MR Imaging Characteristics Associate with Tumor-Associated Macrophages in Glioblastoma and Provide an Improved Signature for Survival Prognostication

BACKGROUND AND PURPOSE: In glioblastoma, tumor-associated macrophages have tumor-promoting properties. This study determined whether routine MR imaging features could predict molecular subtypes of glioblastoma that differ in the content of tumor-associated macrophages. MATERIALS AND METHODS: Seven internally derived MR imaging features were assessed in 180 patients, and 25 features from the Visually AcceSAble Rembrandt Images feature set were assessed in 164 patients. Glioblastomas were divided into subtypes based on the telomere maintenance mechanism: alternative lengthening of telomeres positive (ALT+) and negative (ALT−) and the content of tumor-associated macrophages (with [M+] or without [M−] a high content of macrophages). The 3 most frequent subtypes (ALT+/M−, ALT−/M+, and ALT−/M−) were correlated with MR imaging features and clinical parameters. The fourth group (ALT+/M+) did not have enough cases for correlation with MR imaging features. RESULTS: Tumors with a regular margin and those lacking a fungating margin, an expansive T1/FLAIR ratio, and reduced ependymal extension were more frequent in the subgroup of ALT+/M− (P < .05). Radiologic necrosis, lack of cystic component (by both criteria), and extensive peritumoral edema were more frequent in ALT−/M+ tumors (P < .05). Multivariate testing with a Cox regression analysis found the cystic imaging feature was additive to tumor subtype, and O6-methylguanine methyltransferase (MGMT) status to predict improved patient survival (P < .05). CONCLUSIONS: Glioblastomas with tumor-associated macrophages are associated with routine MR imaging features consistent with these tumors being more aggressive. Inclusion of cystic change with molecular subtypes and MGMT status provided a better estimate of survival.

G lioblastomas are the most common and aggressive primary malignant brain tumor, with an incidence of 5.26 per 100,000 population. 1 With temozolomide and radiation therapy, the median survival is 14.6 months. 1 Patients' response to treatment and prognosis are notoriously variable. Many studies have attributed heterogeneous outcomes to molecular differences among glioblastomas, such as different means to maintain telomere integrity (telomerase activity or an alternative method known as the alternative lengthening of telomeres [ALT]). 2 In 2003, the presence of the ALT telomere maintenance mechanism was associated with longer patient survival compared with ALT negative (ALTϪ) tumors, including those that were telomerase positive (ALTϩ) and those where the telomere maintenance mechanism was unknown. 3 It has been recently found that most tumors without a defined telomere maintenance mechanism contain a high con-tent of tumor-associated macrophages (Mϩ) that affects assays assigning telomere maintenance. 4 The revised subgroups were ALT positive tumors with and without (MϪ) a high content of macrophages (ALTϩ/Mϩ and ALTϩ/MϪ) and ALT negative tumors with and without a high content of macrophages (ALTϪ/Mϩ and ALTϪ/MϪ). With temozolomide treatment, patients with ALTϪ/MϪ tumors have improved survival compared with patients with ALTϪ/Mϩ tumors, 4 suggesting that identifying the increased macrophage content will distinguish those with a poor outcome.
MR imaging is a diagnostic tool that evaluates tumor as well as peritumoral characteristics. Improving prognostic determination based on MR imaging features alone has associated radiologic necrosis and peritumoral edema (PTE) with poorer survival. [5][6][7][8] Tumors with a cystic component have been associated with improved patient survival. 9,10 However, inconsistencies occur between studies, and MR imaging features are not always found as independent factors associated with survival. [11][12][13][14] Imaging features have been attributed to differences in tumor biology, suggesting an analysis of MR imaging features combined with molecular characteristics may improve prognostic prediction. Isocitrate dehydrogenase 1 (IDH1) mutations, O6-methylguanine methyltransferase (MGMT) promoter methylation, and epidermal growth factor receptor overexpressing tumors have been correlated with MR imaging features. 13,[15][16][17][18][19] That tumor biology influences imaging features is further evident in studies that investigate subgroups of glioblastomas based on multiple gene expression differences. 20 This study aimed to determine whether MR imaging features were associated with the telomere maintenance mechanism and tumor-associated macrophage content-based subtypes. The inclusion of MR imaging parameters with the molecular subtyping and MGMT promoter methylation status 21 was also investigated to determine whether it could better predict outcome for patients with glioblastoma.

Study Participants
One hundred eighty patients from Dunedin, Christchurch, and Waikato hospitals in New Zealand with a diagnosis of glioblastoma between the years 2004 and 2014 were included in the study. The cohort was aged between 16 -82 years (mean, 61 years; 95% CI, 59 -63); 39% were female and 61% male; and 93% were white, 3% Maori, and 4% identified with other ethnic groups. All patients were eligible for standard of care therapy (surgery, radiation therapy, and temozolomide), and no patients received other therapies. The participants were all diagnosed with glioblastoma after the Stupp protocol was used in New Zealand. 1

Participant Selection/Inclusion Criteria
The inclusion criteria were a diagnosis of glioblastoma (as assessed by 2 pathologists independently), no previous lower-grade brain tumor or other brain surgery, and no previous radiation or chemotherapy. No patients received corticosteroids at the time of the preoperative MR imaging scan. Survival time was defined as the time interval (months) between the time of diagnosis (defined as the time of the initial preoperative MR imaging scan) and death.

MR Imaging Features and Interpretation
Preoperative MR imaging scans were evaluated including T1, T2, FLAIR, and postcontrast T1-weighted sequences after intravenous infusion of 10 mL of gadolinium. Images were taken by clinical 1.5T (Siemens, Erlangen, Germany or GE Healthcare, Milwaukee, Wisconsin) scanners. All study scans had routine tumor protocol sequences (T1, T2, FLAIR, DWI, ADC, and postcontrast gadolinium T1), and some had additional susceptibilityweighted sequences. At least 3 reviewers (3 neuroradiologists [M.V.R., B.K.J.W., and D.A.B.] and 1 radiology registrar [J.Z.]) read each MR imaging scan independently in the Public Hospital PACS. All readers were blinded to patient demographics, treatment regimen, and tumor molecular subtypes. The term "overall agreement" was defined as when 3 or more examiners agreed and the term "overall disagreement" was defined as when 2 or more examiners disagreed. In cases with disagreement between reviewers, the cases were reanalyzed in a collaborative fashion by at least 3 examiners, and a consensus score was reached and used in the final analysis.
Based on the current available MR imaging literature and the cumulative experiences from the 2 tertiary centers, 7 imaging features were analyzed, as detailed in Table 1. These features included tumor margin: regular or irregular ( Fig 1A, regular margin being a tumor with a smooth enhancing margin on postgadolinium T1 and irregular margin being a tumor that lacks a smooth enhancing margin on postgadolinium T1), fungating or nonfungating ( Fig 1B, a fungating margin being a tumor with a thick heterogeneous "brush"-like enhancing rim and nonfungating margin being a tumor that lacks a thick heterogeneous "brush"like enhancing rim), cystic component (Fig 1C), radiologic necrosis (Fig 1D), limited or extensive PTE (Fig 1E), multifocality ( Fig  1F), and hemorrhage ( Fig 1G).
The protocol for measuring PTE grade was based on that developed by Wu 5 and Hartmann. 22 The degree of the white matter edema was estimated on the basis of the maximal distance from the tumor margin to the furthest point of the white matter edema. The tumor's maximal dimension was estimated based on the maximal diameter of the tumor on any axis. When the degree of white matter edema was less or similar (no more than 4 mm greater) to the tumor's maximal dimension, the edema was considered low grade, and when the degree of white matter edema was greater (more than 4 mm) than the tumor's maximal dimension, the edema was considered a high-grade.

Telomere Maintenance Mechanism Subtyping
Tumors were classified into 4 subgroups: ALT positive tumors with or without a high content of tumor associated macrophages (ALTϩ/Mϩ and ALTϩ/MϪ, respectively) and ALT negative tumors with or without a high content of tumor associated macrophages (ALTϪ/Mϩ and ALTϪ/MϪ, respectively) based on established methods by using paraffin-embedded tumor sections. 4,25 Tumors from 80 cases were typed for telomere maintenance mechanisms as part of an earlier study, and 100 tumors were new to this study. 4 The tumor-associated macrophage content was identified by immunohistochemistry staining for CD163 positive macrophages. The CD163 antibody used was EPR19518 (Abcam, Cambridge, United Kingdom), and the macrophage content was estimated as described previously. 4

Mutant IDH1 and IDH2 Determination
The presence of the R132H IDH1 mutation was estimated by using immunohistochemistry. 26 To confirm the presence and frequency of IDH1 mutations and to estimate the number of IDH2 mutations, exon 4 of IDH1 and IDH2 were amplified from tumor-extracted DNA and sequenced. 27 Tumor DNA was extracted from paraffin embedded tumor sections or frozen tumor.

MGMT Promoter Methylation Determination
The method to determine whether tumors had a methylated or unmethylated MGMT promoter was based on published methods. 21, 28 In this study, genomic DNA was extracted from paraffinembedded tumors by using the QIAmp DNA FFPE tissue Kit (QIAGEN, Tokyo, Japan), and bisulfite DNA conversion was performed by using the EpiTect Bisulfite kit (QIAGEN).

Statistical Methods
The clinico-demographic and MR imaging features were compared between the 3 telomere maintenance-based subtypes by using Pearson 2 tests and 1-way ANOVA as appropriate. The statistical test was performed to assess the concordance between interobserver agreement. Logistic regression was used to test if imaging could predict the ALTϩ/MϪ and ALTϪ/Mϩ subtypes and used imaging features as covariants. Survival was compared between the subtypes by using the logrank test and the independent association of the MR imaging features in addition to the subtype effects. To test multiple variants on survival, Cox proportional hazards regression models were used. A 2-tailed P value Ͻ.05 was taken to indicate statistical significance.

RESULTS
Whether the 7 internally derived MR imaging features could be consistently identified was investigated by comparing MR imaging scores from the 3 different readers and calculating the percentage of scans on which all 3 blinded reviewers agreed on the presence or absence of each MR imaging feature. statistical analyses were performed. The interobserver data including value demonstrated reproducible results ( Table 2). The interobserver agreement rate was highest for the glioblastoma cystic component feature (86.2%, ϭ 0.71), followed by multifocal lesion (85.6%, ϭ 0.71). The margin-based parameters had the lowest interobserver agreement percentages (regular margin, 62.9%, ϭ 0.23; fungating margin, 64.6%, ϭ 0.28).

MR Imaging Features Predictive of ALT؉ and M؉ Status
In the 7 internally derived MR imaging feature groups, the clinico-demographic features and the frequency of IDH1 and IDH2 mutations and MGMT promoter methylation among the telomere maintenance subtypes are given in Table 3. Three telomere maintenance-based subtypes had sufficient numbers to investigate MR imaging features (ALTϩ/MϪ, n ϭ 26; ALTϪ/MϪ, n ϭ 92; and ALTϪ/Mϩ, n ϭ 56). Because of the lack of glioblastomas in the ALTϩ/Mϩ group (n ϭ 6), these cases were not included in the subsequent analyses. The frequency of each MR imaging feature among the telomere-based subtypes is given in Table 4. Five MR imaging features-regular margin, fungating margin, tumor cystic component, radiologic necrosis, and limited The ALTϩ/MϪ subgroup had the highest incidence of tumors with a regular margin (58% versus 30% in both the ALTϪ/MϪ and ALTϪ/Mϩ groups; P ϭ .026) and the lowest incidence of tumors with a fungating margin (19% versus 53% and 61% in the ALTϪ/MϪ and ALTϪ/Mϩ groups, respectively; P ϭ .002).
With the exception of PTE, the other MR imaging features were strongly associated with each other. Cystic tumors were more likely to have regular margins with no necrosis, and noncystic tumors were more likely to have irregular margins with necrosis (cystic association with regular margin, P Ͻ .0001; cystic with no necrosis, P Ͻ .0001).
To test whether imaging could predict ALTϩ/MϪ or an ALTϪ/Mϩ status, a logistic regression analysis was performed. Using the VARSARI feature set, 6 VASARI features-side of lesion center (F2), proportion of necrosis (F7), cysts (F8), T1/FLAIR ratio (F10), proportion of edema (F14), and ependymal extension (F19)-were found to be correlated with the 3 tumor subgroups. The full set of VASARI features and the associations with telomere maintenance glioblastoma subtypes is detailed in On-line Table 1.  The ALTϪ/Mϩ subgroup had the lowest incidence of tumors with a cyst (4%) compared with the ALTϩ/MϪ (30%) and the ALTϪ/MϪ (12%) groups (P ϭ .007), the highest incidence of tumors with a higher proportion of necrosis (71%) versus 49% for ALTϪ/MϪ and 42% for ALTϩ/M (P ϭ .026), and the highest incidence of tumors with the highest grade of edema (grade 5; 53%) compared with 32% in the ALTϪ/MϪ and 14% in the ALTϩ/MϪ group (P ϭ .002).
The ALTϩ/MϪ subgroup had the highest incidence of tumors with an expansive T1/FLAIR ratio (79%) compared with 41% in the ALTϪ/MϪ and 37% in the ALTϪ/Mϩ group (P ϭ .0003), the lowest incidence of tumors located in the right side of the brain (17%) compared with 41% in the ALTϪ/MϪ and 54% in the ALTϪ/Mϩ group (P ϭ .007), and the lowest incidence of tumors with ependymal extension (30%) compared with 60% in the ALTϪ/MϪ and 49% in the ALTϪ/Mϩ group (P ϭ .036).

MR Imaging Features Predict Survival Beyond Molecular Status
Consistent with earlier studies, the ALTϩ/MϪ subgroup patients had the best survival (Fig 2), and the ALTϪ/MϪ subgroup patients had an improved overall survival compared with the    3,4,29 There were no significant differences (P Ͼ .05) among telomere maintenance-based subgroups in relation to the amount of patients treated with radiation therapy, concurrent temozolomide, or adjuvant temozolomide.
In an earlier study and in this study (Table 4), the ALTϪ/Mϩ subgroup was associated with an increased frequency of partial resections compared with near-total resections. 4 A Cox regression analysis by using forward and backward stepwise models was used to determine whether 4 imaging features (margin regularity, cystic, necrosis, and PTE grade) added to telomere maintenance-based subtype and MGMT promoter methylation status to better determine patient prognosis. Cystic feature was significant (P ϭ .045) in predicting patients' survival independent of tumor subtype and MGMT promoter methylation status (Table 6).

DISCUSSION
An assessment of MR imaging features is an advantage to using histologic-based subtypes alone because the characteristics of whole tumor in situ can be assessed before surgery. In this study, we found radiologic features (including the VASARI MR imaging) associated with telomere maintenance and tumor-associated macrophage content-based subtypes, consistent with MR imaging features being informative in identifying differences in tumor biology. A cystic component was additive in predicting an improved patient outcome, suggesting that inclusion of this MR imaging feature along with an assessment of telomere maintenance subtypes and MGMT promoter methylation status will improve patient outcome prediction. Both feature sets had imaging parameters that predicted the ALTϩ/MϪ and the ALTϪ/Mϩ subgroups.
Because temozolomide use ALTϪ/Mϩ tumors are associated with the poorest survival independent of patient age, extent of surgery, and treatment received. 4 ALTϪ/Mϩ tumors were associated with a higher proportion of necrosis, extensive edema, and lack of a cystic component. Whether the increased macrophage content directly contributes to the associated MR imaging features is unknown. Tumor-associated macrophages have many tumor-promoting activities including proinvasive and proangiogenesis properties. 30 Increased edema may result from increased vascular permeability. Increased vascular endothelial growth factor and angiogenesis correlated with edema, 31,32 and a molecular marker found on endothelial cells (deltalike ligand 4) that functions in angiogenesis was associated with PTE and poorer patient survival. 33 Consistent with a link between PTE and angiogenesis, patients treated with the vascular endothelial growth factor inhibitor bevacizumab had reduced PTE. 34 Increased edema has been associated with poorer prognosis. [6][7][8] In the study by Carillo et al, 13 edema was additive for survival for MGMT methylated but not unmethylated tumors. In the current study, edema was not additive to survival by tumor subtype; this is likely attributable to edema being closely associated with the ALTϪ/Mϩ subtype. In a future study, we aim to incorporate quantitative MR imaging techniques, more specifically DTI in vivo to demonstrate the PTE grade. 35 The imaging features associated with the ALTϪ/Mϩ group are consistent with this subgroup comprising aggressive tumors. The increased content of macrophages in the ALTϪ/Mϩ subtype, combined with these tumors having a poorer prognosis that has not changed with temozolomide, suggests this subtype would be an ideal target for immunotherapies, particularly those that could activate macrophages to remove the malignant cells or those that inhibit the tumor-promoting activities of tumor-associated macrophages. 4,36 Alternative lengthening of telomere-positive tumors is associated with the best overall patient survival independent of patient age, extent of surgery, treatment received, and IDH1 mutant status. 3,4,29 In this study, ALTϩ tumors were more likely to have a regular margin, lack of a fungating margin, cystic component, expansive T1/FLAIR ratio, and lack of ependymal extension, consistent with ALTϩ tumors being less aggressive. The MR imaging data from other studies support ALTϩ tumors being less aggressive. 29,37,38 Sharper margins and less contrast enhancement by MR imaging have been associated with features of ALTϩ tumors, including mutant IDH1, mutant tumor protein p53 (TP53), or proneural tumors. 13,16,18,20,39,40 Considering that most ALTϩtumors in this study had mutant IDH1 or IDH2, MR imaging features of ALT may be due to mutant IDH. The expansive T1/FLAIR VASARI feature had greater significance when tumors were compared based on IDH mutant status rather than telomere maintenance subtype.
The presence of the cystic component was associated with improved survival. A cystic component has been associated with improved survival in some, 9,10 but not all studies. 6,12,13 If molec- The 3 telomere maintenance mechanism subtypes are associated with differences in patient survival. Eighty cases were those used in a previous study, 4 and 100 cases were new to this study. , and most are thought to arise de novo. The cystic component in ALTϪ/MϪ tumors may arise because of a different mechanism to that in ALTϩ tumors. Why ALTϪ/MϪ tumors with a cystic component are associated with improved survival is unclear. The cystic component may represent a distinct tumor biology that responds well to treatment, or it may directly aid in the acquisition or detainment of temozolomide. The cystic component was reproducibly identified among different radiologists. A larger study to investigate whether ALTϪ/MϪ tumors with a cystic component are those that respond well to temozolomide is warranted.
Limitations of this study include its retrospective nature. The study excluded patients previously treated with radiation therapy to prevent factors that would affect MR imaging interpretation. This would exclude more patients with secondary glioblastoma that are more likely to be ALTϩ. Thus, the MR imaging features of the ALTϩ group will be representative of a subset of ALTϩ tumors.

CONCLUSIONS
The radiologic feature of cystic component was a predictive factor for survival and could be combined with telomere maintenance mechanism and macrophage content-based subtypes and MGMT promoter methylation status to provide a better estimation of survival. A lack of cystic component, a lack of left-sided tumor epicenter, higher necrosis and edema, along with a higher ependymal extension are more often present in tumors with a high tumor-associated macrophage content, suggesting that the presence of macrophages leads to a more invasive tumor behavior and a greater difficulty in obtaining near total resection, which could be predicted by MR imaging.