All fluorescent probes were purchased from Thermo Fisher Scientific (USA) and other common chemicals from Sigma Inc. (St. Louis, MO, USA); cyclophilin A inhibitor was from Calbiochem (Cat no-239836). LN229 cell line used here is a p53 mutant, PTEN wild type, and was purchased from ATCC, USA. It is derived from aggressive, metastatic and grade IV (WHO) glioblastoma. It is an often used GBM cell line in acidosis and other glioblastoma-related studies (20–23). Other normal and tumor cell lines used in this study were also from ATCC, USA. All antibodies used in this study are detailed in Supplementary Material.

A series of atomic-scale molecular dynamic simulations were carried on the lipid bilayer comprising of 16 AcGM3 and 146 POPC molecules for three different pHs: 7.4, 6.2, and 3.4. This allowed us to study the conformation and organization of the glycan headgroup of GM3 on bilayer normal at different pHs in a systematic manner. The simulation was performed using the GROMACS molecular dynamics package (for further details, please see Section ‘Methods and Materials’ in Supplementary Material) (24–27).

Briefly, cells were seeded for 16 h in DMEM (high glucose, Invitrogen, Cat No-12100-046) medium with 10% fetal bovine serum (Cat no. RM9955-500mL, HiMedia) and 1X antibiotic–antimycotic solution (HiMedia-A002A). For 8-well chamber slides, seeding density was kept at 8 × 10⁴ cells per well; for 6-well plates, initial cell density was 2.8 × 10⁶, and for 96- and 24-well plates, the seeding densities were 10⁴ and 4.8 × 10⁴, respectively. The cells were subsequently incubated in serum-free DMEM (high glucose) medium with 1X antibiotics for 2 h for serum-derived signal downregulation. The medium was then replaced with DMEM (high glucose) supplemented with 100 ng/ml of recombinant epidermal growth factor and adjusted to pH 7.4, 6.2, and 3.4. pH of the medium was adjusted with 2 N HCl. Cells were incubated at the specified pH units for 3 h, and then experimental sets were either treated with 10 µM methyl-beta-cyclodextrin (+veCD condition) or were left untreated in the respective pHs (−veCD condition). Cells were given these treatments for 2 h and were either then fixed with 1.5% PFA for immunocytochemistry or were processed for other assays as described in Supplementary Material. Each condition in the experimental sets (−veCD and +veCD) was kept in triplicate for the derivation of statistics and significance of data.

All above experiments were performed in triplicates. Each independent experiment had three technical replicates. Error bars are indicative of average SDs obtained across three independent experimental sets. The p values were obtained through Bonferroni’s t-test between pH 7.4 units −veCD condition and other pH conditions and were represented as *p ≤ 0.05, **p ≤ 0.01, and ***p ≤ 0.001, respectively. In individual situations, other comparisons of significance were also indicated. Image analysis was done on over 200 cells in each condition using Fiji image processing software. Calibration bar for LUT converted images were shown in the respective figures.

Please see Supplementary Material for further details on image analysis schema and other common methods used in this study.

Acid stress experiments have been successfully employed to understand cellular responses to both the intracellular (pHi) and extracellular pH (pHe) shifts (28–30).

We chose pH of 7.4, 6.2, and 3.4 units as representatives of physiological (pH 7.4), low (pH 6.2), and very low (pH 3.4) proton concentration ranges for further studies on pH-associated oncogenic vs. non-oncogenic transformations. Briefly, by about 6 h, physiological pH range 7.4 showed well spread out and spindle-shaped morphologies (Figure 2A). Low pH range of 6.2 units showed both spindle-shaped cells and rounded phenotypes (associated with proliferative morphologies) (31, 32) and ‘very low’ pH range of 3.4 units showed progressive cellular aggregations, formation of pseudo-luminal areas indicated by L, and release of small vesicles-indicated by yellow arrows. Cells in pH range 3.4 also showed extensive liberation of micrometer-sized blebs that got detached from the cells and were seen floating freely into the medium, indicating buildup of high hydrostatic pressure at very low pH (cyan and yellow arrows in Figure 2B, white arrows and insets in Figure 2A) (33, 34). It is to be noted that in pH 3.4, the cells remained adhered for 3 days and beyond this, a slow peeling began to take place. However, when the tumor cells growing at low and very low pH were exposed to physiological pH medium, the cells in pH 3.4 showed 100% rapid anoikis, suggesting that very low pH-adapted cells cannot recover to physiological phenotype (Figure 2C). This observation supports the reports that buffer therapy may kill low pH-adapted cells (35). It is to be further noted that tumor cells adapted to pH 6.2, did not show any signs of anoikis in the recovery experiments. When such peeled/anoikis cells (pH 3.4 cultures re-exposed to pH 7.4 medium) were collected and re-plated in the physiological medium, they failed to show any residual growth or anastasis when followed for several days (36, 37).

Pyknotic nuclear morphology, an indicator of apoptotic cell death, was not observed in tumor cells adapted to different pHs (Figure 2D). Lack of ladder formation in DNA fragmentation assay also indicated that cells were not undergoing apoptosis even though cleaved caspase-3 levels were slightly high at pH 3.4 in comparison to other pH ranges (Figures S2A,D in Supplementary Material). Furthermore, cleaved PARP1, cleaved caspase-8, cell senescence, and autophagy markers also did not indicate prominent activation of any of these processes in different pH ranges (Figures S2B,C and S3A–C in Supplementary Material). Note that cleaved caspases were observed to be sequestered near the inner surface of the plasma membrane at low and very low pHs, a localization that cannot induce apoptosis (38).

Localization of LAMP2 to the plasma membrane is recently evidenced to protect breast cancer cells from acidosis and surface hydrolysis (18). It is a new histopathological marker for progressive tumor acidosis, and its expression levels are seen to be much higher in low pH tumor zones. We found a proportionate relocation of LAMP2 to the plasma membrane of LN229 acid-adapted cells (Figure 2E), wherein high surface localization was observed in tumor cells exposed to low and very low pH. Human Protein Atlas (39, 40) data also showed high expression of LAMP2 protein (Figure 2F) in necrotic and peri-necrotic pseudo-palisading zones (cells resident in very low pH). ‘Cellular Tumor’ zone, generally at low pH, showed moderate expression, while the ‘Leading Edge,’ generally at near physiological pH, had low levels. Interestingly, HMGCR, a rate-limiting enzyme in cholesterol biosynthesis, showed high corroboration with the expression trend lines of LAMP2 in different tumor zones, when compared to the same patient samples (Figure 2F, for more patient data, please see Figure S4 in Supplementary Material). Similar, correlation was observed for LAMP2 and SREBF2 (a transcription factor that regulates HMGCR mRNA synthesis), Figure S5 in Supplementary Material. These observations cumulatively suggested that low and very low pH zones may upregulate cholesterol for efficient acid adaptation both in vitro and in vivo.

As mentioned earlier, in physiologically pH-adapted cells, high levels of cholesterol and GM3 were detected on the surface (Figure S1 in Supplementary Material). This adaptation is proposed to prevent the cell membranes from acid hydrolysis, which is akin to mechanism suggested for LAMP2 function in the acid-adapted tumor cells. Hence, we first wanted to understand the role of cholesterol in tumor acid adaptation and cell fate transformations and then follow-up on GM3 glycosphingolipid.

TCGA mRNA transcript data from 538 GBM patients showed high upregulation of cholesterol synthesizing genes (Figure 2G, see more than twofold increase in expression levels of HMGCR and SREBF2). Further, significantly high expression and colocalization of LAMP2 and surface cholesterol were detected in necrotic and pseudo-palisading zones of GBM patient samples, followed by moderate expression and colocalization in ‘Cellular Tumor’ zone vs. the leading edge areas (Figure 3).

This initial analysis, hence suggested that low and very low pH zones may transport more cholesterol to the surface for acid adaption and survival.

As discussed in the earlier section, there may exist a cross talk between the low pH interfacing cell surfaces and high surface cholesterol to shield tumor cells from acidosis. A significant rise in cholesterol was also observed when LN229 glioblastoma cell line was exposed to lower pH ranges of 6.2 and 3.4 units (Figure 4A; −veCD). The plasma membrane cholesterol quantitation was performed via imaging of signal from cholesterol binding nystatin dye that was incubated on the surface of (i) live cells (kept on ice, to prevent endocytosis; cells were fixed after staining and imaged) and (ii) also post-fixation (Figure S6B in Supplementary Material). Similar results were obtained when measurements were made through fluorescence spectroscopy (Figure S6C in Supplementary Material).

When the glioblastoma cells in different pH microenvironments were further perturbed with a very low concentration of methyl-beta-cyclodextrin (+veCD 10 µM) (37), a significant rise in cholesterol from its initial values was observed at all pH ranges (Figure 4A). This may be a feedback response to the initial loss in cholesterol by CD treatment, highlighting the need to maintain cholesterol levels in pH-graded tumor microenvironments. Hence, by this strategy, we were able to generate tumor cells with enhanced levels of cholesterol. This enabled us to study the effects of much higher levels of cholesterol on cell fate competencies in response to differential proton concentrations.

Cholesterol levels crucially determine surface rigidity/anisotropy (41, 42). Surface rigidity is known to determine tumor cell fates (43). The low (6.2) and very low pH (3.4) microenvironments in the unperturbed cholesterol condition (−veCD) showed enhanced surface anisotropy or rigidity, and this was further elevated in the high cholesterol conditions (+veCD) (Figure 4B).

Higher cholesterol levels increase the substrate adhesion energies for efficient mechanical load dissipation, and the gel contraction assay suggests that indeed high cholesterol pH 6.2 +veCD gels contracted more and hence were far competent in the dissipation of the mechanical force caused due to ECM–tumor cell interactions (Figure 4C) (44). Of note was that gels in pH 3.4 microenvironments also contracted moderately, again suggesting that the cells in this pH condition were not dead but were under high membrane tension and could manage to moderately release tension by contracting. Further, gels devoid of cells, failed to contract in pH adjusted medium, confirming that contractility was produced by cells and not by the medium.

Surface rigidities and mechanical homeostasis directly determine tumor metabolic milieu (45, 46). The surface mechanical states are communicated to the chromatin via nuclear translocation of several surface resident mechanosensitive transcription factors that then enable appropriate adaptive modulations in cellular metabolic states. The cortical actin also relays the information on the surface stiffness to the chromatin through stress fibers that are focally linked to the nuclear envelope. This further enables adaptive modulations in metabolic gene expressions (47). Besides, the surface stiffness also induces conformational changes in several metabolic enzymes to prevent their degradation. Surface rigidities crucially determines the trafficking of channels that maintain the pHi of the tumor cells to effect the functions of energy metabolism-associated proteins and enzymes, even in low extracellular pH conditions (48–51).

We find that low and very low pH microenvironments in −veCD and +veCD conditions showed higher intracellular alkalinity (Figure 5A) in comparison their respective external pH environments. Since most of the glycolysis-associated enzymes become inactive at very low pHi; this rise in intracellular pH was particularly important adaptation for tumor survival in low pH conditions. Surprisingly, cAMP, a known soluble mediator of bicarbonate channels, also showed high levels in ‘very low’ pH environment; hence, these tumor niches were rather quite successful in maintaining very high pHi and cAMP pools for their survival (Figure 5B) (52, 53). We find that pH 6.2 had high glucose uptake which was significantly increased in higher cholesterol conditions at both pH 7.4 and 6.2 (Figure 5C). Cells at very low pH (3.4 units) had reduced glucose uptake and negligible levels of LDHA, a crucial enzyme in anaerobic glycolysis (Figure 5D). Although the observation on the absence of anaerobic glycolysis-associated marker fits well with increased intracellular alkalinity at ‘very low’ pH, we wondered how tumor cells at pH 3.4 were able to generate high levels of ATP to sustain high levels of cAMP, if LDHA levels were low.

It came to our notice that macropinocytosis is an alternative mechanism by which cells drive excess nutrients from the extracellular microenvironment (54–56), and we observed that this phenomenon was enhanced at both ‘low’ and ‘very low’ pH microenvironments in −veCD conditions and at all pH ranges in +veCD-treated cells (Figure 5E) (as measured by the uptake of dextran–TMR, an assay to detect macropinosomes). Hence, high cholesterol and low pH microenvironments used multiple and heterogeneous pathways for active metabolism in comparison to tumor cells resident in pH 7.4 −veCD (basal cholesterol) condition.

The physicochemical and metabolic remodeling via differential proton concentration may act as a stimulus for generation of heterogeneous tumor cell fate, a possibility that was further explored.

The BrdU-labeling index showed that high cholesterol tumor cells were proliferation competent and the high cholesterol pH 6.2 condition was most proliferative (Figure 6A). However, to further examine whether different pH ranges also impact substrate anchorage-independent proliferation and growth (that gradually evolves stem-like cells), we ran the soft agar anchorage-independent growth assay (Figure 6B). Gliomaspheres incubated in the unperturbed cholesterol physiological pH condition were 100% anchorage independent and lower pH treatments flattened the spheres, which de-promoted long-term anchorage-independent growth. The −veCD pH 6.2 condition showed approximately 60% sphere flattening, probably due to an increase in cholesterol-mediated surface rigidity. The +veCD conditions showed heterogeneous behavior with 60–70% flattening of spheres at pH of 7.4 and 6.2 units, respectively. Notably, spheres at the ‘very low’ pH in all conditions showed flattened morphology without any further growth when monitored for three consecutive days. Hence, high cholesterol containing tumor cells, though more proliferative in 2D cultures, were diminished of substrate-independent growth properties vs. the tumor cells growing in the unperturbed cholesterol physiological pH condition (7.4 −veCD).

We further found that Oct4 (Figure 6C), an often used marker for substrate-independent growth, was diminished nuclearly in pH 6.2 −veCD condition and its total levels also dropped in the high cholesterol physiological and pH 6.2 conditions. Oct4 showed reduced cytoplasmic accumulation in +veCD pH 7.4 and 6.2 conditions but high accumulation in the cytoplasm at very low pH in all conditions, a phenotype also observed in the gastric cancers (57, 58) and glioblastoma biopsies (see expression in different tumor zones of GBM patient samples—Figure S7 in Supplementary Material). Reduction in nuclear Oct4 in high cholesterol conditions may be one of the major reasons for diminished anchorage-independent growth. Hence, Oct4 nuclear vs. non-nuclear localization in the response to pH and cholesterol levels may crucially determine differential cell fates. In fact, the cells at very low pH range showed very high upregulation of glial acid fibrillary protein, GFAP, a well-known marker of astrocyte differentiation (Figure 6D; see expression in different tumor zones of GBM patient samples—Figure S8 in Supplementary Material). Also, the distribution and concentration of GFAP protein at this pH were membranous and sub-membranous, while it was predominantly cytoplasmic in the other pH ranges. This suggests that ‘very low pH’ can program tumor cells toward differentiated and non-oncogenic fate.

While high cholesterol pH environment of 6.2 and 7.4 units excelled in proliferation and pH 3.4 in differentiation-like status, the wound healing assay showed a different ranking in migration competence of pH-graded microenvironments, with very low pH conditions being highly migration incompetent (Figure 6E). Tumor cells at physiological pH were most migration competent closely followed by cells growing in high cholesterol physiological pH microenvironments. However, tumor cells at pH 6.2 units migrated more slowly than physiological pH exposed cells (Figure 6E).

The migration competence showed high coroborration with enhanced colocalization of cortactin and vinculin at the focal adhesion structures in pH 7.4, as indicated by Pearson’s coefficient (R) (Figure 6F, white arrows, required for efficient migration). High R values indicated that the pH 7.4 microenvironment was intrinsically highly ‘metastasis competent.’ These observations are much supported by the evidence that tumor margins, essentially at near physiological pH are most migration competent (59, 60). Similar corroboration was derived from the other components of the focal adhesion such as the α-actinin–F-actin association, the levels of surface integrins, and its colocalization with the underlying alpha-tubulins and juxtapositioning of Rac1 to the inner leaflet of the plasma membrane, all of which generate an efficient migration competence (Figure S9 in Supplementary Material).

Knowing that cholesterol levels in different pH microenvironments can crucially impact cell fates and growth, we decided to deplete the surface cholesterol from acid-adapted and non-acid-adapted tumor cells by treatment with 1 mM methyl-beta-cyclodextrin (MBCD/CD). Unlike 10 µM CD that enhances surface cholesterol levels by a feedback mechanism, 1 mM and above concentrations of CD are shown to sequester cholesterol from surface permanently, in a dose-dependent manner (61). However, 1 mM MBCD/CD treatment is shown to be mild in normal cells (also tested by us, Figures 7D,E). We find (Figures 7A–C) that tumor cells resident in low pH (6.2) and very low pH (3.4) microenvironments began to peel off within 30 min of treatment and by about 1 h all the cells in pH 3.4 microenvironment had peeled off and showed necrosis. By about 3 h, all the cells in pH 6.2 microenvironment showed the same effect while cells growing in physiological pH were unharmed. We collected the floating necrotic cells and upon centrifugation, pelleting and resuspension in the physiological medium, we re-plated the suspension in plastic dishes and on the soft agar. We did not find any evidence of residual cell revival/anastasis or recovery upon monitoring the cultures for next 7 days. Hence, the depletion of surface cholesterol in acid-adapted cells removed the protective shield and made cells vulnerable to acid-mediated toxicity and permanently killed them by anoikis. The very low pH (3.4)-adapted cells showed more rapid vulnerability to this treatment vs. the low pH (6.2)-adapted tumor cells.

Since physiological pH-adapted tumor cells were unharmed by moderate cholesterol depletion (Figure 7A), we questioned whether GM3 glycosphingolipid, which is also seen to be high in acid-adapted cells, can be exploited to eradicate tumor cells in all pH microenvironments, including cells growing at the physiological pH.

We began by running atomic level molecular dynamic simulations of GM3–POPC in an asymmetric bilayer, to understand whether GM3 can respond to microenvironmental pH levels and thereby can act as a pH sensor and cell fate regulator. The density of hydrogen (H) bonds [Figure 8A: a–c (iii) shown as yellow dotted lines] in the outer leaflet containing GM3 was found to be progressively increased with a decrease in pH (from physiological: pH 7.4, low pH: pH 6.2, to very low: pH 3.4). The H-bonds at pH 6.2 and 3.4 showed reduced vertical projection [Figure 8A: a–c (iii)] probably due to compactness of H-bonds. GM3 at the physiological pH, in contrast, basically had more vertical projections from the surface. This explanation was supported by the observations that glycan headgroup mainly stuck out of the plane of the bilayer at the physiological pH [Figure 8A: a (ii), see white arrow] which gradually tilted toward the bilayer with the decreasing pH values [Figure 8A: b,c (ii) indicated by white arrows]. The x–y projection of the simulations showed that due to increased H-bonding between glycans of GM3, there was an evolution of dimers at pH 6.2 and more clustered organization at pH 3.4 (Figure 8A: e vs. f,g). A detailed analysis of the H-bonds between different glycans of the neighboring as well as self in GM3 revealed that GM3 H-bonds were predominantly established by the sialic acid [N-acetylneuraminic acid (NANA)] moieties at pH 6.2 and 3.4, followed by intermolecular interaction via galactose (Gal) (see Table 1). However, at pH 7.4, the H-bonds were formed predominantly with water and the total numbers of intermolecular H-bonds per nanosecond were particularly low.

The results obtained by us in bi-component membrane simulations at pH 7.4 are much in concert with the recent report on MSD data of GM3 clusters on complex lipid membranes (62). However, since our focus was also on the GM3 conformational patterns at the lower pH units, our MSD results suggests that acidic pHs can trigger a lipid shielding effect of Gal and NANA sugar moieties toward the POPC bilayer.

To further understand the basis of competence of different pHs in evolving GM3 receptor heterogeneity by local spatial changes in its glycan headgroups, we studied the tilt angles (Figure 8B; Figure S10 in Supplementary Material). While preferred tilt angle of glycans at pH 7.4 peaked at 50°, it peaked to 62° and 69° in pH 6.2 and 3.4, respectively. This suggests that at low pH, sialic acid can tilt more to find its surrounding binding partners. A look at the net effects of this pH-driven GM3 glycan conformational heterogeneity on its acyl chain orientation, that guides the membrane thickness, suggests that the membranes were progressively thicker at the lower pH units, indicative of stretched acyl chains in the membrane (Figure 8C).

We further investigated the compression of lipids per unit area in response to the decreasing pH. Results show that ‘area per lipid’ was reduced at low and very low pHs, suggesting an evolution of high lipid packing density. Hence, enhanced lipid packing at low pH and very low pH indicated the probability of high compressional stress on the membrane (Figure 8D).

GM3 is highly expressed in gliomas and is shown to be a crucial component of the migratory complex (63, 64). Immunohistochemistry from glioma patient revealed that GM3 is highly expressed in malignant glioblastomas and was well colocalized with cholesterol-rich regions in the WHO grade IV (Figure 9) cancers in both pediatric and adult patients vs. grade II (Figure S11 in Supplementary Material, see table embedded in the figure for comparative R values). GM3 was highly enriched in the necrotic and luminal zones where it showed large punctate/clustered patterns on the cellular surface (phenotypes designated by numbers 6 and 3) (Figures 9B,C, white arrow points to very low pH-associated nuclear morphology as observed in Figure 2D). However, the non-luminal zones showed differential patterns where either a mix of large and small puncta were observed all over the cells (a phenotype designated by number 5) or small punctate and polarized clustered assemblies were noticed, suggestive of a polarized migratory phenotype (a phenotype designated by number 4) (Figures 9D,E). Further, small phase separated clusters were observed in some tumor cells (phenotypes designated by numbers 1 and 2). All these patterns of GM3 were well correlated with the expression of cholesterol in the tumor cells and also with GM3 and cholesterol co-expression patterns in the human kidney and the gastric tissues which are the naturally acid-adapted organs (Figure S1 in Supplementary Material). Hence, it appears that in a very low acidic environment, the tumor pH induces conformational changes in GM3 and simultaneously upregulates cholesterol to induce a lipid shielding effect against harsh acidic environment, which is generated by the extrusions of proton from the nutritionally starved and over-active tumor cells.

We were rather excited to see that the patterns obtained on the GBM patient tumor tissue sections strongly corroborated with the GM3 surface organization in our pH range in vitro studies (Figures 10A,B).

Overall, the unperturbed cholesterol conditions (−veCD) at physiological and low pH had reduced size and number of clusters, in comparison to the high cholesterol pH conditions, but pH 6.2 (−veCD) showed more number of clusters than pH 7.4 (−veCD), which was also predicted by our simulation data. Both GM3 cluster size and cluster fluorescence intensity that essentially denoted the number of expressing pixels in a cluster were found to be highly correlated in each condition (Figure 8B).

The high cholesterol tumor cells (+veCD) at physiological pH 7.4 showed numerous small sized GM3 clusters (0.17 μm²) that were organized in a polarized pattern. However, the pH 6.2 exposed cells had comparatively fewer punctate clusters in the range of 0.17 μm² but rather showed some larger sized clusters (0.34 μm² and above), probably due to the high levels of cholesterol in these conditions. The few but larger clusters observed in pH 6.2 +veCD condition could have exerted a high compressional force on the tumor cells, forcing a spherical mitotic geometry, which might be the reason for the observed highest proliferative index of this condition (Figure 6A). The very low pH 3.4 exposed cells showed an evolution of very large sized clusters that almost appeared to cover the entire cell surface. Such very large numbers and sizes of GM3 clusters can exert very high compressional stress on tumor cells (as predicted in MSD data) which could have been responsible for the growth arrested phenotype at this pH.

The GM3 cluster size and intensity were highly diminished at all pH conditions from their original values in HQ (cytoskeleton disrupter) treated cells, suggesting that the major micron-sized GM3 clustering was cytoskeleton dependent. Since +veCD conditions also showed enhanced clustering, it can be assumed that micron-sized clusters were cholesterol and cytoskeleton dependent.

We confirmed that the clustering patterns obtained were not due to the cluster detection assay performed at a low temperature because in experiments where the cells that were first fixed at the room temperature and then probed for surface GM3 also showed similar surface patterns as seen when the GM3 antibody was incubated on the surface of live cells that were kept on ice, to prevent antibody endocytosis (Figure S12A in Supplementary Material). When live cells in different conditions were treated with saponin to deplete the surface cholesterol and then probed for surface GM3, the GM3 clustering pattern was highly reduced in all conditions at all pH units, again suggesting that clustering was sensitive to the levels of cholesterol (Figure S12B in Supplementary Material). We find that pH and cholesterol-driven GM3 differential clustering was specific and its precursor lipid species, lactosylceramide or another ganglioside SSEA4 (65) did not exhibit this type of clustering phenomena when exposed to the similar conditions (Figures S12C,D in Supplementary Material).

We further considered that the cholesterol-independent GM3–self-glycan interactions, as seen in lactonized forms, may also exist in low pH conditions, at least to some extent, and may contribute to observed GM3 clustering (17, 66). For this, we investigated the extent of contribution of low pH-induced GM3 lactones (intra-glycan bonding between sialic acid and other sugar moieties) vs. GM3–GM3 glycan ligated clusters in our pH conditions. Data showed that GM3 lactonization was rather minimal as the digestion of the sialic acid by sialidases, dramatically reduced surface GM3 detection at all pHs, which would not have been the case if detected clusters were predominantly lactones, as GM3 lactones are resistant to such enzymatic treatment (Figure S13 in Supplementary Material) (67).

From the insights gained, we wanted to further test whether incubation of LN229 GBM cells with multivalent anti-GM3 IgM antibody can partially mimic the GM3 clustered phenotype as observed at the very low pH. We wanted also to test whether such treatment can enable differentiation of the tumor cells in the −veCD and +veCD, pH 7.4 and 6.2 oncogenic conditions.

For this, we incubated the confluent tumor cell cultures with 20 µg of antibody in 300 µl of medium per well in the eight-well chamber slides. In 24 h, no visible morphological effects were observed over the isotype antibody control. When the next antibody shot was given, within 1.5 h, we began to observe massive cellular blebbing that was further enhanced by the next 3 h (Figure 11A, DIC image panel). The blebs were akin to what were seen at very low pH in our acid stress test (Figure 2B), but the number of levitated blebs and cells were far more in any field of microscopic view suggesting a buildup of high membrane tension which was released by the cells via blebbing.

The antibody-incubated cultured cells were fixed and probed for GM3 surface clusters, and the impact of this organization on the cortical and intracellular actin networks as well as on the status of differentiation-associated gene ‘GFAP’ were examined. The cells were fixed within 26 h of antibody treatment (for immunostaining-based analysis) as almost 100% anoikis was observed in 50–60 h post-treatment.

We find that in blebbed conditions (Figure 11), GM3 clusters were formed and were akin to those seen at the very low pH in the acid stress experiments (Figure 10A). These results also suggest that anti-GM3 antibody may have generated GM3 tilts akin to those observed in our simulation studies for the very low pH condition.

The cytoskeleton too was similarly disrupted as observed in the acid stress experiments (Figure S9 in Supplementary Material), and an appreciable increase in GFAP staining was seen (Figure 11). However, the GFAP appeared to be bundled and accumulated in certain regions of the cells, which was not the case when experimental conditions were incubated with isotype-specific IgM antibody alone (Figure S14 in Supplementary Material). These clustered GFAP rich regions were found to be intriguingly enriched in GM3 clusters that were generated by the anti-GM3 antibody-mediated cross-linking. Hence, the cells could be effectively forced into a differentiated state via the GM3 antibody-mediated clustering at both physiological and low pHs. The loss of protumorigenic receptors and signaling lattices through extensive surface blebbing could be one of the possible causes for this differentiation-like state.

Indeed, necrotic and peri-necrotic zones (associated with low pH environments) in glioblastoma patient samples showed extensive GM3 and GFAP colocalization and supraclustering (Figure 12), which was not seen in the non-necrotic zones of the tumor mass (see R values). The clusters in the non-necrotic zones were small and spread throughout the cells or were small and polarized. Hence, these results show that not just GM3 but GM3 supraclustered conformations must be associated with the tumor cell death.

Inhibition of cyclophilin binding to its surface receptors by cyclosporin A has been evidenced to impact GFAP expression (68). Cyclophilin A is found to be highly expressed in gliomas (Figure S16 in Supplementary Material, GBM patient data on CyPA expression). In our acid stress experiments, cyclophilin A (CyPA) release or secretion was found to be highly reduced at the very low pH and GFAP expression was enhanced (Figure S15 in Supplementary Material; Figure 6D).

We observed that a potent cyclophilin A inhibitor (69) at a very low concentration of 5 nM could inhibit CyPA release and induced appreciable tumor cell blebbing (Figure S17 in Supplementary Material, DIC image inset panels). This was accompanied by an increase in membranous GFAP and GM3 supraclustering; however, the cytoskeleton did not show an extensive rupturing as seen with anti-GM3 antibody treatment. The inhibitor treated cells showed massive stress fibers, indicative of high surface mechanical stress (Figure S17 in Supplementary Material). It is to be noted that the inhibitor treatment was observed to increase the intracellular concentration of cyclophilin A, suggesting an inhibitory action on the extracellular release of CyPA. The time frame for the recording of this observation was as follows: the first inhibitor shot: 24 h; second shot: 3 h.

To enhance the supraclustering effects, we incubated the anti-GM3 antibody and cyclophilin A inhibitor together and found far prominent cell surface blebbing accompanied by an extensive rupture of the cytoskeleton (Figure 11). In this combination treatment, the GFAP expression was more enhanced than in the conditions with the anti-GM3 antibody alone and GFAP showed both surface anchored and sub-surface accumulation in the corresponding GM3 clustered areas.

As GFAP is a membranous and also cytoplasmic cytoskeletal protein, high surface mechanical stress probably influenced its integrity; however, GFAP enhanced expression indicated that cells might have switched their fate to more differentiated phenotypes. Thus a combination of anti-GM3 antibody and cyclophilin A inhibitor was more potent than anti-GM3 antibody alone in first causing cell differentiation and then forcing it to die by anoikis over the course of time (about 30–50 h post treatments). We further find that other GBM cell lines, U87MG and U373, when treated in a like manner, showed similar anoikis phenotype while normal cells such as epidermal keratinocytes (HaCaT), mouse astrocytes, and neurons were unaffected (Figures 13A–C). Hence, indeed manipulation of surface GM3 via cross-linking served as an effective therapeutics to eradicate not only low pH zones but also the tumor cells that were growing in physiological pH microenvironments. This combination treatment hence converged the tumor cell heterogeneity into anoikis zones (Figure 13D).