Transcriptome signatures associated with meningioma progression

Meningioma samples were obtained at the Hospital of the University of Pennsylvania (HUP) and banked after intraoperative examination under IRB protocol approved by the University of Pennsylvania. After report review, each diagnosis was verified via histopathological review by a board-certified neuropathologist (MML and/or AV). Tumors with intermediate features (incomplete atypical features) and tumors with grade-defining histology (i.e. choroid or clear cell meningioma) were specifically excluded to amplify the effect of potential pathways implicated in meningioma progression in a more homogeneous cohort. Clinical and demographic information was obtained within IRB specifications, including prior history of radiation therapy. The discovery set consisted of 25 meningioma samples from 20 patients, which included de novo tumors (WHO I = 9, WHO II = 7, WHO III = 3), and progressive tumors (n = 6). Table 1 shows all selected patient samples along with corresponding information on WHO grade, gender, brain invasion, tumor location, and length of follow-up. The validation cohort, used for quantitative reverse transcription (qRT)-PCR only, was selected with similar criteria and consisted of 38 samples from 38 different patients (WHO I = 20, WHO II = 12, WHO III = 6). Tables 2 and Table 3 show all selected patient samples along with corresponding information on WHO grade, gender, brain invasion, and tumor location. For a subset of these (WHO I = 7) included in Table 3, length of follow up was known and superior to 5.4 years, as shown.

Total RNA was prepared from frozen meningioma samples using TRIzol (Invitrogen, Carlsbad, CA, USA). RNA was reversed transcribed into cDNA using Superscript III reverse transcriptase (Invitrogen) according to manufacturer’s recommendations. Real-time PCR was performed with a SYBR Green probe using a ViiA 7 (Applied Biosystems). Each sample was run in triplicate, and the RNA level for each gene was assessed by normalization to the expression of the housekeeping gene GAPDH or RN18S1. The primer sequences for the various genes used in this study are available on request.

The library preparation and sequencing process were performed at the Beijing Genomics Institute (BGI) facility located in the Children’s Hospital of Philadelphia (CHOP). Library construction was performed by following Illumina stranded RNA-seq workflow (TruSeq Stranded Total RNA Library Prep Kit, Cat# RS-122-2201). Briefly, 200 ng of total RNA is treated with Ribo-zero kit to remove ribosomal RNA, and then purified. RNA is then fragmented and converted to cDNA with RT reaction. Subsequent steps include end repair, addition of an “A” overhang at the 3′ end, and ligation of the indexing-specific adaptor, followed by purification with Agencourt Ampure XP beads. The library is then amplified and purified with Ampure XP beads. Size and yield of the bar-coded libraries are assessed on the LabChip GX, with an expected distribution around 260 bp. Concentration of each library is measured with real-time PCR. Pools of indexed library are then prepared for cluster generation and 100 bp by 100 bp paired end sequencing on the Illumina HiSeq 2000.

The samples were sequenced at the sequencing core BGI@CHOP. Randomly fragmented DNA sequences were run through libraries prepared for paired end sequencing on an Illumina HiSeq 2000. The raw RNA sequencing reads were run through the QC checks by FastQC and then mapped to reference genome (h19) by aligner STAR. After that, HTSeq was applied to detect the sequencing read count for each gene. DESeq2 was applied to detect the differential expression level for each gene between different groups. Principal component analysis (PCA) was used to characterize the RNA-seq data set; Hierarchical clustering was performed using thousands of genes with the highest variability across the samples. For the discovery set, the differential expression between different groups was considered for the Gene Set Enrichment Analysis (GSEA) [49]. STAR-Fusion [21] was applied for the RNA-seq transcript fusion detection.

Diagnostic slides were reviewed to ensure all tumors were graded according to the 2016 WHO classification of meningiomas [33]. A total of 71 meningiomas were used in the construction of three tissue microarrays (TMA) including 15 patients whose tumors recurred with progression to a higher grade (13 patients with progression from grade I to grade II and 2 patients with progression from grade II to grade III) and 42 patients without tumor progression (18 WHO grade I, 17 WHO grade II, and 7 WHO grade III). Representative area(s) of each tumor was/were selected and circled on H&E stained-slides, and 1 mm cores were cut from the corresponding paraffin-embedded blocks. In the majority of cases, multiple cores were taken from the same tumor. Control tissues used in the construction of the TMA were obtained from histologically unremarkable brain autopsy specimens (cortex, cerebellum and meninges controls) and from surgical tonsillectomy specimens. Sections were cut at 5 μm from each TMA and used for hematoxylin and eosin staining and immunohistochemistry.

Immunohistochemical staining was performed on 5 µm sections using standard methods. Briefly, slides were deparaffinized in xylene and rehydrated through graded alcohols. Antibodies to CD45 (leukocyte common antigen) were used (1: 200, Dako), and detection of the antibodies was performed using a chromogenic substrate, diaminobenzene (Dako). Slides were counterstained with hematoxylin, dehydrated through a series of ascending concentrations of ethanol and xylene, and then coverslipped.

Immunohistochemical staining for CD45 was evaluated by counting the number of immunopositive cells within tumor areas in each 1 mm tissue core. The pathologist performing the counting was blinded to tumor grade and progression/non-progression status. A core was excluded if it contained less than 90% solid tumor. In instances where more than one core was present from a tumor, counts were averaged across all cores from the same tumor to give a single count. Counts were then averaged for grade I and grade II/III tumors; de novo tumors were grouped separately from tumors that progressed.

The discovery set chosen for RNA-seq (Table 1) consisted of 25 meningioma samples from 20 patients, which included: samples from grade I tumors that did not progress to higher grade (grade I never progressed or “I NP”), samples from de novo grade II or III tumors, samples from patients with tumors which progressed from grade I (grade I progressed or “I P”) to grade II (secondary or “II S”) (patients #1–3) and samples from a patient with a tumor progressing from grade II to III (patient #4). The majority of patients (76%) were female, which is consistent with the well-known higher incidence of meningioma in women. Median age at surgery was 47.5 years (range 27–86). Eight tumors (38%) had brain invasion (WHO grade II or III tumors). Among the 20 patients selected, 5 (25%) had radiation therapy for a prior malignancy.

The RNA-seq data were analyzed using unsupervised clustering approaches [2, 42, 53]. Interestingly, 4 of the 6 grade I tumors that did not progress (I NP) clustered together and separately from the other tumor samples (Fig. 1a). Grade I tumors that progressed (I P) clustered together with their respective secondary grade II tumor, suggesting that secondary tumors mostly resemble their primary. Generally, grade II tumors that arose de novo (DN) tended to mix with grade II tumors that progressed from grade I (grade II S). Grade III tumors did not cluster together, suggesting that they represent a more heterogeneous class (Fig. 1a). This pattern was observed when differential expression level across 2000 genes was measured and compared (Fig. 1b). When only grade I tumors (i.e. both I P and I NP) were analyzed via unsupervised clustering to avoid possible confounding results from other grade tumors, 4/6 I NP were clearly separate from the remainder of the cohort; while 2/6 clustered more closely to I P (Additional file 1: Figure S1a). When grade II tumors were analyzed independently, a clustering pattern was harder to discern, albeit 5 of the 7 grade II DN tumors formed 2 clusters separately from the grade II S ones (Additional file 1: Figure S1b).

These results suggest, at least in part, that grade I tumors that progress to higher grades are molecularly different from grade I tumors that did not progress. Therefore, we identified a set of genes (Additional file 2: Table S1, Fig. 2a) that differ between these two prelabelled subclasses of grade I tumors, supporting the findings of unsupervised clustering. We further performed Gene Set Enrichment Analysis (GSEA) [34] and found that the differentially expressed genes between I P and I NP tumors were significantly overrepresented in signatures associated with ‘Hypoxia’, ‘EGF signaling’, ‘HRAS oncogenic’ and ‘Tumor angiogenesis’ (Additional file 1: Figure S2). These results shed light on the numerous pathways that distinguish these two groups. For instance, hypoxia-associated pathways and HIF1 signaling (a downstream effector involved in the response to hypoxia [31]) are found in I P samples. On the other hand, reduced TGFβ signaling is found in I NP samples (Additional file 1: Figure S2).

Grade I NP tumors also differed from grade II or/and III meningiomas in a set of differentially expressed genes, as determined by supervised clustering (Additional file 1: Figure S2c, Additional file 3: Table S2, Additional file 4: Table S3 and Additional file 5: Table S4).

The unsupervised clustering analysis pointed towards de novo grade II tumors being potentially different from secondary grade II that arise from a prior grade I tumor, as specific transcriptome signatures for each subtype were identified with supervised clustering (Additional file 1: Figure S2b, Additional file 6: Table S5).

To validate the results identifying differentially expressed genes between grade I and grades II/III tumors, we selected a new set of meningioma samples that constitute a second (validation) cohort (Tables 2 and 3); q RT-PCR was performed on these samples. The validation set consisted of 38 meningioma samples, which included grade I through III tumors with no progression. A special group consisted of grade I NP meningiomas for which follow-up was at least 5.4 years (average follow-up of 6.4 years), given the importance of lengthy follow-up to exclude recurrence and/or progression (Table 3). Out of the 38 patients, 28 (74%) were female. Median age at surgery was 55 years (range 34–88). Six (16%) tumors had brain invasion, three with grade II meningiomas and three with grade III.

Among the numerous genes differentially expressed across meningioma grades, we identified GREM2, a regulator of the BMP pathway, and the snoRNAs, SNORA46 and SNORA48, as being significantly reduced in meningioma progression by RNA-seq analysis (Fig. 3a-c). In particular, GREM2 levels in grade I tumors were significantly higher than in grades II and III (p < 0.0001); similar differential expression was found for SNORA46 (p < 0.0001) and SNORA48 (p < 0.0001). We confirmed this differential expression in the validation cohort via qRT-PCR (Fig. 3d-f). GREM2, SNORA46, and SNORA48 were each expressed at significantly higher levels in grade I meningiomas than grade II and III (p = 0.047, p = 0.017, and p = 0.038, respectively; one-tailed t-test). When analyzing these genes’ expression across each meningioma grade, we observed how GREM2, SNORA46, and SNORA48 were expressed at significantly higher levels in I NP when compared to I P, secondary or de novo grade II, and grade III. On the other hand, no difference was found among other grades (Fig. 3 g-i).

Gene fusions have been known to play a major role in tumorigenesis since the discovery of the Philadelphia chromosome in chronic myeloid leukemia. More recently, their role has been investigated in brain neoplasms as well. In ependymomas, C11orf95-RELA fusions drive oncogenic nuclear factor-κB (NF-κB) to transform neural stem cells into tumor cells [40]. In IDH-wildtype gliomas, FGFR-TACC fusions have been investigated as possible therapeutic targets, with inhibition of the FGFR fusion transcript yielding promising preliminary results [13]. The role of fusions in meningiomas, however, is still under investigation [1, 17].

We used STAR-fusion [21] to identify potential novel fusion transcripts and to determine if fusion events were also differentially present in meningiomas of different grades [14]. We observed that grade I meningiomas that never progressed have a significantly higher number of rearrangements as identified by sequencing than grade I meningiomas that eventually did progress or grade II (both de novo and secondary) and grade III meningiomas, which had a smaller fusion burden (one-way ANOVA with Tukey’s multiple comparison test; Grade I NP vs. Grade I P, p = 0.0003; Grade I NP vs. Secondary grade II, p = 0.0006; Grade I NP vs. de novo Grade II, p = 0.0002; Grade I NP vs. Grade III, p = 0.0013) (Fig. 4a-b). No significant difference was found among grade I meningiomas that progressed, grade II (both de novo and secondary), and grade III. Among the identified fusion events (Table 4), we selected two novel NF2-involved fusion products not observed so far in meningioma or other tumors: NF2—ZPBP2 (Zone Pellucida Binding Protein 2) (chromosomes 22q and 17q) and NF2—OXCT1 (3-oxoacid CoA-transferase) (chromosomes 22q and 5p) (Fig. 4c and d, respectively) which led to a truncated and non-functional NF2 transcript. Of note, the NF2—OXCT1 fusions all occurred in meningiomas that progressed to higher grade or were secondary to progression. To validate these new fusions, we designed primer pairs specific for each fusion transcript and analyzed our samples with RT-PCR. We found that, indeed, there was clear concordance between the RNA-seq data and RT-PCR analyses (Fig. 4e and f for NF2—ZPBP2 and NF2—OXCT1, respectively). Of note, we observed that the novel NF2—OXCT1 fusion was found in case #1818 and its two instances of recurrence, i.e. cases #3254 and #3526 (Fig. 4f), suggesting the importance of this fusion event that was maintained in all 3 resected tumors from the same patient.

In addition to fusions implicating the NF2 gene, other fusion transcripts were observed in more than one meningioma sample including C10orf112-PLXDC2 (found in I NP, I P, and grade III); GAB1-HHIP-AS1 and HHIP-AS1--GAB1 found in a I NP sample; KANSL1-ARL17A (found in two I NP, two de novo grade II, one secondary grade II, and one grade III); MLLT3-CNTLN (one de novo grade II and two secondary grade III); RP11-444D3.1-SOX5 (two grade I NP, one grade III); and SAMD5-SASH1 (found in four grade I NP samples) (as summarized in Table 4). Of note, NF2-OXCT1 and MLLT3i-CNTLN fusions are found across all samples from the same patient (patient #4), i.e. for primary and secondary samples. None of these fusion transcripts was observed in pediatric brain tumors sequenced within the Children Brain Tumor Tissue Consortium (CBTTC) (data available on CAVATICA).

To determine if gene signature associated with different biological processes may be linked to meningioma progression, we performed GSEA analysis on the differentially expressed genes between grade I and grade II/III tumors. Surprisingly, the genes that distinguished grade I from grade II/III tumors were significantly overrepresented in gene signatures associated with different immune responses; this association was much reduced in grade II and III meningiomas. In particular, a strong association was found with ‘allograft rejection,’ ‘interferon gamma response,’ and ‘inflammatory response’ gene sets (p < 0.0001) (Fig. 5a).

To further validate the finding that the host immune response varies among different meningioma grades, we set out to analyze if the number of infiltrating immune cells was different between I NP and grade II/III tumors. We therefore generated 3 tissue microarrays (TMAs) from 71 meningiomas. Included within these TMAs were 15 patients whose tumors recurred with progression to a higher grade (13 patients with progression from grade I to grade II and 2 patients with progression from grade II to grade III) and 42 patients without evidence of tumor progression at an average follow-up of 6 years (18 WHO grade I, 17 WHO grade II, and 7 WHO grade III). These TMAs were stained for CD45 (leukocyte common antigen), a marker widely used to examine general immune infiltration and inflammation [37]. Grade I NP meningiomas had a noticeable CD45 positive infiltrate, with significantly higher numbers of CD45 positive inflammatory cells compared to grade II (secondary or de novo) and III tumors. This result was consistent with the GSEA analysis indicating increased immune activity in low grade meningioma microenvironment (one-way ANOVA with Tukey’s multiple comparison test; grade I NP vs. grade I P, n.s.; grade I NP vs. secondary grade II, p = 0.05; grade I NP vs. de novo grade II, p = 0.013; grade I NP vs. grade III, p = 0.006) (Fig. 5b-c).