prognostic role of mitochondrial pyruvate carrier in

CLINICAL ARTICLEJ Neurosurg 130:56–66, 2019

Gliomas are one of the most common types of pri-mary brain tumors, accounting for 28% of all brain tumors and having an annual age-adjusted

incidence of 2.12 per 100,000.27 The 2016 WHO classifi-cation categorizes diffuse astrocytic and oligodendroglial tumors with key diagnostic molecular features, including isocitrate dehydrogenase (IDH) mutation and 1p19q code-letion.1,4,12,21,25 IDH mutations are seen in > 70% of WHO grade II and grade III (referred to collectively hereafter as

lower-grade) gliomas or secondary glioblastoma and can be used to predict overall survival (OS), progression-free survival (PFS), and treatment response.9,30,37 Patients with IDH-mutant gliomas have much longer median OS than those with IDH–wild type gliomas (2- to 3-fold longer in glioblastomas and 3- to 5-fold longer in lower-grade glio-mas).30,37

Metabolic reprogramming has been recognized as a crucial process to the survival and proliferation of can-

ABBREVIATIONS FDR = false-discovery rate; KPS = Karnofsky Performance Scale; OS = overall survival; PFS = progression-free survival; TCGA = The Cancer Genome Atlas.SUBMITTED August 17, 2017. ACCEPTED September 20, 2017.INCLUDE WHEN CITING Published online March 16, 2018; DOI: 10.3171/2017.9.JNS172036.

Prognostic role of mitochondrial pyruvate carrier in isocitrate dehydrogenase–mutant gliomaMichael Karsy, MD, PhD,1 Jian Guan, MD,1 and L. Eric Huang, MD, PhD1,2

Departments of 1Neurosurgery and 2Oncological Sciences, University of Utah, Salt Lake City, Utah

OBJECTIVE Gliomas are one of the most common types of primary brain tumors. Recent studies have supported the importance of key genetic alterations, including isocitrate dehydrogenase (IDH) mutations and 1p19q codeletion, in glioma prognosis. Mutant IDH produces 2-hydroxyglutarate from a-ketoglutarate, a key metabolite of the Krebs cycle. The mitochondrial pyruvate carrier (MPC) is composed of MPC1 and MPC2 subunits and is functionally essential for the Krebs cycle. The authors sought to explore the impact of MPC1 and MPC2 expression on patient prognosis.METHODS Genomic and clinical data in patients with lower-grade glioma (WHO grades II and III) from The Cancer Genome Atlas (TCGA) were evaluated using Kaplan-Meier analysis and hazards modeling. Validation was conducted with additional data sets, including glioblastoma.RESULTS A total of 286 patients with lower-grade glioma (mean age 42.7 ± 13.5 years, 55.6% males) included 54 cases of IDH–wild type (18.9%); 140 cases of IDH-mutant, 1p19q-intact (49.0%); and 85 cases of IDH-mutant, 1p19q-co deleted (29.7%) tumors. Kaplan-Meier analysis showed that an MPC1 z-score > 0 distinguished better survival, particularly in IDH-mutant (p < 0.01) but not IDH–wild type tumors. Conversely, an MPC2 z-score > 0 identified worsened survival, particularly in IDH-mutant (p < 0.01) but not IDH–wild type tumors. Consistently, neither MPC1 nor MPC2 was predictive in a glioblastoma data set containing 5% IDH-mutant cases. Within the IDH-stratified lower-grade glioma data set, MPC1 status distinguished improved survival in 1p19q-codeleted tumors (p < 0.05), whereas MPC2 expression delineated wors-ened survival in 1p19q-intact tumors (p < 0.01). A hazards model identified IDH and 1p19q status, age (p = 0.01, HR = 1.03), Karnofsky Performance Scale (KPS) score (p = 0.03, HR = 0.97), and MPC1 (p = 0.003, HR = 0.52) but not MPC2 (p = 0.38) as key variables affecting overall survival. Further validation confirmed MPC1 as an independent predictor of lower-grade glioma. A clinical risk score using IDH and 1p19q status, age, KPS score, and MPC1 and MPC2 z-scores defined 4 risk categories for lower-grade glioma; this score was validated using a secondary glioma data set.CONCLUSIONS These results support the importance of MPC, especially MPC1, in improving prognostication of IDH-mutant tumors. The generation of a risk score system directly translates this finding to clinical application; however, fur-ther research to improve the molecular understanding of the role of MPC in the metabologenomic regulation of gliomas is warranted.https://thejns.org/doi/abs/10.3171/2017.9.JNS172036KEY WORDS biomarker; glioblastoma; isocitrate dehydrogenase; IDH; lower-grade glioma; mitochondrial pyruvate carrier; overall survival; prognosis; risk gene; oncology

J Neurosurg Volume 130 • January 201956 ©AANS 2019, except where prohibited by US copyright law

Unauthenticated | Downloaded 04/22/22 03:00 AM UTC

J Neurosurg Volume 130 • January 2019 57

M. Karsy et al.

cer cells.11,34 Cancer cells switch to aerobic glycolysis (the Warburg effect) and divert glycolytic metabolites to fuel anabolic processes for cell proliferation. Glucose oxida-tion through the Krebs cycle requires the transport of pyru-vate into the mitochondria for oxidative phosphorylation.29 Recent studies have revealed that the mammalian mito-chondrial pyruvate carrier (MPC) is a 150-kD multimeric protein complex composed of MPC1 and MPC2 subunits and is embedded in the mitochondrial inner membrane.2,18 MPC2 was previously known as brain protein 44 (BRP44) and MPC1 as brain protein 44-like (BRP44 L);2 however, their gene loci differ: MPC1 at 6q27 and MPC2 at 1q24.2.

Although MPC is evolutionarily conserved and es-sential for mitochondrial pyruvate transport, MPC1 is frequently deleted or downregulated, thereby resulting in low MPC activity in a variety of human cancers, including colorectal, prostate, and esophageal squamous cell carci-nomas.2,23,24,33 Experimental studies have indicated that MPC acts as a repressor of cancer cell growth by inhibit-ing aerobic glycolysis, which is consistent with the concept of the Warburg effect. Given the importance of metabolic adaptation in malignant glioma,7,13,22,32 we hypothesized that levels of MPC1 and MPC2 expression may influence glioma patient survival in an interaction with IDH status.

MethodsData Sets

The Brain Lower Grade Glioma (The Cancer Genome Atlas [TCGA], Provisional) data set, which included 286 cases of sequenced tumors with genomic and clinical data, was obtained from cBioPortal (http://www.cbioportal.org/)5,16 as described previously.19 The genomic data set contained IDH status and MPC1 and MPC2 mRNA ex-pression z-scores (RNA Seq V2 RSEM), and the clinical data set contained demographic, clinical, and pathological information. Cases with either IDH1 or IDH2 mutations were grouped into a single group, IDH-mutant. The TCGA glioblastoma data set containing 136 cases was down-loaded in the same way. The publicly available GSE16011 cohort contained 109 cases of lower-grade gliomas.17 The TCGA lower-grade glioma data set was used for analysis of gene expression and patient survival and for creation of a prognostic score. Subsequently, the 21 TCGA data sets of various cancer types including glioblastoma, lower-grade glioma, and GSE16011, were used for validation.

Comparison of Gene ExpressionColumn analyses were performed using GraphPad

Prism 7 to compare the differences in MPC1 and MPC2 expression among different groups with unpaired t-tests or ANOVA. The results are presented in scatterplots. Two-tailed p values were used for statistical significance.

Survival AnalysisAll continuous variables were analyzed as mean ±

standard deviation, and continuous variables were ana-lyzed as count and percentage of total. Continuous vari-ables were analyzed using the t-test, whereas discrete variables were analyzed using the chi-square test. Kap-lan-Meier curves with log-rank testing were generated.

Kaplan-Meier survival analysis was also performed using GraphPad Prism.7

ValidationWe entered known variables affecting patient survival

into a Cox proportional hazards logistic regression model with interaction term between MPC1 and MPC2. Boot-strapping with sample replacement of the hazards model was performed with 1000 iterations, and the results were compared with those of the single-run logistic regression.

Cox coefficients based on MPC1 and MPC2 expres-sion, sex, age, tumor, and grades were compared among 21 TCGA cancer types using OncoLnc (http://www.oncolnc.org); p values of all the cancer types were corrected by false-discovery rate (FDR) and ranked according to statis-tical significance. Cox coefficients with FDR < 0.25 were considered statistically significant.

A clinical scoring system was devised to predict OS and PFS by using MPC1 and MPC2 z-scores as well as glioma scores from previous studies.6,31 Points were as-signed according to relative hazard ratios (HRs) in the Cox proportional hazards model. For molecular subtypes, a score of 1 was given to IDH-mutant, 1p19q-codeleted tumors; 2 to IDH-mutant, 1p19q-intact tumors; and 3 to IDH–wild type tumors. Patients with unclassified 1p19q status were excluded. Patients 50 years and older or with a Karnofsky Performance Scale (KPS) score of 80 or lower were assigned a point value of 1. Patients with an MPC1 z-score < 0 or MPC2 z-score > 0 were also assigned a value of 1. The cumulative score was summed to identify the clinical score.

Statistical AnalysisStatistical analysis was performed using IBM SPSS

(version 20.0, IBM Corp.) and GraphPad Prism 7. Kaplan-Meier survival analysis was performed using GraphPad Prism 7. A p value < 0.05 was considered statistically sig-nificant.

ResultsBaseline Characteristics

The Brain Lower Grade Glioma data set from TCGA consisted of 286 patients with defined IDH and 1p19q status (Table 1). The mean age of the cohort was 43 ± 13 years, and 159 patients (55.8%) were male. The most common Karnofsky Performance Scale (KPS) score was 90 in this group. This data set included IDH–wild type astrocytomas (n = 54, 18.9%); IDH-mutant, 1p19q-intact astrocytomas (n = 140, 49.0%); and IDH-mutant, 1p19q-codeleted oligodendrogliomas (n = 85, 29.7%); 7 tumors were unclassified. Histological diagnoses according to the 2007 WHO criteria identified oligodendroglioma (23.8%), oligoastrocytoma (26.2%), anaplastic oligoastrocytoma (15.7%), astrocytoma (10.8%), and anaplastic astrocytoma (23.4%). Tumor locations were equally divided between left (49.1%) and right (50.9%) sides. Most tumors were in the frontal lobes (61.2%), followed by the temporal lobes (26.9%). Other locations included the parietal (8.0%), oc-cipital (1.0%), and cerebellar (1.0%) lobes; the areas were unspecified in 1.7%. Seizure was the most common pre-


M. Karsy et al.

J Neurosurg Volume 130 • January 201958

sentation (49.3%); headache was the second most common presentation (22.0%), followed by motor/movement chang-es (7.3%), mental status changes (7.3%), visual changes (2.4%), and sensory changes (2.1%). Documented treat-ments in patients included preoperative corticosteroids in 32.9%, neoadjuvant therapy prior to resection in 54.2%,

and adjuvant postoperative target therapy in 14.0%. The OS was 30.5 ± 30.6 months (range 0–182.2 months), and the PFS was 24.2 ± 24.2 months (range 0–172.6 months).

The mean MPC1 mRNA z-score was -0.23 ± 1.02 and the mean MPC2 mRNA z-score was -0.12 ± 1.08 (Table 1). Although there was no significant difference in the mean MPC1 mRNA expression between IDH–wild type and IDH-mutant tumors (z-scores -0.24 ± 0.14 vs -0.23 ± 0.07, p = 0.96), the mean MPC2 mRNA level for IDH–wild type tumors was significantly higher than that of IDH-mutant tumors (z-scores 0.30 ± 0.20 vs -0.21 ± 0.06, p = 0.002; Fig. 1A). IDH-mutant, 1p19q-codeleted tumors, however, showed a greater mean MPC1 mRNA level compared with 1p19q-intact tumors (z-scores 0.24 ± 0.99 vs -0.57 ± 0.88, p < 0.0001; Fig. 1B) and a lower mean MPC2 mRNA level compared with IDH–wild type tumors (z-scores -0.42 ± 0.85 vs -0.11 ± 0.99, p < 0.001; Fig. 1C).

MPC1 scores were distinctly clustered for IDH-mutant, 1p19q-intact tumors and IDH-mutant, 1p19q-codeleted tu-mors compared with MPC2 scores (Fig. 2). For IDH–wild type tumors, a wider distribution of MPC1 compared with MPC2 scores was identified. MPC1 values as a group did not show a statistically significant correlation with OS (R = -0.5, p = 0.42), whereas MPC2 showed a significant negative correlation with OS (R = –0.1, p = 0.02). Nev-ertheless, MPC1 z-scores significantly correlated with MPC2 z-scores (R = 0.1, p = 0.02).

Effect of MPC on Lower-Grade Glioma Patient SurvivalKaplan-Meier survival analysis was used to compare

the effect of MPC cutoffs on survival. For a cutoff of MPC1 z-scores < 0, a median OS of 64 months was ob-served; for z-scores > 0, the median OS was improved to 134 months (p < 0.05) (Fig. 3A), which was 154% great-er than the median survival (87 months) of IDH-mutant patients from the same cohort. It is noteworthy that no significant effects were observed with the analysis of the TCGA glioblastoma data set (Supplementary Fig. S1A), suggesting the specificity of MPC1 effect on survival in lower-grade glioma. Given IDH mutations in 81% of the lower-grade glioma data set versus 5% of the glioblastoma data set, we reasoned that IDH status might account for the difference. In keeping with this, IDH-mutant, but not IDH–wild type, tumors also showed a much longer sur-vival with higher MPC1 levels (p < 0.01) (Fig. 3B). On the other hand, with MPC2 scores > 0, the median survival was significantly shorter for the entire cohort (50 months, p < 0.05) (Fig. 3C) and patients with IDH-mutant tumors (64 months, p < 0.01) (Fig. 3D). Again, no significant ef-fect of MPC2 on survival was seen in glioblastoma (Sup-plementary Fig. S1B).

Survival Effect of MPC Associated With 1p19q StatusTo determine whether the MPC survival effect was as-

sociated with glioma molecular subclassification, we fur-ther stratified IDH-mutant gliomas on the basis of 1p19q status. Segregation of patients with an MPC1 z-score cutoff of 0 showed a significant positive effect on OS in 1p19q-codeleted (p < 0.05, Fig. 4A) but not 1p19q-intact

TABLE 1. Analysis of patient demographics, clinical variables, and tumor types in 286 patients

Variable Value

Mean age in yrs 42.7 ± 13.5No. of male patients 159 (55.6)KPS score 40 50 60 70 80 90 100 Unknown

1 (0.3)3 (1.0)4 (1.4)

13 (4.5)25 (8.7)61 (21.3)29 (10.1)

150 (52.4)Mutation status IDH–wild type IDH−mutant/1p19q−intact IDH−mutant/1p19q−codeleted

54 (18.9)140 (49.0)85 (29.7)

Mean MPC1 z-score −0.23 ± 1.02Mean MPC2 z-score −0.12 ± 1.08Histological diagnosis Oligodendroglioma Oligoastrocytoma Anaplastic oligoastrocytoma Astrocytoma Anaplastic astrocytoma

68 (23.8)75 (26.2)45 (15.7)31 (10.8)67 (23.4)

Tumor side Lt Rt

136 (49.1)141 (50.9)

Tumor site Frontal lobe Occipital lobe Parietal lobe Temporal lobe Cerebellum Not specified

175 (61.2)3 (1.0)

23 (8.0)77 (26.9)

1 (1.0)5 (1.7)

Presenting symptoms Headache Seizure Motor/movement changes Visual changes Mental status changes Sensory changes

63 (22.0)141 (49.3)21 (7.3)7 (2.4)

21 (7.3)6 (2.1)

Preop corticosteroids 94 (32.9)Neoadjuvant therapy prior to resection 155 (54.2)Adjuvant postop targeted therapy 40 (14.0)Mean OS in mos 30.5 ± 30.6Mean PFS in mos 24.2 ± 24.2

Values are presented as the number of patients (%) unless stated otherwise.


https://thejns.org/doi/suppl/10.3171/2017.9.JNS172036




M. Karsy et al.

FIG. 2. Comparison of MPC1 and MPC2 z-scores with OS. A: On scatterplot of MPC1 z-scores and OS along with IDH and 1p19q status, there was no significant correlation of MPC1 z-scores with OS for all patients (R = -0.5, p = 0.42). B: A scatterplot of MPC2 z-scores and OS indicated significant correlation for all patients (R = -0.1, p = 0.02). A wider spread of data is noted for MPC1 compared with MPC2. With MPC1 but not MPC2, IDH-mutant (mut), 1p19-intact patterns also cluster from IDH-mutant, 1p19q-codeleted (codel) patterns. Expression of MPC1 and MPC2 remains variable for IDH–wild type (wt) tumors. C: Com-parison of MPC1 and MPC2 clustering for various IDH and 1p19q status in gliomas is shown. A significant correlation was seen between MPC1 and MPC2 expression for all patients (R = 0.1, p = 0.02). Figure is available in color online only.

FIG. 1. Comparative analysis of MPC1 and MPC2 mRNA levels with regard to IDH and 1p19q status. A: The mean MPC2 mRNA level was significantly greater in IDH–wild type glioma than in IDH-mutant glioma, as shown in scatterplot with mean ± SD. B: The mean MPC1 level was statistically greater in 1p19q-codeleted tumors than in 1p19q-intact tumors. C: The mean MPC2 level was statistically smaller in 1p19q-codeleted tumors than in IDH–wild type tumors. *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001.


M. Karsy et al.


(Fig. 4B) tumors. Conversely, segregation of patients using an MPC2 z-score cutoff of 0 showed a negative impact on survival in 1p19q-intact tumors (p < 0.01, Fig. 4D) but not in 1p19q-codeleted tumors (Fig. 4C). Further analysis re-vealed that the frequency of MPC1 z-scores > 0 was 56% in IDH-mutant, 1p19q-codeleted, but 21% in 1p19q-intact glioma compared with 43% in IDH–wild type glioma (Fig. 5). By contrast, the frequency of MPC2 z-scores < 0 was 75% in 1p19q-codeleted and 64% in 1p19q-intact glioma compared with 51% in IDH–wild type glioma. These results are not only consistent with the best OS in oligodendroglioma (i.e., 1p19q-codeleted) but also sug-gest a predictive role of MPC1 for oligodendroglioma and MPC2 for astrocytomas (i.e., 1p19q-intact).

Hazards Model of MPC in GliomaA Cox proportional, multivariate hazards model was

developed to assess the effect of predictive factors on OS after adjusting for various factors (Table 2). With IDH mutation and 1p19q codeletion as a baseline factor (p = 0.0001), patients with IDH–wild type tumors (p = 0.0001, HR 13.3) had shorter survival than patients with IDH-mu-tant, 1p19q-intact tumors (p = 0.5, HR 1.4). As expected, age played a role in shorter survival (p = 0.01, HR 1.04), whereas KPS score predicted improved survival (p = 0.03, HR 0.97). Although increased MPC1 expression predicted improved survival (p = 0.003, HR 0.52), neither MPC2 (p = 0.39) nor the interaction between MPC1 and MPC2 (p = 0.59) showed an effect. PFS was also analyzed using a multivariate hazards regression (Table 2). PFS was af-fected by molecular classification, namely IDH-mutant,

1p19q-codeleted (p = 0.0001), and IDH–wild type (HR = 10.2, p = 0.0001), but not IDH-mutant, 1p19q-intact (HR = 2.0, p = 0.12) tumors. However, age (p = 0.71, HR 1.00), KPS score (p = 0.26, HR 0.98), MPC1 (p = 0.39, HR 0.87), MPC2 (p = 0.62, HR 0.92), or MPC1 × MPC2 (p = 0.68, HR 1.04) did not affect PFS. OS and PFS were also ana-lyzed by bootstrapping; this analysis showed results that were equivalent to those of logistic regression. Bootstrap-ping of the model did not show significant differences in confidence intervals for the individual variable HRs com-pared with our base model (Supplementary Table S1).

Validation of Prognostic Values of MPC in GliomaTo obtain supporting evidence for the role of MPC in

glioma prognostication, we performed Cox regression analysis of 21 TCGA data sets of different cancer types, including 508 cases of lower-grade glioma and 152 cases of glioblastoma, as well as colon adenocarcinoma, renal clear-cell carcinoma, and renal papillary-cell carcinoma (Table 3). By taking into consideration sex, age, tumor grades, and gene expression, Cox regression analysis confirmed MPC1 as an independent predictor of survival in lower-grade glioma, with a protective Cox coefficient of -0.211 (Table 3). Furthermore, among all the cancer types, lower-grade glioma was ranked statistically second to renal papillary cell carcinoma according to the FDR correction, followed by pancreatic, lung, cervical and en-dometrial, and colon cancers. The Cox coefficients for the remaining cancer types, including glioblastoma, indicated that MPC did not predict survival in these types. Further-more, the Cox coefficients of MPC2 were insignificant for

FIG. 3. Impact of MPC mRNA levels on survival in lower-grade gliomas. A: Evaluation of an MPC1 cutoff > 0 (thick line) showed better survival compared with MPC1 < 0 (thin line). B: An MPC1 cutoff > 0 also separated better survival in IDH-mutant (solid lines; top 2 survival curves) but not IDH–wild type (dashed lines; bottom 2 survival curves) tumors. C: A cutoff for MPC2 > 0 (thick line) showed a significant worsening of survival. D: For IDH-mutant tumors (solid lines; top 2 survival curves), an MPC2 cutoff > 0 showed worsened survival; however, this effect was not seen for IDH–wild type tumors (dashed lines; bottom 2 survival curves). *p < 0.05; **p < 0.01.




M. Karsy et al.

lower-grade glioma and glioblastoma but still suggestive of a more harmful effect than in colon cancer. Moreover, log-rank survival analyses of the validation data sets showed that MPC1 expression significantly predicted OS when comparing the top 25th percentile and the bottom 25th percentile of the lower-grade glioma, but not glioblastoma, data set (Supplementary Fig. S2A and B). As expected, no survival effects were seen with MPC2 expression in either data set (Supplementary Fig. S2C and D). Therefore, we conclude that MPC1 is an independent predictor specific to lower-grade glioma as well as a few other cancer types (Table 3), but not to glioblastoma and many other cancer types.

Scoring SystemWhen age, KPS, and MPC1 and MPC2 z-scores as well

as IDH and 1p19q status were factored in, the cumulative score segregated patients by OS (log-rank test, p = 0.001) but not PFS (log-rank test, p = 0.45). Mean OS of 95.7 ± 0.1 months for scores of 1–2, 136.2 ± 17.7 months for a score of 3, 90.4 ± 9.1 months for a score of 4, and 61.7 ± 7.0 months for a score > 4 were observed (Tables 4 and 5, Fig. 6A and B). We then validated the risk score with an inde-pendent glioma data set GSE16011 (Table 6). Demograph-ic features in terms of age, KPS score, IDH/1p19q status, MPC1 and MPC2 levels, as well as histological diagnosis are reported. Generated risk scores showed a significant difference in survival (p = 0.0001, log-rank test, Fig. 6C). Mean OS of 115.5 ± 15.8, 59.8 ± 13.8, 57.7 ± 9.9, and 14.6 ± 3.1 months were generated for risk scores of 1–2, 3, 4, and

> 4, respectively. PFS was not available in the secondary data set for evaluation.

DiscussionThis study supports the potential benefit of using MPC

status in the prognostication of lower-grade gliomas. In our Kaplan-Meier analysis of the 286 patients from the TCGA data set, MPC1 expression was significantly associated with improved OS of patients with IDH-mutant, specifi-cally 1p19q-codeleted gliomas. Hazards modeling showed that MPC1 expression resulted in significantly improved OS after accounting for other factors, including IDH mu-tation, 1p19q codeletion, age, and KPS score. By contrast, no such survival effects were seen in glioblastoma. Cox regression analysis of 21 TCGA cancer types confirmed an independent predictive role for MPC1 in lower-grade glioma but not glioblastoma or many other cancer types. Furthermore, our risk score model was able to predict OS using MPC status and was validated in a secondary data set (GSE16011). Although the use of MPC1 expression did not improve prediction of PFS, these results support the importance of MPC1 as an independent predictor of OS in lower-grade glioma in addition to IDH and 1p19q status.

Warburg Effect in GliomaCancer metabolism has been recognized as a key path-

way toward proliferation and survival (Fig. 7). The War-burg effect was initially described as cancer cells pre-dominantly using aerobic glycolysis rather than oxidative phosphorylation for bioenergetics;3 however, recent studies

FIG. 4. Association of survival effect of MPC with 1p19q status of IDH-mutant glioma. A and B: MPC1 expression separated survival of IDH-mutant glioma patients with 1p19q codeletion (A) but not those without (B); elevated MPC1 levels significantly improved survival only in 1p19q-codeleted tumors. C and D: Conversely, MPC2 expression differentiated survival in 1p19q-intact tumors (D) but not in 1p19q-codeleted tumors (C). These results support a prognostic role for MPC1 in oligodendrogliomas (1p19q-codeleted) and MPC2 in astrocytomas (1p19q-intact). *p < 0.05; **p < 0.01.




M. Karsy et al.


support the concept that cancer cells use aerobic glycoly-sis to divert pyruvate and other glycolytic metabolites to fuel anabolic processes for biosynthesis.11,34 Specifically in glioma, the Warburg effect has been suggested as key to maintaining the nondifferentiation of the cancer stem cell population through upregulation of hypoxia-inducible fac-tors.38

Although the role of MPC in cancer metabolism remains to be fully elucidated, several lines of research suggest its importance in cancer biology. Schell et al.33 showed that the MPC1 locus is frequently lost in a variety of cancers, including ovarian, colon, rectal, kidney, lung, bladder, and brain cancers. In addition, restoration of MPC1 expression inhibited cancer cell stemness as the potential mechanism

FIG. 5. Distribution of MPC1 and MPC2 z-scores in molecular subtypes of lower-grade glioma. Waterfall graphs of MPC1 (left) and MPC2 (right) z-scores in IDH–wild type glioma (A), IDH-mutant, 1p19q-intact glioma (B), and IDH-mutant, 1p19q-codeleted glioma (C). Distribution percentages are indicated on the y-axis. Figure is available in color online only.

TABLE 2. Prediction of OS and PFS for gliomas using a Cox proportional hazards model

VariableOS PFS

β p Value HR (95% CI) β p Value HR (95% CI)

IDH–wild type 2.59 0.0001 13.3 (4.5–39.0) 2.32 0.0001 10.2 (3.9–26.5)IDH-mutant, 1p19q-intact 0.36 0.5 1.4 (0.5–4.0) 0.71 0.12 2.0 (0.8–5.0)IDH-mutant, 1p19q-codeleted (reference) 0.0001 0.0001Age 0.04 0.01 1.04 (1.01–1.07) 0.004 0.71 1.00 (0.98–1.03)KPS score −0.03 0.03 0.97 (0.94–1.0) −0.02 0.26 0.98 (0.95–1.01)MPC1* −0.65 0.003 0.52 (0.34–0.80) −0.14 0.39 0.87 (0.63–1.20)MPC2* −0.28 0.39 0.76 (0.41– 1.4) −0.08 0.62 0.92 (0.68–1.27)MPC1 × MPC2* −0.13 0.59 0.88 (0.54–1.4) 0.04 0.68 1.04 (0.87–1.24)

* z-scores for MPC1 and MPC2.



M. Karsy et al.

tying cancer metabolism to differentiation. Blocking of MPC activity with the drug UK5099 in an esophageal cancer cell model triggered aerobic glycolysis, increased hypoxia-inducible factor 1a (HIF-1a) expression and reac-tive oxygen species production, and induced resistance to chemotherapy and radiotherapy.24 Higher MPC levels cor-

related with improved OS in prostate cancer23 and esopha-geal squamous cell carcinoma.24

Interestingly, our results suggest that mutant IDH in-teracts with MPC in glioma for metabolic adaptation (Fig. 7). Physiologically, IDH converts isocitrate to a-ketoglu-tarate, whereas mutant IDH reduces a-ketoglutarate to 2-hydroxyglutarate.3,10,35 2-hydroxyglutarate competitively inhibits multiple a-ketoglutarate-dependent dioxygenas-es, which leads to global DNA hypermethylation14,36 and a CpG island methylator phenotype (G-CIMP),8 induces histone methylation to block cell differentiation,26 and increases HIF-1a levels regardless of oxygen status (e.g.,

TABLE 3. FDR-ranked Cox coefficients of MPC1 and MPC2 among 21 cancer typesMPC1 MPC2

Cancer Cox Coefficient p Value FDR Corrected p Value Cancer Cox Coefficient p Value FDR Corrected p Value

KIRP −0.617 0.00024 0.00383 KIRC −0.147 0.073 0.147LGG −0.211 0.038 0.0741 SARC −0.222 0.037 0.202PAAD −0.271 0.014 0.0904 PAAD −0.197 0.065 0.211LUAD −0.187 0.015 0.104 COAD −0.27 0.0064 0.217CESC −0.357 0.006 0.155 LAML 0.157 0.17 0.567COAD −0.275 0.0059 0.213 BLCA −0.073 0.32 0.585LIHC −0.155 0.098 0.316 OSC −0.107 0.14 0.712KIRC −0.095 0.21 0.328 KIRP 0.093 0.56 0.721BLCA −0.063 0.39 0.647 LGG 0.041 0.66 0.74BRCA 0.086 0.34 0.696 SKCM 0.035 0.63 0.779LAML −0.101 0.36 0.728 CESC −0.073 0.59 0.841SARC −0.076 0.47 0.728 STAD 0.039 0.65 0.885SKCM −0.036 0.61 0.765 LIHC −0.028 0.76 0.895OSC 0.089 0.23 0.79 LUAD 0.019 0.81 0.907GBM –0.122 0.15 0.807 LUSC −0.024 0.74 0.94HNSC −0.043 0.55 0.811 GBM 0.043 0.61 0.945LUSC −0.019 0.79 0.953 BRCA 0.016 0.86 0.951STAD −0.016 0.85 0.955 READ 0.169 0.37 0.953ESCA 0.172 0.24 0.972 ESCA 0.139 0.29 0.973READ −0.121 0.53 0.972 HNSC −0.002 0.97 0.989UCEC −0.009 0.93 0.995 UCEC −0.087 0.38 0.992

BLCA = bladder urothelial carcinoma; BRCA = breast invasive carcinoma; CESC = cervical squamous cell carcinoma and endometrial adenocarcinoma; COAD = colon adenocarcinoma; ESCA = esophageal carcinoma; GBM = glioblastoma; HNSC = head and neck squamous cell carcinoma; KIRC = kidney renal clear cell carcinoma; KIRP = kidney renal papillary cell carcinoma; LAML = acute myeloid leukemia; LGG = brain lower grade glioma; LIHC = liver hepatocellular carcinoma; LUAD = lung adenocarcinoma; LUSC = lung squamous cell carcinoma; OSC = ovarian serous cystadenocarcinoma; PAAD = pancreatic adenocarcinoma; READ = rectum adenocar-cinoma; SARC = sarcoma; SKCM = skin cutaneous melanoma; STAD = stomach adenocarcinoma; UCEC = uterine corpus endometrial carcinoma.Cox regression analysis was performed using OncoLnc. FDR < 0.25 is considered significant. FDR-insignificant cancer types are shaded. LGG and GBM are in bold-face type.

TABLE 4. Risk score for lower-grade gliomas

Variable Score

Mutational status IDH-mutant/1p19q-codeleted IDH-mutant/1p19q-intact IDH–wild type

123

Age in yrs <50 ≥50

01

KPS score ≥80 <80

01

MPC1 negative z-score 1MPC2 positive z-score 1

TABLE 5. Sample validation patient outcomes for cumulative risk scores listed in Table 4

Cumulative Score

Mean Survival in Mos ± SDSample OS Sample PFS Validation OS

1–2 95.7 ± 0.1 68.3 ± 12.2 115.5 ± 15.83 136.2 ± 17.7 82.7 ± 15.2 59.8 ± 13.84 90.4 ± 9.1 57.0 ± 4.6 57.7 ± 9.9

>4 61.7 ± 7.0 54.8 ± 9.2 14.6 ± 3.1


M. Karsy et al.


pseudohypoxia).15,21 To maintain 2-hydroxyglutarate pro-duction, an adequate supply of a-ketoglutarate is required, as supported by the overexpression of Krebs cycle genes in IDH-mutant glioma.22 Therefore, the requirement of Krebs cycle in 2-hydroxyglutarate production supports the notion that increased MPC1 expression predicts improved surviv-al of patients with IDH-mutant gliomas, perhaps through the inhibition of the Warburg effect.

Biomarkers of GliomasThe use of mutational biomarkers for diagnosis and

prognosis in gliomas has been a paradigm.20 The mean age of diagnosis ranges from 37 to 59 years of age, be-ing higher in IDH–wild type than in IDH-mutant glioma patients. In addition, survival differs greatly among tumor types, with a median survival > 10 years for IDH-mutant, 1p19q-codeleted tumors (oligodendroglioma), > 10 years for IDH-mutant, TERT mutated tumors (astrocytoma), 8 years for IDH-mutant, ATRX-mutated tumors (astrocytic or oligoastrocytic), 3.5 years for IDH–wild type, 1p19q-intact, TERT–wild type (primary glioblastoma), and 1.8 years for TERT-mutated tumors (primary glioblastoma). Our risk score reflects some of these findings, supporting the importance of IDH and 1p19q status6,31 but also sug-gesting the potential clinical role of MPC. In addition, age and KPS were factored in the risk score; however, other significant tumor-specific features seen in prior stud-ies, such as size and eloquent cortex, were not evaluated. MPC1 (HR 0.52) was significant in our multivariate haz-ard model but showed a lower survival benefit than IDH or 1p19q status (HR 0.08–0.11) as assessed by the com-parative HR in the model. The use of gene expression plat-forms greatly amplifies the ability to segregate tumors but is not yet practical for widespread clinical use. Antibodies for MPC1 and MPC2 do currently exist, and immunohis-tochemical confirmation in a larger set of glioma patients may be achievable. Additional refinement of the model could be possible with additional data, but the goal should be to make the tool clinically relevant.

TABLE 6. Validation data set demographics

Variable Value

No. of patients 109Mean age in yrs 46.7 ± 13.2No. of male patients 72 (66.1)KPS score 10 20 30 40 50 60 70 80 90 100 Unknown

1 (0.9)3 (2.8)1 (0.8)4 (3.7)2 (1.8)8 (7.3)8 (7.3)

16 (14.7)19 (17.4)39 (35.8)

8 (7.3)Mutation status IDH–wild type IDH-mutant/1p19q-intact IDH-mutant/1p19q-codeleted

28 (25.7)23 (21.1)39 (35.8)

Mean MPC1 z-score 0.08 ± 0.95Mean MPC2 z-score −3.03 ± 0.91Histological diagnosis Oligodendroglioma Anaplastic oligodendroglioma Oligoastrocytoma Anaplastic oligoastrocytoma Astrocytoma Anaplastic astrocytoma

8 (7.3)44 (40.4)

3 (2.8)25 (22.9)13 (11.9)16 (14.7)

Mean OS in mos 53.3 ± 52.1

Values are presented as the number of patients (%) unless stated otherwise.

FIG. 6. Kaplan-Meier survival analysis utilizing risk score for gliomas incorporating MPC status. Survival analysis was performed utilizing log-rank tests and a scoring system. A and B: Statistical correlation of the risk scores was observed with OS (A, p = 0.001) but not with PFS (B, p = 0.45). C: Predicted risk scores were validated on an independent lower-grade glioma data set (p = 0.0001). Figure is available in color online only.



M. Karsy et al.

Although higher MPC1 levels correlated with im-proved survival in IDH-mutant glioma, the role of MPC2 remains less clear. MPC2 expression predicted poorer OS in IDH-mutant, 1p19q-intact gliomas, as supported by a positive Cox coefficient, albeit statistically insignificant, in lower-grade glioma. Schell et al.33 showed that MPC1 and MPC2 levels correlated positively with each other; however, MPC1 transcript levels were frequently down-regulated in cancer whereas MPC2 transcript levels had no clear pattern. Similarly, our results showed a correlation between MPC1 and MPC2 mRNA levels but revealed an opposing impact of MPC1 and MPC2 on survival. Fur-ther studies are warranted to elucidate the oligomeric ratio of MPC1 and MPC2 subunits required for the formation of MPC complex.2 Alternatively, MPC2 may participate independently in other biological processes, as suggested recently.28

LimitationsWe validated the survival effects of MPC1 and MPC2

with multiple data sets involving glioblastoma, additional cases of lower-grade glioma, and 19 other cancer types. Validation of the glioma risk score was conducted using

a secondary data set. Further validation via immunohisto-chemical analysis and prospective study may improve the utility of these markers. Limitations of the primary TCGA data set used for this study include censorship of data in our Kaplan-Meier survival analysis and limited follow-up in some risk score groups. These limitations may account for why a longer survival was seen for the third risk group compared with groups 1 and 2. In addition, limited knowl-edge of the roles of MPC1 and MPC2 in cancer hinders the ability to fully understand the altered signaling path-way in glioma. Therefore, further investigation of MPC biology in glioma metabolism is an avenue of exploration.

ConclusionsOur results support a role for MPC1 overexpression and

MPC2 underexpression in improving OS for patients with IDH-mutant but not IDH–wild type glioma. MPC1 over-expression improved OS of patients with 1p19q-codeleted tumors, whereas MPC2 underexpression improved OS of patients with 1p19q-intact tumors. Using a hazards model, a risk score can be generated to allow clinical use of these multiple signaling pathways. Our study suggests that addi-tional classification of IDH-mutant glioma patients is pos-sible with novel genetic markers to improve clinical treat-ments and to promote hypothesis-driven research.

AcknowledgmentsWe thank Kristin Kraus, MSc, for her editorial assistance and

Howard Colman, MD, PhD, for his thoughtful comments on the manuscript. This work was supported in part by the Huntsman Cancer Foundation.

References 1. Brat DJ, Verhaak RG, Aldape KD, Yung WK, Salama SR,

Cooper LA, et al: Comprehensive, integrative genomic analy-sis of diffuse lower-grade gliomas. N Engl J Med 372:2481–2498, 2015

2. Bricker DK, Taylor EB, Schell JC, Orsak T, Boutron A, Chen YC, et al: A mitochondrial pyruvate carrier required for pyruvate uptake in yeast, Drosophila, and humans. Science 337:96–100, 2012

3. Cairns RA, Harris IS, Mak TW: Regulation of cancer cell metabolism. Nat Rev Cancer 11:85–95, 2011

4. Ceccarelli M, Barthel FP, Malta TM, Sabedot TS, Salama SR, Murray BA, et al: Molecular profiling reveals biologi-cally discrete subsets and pathways of progression in diffuse glioma. Cell 164:550–563, 2016

5. Cerami E, Gao J, Dogrusoz U, Gross BE, Sumer SO, Aksoy BA, et al: The cBio cancer genomics portal: an open platform for exploring multidimensional cancer genomics data. Can-cer Discov 2:401–404, 2012

6. Chang EF, Smith JS, Chang SM, Lamborn KR, Prados MD, Butowski N, et al: Preoperative prognostic classification sys-tem for hemispheric low-grade gliomas in adults. J Neuro-surg 109:817–824, 2008

7. Chaumeil MM, Larson PE, Woods SM, Cai L, Eriksson P, Robinson AE, et al: Hyperpolarized [1-13C] glutamate: a metabolic imaging biomarker of IDH1 mutational status in glioma. Cancer Res 74:4247–4257, 2014

8. Christensen BC, Smith AA, Zheng S, Koestler DC, House-man EA, Marsit CJ, et al: DNA methylation, isocitrate dehy-drogenase mutation, and survival in glioma. J Natl Cancer Inst 103:143–153, 2011

FIG. 7. Molecular pathway of MPC and IDH1/2. A molecular sche-matic of the glycolytic and Krebs cycle pathways depicts the potential interaction between MPC and IDH. The MPC1/2 complex mediates pyruvate transport through the inner mitochondrial membrane to provide a-ketoglutarate, which can ultimately serve as substrate for cytosolic conversion to 2-hydroxyglutarate (2-OG) by mutant IDH. Increased 2-hydroxyglutarate production results in a global hypermethylation phe-notype in gliomas along with increased H3K9/H3K27 methylation and HIF-1 stabilization. GLUT = glucose transporter; MCT = monocarboxylic acid transporter. Figure is available in color online only.


M. Karsy et al.


9. Cohen AL, Colman H: Glioma biology and molecular mark-ers. Cancer Treat Res 163:15–30, 2015

10. Dang L, White DW, Gross S, Bennett BD, Bittinger MA, Driggers EM, et al: Cancer-associated IDH1 mutations pro-duce 2-hydroxyglutarate. Nature 462:739–744, 2009

11. DeBerardinis RJ, Lum JJ, Hatzivassiliou G, Thompson CB: The biology of cancer: metabolic reprogramming fuels cell growth and proliferation. Cell Metab 7:11–20, 2008

12. Eckel-Passow JE, Lachance DH, Molinaro AM, Walsh KM, Decker PA, Sicotte H, et al: Glioma groups based on 1p/19q, IDH, and TERT promoter mutations in tumors. N Engl J Med 372:2499–2508, 2015

13. Esmaeili M, Hamans BC, Navis AC, van Horssen R, Bathen TF, Gribbestad IS, et al: IDH1 R132H mutation generates a distinct phospholipid metabolite profile in glioma. Cancer Res 74:4898–4907, 2014

14. Figueroa ME, Abdel-Wahab O, Lu C, Ward PS, Patel J, Shih A, et al: Leukemic IDH1 and IDH2 mutations result in a hy-permethylation phenotype, disrupt TET2 function, and im-pair hematopoietic differentiation. Cancer Cell 18:553–567, 2010

15. Fu Y, Zheng S, Zheng Y, Huang R, An N, Liang A, et al: Gli-oma derived isocitrate dehydrogenase-2 mutations induced up-regulation of HIF-1a and b-catenin signaling: possible impact on glioma cell metastasis and chemo-resistance. Int J Biochem Cell Biol 44:770–775, 2012

16. Gao J, Aksoy BA, Dogrusoz U, Dresdner G, Gross B, Sumer SO, et al: Integrative analysis of complex cancer genomics and clinical profiles using the cBioPortal. Sci Signal 6:pl1, 2013

17. Gravendeel LA, Kouwenhoven MC, Gevaert O, de Rooi JJ, Stubbs AP, Duijm JE, et al: Intrinsic gene expression profiles of gliomas are a better predictor of survival than histology. Cancer Res 69:9065–9072, 2009

18. Herzig S, Raemy E, Montessuit S, Veuthey JL, Zamboni N, Westermann B, et al: Identification and functional expression of the mitochondrial pyruvate carrier. Science 337:93–96, 2012

19. Huang LE, Cohen AL, Colman H, Jensen RL, Fults DW, Couldwell WT: IGFBP2 expression predicts IDH-mutant glioma patient survival. Oncotarget 8:191–202, 2017

20. Karsy M, Guan J, Cohen AL, Jensen RL, Colman H: New molecular considerations for glioma: IDH, ATRX, BRAF, TERT, H3 K27M. Curr Neurol Neurosci Rep 17:19, 2017

21. Karsy M, Guan J, Jensen R, Huang LE, Colman H: The im-pact of hypoxia and mesenchymal transition on glioblastoma pathogenesis and cancer stem cells regulation. World Neuro-surg 88:222–236, 2016

22. Khurshed M, Molenaar RJ, Lenting K, Leenders WP, van Noorden CJF: In silico gene expression analysis reveals gly-colysis and acetate anaplerosis in IDH1 wild-type glioma and lactate and glutamate anaplerosis in IDH1-mutated glioma. Oncotarget 8:49165–49177, 2017

23. Li X, Ji Y, Han G, Li X, Fan Z, Li Y, et al: MPC1 and MPC2 expressions are associated with favorable clinical outcomes in prostate cancer. BMC Cancer 16:894, 2016

24. Li Y, Li X, Kan Q, Zhang M, Li X, Xu R, et al: Mitochon-drial pyruvate carrier function is negatively linked to War-burg phenotype in vitro and malignant features in esophageal squamous cell carcinomas. Oncotarget 8:1058–1073, 2017

25. Louis DN, Perry A, Reifenberger G, von Deimling A, Fig-arella-Branger D, Cavenee WK, et al: The 2016 World Health Organization Classification of Tumors of the Central Nervous System: a summary. Acta Neuropathol 131:803–820, 2016

26. Lu C, Ward PS, Kapoor GS, Rohle D, Turcan S, Abdel-Wa-hab O, et al: IDH mutation impairs histone demethylation and results in a block to cell differentiation. Nature 483:474–478, 2012

27. Ostrom QT, Gittleman H, Fulop J, Liu M, Blanda R, Kromer C, et al: CBTRUS statistical report: primary brain and cen-

tral nervous system tumors diagnosed in the United States in 2008–2012. Neuro Oncol 17 (Suppl 4):iv1–iv62, 2015

28. Otero JM, Papadakis MA, Udatha DB, Nielsen J, Panagiotou G: Yeast biological networks unfold the interplay of antioxi-dants, genome and phenotype, and reveal a novel regulator of the oxidative stress response. PLoS One 5:e13606, 2010

29. Papa S, Francavilla A, Paradies G, Meduri B: The transport of pyruvate in rat liver mitochondria. FEBS Lett 12:285–288, 1971

30. Parsons DW, Jones S, Zhang X, Lin JC, Leary RJ, Angenendt P, et al: An integrated genomic analysis of human glioblas-toma multiforme. Science 321:1807–1812, 2008

31. Pignatti F, van den Bent M, Curran D, Debruyne C, Sylves-ter R, Therasse P, et al: Prognostic factors for survival in adult patients with cerebral low-grade glioma. J Clin Oncol 20:2076–2084, 2002

32. Reitman ZJ, Jin G, Karoly ED, Spasojevic I, Yang J, Kinzler KW, et al: Profiling the effects of isocitrate dehydrogenase 1 and 2 mutations on the cellular metabolome. Proc Natl Acad Sci U S A 108:3270–3275, 2011

33. Schell JC, Olson KA, Jiang L, Hawkins AJ, Van Vranken JG, Xie J, et al: A role for the mitochondrial pyruvate carrier as a repressor of the Warburg effect and colon cancer cell growth. Mol Cell 56:400–413, 2014

34. Vander Heiden MG, Cantley LC, Thompson CB: Understand-ing the Warburg effect: the metabolic requirements of cell proliferation. Science 324:1029–1033, 2009

35. Ward PS, Patel J, Wise DR, Abdel-Wahab O, Bennett BD, Coller HA, et al: The common feature of leukemia-associated IDH1 and IDH2 mutations is a neomorphic enzyme activity converting a-ketoglutarate to 2-hydroxyglutarate. Cancer Cell 17:225–234, 2010

36. Xu W, Yang H, Liu Y, Yang Y, Wang P, Kim SH, et al: On-cometabolite 2-hydroxyglutarate is a competitive inhibitor of a-ketoglutarate-dependent dioxygenases. Cancer Cell 19:17–30, 2011

37. Yan H, Parsons DW, Jin G, McLendon R, Rasheed BA, Yuan W, et al: IDH1 and IDH2 mutations in gliomas. N Engl J Med 360:765–773, 2009

38. Yuen CA, Asuthkar S, Guda MR, Tsung AJ, Velpula KK: Cancer stem cell molecular reprogramming of the Warburg effect in glioblastomas: a new target gleaned from an old concept. CNS Oncol 5:101–108, 2016

DisclosuresThe authors report no conflict of interest concerning the materi-als or methods used in this study or the findings specified in this paper.

Author ContributionsConception and design: Huang. Acquisition of data: Huang. Anal-ysis and interpretation of data: all authors. Drafting the article: Huang, Karsy. Critically revising the article: all authors. Reviewed submitted version of manuscript: all authors. Approved the final version of the manuscript on behalf of all authors: Huang. Statisti-cal analysis: Karsy, Guan.

Supplemental Information Online-Only ContentSupplemental material is available with the online version of the article.

Supplementary Table and Figures. https://thejns.org/doi/suppl/ 10.3171/2017.9.JNS172036.

CorrespondenceL. Eric Huang: University of Utah, Salt Lake City, UT. [email protected].




prognostic role of mitochondrial pyruvate carrier in

Documents