Identification of New Genetic Clusters in Glioblastoma Multiforme: EGFR Status and ADD3 Losses Influence Prognosis

Glioblastoma multiforme (GB) is one of the most aggressive tumors. Despite continuous efforts to improve its clinical management, there is still no strategy to avoid a rapid and fatal outcome. EGFR amplification is the most characteristic alteration of these tumors. Although effective therapy against it has not yet been found in GB, it may be central to classifying patients. We investigated somatic-copy number alterations (SCNA) by multiplex ligation-dependent probe amplification in a series of 137 GB, together with the detection of EGFRvIII and FISH analysis for EGFR amplification. Publicly available data from 604 patients were used as a validation cohort. We found statistical associations between EGFR amplification and/or EGFRvIII, and SCNA in CDKN2A, MSH6, MTAP and ADD3. Interestingly, we found that both EGFRvIII and losses on ADD3 were independent markers of bad prognosis (p = 0.028 and 0.014, respectively). Finally, we got an unsupervised hierarchical classification that differentiated three clusters of patients based on their genetic alterations. It offered a landscape of EGFR co-alterations that may improve the comprehension of the mechanisms underlying GB aggressiveness. Our findings can help in defining different genetic profiles, which is necessary to develop new and different approaches in the management of our patients.


Introduction
Glioblastoma multiforme, IDH wild-type (GB-IDHwt), is the most frequent malignant brain tumor in adults and the most aggressive in nature, with an average survival of around 15-18 months [1,2]. GB displays, in addition to morphological heterogeneity, a wide genetic heterogeneity [3,4]. This fact is, despite constant efforts, one of the main causes of the absence of effective treatment and thus, survival analysis was performed. Overall survival (OS) was calculated as time from surgery to death. Event times were censored if the patient was alive at the time of last follow-up.
Multiplex Ligation-dependent Probe Amplification (MLPA) using the MLPA KIT P105 (version C1-C2) and ME024 (version A1-B1) was performed in accordance with the manufacturer's protocol (MRC Holland, Amsterdam, Netherland) [30,31]. Amplification products were separated on an ABI 310 Sequencer (Applied Biosystems, Inc, Foster City, CA, USA) and data analysis was made with Coffalyser excel-based software (MRC-Holland) [32], where relative probe values of probe-amplified products are compared with normal controls. These kits included multiple glioma and cancer related genes that were evaluated depending on their probes based on previous reports [30,31,33,34], as explained below. Both kits are approved for investigational-use only.

Status of EGFR: EGFRvIII, Copy Number Alterations and Interphase Fluorescence In Situ Hybridization
EGFR was studied by MLPA using the P105 kit (MRC-Holland). It includes 11 probes for exons 1-8, 13, 18 and 24 of EGFR. This design allowed us to determine on one hand, the presence of the variant III and on the other hand, somatic copy number alterations (SCNAs) of EGFR in the samples analyzed. Following previously published descriptions, to identify EGFRvIII we determined the average value for exons 2-7 probes and established the ratio with the average value of probes for exons 1, 8, 13, 18 and 24. Patients with ratios below 0.8 were considered to harbor the EGFRvIII [31,35]. SCNAs of EGFR were determined based on the average value of exons 1, 8, 13, 18 and 24 in order to exclude the ones that are frequently involved in EGFR variants. The thresholds applied classified the samples as no amplified (0.7 < x <1.3) or gained (x ≥ 1.3) for downstream statistical analysis. We also analyzed the EGFR gene status in interphase cells by dual-color FISH probes using interphase Fluorsence in situ hybridization (iFISH). For that purpose, we used tissue microarrays (TMAs) and the probe LSI EGFR SpectrumOrange/CEP-7 SpectrumGreen Probe from Vysis (Abbott Laboratories, IL, USA). Hybridizations were performed according to the manufacturer's instructions, and signals were counted in two different regions of 200 non-overlapping nuclei. We calculated the ratio between the average signal count of EGFR and the control probe CEP-7 (EGFR/CEP7 ratio). EGFR was considered to be amplified when the EGFR/CEP-7 signal ratio was >2 [36]. Cases were subclassified according to previous descriptions as GBs with high level of amplification (H-amp) when more than 20% of the cells showed more than 20 copies of EGFR. GBs with low levels of amplification (L-amp) included cases with 5-20% of cells with 4-12 copies of EGFR. Cases without amplification (N-amp) showed two copies of EGFR. The exact ratio was not calculated in cases with high amplification levels [19,36]. The aim of this validation study was to determine the concordance rates of these techniques. Concordance between both techniques was assessed by the Cohen's Kappa statistic. It has a range of 0-1.0 and κ values <0.2 means a poor agreement, 0.21 < κ < 0.4 means a weak agreement, 0.41 < κ < 0.6 means a moderate agreement, 0.61 < κ < 0.8 means a good agreement and 0.81 < κ < 1 means a very good agreement between techniques.

Analysis of Locus 9p21 and Other Glioma and Cancer-Related Genes
SALSA MLPA P105 and ME024 included a collection of probes to determine PTEN status and genes located on 9p21. Many probes in order to characterize CDKN2A-CDKN2B (p15INK4B-p14ARF-p16INK4A) were included. These kits also comprised probes for other genes located up and downstream on 9p21.3 (KLHL9, MIR31, MLLT3 and MTAP) and for genes located near on 9p (DOCK8 9p24.3 and GLDC on 9p24.1, and PAX5 on 9p13.2). In addition, a wide collection of probes addressed to many different genes were assayed, including: The variety of loci explored offered a detailed landscape of GB SCNAs.
To categorize MLPA call to SCNA value, we established deletions as x < 0.7, normal as 0.7 < x < 1.3 and gain as x > 1.3. When probe values within the different exons of one locus were heterogeneously distributed across some categories such as heterozygous and homozygous deletions, we defined the category in which more than 70% of probe values belonged. For convenience, homozygous and/or heterozygous deletions collectively were referred to as deletion, while amplification and/or gain as gain/amp. Non-canonical SCNAs such as gain/amp in CDKN2A, PTEN or TP53 and deletion in EGFR or CDK4 were not considered in the data analysis.

Statistical Analysis
The statistical analysis of the data was carried out according to the type of variable. Quantitative variables were evaluated using the Kolmogorov-Smirnov and Levene tests; depending on their results and their characteristics, Student's t-test, ANOVA, Mann-Whitney-Wilcoxon test or Kruskal-Wallis test were performed. For comparisons among categorical variables, Fisher's exact, Pearson's chi-squared, and Kruskal-Wallis test were used depending on the number or rows/columns and the expected frequencies. A survival analysis using the Kaplan-Meier method was also done. The statistical significance of these curves was calculated using the log-rank (Mantel-Cox) test. Significance was accepted at least at p < 0.050 level. Data were analyzed with SPSS (version 26) software (IBM, Madrid, Spain). To perform the clusters, the average-linkage method was used. Therefore, distance between clusters was obtained as the average distance between all possible pairs of cases from both clusters, providing robust groups. Euclidean distance was used because of the binary scale of the genetic variables. The choice of three clusters was decided from the hierarchical tree, according to the level of heterogeneity at which the clusters were combined. Cluster combination stopped when this level was roughly 70% of the total amount.

TCGA Analysis and Functional Protein Associations
We accessed data for GB samples from The Cancer Genome Atlas (TCGA) by using cBioPortal for Cancer Genomics (www.cbioportal.org) [12,37,38] to validate whether the SCNAs detected in our cohort were associated with EGFR amplification there. We studied the Genomic Profile "Putative copy-number alterations from GISTIC" for the latest dataset available in cBioportal for GB (TCGA, Provisional 604 samples). Copy number alteration data from 577 cases were obtained and further analyzed. A "User-Defined List", including EGFR, CDKN2A, MTAP, TRAF4, JAG1 and MSH6, was entered into the "Enter Gene" box. Samples were classified for each gene according to their putative copy number variation calculated by GISTIC with default cBioportal thresholds 33. The groups were Diploid (0), Shallow Deletion (−1), Deep deletion (−2), Gain (1) and Amplification (2). The associations between EGFR amplification and the copy number profile were analyzed using the "Plots" tool and retrieving the raw data. Outlier values (n < 10 from 577 cases) were negligible. Statistical significance for amplification vs. diploid and deletion (shallow or deep) vs. diploid for each gene was assessed using Fisher's exact test.
We also used the Search Tool for the Retrieval of Interacting Genes/Proteins (STRING) database of known and predicted protein-protein interactions (https://string-db.org) to establish associations among the proteins encoded by the genes affected by SCNAs in this series. This database contains information from numerous sources, including KEGG, Reactome or Pubmed, among others [37]. To make a restrictive analysis, we explored both evidence and confidence meaning of network edges and we increased the minimum required interaction score to 'high confidence (0.700)' to reach a greater reliability.

Clinical and Histopathological Data
Our study included 137 patients. Clinical data, including age, sex, tumor location, size, initial symptom and Karnofsky performance status scale (KPS), are summarized in Table 1. Mean age at diagnosis was 57.7 years and male/female ratio was 1.17. OS was 210 days and did not reflected statistical differences depending on sex nor age. Histologically, all tumors showed features of GB with pleomorphic astrocytic tumor cells, prominent microvascular proliferation and necrosis ( Figure 1). From confirmed primary GB (n = 135), the majority were IDHwt (n = 128) but a little subgroup of tumors displayed IDH mutations (n = 7) despite being primary GB (GB-IDHmut). In agreement with the WHO 2016 classification, these GB-IDHmut-affected patients were significantly younger (40.9 years in IDHmut vs. 59 years in IDHwt, p < 0.001 **). OS was also significantly higher in IDHmut than in IDHwt patients (3300 days vs. 180 days, p < 0.001 **). From the 128 cases, 63.9% of patients were >55 years old at diagnosis, the 56.3% were male and 76.9% had a KPS ≤ 85 preoperatively. Interestingly, tumor Pearson correlation test demonstrated an association between tumor size and OS (p = 0.020 *).

EGFR Status Assessment by iFISH and MLPA
iFISH analysis showed a distribution of cases among N-amp, L-amp and H-amp of 35.6%, 12.7% and 51.7%, respectively ( Figure 1). Similarly, MLPA analysis, which only separates two categories, showed that 29.7% of cases had no EGFR gains, whereas 70.3% of the cases displayed gain of EGFR copies. Comparing FISH H-amp and L-amp with MLPA "gain", and FISH N-amp with MLPA "no gain", concordance between both techniques was good, with a coincident result in 83.05% of the samples (Cohen's kappa index: 0.610). Agreement was 93.4% for L-amp and H-amp GBs assessed by iFISH that were detected as gains by MLPA. Regarding N-amp tumors, the concordance dropped to 64.3% (Supplementary Table S1 shows the result from both techniques for each case). The visual determination of EGFR made iFISH more suitable to categorizing the cases according to their amplification status. Thus, the following analyses regarding EGFR amplification status were established according to the iFISH data. Among the 128 GB patients, we found no significant associations between EGFR amplification and survival (p = 0.387). Clinical data did not reveal any strong association to the amplification status either: 50.8% of men and 49.2% of women displayed H-amp EGFR, reflecting no differences depending on the basis of sex. We did not find significant associations between EGFR amplification and patient age.

EGFR Variant III Is More Frequent in Women and Is Associated with Shortened Survival
MLPA allowed us to determine the presence of EGFRvIII. IDHwt GBs (n = 128) exhibited EGFRvIII in 35.2% of the cases (n = 45). Clinical data revealed no differences between both groups regarding age, tumor location nor size. Nevertheless, there was a significant increase in the presence of this variant in women (55.6% EGFRvIII vs. 37.3% EGFRwt, p = 0.047 *). Interestingly, Kaplan-Meier analysis revealed statistical differences in OS ( Figure 2, p = 0.014 *), accounting 150 days in EGFRwt vs. 90 days in EGFRvIII (Student t-test p = 0.027 *). In addition, EGFRvIII was significantly more frequent in GBs with EGFR amplification. We found it in 10 cases of N-amp GBs (23.8%), 2 cases of L-amp GBs (13.3%) and 31 cases of H-amp GBs (50.8%, p = 0.003 **).

EGFR Variant III Is More Frequent in Women and Is Associated with Shortened Survival
MLPA allowed us to determine the presence of EGFRvIII. IDHwt GBs (n = 128) exhibited EGFRvIII in 35.2% of the cases (n = 45). Clinical data revealed no differences between both groups regarding age, tumor location nor size. Nevertheless, there was a significant increase in the presence of this variant in women (55.6% EGFRvIII vs. 37.3% EGFRwt, p = 0.047 *). Interestingly, Kaplan-Meier analysis revealed statistical differences in OS ( Figure 2, p = 0.014 *), accounting 150 days in EGFRwt vs. 90 days in EGFRvIII (Student t-test p = 0.027 *). In addition, EGFRvIII was significantly more frequent in GBs with EGFR amplification. We found it in 10 cases of N-amp GBs (23.8%), 2 cases of L-amp GBs (13.3%) and 31 cases of H-amp GBs (50.8%, p = 0.003 **).

MLPA Analysis Showed a Great Heterogeneity in GB
We identified that 100% of the tumor cases showed SCNA in at least two of the genes analyzed. In addition, from all the loci explored, we found SCNA in all of them in at least seven cases. An overview of the prevalence of genetic alterations identified in 128 GB-IDHwt patients is shown in Table 1. A summary of all the SCNAs detected among the studied genes can be seen in Figure 2A which offers a landscape of the high genetic heterogeneity found. We detected alterations affecting more than 45% of the cases on EGFR (70.1% of the cases), CDKN2A (65.6%), TIMP3 (64.1%), MEN1 (57.8%), CDKN2B (56.3%), MVP (56.3%), PTEN (54.8%), MTAP (53.5%), ADD3 (46.9%) and PCCA (46.6%). From those loci, only ADD3 tend to be more altered in women (55.4% of the cases) than in men (40.3% of the cases). However, ADD3 demonstrated to be associated with OS, as patients with SCNAs on ADD3 showed an OS of 6.98 ± 1.17 months while patients with no-SCNA on ADD3 showed an OS of 13.46 ± 2.24 months (p = 0.012 *). Long Rank (Mantel-Cox) analysis demonstrate a statistic association (p = 0.014 *, Figure 2B).

EGFR Amplified GBs Displayed Different SCNAs to Non-EGFR Amplified Cases
In this series, four genes revealed statistical differences on their affectation, depending on the amplification status of EGFR: losses/gains of MSH6 on 2p16.3, losses of CDKN2A and MTAP, both on 9p21 and gains of JAG1 on 20p12.2 ( Figure 2C). TCGA analysis through cBioportal supported our data, as it showed strong associations between both CDKN2A and MTAP losses and EGFR gain/amp (p < 0.0001 ***). Of note is that both genes are located in 9p21. TCGA data also showed statistical association between JAG1 gain/amp and EGFR gain/amp (p < 0.0001 ***). MSH6 showed alterations in a little number of GBs and did not reach a significant result (Table 2). STRING analysis provided a PPI enrichment p-value of 0.000987 and association to NOTCH3 activation and to a negative regulation of cell-matrix adhesion. Based on previous reports [7], we looked for GB with triple SCNA (EGFR, CDKN2A and PTEN); in our series, this appeared in 25.0% of the cases and they showed an OS of 7.64 ± 1.97 months, which was lower than the 11.25 ± 1.71 months in cases with no triple SCNA.

Clustering Analysis Revealed Different Genetic Glioblastoma Groups
Hierarchical cluster classification distinguished three groups (C1, C2 and C3) depending on the frequencies of alteration within the different loci explored (Table 3). In this analysis we excluded 37 patients because some markers were not available. From the 91 patients to classify, 18 were unclassifiable subjects due to the diversity of the SCNAs found and 73 were distributed among the three different groups performed. The clusters showed partially overlapped changes and others completely differentiated among them (Figure 3). The Chi-squared test was used to assess the dependence between the normal/altered presence of a gene and the membership cluster in order to identify which genes have a higher power of discrimination. When the expected frequency in cells of the cross-table was too small (n < 5) in more than 33% of cells, the Kruskal-Wallis test was used as an alternative to Chi-squared. Regarding clinical data, size was similar in the different clusters, showing an average of 5.7 cm in C1, and 5.3 cm in C2 and 4.9 in C3. Age at diagnosis also showed similar averages (59, 57 and 61 years, respectively). Overall survival was 7.2 months for C1, 5.9 for C2 and 10.7 months for C3. Genetically, the analysis showed that C1 was the least affected, C2 showed alterations of near half of the loci explored in more than 50% of the cases, and C3, displayed an intermediate situation, with near 30% of the genes included affected in more than 50% of the cases. When we analyzed the EGFR amplification status by iFISH in relation to these clusters, we found statistically significant differences (p = 0.007 **): most cases from C1 were N-amp (63.0%), compared to 28.0% and 19.0% in C2 and C3, respectively. Most cases from C2 and C3 where H-amp (64.0% and 61.9%, respectively), compared to 22.2% in C1. This distribution of the cases for the L-amp group was more homogeneous, accounting 14.8%, 8.0% and 19.0% in C1, C2 and C3, respectively.

Genetic Changes According to Clustering Analysis Point to Differentially Altered Pathways
EGFRvIII and losses in ADD3, associated with survival, were concentrated in cluster 2. In concordance, C2 displayed as aforementioned, the shortest OS. Cluster 3, in common with C2, showed losses in CDKN2A in 100% of the cases, a high frequency of EGFR amplification and SCNA in MTAP and MSH6. Additionally, this C3 displayed frequent SCNAs in TP53, IL4, PCCA and SIX3. STRING analysis showed a PPI enrichment p-value of 0.0019 and an association to the biological processes 'regulation of cell cycle phase transition' with an FDR of 0.0060. However, no specific cellular component was revealed. Finally, C1 was the less altered cluster. It showed a statistically significant lower level of CDKN2A, MSH6, MTAP and EGFR alterations compared with its counterparts.
Overall, GB-IDHwt displayed a wide genetic heterogeneity. However, cluster analysis allowed a separation based on the frequency of alterations detected by MLPA into three groups. From them, the two that displayed EGFR amplification offered completely different outcomes, depending on the Cluster 3, in common with C2, showed losses in CDKN2A in 100% of the cases, a high frequency of EGFR amplification and SCNA in MTAP and MSH6. Additionally, this C3 displayed frequent SCNAs in TP53, IL4, PCCA and SIX3. STRING analysis showed a PPI enrichment p-value of 0.0019 and an association to the biological processes 'regulation of cell cycle phase transition' with an FDR of 0.0060. However, no specific cellular component was revealed. Finally, C1 was the less altered cluster.
It showed a statistically significant lower level of CDKN2A, MSH6, MTAP and EGFR alterations compared with its counterparts.
Overall, GB-IDHwt displayed a wide genetic heterogeneity. However, cluster analysis allowed a separation based on the frequency of alterations detected by MLPA into three groups. From them, the two that displayed EGFR amplification offered completely different outcomes, depending on the presence of additional alterations, as SCNA on ADD3 and the variant III of EGFR. These two changes were shown to be independent biomarkers for bad prognosis, and their statistical association to C2 highlights the interest of exploring the aggregation of genetic alterations in it.
Cells 2020, 9, x FOR PEER REVIEW 13 of 19 presence of additional alterations, as SCNA on ADD3 and the variant III of EGFR. These two changes were shown to be independent biomarkers for bad prognosis, and their statistical association to C2 highlights the interest of exploring the aggregation of genetic alterations in it.

Figure 4.
Pathways outlined by clustering analysis. The introduction of genes that differentially contributed to the clusters on the STRING database analysis platform offered a variety of genetic pathways and processes that were especially damaged in C2 and C3. Genes in blue displayed gains, genes in orange displayed losses, EGFRvIII includes loss from exons 2-7. Genes in bold were the most distinctive among the clusters. It needs to be mentioned that MSH6, in addition to gains, showed losses in < 10% of the cases. The main difference between C2/C3 and C1 is that on C1, EGFR alterations were found only in half of the cases. Despite C2 and C3 share some alterations, the downstream connections lead to a subtly quickest disease on C2 compared to the slowest situation on C3.

Discussion
EGFR genetic alteration is an essential component of the portrait of most GBs occurring in 57% of tumors [4,38]. This frequency is similar to what we find for EGFR amplification and EGFR mutation in this work. This fact, along with the devastating outcome of GB, justifies the continuous search for associations in relation to EGFR changes and the therapeutic responses of patients. In light of the TCGA project, RTK alterations and their downstream effectors are of potential interest as targetable driver mutations [39]. However, it could be especially interesting to be able to classify our patients depending on specific common changes or even better, different pathways affected, to understand what makes them undergo a quicker or slower disease.
The use of iFISH to precisely determine the EGFR amplification status is the current goldstandard in GB [19,40]. Previous works of our and others groups, demonstrated that this method allows the detection of intermediate levels of amplification [19,24]. Lassman and his colleagues deepened the potential of different new techniques in comparison to iFISH with positive results [40]. Pathways outlined by clustering analysis. The introduction of genes that differentially contributed to the clusters on the STRING database analysis platform offered a variety of genetic pathways and processes that were especially damaged in C2 and C3. Genes in blue displayed gains, genes in orange displayed losses, EGFRvIII includes loss from exons 2-7. Genes in bold were the most distinctive among the clusters. It needs to be mentioned that MSH6, in addition to gains, showed losses in <10% of the cases. The main difference between C2/C3 and C1 is that on C1, EGFR alterations were found only in half of the cases. Despite C2 and C3 share some alterations, the downstream connections lead to a subtly quickest disease on C2 compared to the slowest situation on C3.

Discussion
EGFR genetic alteration is an essential component of the portrait of most GBs occurring in 57% of tumors [4,38]. This frequency is similar to what we find for EGFR amplification and EGFR mutation in this work. This fact, along with the devastating outcome of GB, justifies the continuous search for associations in relation to EGFR changes and the therapeutic responses of patients. In light of the TCGA project, RTK alterations and their downstream effectors are of potential interest as targetable driver mutations [39]. However, it could be especially interesting to be able to classify our patients depending on specific common changes or even better, different pathways affected, to understand what makes them undergo a quicker or slower disease.
The use of iFISH to precisely determine the EGFR amplification status is the current gold-standard in GB [19,40]. Previous works of our and others groups, demonstrated that this method allows the detection of intermediate levels of amplification [19,24]. Lassman and his colleagues deepened the potential of different new techniques in comparison to iFISH with positive results [40]. In the present work, we compared the potential of MLPA to that aim because it is a user-friendly and cost-effective technique, and the Cohen agreement we got was also good. The point of iFISH remains the possibility of separating that intermediately amplified group, for which it is not entirely clear whether it represents a progression via the amplification status or a separate path for tumor progression in GB. In any case, MLPA proves to be an easy and fast technique, with the additional advantage of being able to determine mutant variants such it is EGFRvIII, in agreement with previous works [21,24,31,35].
The wide heterogeneity of cancer cells is a common challenge in terms of learning how to stop tumor growth. The clonal evolution of GB IDHwt and the acquisition of new mutations represent a major problem for finding effective therapies [1,4,8,23]. This heterogeneity is patent in our work; from all the loci explored, we found SCNA in all of them in at least one case. The genetics of H-amp GB and GB displaying EGFRvIII offers landscapes that are statistically different from their N-amp/L-amp or EGFRwt counterparts, respectively. STRING analysis correlates the SCNA statistically associated with amplified-EGFR GBs with NOTCH3 activation and negative regulation of cell-matrix adhesion. Crespo et al., using high-density (500K) single-nucleotide polymorphism arrays, achieved similar relevance for the axis established with MTAP [16]. In concordance with that work, our group of patients that displayed EGFRvIII, which is the one with the worst outcome, also showed the impairment of cell-matrix adhesion processes. Despite the alteration of different sets of genes, we found a higher frequency of EGFRvIII among the EGFR amplified GBs, in concordance with previous works [9,24], and interestingly, we found a significant association between EGFRvIII and a shortening in survival, supporting previous descriptions [22,27,41].
An outstanding finding when analyzing the genes whose alterations were associated with EGFRvIII is that ADD3 SCNAs is associated with bad prognosis, with a marked reduction in OS. ADD3 codifies the γ-adducin, which build heterotetramers with its counterparts αand β-adducin, and has been widely studied in red-cell membranes [42]. Different works put a spotlight on adducin's controversial role as either oncogene or tumor suppressor in cancer [42][43][44][45]. Our findings support a recent report of Kiang KM et al. that points to the downregulation of ADD3 in GB, but not in less malignant gliomas, as a critical event during malignant progression [46]. While most cell-based studies suggest an oncogenic behavior, different papers on glioma tumor specimens, in agreement with our data, relate ADD3 downregulation to progression [43,44] and to migration [45]. Other cancer that shows EGFR amplification, such as non-small cell lung cancer, shares this ADD3 infraexpression associated with cell migration [47]. We suggest that the use of such easy techniques as MLPA to assess ADD3 SCNA could be considered for diagnostic routines to better tailor clinical decisions as an independent prognostic factor and to delve deeper into the GB classification of patients.
The frequency of ADD3 SCNAs and its co-occurrence with EGFRvIII led us to look for genetic clustering. Thus, according to the genetic probes analyzed by MLPA, clustering analysis causes GBs to separate into three different groups. The first cluster (C1) offers a low rate of SCNAs compared with the rest. Cluster 2 (C2) and cluster 3 (C3) share a high rate of SCNAs in MTAP, CDKN2A, MSH6 and EGFR, in agreement with previous descriptions [16], and suggesting that GBs from both clusters display alterations in DNA metabolic processes (by STRING). However, the second cluster (C2) concentrates the co-existence of EGFR alterations through EGFRvIII or EGFR amplification and the highest rate of SCNAs. Of these, it is worth mentioning that JAG1, which is involved in angiogenesis [48], MVP, was recently implicated in vesicle trafficking [21] or the aforementioned ADD3. Coherent with these data, this cluster displayed the shortest OS. It is curious that MSH6, in addition to showing gains as was previously reported in association with resistance to chemotherapy [49], displays losses in a short set of cases in this series. These cases with losses do not show an increase on the tumor genetic burden consequence of the defect on DNA repair, but interestingly, they show a really subtle increase in survival. It is not significant, but it deserves to be further studied, as it resembles the protective effect of the promoter methylation of MGMT, which improves responsiveness to temozolomide treatment [1,50]. Regarding MTAP, its deficiency is usually seen as a collateral effect of CDKN2A deletion, because of its location next to this gene [51,52]. Nevertheless, it is necessary to improve our understanding of the consequences of this loss because it could offer different insights into therapy. It is known that the loss of MTAP influences the metabolism of ATP: both adenine and adenosine are disturbed as a consequence of the disruption of the polyamine salvage pathway [53,54]. This fact could be related to the metabolic reprogramming strategy to actively modulate the immune landscape of GB [55]. Contrary to what was expected, it was shown that adenosine did not significantly accumulate in GB [55]. This fact would be in line with previous descriptions of N6-isopentenyladenosine and other modified nucleosides, with important anti-proliferative and pro-apoptotic effects in GB cases that display amplification on EGFR [56,57]. Interestingly, in our study, cluster 2 and 3 showed amplifications on EGFR and losses on CDKN2A in all cases, along with a high proportion of cases displaying MTAP losses. It would be desirable to further study whether the efficacy of those molecules could represent a therapeutic benefit for these specific patients with adenosine metabolism disturbances.
A noteworthy fact is that ADD3 and EGFRvIII, both previously reported as alterations associated with shortened survival in GB [27,46] and confirmed here, are a signature in this C2. On the other hand, C3 involves TP53 alterations in addition to the shared ones. This fact completely makes sense considering that TP53 mutations are, according to the WHO, more frequent in the group of GB with IDH mutations, characterized by a higher survival than IDHwt GBs [1]. Moreover, EGFRvIII has been broadly associated with a poorer outcome [22,27]. These findings reflect that, not only the amplification status of EGFR could be decisive in the comprehension of GB progression, but also the interconnection established with parallel genetic pathways.
A comparison between the main findings in the clusters defined here and the previously reported TCGA expression subtypes [10,38] is worthy of some comments: the small set of GB, IDH-mutant cases we studied may fit in the proneural TCGA group, with mutations in TP53 in half of the samples and loss of PTEN in 2/3 of the cases. Similarly, lifespan is higher, as we expected for being IDH1-mutant [1]. Cluster 2 fits quite well in the "Classical" subtype [10,38]. EGFR amplification is the main characteristic accompanied by losses on CDKN2A in all the cases and losses on PTEN in a high proportion (88%). However, cluster 3 seems to be a variant of this classical subtype. It is also characterized by EGFR amplification and losses on CDKN2A, but losses on PTEN drop down to a half. However, other alterations define it better than PTEN status, such as gains on TP53 or losses on SIX3. It is worth mentioning that TP53 stands out, but it does so because of SCNA, and not mutations, as would be more characteristic of the mesenchymal TCGA subtype [10,38]. Thus, between these two clusters, a better genetic definition of the patients is offered. It is of note that C1 seems to have little to do with EGFR, in contrast to C2 and C3. C1 agrees with the mesenchymal subtype, being the only set of patients that are not characterized by EGFR amplification. It displays a complex mix of genetic changes without dominance of any specific feature, all them in lower proportions than its counterparts cluster 2 and 3, and in a similar way as happens in the mesenchymal TCGA subtype [10,38]. On both sides, EGFR status-dependent subgroups of GBs, genetically and clinically different, can be separated. The identification of alterations in shared nodes of convergence downstream of RTKs hast been an interesting approach in cancer [4,8,58].
The present work sets out MLPA as an advantageous methodology, simple and useful for FFPE specimens. In addition, it may provide new insights into the molecular underpinnings of GB pathogenesis in a comprehensive manner. Clustering the genetic alterations of GB highlights the importance of EGFR in this very aggressive tumor type and could represent a strong step towards precision medicine: the aggregation of changes depending on the presence of the amplification of EGFR, the mutation variant III or both simultaneously, lead to different pathways to analyze. Our results underline the importance of EGFRvIII and ADD3 SCNAs as markers of poor prognosis that need further consideration in GB. The presence of a group of GB-IDHwt without alterations in EGFR may explain part of the absence of effect of RTK inhibitors in this type of tumor. Furthermore, the clear separation of EGFR-amplified related GBs, showing sets of genes that are differentially altered, points to the need to rethink the possibilities of personalized therapy in future clinical settings. The differential groups that can be established could be used for a more accurate therapy.