.org.cn/). The RNA sequencing data presented as
   STAR-counts of two batches (mRNAseq_325 and mRNAseq_693) and
   corresponding clinical information of LGG samples were combined into a
   single CGGA-LGG dataset containing 408 samples (164 dead and 244
   alive). For additional verification, we also considered RNA-seq data of
   glioblastoma multiforme (GBM) from the TCGA-GBM dataset (151 samples:
   122 dead, 29 alive) and CGGA-GBM dataset (218 samples: 183 dead, 35
   alive), combining GBM samples from mRNAseq_325 and mRNAseq_693 CGGA
   batches.

   The sequence of TCGA, CGGA and GEO datasets were filtered to leave only
   protein-coding genes. In addition, to correct for a batch effect when
   combining data from different batches (datasets of healthy samples or
   mRNAseq_693 and mRNAseq_325 batches from CGGA) the ComBat-seq algorithm
   of the sva v3.42.0 software [[269]72, [270]73] R package was used.
   Finally, the expression count data were normalized by the transcripts
   per million (TPM) method and the normalized expression values were
   transformed to log[2] values.

Differential expression and functional annotation analysis

   Differential expression analysis was conducted in R software v4.1.2 and
   calculated using negative binomial generalized linear modeling
   implemented in the DESeq2 package v1.34.0 [[271]74]. Genes were
   considered differentially expressed when the q-value cutoff (FDR
   adjusted p-value using Benjamini–Hochberg mode) [[272]75] was <0.05. To
   identify genes with significant differential expression, we set the
   following selection criteria: (i) the absolute factor of change in
   expression between the groups is ≥2 (|Fold change| ≥ 2); (ii) the
   average of the normalized count values for all samples is >100 (base
   Mean > 100). To identify differentially expressed genes in the
   TCGA-LGG, [273]GSE80336 and [274]GSE78936 datasets, the second
   condition was relaxed (base Mean > 50) due to the presence of more
   genes with very low read count.

   Biological processes were analyzed using the PANTHER functional
   classification system ([275]http://www.pantherdb.org) [[276]55].

Gene expression correlation heatmaps

   To determine the co-expression relationships between genes, Pearson
   correlation between the expression profiles of a pair of genes was
   calculated using the pearsonr function from scipy.stats v1.7.3 Python
   package. The correlation values were computed using log[2](TPM + 1)
   normalized gene expression. For each resulting correlation matrix,
   heatmaps were built using a Heatmap function from the ComplexHeatmap
   v2.10.0 R Bioconductor package [[277]76].

Determination of metagene specific to the Th17 cells and its prognostic
efficacy

   To evaluate the prognostic efficacy of Th17 cells immune contexture in
   LGG patients, we considered Th17-signature [[278]40] and assessed the
   relationship between the expression of the Th17-associated metagene and
   overall survival of patients. The TCGA-LGG dataset was used to generate
   a correlation matrix of gene expression levels from the respective
   signature by estimating the Pearson’s correlation coefficients. The
   correlation matrix was subjected to hierarchical clustering (Euclidean
   distance, average linkage). The metagene associated with LGG-specific
   Th17 cells was chosen as a cluster of highly correlated genes that
   included the reliable Th17 cells marker (IL17A) [[279]42]. We then
   defined metagene expression as the average value of the expression of
   the genes composing a metagene and assessed the association of the
   metagene with overall survival. We stratified patients on the basis of
   the 75th percentile of metagene expression into two groups (high or low
   expression level). The resulting groups were plotted with respect to
   overall survival to produce respective Kaplan–Meier curves. Statistical
   comparison of survival by log-rank Mantel–Cox test was performed
   between groups. In addition, the median survival (in days) was
   calculated for each group. We calculated the percent change in median
   survival (%ΔMS) between the high and low metagene expression level
   groups, as previously reported [[280]14], using the formula
   [MATH: <mrow><mi>%</mi><mi
   mathvariant="normal">Δ</mi><mi>M</mi><mi>S</mi><mo>=</mo><mfrac><mrow><
   mi>M</mi><msup><mrow><mi>S</mi></mrow><mrow><mi>H</mi><mi>i</mi><mi>g</
   mi><mi>h</mi></mrow></msup><mo>−</mo><mi>M</mi><msup><mrow><mi>S</mi></
   mrow><mrow><mi>L</mi><mi>o</mi><mi>w</mi></mrow></msup></mrow><mrow><mi
   >M</mi><msup><mrow><mi>S</mi></mrow><mrow><mi>H</mi><mi>i</mi><mi>g</mi
   ><mi>h</mi></mrow></msup></mrow></mfrac><mo>×</mo><mn>100</mn></mrow>
   :MATH]
   , where MS^High is the median survival in the group with high
   expression level and MS^Low is the median survival in the group with
   low expression level.

Prognostic model construction

   A special feature of survival data is right censoring when the
   observation period expires before death occurs. In this case, the Cox
   proportional hazards regression model [[281]77] is the most common
   approach for studying the dependency of a patient’s survival time on
   several predictor variables.

   To identify prognostic genes that affect the survival of patients,
   univariate Cox proportional hazards regression analysis was performed
   using CoxPHFitter function from lifelines v0.26.0 Python package
   [[282]78]. This method enables the evaluation of the correlation
   between the expression level of each gene and overall survival in the
   cohort. The Wald statistic is used to estimate the statistical
   significance for each of the covariates in relation to overall
   survival. Only those genes with a p-value < 0.05 were considered as
   significant predictors and entered into the final multivariate Cox
   regression model. For evaluating the performance of the prediction
   model, the index of concordance (C-index) was calculated, which is a
   generalization of the receiver operating characteristic area under
   curve to survival data that include censored data. The C-index values
   range from 0 to 1, and the larger value, the better the prediction
   [[283]79].

   Next, we calculated the individual risk score of each patient with
   coefficient-weighted gene expression and constructed a predictive model
   with the following formula:
   [MATH: <mrow><mi>R</mi><mi>i</mi><mi>s</mi><mi>k</mi><mspace
   width="0.25em"></mspace><mi>S</mi><mi>c</mi><mi>o</mi><mi>r</mi><mi>e</
   mi><mo>=</mo><munderover accent="false"
   accentunder="false"><mrow><mo>∑</mo></mrow><mrow><mi>i</mi><mo>=</mo><m
   n>1</mn></mrow><mrow><mi>k</mi></mrow></munderover><mfenced close=")"
   open="("><mrow><mi>C</mi><mi>o</mi><mi>e</mi><msub><mrow><mi>f</mi></mr
   ow><mrow><mi>i</mi></mrow></msub><mo>×</mo><mi>E</mi><msub><mrow><mi>V<
   /mi></mrow><mrow><mi>i</mi></mrow></msub></mrow></mfenced><mo>,</mo></m
   row> :MATH]

   where k is the number of prognostic genes, Coef[i] is the coefficient
   of the i-th gene in the multivariate Cox regression model, and EV[i] is
   the log[2](TPM + 1) normalized expression value of the i-th gene. If
   Coef[i] > 0 the i-th gene is defined as a high-risk signature, and if
   it is < 0, the gene is defined as protective. The patients were divided
   into high-risk and low-risk groups according to the median risk score
   calculated based on the prognostic gene signature.

   To compare the differences in overall survival time between the
   low-risk and high-risk patient groups, survival curves were constructed
   using the Kaplan-Meier method, and the log-rank test was employed to
   assess the statistical significance of the difference. For this,
   KaplanMeierFitter and logrank_test functions from lifelines Python
   package were used. The sensitivity and specificity of the prognostic
   model for predicting the clinical outcome were evaluated by calculating
   the area under curve of the time-dependent receiver operating
   characteristic curve using survivalROC function from survivalROC
   v1.0.3 R package [[284]80]. Receiver operating characteristic curves
   are widely used for presenting the sensitivity and specificity of
   continuous diagnostic markers for a binary disease outcome. This
   approach estimates how well the risk score can distinguish those who
   had an event (died) by a pre-specified time (e.g., 1, 3, 5 years) from
   those who remained alive.

Immune-deconvolution: estimation of glioma-infiltrating cells

   Immune cell proportions in tissue were estimated using EPIC (Estimating
   the Proportions of Immune and Cancer cells) deconvolution method
   [[285]81]. The EPIC method estimates the fraction of five types of
   immune cells (B cells, CD4^+ T cells, CD8^+ T cells, macrophages and NK
   cells), cancer-associated fibroblasts (CAFs), endothelial cells and
   uncharacterized cells (mostly cancer cells) from bulk gene expression
   data. The method is based on the expression profiles of the reference
   genes for the cell types under consideration and predicts the
   proportion of these cells and the remaining uncharacterized cells for
   which no reference profile is given, using constrained least squares
   regression. Although EPIC is more limited in the number of tested
   immune cell types compared to other methods [[286]82, [287]83], it
   allows for the quantification of non-immune cell types such as CAFs and
   endothelial cells and was also specially designed for RNA-seq data, not
   microarray data.

   The TPM normalized expression data of the TCGA-LGG dataset were used
   for estimation of the fractions of cell types in the tumor for each
   individual, which were deconvoluted by the EPIC v1.1.5 R package
   ([288]https://github.com/GfellerLab/EPIC) [[289]81]. The correlation
   between risk score of prognostic signature and cell infiltration was
   calculated using Pearson’s correlation.

Statistical Analysis

   Statistics were calculated in GraphPad Prism (V.9.2). The samples or
   mice have never been excluded from the analysis. The method of
   randomisation has not been used in the manuscript. The results of the
   phagocytosis assay and DC activation and maturation assay were analyzed
   by two-way ANOVA with Tukey’s multiple comparisons test. Kaplan–Meier
   survival curves show the timeline of tumor development. Survival in the
   low-risk and high-risk groups was analyzed by log-rank Mantel–Cox test.
   The similarity of the variance between the samples in the large groups
   has been pre-checked with the Levene test.

Supplementary information

   [290]Legend for the suppl. figures^ (17.1KB, docx)
   [291]Suppl.Figure 1^ (395.9KB, pptx)
   [292]Suppl.Figure 2^ (519.5KB, pptx)
   [293]Suppl.Figure 3^ (40.4MB, pptx)
   [294]Suppl.Figure 4^ (392.9KB, pptx)
   [295]checklist^ (665KB, pdf)

Acknowledgements