Introduction

High-grade infiltrating gliomas in the spinal cord are relatively rare compared to their intracranial counterparts, accounting for 7.5% of all intramedullary gliomas and only 1.5% of all spinal cord tumors [1, 2]. Only a handful studies on outcomes of high-grade infiltrating gliomas of the spinal cord have been reported in the literature, predominantly as case reports, case series, or as part of a larger cohort of spinal intramedullary tumors [3,4,5,6,7,8]. However, similar to their supratentorial counterpart, high-grade infiltrating gliomas of the spinal cord have been shown to have an extremely poor prognosis, with reported median survival duration ranging from 12 to 14 months [4, 5]. Spinal cord surgical procedures typically provide small amounts of tissue even when a resection rather than a biopsy procedure is performed. Owing to this poor prognosis, rare occurrence, and limited availability of tumor tissue, high-grade infiltrating glioma of the spinal cord remains incompletely characterized.

The most recent World Health Organization (WHO) Classification of Tumors of the Central Nervous System introduced the new entity of “Diffuse Midline Glioma, H3K27M- mutant” based on histological and molecular characterization, which corresponds to a grade IV designation [9, 10]. The current understanding regarding this mutation initially came from molecular analysis of midline high-grade gliomas in pediatric and adolescent patients, which revealed a characteristic recurrent lysine to methionine substitution at codon 27 (K27M) in histone H3 variants encoded predominantly by H3.3 gene H3F3A (75–80% cases) and in H3.1 gene HIST1H3B (25%). The K27M mutation is located in a critical posttranslational modifiable region of the N-terminal histone tail and seems to act in a dominant-negative manner through inhibition of EZH2 enzymatic subunit of the Polycomb repressive complex 2 (PRC2). H3F3A K27M-mediated PRC2 interaction results in reprogramming of the epigenetic landscape with global reduction of histone H3.3 K27 methylation, including the trimethylated H3K27 (H3K27-me3). The H3K27-me3 normally serves as a transcriptional repressor; hence, its reduction has been thought to lead to extensive transcriptional modification of the tumor cell regulation, eventually leading to increased expression of cancer-related genes [11,12,13,14,15,16,17,18,19,20]. These H3K27M-mutant gliomas have been associated with an extremely aggressive behavior, regardless of histologic features in pediatric populations [18,19,20,21,22,23]; however, within the adult population, the frequency of the mutation and its impact on prognosis is less clear. Within an admixed cohort of 36 pediatric and adult patients with diffusely infiltrating gliomas of the spinal cord, the H3K27M mutation was detected by immunohistochemistry in 16 cases, including 6 of the 11 adults and 10 of the 14 children [24]. A recent study assessing midline gliomas in adults [25] for these common alterations included eight spinal cord gliomas, of which five were found to harbor the H3K27M mutation and none were positive for an IDH1 mutation.

Other gene modifications have also been reported associated with H3K27M mutations, including TP53 gene mutations, which have been reported in 50% of pediatric gliomas [20, 26], and in the histone chaperone alpha-thalassemia/mental retardation syndrome X- linked (ATRX) that affects the loading of histone H3.3 in heterochromatic telomeric regions [26, 27]. Mutations of IDH and H3F3A are considered mutually exclusive, demonstrating a potential difference in origin and subsequent behavior of the latter [28]. The telomerase reverse transcriptase gene promoter (TERTp) has been shown to be useful for prognostication, with poor outcomes reported among patients with IDH-wild-type infiltrating gliomas [29, 30]. Molecular alterations with potential therapeutic predictive value have been identified, including the BRAF V600E and PPM1D mutations [31,32,33].

In this study, we expand on these findings with a comprehensive molecular analysis of 13 adult patients with high-grade infiltrating glioma of the spinal cord, whose surgery was performed at our institution over a 26-year period—one of the largest single institutional series to date.

Methods

Patient cohort

After institutional review board approval, the institutional cancer registry was queried for all patients with high-grade infiltrating glioma of the spinal cord who underwent surgery performed at our institution, identifying 20 cases with tissue specimens available for 19 cases. Of these, 3 patients were aged <18 years and thus were excluded. Another 3 patients were excluded because they were found to have supratentorial infiltrating gliomas, and their spinal cord tumors could not be definitely deemed to represent primary tumors. In total, 13 high-grade infiltrating glioma of the spinal cord in adult patients were included for analysis. All cases had research authorization reported. The following variables were reported: (1) patient demographics; (2) presenting symptoms; (3) tumor location; (4) treatment; (5) histological diagnosis; (6) preoperative imaging characteristics, including the presence of post-contrast enhancement and extent of FLAIR signal, if available; (7) presence of hydrocephalus and mass effect; (8) presence of dissemination; (9) follow-up duration; and (10) vital status at last follow-up.

Immunohistochemistry

Formalin-fixed, paraffin-embedded tissue sections were stained with hematoxylin and eosin for routine histological review. Immunohistochemical staining was performed on 4-μm-thick formalin-fixed, paraffin-embedded sections using antibodies against H3K27M (polyclonal, EMD Millipore, Billerica, MA), trimethylated H3K27 (clone C36B11, Cell Signaling Technology, Danvers, MA), IDH1 R132H (clone H09, Dianova, Hamburg, Germany), ATRX (polyclonal, Sigma-Aldrich, St. Louis, MO), and BRAF V600E (clone VE1, Spring Bioscience, Pleasonton, CA).

Molecular testing

DNA was extracted from 5-micron-thick unstained formalin-fixed, paraffin-embedded tissue sections using the QIAamp ® DSP DNA formalin-fixed, paraffin-embedded Tissue Kit (Qiagen Inc., Germantown, MD) with slight modifications. Next-generation sequencing library preparation was performed using a QIAseq Targeted DNA custom amplicon-based panel that was designed to interrogate 150 genes associated with central nervous system tumors, including H3F3A, HIST1H3B, HISTI1H3C, IDH1, IDH2, TERT (including promoter region), ATRX, BRAF, CDKN2A, CDKN2B, CDKN2C, CIC, DAXX, EGFR, FGFR1, FGFR2, FGFR3, FUBP1, MYBL1, and MDM2. Libraries were sequenced on an Illumina HiSeq 2500 (Illumina, Inc., San Diego, CA). Sequencing data were processed through a custom bioinformatics pipeline that was developed to detect single-nucleotide polymorphisms and small insertions–deletions (<50 base pairs). Variants at allele frequency of at least 10% with a minimum of 100× depth of coverage were evaluated and classified as benign polymorphism, variant of unknown significance, or pathogenic mutation. Only pathogenic mutations were considered for analysis.

Statistical analysis

Continuous variables have been summarized using mean and standard deviation or median and interquartile ranges and analyzed between patients with and without a mutation using t test or Wilcoxon rank-sum test. Categorical variables have been summarized using frequencies and proportions and compared between patients with and without a mutation using Pearson chi-square or Fischer’s exact test.

Kaplan–Meier curves and log-rank test were used to compare survival between patients with and without a mutation.

Results

Clinical and histological findings

The 13 patients identified included 11 men and 2 women, with a median age of 38 years (range 18–69 years). The thoracic spinal cord was the most commonly affected region (46%, n = 6) followed by the cervical (31%, n = 4), thoraco-lumbar (15%, n = 2), and cervico-thoracic (8%, n = 1) regions. Radiologically, regions of post-contrast enhancement were observed in 9 cases (Fig. 1a–c). In terms of surgical management, 8 patients underwent a biopsy, 1 patient underwent a gross total resection and 4 patients underwent a subtotal resection.

Fig. 1
figure 1

Illustrative images from case no. 5. a T2-weighted magnetic resonance images (MRIs) reveal an intramedullary mass affecting C2–C6 levels. b T1-weighted pre-contrast and c T1-weighted post-contrast MRIs demonstrate enhancement. d Histological sections show a cellular, mitotically active infiltrating glioma with microvascular proliferation (hematoxylin and eosin). By immunohistochemistry, e the glioma cells are positive for H3K27M and f show loss of H3K27-me3 expression (all histological images at ×200 magnification)

Histological review of the cases was performed while blinded to the results of immunohistochemical and molecular findings. All cases contained atypical cells with irregular, elongated nuclei arranged in an infiltrative growth pattern characteristic of a diffusely infiltrating astrocytoma. Applying the histological criteria described in the 2016 WHO Classification of Tumors, 8 of the 13 cases (62%) showed morphologic features of glioblastoma (WHO grade IV; Fig. 1d) [9]. Five cases showed elevated mitotic activity (3–12 mitoses identified per 10 high-power fields) while lacking definitive microvascular proliferation and necrosis and therefore classified morphologically as anaplastic astrocytoma (WHO grade III).

By immunohistochemistry, six cases were positive for H3K27M (46%; Fig. 1e), all with concomitant loss of H3K27-me3 expression (Fig. 1. f) and one demonstrating loss of expression of ATRX (the only case in the cohort to show loss of ATRX). All cases were negative for IDH1-R132H and BRAF-V600E by immunohistochemistry.

Eleven patients underwent radiotherapy, while eight patients underwent chemotherapy. The most common agent used for chemotherapy was Temozolomide (n = 4). Other agents included Cytoxan and Prednisone, CCNU (Lomustine), and Panobinostat (used as off-label), while the agent was unknown in one. One patient also underwent immunotherapy. Only one patient underwent trimodality therapy. The mean follow-up for the cohort was 35 months (range: 0.3–212 months; SD = 57.6). At last follow-up, 10 out of 13 patients had expired, with a mortality rate of 77%, a median survival of 13 months, and 1-year and 2-year survival rates of 46% and 50%, respectively.

The clinical and histological data are summarized in Table 1.

Table 1 Summary of clinical, radiological and histological features

Molecular findings

Next-generation sequencing was successfully performed in 10 (of the 13) cases (specimens obtained within 2000–2017), with 3 cases yielding insufficient quantity of DNA for analysis due to the small size (<1 cm2) of the available tissue. The results are summarized in Fig. 2. Among these, four of the six cases positive for H3K27M-mutant by immunohistochemistry were found to harbor the H3F3A K27M mutation; the remaining two H3K27M-mutant cases failed sequencing. No HIST1H3B, HIST1H3C, IDH1, IDH2, FGFR1, or BRAF mutations were detected.

Fig. 2
figure 2

Oncoprint of mutations identified by 150-gene next-generation sequencing panel [38, 39]. Each vertical bar corresponds to a case and all aligned vertical bars across the genes correspond to the same case

Overall, three distinct molecular groups were apparent: H3K27M-mutant (n = 6), TERT promoter-mutant (n = 3), and those lacking mutations in both (H3K27M/TERT promoter wild type; n = 4). In addition to an H3F3A mutation, all H3K27M-mutant cases frequently showed mutations in TP53, PPM1D, and NF1, as well as a single case each showing mutations in ATRX (in conjunction with an NF1 mutation) and in BCOR mutation (in conjunction with a TP53 mutation). All TERT promoter-mutant cases showed the C228T mutation. In this group, the only recurrently mutated gene was PIK3CA. A single case also showed TP53, EGFR, and SETD2 mutations in addition to a PIK3CA mutation. Among the tumors which were H3K27M and TERT promoter wild type, TP53 mutation was the sole abnormality in two cases, while ATRX, NF1 and PPM1D mutations co-occurred in one case. These data are summarized in Table 2.

Table 2 Status of the selected immunohistochemical and molecular markers

Overall, PPM1D mutations were mutually exclusive with TP53 mutations and observed in conjunction with NF1 mutations in all cases. Both PIK3CA mutations occurred in conjunction with a TERT promoter mutation and both ATRX mutations co-occurred with NF1 and PPM1D mutations. Of note, the two cases with an ATRX mutation, which were predicted to result in protein truncation in both instances, did not show loss of protein expression by immunohistochemistry (Fig. 2), whereas the single case with loss of ATRX protein expression did not have a detectable sequence alteration by next-generation sequencing and occurred in an H3 K27-mutant tumor (Table 2).

Compared to TERT promoter-mutated tumors, H3K27M-mutant tumors appeared to occur in younger patients and were associated with longer overall survival (30.5 ± 9.8 vs 56 ± 7, p = 0.002; median survival = 48.5 months vs 1 month, p = 0.033, respectively) [Table 3 and Fig. 3]. On pair-wise comparison, no significant difference in the overall survival was found between patients with H3K27M-mutant and H3K27M-wildtype tumors (p = 0.45) or between H3K27M-mutant and TERT promoter -mutant tumors (p = 0.13). However, patients with TERT promoter mutant high-grade infiltrating gliomas of the spinal cord had significantly lower overall survival compared to those with TERT wild-type tumors—a group that includes all H3K27M-mutant cases (p = 0.01).

Table 3 Summary of the results by molecular subgroup
Fig. 3
figure 3

Kaplan–Meier survival analysis by H3K27M and TERT promoter mutation status

Discussion

High-grade gliomas of the spinal cord in adults are rare and carry a poor prognosis [4, 5]. Although adult diffuse gliomas involving the cerebral hemispheres have undergone extensive molecular characterization, tumors involving the spinal cord have a relative paucity of molecular data available, likely due to their rarity, poor prognosis, and limited availability of tumor tissue. It is not yet well established whether molecular alterations impact prognosis in adult patients with high-grade infiltrating glioma of the spinal cord. The current study is one of the largest reported single-institutional series of high-grade infiltrating glioma of the spinal cord in adults spanning a period of 26 years. While treatment modalities have evolved over this 26-year period, high-grade infiltrating glioma of the spinal cord has remained a clinically aggressive entity, with the median survival of 13 months and 1-year survival of 46.15% in our cohort.

In our series, we found an incidence of 46.2% of H3K27M-positive cases (n = 6). Two recent series that included pediatric cases reported a rate of 53% (n = 9/17) [21, 24] among all spinal cord gliomas and 80% among grade IV gliomas carrying the H3K27M mutation [21, 34]. Picca et al. reported that 5 of the 8 spinal cord gliomas in adults were positive for H3K27M, while Gessi et al. identified the H3K27M mutation in 6 of the 11 adult spinal cord diffuse gliomas. While, in pediatric cases, diffuse midline gliomas (arising in any location) with the H3K27M mutation carry a worse prognosis compared to wild-type counterparts [35], it is currently uncertain as to what prognostic impact, if any, the presence of this mutation has in spinal cord diffuse gliomas affecting adults. Picca et al., focusing on adult tumors, found no survival difference between H3K27M-mutant vs wild-type tumors when including all midline locations [25]. On the other hand, Yi et al. found that the presence of the H3K27M mutation (detected among 20 of the 25 WHO grade IV gliomas) was associated with longer overall and disease-free survival (40 months) [25, 34]. While this finding was certainly novel, given that almost all studies to date have reported a poor prognosis associated with H3K27M mutation, the authors noted that only five cases in their cohort did not have the mutation. Moreover, the treatment characteristics of the patients were not uniform, with only a subset of patients receiving trimodal therapy (surgery, radiation, and Temozolomide) [34]. The median overall survival for H3K27M-positive cases (n = 6) in our cohort was found to be 48 months, compared to 77 months in tumors without the mutation (n = 4), which are in contrast to those reported by Yi et al., who reported a longer survival among those with H3K27M mutations [34].

Among adult diffuse gliomas, TERT promoter mutation occurring in the absence of IDH mutations and 1p/19q-codeletion is associated with poor prognosis [21]. In the current series, we identified 3 (of the 13) cases harboring a TERT promoter mutation. Among this group, the median survival was found to be just 1 month, as compared to 75.7 months in tumors without an H3K27M/TERT promoter mutation. Picca et al. showed that TERT promoter mutations are also associated with shorter survival. In their cohort, TERT promoter mutations are significantly enriched among H3K27M/IDH-wild-type adult midline gliomas, with concurrent TERT promoter mutation and H3K27M mutation in 2 (of the 8) spinal cord gliomas, and in overall 4 (of the 116) adult midline gliomas from various sites [25]. In our cohort, all three TERT promoter mutant cases were negative for H3K27M. The significance of concurrent H3K27M and TERT promoter mutation among midline gliomas, and spinal cord gliomas in particular, remains to be determined. While data regarding the presence of additional alterations is unavailable in the report by Picca et al., we have been able to identify alterations occurring in the presence and absence of the H3K27M mutation (Table 2), which may help provide some insight into the molecular landscape present in these rare tumors.

All three PPM1D mutations identified in this study were novel truncating alterations within the regulatory C-terminal domain, which is a mutational hotspot region in human cancer [36]. As previously described in brainstem infiltrating gliomas [33], PPM1D mutations were mutually exclusive with TP53 mutations in our series of adult spinal cord high-grade gliomas. In contrast to brainstem infiltrating gliomas, wherein PPM1D mutations were only observed in H3K27M-mutant tumors, PPM1D mutations were seen in both H3K27M-mutant and H3/IDH/TERT wild-type cases in our series. It was also noteworthy that the ATRX status by immunohistochemistry and sequencing did not correlate in this study: the two cases with an ATRX truncating mutation did not show loss of protein expression by immunohistochemistry, whereas the single case with ATRX loss by immunohistochemistry did not have an ATRX sequence alteration detectable by next-generation sequencing. While immunohistochemical loss of ATRX expression in the absence of a detectable ATRX mutation has been observed [37], preserved ATRX protein expression in the presence of an ATRX truncating mutation is somewhat unusual. The identified ATRX mutations occurred within exons 26 and 30 (out of 35 total exons) and seem to have escaped nonsense-mediated decay; the expressed mutant proteins, however, are likely defective as at least one of them lacks the helicase functional domain.

Our findings suggest that high-grade infiltrating glioma of the spinal cord of adult patients are a heterogeneous group of tumors, with main molecular groups including H3K27M-mutant, TERT promoter-mutant, and those harboring neither. Among adult high-grade infiltrating gliomas of the spinal cord, tumors with either an H3K27M or a TERT promoter mutation appear to represent clinically and prognostically distinct groups, with H3K27M-mutant tumors appearing to affect younger patients and being associated with longer overall survival when compared to TERT promoter-mutant tumors. Similar to the findings of Picca et al. pertaining to supratentorial midline tumors among adults, IDH-wildtype/H3K27M-wild-type infiltrating gliomas seems to delineate a group of clinically aggressive tumors, and this group frequently shows TERT promoter mutation in the spinal cord.

In summary, we have characterized 13 high-grade infiltrating gliomas affecting the spinal cord in adult patients. Although the spinal cord is midline in location, these tumors appear to be share similar molecular profiles as diffuse gliomas affecting midline and non-midline locations at other sites. Additionally, the differences in molecular profiles may also have prognostic significance for these patients. Given the small size of our study cohort, evaluation of clinico-pathological and molecular features in larger multi-institutional cohorts may help confirm the prognostic significance of this molecular heterogeneity. Further, prospective studies may help determine whether adult patients with high-grade infiltrating glioma of the spinal cord would indeed benefit from targeted therapies in the similar manner as molecularly defined gliomas affecting other sites.