Targeting the mesenchymal subtype in glioblastoma and other cancers via inhibition of diacylglycerol kinase alpha

Abstract

Background

The mesenchymal phenotype in glioblastoma (GBM) and other cancers drives aggressiveness and treatment resistance, leading to therapeutic failure and recurrence of disease. Currently, there is no successful treatment option available against the mesenchymal phenotype.

Methods

We classified patient-derived GBM stem cell lines into 3 subtypes: proneural, mesenchymal, and other/classical. Each subtype’s response to the inhibition of diacylglycerol kinase alpha (DGKα) was compared both in vitro and in vivo. RhoA activation, liposome binding, immunoblot, and kinase assays were utilized to elucidate the novel link between DGKα and geranylgeranyltransferase I (GGTase I).

Results

Here we show that inhibition of DGKα with a small-molecule inhibitor, ritanserin, or RNA interference preferentially targets the mesenchymal subtype of GBM. We show that the mesenchymal phenotype creates the sensitivity to DGKα inhibition; shifting GBM cells from the proneural to the mesenchymal subtype increases ritanserin activity, with similar effects in epithelial-mesenchymal transition models of lung and pancreatic carcinoma. This enhanced sensitivity of mesenchymal cancer cells to ritanserin is through inhibition of GGTase I and downstream mediators previously associated with the mesenchymal cancer phenotype, including RhoA and nuclear factor-kappaB. DGKα inhibition is synergistic with both radiation and imatinib, a drug preferentially affecting proneural GBM.

Conclusions

Our findings demonstrate that a DGKα–GGTase I pathway can be targeted to combat the treatment-resistant mesenchymal cancer phenotype. Combining therapies with greater activity against each GBM subtype may represent a viable therapeutic option against GBM.

Importance of the study.

In this study, we demonstrate for the first time a clinically viable agent with preferential activity against the aggressive and treatment-resistant mesenchymal GBM subtype. We show that DGKα inhibition with ritanserin, a serotonin receptor inhibitor we repurpose as a novel DGKα inhibitor, has clearly superior activity against the mesenchymal phenotype in GBM and other cancers. Mechanistically, we identified an interaction between DGKα and geranylgeranyltransferase I leading to increased preferential activity against the mesenchymal subtype. Furthermore, we demonstrated that ritanserin treatment induced mesenchymal-proneural transition and—combining ritanserin with radiation or imatinib, agents with higher activity against proneural GBM—showed synergy. Our results suggest combined targeting of GBM subtypes as a response to cancer heterogeneity.

Glioblastoma (GBM) is the most common aggressive brain tumor, with an extremely grim prognosis.¹ Standard therapy, consisting of surgery and the combination of temozolomide chemotherapy and radiation, offers only 15–18 months of median survival from initial diagnosis.² GBM is a highly heterogeneous tumor due to the diversity of underlying genotypes and to pronounced adaptability to standard or targeted therapies.³ Recent gene expression studies have shown that GBM can be classified into 3 subtypes: classical, proneural, and mesenchymal.^4,5 In line with this, recent studies have focused on identifying key signaling drivers of the subtypes and subtype-targeted therapies.^6–8

In epithelial tumors, cancer cells acquire mesenchymal features through epithelial-mesenchymal transition (EMT), marked by treatment resistance, invasiveness, and increased metastasis.⁹ Recent evidence suggests that upon treatment, proneural GBM cells acquire mesenchymal features as an escape/resistance mechanism in a manner analogous to the EMT process^6,10 and the similarly dubbed proneural-mesenchymal transition (PMT). Acquisition of mesenchymal features in GBM confers aggressiveness and treatment resistance—pointing to the importance of efficiently targeting the mesenchymal phenotype. In keeping with this, it has also been shown that GBM originally presenting as primarily another subtype can recur with a mesenchymal phenotype, resulting in worse prognosis.⁵ Overall, conversion to a mesenchymal phenotype may represent one of the most common and powerful mechanisms through which GBM and other cancers survive therapy.

Given the significance of the mesenchymal GBM subtype, a number of studies have recently been done to understand its molecular mechanisms of treatment resistance, potential drivers, and markers.^6,8,10 Despite their significant contributions to our understanding of the mesenchymal GBM subtype, these studies have not yet yielded agents that can be translated easily into the clinic. At present there is no viable therapeutic approach to the mesenchymal cancer phenotype, and this represents a critical unmet need in oncology.

Our group and others have shown that the diacylglycerol kinase alpha (DGKα) pathway has extensive interactions with key signaling pathways/proteins and that it can be a potential therapeutic target in the treatment of cancer.^11,12 In this study we identify inhibition of DGKα as a preferential vulnerability for the mesenchymal phenotype in GBM and other cancers. From a mechanistic perspective, we demonstrate for the first time that DGKα inhibition results in inhibition of geranylgeranyltransferase I (GGTase I), which in turn inhibits Ras homolog gene family, member A (RhoA) and nuclear factor-kappaB (NF-κB), which are preferential dependencies of the mesenchymal phenotype. We also demonstrate synergistic combinations of DGKα inhibition with radiation and imatinib, therapies proven to be more active against proneural GBM, suggesting the potential of combination regimens targeting distinct GBM subtypes. This approach—using combinations that effectively target the very limited number of cancer phenotypes—may offer an alternative to personalized oncology strategies targeting individual genetic lesions, which have been hampered by issues such as extreme genetic variability and adaptive switching of driver mutations.

Materials and Methods

For detailed experimental procedures, see the Supple mentary material.

Cell Culture

Established and previously validated GBM stem cell (GSC) lines were cultured as previously described.^6,13 At the beginning of this study, all cell lines were tested negative for mycoplasma contamination by PCR, verified as human cells with short tandem repeat profiling, and classified as mesenchymal, proneural, and classical as previously described.⁶

Animal Studies

All mouse studies were approved by the Institutional Animal Care and Use Committee at the University of Virginia. Five thousand mesenchymal (G2 and G88), 500000 classical (G528), or 500000 proneural (G464) GSCs were stereotactically injected into the striatum of 6- to 8-week-old female BALB/c SCID NCr mice. Beginning 6 days after surgery, 100 µL of ritanserin (10 mg/mL in corn oil) was given once daily via oral gavage.

Preparation of Liposomes

The preparation of liposomes followed a previously reported protocol from MacDonald et al.¹⁴ To measure DGKα activity, phosphatidylcholine (PC), diacylglycerol, and phosphatidylserine (Avanti Polar Lipids) were dissolved in CHCl₃, combined, and dried in vacuo to remove all solvent. The surface concentration of lipids was as follows: 5 mol% diacylglycerol, 40 mol% phosphatidylserine, and 55 mol% PC. The lipids were hydrated to 19 mM in 50 mM (3-(N-morpholino)propanesulfonic acid) (MOPS), pH 7.5, 100 mM NaCl, and 5 mM MgCl₂ (buffer B) and were subjected to 5 freeze-thaw cycles in liquid nitrogen, followed by extrusion through a 100 nm polycarbonate filter 11 times.

To measure GGTase I binding, we combined PC, phosphatidic acid (PA), and phosphatidylethanolamine (PE) in CHCl₃ as above. The surface concentrations of lipids were as follows: 100 mol% PC, 80 mol% PE, and 20 mol% PA, or 50 mol% PC, 20 mol% PA, and 30 mol% PE. The lipids were hydrated to 1 mM in 20 mM Tris-HCl, pH 7.4, 150 mM NaCl (buffer C). Liposomes were prepared as described above.

Liposome Binding

Binding of GGTase I to 100 nm liposomes was measured using a modified version of the Höfer et al method. ¹⁵ Briefly, green fluorescent protein–tagged GTTase I was incubated for 30 min at 30°C with liposomes comprising the indicated phospholipid concentrations as well as 0.1 mol% pyrene-PC, then mixed with an equal volume of 80% (w/v) sucrose in buffer C. This mixture was overlaid onto 270 µL of 80% (w/v) sucrose in buffer C in 5 × 41 mm Beckman tubes, followed by 150 µL of 20% (w/v) and 100 µL of buffer C. The gradient was centrifuged in an SW 55-Ti swinging bucket rotor containing nylon inserts at 240000 × g for 1 h. The top 100 µL fraction was collected with a micropipette. Pyrene emission was measured to correct each sample for the percent liposome recovery after flotation and green fluorescent protein emission was measured to determine the fraction of GGTase I bound to the liposomes. The binding of GGTase I to PC:PA and PC liposomes was normalized to the binding to PA:PE:PC liposomes. The binding assay was performed 3 times.

Soft Agar Colony Formation and In Vitro Radiation

A total of 3 × 10⁴ cells per well were seeded in a 24-well plate in 0.3 mL of 0.4% agar medium over 0.5 mL of 0.8% agar medium. Dimethyl sulfoxide or ritanserin was added to the top agar at the desired concentration. After top agar was dried, liquid medium (1 mL) was added over the top agar with the same concentration as the top agar. Ritanserin was replaced every 5 days. After 22 days, medium was removed and the colonies were stained with 0.005% crystal violet for more than one hour. Pictures of colonies were taken at 0.5 × 10 magnification and the number of colonies determined with the ImageJ program. The data are given as mean ± SE of 3 independent wells. Cells were irradiated using a SARRP (small animal radiation research platform) (Xstrahl Life Sciences).

EMT, PMT, Mesenchymal-Proneural Transition, and Quantitative Real-Time PCR

Three-dimensional multicellular spheroid cultures were generated and induced to undergo EMT with exposure to tumor necrosis factor (TNF)-α and transforming growth factor-β. PMT was induced by treating proneural GSCs with TNF-α. In an assay testing for mesenchymal-proneural transition (MPT), mesenchymal GSCs were treated with ritanserin over 5 days. Following treatment, quantitative real-time PCR was performed to verify transformation. Primer sequences can be found in the Supplementary material.

Immunoblotting, Small Interfering RNA Transfection, and PA Rescue Assay

Immunoblotting was performed as previously described.¹⁶ Lipofectamine RNAiMAX transfection reagent (#13778150, Thermo Fisher Scientific) was utilized for small interfering (si)RNA transfection according to the producer’s instructions with final siRNA concentration of 10 nmol/L. DGKA siRNAs were as follows: (i) custom DGKA siRNA: 5′-GGAUUGACCCUGUUCCUAA-3′, (ii) Dharmacon SMARTpool ON-TARGET plus (L-006711-00-0005). The data using the custom DGKA siRNA was generated with double transfection of GSCs 3 days apart. PA solution was prepared as described before.¹⁷ One hundred micromolar of PA was added to the respective wells twice daily for 4 days.

Luciferase Reporter Assay, RhoA Activation Assay, and Caspase-3/7 Assay

NF-κB luciferase reporter and control empty vectors were transfected into GSCs using Fugene HD (Promega) according to the manufacturer’s instructions. Forty-eight hours after transfection, luciferase activity was measured with the Dual-Luciferase Reporter assay system kit (Promega) and a Promega GloMax 20/20 luminometer. RhoA activity was measured with the G-LISA RhoA activation assay kit (Cytoskeleton) per the manufacturer’s instructions. Caspase-3/7 assay was performed using the Caspase-Glo 3/7 Assay kit (G8090, Promega) according to the manufacturer’s instructions following 3 days of ritanserin treatment.

Statistics

GraphPad Prism 6 and CompuSyn (ComboSyn) were utilized for statistical and synergy analyses. Student’s t-test was used for 2-group comparisons. For multiple comparisons, both 1-way ANOVA with post-hoc Tukey analysis and 2-way ANOVA with Bonferroni posttest were utilized. P-values less than 0.05 were considered significant using an error rate α = 0.05. For synergy, the Chou–Talalay method was used to generate combination indices (CIs). Kaplan–Meier analysis was used for mouse survival curve generation using Prism software.

Results

A Novel DGKα Inhibitor, Ritanserin, Preferentially Targets the Mesenchymal Cancer Phenotype

We recently reported the abandoned serotonin reporter ritanserin to be a novel DGKα inhibitor.¹⁸ Ritanserin has been found to be safe and well tolerated in several clinical trials, with favorable pharmacologic features, including a 40-hour plasma half-life and blood–brain barrier penetration.¹⁹ In pharmacokinetic studies of mice, even a single low oral ritanserin dose (10 mg/kg) in corn oil was able to reach micromolar concentrations in the brain (Supplementary Figure S1 and Supplementary Table S1). Given its favorable properties and prior clinical trial history, we employed ritanserin as a DGKα inhibitor for subsequent studies.

With data from The Cancer Genome Atlas we determined that the DGKα mRNA expression level is significantly higher in the mesenchymal GBM subtype compared with other GBM subtypes (Fig. 1A), though higher DGKα expression correlates with improved survival in this subtype (data not shown). This prompted us to assess whether DGKα played a greater functional role in mesenchymal GBM. We showed with a self-renewal assay that a mesenchymal GSC line—subtyped using a previously reported approach⁶—was significantly more sensitive to ritanserin than a non-mesenchymal GSC line (Supplementary Figure S2A and S2B). We then extended our findings by comparing sensitivity in 4 mesenchymal GSC lines versus 5 non-mesenchymal lines. DGKα inhibition with ritanserin revealed markedly different sensitivity across subtypes, with a roughly 2- to 3-fold decrease in half-maximal inhibitory concentration of mesenchymal lines versus non-mesenchymal lines (Fig. 1B). Our results with a caspase-3/7 assay and significant reversal of ritanserin-induced cytotoxicity with a pan-caspase inhibitor suggested that DGKα inhibition triggered cell death partially through induction of apoptosis (Supplementary Figure S3A and S3B). Testing whether the higher sensitivity of mesenchymal lines was mainly due to DGKα inhibition, we recapitulated our findings by knocking down DGKα expression with 2 different siRNAs (Fig. 1C and Supplementary Figure S4A and S4B). To further demonstrate that ritanserin’s effects were due to DGKα inhibition and not serotonin receptor inhibition, we showed that the 5-hydroxytryptamine 2 receptor antagonist ketanserin, which does not inhibit DGKα, caused minimal cytotoxicity even at very high doses (Supplementary Figure S5); furthermore, a serotonin receptor agonist, α-methyl-5-hydroxytryptamine maleate, also failed to rescue ritanserin’s cytotoxicity (data not shown). As a third line of evidence that ritanserin’s effects are through DGKα inhibition and to confirm that PA is an important mediator, we performed a PA rescue assay showing that exogenous PA could largely reverse ritanserin-induced cytotoxicity in a mesenchymal GSC line (Fig. 1D). In an orthotopic GBM model, once-daily oral ritanserin (50 mg/kg) increased survival significantly in mice with established intracranial xenografts of a mesenchymal GSC line, including long-term survivors. In contrast, there was no in vivo survival benefit with a non-mesenchymal GSC line (Fig. 1E).

Fig. 1.

DGKα inhibition preferentially targets the mesenchymal cancer phenotype. (A) Comparison of DGKα mRNA levels across glioblastoma subtypes: classical subtype (blue), mesenchymal subtype (red), proneural subtype (purple), and neural subtype (green), with significantly higher expression in the MES subtype. The neural subtype is now believed to represent excessive contamination with normal brain. Figure was taken from the GBM Bio Discovery Portal (https://gbm-biodp.nci.nih.gov/), with data from TCGA. (B) Dose-response curve showing higher sensitivity of mesenchymal GSC lines to 5 days of ritanserin treatment versus non-mesenchymal lines. (C) DGKA siRNA exhibits greater cytotoxicity against mesenchymal GSC lines versus non-mesenchymal lines (**P < 0.001; one-way ANOVA with post-hoc Tukey analysis). (D) In a PA rescue assay, the addition of PA (100 µM) to ritanserin (5 µM) treatment reverses most of the ritanserin-induced cytotoxicity in mesenchymal GSC lines. Cell viability was detected at 96 hours of treatment (***P < 0.0001; one-way ANOVA with post-hoc Tukey analysis. Values are mean ± SEM of triplicates). (E) Kaplan–Meier survival curves of mouse xenografts with MES and non-MES GSC lines treated with vehicle vs ritanserin (50 mg/kg via oral gavage), showing prolonged survival only with MES GSC xenografts (*P < 0.05, n = 8 mice per cohort). Ctrl: vehicle, Rit: ritanserin, NS: nonsignificant, MES: mesenchymal. All values are mean ± SEM of triplicates. Cell viabilities were determined via cell counts following 5 days of treatment. Each experiment was performed at least 3 times using separate samples.

Acquisition of the Mesenchymal Phenotype Sensitizes Cancer Cells to Ritanserin Treatment

Short-term in vitro TNF-α treatment of proneural GSCs has been shown to induce PMT, leading to activation of NF-κB and subsequent resistance to radiation.¹⁰ Utilizing this GBM model of PMT, we treated 2 proneural GSC lines with TNF-α in vitro and demonstrated upregulation of mesenchymal markers (Fig. 2A). We subsequently showed that these transformed cells became more sensitive to ritanserin treatment (Fig. 2B). Additionally, using established EMT models we demonstrated that both lung (A549) and pancreatic (MPanc-96) carcinoma cells became more sensitive to ritanserin post-EMT (Fig. 2C).

Fig. 2.

Acquisition of the mesenchymal phenotype sensitizes cancer cells to ritanserin treatment. (A) Following 5-day TNF-α exposure, 2 proneural GSC lines expressed higher mRNA levels of mesenchymal markers (*P < 0.05; ***P < 0.001; 2-tailed t-test). (B) The proneural lines became more sensitive to ritanserin treatment (8 µM) after 5-day TNF-α exposure (***P < 0.001; 2-way ANOVA with Bonferroni correction). (C) Lung and pancreatic carcinoma cells became more sensitive to ritanserin treatment (15 µM and 10 µM, respectively) after EMT induction with TNF-α and transforming growth factor-β (**P < 0.01; ***P < 0.0001; 2-way ANOVA with Bonferroni correction). All values are mean ± SEM of triplicates. Each experiment was performed at least 3 times using separate samples. PN: proneural.

Mesenchymal Preference of DGKα Inhibition Acts Through Inhibition of GGTase I and Its Downstream Mediators

Rho family kinases are known to play a role in the mesenchymal cancer phenotype, in part through activation of NF-κB.^20–22 Given prior reports indicating regulatory roles of other DGKs on RhoA function,²³ we tested whether DGKα inhibition affected RhoA in the mesenchymal subtype of GBM. We found that ritanserin largely ablated RhoA activation, which was largely rescued with DGKα overexpression (Fig. 3A). We assessed whether DGKα might be acting upstream, noting that RhoA protein needs to undergo posttranslational modification through geranylgeranylation by GGTase I.²⁴ Based on the links between DGKα and the targets of GGTase I, as well as both being localized on the membrane, we hypothesized a connection between DGKα and GGTase I. DGKα inhibition indeed resulted in diminished geranylgeranylation of proteins, as indicated by an increase in ungeranylgeranylated Rap1A protein with unchanged total Rap1 levels (Fig. 3B). To verify ritanserin effects in vivo, we demonstrated suppression of prenylation of the GGTase I target Rap1a in tumors in ritanserin-treated mice (Fig. 3C). Further supporting DGKα regulation of GGTase I, we demonstrated that GGTase I shows increased association with liposomes containing the DGKα product PA (Fig. 3D). We also demonstrated that protein farnesylation is not affected by DGKα inhibition (Supplementary Figure S6), suggesting specific action on protein geranylgeranylation. Since RhoA is an important downstream mediator of lysophosphatidic acid (LPA), we demonstrated partial rescue of cytotoxicity from ritanserin and a GGTase I inhibitor (GGTI-298) with exogenous LPA (Supplementary Figure S7). Exogenous mevalonate failed to rescue ritanserin-induced cytotoxicity, indicating that upstream regulators of GGTase I such as 3-hydroxy-3-methylglutaryl-coenzyme A reductase are not involved (Supplementary Figure S8). Supporting GGTase I as a mediator of DGKα, GGTI-298 also showed mesenchymal preferential cytotoxicity (Fig. 3E). One of the major drivers of the mesenchymal phenotype, NF-κB, has been particularly implicated in GBM resistance to conventional therapy. Given that GGTase I–dependent Rho family kinases activate NF-κB,²¹ we tested both ritanserin and GGTI-298 against an NF-κB activity reporter and noted comparable levels of inhibition with both agents (Fig. 3F and 3G). Overall, we have shown for the first time a direct connection between DGKα and GGTase I, and these results suggest that DGKα inhibition preferentially targets the mesenchymal phenotype at least in part through inhibition of GGTase I.

Fig. 3.

The mesenchymal preference of DGKα inhibition acts through inhibition of GGTase I and its downstream mediators. (A) Three days of ritanserin (4 µM) treatment suppresses RhoA activation detected with an enzyme-linked immunosorbent–based assay, with activity largely rescued by a DGKα expression plasmid (***P < 0.0001; **P < 0.01, one-way ANOVA with post-hoc Tukey analysis. Values are mean ± SEM of triplicates). (B) Forty-eight hours of ritanserin treatment blocks geranylgeranylation of Rap1A. Shown is an immunoblot using antibodies specific for unprenylated Rap 1A and total Rap 1. A commercially available GGTase I inhibitor (GGTI-298) was used as positive control. (C) Ritanserin suppresses the DGKα–GGTase I pathway in vivo. Shown is an immunoblot for unprenylated Rap 1A and total Rap 1 with control and ritanserin-treated mouse brains. (D) GGTase I binds PA-containing liposomes comparably to lipin 1, a known PA-binding enzyme. The normalized relative binding of purified GGTase I and lipin 1 to liposomes containing phosphatidylcholine (PC) alone or PC + PA (**P < 0.005; one-way ANOVA with post-hoc Tukey analysis. Plot shows mean ± SEM of 3 repeats). (E) A GGTase I inhibitor (GGTI-298, 5 µM) also exhibits greater cytotoxicity against mesenchymal GSC lines versus non-mesenchymal lines (***P < 0.0001, one-way ANOVA with post-hoc Tukey analysis. Values are mean ± SEM of triplicates). (F and G) Both ritanserin (4 µM) and GGTI-298 (5 µM) suppress NF-κB activity detected with a luciferase reporter assay at 48 hours of treatment (***P < 0.0001; 2-tailed t-test. Values are mean ± SEM of triplicates). Each experiment was performed at least 3 times using separate samples.

DGKα Inhibition Synergizes with Radiation In Vitro Through Inhibition of the DNA Damage Response

Radiation therapy is a key component of GBM treatment. Following initial radiotherapy, GBM has been found to develop resistance at least in part through acquisition of mesenchymal features—with NF-κB one of the major drivers of radioresistance.^6,10 We therefore tested the DGKα inhibitor ritanserin as a radiosensitizer. We showed that the combination of ritanserin with radiation exhibited significant synergy against GSCs in vitro (Fig. 4A and 4B; note that the response after a single ritanserin dose is greater at 5 days, as in Fig. 4B, than at 3 days, as in Fig. 4A). Studies have shown that the relative radioresistance of GSCs stems from a more pronounced DNA damage response (DDR) than that of differentiated GBM cells, as well as NF-κB being a crucial regulator of DDR pathways.^25,26 We further evaluated downstream mediators for the radiosensitizing effect of ritanserin, finding that DGKα inhibition diminishes activation of DDR mediators such as ataxia telangiectasia mutated kinase and checkpoint 1 and checkpoint 2 kinase post radiation (Fig. 4C). Our results suggest that DGKα inhibition may sensitize to radiation through inhibition of NF-κB–driven DDR, with the combination of radiotherapy and ritanserin having therapeutic potential.

Fig. 4.

DGKα inhibition synergizes with radiation in vitro through inhibition of the DNA damage response. (A) The bar graph shows synergy with the combination of ritanserin and radiation treatments. Using adherent cultures of GSC lines G88, G1005, and G816, ritanserin treatment was started 24 hours before radiation, and cell viability was assayed 48 hours after radiation treatment (***P < 0.001; one-way ANOVA with post-hoc Tukey analysis. Values are mean ± SEM of triplicates. Combination index (CI) values calculated with the Chou–Talalay method are 0.33, 0.53, and 0.4 for G88, G1005, and G816, respectively). (B) Soft agar colony formation showing synergy combining ritanserin (1.25 µM) and radiation (3 Gy) treatments, both beginning on the same day, in the mesenchymal G88 line (***P < 0.001; one-way ANOVA with post-hoc Tukey analysis. Values are mean ± SD of triplicates. CI value is 0.2). (C) Ritanserin blocks the activation of the DNA damage response upon radiation treatment. Shown is an immunoblot for key proteins activated upon radiation treatment as part of the DNA repair process. Cells were treated with ritanserin (4 µM) for 2 days prior to radiation (3 Gy) treatment and lysed 3 hours after radiation. XRT: radiation. Each experiment was performed at least 3 times using separate samples.

DGKα Inhibition Induces the Acquisition of Proneural Features and Synergizes with a Drug with Preferential Anti-Proneural Activity

Recent reports have identified a shift between GBM subtypes as a major mechanism of treatment resistance.^6,10 Given the preferential anti-mesenchymal activity of DGKα inhibition, we hypothesized that such a subtype shift might play a role in GBM resistance to ritanserin. We noted that after 5 days of ritanserin treatment, mesenchymal GSCs showed increased expression of certain proneural markers and decreased expression of some mesenchymal markers (Fig. 5A), suggesting potential MPT as a drug resistance mechanism. Activation of the platelet derived growth factor receptor alpha (PDGFRα) pathway is another hallmark of the proneural GBM subtype,^4,5 and we identified greater PDGFRα activation with ritanserin treatment (Fig. 5B). We subsequently combined ritanserin with imatinib—an inhibitor of PDGFRα and other receptor tyrosine kinases found to have preferential activity against proneural GBM²⁷—against mesenchymal GSCs in vitro and identified significant synergy (Fig. 5C). Our results and those of others underscore that GBM may resist therapies through shifting from one subtype to another, and we propose combining therapies with greater activity against each subtype. A schematic incorporating this hypothesis and the other findings of this study is shown in Fig. 6.

Fig. 5.

DGKα inhibition induces the acquisition of proneural features and synergizes with a drug with preferential anti-proneural activity. (A) Two mesenchymal GSC lines markedly increased expression of proneural markers following 5 days of ritanserin treatment (3 µM) (*P < 0.05; Student’s t-test). (B) Immunoblot of 2 mesenchymal GSC lines shows the activation of PDGFRα following 4 days of ritanserin treatment (3 µM). (C) The bar graph shows the synergy between ritanserin (2 µM) and imatinib (0.75 µM) following 5 days of treatment (**P < 0.005; ***P < 0.0001; one-way ANOVA with post-hoc Tukey analysis. Values are mean ± SEM of triplicates. CI values for G20 and G88 are 0.44 and 0.25, respectively). Each experiment was performed at least 3 times using separate samples.

Fig. 6.

Schematic depicting the DGKα–GGTase I pathway.

Discussion

Given the association of the mesenchymal GBM subtype with aggressive behavior and treatment resistance, identifying easily translatable approaches to target it is a key therapeutic goal in neuro-oncology. In fact, a recent report indicates that PMT is a critical pan-resistance mechanism in glioblastoma.²⁸ Prior efforts to characterize and find therapeutic leverage against the mesenchymal GBM subtype have made significant progress but have left important gaps. An earlier report identified the transcription factors signal transducer and activator of transcription 3 and CCAAT-enhancer binding homologous protein beta as mesenchymal GBM drivers.⁸ NF-κB signaling has also been implicated in GBM PMT and radioresistance, in keeping with prior studies on the role of NF-κB in mesenchymal cancers.¹⁰ We have previously described increased glycolysis and MLK4 expression as vulnerabilities in mesenchymal GBM, with in vitro demonstration of preferential effects on mesenchymal GBM lines.⁶ While these reports have contributed to our understanding of the mesenchymal GBM subtype, none have demonstrated in vivo efficacy and yielded therapies that can be translated easily to the clinic. This work demonstrates for the first time a clinically viable agent with preferential anti-mesenchymal GBM activity in vitro and in vivo. We show that DGKα inhibition with either ritanserin or RNA interference has clearly superior activity against mesenchymal GBM stem cell lines versus non-mesenchymal GSC lines. Underscoring mesenchymal targeting through DGKα inhibition, we demonstrate that driving PMT in GSC lines increases their sensitivity to ritanserin. Importantly, this report also shows the same for 2 non-GBM lines; enforced EMT in lung and pancreatic cancer lines increases their sensitivity to ritanserin. This suggests the therapeutic potential of DGKα inhibition to target not only the mesenchymal GBM subtype but also the treatment-resistant post-EMT mesenchymal phenotype in many other cancers, with much broader implications.

While this work represents the first findings of the preferential anti-mesenchymal cancer activity of DGKα inhibition, a number of other reports have indicated possibly associated roles for DGKα in cancer. Torres-Ayuso and colleagues have shown that DGKα expression correlates with treatment resistance in cancer, which may connect to our own findings given the relationship of the mesenchymal cancer phenotype to treatment resistance.²⁹ Another report shows a role for DGKα in cancer invasiveness, a characteristic also linked to the mesenchymal phenotype.³⁰ Others have connected DGK family members to RhoA and Rac,^23,30 which our work suggests might be secondary to upstream interaction of DGKα/PA with GGTase I. DGKα was previously linked to NF-κB in melanoma, but a mechanism was not established.³¹ Our prior report on DGKα in cancer described mammalian target of rapamycin and hypoxia-inducible factor 1α as critical mediators of the role of DGKα in cancer, but in light of our new findings we now hypothesize that those pathways are linked to baseline toxicity of DGKα inhibition in cancer—while the DGKα–PA–GGTase I–Rho/Rac–NF-κB pathway is responsible for the additional preferential toxicity to mesenchymal GBM and other cancers.

We demonstrated that the DGKα product PA binds to GGTase I and that GGTase I inhibition also has an anti-mesenchymal preference, indicating a DGKα–GGTase I mechanism for the mesenchymal activity. It is perhaps unsurprising that regulation of GGTase I is a mediator of the preferential anti-mesenchymal activity of DGKα inhibition, given that downstream targets of GGTase I such as RhoA, cell control protein 42 homolog, and others have previously been associated with EMT, radioresistance, and activation of NF-κB.^20,21,32 Further studies should also assess whether other DGK family members or other classes of PA-synthesizing enzymes regulate GGTase I and interact with the mesenchymal cancer phenotype. GGTase I inhibitors are also in testing and could potentially be redirected as anti-mesenchymal cancer agents; a selective GGTase I inhibitor, GGTI-2418, has been in phase I trials but has not been reported in subsequent trials of efficacy in patients with cancer. It remains to be seen whether DGKα inhibition or GGTase I inhibition is a better approach against mesenchymal cancers, but DGKα inhibition has other anticancer effects that may also prove beneficial. We have previously reported anti-angiogenic activity of DGKα inhibition, and a patent identified ritanserin among the most effective anti-angiogenic agents in a drug screen¹¹ (US PTO #8,367,737). In addition, DGKα inhibition is receiving increasing attention for its potential to upregulate T cell activity, break cancer-induced T cell anergy, and cooperate with cancer immunotherapies.³³

Repurposing ritanserin offers a potentially rapid route to the clinic for DGKα inhibition, but there are potential drawbacks. Ritanserin is nonspecific, with potent inhibition of serotonin receptors and potentially other targets as well, and its inhibition of DGKα requires relatively high concentrations. That being said, those concentrations are achievable in mice and very likely in humans as well. Ritanserin inhibition of serotonin receptors is unlikely to cause major side effects in patients, given its track record, and may in fact cause beneficial effects such as improved mood and sleep.

Following the identification of distinct GBM subtypes, recent studies have identified another resistance mechanism: transitioning from one subtype to another upon treatment. To combat this phenomenon, as well as initial subtype heterogeneity within individual GBMs,^3,4 we propose the development of combinations of subtype-targeted therapies. While our results demonstrated significant single-agent efficacy of ritanserin against mesenchymal GSC lines in vivo, they also demonstrated the need for combinations with a proneural-targeting agent. To our knowledge, these findings are the first demonstration of a synergistic combination of agents targeting different GBM subtypes. Given the heterogeneity of individual GBMs and the apparent ability of GBM cells to switch subtypes, we believe that such combinations represent a promising approach. Furthermore, the same principle may well hold true for epithelial cancers with contrasting epithelial and mesenchymal phenotypes. Cancers appear to have a very limited phenotypic repertoire, in contrast to a very diverse genotypic repertoire, and combined targeting of cancer phenotypes may provide broadly applicable regimens as an alternative to personalized genetic medicine approaches to cancer. Personalized cancer medicine strategies have focused largely on tailoring therapy to individual genotypes and driver mutations, but these are highly diverse and cancer’s ability to switch its driver genotype often leads to rapid resistance. In contrast, these diverse cancer genotypes converge on a small number of phenotypes, and regimens that simultaneously target the key phenotypes may not provide easy routes to adaptive resistance. Effective therapies against the mesenchymal cancer phenotype, with DGKα inhibition a promising option, are likely to be a key element in such regimens.

Supplementary Material

Supplementary material is available at Neuro-Oncology online.

Funding

This work was supported by the National Institutes of Health (5R01CA180699, 1R01CA189524 to B.P.).

Conflict of interest statement. The authors declare no competing financial interests, but the University of Virginia has filed for a patent on the use of ritanserin and DGKα inhibition alone and in combination for treatment of glioblastoma. The FDA has granted B.P. an Orphan Drug designation for the use of ritanserin for the treatment of glioblastoma.