Introduction

WHO grade II and III cerebral gliomas, also recently termed lower grade gliomas (LGGs), comprise a heterogeneous group of infiltrative neoplasms with astrocytic and oligodendroglial morphology. These tumors have a wide range of both progression-free survival (PFS) and overall survival (OS); some respond to therapy with OS approaching 15 years, while others rapidly progress to glioblastoma [1]. Traditional histopathological classification and grading, the conventional gold standard, is limited by high intraobserver and interobserver variability. Subsequently, this uncertainty in predicting a tumor’s pathogenesis can complicate the selection of optimal treatment and estimation of a patient’s eventual outcome [2, 3].

During the past several decades, tremendous strides have been made to elucidate the underlying molecular alterations present in gliomas [4]. Codeletion of chromosome arms 1p and 19q (henceforth termed 1p/19q codel) was first established as an association with oligodendroglial morphology; subsequently, it was validated as a prognostic marker and shown to be predictive of response to chemoradiation [5,6,7,8]. Following this, isocitrate dehydrogenase gene mutations (referred to as IDH mutations, and reflecting either a mutation in IDH1 or IDH2) were identified; this (collective) mutation is thought to be an early step in gliomagenesis, and estimated to occur in 70–90% of grade II/III gliomas. IDH mutations have also been well-validated as both prognostic and predictive markers of prolonged survival relative to IDH-wildtype tumors [9,10,11,12,13,14]. More recently, explorations into mutations of the alpha-thalassemia/ mental retardation syndrome X-linked gene (termed ATRX), revealed that its loss is more common in astrocytomas, and that it is mutually exclusive with 1p/19q codel [15,16,17]. Furthermore, initial studies suggest it may be associated with a favorable prognosis [17, 18]. Similarly, TP53 mutations are also more frequent in astrocytomas, a marker of clinical outcome and mutually exclusive with 1p/19q codel, thereby commonly co-existing with ATRX mutations [1].

Adding momentum to this field, two recent studies have demonstrated that molecular characterization of glioma including IDH and 1p/19q status is more robust for prediction of clinical outcomes, compared with histological classification. In particular, it was determined that IDH-wildtype gliomas resemble aggressive grade IV gliomas clinically and molecularly, suggesting these should be termed “stealth glioblastomas” [1, 4]. Recently, these efforts were validated in the 2016 World Health Organization Classification of Tumors of the Central Nervous System in which molecular parameters were integrated with histopathology into glioma characterization to create more biologically homogenous groups and improve predictive and prognostic accuracy [19]. As a complement, large-scale sequencing studies have helped define three groups of clinically and biologically distinct gliomas delineated by IDH mutation with 1p/19q codeletion (and TERT promoter mutation), IDH mutation with TP53 mutation and frequent ATRX mutation, and IDH-wildtype status (with TERT promoter mutation and glioblastoma-associated genomic changes) [20]. Subsequently, following from these collective guidelines, we classified grade II/III gliomas into 3 distinct molecular subgroups: IDH-wildtype (IDH-WT), IDH-mutant-1p/19q codeleted-ATRX intact (IDH-MT1p/19qcodel) and IDH-mutant-1p/19q intact-ATRX loss (IDH-MT ATRXloss), with the final group also characteristically having a TP53 mutation.

While molecular analyses can distinguish these subgroups, difficulties in creating uniform genotype testing and reporting can limit its utility, and the technique is also dependent on tissue sampling [21]. MRI is the primary modality for initial diagnostic work-up of gliomas prior to invasive biopsy. As such, distinct imaging features could be defined on MRI reflecting the subgroups underlying molecular differences and yielding essential predictive and prognostic information at the initial diagnostic MRI to help guide clinical decisions. While previous studies have demonstrated that IDH and 1p/19q status have some distinct imaging features in selective subsets of gliomas, there is a dearth of studies evaluating these relationships collectively including both grades. In this study, we hypothesized that these aforementioned grade II/III glioma subgroups, devised from the 2016 WHO classification and large-scale sequencing studies, have unique imaging signatures that may aid in the prediction of patient outcomes.

Methods

Patient cohort

A single institution retrospective study was performed at the Moores Cancer Center at the University of California, San Diego with institutional review board approval. From April 2008 to March 2015, 57 patients were identified who had a pathologically proven grade II or III glioma. Patients were excluded for lack of pre-treatment imaging (defined as prior to surgical resection or initiation of chemoradiotherapy), incomplete imaging (defined as incomplete pre- or post-contrast, FLAIR, ADC and/or DWI sequences), incomplete molecular data (IDH, 1p/19q, and ATRX status) or tumor upstaging to glioblastoma. Upon neuroradiologist review of the cases, one patient was excluded for having a collision lesion: a concomitant arteriovenous malformation and glioma. The final cohort consisted of 40 patients. The specifics are delineated in the STARD chart (Online Resource 1).

Molecular analysis

1p/19q status was determined by either OncoScan microarray analysis (Affymetrix, Santa Clara, CA) or fluorescence in situ hybridization (FISH) performed on formalin fixed paraffin embedded tissue. The majority of IDH and all ATRX status were analyzed by whole exome next-generation sequencing (Illumina, San Diego, CA), as previously described by Dubbink et al. [22]. A minority of IDH status was determined by mutation-specific immunohistochemistry (IHC) for the most common IDH1 mutation (R132H). Following this, the tumors were then classified into three subgroups: Group (1) IDH-WT; Group (2) IDH-MT1p/19qcodel and Group (3) IDH-MT ATRXloss.

Imaging technique and analysis

All of the pre-treatment MR scans were acquired on either a 1.5T or 3T MR unit and included T1 pre- and post-contrast sequences, FLAIR, and DWI/ADC. The exact scanning parameters were variable as some of the scans were performed at outside institutions prior to the patient’s referral to our center. However, all scans were deemed to be of diagnostic quality without significant artifacts and included at least T1W pre- and post-contrast, FLAIR, ADC and DWI sequences. Of the initial cohort, only five scans had perfusion data available.

All MRIs were jointly reviewed by two board-certified neuroradiologists (N.F. with 6 and J.H. with 10 years of post-graduate experience) who were blinded to the clinical history, histopathologic diagnosis, and molecular status. Imaging characteristics were qualitatively evaluated and scored based on previously agreed-upon scales for FLAIR border, contrast enhancement, and diffusion characteristics. A consensus was reached for each category on all 40 patients (Fig. 1). The MR imaging features were based on whole tumor volume and included: (1) tumor location (frontal, parietal, temporal, occipital, multilobar, or cerebellar); (2) frontal only versus non-frontal/involving more than the frontal lobe (based on prior literature demonstrating a frontal lobe predominance of IDH mutant and 1p/19q codeleted grade II/III gliomas); (3) hemisphere (right, left or bilateral); (4) tumor border definition on FLAIR (well-defined: where the border between tumor and normal appearing white matter can be delineated; ill-defined: where the border between tumor and normal appearing white matter is hazy; cannot be defined: where the border between tumor and normal appearing white matter cannot be delineated); (5) volume of contrast enhancement (<25, 25–75 or >75%); and (6) qualitative assessment of diffusion on DWI/ADC (restricted: hyperintense on DWI and hypointense on ADC) [23]. The imaging classification scheme is demonstrated in Fig. 1.

Fig. 1
figure 1

The top panel demonstrates the classification for FLAIR tumor border: the image on the left shows a left frontal tumor with a well-defined border; the middle image shows a left frontal tumor classified as an ill-defined border; the right image shows a left sided multilobar tumor whose border cannot be discerned. The bottom panel demonstrates the classification for degree of tumor contrast enhancement: the image on the left shows a left sided multilobar tumor with no contrast enhancement; the middle image shows a left frontal tumor in which >25% but <50% of the tumor bulk enhances; the right sided image shows a multilobar glioblastoma in which >50% of the tumor bulk enhances (no tumors in this study cohort had this characteristic)

Full size image

Progression free survival (PFS) analysis

PFS was calculated from the date of tissue diagnosis to the date of documented imaging progression. Tumor progression was defined as unequivocal progression on MRI in conjunction with the assessment of the treating neuro-oncologist. Patients without documented progression were censored at the date of the last stable MRI scan.

Statistical analysis

Analysis of variance (ANOVA) and chi square tests were performed to test for group differences in age and gender distribution. Fisher’s exact test was used to evaluate the relationship between imaging characteristics and molecular status. Analysis was initially performed comparing all three groups. Subsequently, to elucidate potential inter-group differences, analysis was performed to compare the collective IDH-MT groups versus the IDH-WT group and the two IDH-MT groups.

Univariate Cox Proportional Hazard (CPH) models and Kaplan–Meier analyses were conducted to assess the contribution of age, gender, and type of surgery (gross total resection vs. subtotal resection vs. biopsy) to PFS. Significant univariate predictors were included as covariates in the main Kaplan Meier analysis, which was performed to assess for differences in PFS among the molecular subgroups and compared using a log-rank test. PFS analysis was also performed based on grade across all three molecular subtypes. (Statistical Package for the Social Sciences Statistics, Version 21. Armonk, New York).

Results

Cohort characteristics, molecular subgroups, and tumor grade

The clinical characteristics of the final cohort are summarized in Online Resource 2.

A significant difference existed in age among the three subgroups [F (2,39) = 7.16, p = 0.002]. A post-hoc analysis revealed that the IDH-MT ATRXloss patients (mean age = 35.4 years) were significantly younger than both the IDH-MT1p/19qcodel (mean age = 46.6 years; p = 0.046) and the IDH-WT (mean age = 51.6 years; p = 0.002) patients. The overall cohort consisted of 13 (32.5%) females and 27 (67.5%) males. The gender breakdown was also statistically significant (p = 0.035) with more males in the IDH-WT group (92.3%) versus the two MT groups (IDH-MT1p/19qcodel: 66.7%, IDH-MT ATRXloss: 46.7%). CPH analysis demonstrated that age was not associated with PFS (95% CI 0.969–1.039, DF 1, p = 0.866). Additionally, gender was not associated with PFS (Log-Rank: Chi-Square 1.938, DF 1, p = 0.164). However, type of surgery was associated with PFS (Log-Rank:Chi-Square 6.9, DF 2, p < 0.05).

MR imaging correlates of molecular subgroups

The radiologic characteristics of the cohort are summarized in Table 1. Tumor location was associated with the molecular subgroup, in that the IDH-MT tumors were predominantly located in the frontal lobe (p = 0.006). There was no significant difference in location between the IDH-MT groups.

Table 1 MRI features of the cohort according to molecular subgroup
Full size table

The tumor border on FLAIR was also highly associated with the molecular subgroup. IDH-WT tumors were more likely to have a non-definable border, while the collective IDH-MT tumors had well-defined or slightly ill-defined borders (p = 0.019). Specifically, of the 27 IDH-MT tumors, only one had a border that could not be defined, whereas this was true for six of the 13 IDH-WT tumors. There was no significant difference in border pattern between the two IDH-MT groups (p = 0.684).

There was no significant difference in contrast enhancement among the three groups (p = 0.336). In fact, none of the tumors (all grade II/III) in this cohort demonstrated marked enhancement. Only five (12.5%) of the tumors demonstrated restricted diffusion. This included two in the WT and three in the MT groups. This was not significantly different across the three groups (p = 0.345). Table 1 demonstrates the MRI features of the cohort according to the molecular subgroup. Figures 2 and 3 compare imaging between IDH-MT and IDH-WT grade II and III tumors respectively.

Fig. 2
figure 2

The top panel demonstrates MR images from a patient with a left-sided frontal lobe IDH-MT1p/19qcodel oligodendroglioma (WHO grade II) who did not experience progression during the study. The bottom panel demonstrates MR images from a right-sided multilobar IDH-WT diffuse astrocytoma (also WHO grade II) who progressed within 8 months following tissue sampling. The mutant tumor has a well-defined border on FLAIR whereas the wildtype tumor border cannot be delineated (arrows). Both tumors have minimal (<25%) contrast enhancement. Neither tumor demonstrates restricted diffusion

Full size image
Fig. 3
figure 3

The top panel demonstrates MR images from a patient with a right-sided frontal lobe IDH-MT ATRXloss anaplastic astrocytoma (WHO grade III) who did not experience progression during the study. The bottom panel demonstrates MR images from a bilateral multilobar IDH-WT anaplastic astrocytoma (also WHO grade III) who progressed less than 14 months following tissue sampling. The mutant tumor has a well-defined border on FLAIR whereas the wildtype tumor border cannot be delineated. Both tumors have minimal (<25%) contrast enhancement. Neither tumor demonstrates restricted diffusion. H&E sections of the two tumors show relatively similar morphology; however, the mutant tumor stains positive for an antibody against IDH confirming the mutation presence, which is negative in the IDH-WT tumor

Full size image

PFS

The median PFS within the IDH-WT subgroup was 8 months with 77% (n = 10) progressing during the study duration, which was significantly shorter compared to the IDH-MT subgroups (Log-Rank: Chi-Square 40.1, DF 5, p < 0.001, Fig. 4). Among the IDH-MT subgroups, the IDH-MT1p/19qcodel had a median PFS of 51 months with 42% (n = 5) progressing during the study and the IDH-MT ATRXloss had a median PFS of 43 months with 40% (n = 6) of patients progressing (Fig. 4). There was no difference in PFS between the two IDH-MT groups. Further analysis revealed no significant differences in PFS among the whole cohort when based on grade (II or III) (Log-Rank: Chi-Square = 7.3, DF 5, p = 0.202).

Fig. 4
figure 4

Kaplan–Meier curve with log-rank tests were used to examine differences in PFS among the different molecular subgroups with type of surgery as a covariate. The red line represents the IDH-WT tumors (n = 13), the blue line represents the IDH-MT1p/19qcodel tumors (n = 12), and the green line represents the IDH-MT ATRXloss tumors (n = 15). The median PFS in the IDH-WT subgroup was significantly shorter (8 months) than PFS in the IDH-MT1p/19qcodel (51 months) or the IDH-MT ATRXloss (43 months: p < 0.001) subgroup. However there was no difference in PFS between the two IDH MT subgroups

Full size image

During the duration of follow up, six of the patients in the cohort died and one went to hospice. Of these seven patients, six were in the IDH-WT group.

Discussion

The delineation of three distinct molecular subgroups of grade II/III gliomas (IDH-WT, IDH-MT1p/19qcodel and IDH-MT ATRXloss) validates years of research linking 1p/19q and IDH status with clinical outcome [6, 10]. Indeed, molecular status can yield more prognostic information than traditional histopathology alone. The novel integration of genotypic information with tumor histology in the 2016 WHO Classification of Tumors of the CNS attests to the significant role radiogenomics can play within the field of neuro-oncology.

Using primarily next-generation sequencing techniques and OncoScan microarray analysis to determine three distinct molecular subgroups of grade II/III gliomas, we demonstrate that on FLAIR imaging, the IDH-MT tumors had more well-defined borders in contrast to the IDH-WT tumors whose borders could not be defined. Given that IDH-WT tumors are known to be clinically and molecularly similar to glioblastomas, it follows that these tumors show a more infiltrative pattern on MRI. This characteristic FLAIR pattern could help distinguish the sentinel molecular classification of a tumor at initial imaging diagnosis.

This differentiating FLAIR border adds to prior work of other investigators including Metellus et al. who found that 100% of IDH-WT tumors in their cohort had indistinct borders in contrast to only 45% of IDH-MT tumors. However, their study was limited to only grade II gliomas and used PCR for assessment of IDH mutations [23]. Qi et al. found similar results, but only evaluated astrocytomas [24]. Our study expanded on these by including grade II and III astrocytomas and oligodendrogliomas; furthermore, employing predominantly a next-generation sequencing technique allowed us to capture rare IDH mutations (including non-R132H IDH1 and all IDH2 mutations) enabling us to more accurately characterize our sample [25]. In fact, less sensitive methods may have hampered classification in previous studies. We did not find a difference in the FLAIR border characteristics between the two IDH mutant groups. This is in contrast to Jenkinson et al., who in 2006 prior to the 2016 WHO guidelines, investigated the differences in tumor borders of histologically classified oligodendroglial tumors and found that 1p/19q intact tumors had sharp/smooth borders while the 1p/19q codel tumors rarely had sharp borders [26].

Our findings that the IDH-MTs collectively had a frontal lobe predilection, echoes previous studies [24, 27]. The current hypothesis is that the IDH-MT tumors arise from a neural precursor population with a defined spatial and temporal location, and since the IDH mutation precedes 1p/19q codeletion in tumorgenesis, it follows that the 1p/19q codeleted tumors are also frontal lobe predominant [11, 28, 29]. Similarly, as we did not find any substantive differences between the IDH-MT1p/19qcodel and the IDH-MT ATRXloss groups, this may be secondary to their common origin path in gliomagenesis (IDH mutation), and more advanced imaging techniques may be needed to elucidate their differences.

Our study is in accordance with prior publications by demonstrating the prognostic significance of these novel molecular subgroups [30]. We found that median PFS was significantly shorter in the IDH-WT group compared to both IDH-MT groups (p < 0.001); whereas, there was no difference in PFS between the two IDH-MT groups. While OS was not a primary endpoint evaluated in this study, it should be noted that of the seven patients who died or went to hospice, six were in the IDH-WT group. Importantly, there were no significant differences in PFS among the three groups when based on grade (II or III). This reinforces prior studies that found no significant prognostic differences between IDH-MT grade II and III tumors [31, 32]. In fact, Suzuki et al. stated that grade III IDH-MT tumors showed OS comparable to corresponding grade II tumors and recommended managing them as low-grade tumors [33]. Additionally, the significant difference in age among the three subgroups (with the IDH-WT being older than the collective IDH-MT groups) is in accordance with Eckel-Pastow et al. and should be expected since IDH-WT tumors are on the spectrum of glioblastomas, which tend to present later in life than less aggressive gliomas [4].

While genotyping is presently the gold standard for classifying patients into different molecular subgroups with established predictive and prognostic value, radiogenomic delineation can add considerable value. First, these genotyping techniques as well as surrogate immunohistochemistry may not be readily available at all institutions and the variability among genotyping assays may lead to imperfect groups when less sensitive modalities are used. Furthermore, certain tumors such as brainstem gliomas may not be amenable to biopsy, and if sampling is attempted, intertumoral regional heterogeneity is a potential problem with possible exclusion of tumor cell subpopulations [33]. In contrast to these shortcomings, imaging features, specifically FLAIR border pattern, can be readily evaluated on the initial diagnostic MRI, potentially providing predictive and prognostic information. The ability to differentiate IDH-mutant and IDH-wildtype tumors at the time of initial imaging may influence decisions regarding the extent and strategy of surgical resection. Additionally, FLAIR border analysis may play a pertinent role in monitoring patients, specifically in assessing for a change in molecular phenotype with progression or recurrence.

While other investigators have found similar associations between IDH-status and FLAIR border pattern, our cohort is unique in that we examined both grade II and III gliomas and we devised groups prescribed by the recently published 2016 WHO Consensus and complementary large-scale sequencing studies, particularly incorporating IDH, 1p/19q and ATRX status. Additionally, we expanded the present knowledge base by assessing for differences among these groups utilizing several common imaging techniques (FLAIR, contrast-enhancement and diffusion) thereby emulating tumor evaluation in daily clinical neuroradiology practice. Although the current clinical utility of these imaging-genomic associations may be primarily reserved for cases where biopsy cannot be performed, given the rapidly evolving field of radiogenomics, these associations may have much more significant implications in the near future such as in vivo assessment of treatment response versus tumor progression.

There are limitations of this study based on its retrospective nature. Additionally, there was heterogeneity in the MRI acquisition parameters, since several of the scans were from outside institutions due to the need to assess pre-treatment MRIs. This precluded performing a formal quantitative assessment of imaging characteristics. However, this is in line with routine clinical practice, which utilizes qualitative analysis of lesions for initial assessment as well as follow-up examinations. It is possible that future studies could incorporate quantitative analysis to further distinguish molecular subgroups. Furthermore, while we accounted for type of surgery in our Kaplan Meier curves, chemotherapy regimens were varied among the groups, which could have impacted PFS (perhaps limiting our ability to find groups differences between the IDH-MT subgroups). However, all of the evaluated MRIs were pre-treatment and therefore the imaging analysis was not affected by treatment heterogeneity.

Conclusion

In this study we found that imaging features, particularly FLAIR border pattern and tumor location, could distinguish molecular subgroups of grade II and III gliomas based on the Updated WHO Consensus Classification, and the ability to distinguish these subgroups on initial diagnostic imaging may impact clinical decision making. Furthermore, these molecular subgroups provide important prognostic information that is not captured by histological classification and grading alone. Future studies should evaluate whether these imaging features can characterize progressive or recurrent disease, and guide therapeutic decisions including surgical and radiotherapy strategies.