Human astrocytomas of WHO grade I-IV [6 WHO I, 5 WHO II, 5 WHO III, 21 WHO IV (not matched, 14 primary and 7 recurrent)] and matched primary and recurrent glioblastomas (14 different pairs, WHO IV) were surgically dissected tissues from the Department of Neurosurgery (Kiel, Germany). All samples were obtained in accordance with the approved ethical standards of the ethics committee of the University of Kiel and the Helsinki Declaration of 1975 (D 536/15) and after informed consent of the donors. The diagnosis was established by a neuropathologist. Commercial human glioma cell lines were obtained from Deutsches Krebsforschungszentrum (Heidelberg, Germany; U118 and U343) or from the American Type Culture Collection (ATCC; Manassas, VA, USA; A172, T98G and U373). Long-term human primary cultures (A764 and A767) we obtained by dissociation of glioblastoma tissue and repetitive subcultivations, whereas short-term human primary cultures (14/07, 25/07, 76/12 and 96/14) were only used for up to three passages after preparation from tumor tissues. When an adequate amount of tumor tissue was available, matched samples of solid/cultured primary human glioblastomas were used for experiments. All cell lines were cultured in Dulbecco's modified Eagle's medium (DMEM; Invitrogen, Carlsbad, CA, USA) supplemented with 10% fetal calf serum (FCS). Cells were routinely checked for mycoplasma contamination by PCR.

RNA was isolated using TRIzol reagent (Invitrogen) following the manufacturer's instructions for tissue or cell cultures. DNA was digested by RNase-Free DNase (Fermentas, St. Leon-Rot, Germany; Promega, Madison, WI, USA) and cDNA was synthesized by reverse transcription (Fermentas). Real-time PCR was performed using TaqMan Master Mix and Primer Probes (Applied Biosystems, Foster City, CA, USA): hGAPDH (Hs99999905_m1), hGFAP (Hs00157674_m1), hOCT4 (Hs00999632_g1), hNanog (Hs02387400_g1), hMusashi-1 (Hs00159291_m1), hSOX2 (Hs00602736_s1), hKLF4 (HS00358836_m1) and hMYC (Hs00153408_m1). Cycle of threshold values (C_T) were measured by an ABI 7500 Fast cycler (Applied Biosystems) or the MyiQ™ Single Color Real-Time PCR detection system (Bio-Rad Laboratories, Munich, Germany). All genes of interest were normalized to the house-keeping gene glyceraldehyde-3-phosphate dehydrogenase (GAPDH) yielding ΔC_T values: C_T (gene of interest) - C_T (GAPDH). As ΔC_T values mirror the exponential course of the PCR, a ΔC_T of 3.33 corresponds to a 10-fold lower expression compared to GAPDH, a ΔC_T value of 6.67 corresponds to a 100-fold lower expression and so on. For statistical analysis, the relative gene expression compared to GAPDH (2^−ΔCT) was employed. To include undetectable samples in mean expressions, these sample values were defined as 21 (= ΔC_T that would be obtained after the maximum cycle number). The expression differences between primary and recurrent glioblastomas were calculated as n-fold ΔΔC_T values = 2^{−ΔCT primary - DCT recurrent} and displayed in a grey shade encoded manner. A relative expression value of 1 (equal expression in primary and recurrent tumor) was assigned to 30% grey. Expression values >1 (higher expression in recurrent than primary tumor) were assigned with increasing grey shading, reaching black at 3-fold (or higher) induction. Expression values <1 (indicating lower expression in recurrent than primary glioblastomas) were assigned with lighter grey shadings, increasing to white shading for values reaching 0. The induction of gene expression upon stimulation is displayed as relative gene expression, n-fold expression changes were calculated as ΔΔCT values = 2^{−ΔCT control - ΔCT stimulus}.

For immunohistochemistry, 10-μm cryo-sections were fixed with an ice-cold acetone-methanol mixture (1:1) for 10 min, blocked for autofluorescence with sudan black (1% in 70% ethanol) and for unspecific antibody-binding with 0.5% glycine/0.5% bovine serum albumin. Primary antibodies were applied overnight at 4°C, secondary antibodies were incubated at 37°C for 1 h, and nuclei were counterstained with DAPI (Sigma-Aldrich, Hamburg, Germany). Slides were embedded with Thermo Scientific™ Shandon™ Immu-Mount™ (Thermo Shandon, Inc., Pittsburgh, PA, USA). Between each step, slides were washed 3× for 5 min with Tris-buffered saline with 0.1% Tween (TBS-T). Primary antibodies were anti-GFAP (mouse, 1:100; Dako, Glostrup, Denmark), anti-MIB-1 (Ki-67 clone, mouse, 1:100; Dako), anti-OCT4 (rabbit, 1:200; Cell Signaling Technology, Danvers, MA, USA), anti-SOX2 (rabbit, 1:100; Santa Cruz Biotechnology, Santa Cruz, CA, USA), anti-Musashi-1 (1:100, mouse; R&D Systems, Minneapolis, MN, USA), anti-Nanog (1:500, rabbit; Thermo Fisher Scientific, Rockford, IL, USA) and anti-KLF4 (1:250, mouse; Pierce/Thermo Fisher Scientific). When antibodies were derived from different species, they were applied simultaneously and for the antibody controls, both primary antibodies were omitted. If the primary antibodies were derived from the same species, first one primary antibody and its secondary antibody were stained, then unspecific binding was blocked by F(ab) fragments derived from the very same species [donkey anti-mouse and anti-rabbit F(ab) fragments, 1:100, from Dianova, Hamburg, Germany] and the second primary and secondary antibodies were applied. In this case, antibody controls were performed by omitting both primary antibodies or, in order to control F(ab) blocking efficiency, one of them, respectively. Secondary antibodies were donkey anti-mouse or anti-rabbit IgGs labelled with Alexa Fluor 488 or Alexa Fluor 555 (1:800-1:1000; Invitrogen).

For stimulation and transfection experiments, T98G and A172 glioblastoma cells were seeded subconfluently on 6- or 12-well plates or 75-mm² culture flasks and grown overnight. Media were changed to DMEM + 10% FCS supplemented with temozolomide or camptothecin (both Sigma-Aldrich) or corresponding concentrations of DMSO as solvent control, cells were incubated for the indicated times and harvested for RNA isolation. For transfection, media were changed to Opti-MEM (Invitrogen), and transfection was performed using Lipofectamine (Invitrogen). The vectors pcDNA3.3eGFP (mock, Addgene plasmid 26822), pcDNA3.3_KLF4 (Addgene plasmid 26815), pcDNA3.3_OCT4 (Addgene plasmid 26816) and pcDNA3.3_SOX2 (Addgene plasmid 26817) were generated by Derrick Rossi (Harvard University Department of Stem Cell and Regenerative Biology, Boston, MA, USA) (29), and kindly provided by deposition at Addgene, Inc., (Cambridge, MA, USA). Media were changed to DMEM + 10% FCS 6 h after transfection, and cells were harvested for RNA isolation after the indicated times.

To test significance of expression increase or decrease the Student's t-test was performed. For matched samples (primary-recurrence from the same patient; samples from the same experiment) a paired test was used. Astrocytoma samples of different grades from different patients were analyzed by an unpaired test. For the calculation of mean expression values of tumor samples, undetectable samples were set ΔC_T = 21, as this would commonly represent the detection limit if GAPDH is used as house-keeping gene and 40 amplification cycles are run. For Pearson correlation analysis, expression levels of all investigated stem cell markers of the different astrocytoma malignancy grades, a cohort of mixed primary and recurrent glioblastomas (not matched) and the respective groups of primary and recurrent glioblastomas of matched patient samples were correlated using GraphPad Prism, P-values indicate as follows: ^*p<0.05, ^**p<0.01, ^***p<0.001 and ^****p<0.0001.

We first investigated the expression of the stem cell factors in astrocytomas of WHO grade I–IV, with IV representing the glioblastoma multiforme (GBM), and matched primary and recurrent glioblastomas (Fig. 1). However, as increasing expression levels of SOX2 in astrocytomas with higher malignancy have already been described earlier by us and others (7,30), we restricted our analysis in astrocytomas to OCT4, Nanog, KLF4 and MYC. These were detectable on moderate to high levels in astrocytomas of all malignancy grades. However, for Nanog and KLF4, we found also a lack of expression in distinct samples. Comparable results were obtained for the transcription of Nanog, KLF4 and MYC in matched primary and recurrent glioblastoma samples. In these samples, we also investigated the transcription of SOX2, which was considerably lower, and could not be detected in 9/28 primary/recurrent glioblastoma samples. For OCT4 and MYC we observed a significant increase in mRNA expression in highly malignant astrocytomas (grade III and IV for OCT4 and grade IV for MYC) in comparison to the benign grade I astrocytomas. A trend to higher transcription levels in high grade astrocytomas was also visible for KLF4. However, due to high variances within the groups, these results were not statistically robust. In contrast, Nanog mRNA expression was hardly altered in astrocytomas of different grades in our sample cohort. In matched samples of primary and recurrent glioblastomas mean expression levels were not altered during progression from primary to recurrent glioblastomas except for MYC, which was found slightly decreased in recurrent samples. However, regarding the regulation profile of each patient's primary-recurrent pair, we found distinct patterns of upregulation and downregulation of stem cell markers (Table I, n-fold expression changes between individual primary-recurrent pairs, which are numbered consecutively). Based on these observations, we distinguished one group of primary/recurrent pairs with one or more markers being clearly upregulated (>2-fold; Table I, upper part, pairs 1–6) from a second group showing no clear upregulation, or even downregulation of one or more stem cell markers in the recurrent tumor (Table I, lower part, pairs 7–14).

Interestingly, when analyzing a possible correlation of mRNA expression of individual investigated stem cell markers in both astrocytomas of WHO grade I–IV and matched primary and recurrent glioblastomas (Table II), significant positive correlations were restricted to individual marker combinations for most grades and the primary glioblastomas of the primary/recurrent pairs. In detail, astrocytomas grade I showed significant correlations between MYC and Nanog as well as MYC and KLF4 expression. In astrocytomas grade II we also observed a significant positive correlation between MYC and Nanog, whereas in astrocytomas grade III there was no correlation between the markers. In astrocytomas grade IV (glioblastomas) a significant positive correlation between KLF4 and MYC expression was detectable (Table IIA). Thus, we could not observe a specific correlation pattern consistent to all astrocytoma malignancy grades. However, one has to keep in mind that the sample groups of astrocytomas grade I–III are quite small due to the limited availability of tumor tissue. Our further analysis of the cohort of matched primary/recurrent glioblastoma pairs revealed in the primary samples only a significant correlation between OCT4 and KLF4 expression, but remarkably, in recurrences several stem cell markers showed statistically significant correlations between their mRNA expression levels (e.g. OCT4-MYC, Nanog-MYC and Nanog-OCT4; Table IIB).

Taken together, embryonal and neural stem cell markers are expressed on mRNA level in astrocytomas of different WHO grades, and primary and recurrent glioblastomas. While OCT4 and MYC expression was elevated with increasing malignancy of astrocytomas, MYC expression was slightly reduced in recurrent tumors compared to the corresponding primary tumor. In addition, a correlation analysis of individual stem cell marker mRNA expression revealed sporadic significant positive correlations between Nanog-MYC or KLF4-MYC expression in glioma progression. Remarkably, significant positive correlations between Nanog or OCT4 with other embryonic stem cell factors were frequently found in glioma recurrence, potentially identifying these markers as highly important in glioma progression and recurrence. Apart from this, the regulation pattern of the matched patient samples revealed that upregulation or downregulation of stem cell markers in the progress of primary to recurrent glioblastoma occurs in an individual, probably patient-specific manner.

To investigate the localization and regional distribution and their potential alteration in glioblastoma progression, we prepared fluorescence stainings of OCT4, SOX2, KLF4 and Nanog using 3 pairs of matched primary and recurrent glioblastoma samples. The protein expression of MYC has carefully been described elsewhere to be evenly throughout the tumor (31). To identify tumor cells in more differentiated tumor regions, we co-stained for GFAP (glial fibrillary acidic protein). As exemplarily shown in Fig. 2, all investigated markers could be detected on protein levels in glioblastoma cells. Remarkably, only a small subpopulation of tumor cells showed positive staining for the respective stem cell markers, and these cells were commonly found in small groups in GFAP positive tumor regions. To characterize the proliferative status of glioma cells expressing embryonic stem cell markers, we performed fluorescence double-stainings of OCT4, KLF4 and Nanog with MIB-1/Ki-67, which is a nuclear marker for all active cell cycle phases. Cells with positive staining for OCT4, Nanog or KLF4 did hardly show any MIB-1-signal in the nucleus. Very rare examples of double positive cells are shown in Fig. 3. However, the vast majority of stem cell marker positive cells did not show staining of the proliferation marker MIB-1 (compare also Table III).

To comprehensively analyze the co-expression patterns of embryonic stem cell markers and also compare it to known neural/glioma stem cell markers, we performed fluorescence double-stainings of combinations of the markers OCT4, KLF4 and Nanog with SOX2 and Musashi-1, which are well documented neural/glioma stem cell markers. Representative examples are shown in Fig. 4, and results are summarized in Table III, which also includes the co-staining results with GFAP and MIB-1 in a semi-quantitative evaluation of 4 different glioblastoma samples, respecting the intra- and inter-individual variances. In fact, OCT4, Nanog and KLF4 were partly co-expressed with the other stem cell markers (OCT4, Nanog, KLF4, SOX2 and Musashi-1), but were also found to be expressed in single positive cells. Remarkably, most stem cell marker-expressing cells were also SOX2-positive, but KLF4-negative. When comparing these data to mRNA correlation analysis of investigated stem cell markers (Table IIB), it became apparent that Nanog and OCT4, which showed a positive mRNA correlation to KLF4, respectively, were also frequently co-expressed by the same cell types. Additionally, the missing correlation between KLF4 and SOX2 mRNA expression was underlined by an infrequent KLF4-SOX2 co-staining. Contrasting these findings, Nanog and OCT4, which positively correlated in mRNA expression, were rarely co-expressed. Additionally, a frequently observed co-staining of OCT4 and Nanog with SOX2 was not reflected in a positive correlation of the mRNA expression. Summarized, a heterogeneous picture results which leads to the statement that despite the evidence of a specific role of individual stem cell markers in glioma progression and recurrence, a ‘master marker’ which is common to all glioblastoma cells with stem cell characteristics, could not be identified.

For further in vitro investigations, we analyzed the mRNA expression of the stem cell markers OCT4, SOX2, KLF4, MYC, Nanog and Musashi-1 in five commercially available glioma cell lines (A172, U118, U343, U373 and T98G) and two long-term cultures (A764 and A767) as well as four short-term human glioblastoma cultures (14/07, 25/07, 76/12 and 96/14). We found these markers to be expressed on moderate levels in nearly all investigated cell lines as well as in long- and short-term cultures, apart from T98G and 14/07 that rarely expressed Nanog (T98G and 14/07) and SOX2 (T98G), and also comparably low levels of Musashi-1 (T98G; Fig. 5A and B). Interestingly, mRNA expression levels of different stem cell markers showed a surprising range of variation within commercial cell lines and short- and long-term cultures, but within this range cell lines were comparable to long- and short-term cultures. Thus, we chose established glioma cell lines, which are available unlimitedly and therefore solve the problem of limited access to tumor tissue, for further experiments. Experiments mainly focused on the ability to regulate stem cell factors upon extrinsic chemotherapeutic stress and upon intrinsic overexpression of respective stem cell transcription factors.

To investigate the influence of chemotherapeutics on the expression levels of stem cell markers, we chose A172 and T98G as exemplary glioma cell lines because A172 showed moderate to high expression levels of stem cell markers, and T98G showed low expression levels for Nanog and SOX2 (comparable to low expression levels of the short-term cultures 14/07 and 76/12). These cell lines were stimulated for 24 or 48 h with 50 μg/ml camptothecin, or 400 μg/ml temozolomide, the most commonly used chemotherapeutic in glioblastoma treatment; corresponding DMSO concentrations were used for stimulation of solvent controls and used to determine expression changes. For MYC, SOX2 and Musashi-1 we did not observe any significant regulation of the mRNA expression level (data not shown). In contrast, in T98G cells, the expression level of OCT4 was significantly elevated 4.2-fold (±4.3 SD) after 48 h (Fig. 5C). This effect was even more pronounced upon stimulation with camptothecin, but due to high variations not significant. In A172 cells, OCT4 mRNA expression was regulated neither by temozolomide nor by camptothecin. Nanog, which was hardly detectable in unstimulated T98G cells, was found to be increased. However, the proximity to the detection limit has to be kept in mind, thus the data are transparently shaded. In temozolomide-treated A172 cells (48 h), Nanog was significantly but only slightly increased (2.6-fold±0.7 SD), whereas camptothecin yielded no significant effect. Contrasting these slight or little robust expressional changes of OCT4 and Nanog, KLF4 was significantly induced by both temozolomide and camptothecin in T98G and A172 cells. Showing a significant induction after 24 h already, temozolomide yielded a 17.1-fold (±14.0 SD), and camptothecin a 40.9-fold (±37.8 SD) upregulation after 48 h. In A172 cells, an upregulation to 17.7-fold (±24.4 SD) by temozolomide, and to 13.1 (±13.2 SD) by camptothecin was induced after 48 h.

Summarized, we observed an induction of the embryonic stem cell markers Nanog and KLF4 (and partly OCT4) in glioma cell lines upon chemotherapeutic stress. Surprisingly, this effect is more pronounced in T98G cells, where baseline expression of stem cell markers is comparably low. Among the investigated stem cell markers, KLF4 is most robustly regulated in both glioma cell lines upon extrinsic cytotoxic challenge with camptothecin or temozolomide.

Having observed this limited but specific plasticity regarding stem cell marker regulation upon extrinsic challenge, we investigated, if stem cell marker expression might be regulated by intrinsic alteration of stem cell transcription factors SOX2, KLF4 and OCT4. Thus, we performed overexpression experiments with stem cell marker low expressing T98G cells using expression vectors for the respective transcription factors (Fig. 6). Successful transfection was proven either by quantitative RT-PCR (OCT4, 43.6-fold induction ±34.2 SD; KLF4, 50.4-fold induction ±36.8 SD) or in case of SOX2 by immunocytochemistry (insert in Fig. 6).

The overexpression of these transcription factors did not yield a significant regulation (induction or reduction) of other stem cell markers investigated (SOX2, KLF4, OCT4, MYC, Nanog and Musashi-1), apart from Musashi-1, which was only very slightly induced (1.8-fold±0.7 SD) upon KLF4 overexpression.

Taken together, the embryonic stem cell transcription factors SOX2, MYC, OCT4, KLF4 and Nanog are expressed in astrocytomas of different malignancy grades as well as in primary and recurrent glioblastomas in a very distinct, probably patient specific expression pattern. Positive correlations between mRNA expression levels of stem cell markers were more pronounced in recurrent glioblastomas, while in situ co-expression occurs in a complex manner, which did not necessarily match the mRNA correlation results. In vitro, stem cell markers were expressed in established cell lines as well as in glioblastoma short- and long-term cultures. The factors were partly regulated by extrinsic chemotherapeutic challenge but not by intrinsic overexpression of the respective transcription factors.