Glioblastomas are composed of genetically divergent clones with distinct tumourigenic potential and variable stem cell-associated phenotypes

The development of genomic instability leading to the evolution of aneuploid cell lineages is a hallmark of many cancers and was early recognized to be associated with a poor prognosis [7, 26, 27, 34]. Therefore, difference in DNA content is a first layer of genomic individuality of clones where differences in ploidy can be used as a proxy for genetic heterogeneity. To assess the intra-tumoural clonal architecture and the extent of genetic heterogeneity in primary GBMs, we performed flow cytometric analysis based on DNA staining (i.e. using DAPI or Hoechst) on 36 patient biopsies to discriminate the DNA content present in individual nuclei of a complex tumour sample (supplementary Fig. 1). Assuming that each patient biopsy also contained normal stromal cells, the first histogram peak in the DAPI channel typically was considered diploid and each successive peak represented either the diploid 4N (G₂/M)fraction (Fig. 1a, top) or a distinct aneuploid population within the tumour (Fig. 1a, bottom; supplementary Fig. 1a). In this context, we considered ploidies to be unique when differences in the DNA index (DI) of DAPI stained nuclei were ±0.1 (differences in ploidy of ±0.2N). Analysis of 36 tumour biopsies showed that GBMs display heterogeneous profiles based on DNA content and corresponding copy number variation as determined by FS-array CGH (Fig. 1a, b; supplementary Table 1). 23 GBMs (64 %) were monogenomic and 13 GBMs (36 %) were heterogenomic (polygenomic) having at least one additional aneuploid peak (Fig. 1c; supplementary Table 1).In two patients (T113 and T145) a second aneuploid peak was detected (Fig. 1c; supplementary Table 1). We found no association between the ploidy profile and the patient age (student t test, p = 0.39) and sex (Chi squared test, p = 0.63) (supplementary Table 1). Of note, five patients underwent a second surgery after recurrence, of which four tumours were monogenomic and one was polygenomic in the primary biopsy, which upon recurrence retained their ploidy level (supplementary Table 1).

To assess the genomic profiles of the ploidy-based clones in patient biopsies, each individual DNA peak was FACS-sorted and sorted nuclei were analysed with array CGH to obtain high-definition clonal profiles of the copy-number aberrations in each tumour genome (Fig. 1b; supplementary Fig. 2). The aberrant genomic intervals in each population were determined by the aberration detection algorithm ADM2 [31]. All the monogenomic tumours exhibited genetic aberrations typical of a GBM genome (Fig. 1b upper panel, additional examples in supplementary Fig. 2a), confirming the presence of tumour cells in each patient biopsy. In polygenomic tumours, all sorted aneuploid clones purified from patient biopsies also exhibited a typical GBM profile [11] (Fig. 1b lower panel; supplementary Fig. 2b ‘red’). Interestingly, in all cases of polygenomic samples the FACS-sorted diploid peak also exhibited genomic aberrations (Fig. 1b middle panel; supplementary Fig. 2b ‘blue’). The lower resolution of aberrations detected by array CGH in certain cases in the diploid fractions compared with aneuploid fractions were likely due to the presence of admixed non-tumour cells within the same DAPI peak, thereby diluting the tumour clone and obscuring tumour-specific genomic aberrations (e.g. T101GBM; supplementary Fig. 2b).

For all monogenomic and polygenomic tumours analysed, aberrations were seen at the level of large-scale structural aberrations such as gains or losses of whole chromosomes or chromosome arms as well as at the level of more focal events including high-level (log² ratio >1) amplifications and homozygous(log² ratio < −3) deletions. Importantly, the number of ADM2-defined intra-chromosomal copy-number aberrations in sorted populations varied widely between individual patient biopsies (from 10 in several samples up to 129 in the aneuploid fraction of T221) (supplementary Fig. 2) illustrating the individuality of GBM genomes. Although the same GBM associated genes were often targeted by recurrent aberrations (e.g. amplification of EGFR, homozygous deletion of CDKN2A or chromosome 10; Fig 1c), we did not detect any common breakpoints between different tumours for a given targeted gene, highlighting the uniqueness of each patient’s tumour genome (not shown). Moreover, we did not detect any obvious correlation between typical GBM genetic aberrations and the ploidy profiles of the tumours (Fig. 1c).

In summary, we show that individual GBM patients harbour unique tumour genomes both at the level of their DNA content and at the level of their chromosomal structure. A small majority of GBMs (64 %) were found to be diploid by flow cytometry, versus 36 % of polygenomic tumours containing diploid and aneuploid tumour cell populations. Importantly, recurrence of the disease did not necessarily involve aneuploidization and appearance of new divergent clones.

Since all of the monogenomic tumours displayed only one diploid DNA peak, we assumed that they were composed of an abnormal tumour cell population with the approximate (eu)ploidy of normal stromal cells. Indeed all monogenomic tumours were confirmed by array CGH to contain a rearranged genome (Fig. 1a; supplementary Fig. 2a). To distinguish admixed tumour and stromal cell populations within one DNA peak, we performed ploidy analysis based on Hoechst staining combined with cell membrane phenotyping in viable cells (Fig. 2a). Hoechst 33342 is regularly used for assessment of efflux properties [Side Population (SP) phenotype] in (cancer) stem cells and we have previously shown that the gating strategy for the SP phenotype relies strongly on the ploidy of tumour cells [20]. Therefore, to assess ploidy in viable cells carrying the SP phenotype it is essential to inhibit Hoechst efflux. This does not, however, apply to GBM, since we [19] and others [9] have recently shown that GBM tumour cells do not carry efflux capacities. Indeed in GBM tissue only stromal endothelial cells were found to display the SP phenotype [19]. Therefore, Hoechst staining could be used directly for ploidy assessment and FACS-sorting of viable cells. Interestingly, in all tested biopsies EGFR⁺ tumour cells displayed a significantly higher DI than stromal haematopoietic cells (CD45⁺/CD44⁺; Fig. 2a; supplementary Fig. 4a). Tumour cells could thus be designated as “pseudodiploid” [36] as they carry an apparently diploid genome (by DNA content) despite harbouring multiple chromosomal aberrations including amplifications and deletions (see T251 GBM example on Fig. 2b arrows and inset).

To further confirm an admixture of tumour and stromal cells within monogenomic biopsies, we performed array CGH on single nuclei sorted from the same uniform DAPI peak of the T16 patient biopsy (Fig. 2c). DNA of each sorted single nucleus was amplified by WGA before array CGH was performed. Ultra-diluted and amplified female DNA was used as a reference. For the T16 GBM, 16 independent nuclei from the detected DNA peak were analysed and compared with the “bulk” sorted DNA peak. A majority of the single nuclei (14/16) showed aberrations also detected in the bulk tumour peak (Fig. 2c, red versus blue line), but occasional nuclei (2/16) displayed a flat profile without any of the aberrations detected in the bulk tumour sample (Fig. 2c, green), representing admixed stromal cells in the biopsy material. Although the resolution of single nucleus array CGH is limited owing to the very small amount of starting DNA, our data indicate that the technique allows to successfully assess clonal diversity within one DNA peak (Fig. 2c, arrows indicating the ‘major’ chromosomal aberrations established previously in the ‘bulk’ tumour).

Thus by applying two different methods we demonstrate that monogenomic diploid GBMs consist of an aberrant pseudodiploid tumour cell population admixed with normal stromal cells. Hoechst-based analysis combined with phenotyping in viable cells and single nucleus sort combined with array CGH analysis are powerful tools to analyse the heterogeneity of cells present within a complex tumour sample displaying similar ploidy level.

Since all polygenomic GBMs contained a diploid fraction, we first confirmed the presence of tumour cells in the diploid fraction and validated the purity of the nuclei sort by applying Hoechst-based ploidy analysis (see example of T238 GBM on Fig. 2d). EGFR⁺ tumour cells showed two distinct clones with varying ploidy (Fig. 2d, right). Similar to the monogenomic GBMs, tumour cells within the diploid peak appeared pseudodiploid, with a shift at the DNA level compared with diploid control cells, further confirming the genomic aberrations present in this fraction.

To understand the clonal progression between diploid and aneuploid clones in GBM we performed an in-depth comparison of aberrations present in diploid and aneuploid tumour cells. We have shown in Fig. 1b and supplementary Fig. 2b that aberrations detected by array CGH in the diploid fractions were also present in the aneuploid clones of the same tumours, suggesting that tumour populations are clonally related. The same chromosomal aberrations were detected at the level of large-scale structural aberrations (e.g. chromosomal loss) as well as at the level of more focal events (high level amplifications and homozygous deletions) (Fig. 1b, additional examples on supplementary Fig. 2b). Moreover, different clones present within one patient biopsy carried the exact same chromosomal breakpoints for any given aberration. Figure 2e shows the chromosomal breakpoints for the PTEN deletion and the MET amplicon in the T176 GBM (additional examples on supplementary Fig. 3). This was also true for all other polygenomic GBMs analysed interrogating, e.g. amplifications of MDM2, CDK4, EGFR or deletions of CDKN2A and PTEN (supplementary Fig. 3).

Taken together, our data indicate that aneuploid GBM cells carry the genomic aberrations already present in the pseudodiploid tumour cells. Since the genetic aberrations were shared between clones of the same biopsies, and at the same time aberrations varied widely between patients, it can be inferred that tumour populations within one patient biopsy are clonally related and likely share a common ancestor. By applying the principle of non-reversibility of acquired somatic events and by taking into account that all GBMs contained a pseudodiploid fraction, our data suggest that aneuploidization is a late event in GBM, which occurs once initial driver aberrations/mutations have been acquired by the tumour genome.

To characterize the phenotypic properties of divergent genetic clones present within the same GBM we established intracranial xenografts in NOD/Scid mice from a number of mono and polygenomic GBMs (supplementary Table 2). We and others have previously shown that xenografts based on organotypic spheroids closely maintain the genetic, phenotypic and behavioural profiles of the parental patient tumours [15, 28, 57]. DAPI-based analysis revealed that the heterogeneity present at the ploidy level in patient biopsies was recapitulated in the respective xenografts (Fig. 3a; supplementary Table 2). Importantly, we were able to discriminate tumour nuclei (‘black’) from the stromal compartment (‘green’) by human-specific lamin A/C positivity (Fig. 3a). Stromal nuclei were used as an internal diploid control as human and mouse nuclei have the same DNA index (supplementary Fig. 4b). An example of a monogenomic (T331) and a polygenomic (T341) tumour is depicted in Fig. 3a. The presence of distinct pseudodiploid and aneuploid clones was further confirmed with Hoechst-based ploidy discrimination in eGFP expressing NOD/Scid mice in multiple GBM xenografts (supplementary Fig. 4c, d). The exact recapitulation of the ploidy clones from patient biopsies in spheroid-based xenografts was true for all tumours analysed, except for T16 GBM, where the aneuploid population was only detected in the xenografts (supplementary Table 2). This may be explained by a very small aneuploid population in the biopsy or, more likely, by the spatial heterogeneity of the tumour, where the aneuploid fraction was present only in the specimen part used for spheroid derivation and not in the part used for initial ploidy analysis of the parental tumour. However, the possibility of cell aneuploidization occurring in the xenograft cannot be completely ruled out.

We then determined whether in xenografts divergent clones from polygenomic tumours retained the genetic aberrations present in the patient biopsies by applying DAPI-based FS-array CGH. Both diploid (2N in blue) and aneuploid (AN in red) clones were purified from the patient biopsy (T101) and from two serially transplanted xenografts (generation 1 and 6) (Fig. 3b). Diploid and aneuploid clones remained remarkably similar upon serial xenotransplantation and retained their initial ploidy as well as all major aberrations and identical chromosomal breakpoints as seen in the parental biopsy (Fig. 3b, arrows).

Taken together, we show that GBM ploidy-based genomic heterogeneity can be recapitulated in organotypic spheroid-derived xenografts. Importantly, tumour cells retain their initial ploidy and the genetic aberrations upon serial transplantation in the mouse brain. This is in contrast to in vitro cultures of patient-derived glioma cell lines, which are known to undergo strong selection, aneuploidization and acquisition of new genetic aberrations in culture [30]. Of note in this context, GBM stem-like cell lines grown under serum-free conditions were found, similar to classic adherent glioma cultures, to be exclusively aneuploid, with only one out of seven lines (TB107) containing both diploid and aneuploid cells (supplementary Table 2).

We then assessed whether aberrant clones were independently capable of initiating tumours in mice. Using Hoechst-based discrimination of viable cells, distinct tumour clones (pseudodiploid and aneuploid) were isolated from a polygenomic GBM transduced with DsRed expressing lentiviral vectors (Fig. 4a, T16 xenograft). Diploid (2N, blue) and aneuploid (AN, red) tumour clones were separately tested for their spheroid formation capacity in vitro and tumourigenic potential in vivo. Both diploid and aneuploid cells formed spheroids in vitro with a similar efficacy and similar size within 10 days of culture (Fig. 4b). To assess the tumourigenic potential in vivo, six spheroids of each clone were implanted intracranially into eGFP⁺ NOD/Scid mice (n = 5 per group). All mice developed tumours as observed by fluorescence imaging (Fig. 4c). However, mice carrying aneuploid spheroids displayed significantly shorter survival times compared to those engrafted with either diploid or unsorted bulk spheroids (98.6 ± 1.56 versus 109 ± 1.47 and 112.4 ± 1.53 days, respectively, p = 0.021 2N versus AN, p = 0.0002 AN versus bulk unsorted) (Fig. 4d). Subsequent Hoechst-based ploidy analysis of the tumours showed that within one xenograft generation clones were genetically stable and had retained their initial ploidy (Fig. 4e). Interestingly, during serial transplantation of unsorted T16 spheroids the aneuploid cells eventually outgrew the diploid population (Fig. 4f), confirming the differential in vivo growth potential of the two distinct populations. A similar behaviour was seen for T238 with a decrease in the diploid population and a shortened survival time over three serial transplantations (not shown).

In summary, our data suggest that both pseudodiploid and aneuploid tumour clones can establish tumours in mice; however, the aneuploid clones lead to more aggressive tumours. The aneuploid fraction appears to have a growth advantage over the pseudodiploid population, which is also reflected by the clonal outgrowth and apparent selection over time. The present in vivo GBM model based on serial transplantation of biopsy spheroids could thus be considered as a proxy for the dynamic behaviour of distinct individual clones in a patient tumour.

In GBMs several markers have been proposed to enrich for putative glioma CSCs including CD133, CD15, CD44 and A2B5 [42, 49, 51]. However, CSC discrimination approaches largely rely on cell membrane phenotyping without taking into account the genetic characteristics of the tumour cell subpopulation. Based on the differential tumourigenicity of pseudodiploid and aneuploid GBM clones, we asked whether the stem-like associated marker expression was specific to the aneuploid cell population. We, therefore, combined cell membrane phenotyping with Hoechst-based ploidy measurement in our GBM xenografts. As most of the putative glioma CSC associated markers (e.g. CD133, CD15, A2B5 and CD44) are not unique to tumour cells but are also expressed by stromal cells [19], we used spheroid-based xenografts in eGFP⁺ NOD/Scid mice for a distinct analysis of pure tumour cell populations. In agreement with earlier studies, different GBMs displayed a strong heterogeneity in the number of marker-positive cells, e.g. CD15 or A2B5 positive cells ranged from 0 to 100 % in different GBM samples (Fig. 5a). Additionally, some biopsies displayed a heterogeneous staining intensity with the presence of low and high expressing cells. This was equally true for monogenomic (e.g. CD15 expression in T185 GBM, Fig. 5a) and polygenomic tumours (Fig. 5d, bulk tumour represented by ‘black’ histogram). To gauge the influence of genetic heterogeneity on CSC-associated marker expression profiles, we closely analysed two polygenomic GBMs (T16 and T238) containing distinct genetic clones with significant amounts of diploid (2N, blue) and aneuploid (AN, red) tumour cells (Fig. 5b). Several cell membrane markers such as CD56 (Fig. 5c), CD90 and CD29 (not shown) were found to be expressed at similar levels in diploid and aneuploid clones. Interestingly, a subset of cell membrane markers was differentially expressed between clones in one or both cases contributing to a heterogeneous intra-tumoural expression profile in the tumour bulk (Fig. 5d, ‘black’). For example in the T16 pseudodiploid fraction, A2B5 was strongly enriched in the diploid fraction (21 %) compared with the aneuploid population (2.8 %), whereas CD133 was enriched in the aneuploid fraction (47 versus 12.5 % in the diploid fraction) (Fig. 5d). However, the inverse was true for CD133 expression in the T238 tumour (16.5 % positive cells in diploid versus 10 % in aneuploid fraction), indicating that the expression of a particular marker could not be directly correlated with cellular ploidy. Similarly, CD15⁺ cells were present almost exclusively in T16 diploid (2.5 versus 0.3 %) and T238 aneuploid clones (10 versus 1.2 %) (Fig. 5d). Other markers such as CD44 (Fig. 5d) and CD184 (not shown) varied in distinct clones only in one of the patient biopsies. While the two distinct clones were present in all marker-expressing populations analysed, the proportion of each of the clones compared with the respective tumour bulk significantly changed for each marker (Fig. 5e). This indicates that the isolation of tumour cells based on CSC associated marker expression leads to genetically heterogeneous subpopulations with different tumourigenic potential.

Taken together, we show that distinct ploidy-based tumour clones in polygenomic GBMs are heterogeneous at the phenotypic level and that CSC identification based on marker expression may be biased due to the changing ratio of genetically distinct clones. We, therefore, suggest that functional assays on tumour cell subpopulations separated by marker expression should consider the genetic landscape of the cells.