Metabolic adaptations of micrometastases alter EV production to generate invasive microenvironments

Abstract

Altered cellular metabolism has been associated with the acquisition of invasive phenotypes during metastasis. To study this, we combined a genetically engineered mouse model of mammary carcinoma with syngeneic transplantation and primary tumor resection to generate isogenic cells from primary tumors and their corresponding lung micrometastases. Metabolic analyses indicated that micrometastatic cells increase proline production at the expense of glutathione synthesis, leading to a reduction in total glutathione levels. Micrometastatic cells also have altered sphingomyelin metabolism, leading to increased intracellular levels of specific ceramides. The combination of these metabolic adaptations alters extracellular vesicle (EV) production to render the microenvironment more permissive for invasion. Indeed, micrometastatic cells shut down Rab27-dependent production of EVs and, instead, switch on neutral sphingomyelinase-2 (nSM2)-dependent EV release. EVs released in an nSM2-dependent manner from micrometastatic cells, in turn, influence the ability of fibroblasts to deposit extracellular matrix, which promotes cancer cell invasiveness. These data provide evidence that metabolic rewiring drives invasive processes in metastasis by influencing EV release.

Figure 1.

Cells from lung micrometastases display increased invasiveness. (A) Parental (P and P’) cell lines were established from mammary tumors spontaneously arising in MMTV-PyMT female mice. These cells were then transplanted into the fourth mammary FP of syngeneic mice (FVB/N), and tumors grew to a defined size (8–10 mm). Tumors resected from the mammary FP were used to establish “fat pad” (FP and FP’) cell lines. Following tumor resection, mice were maintained for sufficient time (1 mo) to allow seeding of micrometastases in the lung. Lungs were then removed, and PyMT-positive “micrometastatic” (M and M’) cell lines were established from lung homogenates. (B) P, FP, or M and P’, FP’, or M’ cells (as described in A) were plated onto 6-well dishes, and their growth was determined using the IncuCyte ZOOM live-cell imaging system. Values are mean ± SEM, n = 3 technical replicates/cell line. (C) P, FP, or M and P’, FP’, or M’ cells were seeded into the upper chambers of Transwells (8-µm pore size), and transmigration over a 2-h period toward a gradient of serum and fibronectin (applied to the lower chamber) was determined. Quantification of the number of cells adherent to the upper surface of the lower chamber are displayed. Values are mean ± SEM, n = 3 independent experiments (colored dots); data were analyzed using a paired t test. (D and E) P, FP, or M cells were plated onto plugs of rat tail collagen that had previously been conditioned by TIFs for 4 days. Cancer cells were allowed to invade preconditioned plugs for 7 days, followed by fixation and visualization of cells using H&E. Representative images (D) and quantification of the total number (E, left graph) and % (E, right graph) of cells remaining on top of or invading collagen organotypic plugs to a depth of >10 µm are displayed. Values are mean ± SEM, n = 3 independent experiments (colored dots), up to 52 fields of view were analyzed per cell line in each experiment; data were analyzed using a paired t test.

Figure 2.

Micrometastatic cells secrete proline. (A) Parental (P), FP, or micrometastatic (M) cells were plated onto 6-well plates and incubated at 37°C for 24 h. Conditioned medium was collected, and levels of the indicated metabolites were determined using LC-MS–based metabolomics. Data are expressed as the difference between metabolite levels detected in cell-conditioned medium and those in medium incubated at 37°C in the absence of cells (Δ peak area). Thus, positive values indicate production and release of a metabolite by cells, whereas negative values indicate consumption of that metabolite during the 24-h period. Values are mean ± SEM, n = 3 independent experiments; data were analyzed by two-way ANOVA with Tukey’s multiple comparison. (B) MMTV-PyMT and non-tumor–bearing FVB control mice were culled at 11–14 wk of age, blood was collected by puncture of the posterior vena cava, sera were prepared by centrifugation, and levels of the indicated metabolites were determined using LC-MS. Lungs of MMTV-PyMT mice (MMTV-PyMT (FVB)) and non-tumor–bearing FVB control mice (FVB) (n = 9 mice) were assessed by histology and categorized according to the presence (Mets; n = 22 mice) or absence (no Mets; n = 7 mice) of lung metastases. Values are mean ± SEM; data were analyzed by ordinary one-way ANOVA with Sidak’s multiple comparison. (C) LC-MS metabolomics was used to determine levels of the indicated metabolites in plasma collected from metastatic breast cancer patients (n = 99) and healthy volunteers (n = 56). Values are mean ± SEM; data were analyzed using an unpaired Student’s t test.

Figure 3.

Proline and glutathione metabolism is altered in micrometastatic cells. (A and B) Parental (P), fat pad (FP), or micrometastatic (M) cells were cultured as for Fig. 2 A, and levels of intracellular metabolites were determined using LC-MS–based metabolomics. The abundance of intracellular metabolites in FP and micrometastatic (M) cells was normalized to cell number and expressed relative to levels of the same metabolites in parental (P) cells. For the heatmap in A, values are log₂-fold changes, and for B, values are mean fold-change; data were analyzed by repeated measures one-way ANOVA with Tukey’s multiple comparison test, n = 5 experiments (each colored dot represents an individual experiment). (GSH, reduced glutathione; GSSG, oxidized glutathione; SAM, S-adenosyl methionine; DHAP, dihydroxyacetone phosphate; G3P, glycerol-3-phosphate; PEP, phosphoenolpyruvate; SAH, S-adenosylhomocysteine). (C) Parental (P) and micrometastatic (M) cells were cultured as for Fig. 2 A, and lysates derivatized with NEM. NEM adducts of cysteine (cys-NEM), γ-glutamylcysteine (glu-cys-NEM), and reduced glutathione (GS-NEM) were then detected using LC-MS and plotted as fold change relative to parental cells, n = 3–4 independent experiments (colored dots); values represent mean ± SEM, paired t test. (D) Cells were cultured as for Fig. 2 A, and levels of the mRNAs encoding GCL catalytic subunit (Gclc; left graph) and Slc7a11 (xCT; right graph) were determined using qPCR. Values were normalized to ARPP P0 mRNA levels and expressed as fold change relative to parental (P) cells. Values are mean ± SEM, n = 3 independent experiments (colored dots); data were analyzed using one-way ANOVA with Tukey’s multiple comparison test. (E) Mammary tissue and lungs were resected from MMTV-PyMT mice at primary tumor endpoint (tumor size 15 mm). Tissues were formalin-fixed, paraffin-embedded, and sectioned for histological examination. Expression of the mRNA-encoding xCT/Slc7a11 was visualized using ISH (RNAScope), and separate serial sections were counterstained with H&E. Sections representing primary mammary tumors, their matched lung micrometastases, and surrounding lung parenchyma are displayed. Red arrowheads highlight the brown dots that indicate hybridization with the xCT probe, bar 100 μm. Quantification of the average optical density of the xCT probe in stained tissue sections was achieved using HALO software; values are mean ± SEM, Kruskal–Wallis with Dunn’s multiple comparisons test, n = 6 mice (each colored dot represents an individual mouse).

Figure S1.

Comparison of metabolite levels of cells from micrometastases and frank metastases with their primary tumor counterparts. (A–D) The abundance of intracellular metabolites in fat pad’ (FP’), micrometastatic’ (M’) (A and B) and frank macrometastatic (maM’) (C and D) cells was determined using LC-MS and expressed relative to levels of the same metabolites in parental (P’) cells. For the heatmaps (A and C), values are log₂-fold changes, and for the graphs (B and D), values are mean fold change ± SEM; statistics are repeated measures one-way ANOVA with Tukey’s multiple comparison test, n = 3 independent biological replicates (each colored dot represents an individual experiment).

Figure 4.

Proline and glutathione synthesis pathways compete for glutamine-derived carbons in micrometastatic cells. (A and B) Parental (P) or micrometastatic (M) cells were cultured for 24 h in the presence of ¹³C₅-labelled glutamine (A). Additionally, micrometastatic (M) cells were cultured with ¹³C₅-labelled glutamine in the presence of the indicated concentrations of PYCR inhibitor (PYCRi) for 24 h (B). The presence of ¹³C₅-labelled glutamine-derived carbons in the indicated metabolites was determined using LC-MS. Values are mean ± SEM, n = 5 independent experiments (A), n = 4 (for cellular metabolites) or n = 3 (for secreted metabolites) independent experiments (B); data were analyzed by paired t test. Each colored dot represents an independent experiment. (C) Parental cells were cultured for 6 h in the presence of ¹³C₅-labelled glutamine in DMEM (full) or cystine-free DMEM supplemented with the indicated concentrations of cystine. Metabolomics were performed as for A. Values are mean ± SEM, n = 3 independent experiments (colored dots); data were analyzed by two-way ANOVA with Dunnett’s multiple comparison test. (D) Schematic representation of the utilization of ¹³C₅-glutamine–derived carbons (denoted with red asterisks) in PyMT-derived primary tumor (P) and micrometastatic (M) cells. Glutamine-derived carbons are primarily used for glutathione production in primary tumor cells, while micrometastatic cells reroute these carbons toward proline synthesis at the expense of glutathione production (Cys-s, cystine).

Figure 5.

Decreased glutathione synthesis in micrometastatic cells is associated with increased EV release. (A and B) Conditioned media were collected from parental (P and P’), FP (FP and FP’), and micrometastatic (M and M’) cells, and EVs were purified from these using differential centrifugation. The size distribution and number of EVs were analyzed by nanoparticle tracking (A), and the CD63, TSG101, and CD81 content of EV pellets was determined by western blotting (B). Values are mean ± SEM, n = 7 (P, FP, and M); n = 5 (P’, FP’, and M’) (A) or n = 4 (P, FP, and M); n = 5 (P’, FP’, and M’) (B) independent experiments; data were analyzed by one-way ANOVA with Tukey’s multiple comparisons test. (C–E) Parental (P) cells were incubated with the indicated concentrations of BSO for 24 h, and the cells lysed for determination of intracellular metabolites by LC-MS (C). Following another 24 h, cell-conditioned medium was collected for isolation of EVs by differential centrifugation (D and E). The heatmap (C) displays levels of the indicated metabolites expressed as log₂-fold change (normalized to cell number) relative to untreated (P) cells, n = 1 (three technical replicates/condition). The size distribution and number of EVs released by cells incubated in the presence or absence of BSO (1.88 μM) were analyzed by nanoparticle tracking (D), and the CD63 and TSG101 content of EV pellets from untreated or BSO-treated (1.25 μM, 1.88 μM, and 2.5 μM) P cells was determined by western blotting. CD63 levels in EV pellets were quantified as for Fig. 5 A (E). Values represent mean ± SEM, n = 4 independent experiments (colored dots); data were analyzed by Friedman ANOVA with Dunn’s multiple comparison. (F and G) Parental (P), FP, or micrometastatic (M) cells were cultured as for Fig. 2 A, and levels of intracellular lipidic metabolites were determined using LC-MS–based metabolomics. Data are expressed as a volcano plot (F) showing the mean differences (M minus P [M-P]; x axis) between the peak areas of lipids identified in M and P cells. The dotted line represents the P value (y axis) above which all the indicated lipid classes display significant differences across the conditions, n = 3 independent experiments, data were analyzed by paired t test. The classes of lipids that were detected at significantly different (P < 0.033) levels between M and P cells are denoted with colored dots. Color coding for the lipid classes is as follows: cholesterol esters (CE); sphingomyelin/ceramide (SM/Cer); PE, phosphatidylethanolamine (PE); phosphatidylcholine (PC); plasmanyl/plasmenyl phosphatidylcholine (Plas.PC); bis(monoacylglycerol) phosphate (BMP). Log₂-fold differences peak areas of the indicated lipid classes (normalized to cell number) between FP and P (FP minus P; blue bars) and M and P (M minus P; red bars) are displayed in the graph in G, n = 3 independent experiments; data were analyzed using two-way ANOVA with Tukey’s multiple comparison test. Source data are available for this figure: SourceData F5.

Figure S2.

Treatment of parental (P) cells with BSO. (A–D) Parental (P) cells were incubated with the indicated concentrations of BSO for 24 h (A and B), 48 h (D), or for the indicated times (C). Following this, cell-conditioned medium was collected for isolation of sEVs by differential centrifugation (D), and the cells lysed for determination of intracellular metabolites by LC-MS (A and C). Levels of oxidized (GSSG) and reduced (GSH) glutathione are expressed as fold change relative to untreated cells (A and C), and the total number of cells was determined (B). Values in A–C are mean ± SEM, three technical replicates per condition. The CD63 and TSG101 content of EV pellets from untreated or BSO-treated (2.5 μM) P cells was determined by western blotting. Each western blot represents an individual experiment. CD63 levels in EV pellets were quantified as for Fig. 5 B, and values represent mean ± SEM, n = 3 independent experiments (each colored dot represents an individual experiment); data were analyzed by paired t test. Source data are available for this figure: SourceData FS2.

Figure 6.

Micrometastatic cells and their EVs are enriched in ceramide species. (A) Parental (P), FP, and micrometastatic (M) cells were cultured for 48 h on glass-bottomed dishes and fixed using paraformaldehyde. Ceramides (green), F-actin (phalloidin; magenta), and nuclei (DAPI; blue) were visualized by immunofluorescence; bar is 10 μm. ImageJ was used to quantify the mean intensity of ceramide (sum of z-stacks, 10 stacks/field of view). Values are mean ± SEM, n = 3 independent experiments (colored dots); data were analyzed by one-way ANOVA with Tukey’s multiple comparisons. (B) The intracellular levels of ceramide species in P, FP, and M cells were determined as for Fig. 5, F and G. Ceramide species found to be present at significantly different levels between M and FP or P cells were normalized to cell number and expressed as fold change relative to (P) cells; values are mean ± SEM; data were analyzed using one-way ANOVA with Tukey’s multiple comparison test, n = 3 independent experiments. Greater-than two technical replicates/experiment were performed, and each technical replicate is represented by a dot. (C) P, FP, and M cells were cultured for 48 h. Conditioned media were harvested from these cultures, and EVs purified using differential centrifugation. LC-MS–based lipidomics was used to determine the levels of the indicated ceramide species in the EV pellets; values are mean ± SEM, n = 4 independent experiments (each dot represents one experiment); data were analyzed by one-way ANOVA. Lipid extracts of EVs were normalized to the number of EV-releasing cells prior to lipidomic analysis. (D) LC-MS metabolomics was used to determine levels of the indicated ceramide species in plasma collected from metastatic breast cancer patients (n = 96) and matched healthy volunteers (n = 55). An internal standard (Splash II, Avanti) was spiked in the lipid extraction buffer and was analyzed to account for technical variation among samples (right graph). Values are mean ± SEM; data were analyzed by unpaired Student’s t test. Each dot represents an individual patient. (E) M cells were transduced with lentiviral vectors bearing gRNAs against non-targeting sequences (NTC) or recognizing sequences in nSMase1 (nSM1) or nSMase2 (nSM2). Ceramides were determined using LC-MS as for (B). Values are mean ± SEM, n = 3 independent experiments; data were analyzed by one-way ANOVA with Tukey’s multiple comparison test. Greater-than two technical replicates/experiment were performed, and each technical replicate is represented by a dot.

Figure S3.

Validation of CRISPR disruption of nSMases and Rab27s and EV release from nSMase-1 CRISPR cells. (A) Parental (P), FP, and micrometastatic (M) cells in which nSMase1 (nSM1) or nSMase2 (nSM2) were disrupted by CRISPR were lysed. Levels of mRNA-encoding nSM1 (left graph) or nSM2 (right graph) were determined using qPCR. All data were normalized to ARPP P0 and presented relative to expression in parental NTC cells; values are mean ± SEM, n = 3, colored dots are independent experiments. (B and C) Parental (P), FP, or micrometastatic (M) cells were transduced with a lentiviral vector bearing gRNAs recognizing non-targeting sequences (CRISPR-NTC) or sequences targeting nSMase1 (CRISPR-nSM1). EVs were purified from conditioned medium collected over a 48-h period and analyzed using nanoparticle tracking (B) as for Fig. 5 A, and levels of CD63 in EV pellets were determined by western blotting (C) as for Fig. 5 B. Values are mean ± SEM, one-way ANOVA, with Tukey’s multiple comparison, n = 3 (B), n = 6 (C), colored dots are independent experiments. (D) Western blotting was used to determine levels of Rab27a and Rab27b in parental (P) and micrometastatic (M) cells in which the genes for the Rab GTPases were disrupted using CRISPR. Actin is used as loading control. Source data are available for this figure: SourceData FS3.

Figure 7.

Micrometastatic cells release EVs via an nSMase2-dependent and Rab27-independent mechanism. Parental (P), FP, or micrometastatic (M) cells were transduced with a lentiviral vector bearing gRNAs recognizing non-targeting sequences (NTC) or sequences in nSMase2 (nSM2), Rab27a, or Rab27b. (See Fig. S3 for confirmation of suppression of nSMase [Fig. S3 A] and Rab27 [Fig. S3B] expression at the gene and protein levels, respectively). (A–F) EVs were purified from conditioned medium collected over a 48-h period and analyzed using nanoparticle tracking (A, C, and E) as for Fig. 5 A, and levels of CD63 in EV pellets were determined by western blotting (B, D, and F). In C–D, P cells transduced with non-targeting (NTC) lentiviral vectors or those targeting nSMase-2 (nSM2) were incubated with BSO (1.88 µM) during the 48-h EV collection period. Values are mean ± SEM; data were analyzed by one-way ANOVA, n = 5 (A, C, and E), n = 4 (B), n = 3 (D) independent experiments. Source data are available for this figure: SourceData F7.

Figure 8.

nSMase2-dependent EV release favors generation of a pro-invasive microenvironment. (A and B) Micrometastatic (M) cells were transduced with lentiviruses bearing gRNAs recognizing non-targeting sequences (NTC) or sequences in nSMase2 (nSM2-CR). Transduced cells were plated into the upper chamber of Transwells and transmigration determined as for Fig. 1 C (A). Values are mean ± SEM, n = 3 independent experiments (colored dots); data were analyzed using a paired t test. Alternatively, transduced cells were plated onto fibroblast preconditioned collagen plugs, and invasion into these was determined as for Fig. 1 E (B). Values are mean ± SEM, two plugs/condition, n = 39–47 fields of view; data were analyzed by mixed effects ANOVA with Tukey’s multiple comparison test. (C) Schematic depiction of protocol for determining EVs’ influence on the invasive microenvironment of organotypic collagen plugs. EVs released by control (NTC) or nSMase-2 CRISPR (nSM2-CR) M cells were incubated with TIFs for 72 h. EV-treated TIFs were then allowed to precondition organotypic plugs of rat tail collagen for 4 days. NTC or nSM2-CR M cells were then plated onto preconditioned plugs and allowed to invade for 7 days. (D) Control (NTC) or nSMase-2 CRISPR (nSM2-CR) M cells were plated onto organotypic collagen plugs that had been preconditioned as outlined in C. Invasion was quantified as for Fig. 1 E. Values are mean ± SEM, n > 22 fields of view/condition (n = 2 independent experiments); data were analyzed using mixed effects ANOVA with Tukey’s multiple comparison test. (E) TIFs were incubated with EVs from control (NTC) or nSMase-2 CRISPR (nSM2-CR) M cells for 72 h or were left untreated. EV pre-treated TIFs were then replated and allowed to generate ECM for 7 days, which was then de-cellularized. MDA-MB-231 breast cancer cells were plated onto de-cellularized ECMs, and time-lapse microscopy (over 16 h) followed by cell tracking (ImageJ) was employed to measure their migration (n > 100 cells, 2 independent experiments). Colored lines indicate representative tracks of individual migrating cells (left panels). Bar, 100 μm. Box and whisker plots are 10–90 percentile, and the line displays median. Data were analyzed by Kruskal–Wallis test with Dunn’s multiple comparisons test. (F and G) Schematic summary of how metabolic control of EV biogenesis in cells from primary mammary tumors and lung micrometastases may influence ECM microenvironments. Cells from primary tumors synthesize glutathione using glutamine-derived carbons and release EVs in a Rab27-dependent/nSMase2-independent manner (F). In micrometastatic cells, more glutamine-derived carbons are used for proline production, leading to decreased glutathione synthesis. This, in combination with increased nSMase2 (nSM2)-dependent ceramide production, promotes nSM2-dependent intraluminal vesicle budding and the release of the resulting EVs in a Rab27-independent way, which encourages fibroblasts to generate a more invasive microenvironment (G) (Gln, glutamine; Glu, glutamate; Cys-s, cystine; GSH, reduced glutathione; Pro, proline; SM, sphingomyelin; Cer, ceramide).

Figure S4.

PYCR activity is not required for transmigration of MMTV-PyMT mammary cancer cells and serine/glycine levels in mammary cancer cells from primary tumor and micrometastases and following treatment with BSO. (A) Parental (P) or micrometastatic (M) cells were treated with PYCR inhibitor (PYCRi; 20 µM) or control vehicle as indicated. Cells were then seeded into the upper chambers of Transwells (8-µm pore size) and allowed to transmigrate over a 2-h period toward a gradient of serum and fibronectin (applied to the lower chamber) in the presence or absence of 20 µM PYCRi or vehicle control (DMSO). Representative images (left panels) and quantification (right panels) of the number of cells adherent to the upper surface of the lower chamber are displayed. Bars, 100 µm. Values are mean ± SEM, n = 3 experiments (each colored dot represents an individual experiment); data were analyzed using a paired t test. (B) Parental (P), FP or micrometastatic (M) cells were cultured as for Fig. 2 A, and levels of intracellular metabolites were determined using LC-MS–based metabolomics. The abundance of intracellular serine and glycine are expressed as mean fold change ± SEM relative to parental cells, n = 5 experiments (each colored dot represents an individual experiment). (C) Parental (P) cells were incubated with the indicated concentrations of BSO for 24 h and then cells were lysed for determination of intracellular metabolites by LC-MS. Levels of glycine and serine are expressed as fold change relative to untreated cells. Values are mean ± SEM, three technical replicates per condition.

Figure S5.

Micrometastatic (M) cells release less EV-associated mtDNA than cells from primary tumors (P), and EVs from micrometastatic cells do not influence the fibronectin filament length or organization in the ECM deposited by fibroblasts. (A–C) EVs were purified from the conditioned media of parental (P) and micrometastatic (M) cells as for Fig. 5 A and B. The number of EVs was determined by nanoparticle tracking (A). Purified EVs were incubated in the presence of DNase immobilized to agarose beads to remove DNA associated with the external surface of the EVs. DNase-treated EVs were then lysed, and mtDNA and nuclear DNA was determined by digital-drop PCR using primers complementary to sequences within the mitochondrial ND1, ND2, and ND5 genes or the nuclear-encoded VDAC1 gene (B). The cellular content of the ND1, ND2, ND5, and VDAC1 genes were confirmed using digital-drop PCR (C). Values are mean ± SEM, n = 3 independent experiments (each colored dot represents an individual experiment); data were analyzed using a paired t test. (D–G) TIFs were incubated with EVs from control (NTC) or nSMase-2 CRISPR (nSM2-CR) M cells for 72 h or were left untreated (−). EV pre-treated or untreated TIFs were then replated and allowed to generate ECM for 7 days, which was then de-cellularized. Fibronectin was then visualized in the de-cellularized ECM using immunofluorescence followed by confocal microscopy. Quantitative image analysis (TWOMBLI) (Wershof et al., 2021) was then used to determine the characteristics of the deposited fibronectin fibers, including the fiber length (D), alignment (E), curvature (F), and branching (G). Values are mean ± SEM, 60 fields of view were quantified per condition (each grey dot corresponds to a field of view).