Abstract

Background

   The metabolic enzyme carbonic anhydrase 12 (CA12/CAXII) emerges as a
   promising cancer therapeutic target with drug development projects
   underway. Previous reports proposed the relevance of CA12 in the
   context of glioma but are limited in patient data quantity, ignore
   ethnic diversity of patients or rely on semi-quantitative, thereby out
   of date, methodology. Moreover, little is known on the association of
   CA12 to brain tumor stemness or on the effect of anti-CAXII-directed
   monotherapies on glioma stem cells (GSCs), in particular their response
   regarding mesenchymal differentiation status.

Methods

   We performed in silico analysis on three independent, large-scale
   patient datasets interrogating state of the art molecular diagnostics
   alongside clinical outcomes. We analyzed CAXII abundance on a
   collection of GSCs and functionally tested their response to exposure
   to CAXII blocking antibody 6A10.

Results

   CA12 is highly expressed in glial tumors compared with normal tissue
   and predicts for poor clinical course of tumor patients. CA12
   expression in glioblastoma significantly correlates with clinically
   established, molecular markers of IDH1^WT DNA, WHO grade IV or absence
   of 1p/19q chromosome arm co-deletion. Furthermore, tumors with elevated
   CA12 cluster into the mesenchymal transcription subclass of the
   disease. CAXII abundance in different GSCs ranges from almost absent to
   high levels and does not correlate to stem cell marker CD133/AC133 cell
   surface expression. Moreover, aiming to pharmacologically block CAXII
   in our cells with antibody 6A10 caused significant functional response
   only in one of the tested GSCs models, featuring suppression of cell
   invasion accompanied by reduction of ZEB1 protein and other stem cell
   markers.

Conclusion

   CA12 represents a clinically relevant and molecular brain tumor-subtype
   specific therapeutic target. Our correlative data from experimental and
   clinical samples does not support CA12/CAXII to be GSC specific. 6A10
   possesses promising potential to impede the invasive capacity of glioma
   cells and supports the emerging concept that CAXII interacts with
   cancer EMT programs. However, further mechanistic studies are required
   to comprehensively assess the therapeutic potential of 6A10 and to
   identify different resistance mechanisms of GSCs.

   Keywords: carbonic anhydrase 12, 6A10, glioma, therapeutic antibody,
   ZEB1, disease modeling

Introduction

   Gliomas, amongst which glioblastoma (GBM) is the most aggressive form,
   represent a clinically challenging class of cancers. Their high
   malignancy is characterized by elevated levels of therapy resistance
   and invasive growth. It is believed that malignant features of GBM are
   promoted by cells with stem cell properties.[44]^1 Recent evidence
   proposes that GBM stem cells (GSCs) are tightly associated with the
   molecular reconversion into mesenchymal state, and
   epithelial-to-mesenchymal-like processes (EMT) are intertwined with GBM
   stemness in order to augment various malignant properties of the
   disease.[45]^2 Over recent years, the prominent transcriptional EMT
   activator ZEB1 (Zinc finger E-box-binding homeobox 1) has emerged as a
   potent regulator of GSCs.[46]^3–7 A pharmacological strategy to block
   the GSCs ZEB1 program is of high clinical need.

   Carbonic anhydrases (CA), a family of isoforms of hypoxia-responsive
   cell metabolic enzymes, are discussed as potent anti-cancer targets in
   a wide variety of cancers,[47]^8 including GBM.[48]^9–13 Previous work
   indicates CA isoform 12 (CA12) may be of particular relevance. However,
   until now our knowledge has been restricted to a low number of analyzed
   patient samples, ignoring ethnic diversity of patients and exclusively
   based on semi-quantitative, thereby out of date, methodology to score
   CA12 activation.[49]^11^,[50]^12 The presented results on over 1700
   patient samples using various state of the art molecular
   diagnostics[51]^14–18 closes all these gaps of current knowledge in
   order to benchmark the clinical relevance of CA12 in the current
   scientific neuro-oncological environment. We also compare the clinical
   prognostic value of CA12 mRNA expression against counterparts of other
   CA isoforms. Overall, our data strongly position CA12 as a molecular
   target with significant clinical negative prognostic value of glioma
   patients, and with significant association with various WHO diagnostic
   guideline-accepted, molecular properties of GBM. Aiming to validate our
   computational-build hypothesis with functional data, we performed
   applied stem cell models of GBM, in order to recapitulate the most
   aggressive subtype of cells of the disease. In total, correlative data
   on clinical and experimental biosamples do not support CA12 to be
   highly stem cell specific nor predict for response to
   anti-CAXII-directed therapeutic antibody 6A10. Interestingly in a
   fraction of tested GSCs, 6A10 treatment caused significant suppression
   of ZEB1 associated with impaired cellular invasion. Since clinically
   feasible strategies to directly inhibit transcription factor activities
   are challenging, our data promote subsequent studies with 6A10 as a
   potential drug candidate. Resistance mechanisms and biomarkers
   predicting response need to be identified to facilitate 6A10’s clinical
   application.

Materials and Methods

Patient Data Resources and in silico Analysis

   Molecular and clinical data of glioma patients were obtained from the
   publicly assessable sources, namely The Chinese Glioma Genome Atlas
   (CGGA, ([52]https://www.cgga.org.cn, two independently acquired
   datasets with n=325 individuals, and a 2019 dataset with n=693
   individuals) and The Cancer Genome Atlas (TCGA, with n=702 individuals,
   for verification). For CGGA, molecular testing of each patient was
   performed at the Molecular Pathology Testing Center of Beijing
   Neurosurgical Institute and the data acquisition was approved by the
   Beijing Tiantan Hospital institutional review board and tumor specimen
   quality control.

   The statistical computations and figure drawing were performed with R
   package ‘ggplot2ʹ. Unsupervised cluster analysis was performed by
   pheatmap package in software environment R (version 4.0.0). The
   analysis processes were carried out under the default parameters.
   Student’s t test was used in the differential analysis of the two
   groups in this study. The drawing was performed by ggplot2 package in
   software environment R (version 4.0.0). The differential analysis of CA
   families between normal brain tissue and brain tumor tissue were
   performed in the GEPIA portal ([53]http://gepia.cancer-pku.cn). The
   most significantly related genes of CA12 were obtained by Pearson
   correlation analysis. Genes with a correlation coefficient greater than
   4.5 were used for enrichment analysis. Biological function and pathway
   enrichment analysis were performed by analysis tools in The Database
   for Annotation, Visualization and Integrated Discovery (DAVID, v6.8)
   ([54]https://david.ncifcrf.gov). The Pearson correlation analysis
   method was used in the correlation analysis of this study. Glioma stem
   cell markers were obtained from the Cell Marker portal
   ([55]http://biocc.hrbmu.edu.cn/CellMarker/). Kaplan–Meier survival
   analysis was used in the survival analysis of this study. Drawing was
   performed using ggplot2 and survival packages in software environment R
   (version 4.0.0). The analysis processes were carried out under the
   default parameters.

Cell Culture and Pharmacological Model

   The in vitro cell models 407 and BTSC233 were kindly provided by M.S.
   Carro, Freiburg University, Germany; GBM1 by A. Vescovi, San Raffaele
   Hospital, Milano, Italy; NCH421k and NCh644 by G. Herold-Mende,
   Heidelberg, Germany; JHH520 by G. Riggins and SF188 by E.H. Raabe (both
   Johns Hopkins Hospital, Baltimore, USA). GSC models were grown under
   neurosphere conditions as described previously.[56]^20 6A10 was
   provided as previously described.[57]^21 Determination of the molecular
   concentration for testing the effect of the substance (25 μg/mL) was
   based on the wide experience with 6A10 from co-author R. Zeidler,
   Munich. Moreover, our experimental design was guided by published data
   of 6A10 impacting cancer cells in vitro using the same
   concentration.[58]^22 Duration for acquiring cell invasion potential
   was limited to time span of overnight to a maximum of 24 hours. For all
   assays, volume-adjusted medium was used as mock treatment. Ethical
   approval to conduct the in vitro studies with the used cell lines was
   granted by the ethics commission of the Medical Faculty of the
   Heinrich-Heine University (study ID 5841R).

Functional in vitro Assays

   Cell growth quantification was executed using MTT assay similarly as
   described before.[59]^7 The following specific parameters were applied:
   assay format: 96 well, 4000 cells each well/100 μL culture media.
   Quantification of cell proliferation, apoptosis and cell cycle
   progression was performed using Muse^® Cell Analyzer (Merck KGaA,
   Germany) instrumentation using relevance kits and following
   manufacturer protocols (Luminex, United States: Muse^® Ki67
   Proliferation Kit, Muse^® Annexin V & Dead Cell Kit, Muse^® Cell Cycle
   Kit). Invasion assay was performed using modified Boyden chamber assay
   similarly as described before.[60]^3 In brief, Matrigel was thawed at 4
   °C overnight. Inserts were incubated with 1 mL Matrigel^® at 37 °C for
   an hour. Harvested cells were adjusted to concentration of 3 x 10^5 in
   - FCS solution. The upper chamber was filled with 500 µL harvested
   cells, while 800 µL + FCS solution was added to the lower chamber.
   Non-invading cells that remained in the upper chamber were removed by
   cotton swab. Invasive cells in the lower chamber were fixed with 500 µL
   precooled methanol. After one wash, cell nuclei were stained with
   hematoxylin and washed with warm water. N=5 randomly selected areas in
   each insert were photographed under inverted light microscope (40×
   objective) and further quantified in Image J by manual counting of
   cells. For the in vitro assays, acquired data were presented as average
   and SEM. Welch’s t-test was used to check for significant differences
   (p≤0.05), with p-values showing to be significant (p<0.05, *), very
   significant (p<0.01, **) and highly significant (p<0.001, ***).

qPCR

   Targeted gene expression quantification was performed according to
   established procedures in the lab using SYBR green-based nucleic acid
   intercalation assay, similar to those described previously.[61]^23
   [62]Suppl. Figure S1 lists all primer specifications relevant for this
   study. Acquired data were presented as average and SEM. Welch’s t-test
   was used to check significant difference (p≤0.05), with p-values
   showing to be significant (p<0.05), very significant (p<0.01) and
   highly significant (p<0.001).

Protein Quantification Assays

Western Blot

   Standard procedures similar to those described previously were
   applied.[63]^24 Non-specific sites were blocked by 5% milk powder
   solution followed by primary antibody incubation at 4 °C overnight.
   Fluorescent secondary antibody was diluted in TBS-T and incubated at
   room temperature for an hour. Detected signals was acquired with
   Odyssey^® CLx (LI-COR Biosciences, Germany). Further data analysis was
   processed in Image Studio Ver2.1. [64]Suppl. Figure S2 lists all
   antibody specifications relevant for this study.

FACS Analysis

   In this flow cytometry experiment, cells were prepared according to the
   direct flow cytometry protocol from Abcam. In brief, 5–10 × 10^5 cells
   were collected in a Falcon round bottom polystyrene test tube. After
   trypsinization into single cells, media was replaced by ice cold 10%
   FBS in PBS. Afterwards, cells were incubated with fluorescent-labelled
   antibodies on ice for 30 min ([65]Suppl. Figure S2). Since there is
   overlap between two fluorescent emission spectra, single-stained
   samples were applied for correction as compensation control. Cells were
   washed with ice cold PBS 3x and re-suspended in 500 µL 10% FBS in PBS.
   Samples were analyzed within one hour on BD FACSCanto^®. Bivariate
   histograms and statistical plots were generated by the BD FACSDiva^®
   Software at the core facility setup.

Results

CA12 Transcription Level is Increased in Gliomas Compared with Non-Tumor
Cells, Predict Poor Clinical Prognosis of Glioma Patients

   We present the landscape of transcription of CA families and their
   relationship between diverse diagnostically relevant,
   clinicopathological factors in the CGGA Database and CGGA 2019 Database
   ([66]Figure 1A and [67]B). Amongst all CA isoforms, expression levels
   of CA12 and CA13 are the only family members that predict for
   significant poor prognosis in terms of overall patient survival time in
   both CGGA datasets ([68]Figure 1C). Moreover, expression levels of
   those negative predictors are significantly increased in tumors
   compared with non-tumor tissue, both in low grade (LGG) and high grade
   gliomas (HGG) respectively ([69]Figure 1D). Poor prognostic value of
   high CA12 mRNA levels was confirmed in additional datasets of
   Gravendeel et al.[70]^25 and Kawaguchi et al.[71]^26 ([72]Suppl. Figure
   S3). To the best of our knowledge, this is the ethnically most diverse
   and largest number of cancer patient samples probed for clinical
   prognostic values of CA12. Given these strong indications, next we
   interrogated the data of state of the art molecular characterization of
   brain tumors to dissect any relevance of CA12 as a potential
   molecular-subtype specific therapeutic target.

Figure 1.

   [73]Figure 1
   [74]Open in a new tab

   CA12, amongst all CA family members, represents the most promising
   therapeutic target in glioma. The landscape of CA families and their
   relationship between diverse diagnostic relevant clinicopathological
   factors in (A) CGGA database and (B) CGGA 2019 database. (C) CA10, CA12
   and CA13 are independent prognostic factors in both CGGA and CGGA 2019
   databases. Red: poor prognostic factors. Blue: good prognostic factors.
   (D) Comparison of the expression differences of CA10, CA12 and CA13 in
   normal brain tissue and brain tumor tissue. Green: significantly high
   expression in normal tissues. Red: significantly high expression in
   tumor tissues. Black: no significant difference.

Elevated CA12 Transcriptional Levels in Glioma Correlate with Clinically
Established, Molecular Properties

   High mRNA levels of CA12 in glioma significantly correlate with
   clinically relevant, negative prognostic pathological and molecular
   markers of brain tumors. Those are tumors defined as WHO grade IV types
   ([75]Figure 2A and [76]F), tumors with IDH1 wildtype DNA according to
   WHO classification of 2016[77]^27 ([78]Figure 2B and [79]G), and tumors
   without 1p19q chromosome co-deletion[80]^28 ([81]Figure 2C and [82]H).
   Of note, in one of CGGA datasets, high CA12 expression was found in
   tumors with methylated DNA promoter of O(6)-methylguanine-DNA
   methyltransferase (MGMT)[83]^29 ([84]Figure 2D and [85]I). In addition,
   the mesenchymal GBMs, known to be particularly aggressive,[86]^15
   present significantly higher expression levels of CA12 than other
   transcriptional subtypes in both Asian datasets ([87]Figure 2E and
   [88]J). We validated our findings in the TCGA dataset and found high
   CA12 expression significantly associated to gliomas with WHO grade IV
   staging, to tumors with IDH1^WT DNA, to tumors with 1p19q chromosome
   arm co-deletion as well as to cases classified as mesenchymal subtype
   of GBM ([89]Figure 3A–[90]F). In the TCGA dataset CA12 transcription is
   enriched in cases with unmethylated MGMT promoter DNA ([91]Figure 3E).

Figure 2.

   [92]Figure 2
   [93]Open in a new tab

   The relationship between CA12 and clinicopathological factors in CGGA
   and CGGA 2019 databases. High CA12 mRNA levels significantly associated
   with WHO IV grade (A and F), IDH1 wildtype DNA status (B and G) and
   tumors without 1p/19q chromosome arm co-deletion (C and H). CA12 mRNA
   levels are upregulated in tumors carrying methylated MGMT promoter
   region (significantly for CGGA) (D and I). CA12 mRNA is significantly
   upregulated in most aggressive mesenchymal subtype of GBM (E and J).

Figure 3.

   [94]Figure 3
   [95]Open in a new tab

   Validation of the identified association of CA12 transcription level
   with clinicopathological factors using TCGA database. The landscape of
   CA families and their relationship between diverse diagnostically
   relevant clinicopathological factors in TCGA Database (A). High CA12
   mRNA levels significantly associate with WHO IV grade (B), IDH1
   wildtype DNA status (C), tumors with 1p/19q chromosome arm co-deletion
   (D) and CA12 tumors carrying de-methylated MGMT promoter region (E).
   CA12 mRNA is significantly upregulated in the most aggressive
   mesenchymal subtype of GBM (F).

Correlation of CA12/CAXII with Stem Cell Markers

   To look for an association with malignancy promoting tumor stem cell
   signals, we performed targeted correlative assays in clinical and
   experimental biosamples. This rationale was further supported by the
   work of others in the context of lung cancer, where CAXII was described
   to serve as a marker for tumor stem cells.[96]^21 Our analysis does not
   unequivocally support this hypothesis to apply in the context of brain
   tumors. First, in silico analysis testing for correlative expression of
   CA12 with mRNA levels of previously reported stem cell molecules shows
   a mixed picture. Although prominent markers such as heme oxygenase
   1,[97]^30 membrane transporter SLC16A1[98]^31 or Nestin[99]^32 show
   positive correlation in all datasets, negative correlation between CA12
   and likewise prominent stem cell markers SOX2[100]^33 or ZEB1[101]^4
   exists ([102]Figure 4A–[103]C). Moreover, high CA12 expression revealed
   association to expression patterns of biological processes and
   molecular signaling pathways (using Kyoto Encyclopedia of genes/KEGG
   consensus items as determinates) both associated to cancer stem cells
   and to those previously not dominantly associated to tumor stemness
   ([104]Suppl. Figure S4).

Figure 4.

   [105]Figure 4
   [106]Open in a new tab

   The relationship between CA12 and glioma stem cell markers. CA12 is
   significantly positively correlated with the expression of most glioma
   stem cell markers in CGGA (A), CGGA 2019 (B) and TCGA (C) databases.
   Blue: significantly correlated. Black: non-significantly correlate.
   Cell surface abundancy of CA12 does not correlate with relative cell
   surface abundance of bona fide brain tumor stem cell marker CD133 in
   our in vitro disease models as expressed by percentage of positive
   cells and mean fold intensity/MFI, one time experiment (D).

   In addition, our protein analysis on GSC in vitro models does not
   support CA12 to be GSC specific. FACS data show that the amount of cell
   surface CAXII does not correlate with the prevalence of cells positive
   for bona fide GSC marker CD133/AC133[107]^34 ([108]Figure 4D). In
   addition, baseline abundance of total CAXII in our stem cell models
   varies from very low to high signals, as detected in Western blot
   ([109]Figure 5). Together, our clinical and experimental data do not
   strongly support that high CA12/CAXII levels correlate with brain tumor
   stem cell signatures.

Figure 5.

   [110]Figure 5
   [111]Open in a new tab

   Pharmacological targeting of CAXII in glioma in vitro functional
   responses. 6A10 treatment did cause mild inhibitory effect on cell
   growth in subset of tested GSC models/GBM1 and BTSC407, and no effect
   on BTSC233, unb = control (A–C). All cells are IDH1wt and MGMT promoter
   methylated, GBM1 and 407 are classical subtype of GBM, BTSC233 is
   mesenchymal subtype of GBM (data not shown). 6A10 treatment caused
   suppression of cellular invasion in responder cell line GBM1 but not in
   non-responder model 407 (D and E), BTSC233 not tested.

6A10 Treatment in vitro Can Cause Moderate Inhibition of Growth and
Significant Reduction of Invasion Capacity

   In our antibody treatment experiments, we observed a mild reduction of
   cell growth for GBM1 and 407 ([112]Figure 5A and [113]B), and no clear
   effect for BTSC233 ([114]Figure 5C). This was accompanied by a trend
   towards induction of apoptosis in GBM1, abut no changes in apoptosis
   rate were observed in the other two models used ([115]Suppl. Figure
   S5). None of the cell models showed a difference in cell proliferation
   or cell cycle analysis in response to the treatment ([116]Suppl. Figure
   S5). Most notably, responder model GBM1 reacted with significantly
   diminished capacity of cellular invasion ([117]Figure 5D), as 407 did
   not reduce its invasive properties ([118]Figure 5E).

6A10 Treatment Can Suppress the Activation of EMT Marker ZEB1 and Leads to
Induction of CA12 Expression and Protein

   Next, we tested the change of stem cell markers in response to 6A10
   exposure on mRNA level. Interestingly, GBM1 showed significantly
   suppressed expression of various stem cell markers, whereas no clear
   trend was seen for the other two tested cell models. All tested models
   showed signs of upregulation of CA12 expression upon antibody
   treatment, with GBM1 showing significant induction ([119]Figure 6A).

Figure 6.

   [120]Figure 6
   [121]Open in a new tab

   Pharmacological targeting of CAXII in glioma in vitro effects on stem
   cell markers, CAXII and ZEB1 protein 6A10 treatment on GBM1 caused
   significant reduction of stem cell molecules Nestin, CD133/PROM1, ZEB1
   mRNA, and trend of reduction of CD44 and SOX2. Moderate effects are
   observed on models 407 and BTSC233 (A). Induction of 6A10 target CA12
   on mRNA and protein was found in all models with significant
   upregulation on protein level in GBM1 (A and B). Reduction of stem cell
   marker ZEB1 in GBM1 upon 6A10 treatment was confirmed on protein level
   with no change detected in 407 and BTSC233 (B) (c – control/mock, T –
   treatment, 25µg/mL 6A10, for more details please see relevant section
   in Materials and Methods section).

   When testing the abundance of relevant factors on total protein level,
   to our surprise we identified CAXII to be virtually absent in naïve
   GBM1 and 407, and comparatively strongly active in BTSC233 ([122]Figure
   6B and [123]C). Moreover, it seems that the cell surface protein level
   of CAXII differs from total CAXII, as 407, but foremost GBM1, are high
   on cells positive for cell surface CAXII ([124]Figure 4D).
   Interestingly, for all models we identified stronger CAXII bands when
   exposed to 6A10, with significant induction in GBM1 ([125]Figure 6B).
   Naïve ZEB1 levels are roughly similar between the tested cell models.
   Interestingly, we see reduction of protein in response to treatment in
   some of the biological repetition experiments in any of the
   investigated cell lines, although significance over all repetitions was
   reached in GBM1 only ([126]Figure 6B and [127]C).

Discussion

   Our data show CA12 to serve as a clinically relevant anti-cancer
   therapeutic target and proposes its potential suitability for the
   development of neuro-oncology precision medicine. By using large-scale,
   molecular and clinical data retrieved from three different datasets,
   thereby ensuring ethnic and gender diversity, to the best of our
   knowledge this is the largest descriptive analysis of CA12 in tumor
   samples. In the context of brain tumor, a previous study on only 22 GBM
   samples identified CA12 mRNA levels to be associated with tumors
   carrying EGFR/RNF139 DNA abnormalities. The authors acknowledged the
   limitation of their results as they were retrieved on a small sample
   size only. Moreover, no data on clinical diagnostic molecular markers
   according to WHO guidelines are presented.[128]^11 Haapasalo et al.
   assessed CAXII protein abundance in a larger series of glioma cases
   (n=370) and found signal enrichment with increasing WHO staging of the
   tumors. Their survival studies (on n=269 GBM samples) revealed that
   high CAXII protein predicts for shorter survival of the
   patients.[129]^12 Although this study is indicative and our data can at
   least be seen as confirmative, we believe the respective part of our
   data is more convincing as Haapasalo et al. applied a semi-quantitative
   method, prone to subjective errors. In addition, we extended the
   analysis of CA12 activation to clinical molecular markers, namely IDH1
   DNA mutation status, chromosome aberration status and MGMT-methylation
   status of the tumors. Moreover, we correlated CA12 expression to total
   transcriptome signatures of the tumors. In total, given our
   multi-method the acquired data are more comprehensive and more precise
   than previous reports, and our CA12 focused results are a superior
   contribution to the field of precision neuro-oncology.

   Given GSCs are thought to be a main cause of tumors’ malignant
   properties, we extended our analysis to probe if CA12 can be associated
   with glioma stem cell maintenance. By and large, both our expression
   analysis in clinical samples, as well as protein analysis in naïve
   experimental models did not present a clear picture regarding this
   question. Expression of stem cell markers in patient samples is both
   positively and negatively associated to high CA12 mRNA abundance. On
   the other hand, inter-case comparison of our stem cell models showed
   that CAXII is activated to different extents, from high activity to
   virtual absence of signals in Western blot or FACS, depending on the
   cell line. We also did not notice any correlation between the level of
   cells positive for bona fide GSC marker of cell surface CD133/AC133 and
   the level of cells positive for surface CAXII. Interestingly, our cell
   line data revealed differences in relative abundance of total CAXII and
   cell surface CAXII, indicating differences in protein stability or
   protein processing amongst different GBMs. As our cell models only
   resemble IDH1^WT cases, it would be interesting to see if there is a
   noticeable difference in correlation of CA12/CAXII with stem cell
   molecules when comparing GSC models with or without IDH1^R132. However
   the field lacks overall availability of accepted models carrying IDH
   mutations[130]^27 and we cannot conduct this study at this point. Of
   note, a recent conference abstract reports that CA12 plays a role in
   CSG maintenance but no original data is assessable, such as how the
   authors define GSCs or what the IDH DNA mutation status of the used
   cells is.[131]^35

   Next, aiming towards a clinical translational perspective for our
   anticipated results, we decided to use a pharmacological strategy to
   target CAXII in vitro. By choosing 6A10 as a previously validated,
   highly specific CA12 blocking
   antibody,[132]^21^,[133]^22^,[134]^36^,[135]^37 we found that only GBM1
   models show significant phenotypic responses to the treatment. Although
   all IDH1^WT, our models represent different molecular subtypes of GBM
   (GBM1 and 407 are classical subtype, BTSC233 is mesenchymal
   subtype,[136]^24 data not shown). Thus, together with the known fact of
   the enormous inter-tumoral heterogeneity of the
   disease,[137]^15^,[138]^28 our data suggest that the therapeutic
   potential of 6A10 may be limited to a subset of tumors only. In the
   context of Notch1 receptor, we recently showed that drug target protein
   abundance in GSCs is predictive for the cells’ responsiveness to the
   blocking antibody intervention.[139]^20 Our correlative data in this
   project do not support this rationale to be true in the context of
   CAXII and 6A10, however we acknowledge our study has severe technical
   limitations. First, we did not include an IgG control arm to test any
   off-target effects of the antibody setup. We acknowledge previous work
   of others that indicates that IgG treatment does not cause severe
   phenotypic effects of responder model GBM1,[140]^38 but differences in
   age of the cells and different IgG specifications may have varying
   influences. Moreover we designed our study on the assumption of
   ultra-high target specificity of 6A10 to apply in our tests, further
   justifying our chosen mock treatment as an acceptable experimental
   design for this project, especially given the main focus of this paper
   lies on the molecular diagnostics of patient samples. Second, a genetic
   model of CAXII suppression could be applied to discern if the
   phenotypic response of 6A10 is target specific

   Our translational data position well in current drug development
   campaigns as the generation of CA-inhibitors is under heavy
   investigation worldwide.[141]^19^,[142]^39–43 Given this active field
   of research, a comprehensive discussion of the therapeutic angle of our
   6A10 data, adequately appreciating the entire activities in the field,
   is challenging. That is why we focus on direct competitor drug
   candidates in the context of neuro-oncology. In this regard, 6A10 has
   to withstand benchmarking against very promising recent results
   reported with natural-derived compound Psammaplin C and its
   derivatives. Those substances impair enzymatic activity of CA12 in
   glioma cells, reducing their resistance to the standard
   chemotherapeutic agent temozolomide (TMZ) that eventually leads to
   prolonged survival of mice bearing GBM models.[143]^44^,[144]^45
   Emergence of therapy resistance against CA12 inhibitors is discussed,
   with reports showing CA9 compensating altered enzymatic activity of
   CA12 to promote cancer malignancy;[145]^9^,[146]^10 we sought to
   correlate the expression patterns of those isoforms in our clinical
   samples. Although we find positive correlation of CA12 with CA9 in our
   clinical datasets ([147]Suppl. Figure S6), further studies testing
   enzymatic activity of CA12 and CA9 upon 6A10 treatment will be relevant
   to assess the potential of this antibody to achieve durable therapeutic
   effects. Moreover, testing the effect of 6A10 treatment on the
   resistance of cells to standard chemotherapy may be an additional set
   of experiments to benchmark the antibody against competitor compounds.
   Although not observed in all studied models, 6A10 can impair GSC growth
   to some extent. This observation is in line with reported effects of
   6A10 being able to block the growth of cells from lung, gastric, ductal
   breast and colon cancers.[148]^37 However, to the best of our
   knowledge, for the first time we provide evidence that anti-CA12
   directed treatment in glioma can effectively block cell invasion marker
   ZEB1 accompanied by significant suppression of cellular invasion. Given
   the relevance of ZEB1 in mediating the highly invasive growth of
   GBM,[149]^3^,[150]^4 and given ZEB1 activation has been suggested to
   mediate the resistance of GBM cells to TMZ,[151]^4^,[152]^24^,[153]^46
   we believe this result is of particular high translational interest
   requiring further exploration. Interestingly, CA12 activation was found
   to be a consequence of breast cancer cell EMT.[154]^47 Moreover, in a
   pan-cancer study, CA12 was prominently positioned as a top candidate of
   metabolic genes that is positively correlated to EMT transcription
   signature[155]^48 and TGF-b signaling in GBM, a known glioma EMT
   activator,[156]^2 augments CAXII signals[157]^49. Taken together, our
   functional data extend evidence that CAXII participates in - or at
   least is a reporter of - cancer EMT programs. We enforce the concept
   that generation of pharmacological anti-EMT directed strategies is
   feasible and that these represent a promising approach to develop
   future tumor-agnostic treatment regimes.

Conclusion

   CA12 activation predicts for negative clinical course in Asian and TCGA
   brain tumor patients and is associated with clinical molecular marker
   of IDH1^WTDNA and to 1p19q chromosome arm co-deletion. GBMs of the
   mesenchymal subtype are enriched for CA12. In total, our data fill the
   hitherto existing gap of knowledge of probing the emerging cancer
   target CA12 in the context of state of the art molecular
   neuropathology. Correlative analysis in clinical samples and
   corresponding protein studies on experimental models does not promote
   CA12/CAXII to present a specific glioma stem cell marker. However,
   anti-CAXII directed pharmacotherapy using 6A10 blocking antibody can
   cause suppression of growth and cellular invasion in a fraction of GSCs
   and supports the growing concept that CAXII interacts with cancer EMT
   programs. Further studies are required to delineate the risk of
   off-target events of 6A10 in GSCs and to identify the molecular
   background responsible for the different response rates of glioma cells
   to the antibody. We enforce the concept that generation of
   pharmacological anti-EMT directed strategies is feasible and that these
   represent a promising approach to develop future tumor-agnostic
   treatment regimes.

Acknowledgments