IDH1R132H confers age-dependent survival to glioma hosts

To examine the impact of IDH1 mutation and activity on age-dependent glioma outcomes, we implanted GL26 transfected with either wild-type IDH1 or IDH1R132H into young and old syngeneic C57BL/6 (B6) mice. Wild-type IDH1 transfectants (GL26-IDH1) exhibited no significant survival difference in young and old hosts, as we similarly reported for parental GL26 (Fig. 1A) [19]. IDH1R132H transfectants (GL26-IDH1R132H), however, exhibited significantly decreased survival in aged relative to young hosts (18–25 months vs. 6–8 weeks, respectively; Fig. 1B). GL26-IDH1R132H and GL26-IDH1 expressed IDH1R132H and over-expressed wild-type IDH1 protein, respectively (Fig. 2A). The dependence of age-dependent survival on CD8 T cell activity in the GL26 model [10, 20] thus suggested a specific impact of IDH1R132H protein on anti-tumor T cells. Since IDH1R132H is expressed only in tumors, this could further exacerbate intrinsic deficits in T cell antigen responses with aging. Such deficits were readily detectable against a number of tumor-associated epitopes (Supplementary Fig. S1A).

Fig. 1: IDH1R132H tumors exhibit vaccine resistance in humans and mice.

A In C57BL/6 mice implanted with GL26-IDH1 (n = 14 young, n = 4 old), differences in survival did not reach significance (p = 0.5757, Log-rank). B In C57BL/6J mice implanted with GL26-IDH1R132H (n = 15 young, n = 5 old), survival of old vs. young was significantly different (p = 0.002, Log-rank test). C, D C57BL/6 mice were implanted with GL26-IDH1, or with GL26-IDH1R132H, and treated with either PBS (n = 14-15), low dose (therapeutic) DC vaccine (n = 9-10), or high dose (detrimental) DC vaccine (n = 5). The survival benefit associated with vaccination (p < 0.01 in GL26-IDH1 by log-rank) is abrogated by the R132H mutation (p > 0.09 in GL26-IDH1R132H by log-rank). E, F Comparison of progression-free survival between IDH1 mutation-negative, or IDH1R132H GBM patients that were either immunological responders (n = 9 for IDH1; n = 3 for IDH1R132H) or immunological non-responders (n = 15 for IDH1; n = 7 for IDH1R132H), after administration of autologous tumor lysate-pulsed DC vaccine. For IDH1 patients, p = 0.036; p = 0.88 for IDH1R132H patients, by Log-Rank.

Fig. 2: Wild-type IDH1 expression and clinical immunotherapy metrics in GBM patients.

A IDH1, IDH1R132H, and control (GAPDH) protein expression in GL26 transfectants prior to implantation. B IDH1 transfection into GL26 prolongs C57BL/6 (B6) host survival after DC vaccine therapy. C Newly diagnosed GBM from 16 DC vaccine therapy patients were subjected to microarray expression analysis before and after treatment (n = 10), or genomic sequencing (n = 6), and stratified by retention and loss of IDH1/IDH2 expression or gene copies (copy number loss = CNL), respectively. IDH1 CNL tumors (n = 3) were excluded from IDH2 analysis, and IDH2 CNL tumors (n = 1) were excluded from IDH1 analyses. Patients retaining IDH1 but not IDH2 exhibited significantly longer overall survival. Survival differences between IDH1-loss and -retained groups was increased when GBM without post-treatment CD8 infiltration were excluded from analysis (n = 4 & 8, respectively; 372 vs. 716 days median; P = 0.006 by Log-Rank). D Post-vaccine IFNγ production by CD8 T cells, and CD8 signal within tumor tissue (Affymetrix HG-U133+2 probeset 205758_at), both failed to significantly correlate with patient survival.

IDH1R132H diminishes vaccine survival benefits in mouse and human gliomas

To address the impact of IDH1 mutation on anti-tumor T cell activity, we examined how IDH1 and IDH1R132H transfection impacted GL26 progression in DC vaccine-treated syngeneic (C57BL/6; B6) hosts. GL26-IDH1R132H and GL26-IDH1 exhibited identical survival in T cell-deficient B6.Foxn1 nude mice (P = 0.95; data not shown), and in untreated young B6 mice (Fig. 1A, B). Treatment with 50,000, GL26 tumor lysate-pulsed DC 2.4 cells in young B6 hosts (low-dose therapeutic DC vaccine) resulted in increased peripheral CD8 T cells reactive to Trp-2, the dominant CD8 T cell epitope in GL26 [10], in both GL26-IDH1 and GL26-IDH1R132H hosts (Supplementary Fig. S1B). Nevertheless, DC vaccination significantly prolonged survival in GL26-IDH1, but failed to do so in GL26-IDH1R132H hosts (Fig. 1C, D). Thus, IDH1R132H in gliomas impaired immunotherapy success without increasing anti-tumor CD8 T cells, as has been shown in separate glioma models [17].

To determine if the impact of IDH1R132H on glioma immunotherapy was uniformly pro-oncogenic, and thus related to its oncometabolite (i.e., 2HG) production, we examined whether it differentially affected host survival upon exposure to detrimental T cell activity. GL26-IDH1 host survival was decreased by administration of a very large vaccine dose (2 × 107 GL26 tumor lysate-pulsed DC 2.4 cells), reminiscent of other models of immunotherapy-induced immunosuppression [21, 22] (Fig. 1C). GL26-IDH1R132H nullified this effect, restoring survival to that seen in untreated hosts (Fig. 1D). Thus, IDH1R132H-mediated modulation of GL26 host survival depended on the specific impact of immune intervention, decreasing survival with beneficial immunotherapy and increasing it with detrimental vaccination. These findings are inconsistent with a dominant influence of IDH1R132H oncometabolite production in this model, and instead suggest the mutation may inactivate an immune-potentiating function of wild-type IDH1.

To further explore this, we examined immune response metrics in human GBM, focusing first on our own immunotherapy patients. As in previous trials, immune responding patients with wild-type IDH1 (those exhibiting ≥50% increased IFNγ tumor antigen response post-vaccine without IDH1R132H) showed significantly increased progression-free survival (PFS) after DC vaccination (Fig. 1E). In contrast, GBM patients with IDH1R132H exhibited no increase in PFS after DC vaccination (Fig. 1F), despite unaltered immune response magnitudes (Supplementary Fig. S2A). Moreover, all patients with IDH1R132H exhibited either endogenous or vaccine-induced immune responses, whereas approximately a third of those lacking the mutation did not (Supplementary Fig. S2B). Thus, IDH1R132H is associated with impaired vaccine clinical efficacy in mice and patients, yet is consistently associated with anti-tumor T cell responses.

Low grade gliomas and GBM with IDH1 mutation have diminished CD8 T cell activity

Reduction of DC vaccine benefits by IDH1R132H despite systemic T cell responsiveness suggests local impairment of anti-tumor T cells. We thus examined the expression of T cell and other genes by IDH1-mutated and IDH1-wild-type gliomas in The Cancer Genome Atlas (TCGA). Low-grade gliomas with IDH1R132H exhibited significantly lower T cell effector gene expression than those without the mutation (Supplementary Fig. S3A–C). Expression of innate and other adaptive immune genes, including GFAP for activated astrocytes and IBA-1 for microgliosis, were comparable between IDH1R132H and IDH1-wild-type gliomas (Supplementary Fig. S4A–E). Similar trends were observed in GBM, but statistical significance was not reached with the smaller numbers of IDH1R132H tumors in that cohort (Supplementary Figs. S3D–F, S4F–J). Expression of most CD4 T subset-associated genes [23] was also not significantly impacted by IDH1 mutation in low grade gliomas (Supplementary Fig. S4K). These data expand on recent reports associating IDH1R132H with reduced CD8 T cell gene expression in TCGA gliomas [17], which implicated 2HG-mediated inhibition of CXCL10 production in lower T cell recruitment and metabolic reprogramming in CD8 T cells leading to anti-glioma immunosuppression [24, 25]. We did not observe reduced GFAP and IBA-1 expression, however, which should also be impacted by CXCL10 inhibition [26]. Thus, data from human GBM corroborate that IDH1R132H may impair anti-tumor T cells independent of 2HG-mediated effects.

IDH1 over-expression enhances DC vaccine in GL26, and its retention predicts GBM vaccine success

To directly test whether wild-type IDH1 possesses an immune enhancing function eliminated by IDH1R132H, we examined post-vaccine host survival in IDH1-over-expressing GL26-IDH1 (Fig. 2B) in greater detail. Host survival more than doubled after DC vaccination in GL26-IDH1 compared to untransfected and control-transfected GL26 (median 52 days; Fig. 2B and Supplementary Fig. S5, respectively), rendering a majority of hosts long-term survivors. Unmodified GL26 tumors that grew intracranially despite therapeutic vaccination also exhibited decreased expression of IDH1 RNA on microarray analysis, relative to GL26 growing in brains of either unvaccinated or T cell-deficient mice (Supplementary Fig. S6A). Thus, IDH1 over-expression increased GL26 vaccine success, while vaccine failure was accompanied by enrichment of IDH1-lo GL26 cells, possibly due to their less efficient destruction by T cells.

In vaccinated GBM with increased CD8 tumor signal post-treatment (i.e., local treatment response), IDH1 DNA or RNA loss predicted significantly shorter patient survival (Fig. 2C). GBM with low IDH1 expression before DC vaccination also failed to increase CD8 signal after vaccine therapy (Supplementary Fig. S6B). By contrast, IDH2 retention was not associated with longer patient survival after vaccination (Fig. 2D), nor was IFNγ production by peripheral T cells, or increased CD8 in tumors after treatment (Fig. 2D). Together, these findings suggest that retention of wild-type IDH1 in gliomas enhances clinical success of T cell-activating immunotherapy, whereas IDH1 loss impairs local anti-tumor T cell responses and leads to immunotherapy failure. The contrasting finding with IDH2, which catalyzes the identical metabolic reaction as IDH1 in cells, further emphasizes that IDH1 may enhance T cell activity independent of its known enzymatic function.

IDH1R132H diminishes and IDH1 enhances glioma lysis by CD8+ cells

To further examine requirements for IDH1-mediated effects on T cell activity, we examined T-mediated killing of GL26 with and without oxalomalic acid (OMA), a competitive inhibitor of IDH1 substrate binding and catabolic activity [27,28,29]. OMA significantly impaired GL26 killing by a CD8+ T cell hybridoma reactive to H-2Kb (HTB-156.7) at 10:1 effector:target cells (E:T ratio; Fig. 3A). This suggests that IDH1 substrate binding and/or catabolic activity is necessary for optimal T cell lysis, but whether IDH1 was tumor-derived was unclear. We thus compared HTB-156.7 killing of GL26 to that of GL26-IDH1 and GL26-IDH1R132H. In addition to expressing transfected genes appropriately (Fig. 2A), GL26-IDH1 and GL26-IDH1R132H exhibited unique alterations in lipids under glucose-starvation and recovery conditions, respectively, confirming distinct metabolic alterations (Supplementary Fig. S7) [30,31,32]. H-2Kb antigen expression was also unaltered by the transfections (Supplementary Fig S8A), and their sensitivity to cell death induced by the caspase 8 agonist, imidazole, was comparable (Supplementary Fig. S8B), consistent with prior studies [33, 34]. Exogenous 2-hydroxyglutarate (2HG) did not significantly alter overall effector function (IFNγ production; Supplementary Fig. S9A) or proliferation (Supplementary Fig. S9B), also consistent with prior studies [35]. Despite this, lysis of GL26-IDH1R132H targets was delayed after 4 h compared to untransfected GL26, whereas GL26-IDH1 lysis was enhanced at later time points (Fig. 3B). This suggests that intracellular IDH1 and IDH1R132H in tumors enhance and inhibit T cell activity, respectively. Because IDH1 is not normally secreted, this may occur through the release of active cytoplasmic IDH1 or IDH1R132H from CTL-lysed or otherwise ruptured tumor cells. Indeed, ample catalytic activity of IDH1/NADP+ was present in supernatants of lysed tumor cells (Supplementary Fig. S8C). More directly, extracellular anti-IDH1 antibody significantly increased GL26-IDH1R132H killing by CTL from tumor-vaccinated mice after 4 h (Fig. 3C). Together, these findings broadly support the notion that glioma killing by CTL is modulated by cytoplasmic IDH1 proteins released from lysed tumor cells.

Fig. 3: IDH1 treatment increases multimer binding and cytokine production in CD8 T cells.

A GL26 target lysis by an H-2Kb reactive T cell hybridoma effectors after 7 h, +/−5 mM oxalomalic acid (OMA); *p < 0.05, ***p < 0.005. B Parental GL26, IDH1 and IDH1R132H transfectant lysis over time by H-2Kb reactive hybridoma effectors (E:T ratio 10:1); *p < 0.05. C Anti-IDH1 antibody increased early lysis of GL26-IDH1R132H targets by CTL from GL26-vaccinated mice. Anti-IDH1 had no impact on CTL lysis of 7 and 16 h targets (4 h with E:T = 10:1 E:T is shown; *p < 0.05). D Enhanced binding of IDH1-treated CD8 T cells to pMHC I dextramers. E IDH1-treated C57BL/6 (B6) mouse CD8 T cells producing IFNγ expand following stimulation with an H-2Kb/TVSEFLKL Survivin dextramer + anti-CD28 antibody. F B6 CD8 T cell IFNγ production over time stimulated by H-2Kb/SVYDFFVWL Trp-2 dextramer + anti-CD28, after treatment with IDH1, neuraminidase, or control (Untreated or No Rx = [dextramer+anti-CD28], no IDH1-treatment). G 10-h compilation of distinct pMHC I dextramer + anti-CD28 stimulation.

Extracellular IDH1 enhances CD8 T cell reactivity

IDH1 could conceivably enhance CTL activity by altering a distinct immunoactive substrate. Sialic acid may represent such a substrate, given the reported structural similarity between NADP+-dependent IDH and a microbial sialic acid-cleaving enzyme (neuraminidase/sialidase) [18]. Indeed, removal of sialic acid from o-linked glycans on the CD8β stalk by Vibrio cholerae neuraminidase (VCN) enhances coreceptor binding to peptide-MHC I (pMHC-I), and increases T cell responsiveness to pMHC-I by mouse and human T cells [10,11,12,13]. We thus examined whether IDH1 outside the cell similarly affects CD8 T cell responsiveness to pMHC-I multimer stimulation. Treatment of mouse CD8 T cells with IDH1 and NADP+ enhanced binding to tumor pMHC-I (Fig. 3D). More importantly, IDH1 treatment of CD8+ HTB-156.7 hybridoma cells led to its increased IFNγ production in response to pMHC-I stimulation (Fig. 3E), which was sustained over time as VCN-treated cells died, and increased with higher IDH1 (Fig. 3F). This was observed with several pMHC-I multimers (Fig. 3G). These results suggest that, like VCN, extracellular IDH1 increases antigen activation of mouse T cells. Moreover, IDH1-mediated response enhancement was independent of T cell epitope specificity, consistent with an impact on CD8 [10, 13].

Treatment of human CD8 T cells treated with exogenous IDH1/NADP+ led to a marked increase in the proportion of cells binding MAGE-1/HLA multimers (Supplementary Fig. S10A). This multimer-bound population also exhibited increased binding to peanut agglutinin (PNA), a lectin that binds quantitatively to desialylated galactosyl (β-1,3) N-acetylgalactosamine glycans on cell surfaces (Supplementary Fig. S10B). The PNA binding increase was prevented by excess CMP-sialic acid, consistent with competitive inhibition of desialylation. Moreover, pHLA binding was prevented by addition of anti-CD3 antibody prior to flow cytometry (Supplementary Fig. S10A), suggesting involvement of both TCR and CD8. Similar augmentation of pMHC-I and PNA binding was seen in IDH1-treated mouse cells responding to a distinct tumor-associated Survivin epitope (Supplementary Fig. S10C, D). Together, these findings indicate that extracellular IDH1 directly enhances T cell activity, and that this enhancement involves T cell desialylation.

IDH1 but not IDH1R132H removes sialic acid from glycoproteins

The possibility that IDH1 can cleave sialic acid from glycans was first examined through computational modeling, which indicated that sialic acid can in theory bind to the catalytic site of IDH1 with a Ki of 13.9 μm. In this analysis, sialic acid binding was dependent on many of the same catalytic site residues that bind isocitrate, and was similarly stabilized by NADP+ (Fig. 4A, B). By contrast, IDH1R132H modeling predicted much weaker sialic acid binding, with or without NADP+ (Fig. 4C).

Fig. 4: IDH1 demonstrates sialidase activity.

A Crystal structure of isocitrate and NADP+ bound in the active site of IDH1. Dashed yellow lines show hydrogen bonds. B Predicted model of sialic acid in the active site of IDH1. C Predicted model of sialic acid in the active site of IDH1R132H. D Sialic acid release assay of IDH1 relative to V. cholera neuraminidase (VCN), demonstrates in vitro IDH1-mediated sialidase activity against purified CD8β monomer that is largely dependent on NADP+. E IDH1 sialidase activity against CD8β is abrogated by the IDH1R132H mutation.

Further consistent with sialic acid-binding, IDH1-mediated production of NADPH from isocitrate + NADP+ was inhibited and by 6’-sialyllactose, a sialidase substrate, as well as by its known catalytic inhibitor, α-ketoglutarate (aKG; Supplementary Fig. S11A). This provided empirical evidence that IDH1 residues involved in isocitrate catalysis competitively interact with sialylated glycans. We next measured release of sialic acid from fetuin and CD8 glycoproteins by IDH1 and IDH1R132H with and without NADP+. Relative to VCN, IDH1 alone displayed minimal activity, but induced significant sialic acid release from CD8β with NADP+ (Fig. 4D). The same concentration of IDH1R132H failed to exhibit detectable sialidase activity against CD8β with or without NADP+ (Fig. 4E). Antibody absorption of IDH1 also decreased sialidase activity against fetuin from GL26-IDH1 lysate, whereas sialidase activity in GL26-IDH1R132H lysate was significantly lower, and slightly increased by IDH1 absorption (Supplementary Fig. S11B). These findings are consistent with substantial reduction of IDH1-mediated sialidase activity released from dead, IDH1R132H-expressing tumor cells.

To determine if IDH1 desialylates surface molecules on cells, we incubated splenocytes with exogenous IDH1, which revealed significantly more PNA binding by flow cytometry to CD8+ compared to CD8-negative cells in the presence of NADP+, whereas no such increase was observed in the absence of NADP+ (Fig. 5A, B). IDH1R132H did not significantly increase PNA binding to either CD8+ or CD8-negative cells. This suggests that NADP+ and IDH1, but not IDH1R132H, preferentially desialylates CD8+ cell surface molecules.

Fig. 5: IDH1 binds to and desialylates CD8+ cells.

A Flow cytometric analysis of splenocytes exogenously treated with FITC-labeled IDH1 demonstrates higher binding of IDH1 to CD8+ cells and selective desialylation of those cells. B Relative PNA fluorescence intensity (MFI) is increased by IDH1 only on CD8+ cells, and is dependent on both NADP+, and a wild-type R132 residue. C Quantification of FITC-labeled IDH1 and IDH1R132H MFI demonstrates preferential binding to CD8+ cells, irrespective of R132 mutation. *p < 0.05; **p < 0.01; ***p < 0.001.

IDH1 specifically binds and desialylates CD8 dimers

We used fluorescein-labeled IDH1 and IDH1R132H in flow cytometric analyses to examine its interaction with discrete surface ligands as a potential mechanism for its preferential desialylation of CD8+ cells. CD8+ cells indeed exhibited elevated binding to fluorescein-IDH1 and fluorescein-IDH1R132H (Fig. 5A, C). We then immunoprecipitated native IDH1 from surface-biotinylated splenocyte lysates, and detected accompanying labeled proteins on blots with streptavidin. IDH1R132H protein was added to some lysates after surface labeling as well. Immunoprecipitation with either anti-IDH1 in native lysates, or anti-IDH1R132H antibody in IDH1R132H-supplemented lysates, pulled down a doublet of 38–45 kDa under reducing conditions, frequently accompanied by a larger 70 kDa species (Fig. 6A). Since this mobility was consistent with the disulfide-linked CD8αβ heterodimer, we repeated immunoprecipitations on surface-biotinylated splenocytes from CD8β-deficient mice, which revealed loss of the fastest migrating species as expected of CD8β (Fig. 6B, left panel). We then blotted and probed IDH1 immunoprecipitates of wild-type and CD8β-deficient lysates with anti-CD8α antibody, which identified a band at 70 kDa not present in control lysates (Fig. 6B, right panel; note: lower bands obscured by immunoglobulin signal). This confirmed that IDH1 specifically binds to both CD8αα and CD8αβ dimers. Indeed, molecular modeling predicted binding of monomeric IDH1 to CD8αβ dimers based on topological complementarity (Fig. 6C).

Fig. 6: IDH1 binds to CD8.

A Surface-biotinylated proteins coimmunoprecipitated with human and mouse IDH1, and human IDH1R132H on mouse CD8 T cells, resemble the CD8αβ heterodimer. B Surface-biotinylated species (left panel) and anti-CD8α Westerns (right panel), of IDH1-immunoprecipitates from wild-type and CD8β-knockout (KO) splenocyte lysates. C Molecular modeling predicts stable binding of IDH1 monomers to CD8αβ heterodimers. D Low (1 µg) concentration neuraminidase and IDH1 sialidase activity against fetuin and CD8 substrates.

Specific binding to CD8 dimers could allow IDH1 to specifically desialylate CD8 under physiological conditions. Accordingly, low concentration IDH1/NADP+ desialylated CD8α as well as CD8αβ mixtures, which both form dimers. Indeed, CD8αβ desialylation by IDH1/NADP+ was even more efficient than by VCN. By contrast, similar concentrations of other substrates, including fetuin and isolated CD8β monomers, were not efficiently desialylated under such conditions (Fig. 6D). Thus, binding and desialylation by limiting amounts of IDH1/NADP+ was selective for CD8 dimers, and CD8αβ heterodimers particularly.