Connexin 43 drives glioblastoma cancer stem cell phenotypes through a WNK lysine-deficient protein kinase 1-c-MYC signaling axis

Summary

The coordination of cellular processes such as growth and survival relies on communication between cells through gap junctions. Connexin proteins comprise gap junctions and also function to mediate protein-protein interactions and communication with the extracellular space via hemichannels. Despite their essential roles, connexin function in cancer is context dependent, with connexin 43 (Cx43) reported to both promote and suppress tumor growth in glioblastoma, the most common primary malignant brain tumor. Here, we detect primarily intracellular expression of Cx43 in glioblastoma patient-derived cancer stem cells and demonstrate that Cx43 is essential for their survival, self-renewal, and tumor initiation. Mechanistically, Cx43 depletion reduces c-MYC expression through reduced levels of the upstream mediator WNK lysine-deficient protein kinase 1 (WNK1). WNK1 depletion phenocopies Cx43 knockdown and reduces MYC expression and tumor initiation. Together, these results define a signaling axis downstream of Cx43 that promotes tumor growth and cancer stem cell phenotypes in glioblastoma.

Graphical Abstract

Introduction

Patients diagnosed with glioblastoma (GBM), the most common primary malignant brain tumor, typically experience a median survival of less than two years after diagnosis¹. Standard-of-care therapy consisting of maximal safe surgical resection followed by radiation and chemotherapy with the alkylating agent temozolomide has produced a minimal extension in survival, and tumors universally recur. Therapeutic resistance and recurrence of GBM are driven by a combination of tumor-intrinsic characteristics such as the presence of therapeutically resistant, self-renewing cancer stem cells (CSCs), in addition to tumor cell-extrinsic interactions with the tumor microenvironment, including with immune cells. The ability to target these tumor-supportive features will aid in future therapeutic development efforts.

In a normal tissue, adjacent cells communicate directly with one another to coordinate complex processes and also utilize cell-cell communication methods to sense and interact with the surrounding microenvironment (reviewed in ^2,3). One major method of cell-cell communication, gap junction intercellular communication (GJIC), is mediated by gap junctions composed of connexin proteins. Six connexin proteins form a hexameric structure, termed a connexon or hemichannel, with a central pore that opens to allow small molecules to pass through. Connexons in the membrane of one cell dock with a connexon on an adjacent cell, allowing the transfer of material, such as metabolites, ions, miRNAs, and some proteins smaller than 1 kD, between cells. In some situations, connexons have been shown to function as hemichannels, allowing the movement of small molecules between a cell and the extracellular space^2,4. Connexins also bind to a number of intracellular proteins, mainly through their C-terminal cytoplasmic tails, and act as scaffolds to link other proteins together. There are 21 connexin proteins in humans, with the main sequence divergence falling within the C-terminal tails. For this reason, specifically targeting connexins has proven challenging.

Connexins and gap junctions have historically been thought of as tumor suppressive, based on reduced GJIC and the frequent loss of connexin expression in tumor tissue, including in GBM^2,5–7. However, we and others have more recently shown that certain cancer types depend on the expression of specific connexins for both gap junction-dependent and gap junction-independent functions³. Specifically, we showed that GBM CSCs from patient-derived xenograft (PDX) models depend on cell-cell communication through connexin 46 (Cx46)^8,9 and that triple-negative breast CSCs depend on the formation of an intracellular gap junction-independent complex with NANOG and focal adhesion kinase (FAK) by Cx26^10,11. In both cases, targeting of connexin activity reduced tumor growth and suppressed CSC self-renewal. Although specific targeting of connexins remains a challenge in vivo, the concept of reducing connexin function, either directly or by inhibiting downstream events, may be therapeutically beneficial.

The most widely expressed and frequently studied connexin, Cx43, has been reported to have both tumor promoting and tumor suppressing roles in a number of cancers, including GBM. Expression levels of Cx43 vary widely among GBM patient samples and tumor cell lines and are frequently – but not universally – reported to be lower than those of normal brain^12–16. Functionally, Cx43 expression in GBM has been shown to drive pro-tumorigenic properties such as therapeutic resistance^17–24, cell proliferation and survival^25–27, and tumor cell invasion and migration^27–29. However, contrasting results have also been reported for Cx43, including reduced proliferation and migration upon introduction of Cx43 into GBM cell lines^30–37. Many of these studies of Cx43 in GBM have been limited by the use of traditional, serum-cultured cell lines, either human or rat, that do not recapitulate the complex cellular heterogeneity and molecular alterations observed in patient tumors³⁸. Studies in CSC models of GBM, while less abundant, have produced similarly contrasting results; specifically, Cx43 has been shown to suppress tumor growth by inhibiting SRC signaling^33,36,39,40 but also to promote chemoresistance to temozolomide, the GBM standard-of-care chemotherapy, by activating PI3K/AKT^18,19.

Here, we identify a non-junctional role for Cx43 in maintaining GBM CSC proliferation, survival, and tumor initiation through expression of the c-MYC proto-oncogene. Furthermore, we identify expression of WNK lysine-deficient protein kinase 1 (WNK1) as a signaling intermediate between Cx43 and C-MYC. Together, this work has revealed a Cx43-WNK1-MYC signaling axis that drives GBM CSC properties and may be applicable to other cancer types.

Results

Cx43 is expressed by GBM PDX CSCs

Our previous work interrogating the role of connexin proteins in GBM indicated that Cx46 (GJA3) is essential for GBM CSC survival, self-renewal, and tumor initiation in a subset of models, with Cx43 primarily expressed in non-stem tumor cells^8,9. A number of studies using high-passage GBM cells lines and PDX models have suggested that Cx43 (GJA1) is also involved in GBM cell proliferation, survival, migration, and therapeutic resistance^18,19,25. To gain insight into global connexin expression in GBM PDX CSCs, we profiled mRNA expression of all connexins in RNA sequencing datasets from three distinct PDX specimens. RNA-sequencing data for T4121 and T3691 was retrieved from GSE79736⁴¹. Of the 21 human connexin genes, we detected low levels of expression of GJA3 (Cx46), GJB1 (Cx32), GJC1 (Cx45), and GJC2 (Cx47), with comparatively high expression of GJA1 (Cx43) (Fig. 1A). The relatively high levels of Cx43 and low levels of Cx46 observed here were not consistent with our previous observations and led us to more closely examine levels of Cx43 protein in GBM patient tissues (see Table 1 for patient information). In the human cortex (non-tumor conditions, used as a control), Cx43 protein was expressed diffusely from layers 1 – 4 (Fig. S1). Staining was restricted to gray matter and presented as an aggregate of particles condensed around the cell bodies of cortical projection neurons. H&E staining of GBM samples (Fig. 1B, left panel for each patient) presented hallmark histological feature of the disease, including perivascular proliferation, diffuse tissue invasion, and pseudopalisading necrosis. In contrast, H&E staining of metastatic colon adenocarcinoma (Fig. 1B, brain metastasis panels) presented focal, well-circumscribed, and non-invasive histology typical of metastatic lesions of the brain. Staining intensity for Cx43 ranged in patients with GBM from strong (Fig. 1B, Patients 2 and 3) to moderate (Fig. 1B, Patient 1) to light/negligible (Fig. 1B, Patient 4; see also Fig. S1). As in the non-tumor cortical samples, Cx43 protein assembled into loose condensations of puncta, in contrast to the condensed, linear staining pattern indicative of gap junctional plaques that present between cardiac myocytes (Fig. 1B, heart panel; see also Fig. S1), which is consistent with previous reports¹⁶. We observed this staining pattern within hyperdense tumor cores, immediately adjacent to and within millimeters of sites of necrosis (Fig. S1, patients 5 and 6). Cx43 coalesced in perivascular spaces (Fig. S1, patient 9) and, unlike the non-tumor cortex, Cx43 protein was evident in the frayed white matter tracts associated with the tumor (Fig. S1, patient 8).

Figure 1. Cx43 is expressed in GBM PDX CSCs.

(A) Expression of all connexin genes was assessed in GBM PDX CSC models T4121 (GSE79736), T3691 (GSE79736), and DI318 (performed in this study). Expression is shown as fragments per kilobase of transcript per million mapped reads (FPKM). Means ± standard deviations from three independent replicates are shown. (B) H&E and Cx43 protein expression were evaluated in brain specimens isolated from GBM patients, non-cancerous control patients, and patients with metastatic colon cancer to the brain. Heart tissue was used as a positive control for gap junction plaques. Scale bar for H&E-stained specimens = 500 μm. Scale bar for Cx43-stained specimens = 50 μm. (C-D) Baseline levels of Cx43 were examined by immunoblotting in a panel of GBM PDX CSC lines (C) and compared those of traditional serum-cultured cell lines and normal human astrocytes (NHAs) (D). For B, samples were spread across two gels that were run, transferred, incubated, and imaged in parallel. Actin was used as a loading control. B and C are representative of n = 3 independent experiments. Also see Fig. S1 and S2.

Table 1.

GBM patient information.

Patient	Sex	Age	Diagnosis	Overall survival (days)
GBM 1	F	84	GBM	4
GBM 2	F	67	GBM	346
GBM 3	F	64	GBM	174
GBM 4	F	69	GBM	118
GBM 5	F	62	GBM	273
GBM 6	M	62	GBM	201
GBM 7	M	48	GBM	148
GBM 8	M	68	GBM	276
GBM 9	M	67	GBM	700
GBM 10	M	64	GBM	332

As we observed a spectrum of levels of Cx43 expression in GBM patient tissue, we expanded our cellular analysis to a larger panel of GBM PDX models than we had considered in previous efforts. Consistently, we detected Cx43 protein expression at varying levels in the majority of the examined samples by immunoblot (Fig. 1C). Compared to normal human astrocytes (NHAs) as a positive control for Cx43 expression and a panel of non-GBM cancer cell models (human embryonic kidney line 293T, breast adenocarcinoma cell line MDA-MB-231, cervical cancer line HeLa, and prostate cancer line LNCaP) as a negative control, GBM PDX samples expressed intermediate levels of Cx43 (Fig. 1D; quantitation provided in Fig. S2A). Using the IvyGAP database to interrogate spatial GJA1 expression in GBM tumor regions, we found a small but significant enrichment of GJA1 mRNA in the leading edge, which is consistent with previous reports that Cx43 is important for GBM cell migration (Fig. S2B)^27–29. Examination of GJA1 mRNA expression within the GBM cellular subsets identified by Neftel et al.⁴² suggested that the highest expression was found in the astrocyte-like cell fraction, with relatively high levels also detected in the two mesenchymal cell types (Fig. S2C). Based on our previous work that described a non-junctional role for Cx26 in driving triple-negative breast cancer^10,11 and the non-membrane appearance of Cx43 in our GBM patient tissue staining, we further interrogated the localization of Cx43 in GBM PDX models. Consistent with previous work in GBM cells^43–45, we found that Cx43 was primarily localized to an intracellular perinuclear compartment, likely the Golgi or endoplasmic reticulum, with small amounts found at the plasma membrane (Fig. S2D). Together, these data indicate that Cx43 is expressed at detectable levels in GBM PDX CSCs with a primarily intracellular localization.

Cx43 is essential for GBM CSC survival and self-renewal

Based on this observed expression of Cx43 in GBM CSCs, we next interrogated whether Cx43 is important for the function of these cells. Cx43 mRNA was depleted from multiple PDX CSC models using three non-overlapping shRNA constructs (Fig. 2A, Fig. S3A–B), and we observed a significant decrease in cell number over a span of 5 days (Fig. 2B, Fig. S3C). This effect was evident even with a relatively modest reduction in Cx43 protein. This reduction in cell number was accompanied by an increase in apoptotic cell death as measured by caspase 3/7 activity (Fig. 2C, Fig. S3D). We then assayed the critical functional stem cell property of self-renewal using limiting-dilution analysis and observed a significant decrease in stem cell frequency (Fig. 2D, Fig. S3E). Based on these results, we hypothesized that reduction of Cx43 would reduce intracranial GBM tumor initiation, and this is indeed what we observed. Implantation of DI318 CSCs expressing Cx43 shRNA into the brains of NSG mice significantly increased animal survival compared to cells expressing non-targeting shRNA (Fig. 2E). Together, these results indicate that Cx43 is required for GBM CSC properties including survival, self-renewal, and tumor initiation and suggest a pro-tumorigenic role for Cx43.

Figure 2. Cx43 is required for GBM PDX CSC survival.

DI318 and T3691 PDX CSC models were transduced with three non-overlapping lentiviral shRNA constructs against Cx43 (GJA1). (A) The degree of knockdown was verified via immunoblotting using an antibody against Cx43. β-actin was used as a loading control. (B) Cell viability of DI318 and T3691 CSCs containing shRNAs against Cx43 was measured via CellTiter-Glo on day 5 after plating. Luminescence values were normalized to day 0 values, and the fold change was calculated relative to the non-targeting control. n = 5 (DI318) and n = 3 (T3691) independent experiments, each performed in technical triplicate. (C) Cell death of DI318 and T3691 CSCs containing shRNAs against Cx43 was measured via CaspaseGlo 3/7 on day 3 after plating. Luminescence values were normalized to cell number measured via CellTiter Glo at the same time point and then to the non-targeting control. n = 5 (DI318) and n = 3 (T3691) independent experiments, each performed in technical triplicate. (D) DI318 and T3691 CSCs containing Cx43 shRNAs were plated in decreasing cell number (20, 10, 5, and 1 cells/well) with 24 replicates per number and evaluated for sphere formation 10–14 days later. n = 5 (DI318) and n = 4 (T3691) independent experiments. The online algorithm outlined in the Methods section was used to calculate stem cell frequency. For B-D, *p < 0.05, **p < 0.01, ***p < 0.001 by one-way ANOVA analysis with Dunnett’s Multiple Comparison Test, and data are shown as means ± standard deviations. (E) A total of 25,000 DI318 cells containing Cx43 shRNA was implanted into the brains of immunocompromised NSG mice, and animals were monitored until neurological endpoint. Median survival, animal numbers, and p-values shown on plot; p-value calculated by log-rank test. Also see Fig. S3.

Cx43 is required for expression of the c-MYC proto-oncogene

To investigate the mechanistic underpinnings of the requirement for Cx43 in CSCs, we performed RNA sequencing of cells containing shRNA against Cx43. One of the top differences we observed compared to cells expressing a non-target shRNA was a decrease in the levels of the proto-oncogene c-MYC (Fig. S4A–B). When the results from both shRNAs were considered together, two of the three top downregulated gene sets were c-MYC target genes (Fig. 3A. Fig. S4C). MYC is a transcription factor that drives transformation of neural stem cells and is essential for GBM CSC proliferation, survival, self-renewal, and tumor initiation due to its role in cell cycle progression^46–49. We validated the decrease in MYC by qRT-PCR in cells expressing Cx43 shRNA (Fig. 3C). This change in mRNA levels correlated with a decrease in MYC protein level (Fig. 3D). Based on this link between Cx43 and MYC, we next investigated potential signaling intermediates using the Dependency Map (DepMap) portal (https://depmap.org/portal ⁵⁰). We considered top genes that were found to exhibit co-dependency with MYC across a panel of cancer cell lines by either RNAi (DEMETER2 data set⁵¹; includes the Achilles⁵⁰, DRIVE⁵², and Marcotte⁵³ data sets) or CRISPR (DepMap 23Q4)⁵⁴. Only four genes were found in both datasets (Fig. 3E), those encoding the MYC dimerization partner MYC-associated factor X (MAX), the meiotic protein meiosis regulator for oocyte development (MIOS), the mitochondrial protein hydroxysteroid 17-beta dehydrogenase 10 (HSD17B10), and the kinase WNK lysine deficient protein kinase 1 (WNK1). WNK1 is a member of the WNK family of protein kinases (also including WNK2–4) and canonically regulates ion homeostasis by phosphorylating ion channels⁵⁵. As we were searching for a signaling intermediate that could control MYC mRNA expression and WNKs drive several cancer-associated signaling pathways, including expression of c-MYC^56–58, we chose to further investigate the role of WNK1 as a candidate to mediate signaling between Cx43 and MYC. We observed a reduction in WNK1 levels by ELISA in GBM CSCs expressing Cx43 shRNA compared to non-target (Fig. 3F), suggesting a role for Cx43 in WNK1 expression and that WNK may serve as an intermediate between Cx43 and MYC expression.

Figure 3. Cx43 is necessary for MYC expression.

(A) Enrichment plot of the genes contained in the HALLMARK_MYC_TARGETS_V1 gene set enrichment dataset (M5926) in DI318 CSCs expressing Cx43 shRNA. (B) Quantitation of mRNA levels of MYC from the RNA sequencing experiment. n=1 experiment with 3 independent replicates, and data are shown as means ± standard deviations. (C) MYC expression was assessed over three Cx43 shRNA constructs using qRT-PCR in DI318 and T3691 and normalized to GAPDH. n = 3 independent experiments performed in technical triplicate for each model, and data are shown as means ± standard deviations. (D) DI318 and T3691 containing Cx43 shRNA were subjected to immunoblot for MYC protein. β-actin was used as a loading control. n ≥ 5 independent replicates. (E) The Cancer Dependency Map (DepMap) portal was used to identify genes that exhibit co-dependency with MYC across a panel of cancer cell lines (see Results section for details). The top approximately 100 hits from both RNAi (DEMETER2 data set) and CRISPR screening (DepMap 23Q4 data set) were selected, and overlapping genes between the two datasets were determined. (F) Levels of WNK1 were measured by ELISA in DI318 containing Cx43 shRNA and compared to non-target. n = 4 independent experiments performed in technical triplicate, and data are shown as means ± standard deviations. *p < 0.05, **p < 0.01, ***p < 0.001, **** p < 0.0001 by one-way ANOVA analysis with Dunnett’s Multiple Comparison Test. Also see Fig. S4.

Disruption of WNK1 compromises GBM CSC survival

Based on these data, we interrogated whether WNK1 is important for GBM CSCs. Using three non-overlapping shRNA constructs (Fig. S5A), we reduced WNK1 levels compared to a non-targeting control (Fig. 4A, Fig. S5B). These WNK1-depleted cells exhibited a reduction in cell number after 5 days in culture (Fig. 4B, Fig. S5C) with a concomitant increase in apoptosis as measured by caspase 3/7 activation (Fig. 4C, Fig. S5D). Self-renewal was also dramatically reduced in cells containing WNK1 shRNA constructs compared to those with a non-targeting control (Fig. 4D, Fig S5E), suggesting that WNK1 is essential for CSC function. Finally, we implanted CSCs expressing non-target or WNK1 shRNA into the brains of immunocompromised mice and monitored the animals until neurological endpoint. We observed a significant extension in survival of mice with tumors derived from WNK1-depleted cells (Fig. 4E), further supporting our observations that WNK1 drives tumor cell growth and tumor initiation by GBM CSCs.

Figure 4. WNK1 is required for CSC viability, survival, and self-renewal.

PDX CSC models DI318 and T3691 were transduced with lentivirus encoding three different non-overlapping WNK1 shRNA constructs. (A) Cells were subjected to immunoblotting for WNK1. β-actin was used as a loading control. (B) Cell viability of DI318 and T3691 CSCs containing shRNAs against WNK1 was measured via CellTiter-Glo on day 5 after plating. Luminescence values were normalized to day 0 values, and the fold change was calculated relative to the non-targeting control. n = 4 (DI318) and n = 3 (T3691) independent experiments, each performed in technical triplicate. (C) Cell death of DI318 and T3691 CSCs containing shRNAs against WNK1 was measured via CaspaseGlo 3/7 on day 3 after plating. Luminescence values were normalized to cell number measured via CellTiter Glo at the same time point and then to the non-targeting control. n = 6 (DI318) and n = 4 (T3691) independent experiments, each performed in technical triplicate. (D) DI318 and T3691 CSCs containing WNK1 shRNAs were plated in decreasing cell number (20, 10, 5, and 1 cells/well) with 24 replicates per number and evaluated for sphere formation 10–14 days later. n = 3 independent experiments for both models. The online algorithm outlined in the Methods section was used to calculate stem cell frequency. For panels B-D, *p < 0.05, **p < 0.01, ***p < 0.001 by one-way ANOVA analysis with Dunnett’s Multiple Comparison Test, and data are shown as means ± standard deviations. (E) A total of 25,000 DI318 cells containing WNK1 shRNAs was implanted into the brains of immunocompromised NSG mice, and animals were monitored until neurological endpoint. Median survival, animal numbers, and p-values shown on plot; p-value calculated by log-rank test. Also see Fig. S5.

WNK1 controls MYC expression in GBM CSCs

Based on our hypothesis that WNK1 may serve as a signaling intermediate between Cx43 and MYC, we tested whether WNK1 is necessary for MYC expression. WNK1 has previously been shown to control MYC expression in developing T cells⁵⁶, hepatocellular carcinoma⁵⁸, and multiple myeloma⁵⁷ but not in B cells⁵⁹, and a link in GBM has yet to be determined. To gain a broad understanding of WNK1-mediated signaling changes, we performed RNA sequencing and observed that MYC (Fig. S6A–B) and MYC targets (Fig. 5A and Fig. S6C) were significantly downregulated in GBM CSCs containing WNK1 shRNA compared to a non-targeting control. We further confirmed this result by qRT-PCR and observed a significant reduction in MYC mRNA levels in the presence of WNK1 shRNA (Fig. 5C). MYC protein showed a similar trend, with decreases in the presence of WNK1 shRNAs compared to a non-targeting sequence (Fig. 5D, Fig. S6D), and a MYC reporter also showed reduced activity (Fig. 5E). As WNK2 has tumor-suppressive effects in GBM cells^60,61, we utilized RNA sequencing data to determine whether WNK1 depletion might act by increasing WNK2 levels. Importantly, RNA sequencing revealed that expression of WNK2, WNK3, and WNK4 was very low in GBM CSCs and did not increase with Cx43 or WNK1 shRNAs (Fig. S6E), suggesting that other WNK proteins do not compensate at least at the transcriptional level upon Cx43 or WNK1 depletion. Based on our observations that depletion of WNK1 phenocopies reduction of Cx43 and Cx43 reduces levels of WNK1, together, our work suggests the presence of a signaling axis containing Cx43, WNK1, and MYC that is essential for GBM CSCs.

Figure 5. WNK1 is upstream of MYC in GBM CSCs.

(A) Enrichment plot of the genes contained in the HALLMARK_MYC_TARGETS_V1 gene set enrichment dataset (M5926) in DI318 CSCs expressing Cx43 shRNA. (B) Quantitation of MYC mRNA level by RNAseq in cells containing WNK1 shRNA compared to non-targeting control. n = 1 experiment of three independent replicates, and data are shown as means ± standard deviations. (C) MYC expression was assessed over three WNK1 shRNA constructs using qRT-PCR in DI318 and T3691 and normalized to GAPDH. n = 5 (DI318) and n = 3 (T3691) independent experiments performed in technical triplicate for each model, and data are shown as means ± standard deviations. (D) DI318 and T3691 containing WNK1 shRNA were subjected to immunoblotting for MYC protein. β-actin was used as a loading control. n ≥ 5 independent replicates. (E) T3691 cells containing a MYC luciferase reporter plasmid were transduced with WNK1 shRNAs, and luciferase was measured. n = 3 independent experiments each performed in technical triplicate, and data are shown as means ± standard deviations. *p < 0.05, **p < 0.01, ***p < 0.001 by one-way ANOVA analysis with Dunnett’s Multiple Comparison Test. Also see Fig. S6.

Discussion

While our understanding of the tumor-suppressive and tumor-promoting roles of connexins continues to evolve, our observations demonstrate a new pro-tumorigenic signaling axis in GBM CSCs linking Cx43 to c-MYC via WNK1. Our previous work showed an essential role for GJIC through Cx46 in GBM CSCs from a subset of patient samples^8,9. However, this work was unable to directly compare levels of Cx46 to those of other connexins. Here, using sequencing approaches, we show that each of the GBM CSC samples assessed expressed higher levels of Cx43 compared to all other connexins at the mRNA level. The question remains as to whether these patient samples exhibit distinct dependencies on an individual connexin or whether cells that require Cx46 may also require Cx43. The levels of Cx43 we detected in GBM cells were higher than those of traditional, serum-cultured cancer cell lines but lower than those observed in normal astrocytes. The low to undetectable levels of Cx43 in breast cancer, cervical cancer, and prostate cancer cell lines are consistent with the traditional hypothesis that Cx43 is tumor suppressive and is lost upon transformation. However, the comparatively robust expression of Cx43 in GBM CSCs contrasts with that hypothesis and lends further support to the idea that context is critical for understanding the role of connexins and gap junctions, with varying expression, localization, dependency, and molecular function depending on the situation. We detected the majority of Cx43 protein localized intracellularly and not at the plasma membrane, which is consistent with the loss of GJIC in transformed cells and has been previously reported^16,43–45. In fact, mutations in Cx43 in non-small cell lung cancer retain the protein in a perinuclear compartment, and this intracellular localization correlates with increased tumor grade⁶². However, in GBM, the mechanisms underlying Cx43 expression and localization require further study. It is possible that Cx43 levels are regulated by culture conditions, as we observed low expression with the presence of serum. However, our normal astrocytes with high Cx43 were also grown in the presence of serum, although a reduced amount, suggesting that serum presence is not a defining factor in Cx43 expression. Alternatively, our interrogation of the CSC fraction of GBM may have enriched for Cx43 expression compared to bulk tumor; the low frequency of CSCs within the tumor bulk may in fact have contributed to the observation in historical studies of reduced levels of Cx43 within GBM. It is also possible that GBM may simply differ in its underlying biology compared to non-neural tumors, or the long-term culture of the traditional cell lines (including in serum culture) has led to reduced Cx43 levels. Additionally, while some studies have proposed a role for Cx43 in communication between GBM cells, our models show an intracellular, non-junctional localization for Cx43 and do not support such a function. While it remains unclear why different GBM models in different laboratories show discrepancies in functional roles for Cx43, this may be due to the heterogeneity of GBM and deserves further study. Questions of how our GBM models express notable levels of Cx43 protein and why Cx43 does not localize to the plasma membrane also warrant additional work.

Here, we present data indicating a pro-tumorigenic role for Cx43 in GBM CSCs. Mechanistically, we observed that expression of both transcript and protein levels of the proto-oncogene c-MYC relies on the presence of Cx43 and WNK1. Our identification of WNK1 hinged on overlap between DepMap datasets utilizing CRISPR and shRNA to determine cancer cell co-dependencies with MYC. While the limited overlap we observed between RNAi and CRISPR screens was striking, this discrepancy was not surprising due to the differential effects of RNAi (hypomorphic effects) vs CRISPR (true loss of function). Each of these approaches has its own limitations to specificity, and in fact, studies testing both methods head to head identified similar discrepancies^63,64. It has been proposed that combinatorial screening approaches using both shRNA and CRISPR, as we do here, are in fact the most effective at identifying truly essential genes⁶⁵.

Here, we show a link between Cx43 and MYC and between Cx43 and WNK1 expression. c-MYC is a critical regulator of cell cycle progression and apoptosis that is frequently overexpressed in GBM⁶⁶. Previous work from our group and others showed this essential role for MYC in GBM CSCs: MYC is necessary for cell survival, cell-cycle progression, self-renewal, and tumor formation^46,49. Deregulation of MYC may be an initiating event in gliomagenesis; it has been shown that epigenetic reprogramming driven by MYC is sufficient to induce a stem cell-like state, leading to tumorigenesis^67,68. Based on these data, we believe that the loss of MYC is likely sufficient to drive the reduced survival we observed upon Cx43 depletion. However, Cx43 likely affects numerous other cellular processes, as supported by our RNA sequencing data, some of which may also be essential for cell survival.

While WNK1 has previously been identified upstream of MYC expression in hepatocellular carcinoma⁵⁸, multiple myeloma⁵⁷, and developing T cells⁵⁶, we are now able to link WNK1 and MYC in GBM, indicating that WNK1 is an important and understudied player in GBM biology. Our data show that WNK1 supports tumor cell growth, and this is supported by the few previous studies on WNK1 in GBM. Inhibition of WNK1 immediate downstream substrates, STE20/SPS1-related proline/alanine-rich kinase (SPAK) and oxidative stress‐responsive kinase 1 (OSR1), reduced glioblastoma proliferation, migration, and tumor growth⁶⁹; however this study did not test for any kinase- or SPAK- independent roles for WNK1. Another study also detected reduced glioma cell migration with WNK1 shRNA, likely due to reduced phosphorylation of Na+/K+/2Cl− cotransporter isoform 1 (NKCC1), but did not assess a role in cellular proliferation, stem cell functions, or tumor growth⁷⁰. While we were unable to detect mRNA expression of any other WNK family member in our cells (Fig. S6E), WNK3 has also been shown to promote GBM cell migration^71,72. Interestingly, while WNK1 and WNK3 appear to be tumor promoting, WNK2 has been shown to suppress glioma cell migration⁷³, expression of matrix metalloproteinase 2 (MMP2)⁶¹, and activation of c-Jun N-terminal kinase (JNK)⁶¹, suggesting that it instead has tumor-suppressive functions. Furthermore, WNK1 has a MYC response element in its promoter, suggesting that the relationship between these two proteins may be more complex than currently understood. Additional work is needed to fully understand the roles of the WNK family members in tumor cell properties such as growth, migration, and tumor initiation and their relationship to MYC expression.

The Cx43-WNK1-MYC signaling axis may present opportunities for therapeutic development. Connexins have historically been difficult to specifically target due to the high degree of sequence homology among family members. However, several Cx43-specific targeting strategies have been developed in recent years. αCT1 is a peptide mimetic of the region of the Cx43 C-terminal intracellular tail that binds to zonula occludens 1, which regulates the balance between docked and undocked connexon hemichannels at the membrane^74,75. Treatment of cells with αCT1 increases gap junction incorporation of Cx43 while reducing hemichannel activity^74,75 and has been tested clinically to increase wound healing⁷⁶. However, as our data support a pro-tumorigenic role for Cx43, increasing the GJIC of Cx43 is likely to be detrimental rather than beneficial to patients. An additional peptide mimetic of Cx43, TAT-Cx43_266–283, has been developed that inhibits the activity of SRC^36,39,40. This peptide is based on the observation that the presence of Cx43 suppresses SRC activity and again mimics the presence of Cx43 rather than reducing its function. Thus, the available Cx43 peptide treatments are not likely to be useful for reducing tumor growth in GBM cells expressing Cx43; however, these types of therapeutic strategies do suggest that it may be possible to develop specific therapies targeting Cx43 in the future. WNK1 and MYC targeting strategies have also encountered roadblocks. In addition to cancers, therapies to target WNK1 have been of interest to treat Gordon’s syndrome (also called pseudohypoaldosteronism type II), a type of familial hypertension caused by WNK mutations. However, WNK inhibitors lack specificity for a single WNK family member, leading to undesirable effects, and development of the WNK inhibitor WNK463 was halted due to unexpected side effects in mice⁷⁷. MYC inhibitors have also been highly sought after for use in cancer treatment, but targeting MYC has thus far been challenging due to its highly disordered regions and lack of an enzymatic active site⁷⁸. While several strategies to target MYC are currently in clinical trials, including the MYC-MAX interaction inhibitor omomyc and a number of indirect inhibitors that target upstream proteins, none have yet been approved for use in the clinic⁷⁸. The increased understanding of additional signaling nodes surrounding poorly targetable proteins such as Cx43, WNK1, and MYC may provide more amenable targets for therapeutic development. This strategy is particularly relevant for proteins such as Cx43 that have multifaceted roles in tumor cells.

Limitations of this study

Together, our work demonstrates a critical role for Cx43 and WNK1 in GBM CSCs for driving MYC expression, leading to cell survival, self-renewal, and tumor formation. However, there are several limitations to our study. While we thoroughly investigated three PDX GBM models, the future investigation of additional models will be beneficial, particularly due to the high degree of heterogeneity present in GBM. In addition, recent work suggests that sex differences may impact GBM cell phenotypes, including cell growth⁷⁹ and immune responses^80,81. While we used both male and female GBM models and male and female patient tissue for this study, a more detailed assessment of this signaling axis with representation of additional male and female models would be useful. Our studies focused on the cell signaling network starting from Cx43, but it is unclear which function(s) of Cx43 (GJIC, hemichannel, or protein-protein interactions) is responsible for the initiation of this signaling network and whether there are any points of feedback. These are key questions to be addressed in future studies, along with whether the WNK1/c-MYC signaling axis can be initiated by any other connexin. Finally, future studies should focus on untangling the role of WNK1 kinase activity in the phenotypes observed here, as many additional components of this signaling pathway remain to be discovered.

Resource availability statement

Lead contact

Requests for further information and resources should be directed to and will be fulfilled by the lead contact, Dr. Justin Lathia (lathiaj@ccf.org).

Materials availability

This study did not generate new unique reagents.

Data and code availability

• RNA sequencing data for DI318 PDX CSCs expressing non-targeting, Cx43, and WNK1 shRNA have been deposited at GEO as GSE272349 and are publicly available as of the date of publication.

• This paper does not report original code.

• Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

STAR Methods

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Patient-derived GBM xenografts

All GBM PDX models have been previously described. The GBM patient derived PDX CSC models T3691 (male), T3832 (female), T387 (female), T4121 (male), and T4302 (male) were obtained from Duke University (originally from Dr. Jeremy Rich) via material transfer agreement^8,82–85. The DI318 model (male) was generated at Cleveland Clinic and used previously as described⁸⁶. BT124 (sex unknown) was obtained from the University of Calgary (Dr. Samuel Weiss)⁸⁷. L1 and L2 (both female) were obtained from University of Florida (originally from Dr. Brent Reynolds)^88,89. GBM23 (male) was obtained from MD Anderson Cancer Center (originally from Dr. Erik Sulman) and used previously as described⁹⁰. The GBM12 model (male) was obtained from the Mayo Clinic (Dr. Jann Sarkaria)⁹¹. All models were de-identified, established in accordance with an Institutional Review Board-approved protocol with informed-consent obtained from patients, and short-tandem-repeat tested to ensure differences in genetic signatures.

GBM PDX models were cultured in suspension in Petri dishes in Neurobasal medium without phenol red (ThermoFisher Scientific) supplemented with 1% penicillin/streptomycin (ThermoFisher Scientific), 2% B27 without vitamin A (ThermoFisher Scientific), 20 ng/mL EGF (R&D Systems), 20 ng/mL FGF (R&D Systems), 1 mM sodium pyruvate (ThermoFisher Scientific), and 2 mM L-glutamine (Cleveland Clinic Research Media Core). Cells were subcultured using Accutase (BioLegend) to obtain a single-cell suspension. When adherence was required, cells were incubated with Geltrex (ThermoFisher Scientific) diluted to approximately 1:1000 in the culture media and cultured on tissue culture-treated dishes. CSCs from T3691, T3832, T387, T4121, and GBM12 were selected by CD133+ magnetic bead sorting (Miltenyi). CSCs from all other PDX lines were enriched through culture in the CSC-promoting media conditions described above.

Additional cell lines

Conventional human cancer cell lines 293T (human embryonic kidney; ATCC: CRL-3216), MDA-MB-231 (female breast adenocarcinoma; ATCC: HTB-26), LNCaP (male prostate carcinoma; ATCC: CRL-1740), and HeLa (female cervical cancer; ATCC: CCL-2) cells were obtained from the Cleveland Clinic Research Cell and Media Production Core and authenticated by the Core before distribution. All cells were maintained in DMEM with 10% FBS and 1% pen-strep in a humidified incubator with 5% CO₂. Cells were passaged using trypsin. Normal human astrocytes (unknown sex) were obtained from Lonza and cultured in supplemented AGM^™ Astrocyte Growth Medium (Lonza) for limited passages according to the vendor’s protocols. All cells were tested for mycoplasma contamination.

In vivo tumor models

All animal usage was performed in accordance with protocols approved by the Cleveland Clinic Institutional Animal Care and Use committee. A total of ten 6-week-old NSG (NOD.Cg-Prkdc^scid Il2rg^tm1Wjl/SzJ; JAX strain #005557) mice per group, five male and five female, were implanted with DI318 cells expressing either a non-targeting shRNA, WNK1 sh491, or WNK1 sh920. For Cx43 shRNA intracranial implantation, DI318 cells expressing either non-target shRNA, sh73, or sh77 were implanted into a total of 10 male mice per group. The mice were anesthetized via 2% isoflurane gas and placed in a stereotaxic instrument. An insulin syringe with a 31-gauge needle was secured to a probe holder, and the needle was passed through the scalp 0.5 mm rostral and 1.8 mm lateral to the bregma. The needle was inserted 3.5 mm underneath the scalp, where 25,000 cells suspended in unsupplemented Neurobasal media were injected slowly into the mice. The needle was held in place for 60 seconds before being slowly removed from the mouse’s scalp. The experiment was blinded to the investigators who injected the mice and to the investigator who measured the endpoints of each mouse.

Histological evaluation of Cx43 in patient tissue

Ten human GBM specimens, along with control tissue, were obtained from Dr. Peter LaViolette as part of the Brain Bank at the Medical College of Wisconsin under IRB approval and deidentified. Brain specimens isolated from GBM, metastatic disease of the brain, and non-cancer patients were preserved in 10% formalin and subjected to paraffin embedding. Tissue sections with 7 μm thickness were stained with hematoxylin and eosin (H&E) using standard protocols. Adjacent sections were subjected to 32 minutes of heat-activated antigen retrieval at 95°C in EDTA before immunohistochemical staining for Cx43. Tissue sections were incubated with a 1:50 dilution of rabbit anti-Cx43 (Cell Signaling Technologies, #3512S) for 1 hour at room temperature using the Ventana BenchMark automated IHC System. A total of 10 GBM specimens, 5 non-cancerous controls, and 5 metastatic brain lesions were investigated. An equal number of male and female patients were evaluated in each cohort. Patient age ranged from 48 to 95 years (see Table 1 for patient information).

METHOD DETAILS

Plasmids and DNA constructs

Predesigned Cx43 and WNK1 shRNA constructs in pLKO.1-puro were purchased from Sigma-Aldrich. Their respective clone ID numbers are as follows: Cx43 sh73 - TRCN0000059773; Cx43 sh76 TRCN0000059776; Cx43 sh77 - TRCN0000059777; WNK1 sh491 - TRCN0000196491; WNK1 sh918 - TRCN0000000918; WNK1 sh920 - TRCN0000000920. The SHC002 Non-Mammalian shRNA Control Plasmid was used as the non-targeting sequence.

Lentiviral preparation and transduction

Lentivirus was prepared by calcium phosphate transfection of 293T cells using psPAX (Addgene # 12260) and pMD2.G (Addgene # 12259), both provided by Didier Trono. Virus was harvested on days 2 and 3 after transfection, concentrated via polyethylene glycol precipitation, and stored at −80°C until use.

For lentiviral transduction, cells were plated in supplemented neurobasal media as described above with 1:1000 Geltrex (ThermoFisher Scientific) into 6-well plates. Cells were then infected with lentivirus at 5 μL/mL 2 days later. Cells were then selected with 2 μg/mL puromycin (Sigma-Aldrich) and collected for assays after two days. Due to the dramatic viability defects observed with shRNA against both Cx43 and WNK1, cells were freshly transduced for each experiment, and any remaining cells were discarded.

Proliferation and apoptosis assays

For each condition, proliferation and apoptosis assays were plated at 2,000 cells per well in triplicate into a 96-well white-walled, clear-bottom plate (BD Biosciences). CellTiter Glo (Promega) was used to measure ATP levels on days 0, 3, and 5 after plating and used as a surrogate for cell viability. Values were normalized to day 0 and then to the non-targeting control to obtain a fold-change value. Apoptosis was measured similarly by determining active caspases 3/7 via CaspaseGlo 3/7 (Promega) on day 3 after plating. Values were normalized to cell number as measured via CellTiter at the same time point.

Limiting-dilution analysis

GBM CSCs were dissociated using Accutase (BioLegend) and subsequently plated into 96-well suspension plates (Sarstedt) in decreasing cell numbers (20, 10, 5, and 1 cells/well) with 24 replicates for each cell number. After 10–14 days, wells were evaluated for the presence of spheres, which was recorded in a binary manner. The stem cell frequency was subsequently calculated using the online Extreme Limiting Dilution Analysis algorithm (https://bioinf.wehi.edu.au/software/elda/)⁹².

MYC reporter assay

The MYC luciferase reporter was purchased from BPS Bioscience, and luciferase was detected on day 3 after selection with puromycin using the ONE-Step^™ Luciferase Assay System (BPS Bioscience) and a Victor Nivo multimodal plate reader. MYC reporter signal was normalized to the number of live cells as measured by CellTiter Glo at the same time point.

Immunoblotting and antibodies

To prepare lysates, cells were washed with PBS and lysed with lysis buffer (0.5% NP-40, 10 mM Tris-Cl pH 7.5, 1 mM EDTA pH 8.0, 150 mM NaCl) supplemented 1% phosphatase inhibitors and 1% protease inhibitors (Sigma-Aldrich). Insoluble debris was removed via centrifugation at 21,000 × g for 10 minutes at 4°C. Protein concentration was determined via the Bio-Rad protein assay using a 1 mg/mL solution of bovine serum albumin (BSA) for standardization. For blotting, 15–50 μg of protein lysate was separated via SDS-PAGE electrophoresis. Pre-cast 4–15% gels (Bio-Rad) were used to resolve WNK1 protein. Gels were subsequently transferred onto PVDF membranes (Millipore), which were blocked for 45 minutes to 1 hour with 5% nonfat milk. Primary antibodies were then incubated on the membranes for 2–7 days at 4°C in 5% BSA in Tris-buffered saline with Tween-20 (TBST) before being washed with TBST. Membranes were then incubated with the appropriate horseradish peroxidase-tagged secondary antibody for 2.5 to 3 hours at room temperature. Primary antibodies used were anti-Cx43 at 1:1000 dilution (Cell Signaling, #3512S), anti-WNK1 at 1:1000 dilution (Cell Signaling, #4979S), and anti-c-MYC at 1:1000 dilution (Cell Signaling, #9402S). The membranes were developed using the Pierce ECL 2 Western Blotting Substrate (ThermoFisher Scientific) and imaged on a Bio-Rad ChemiDoc MP imaging system. Anti-Actin hFAB Rhodamine Antibody at 1:10,000 (Bio-Rad) was used as loading control.

GJA1 expression in GBMap dataset

Single-cell RNA sequencing data from the GBM dataset (GBMap) were utilized for this analysis⁹³. Malignant cells were isolated based on pre-annotated cell type information within the dataset. To assign Neftel et al.⁴² cell states, gene expression signatures for each state were obtained from the original publication. Gene lists were converted to Ensembl IDs (gconvert function from gprofiler2 v.0.2.3, retaining only successfully mapped IDs). UCell (v.2.8.0) was employed to calculate a module score for each cell state signature for each malignant cell. Malignant cells were classified on the basis of the maximal module score across cell states. The expression of the GJA1 gene (Ensembl ID: ENSG00000152661) was visualized across the assigned cell states (Seurat, v. 5.2.1).

Immunofluorescence staining and confocal imaging

Round German glass coverslips (EMS; 12 mm, #1.5) were coated the day prior with Geltrex (ThermoFisher) diluted 1:1000 in ice-cold DMEM/F-12 to enhance cell adherence. Coverslips were incubated for at least 1 hour at 37°C and maintained in media until cell seeding. A total of 200,000 cells were seeded per coverslip in a 24-well plate and cultured overnight at 37°C, 5% CO₂. The next day, when cells reached ~80–90% confluence, they were washed once with PBS and fixed for 15 minutes at room temperature with 4% paraformaldehyde/PBS. Fixed cells were washed twice with PBS and then incubated for 1 hour at room temperature in blocking/permeabilization buffer (PBS containing 0.1% Triton X-100 and 5% BSA). Cells were then incubated overnight at 4°C with rabbit anti-Cx43 primary antibody (Cell Signaling Technology, clone E7N2R, #83649) diluted 1:250 in blocking buffer. The following day, cells were washed three times with PBS and incubated for 1 hour at room temperature in the dark with Alexa Fluor 488-conjugated goat anti-rabbit IgG secondary antibody (ThermoFisher) diluted 1:250, and Phalloidin-Fluor 555 Plus (ThermoFisher) diluted 1:400 in blocking buffer. After staining, cells were washed twice with PBS, followed by a single rinse in Milli-Q water, and mounted with DAPI Fluoromount-G (ThermoFisher).

Images were acquired on a Leica TCS SP8 confocal microscope using a 100x/1.40 NA oil immersion objective. Z-stacks were acquired at 0.6 μm intervals across the full cell height. Acquisition parameters were kept constant across samples. For visualization, images were generated using the free Imaris Viewer (Oxford Instruments).

Enzyme-linked immunosorbent assay (ELISA)

Lentivirus-infected PDX CSC models were used 2 days after selection. Cells were washed with PBS, lysed, and prepared according to the manufacturer’s instructions. Total WNK1 levels were determined using the Human Phospho-WNK1 (T60) & Total WNK1 ELISA Kit (RayBiotech, PEL-WNK1-T60-T). Absorbance was measured on a Victor Nivo multi-mode plate reader.

Quantitative reverse-transcriptase polymerase chain reaction (qRT-PCR)

RNA was isolated from cells via the RNeasy kit (Qiagen), and concentrations were obtained via a NanoDrop Spectrophotometer. cDNA was subsequently synthesized via qScript cDNA synthesis reagent (Quanta Biosciences) with concentrations measured via the NanoDrop Spectrophotometer. cDNA was diluted to 12.5 ng/μl in accordance with TaqMan (ThermoFisher Scientific) protocols. qPCR was performed on a QuantStudio 3 (Applied Biosystems) thermocycler using TaqMan reactions for Cx43 (GJA1; ThermoFisher Scientific, Hs00748445), WNK1 (ThermoFisher Scientific, Hs01013332), and C-MYC (ThermoFisher Scientific, Hs00153408) and TaqMan Universal PCR Master Mix (ThermoFisher Scientific, 4304437). Data were analyzed using the ΔΔCt method and normalized to GAPDH (ThermoFisher Scientific, Hs03929097).

RNA sequencing

RNA from DI318 CSCs was isolated via RNeasy kit (Qiagen) and submitted to MedGenome for library prep and RNA sequencing using the Illumina TruSeq stranded mRNA kit. RNA-sequencing data for T4121 and T3691 was retrieved from GSE79736⁴¹. DI318 RNA sequencing was deposited into the Gene Expression Omnibus (GEO) under accession number GSE272349.

Differential gene expression analysis

Raw gene count data were analyzed using DESeq2 (v.1.44.0) in R (4.4.1). The dataset comprised samples from three experimental groups: control (non-targeting shRNA), WNK1 knockdown (sh491 and sh920), and Cx43 knockdown (sh73 and sh77). Differential expression analyses for the 3 replicates per shRNA or control condition were performed for the two WNK1 knockdowns vs. control and the two Cx43 knockdowns vs. control conditions. The Wald test was used to identify differentially expressed genes, with statistical significance threshold of a Boneferroni-corrected p < 0.05.

Gene set enrichment analysis

Gene Set Enrichment Analysis (GSEA)⁹⁴ was conducted on ranked genes differing between WNK1 knockdown and control and between Cx43 knockdown and control conditions using the fgsea package (1.30.0) in R. The Hallmark gene set collection (MSigDB - h.all.v2024.1.Hs.symbols.gmt) was used with 1000 permutations.

QUANTIFICATION AND STATISTICAL ANALYSIS

All statistical analysis tests were performed in GraphPad Prism 10.2. All statistical details regarding each experiment, including the statistical tests used, the exact value of n, what n represents, and data representation, can be found within the individual figure legends. A p-value < 0.05 was considered statistically significant.

Key resources table.

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Cx43 antibody	Cell Signaling	Cat# 3512S
Cx43 antibody	Cell Signaling	Cat# 83649
WNK1 antibody	Cell Signaling	Cat# 4979S
c-MYC antibody	Cell Signaling	Cat# 9402s
Anti-Actin hFAB Rhodamine Antibody	Bio-Rad	Cat# 12004163
Alexa Fluor 488-conjugated goat anti-rabbit IgG	ThermoFisher Scientific	Cat# A-11008
Phalloidin-Fluor 555 Plus	ThermoFisher Scientific	Cat# A30106
Bacterial and virus strains
Biological samples
Chemicals, peptides, and recombinant proteins
Geltrex	ThermoFisher Scientific	Cat# A1413201
Pierce ECL 2 Western Blotting Substrate	ThermoFisher Scientific	Cat# 80196
Neurobasal medium – phenol red	ThermoFisher Scientific	Cat# 12348017
B27 – vitamin A	ThermoFisher Scientific	Cat# 12587010
Pierce ECL 2 Western Blotting Substrate	ThermoFisher Scientific	Cat# 80196
Neurobasal medium	ThermoFisher Scientific	Cat# 12348017
B27 – vitamin A	ThermoFisher Scientific	Cat# 12587010
EGF	R & D Systems	Cat# 236-EG-01M
FGF	R & D Systems	Cat# 4114-TC-01M
DAPI Fluoromount-G		Cat# 00-4959-52
Bio-Rad protein assay	Bio-Rad	Cat# 5000006
Critical commercial assays
CellTiter-Glo	Promega	Cat# G7572
CaspaseGlo 3/7	Promega	Cat# G8090
Taqman Gene Expression Assay for c-MYC	ThermoFisher Scientific	Cat# 4351370; Assay ID Hs00153408_m1
Taqman Gene Expression Assay for GAPDH	ThermoFisher Scientific	Cat# 4351370; Assay ID Hs03929097_m1
Taqman Gene Expression Assay for GJA1	ThermoFisher Scientific	Cat# 4351372; Assay ID Hs04259536_g1
TaqMan Universal PCR Master Mix	ThermoFisher Scientific	Cat# 4304437
Human Phospho-WNK1 (T60) & Total WNK1 ELISA Kit	RayBiotech	Cat# PEL-WNK1-T60-T
ONE-Step™ Luciferase Assay System	BPS Bioscience	Cat# # 60690
Deposited data
Bulk RNA sequencing for T3691	Xie et al.54	GSE79736
Bulk RNA sequencing for T4121	Xie et al.54	GSE79736
Bulk RNA sequencing for DI318 +/− Cx43 and WNK1 shRNA	This paper	GSE272349
Experimental models: Cell lines
GBM xenograft T4121	Duke University (Jeremy Rich; Guryanova et al.43)	T4121
GBM xenograft T3691	Duke University (Jeremy Rich; Guryanova et al.43)	T3691
GBM xenograft T387	Duke University (Jeremy Rich; Guryanova et al.43)	T387
GBM xenograft T3832	Duke University (Jeremy Rich; Guryanova et al.43)	T3832
GBM xenograft DI318	Mitchell et al.45	DI318
GBM xenograft L1	University of Florida (Brent Reynolds; Deleyrolle et al.48, Siebzehnrubl et al.47)	L1
GBM xenograft L2	University of Florida (Brent Reynolds; Delleyrole et al.48, Siebzehnrubl et al.47)	L2
GBM xenograft GBM23	MD Anderson (Erik Sulman; Silver et al.49)	GBM23
GBM xenograft GBM12	Mayo Clinic (Jann Sakaria; Ramirez et al.50)	GBM12
GBM xenograft T4302	Duke University (Jeremy Rich; Li et al.44)	T4302
GBM xenograft BT124	University of Calgary (Samuel Weiss; Cusulin et al.46)	BT124
MDA-MB-231	ATCC	Cat# HTB-26
LNCaP	ATCC	Cat# CRL-1740
HeLa	ATCC	Cat# CCL-2
293T	ATCC	Cat# CRL-3216
Experimental models: Organisms/strains
NSG mice (NOD.Cg-Prkdcscid Il2rgtm1Wjl/SzJ)	JAX	Strain #005557; RRID: IMSR_JAX:005557
Oligonucleotides
MISSION® pLKO.1-puro Non-Mammalian shRNA Control Plasmid DNA	Sigma-Aldrich	Cat# shc002
MISSION® pLKO.1-puro shRNA for GJA1, sh73	Sigma-Aldrich	TRC# TRCN0000059773
MISSION® pLKO.1-puro shRNA for GJA1, sh76	Sigma-Aldrich	TRC# TRCN0000059776
MISSION® pLKO.1-puro shRNA for GJA1, sh77	Sigma-Aldrich	TRC# TRCN0000059777
MISSION® pLKO.1-puro shRNA for WNK1, sh491	Sigma-Aldrich	TRC# TRCN0000196491
MISSION® pLKO.1-puro shRNA for WNK1, sh918	Sigma-Aldrich	TRC# TRCN0000000918
MISSION® pLKO.1-puro shRNA for WNK1, sh920	Sigma-Aldrich	TRC# TRCN0000000920
Recombinant DNA
psPAX2	Addgene (Didier Trono)	RRID:Addgene_12260; Cat# 12260
pMD2.G	Addgene (Didier Trono)	RRID:Addgene_12259; Cat# 12259
MYC luciferase reporter	BPS Bioscience	Cat# 78628-G
Software and algorithms
Extreme limiting-dilution analysis	Hu and Smyth51	http://bioinf.wehi.edu.au/software/elda/
Molecular Signatures Database	Subramanian et al.55	https://www.gsea-msigdb.org/gsea/msigdb/index.jsp
The Cancer Dependency Map	Tsherniak et al. 63	https://depmap.org/portal
Other