Solute exchange through gap junctions lessens the adverse effects of inactivating mutations in metabolite-handling genes

Microarray data from 79 CRC lines (Wilding et al., 2010) were analyzed for the expression of connexin (Cx) genes. Expression in a subset of CRC cells was detected for GJA1 (coding for Cx43), GJA3 (Cx46), GJB1 (Cx32), GJB2 (Cx26), GJB3 (Cx31), GJB5 (Cx31.1), and GJC1 (Cx45) (Figure 1A). Gaussian mixture modeling (GMM) determined that the distributions of GJA1, GJA3, GJB1, GJB5, and GJC1 message among CRC lines were bimodal, such that cell lines could be grouped as high- or low-expressing. In contrast, messages for GJB2 and GJB3 were unimodally distributed among the cell lines. Analysis of TCGA datasets confirmed bimodality for GJA3, GJB5, and GJC1 expression (Figure 1—figure supplement 1). Single-cell transcriptomics (Fawkner-Corbett et al., 2021) indicated expression of GJB1, GJB2, and GJB3 in normal colonic epithelial cells (Figure 1B). GJB2 and GJB3 had previously been identified in normal colorectal epithelium (Sirnes et al., 2015; Dubina et al., 2002). Overall, GJB2 and GJB3 are constitutively expressed in CRC cell lines, whereas GJA1, GJA3, or GJC3 were found in a subset of lines, presumably due to mutations or stable methylation differences in their promoter regions. Two-by-two table analysis for bimodally distributed connexin genes indicated positive correlations between GJA1 and GJC1, and negative correlations for pairs GJB1-GJA1, GJB1-GJB5, and GJB1-GJC1 (Figure 1C; see also principal component analysis in Figure 1—figure supplement 2).

Figure 1 with 2 supplements see all

Download asset Open asset

Figure 1—source data 1

Full-length scans of blots for Figure 1F.

https://cdn.elifesciences.org/articles/78425/elife-78425-fig1-data1-v2.xlsx

Download elife-78425-fig1-data1-v2.xlsx

Figure 1D shows the expression of connexin genes, ranked by combined GJB2 and GJB3 message. Fourteen lines were selected for further functional measurements (Figure 1E). Since GJA1 expression accounted for considerable variation, cell lines were categorized as GJA1-high and GJA1-low. The GJA1-low group could be subdivided further as GJC1-low and GJC1-high. This classification was confirmed at protein level using antibodies against Cx43 (GJA1) (Figure 1F). RKO cells were chosen as a negative control because of negligible expression of major Cx genes.

Solute diffusion between cells was inferred from fluorescence recovery after photobleaching (FRAP) of fluorescent calcein (Swietach and Monterisi, 2019). A cell in the middle of a coupled cluster was bleached to ~50% of resting signal, and the subsequent recovery of fluorescence (due to dye ingress from neighboring cells) indicated the degree of coupling. Fluorescence recovery was detected in GJB2-expressing SNU1235 but not in connexin-negative RKO cells (Figure 2A). FRAP recordings were quantified in terms of an apparent permeability constant for calcein (Figure 2B), and the role of Cxs was confirmed from the inhibitory effect of carbenoxolone (100 μM), a broad-spectrum Cx blocker (Figure 2C). The acquired images supply data on cell geometry and time courses of calcein fluorescence for estimating permeability to calcein, albeit with assumptions detailed in Appendix 1. Cell-to-cell calcein permeability correlated best with GJB2 expression (Supplementary file 1). Moreover, GJB2 knockdown ablated functional coupling, whereas siRNA against GJB3 did not (Figure 2D; see Figure 2—figure supplement 1 for knockdown confirmation). In GJA1-high Caco2 cells, functional coupling was ablated with siRNA against GJA1 (Figure 2E; see Figure 2—figure supplement 2 for knockdown confirmation). Note that there was a trend for GJB3 knockdown to increase overall connectivity in DLD1 and SNU1235 cells, which may represent compensatory changes in the expression of other connexins or closeness of cell–cell contacts. Western blot analysis indicated some degree of interaction between the expression of particular pairs of connexins, but the effects were, at best, modest (Figure 2—figure supplement 1).

Figure 2 with 2 supplements see all

Download asset Open asset

Figure 2—source data 1

Full-length scans of blots for Figure 2H, I and K.

https://cdn.elifesciences.org/articles/78425/elife-78425-fig2-data1-v2.xlsx

Download elife-78425-fig2-data1-v2.xlsx

Functional estimates of coupling correlated well with levels of Cx26 protein at cell-to-cell contacts (Figure 2F). Furthermore, Cx43 was detected at cell-to-cell contacts in C10, LOVO, and Caco2 but not in RKO, SW1222, or HT29 cells (Figure 2G). Unlike Cx26 staining, which was uniform among cells, Cx43 immunofluorescence appeared more heterogenous, with only some cells in a monolayer appearing strongly positive. To confirm the co-existence of Cx43-positive and negative sub-populations, LOVO and Caco2 were FAC-sorted by Cx43 status, expanded, and then analyzed by Western blot. This analysis identified distinct Cx43-positive and -negative sub-populations, indicating that not all cells of a nominally GJA1-positive line may benefit from Cx43-dependent connectivity (Figure 2H).

Ablation of GJB2 expression by guide RNA (gRNA) or siRNA (Figure 2I) eliminated Cx26 immunofluorescence (Figure 2J). GJB2 KO clones of DLD1 cells, generated by CRISPR/Cas9 (Figure 2K), had dramatically reduced diffusive exchange (Figure 2L). GJB2 knockdown did not affect Cx31 or Cx43 expression patterns in DLD1 cells (Figure 2—figure supplement 1), arguing against compensatory transcriptional responses. In summary, Cx26 channels provide a principal route for cell-to-cell communication in most CRC lines, and the level of this protein is expressed relatively uniformly among cells.

FRAP-based measurements can identify the Cx channels that underpin permeability but cannot predict the extent to which solutes equilibrate at steady state. The latter was evaluated from the degree of fluorescent dye exchange in co-cultures prepared from two populations of cells loaded with spectrally distinct CellTracker fluorophores. The diffusive properties of CellTracker dyes, determined by FRAP (Figure 3A), correlated inversely with molecular weight (Figure 3B). CellTracker Violet and Orange were selected for co-culture experiments (Figure 3C). Monolayers were imaged sequentially for Violet or Orange in confocal mode. In control monolayers prepared with one dye only, fluorescence bleed-through between channels was minimal (Figure 3D). To aid visualization, CellTracker Violet and Orange fluorescence images were pseudocolored as blue and red, respectively, such that pixels containing both dyes (i.e., following diffusive exchange) appear purple (Figure 3D). In eight CRC lines spanning a range of GJB2 expression, dye exchange was confirmed, particularly in areas of high cell confluency (Figure 3E) which points to a role of cell-to-cell contacts in exchange. As confirmation that dye mixing was due to exchange across cell-to-cell contacts, co-culture experiments were repeated on cells seeded at very low density, which reduces the incidence of cell-to-cell contacts. As expected, a lower incidence of cell-cell contacts resulted in reduced dye-exchange (Figure 3F). Furthermore, connexin-negative RKO co-cultures did not show evidence for significant mixing between cells (Figure 3F).

Figure 3 with 1 supplement see all

Download asset Open asset

Figure 3—source data 1

Permeability data from FRAP experiments.

https://cdn.elifesciences.org/articles/78425/elife-78425-fig3-data1-v2.xlsx

Download elife-78425-fig3-data1-v2.xlsx

Figure 3—source data 2

Coupling coefficient data from CellTracker exchange experiments.

https://cdn.elifesciences.org/articles/78425/elife-78425-fig3-data2-v2.xlsx

Download elife-78425-fig3-data2-v2.xlsx

The degree of dye-mixing arising from cell-to-cell exchange was quantified in terms of a coupling coefficient. This pixelwise analysis measures the fluorescence emitted from each of the two CellTracker dyes as a fraction of total fluorescence and reports the lower of these two numbers as the coupling coefficient (see Supplementary file 1 for details). Thus, before dye exchange takes place, the coupling coefficient is zero; if dye exchange takes place across gap junctions, the coefficient increases as cells become dually fluorescent. In mono-cultures, the coupling coefficient was negligible, as expected from a control experiment in which only one dye is present (Figure 3—figure supplement 1). In contrast, co-cultures produced significant coupling coefficients in the case of cells expressing functional gap junctions seeded at adequate density to form close cell–cell contacts (Figure 3G). Connexin-negative RKO cells, DLD1 GJB2 KO cells, or DLD1 and OXCO1 cells plated at low density produced low coupling coefficients. Strikingly, there was no correlation between FRAP-measured calcein permeability and coupling coefficient determined at steady state for CellTracker dyes. Thus, even monolayers expressing low-conductance gap junctions may produce a uniform distribution of diffusible substances at steady state. Coupling coefficients showed considerable heterogeneity when plotted as a frequency distribution, with weakest coupling in RKO cells and low-density DLD1 and OXCO1 monolayers (Figure 3H). In the other cell lines tested, around half of the cells showed evidence of significant coupling. This number is likely to underestimate the extent of coupling in tissues, where contacts can be established across three dimensions.

Some degree of dye leakage from cells is inevitable during long-term culture, but this is unlikely to result in the transfer of dye into neighboring cells. Firstly, CellTracker dyes are chemically rendered to be membrane-impermeable inside cytoplasm, thus any molecules that leak across damaged membranes cannot (re)enter neighboring cells. Secondly, the volume of extracellular space is far greater than the intracellular volume, thus any released dye would become vanishingly diluted outside cells. For these reasons, the above dye-exchange measurements report the consequences of cell-to-cell diffusion across gap junctions, rather than transfer across the extracellular milieu.

Dye exchange was tested further using flow cytometry (Figure 4A). Co-cultures were prepared using various pairs of spectrally resolvable CellTracker dyes. The gain of detection channels was calibrated using mono-cultures prepared using one dye only. After gating-out doublets, fluorescence was measured on two detection channels. DLD1 cells grown as mono-cultures emitted fluorescence detected on one channel only (Figure 4B). Co-cultures, in contrast, produced a cluster of cells emitting fluorescence in both channels, indicating that dye exchange had occurred during culture. This can be visualized by overlaying pseudocolored density histograms for co-cultures (as green) against the two mono-cultures (as red or blue). To confirm that dye exchange had taken place whilst cells were connected as part of an intact monolayer, control experiments mixed two suspensions of cells harvested from distinctly labeled mono-cultures immediately prior to cytometry (i.e., without a period of culture when cells are coupled to one another). These experiments produced two separate clusters (Figure 4C), confirming that dual labeling arises only in intact co-cultures. Similar experiments were performed in other CRC cells (Figure 4D–G). The extent of dye exchange between coupled cells was calculated to be ~60% (DLD1 or HT29 cells; Figure 4H).

Figure 4

Download asset Open asset

Figure 4—source data 1

Quantification of flow cytometry results.

https://cdn.elifesciences.org/articles/78425/elife-78425-fig4-data1-v2.xlsx

Download elife-78425-fig4-data1-v2.xlsx

In summary, coupled CRC cells can freely exchange fluorescent dyes across connexin channels, which argues that small metabolites are expected to equilibrate across the continuous cytoplasmic compartment. The next series of experiments sought evidence that diffusive exchange can functionally rescue cells carrying a genetic inactivation of a specific metabolite-handling process.

The Na⁺/H⁺ exchanger NHE1, coded by SLC9A1, is a prominent regulator of intracellular pH (pHi), expressed in many CRC cells including HCT116. Genetic inactivation of NHE1 affects pHi control and is detrimental to general cellular physiology (Hulikova et al., 2013). If gap junctions were able to conduct a sufficient flow of H⁺ ions between coupled cells, then it is conceivable for an SLC9A1-deficient cell to have its pHi-regulatory needs serviced by a neighboring WT cell. To test this, two SLC9A1 KO clones were generated using CRISPR/Cas9 (Figure 5A). Confirmation that these KO cells lack NHE1 activity was sought in monolayers loaded with the pH-reporter dye cSNARF1. In WT HCT116 cells, NHE1 activity was measured from the rate of pHi recovery following an acid-load (ammonium prepulse) in the absence of CO₂/HCO₃^- to eliminate HCO₃^--dependent transporters. NHE1 activity was confirmed from the inhibitory effect of cariporide (Figure 5B). KO mono-cultures, in contrast, produced no cariporide-sensitive pHi recovery. To test whether pHi regulation in KO cells could be rescued by coupling onto WT cells, co-cultures were established, wherein KO cells were identified by transfected GFP fluorescence (Figure 5C). pHi recovery, following an ammonium prepulse, was tracked simultaneously in WT and KO cells (Figure 5D). Strikingly, KO cells showed evidence for pHi recovery when co-cultured with WT neighbors, despite the absence of SLC9A1 expression. WT cells in co-culture produced slower pHi recoveries than in mono-cultures because of the additional burden of having to service KO cells. Similar observations were made with either KO clone. These experiments illustrate phenotypic blurring: apparent function is more homogenous than the underlying genotype.

Figure 5 with 1 supplement see all

Download asset Open asset

Figure 5—source data 1

Full-length scans of blots for Figure 5A.

https://cdn.elifesciences.org/articles/78425/elife-78425-fig5-data1-v2.xlsx

Download elife-78425-fig5-data1-v2.xlsx

To test how diffusive exchange impacts steady-state pHi, measurements were performed on cells equilibrated in HEPES/MES-buffered medium over a range of extracellular pH (pHe). In WT cells, the pHe–pHi relationship was linear and not affected by KO of GJB2 (Figure 5E). Steady-state pHi was reduced in KO mono-cultures, as expected from impaired pHi control. However, the pHi difference between WT and KO cells collapsed in co-cultures. This effect was Cx26-dependent because SLC9A1-expressing HCT116 cells that had genetically inactivated GJB2 were unable to raise the pHi of co-cultured KO cells (Figure 5E). Since the readout of pHi regulation was recorded with single-cell resolution, the distribution of pHi among cells was analyzed to seek an effect of coupling on cell-to-cell variation. SLC9A1-deficient cells have weaker pHi control, which manifests as a wider pHi distribution in mono-cultures (Figure 5F). When, however, these KO cells were co-cultured with WT cells, the pHi distribution returned to normal levels. This observation indicates that NHE1 activity in WT cells was able to service the pHi-regulatory needs of KO cells, manifesting as a narrowing of the population-wide spread of pHi. This rescue effect was Cx26-dependent because knockdown of GJB2 resulted in a wider pHi spread (Figure 5F).

To test whether diffusive exchange benefits the growth of SLC9A1-deficient cells, co-cultures were established from different seeding ratios of WT and KO cells. Parallel plates were set-up to record growth terminated at various time points. Total biomass (i.e., growth of WT and KO cells) was measured by the sulforhodamine B (SRB) absorbance assay and growth of the KO compartment was determined from GFP fluorescence. In the case of KO mono-cultures, the GFP and SRB signals measure an equivalent quantity and therefore provide a calibration curve for converting between the two types of measurements. When applied to co-culture experiments, this calibration can convert the measured SRB signal into an expected GFP signal for comparison against the measured GFP signal (i.e. actual growth of the labelled compartment). If WT and KO cells had indistinguishable survival prospects, then the time courses of GFP and SRB should be stoichiometrically related to the seeding ratio and measured GFP should equal expected GFP (i.e. SRB scaled by the seeding density, e.g., by 50% for a 1:1 co-culture). If, however, there is a growth advantage for KO cells in the presence of WT cells, then measured GFP will exceed that predicted from SRB. To test for such rescue, co-cultures were established over a range of seeding ratios. Experiments were performed in media at pH 7.4 or 6.6 to stimulate NHE1 over a range of pH. For co-cultures seeded from 25% KO or 50% KO cells, growth of the GFP-tagged KO compartment exceeded the prediction based on equal WT/KO survival (Figure 5G and H). Similar findings were observed using the second KO line (Figure 5—figure supplement 1). Thus, co-cultured SLC9A1-deficient cells grew faster than expected at the expense of coupled WT cells. Since pHi regulation consumes a substantial amount of energy, this effect could be explained by the exploitation of WT resources by KO cells.

In summary, we show that cells lacking an important ion exchanger at the membrane can hijack the equivalent activity from neighboring WT cells, provided that the relevant ions are able to diffuse across connexin channels.

The next test related to glycolysis, a critically-important metabolic pathway that handles small molecules in cytoplasm. DLD1 cells were selected for these experiments based on their high glycolytic rate. A recent whole-genome CRISPR-Cas9 screen identified ALDOA, a gene coding for the glycolytic enzyme aldolase A, as essential for CRC cell survival under physiological pH (Michl et al., 2022). Ablation of ALDOA using virally transduced gRNAs reduced the expression (Figure 6A) and glycolytic rate, assayed in terms of medium acidification (Figure 6B) and lactate production (Figure 6C). It was not possible to produce a stable ALDOA KO clone, ostensibly because of the enzyme’s critical role for survival, thus experiments were performed within 6 days after transduction with one of the two gRNA constructs.

Figure 6 with 3 supplements see all

Download asset Open asset

Figure 6—source data 1

Full-length scans of blots for Figure 6A.

https://cdn.elifesciences.org/articles/78425/elife-78425-fig6-data1-v2.xlsx

Download elife-78425-fig6-data1-v2.xlsx

ALDOA-deficient DLD1 cells grew four times slower than WT cells (Figure 6—figure supplement 1), most likely due to compromised provision of glycolytic ATP and build-up of fructose-1,6-bisphosphate (the substrate for aldolase A). These metabolites are expected to diffuse freely between coupled cells, thus any fructose-1,6-bisphosphate that accumulates in an ALDOA-deficient cell should diffuse to neighboring cells for processing, whereas ATP will diffuse in the opposite direction. Diffusive rescue would allow ALDOA-deficient cells in co-culture with WT cells to grow faster than expected. To test this, co-culture experiments were performed for various ratios of WT and GFP-labeled ALDOA-deficient cells (Figure 6D). After 6 days of culture at pH 7.4, GFP fluorescence (a measure of ALDOA-deficient cells) and SRB absorbance (total cell biomass) were recorded. If ALDOA-deficient cells received no benefit from being coupled onto WT cells, then the endpoint GFP signal would be proportional to the initial seeding density (Figure 6E, gray line). However, GFP-labeled cells became relatively enriched in co-cultures with WT cells, which was most prominent at 3:1 seeding ratio. Thus, the survival of ALDOA-deficient cells improved when these connected onto WT counterparts. A similar effect was seen using the other gRNA (Figure 6F). Rescue of ALDOA-deficient cells in co-culture with WT cells was absent when GJB2 was knocked out in WT cells with gRNA (Figure 6G), indicating a critical role for Cx26-mediated metabolite exchange in this rescue. As a negative control, co-culturing ALDOA-deficient DLD1 cells with WT RKO cells, a line that does not support significant connexin connectivity, failed to rescue ALDOA-deficient cells (Figure 6—figure supplement 2).

To test whether connexin isoforms other than Cx26 are able to support metabolic rescue, experiments were performed on two GJB2-low cell lines: C10 and NCIH747. C10 cells express Cx43 channels, which produce good cell-to-cell coupling, albeit in a sub-population of cells due to its heterogeneous expression pattern (see Figure 2G). Co-culturing of ALDOA-deficient C10 cells with WT counterparts did not result in significant metabolic rescue (Figure 6H; see Figure 6—figure supplement 3 for knockdown confirmation). This lack of large-scale rescue may relate to the apparent heterogeneity of Cx43 levels across cells, which allows only Cx43-positive cells to benefit from metabolic rescue. NCIH747 cells express isoforms Cx31 and Cx45 but neither were unable to rescue ALDOA-deficient cells in co-culture with WT cells (Figure 6I; see Figure 6—figure supplement 3 for knockdown confirmation).

In summary, we demonstrate that the genetic ablation of a metabolite-handling cytoplasmic enzyme in one cell can be rescued by diffusive access to the required catalysis in neighboring cells via connexin channels. Thus, despite major differences in genotype, the apparent phenotype of KO and WT cells is more similar. Cx26-assembled channels emerge as particularly effective in supporting metabolic rescue.

The third test for metabolite exchange used cells with genetically inactivated mitochondrial respiration on the basis that this organellar process influences the levels of diffusible substances in cytoplasm, including ATP. SW1222 cells were selected for this experiment because of their high respiratory rate (Figure 7B) that could be genetically inactivated by knockout of NDUFS1, a gene coding for a component of complex I (Figure 7A; Michl et al., 2022). Oxidative phosphorylation was blocked in NDUFS1 KO cells, and a compensatory increase in glycolytic rate ensued (Figure 7B). Cells deficient in NDUFS1 grew three times slower than WT cells (Figure 7—figure supplement 1), likely due to ablated mitochondrial metabolism that is critical for delivering substrates and energy. To test whether junctional coupling could rescue NDUFS1-deficient cells, WT and GFP-labeled KO cells were co-cultured in medium at pH 7.7 (i.e. alkaline conditions that activate glycolysis fully; Figure 7C). If KO cells received no benefit from being coupled onto WT cells, the GFP signal measured at endpoint would be proportional to the seeding ratio. However, co-culture with WT cells promoted the growth of GFP-labeled KO cells beyond the expected level, particularly for seeding ratios of 3:1 KO:WT (Figure 7D). This rescue related to the density of cell-to-cell contacts because KO cells could no longer receive a growth benefit in sparse co-culture with WT cells (Figure 7E). Furthermore, the rescue was linked to Cx26-dependent coupling as co-culture with WT cells that had genetically knocked down GJB2 by siRNA yielded no benefit to co-cultured NDUFS1-deficient cells (Figure 7F). Ultimately, the rescue effect of WT cells relates to oxidative phosphorylation, which requires oxygen and should be lessened under hypoxic conditions. Indeed, this was the case in co-cultures incubated in 2% O₂ (Figure 7G).

Figure 7 with 1 supplement see all

Download asset Open asset

Figure 7—source data 1

Full-length scans of blots for Figure 7A.

https://cdn.elifesciences.org/articles/78425/elife-78425-fig7-data1-v2.xlsx

Download elife-78425-fig7-data1-v2.xlsx

The importance of Cx26-assembled channels in providing metabolic rescue to NDUFS1 KO cells was tested in co-cultures with GJB2-low/negative lines: RKO, C10, and NCIH747 (Figure 7H). Connexin-negative RKO cells were unable to rescue NDUFS1-deficient SW1222 cells. GJA1-positive C10 cells, which form Cx43-dependent connections, did not rescue NDUFS1-deficient cells, indicating that Cx26 and Cx43 channels are unlikely to form heterotypic channels (consistent with previous findings Koval et al., 2014). Finally, NCIH747, which produce Cx31 and Cx45, also did not rescue NDUFS1-deficient cells, further supporting the notion that heterotypic channels are less efficacious than Cx26-assembled conduits.

Whilst in vitro assays provide a well-controlled environment for evaluating metabolic rescue, their findings must be verified in vivo, wherein cells are able to connect extensively in three dimensions and grow over many cycles of division. Evidence for connexin-dependent metabolic rescue in vivo was tested in xenografts seeded from a 1:1 mixture of GFP-labeled NDUFS1-deficient (KO) SW1222 cells and DLD1 cells. In this system, DLD1 cells lay down a cellular network for NDFUS1 KO cells to grow in. If metabolic rescue took place, then actively respiring DLD1 cells would promote the growth of the mitochondrially-defective (GFP-labelled) cellular population. Xenografts of WT DLD1 plus NDUFS1 KO SW1222 cells were established subcutaneously on the left flank of mice. As a control, the opposite flank of the mouse was injected with a mixture of NDUFS1 KO SW1222 cells plus Cx26 knockout (GJB2 KO) DLD1 cells (Figure 8A), that is, a combination that cannot establish significant strong DLD1-SW1222 coupling. In DLD1 cells, genetic inactivation of GJB2 had no substantial effect on cell growth (Figure 8B) or ability to form 3-D spheroids (Figure 8C), thus both types of DLD1 cells were deemed suitable for growing xenografts.

Figure 8 with 1 supplement see all

Download asset Open asset

Figure 8—source data 1

Full-length scans of blots for Figure 8A.

https://cdn.elifesciences.org/articles/78425/elife-78425-fig8-data1-v2.xlsx

Download elife-78425-fig8-data1-v2.xlsx

In terms of bulk volume, xenografts grew faster on the left flank, wherein DLD1 and SW1222 cells can establish Cx26-assembled gap junctions (Figure 8D). At the endpoint (a combined left and right tumor burdenof >1 cm³ or 30 days of growth), xenografts were excised for histology and stained with the nuclear dye Hoechst and GFP antibody to enhance visualization of GFP-labelled NDUFS1-deficient cells (Figure 8E). GFP-positive cells were identified by an image analysis pipeline (see ‘Materials and methods’) that quantifies the number of fluorescent clusters and their combined area, a process repeated across multiple slices through the tumor height. This analysis confirmed larger tumors on the left flank that contained a greater number of GFP-positive clusters and a larger combined GFP-positive area (Figure 8F). These inferences were unaltered by changing the threshold for defining GFP-positive areas (Figure 8—figure supplement 1). Overall, these in vivo findings confirm that DLD1 cells are able to rescue NDUFS1-deficient SW1222 cells in a Cx26-dependent manner, consistent with in vitro co-culture observations.

In summary, we show that genetically-inactivated mitochondrial metabolism in one cell can be rescued by coupling onto a WT cell in vitro as well as in vivo. These findings demonstrate that phenotypic differences between cancer cells can be reduced ('blurred') by diffusive coupling (Figure 8G and H).

Human CRC cell lines were provided by the Bodmer Laboratory and authenticated by a panel of informative SNPs. Cell lines are regularly tested for mycoplasma contamination by antigen test. Cells were cultured using DMEM (Sigma-Aldrich, D7777) supplemented with 10% FBS and 1% PenStrep (10,000 U/mL) at 37°C and 5% CO₂. To adjust medium pH, the content of NaHCO₃ was varied (Michl et al., 2019).

Cells were seeded at a density of 200,000 cells/well in a 6-well plate and transfected with either siRNA (20 nM), for example, for GJB2, GJB3, and GJA1, or with a scramble non-targeted siRNA (Dharmacon, siGENOME smart pool), using Lipofectamine RNAiMAX (Invitrogen, Cat# 2373383). After 72 hr, cells were harvested and seeded for experiments.

Gene KOs were made for DLD1, SW1222, and HCT116 cell lines using DMEM, 10% FBS, 1% Pen/Strep. gRNA sequences were cloned into LentiCRISPR v.2 backbone as previously described (http://genome-engineering.org/gecko/wp-content/uploads/2013/12/lentiCRISPRv2-and-lentiGuide-oligo-cloning-protocol.pdf). Two gRNA sequences were cloned for each gene, using sequences below. Concentrated virus aliquots were prepared by the virus production facility at WIMM, University of Oxford. Cells were plated in clear, flat-bottom 6-well plate at a density of 200,000 cells/well and transduced using a 500 μL aliquot of lentivirus carrying the LentiCRISPR v2 construct encoding for a gRNA sequence targeting one individual gene. Polybrene was added at a concentration of 4 µg/mL. The 6-well plate was incubated for 2 days before puromycin (5 μg/mL) was added for selection, and it was incubated for 3 days before the transduced cells were used for further experiments. Single-cell clones with stable deletion of NDUFS1 and SLC9A1 were obtained in SW1222 and HCT116 cells, respectively. Note that for ALDOA gRNA-treated DLD1 cells no stable KO clones could be obtained. Therefore, lentivirus pools of KO cells were mixed populations of cells with different genomic edits and unedited cells. All growth curves were performed within 1 week of initial viral infection to avoid hypomorph or unedited cells outcompeting those that have loss-of-function mutations.

Cells were seeded on Petri dishes and harvested when confluent. Cells were resuspended in FACS-compatible solution containing 2% FBS in PBS and transferred to tubes for incubating with Ab for Cx43 APC-conjugated (R&D Systems, Cat# FAB7737A) for 45 min at room temperature (RT). Cells were washed twice with FACS solution and sorted immediately thereafter. Positive and negative cells were collected, expanded, and prepared for Western blotting for Cx43.

WT and GJB2 KO (Cx26 KO) DLD1 spheroids were prepared by the hanging drop method. Then, 20 μL-drops with 400 DLD1 cells were allowed to form spheroids and grow for 72 hr. Spheroids were then collected and transferred into sterile non-TC flat-bottom 96-well plate (Falcon) pretreated with anti-adherence rinsing solution (STEMCELL Technologies, Cat# 07010) and imaged in brightfield setting using Cytation 5 plate reader to calculate radius.

Lysates were prepared with radioimmunoprecipitation assay (RIPA) buffer. Protein concentration in the samples was measured using bicinchoninic acid (BCA) protein assay kit and adjusted using water. Samples were loaded onto a 10% acrylamide gel. The gel was run at 120 V for 90 min. Afterward, membrane transfer was performed at 250 mA for 70 min. Membranes were incubated with primary antibodies against NDUFS1 (Thermo Fisher, Cat# PA5-22309), aldolase A (Novus Biologicals, Cat# NBP1-87488), NHE1 (BD Biosciences, Cat# 611775), β-actin (Proteintech, Cat# HRP-60008), and GAPDH (Proteintech, Cat# HRP-60004) used as a loading control; connexin 26 (Thermo Fisher, Cat# 13-5800), connexin 31 (Proteintech, Cat# 12880-1-AP), connexin 43 (Thermo Fisher, Cat# 13-8300), and HRP-conjugated goat anti-rabbit and anti-mouse secondary antibodies were applied. The membrane was visualized using ECL.

Cells were grown to 50–80% confluency, fixed with 4% paraformaldehyde in PBS (Pierce; Life Technologies) and permeabilized with 0.2% Triton X-100 in PBS. After blocking with 3% BSA in PBS for 1 hr, cells were incubated with rabbit antibodies against Cx26 (Proteintech, Cat# 14842-1-AP) and Cx43 (Cell Signaling Technology, n.3512) for 1.5 hr at RT. Cells were then washed and incubated with Alexa Fluor 488 secondary antibody (Life Technologies) for 1 hr. Cells nuclei were co-stained with Hoechst 33342 (Life Technologies) applied for 1 min following washing with 1× PBS. The mounting medium used was ProLong Gold Antifade reagent (Invitrogen, Cat# P36930).

Monolayers in 4-well ibidi imaging slides were loaded with calcein AM (Invitrogen, C1430) for 10 min in HEPES RPMI, replaced to remove unloaded dye. Confluent monolayers were imaged using Zeiss LSM 700 confocal microscope. A cell in the middle of a confluent cluster was selected for bleaching (high-power 488 nm) until fluorescence (488 nm excitation, emission >510 nm) decreased by 50%. Signal was measured in the central cell, neighbors and more remote cells, and normalized to the initial intensity. Recovery was fitted to a monoexponential to calculate calcein permeability (units: μm/min) from product of the geometric perimeter and area of the central cell, divided by the time constant of fluorescence recovery. See Appendix 1 for details of calculation.

Cells were loaded with Violet (BMQC 5 µM, excitation 405 nm, Invitrogen, Cat# C10094), Green (CMFDA 20 µM, excitation 492 nm, Cat# C2925), Orange (CMRA 20 µM, 555 nm, Cat# C34551), or DeepRed (20 µM, 630 nm, Cat# 34565) for 15 min in RPMI HEPES medium. FRAP was limited to a 3 × 3 μm region of cytoplasm.

A suspension of cells was split equally, and each loaded with either CellTracker Violet or Orange for 15 min. Cells were pelleted, washed (PBS), combined in 1:1 ratio, and then seeded onto 4-well imaging slides at 300,000 cells/well. As control, cells were loaded with one type of dye. After 48 hr, monolayers were imaged confocally using sequential acquisition that minimizes bleed-through between channels: 405 nm excitation and emission 490–555 nm for the Violet channel and 555 nm excitation and emission 560–600 nm for the Orange channel. Orange and Violet fluorescence were normalized to the mean signal in paired control. The degree of dye mixing was quantified in terms of a coupling coefficient (CC):

where F_O and F_V are the normalized Orange and Violet fluorescence signals. See Appendix 1 for details of calculation.

Cells were loaded with combinations of CellTracker dyes, co-cultured for 48 hr, and processed for flow cytometry (Attune NXT Analyser). Gain-of-detection channels were optimized by analyzing mono-cultures loaded with one dye. Data were presented as pseudocolored bivariate density plots representing mono-cultures with the low-wavelength dye as blue, mono-cultures with the high-wavelength dye as red, and co-cultures as green.

Cells seeded on ibidi 4-well slides were loaded with 10 μM cSNARF1 (Thermo Fisher, Cat.# C1271, excitation, 555 nm; emission, 580 and 640 nm) for 7 min. Cells were then washed by superfusion with HEPES-buffered normal Tyrode containing (in mM) NaCl (135), KCl (4.5), CaCl₂ (2), MgCl₂ (1), HEPES (20), glucose (11), pH adjusted to 7.4 with 4 M NaOH heated to 37°C. After 2 min to allow equilibration, the superfusion was switched to ammonium containing Tyrode (in mM): NaCl (105), NH₄Cl (30), KCl (4.5), CaCl₂ (2), MgCl₂ (1), HEPES (20), glucose (11), pH adjusted to 7.4 with 4 M NaOH at 37°C. After 6 min, the superfusate was returned to normal Tyrode. In some experiments, the solutions contained 30 μM cariporide to block NHE1 (Tocris, Cat# 5358), as control. In separate experiments, cSNARF1 fluorescence ratio was calibrated using the nigericin method and this curve was used to convert measured cSNARF1 ratio to pHi.

Cells were seeded at a density of 80,000 cells/well in ibidi flat-bottom 96-well plates to produce a monolayer after 24 hr. On the measurement day, cells were loaded dually with cSNARF1 (5 μg/mL) and Hoechst 33342 (1:1000) for 15 min, followed by replacement with bicarbonate-free, Phenol Red-free medium based on D5030, containing 10 mM HEPES and MES and titrated at pH over the range 6.2–7.7. Plates were imaged at Cytation 5 plate reader and each cell, identified from the location of its nucleus, was analyzed for pHi (Michl et al., 2019). To measure variation of pHi, the frequency distribution was offset to the mode pHi for each experimental condition.

Cells were seeded at densities of 2000 cells/well on clear, flat-bottom 96-well plate (ibidi). Cells were cultured in bicarbonate-buffered media (D7777, Sigma) set to a pH of 7.4 by adjusting NaHCO₃ accordingly. The cells were incubated for 4 days at 37°C with 5% CO₂ before media were removed for glucose and lactate measurements.

Cells were seeded onto 96-well plates and grown in DMEM (D7777, Sigma) at pH 7.4. After 5 days, medium was collected, spun to remove residue, and analyzed in a Pentra C400 for glucose and lactate concentrations. Calibrations used standard solutions.

Two protocols were performed to measure metabolic rate. The first interrogated glycolytic flux, and the second interrogated glycolysis and respiration simultaneously. For the first, cells were cultured at high density (70,000 cells/well) in flat-bottom, black 96-well plates. To report extracellular pH, media contained 50 µM cSNARF1-dextran. Media, based on DMEM D5030, contained 25 mM glucose, 10% FBS, 1% PS, 1 mM pyruvate, 1% GlutaMAX, and 2 mM HEPES and 2 mM MES to provide a low but constant buffering power over the pH range studied. Fluorescence was monitored for 17 hr using a Cytation 5 device (BioTek, Agilent, Winooski, VT). Excitation was provided by a monochromator, and fluorescence emission was detected sequentially at wavelengths optimized as described previously (Blaszczak et al., 2021). For the second method, cells were cultured at high density (70,000 cells/well) in flat-bottom, black 96-well plates. To report extracellular pH and O₂, media contained 2 µM HPTS (8-hydroxypyrene-1,3,6-trisulfonic acid trisodium salt) and 50 µM RuBPY (tris(bipyridine)ruthenium(II) chloride). Media, based on DMEM D5030, contained 25 mM glucose, 10% FBS, 1% PS, 1 mM pyruvate, 1% GlutaMAX, and 2 mM HEPES and 2 mM MES to provide a low but constant buffering power over the pH range studied. Prior to measurements, each well was sealed with 150 µM mineral oil to restrict O₂ ingress. HPTS and RuBPY fluorescence were monitored for 17 hr using a Cytation 5 device (BioTek, Agilent). Excitation was provided by a monochromator, and fluorescence emission was detected sequentially at five wavelengths, which were optimized as described previously (Blaszczak et al., 2021).

GFP-labeled cells (pLV-eGFP addgene plasmid# 36083) were transduced with lentiviral constructs. Then, 48 hr later, cells were seeded at 2000 cells/well on flat-bottom 96-well plates (ibidi). For co-cultures, the ratio of the two cellular populations was varied. The following day, medium was replaced with bicarbonate-buffered media (D7777, Sigma) and incubated at 37°C with 5% CO₂ for 7 days; in some experiments, atmospheric O₂ was reduced to 2% to produce hypoxia. For some experiments, incubations were terminated on days 2–7 to obtain a time course. Plates were analyzed for GFP fluorescence and SRB absorbance (Cytation 5). The GFP signal was measured in intact monolayers (excitation 490 nm, emission 520 nm). Next, wells were prepared for the SRB assay by fixing with PFA 4% for 30 min, washing with water and staining with 100 μL 0.057% SRB (in 1% acetic acid) for 30 min. Residual SRB was removed by washing with 1% acetic acid four times. Finally, 200 μL/well 10 mM Tris base was added to release SRB and measure absorbance (520 nm).

Six-week-old female athymic Nude Crl:NU(NCr)-Foxn1nu mice were injected subcutaneously with a 1:1 mixture of either WT DLD1 cells and GFP-labeled NDUFS1 KO SW1222 cells or a 1:1 mixture of GJB2 KO DLD1 cells and GFP-labeled NDUFS1 KO SW1222 cells. CRCs cells were resuspended in 100 µL of a 1:1 mixture of Matrigel and serum-free DMEM medium before injection. Each mouse was injected with 2 million cells to each flank: WT DLD1 + NDUFS1 KO SW1222 on left and GJB2 KO DLD1 cells + NDUFS1 KO SW1222 on right. Mice were weighed and tumors were measured three times a week. At the end of the experiments, when tumors reached the scientific or humane endpoint, mice were sacrificed and tumors excised and processed for imaging. Tumor cryo-sections were cut (Thermo Fisher Scientific cryostat) across the tumor height to slices of thickness 5 microns. Sections were then stained with Hoechst and anti-GFP polyclonal antibody (Thermo Fisher #A-11122) to enhance the signal from GFP-expressing NDUFS1-deficient SW1222 cells. Briefly, frozen slides were air-dried for 30 min, relevant sections marked with a PAP pen ring, and then slides were dip-washed in PBS for 10 min (staining jar), dried with tissue and placed in staining chamber tray, blocked with PBS Triton 0.1% with 5% goat serum and 2% BSA for 30 min, rinsed with PBS 3× (using 3 mL pipette), incubated with primary anti-GFP antibody in PBS Triton 0.1% at 1:250 dilution and 2% goat serum for 1 hr, rinsed 3× with PBS, incubated with secondary antibody (goat anti-rabbit Alexa 488) at 1:1000 and Hoechst-33258 in PBS Triton 0.1% for 30 min, rinsed 1× with PBS and then dip-washed in PBS for 2 min, then dried with tissue. Approximately 35 μL of slow-fade diamond (Thermo Fisher, S36963) was added and a coverslip applied and sealed with nail polish. Every fifth slides was imagined using a Zeiss Axioscan slide scanner using a ×20 0.8 Plan-Apochromat objective. The montage of images was constructed for the Hoechst and GFP channels. To demarcate the area of the tumor, the Hoechst signal was dilated (rolling ball, 50 pixels) to expand beyond nuclei and fused to form a mask, with smaller particles ignored. The background of GFP images was calculated from the mean of fluorescence within the Hoechst-defined tumor mask. To produce a thresholded GFP image, pixels were defined as GFP-positive if their fluorescence exceeded 4, 5, or 6 standard deviations from the mean GFP signal within the masked region; all other pixels were reassigned a GFP value of zero. Next, all GFP-positive clusters were counted (ignoring large particles that are likely noncellular stains) and the total area calculated. Each slide also provided a measure of tumor area. The total GFP-positive cluster count and area, and overall tumor area were summed over all slices. The analyses were repeated for the three levels of thresholding to ensure the findings are consistent.

Data are expressed as mean ± SEM. Significance *p<0.05, **p<0.01, ***p<0.001. GMM was performed using GMMchi (https://github.com/jeffliu6068/GMMchi; Liu, 2022). Before selecting tests, data were tested for normality by the Kolmogorov–Smirnov test. One-sample t-tests were performed to compare data against an expected value (for data failing normality test, the nonparametric test was the Wilcoxon signed-rank test). For comparisons between two or more samples, one-way ANOVA was performed.

Permeability to calcein across cell-to-cell junctions was measured in monolayers where cells made contacts between one another. A cell within a confluent structure was selected for bleaching. Permeability to calcein, in units of μm/s, was measured from the FRAP recovery time course measured in the bleached cell, and geometrical information obtained from the image of the monolayer.

The principle behind calculating permeability P is equation 1:

where subscripts ‘cell’ and ‘surround’ refer to the central cell and its cellular neighbors, respectively, and ρ is the ratio of the surface area of the barrier (i.e., membrane) and the cell volume.

In the experiment, fluorescence (F) is a readout of concentration (C), normalized by a scalar α:

Thus, the rate of recovery (LHS of Equation 1) is

And the RHS of Equation 1 becomes

Putting 2 and 3 together cancels out α:

Prior to bleaching, there are no net gradients, thus initial F_cell and F_surround are equal. To account for this, fluorescence F in the cell and its surroundings can be normalized to starting signals:

Thus, Equation 1 becomes

P can therefore be expressed as:

where df_cell/dt is measured from the fluorescence recovery time course. This is the initial rate of fluorescence calculated by best-fitting the recovery time to a monoexponential curve and calculating the initial gradient at the moment of bleaching. Gradient f_cell-f_surround is measured experimentally; in fact, it is set by the protocol to be numerically 0.5 because the central cell is bleached until fluorescence decreases to 50%.

An accurate calculation of ρ would require volume rendering. However, it can be approximated by a reasonable assumption that the monolayer is cuboidal and contracts are made at the sides of cells (and not above or below the x–y plane). The monolayer image provides information about the area of the cell in the x–y plane, and its perimeter then makes contact with neighboring cells. The cell’s volume is thus area × height, and the cell’s surface (at which its contacts are made) is the cell’s perimeter × height. Thus, for the bleached cell,

All these values are inferred from the data. The permeability refers specifically to calcein.

Monolayers are imaged sequentially to obtain Violet and Orange fluorescence bitmaps (512 × 512 pixels) at the optimal wavelength settings for CellTracker Violet and Orange, respectively. Each experiment consisted of three conditions: (i) cells that were loaded with Violet only, (ii) cells that were loaded with Orange only, and (iii) a co-culture of cells that had been loaded with Violet or Orange, mixed in 1:1 ratio. Thus, (i) and (ii) are a mono-culture, whereas (ii) is expected to be a well-mixed co-culture. Diffusive exchange across gap junctions is expected to lead to mixing of dyes between cells. This will have no effect on the mono-cultures as like is exchanged for like, but in co-cultures, some cells would contain Violet and Orange pixels. The degree of dye mixing was calculated by relating the Violet and Orange signal on a pixel-by-pixel basis. To ensure accuracy, several image processing steps are necessary.

Violet mono-culture image processing:

1. In the Violet fluorescence image:

   1. Background pixels are identified by thresholding. This is used to estimate confluency and the inverse to identify cells.

   2. The signal in nonbackground pixels is averaged to give a mean Violet fluorescence value in cells. This is then used for normalizing the degree of Violet dye loading, which may be different to the Orange signal. This is called F_V^mono.

2. In the paired Orange fluorescence image, the 95% percentile of signal in nonbackground pixels is calculated to describe background Orange fluorescence. This represents autofluorescence in the Orange range in the absence of CellTracker Orange. This value is called B_O^mono.

3. The process is repeated for all images taken in the same back (biological replicate) to obtain a mean F_V^mono and B_O^mono.

Orange mono-culture image processing:

1. In the Orange fluorescence image:

   1. Background pixels are identified by thresholding. This is used to estimate confluency and the inverse to identify cells.

   2. The signal in nonbackground pixels is averaged to give a mean Orange fluorescence value in cells. This is then used for normalizing the degree of Orange dye loading, which may be different to the Violet signal. This is called F_O^mono.

2. In the paired Violet fluorescence image, the 95% percentile of signal in nonbackground pixels is calculated to describe background Violet fluorescence. This represents autofluorescence in the Violet range in the absence of CellTracker Orange. This value is called B_V^mono. Typically, B_V^mono is larger than B_O^mono.

3. The process is repeated for all images taken in the same back (biological replicate) to obtain a mean F_O^mono and B_V^mono.

Once complete, means F_V^mono is background-subtracted by B_V^mono, and similarly for Orange.

Co-culture image processing:

1. Background pixels are identified as those that are below the threshold for Violet and Orange fluorescence.

2. The nonbackground pixels are taken as areas within cells. Each pixel is annotated with a Violet and Orange fluorescence. These signals are background-subtracted (B_V^mono and B_O^mono, respectively), and then normalized to mean signals (F_V^mono and F_O^mono, respectively). Each pixel is now described by a background-subtracted and normalized fluorescence array, F_V and F_O.

3. To quantify exchange, Violet fluorescence is expressed as a fraction of total fluorescence (F_V + F_O) and, similarly, Orange fluorescence is expressed as a fraction of total fluorescence (F_V + F_O). The smaller of the two values is taken to represent coupling coefficient, as the smaller value relates to fluorescence from the ingress of CellTracker dye from neighboring cells, rather than dye that had been loaded originally in the cell. To scale this coefficient on a scale from 0 to 1, the final value is doubled.

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

In memory of Robert J Gillies (1953-2022) who made milestone contributions to the field of cancer physiology and would have enjoyed discussing the present work with us. The work was supported by the European Research Council, SURVIVE #723997. We thank Professor Alison Simmons and Dr. Agne Antanaviciute for access to the single-cell RNA GSE datasets for colonic epithelium.

This study was performed in accordance with the provisions of the Animals (Scientific Procedures) Act 1986 and recommendations set by the Biomedical Services unit at Oxford University. All work involving mice obtained approval of ethics and welfare board instructions, and was authorized by Project Licence PPL P01A04016 issued by the UK Home Office. All surgery was performed under anesthesia, and appropriate post-recovery care was provided to minimize suffering.

© 2022, Monterisi et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.