Enhanced mitochondrial fission suppresses signaling and metastasis in triple-negative breast cancer

We purchased MDA-MB-231 cells from the ATCC (Manassas, VA) and cultured cells in Dulbecco’s modified Eagle medium (DMEM) supplemented with 10% fetal bovine serum (FBS), 1% penicillin/streptomycin (Pen/Strep) (Thermo Fisher Scientific, Waltham, MA), and 1% GlutaMAX (Thermo Fisher Scientific, Waltham, MA). We obtained SUM159 cells from Dr. Stephen Ethier (now at The Medical University of South Carolina, Charleston, SC) and cultured cells in F-12 media supplemented with 10% fetal bovine serum, 1% Pen/Strep, 1% glutamine, 5 μg/mL hydrocortisone, and 1 μg/mL insulin. We authenticated all cells by analysis of short tandem repeats and characterized cells as free of Mycoplasma at the initial passage. We used all cells within 3 months after resuscitation and maintained all cells at 37 °C in a humidified incubator with 5% CO₂.

We used MDA-MB-231 and SUM159 cells stably expressing PISD-mCherry as described in our recent study [15] for initial experiments. Plasmid pLV-mito-GFP was a gift from Pantelis Tsoulfas (Addgene plasmid #44385), and we also used this mitochondrial targeted GFP (Mito-GFP) to visualize mitochondrial morphology in our initial experiments as described previously [15, 16].

To identify the activity of ERK and Akt, we used the kinase translocation reporter (KTR) for ERK and Akt as previously described [17, 18]. This KTR construct contains H2B fused to mCherry (H2B-mCherry), the Akt-KTR reporter (Aquamarine), the ERK-KTR reporter (mCitrine), and a puromycin selection marker all separated by P2A linker sequences cloned into the Piggyback transposon vector as described previously (pHAEP) [18]. To stably express PISD in cells with the KTR construct, we generated a construct with unlabeled PISD and a hygromycin selection marker separated by a P2A sequence (PISD-hygromycin). To visualize mitochondria concurrently in cells with the KTR construct, we replaced GFP in the Mito-GFP plasmid with mTagBFP2 (Mito-BFP) and added a neomycin selection marker (Mito-BFP-neomycin). For stable expression of Drp1, we used the pEYFP-C1-Drp1 plasmid [19], a gift from Richard Youle (Addgene plasmid #45160; http://​n2t.​net/​addgene:​45160; RRID: Addgene_45160). We cloned the Drp1 reading frame from this plasmid into the pLVX vector and added a hygromycin selection marker separated by a P2A sequence (Drp1-hygromycin). We produced recombinant lentiviral vectors for PISD-hygromycin, Mito-BFP-neomycin, and Drp1-hygromycin as previously described [20].

We first generated cells stably expressing click beetle green luciferase (SUM159-CBG and MDA-MB-231-CBG) as described previously through selection with blasticidin [21]. We next transfected cells with the Piggyback transposon vector containing both KTRs (pHAEP) using FuGENE HD (Promega). We selected stably expressing cells using puromycin and confirmed expression by fluorescence as previously described [17]. We next transduced cells (SUM159-CBG-pHAEP and MDA-MB-231-CBG-pHAEP) with the Mito-BFP-neomycin lentiviral vector and selected cells using neomycin, generating our wild-type (WT) cells (SUM159- and MDA-MB-231-CBG-pHAEP Mito-BFP-neomycin). Next, we transduced cells with either PISD- or Drp1-hygromycin and selected for stable expression using hygromycin. For cells expressing only Mito-BFP only cells, we transduced cells already expressing CBG with the Mito-BFP construct and selected using neomycin. After selection, we transduced cells with PISD-hygromycin and selected for a stably expressing population. For cells with only PISD or Drp1, after we added and selected for CBG stably expressing cells, we introduced the PISD- or Drp1-hygromycin construct into cells already stably expressing CBG and then selected with hygromycin.

To analyze levels of PISD, we performed qRT-PCR for PISD and β-actin using SYBR Green detection as described previously [22]. Primers for PISD (Sigma Aldrich) were 5′-CTCCATTCGCATCTACTTTG-3′ and 5′-AGCTGAAGTCATTGTAGGAG-3′ and β-actin 5′-TGTACGTTGCTATCCAGGCTGTGC-3′ and 5′-CGGTGAGGATCTTCATGAGGTAGTC-3′.

We purchased l-α-lysophosphatidylethanolamine (LPE, cat 860081) from Avanti Polar Lipids Inc. (Alabaster, AL) and prepared a 50-mM stock as described previously [23].

To visualize mitochondrial morphology, we seeded 2 × 10⁴ cells onto 35-mm dishes with a 20-mm glass bottom (Cellvis, Mountain View, CA) in FluoroBrite DMEM media (ThermoFisher Scientific, cat. A1896701), containing 10% FBS, 1% Pen/Strep, 1% GlutaMAX, and 1% Sodium Pyruvate (Thermo Fisher Scientific, Waltham, MA). Two and 3 days after seeding, we changed media to FluoroBrite DMEM media as described above containing either LPE (50 μM) to drive fragmentation, leflunomide (Cayman Chemical Company, cat: 14860; 50 μM) to drive fusion, or vehicle control. For visualization, we acquired images on an Olympus IX73 microscope with a DP80 CCD camera (Olympus) at the time points indicated in the figure.

To determine effects of mitochondrial morphology on migration, after treatment of cells with lipids as described in the “Mitochondrial morphology assays” section, we loaded cells into our microfluidic migration device as previously described and quantified the distance each cell moved within the device [15].

To study the effects of PISD on cell migration in a complex 3D environment, we used tissue-engineering constructs consisting of a SU8 construct coated with networks of engineered fibrillar fibronectin [24]. We seeded 5 × 10⁴ SUM159-CBG-WT or PISD cells stably expressing pHAEP onto the construct while rotating for 1 h at 37 °C. After 1 h, we washed the construct with PBS and placed it in a 35-mm dish with a 20-mm glass bottom. We covered the construct with 2 mL of FluoroBrite DMEM containing the same additives as described in the “Cell culture” section and placed the dish in the incubator for 3 h. After 3 h, we imaged the dish every 20 min for 3 h on an Olympus FVMPE-RS upright two-photon microscope using settings as described previously [18]. We identified and tracked cells using custom MATLAB code that tracked the H2B-mCherry nucleus of cells on the construct.

We extracted total lipids from SUM159-WT and PISD cells based on a modified protocol described previously [25]. We homogenized SUM159-WT and PISD pellets containing 1 × 10⁷ cells in 0.75 mL of methanol. We transferred the homogenate into 1.5-mL glass vials and added 0.75 mL of chloroform. We sonicated the mixture for 30 min and then centrifuged the mixture at 9000 rpm for 5 min at RT, transferring the liquid layer to another tube. We performed an extraction on the remaining solid phase again using a 2:1 (v/v) chloroform to methanol solution, followed by sonication for 30 min, centrifugation, and then combining the two extracted liquid phases. We repeated this procedure twice more, combining the extracted lipids in the liquid phase each time (total of 4 extractions).

For MALDI mass spectrometry (MS), we re-dissolved the lipid extract in 2 mL of chloroform to methanol (3:2 v/v) containing 1 mM NaCH₃COO. We mixed equal amounts of the re-dissolved lipid extract and the matrix solution (10 mg/mL 2,5-dihydroxybenzoic acid (DHB) in acetonitrile to water (70:30% v/v)) and spotted 1 μL of the mixture on a stainless steel MALDI-plate. We dried the sample in a dessicator and analyzed the spot by MALDI-MS and MS/MS in positive ion modes using a MALDI-TOF-MS instrument (SYNAPT G2-Si, Waters US, Milford, MA). We acquired and analyzed data using Masslynx (Waters, Milford, MA). We identified lipids using MALDI-MS/MS and the LIPID MAPS database for reference (https://​www.​lipidmaps.​org/​).

For MALDI imaging mass spectrometry (IMS), we injected 1 × 10⁶ SUM159- and MDA-MB-231-WT and PISD cells orthotopically into the fourth mammary fat pad of 16–23-week-old female NSG mice. After tumors reached ~ 1 cm in diameter, we removed tumors, mounted tissue on a cryotome tissue block using Shandon M-1 Embedding matrix (ThermoFisher, 1310TS), and placed samples on dry ice. We then stored the sample at − 80 °C until sectioning. We cryosectioned 10-μm sections of each tumor and placed them on clean, untreated microscope slides. We dried slides under vacuum for 30 min, and then we sprayed the samples with 40 mg/mL DHB in 1:1 water to methanol (v/v) containing 0.1% trifluoroacetic acid using a HTX M5 matrix sprayer (HTX Technologies, Chapel Hill, NC). We collected IMS data in positive ion mode and analyzed data using HDImaging software (Waters, Milford, MA).

We analyzed effects of PISD on cellular ATP levels using an ATP colorimetric assay kit (Sigma Aldrich, cat. MAK190) following the manufacturer’s instructions. To determine effects of LPE and leflunomide on mitochondrial mass and membrane potential, we utilized MitoTracker® Green FM (Invitrogen, M7514) and MitoProbe™ JC-1 Assay Kit (Thermo Fisher, M34152) as per the manufacturer’s protocols, respectively [15].

We formed spheroids as previously described [21], co-culturing WT or PISD cells expressing mitochondrially targeted GFP with human mammary fibroblasts (HMFs) expressing mCherry (1:1 ratio of cancer cells to HMFs). We embedded spheroids in fibrin gels (final concentration 4 mg/mL fibrin, 2.5 U/mL thrombin, and 0.01 U/mL aprotinin [26]) and covered in media containing aprotinin (0.01 U/mL) and LPE (50 μM) or leflunomide (50 μM). On the day after embedding (day 1), we added LPE (50 μM) and leflunomide (50 μM) to fresh culture media covering gels. Two days after seeding spheroids in fibrin gels, we imaged spheroids using two-photon microscopy. We quantified the radius and points on the perimeter of the spheroid using custom MATLAB code [17].

To quantify the activation of both ERK and Akt kinases in single cells, we used previously validated kinase translocation reporters (KTR). KTRs measure activities of ERK and Akt by utilizing a known downstream substrate specific for each kinase fused to a fluorescent protein (citrine and aquamarine for ERK and Akt, respectively). Phosphorylation and dephosphorylation of a KTR reversibly shifts localization from cytoplasm to nucleus, respectively [27, 28]. For 2D signaling experiments, we seeded 1.2 × 10⁵ (bulk) or 1.0 × 10⁵ (single cell mitochondrial analysis) MDA-MB-231- and 8.5 × 10⁴ (bulk) or 8.0 × 10⁴ (single cell mitochondrial analysis) SUM159-WT or PISD cells into 35-mm dishes with a 20-mm glass bottom (Cellvis, Mountain View, CA) in FluoroBrite DMEM media (ThermoFisher Scientific, A1896701). After 2 days, we changed medium to Fluorobrite DMEM containing either LPE (50 μM), leflunomide (50 μM), or vehicle control. The following day, we changed media to low (1%) serum FluoroBrite DMEM media containing either LPE (50 μM), leflunomide (50 μM), or vehicle control. After overnight incubation, we treated cells with either serum (10%), EGF (50 ng/mL, R&D Systems, Minneapolis, MN), or vehicle control and acquired images at the times listed in the figure. For bulk signaling experiments, we presented changes in activation of Akt and ERK as the earth mover’s distance (EMD) relative to the initial time point (0 min) [29, 30]. The EMD is a mathematical measure of the distance between two samples. It thinks of two histogram datasets as piles of dirt in multivariate space, and the EMD score is the minimum cost of moving one pile into the other. The higher the EMD score, the greater the difference between the two histograms [30]. There is no p value associated with the EMD score, as it is a statistical test itself.

We analyzed kinase translocation reporter images for activation of Akt and ERK and identified mitochondrial morphology using an adaptive thresholding method in MATLAB as described previously [17, 18]. For individual fluorescence microscopy images of mitochondria, we used MATLAB to skeletonize the mitochondria, measure total pixel area and major axis length using the regionprops command, and assign each mitochondrion a pixel ID. We separated mitochondria into fused or fissioned groups by area and major axis length using a user-defined cutoff that we applied to every cell group analyzed in each experiment. We categorized an identified mitochondrion that had total area/majoraxislength > 2 as networked. For dynamic mitochondrial analysis in cells expressing KTRs, we first took the two-photon microscopy images and made a Z-stack, correcting for any dish movement, and analyzed the mitochondria as described above. We will provide MATLAB code upon request.

In addition to KTRs, we identified changes in active and total ERK and Akt kinases using immunofluorescence staining. We seeded cells as described in the “Kinase translocation reporter” section and stained cells in response to serum at the time points indicated in the figure. We used the following primary antibodies from Cell Signaling Technologies (Danvers, MA, USA): anti-phospho-Akt (Ser473) (cat. 4058), anti-pan-Akt (cat. 2920), anti-phospho-ERK (Thr202/Tyr204) (cat. 4370), and anti-pan-ERK (cat. 4696). We detected primary antibodies with fluorescent secondary antibodies from Jackson ImmunoResearch Laboratores Inc. (West Grove, PA, USA): anti-rabbit AlexaFluor 488 (cat. 111-545-003) and anti-mouse AlexaFluor 594 (cat. 115-585-003).

We analyzed differences in membrane order among cells with the fluorescent dye laurdan. After treating various cells as described in the “Mitochondrial morphology assays” section, we stained cells (~ 70% confluency) with 5 μM Laurdan (Cayman Chemical, cat. 19706) as previously described [31] and imaged cells using two-photon microscopy (excitation 800 nm, emission 400–460 nm and 470–530 nm) [32]. We calculated general fluorescence polarization (GP) of laurdan using custom MATLAB code as described previously [31].

The University of Michigan IACUC approved all animal procedures (protocol 00008822). The animals used in this study received humane care in compliance with the principles of laboratory animal care formulated by the National Society for Medical Research and Guide for the Care and Use of Laboratory Animals prepared by the National Academy of Sciences and published by the National Institute of Health (Publication no NIH 85-23, revised 1996).

We established orthotopic tumor xenografts in the fourth inguinal mammary fat pads of 32-week-old female NSG mice as described previously [22], implanting 2 × 10⁵ SUM159- or MDA-MB-231-CBG-pHAEP Mito-BFP (WT) or PISD stably expressing cells.

To generate systemic metastases, we injected 1 × 10⁵ MDA-MB-231-WT or PISD breast cancer cells or SUM159 cells expressing different mitochondrial morphologies directly into the left ventricle of the heart of 6–8-week-old female NSG mice (Jackson Laboratory, Bar Harbor, ME, USA) as previously described [33]. All injected cells stably expressed click beetle green (CBG) luciferase for bioluminescence imaging. We quantified metastatic burden over 20–21 days after injection by bioluminescence imaging as previously described [21]. Due to spontaneous mortality immediately following intracardiac injection (one in the WT group and three in the PISD group), we present data only for mice that reached the experimental endpoint. Immediately after euthanizing mice, we imaged lung metastases ex vivo by two-photon microscopy. We used the excitation and emission filters described in our previous publication and quantified activation of Akt and ERK using custom MATLAB code [18]. For analysis of metastases to the bone marrow from intracardiac injection, we flushed the bone marrow with PBS from the femur and tibia and measured bioluminescence from cancer cells as described previously [34].

To identify effects of PISD on bone metastasis directly, we injected 2.5 × 10⁴ MDA-MB-231-WT or PISD cells into the left femoral artery of female 15–18-week-old NSG mice as previously described [35]. We tracked tumor progression by bioluminescence imaging [21] and quantified bone loss using a MATLAB code to measure the total volume of the proximal tibia at the time point identified in the figure [35, 36]. For imaging Akt and ERK activation in the bone marrow, we euthanized mice, immediately used a Dremel tool to think cortical bone of excised tibias, and imaged KTR cells ex vivo by two-photon microscopy as described for lung metastases. We quantified the activation of ERK and Akt in single cells using custom MATLAB code [18].

We analyzed the expression of PAI1, EPCAM, and PISD in primary CTCs as previously described [37]. For metastasis-free survival plots, we downloaded gene expression data from three different cohorts of patients with breast cancers [38‐40] (572 total patients with outcomes) from the National Center for Biotechnology Information (NCBI) Gene Expression Omnibus (GEO) website. We pooled data from GSE2034, GSE2603, and GSE12276 and normalized them together using the Robust Multi-chip Average (RMA) algorithm. We analyzed annotated patient data for survival by Kaplan–Meier analysis. To derive optimal cutoff values, we ordered gene expression levels from low to high and computed a rising step function to define a threshold (middle of the step) by the StepMiner algorithm [41]. We converted gene expression values (PISD: 202392_s_at; MFN1: 211801_x_at; MFN2: 216205_s_at) to high and low levels based on the StepMiner threshold. We estimated time-dependent survival probabilities with the Kaplan–Meier method and compared groups using the log-rank test. We obtained expression data of PISD, Drp1, Mfn1, and Mfn2 at 5 years from initial diagnosis from Oncomine (www.​oncomine.​org) using the TCGA Breast dataset.

We used a non-parametric Mann–Whitney U test for comparisons of cell migration and motility in the microfluidics device. For experiments comparing only two groups, we used a two-tailed, unpaired student’s t test. For experiments comparing multiple groups, we used one-way ANOVA and Tukey’s multiple comparisons test. We considered a significance level of p < 0.05 statistically significant. We prepared bar graphs (mean values + SD or SEM as denoted in figure legends) and box plots and whiskers using GraphPad Prism 7 or Origin 9.0. For box plots and whiskers, the bottom and top of a box define the first and third quartiles, and the band inside the box marks the second quartile (the median). The ends of the whiskers represent the 10th percentile and the 90th percentiles, respectively. For cell migration in our microfluidics device, the “□” inside the box indicates the mean, dots outside the box and whiskers indicate outliers, and the “x” refers to the maximum and minimum of all data. For all other box plots and whiskers, the “+” within the box refers to the mean. Dots outside the box and whiskers for the clinical data indicate the maximum and minimum values.

Using an orthotopic xenograft model, we previously showed that overexpression of phosphatidylserine decarboxylase (PISD) in breast cancer cells significantly reduced local tumor growth [15]. To investigate effects of PISD on metastasis independent of local tumor growth, we used two complementary models that bypass local tumor growth and/or invasion and selectively interrogate tumor extravasation and growth at sites of metastatic seeding. First, we injected MDA-MB-231 human breast cancer cells stably expressing PISD (MDA-MB-231-PISD) or control (WT) breast cancer cells (Figure S1A) directly into the left ventricle of female NSG mice. Breast cancer cells expressed click beetle green (CBG) luciferase for bioluminescence imaging of experimental metastases. Twenty-one days after injection, whole-body bioluminescence imaging revealed significantly reduced total metastatic burden in mice injected with MDA-MB-231-PISD cells (Fig. 1a). Liver, but not lung, metastases occurred in significantly fewer mice injected with MDA-MB-231-PISD cells (Table 1). Additionally, bioluminescence imaging of the flushed bone marrow from the femur and tibia demonstrated a trend toward reduced metastatic burden of MDA-MB-231-PISD cells relative to control (p = 0.074, Figure S1B).

As a second experimental model of metastasis, we focused on metastatic disease in the bone, one of the most common sites of breast cancer metastases. We injected MDA-MB-231-PISD or WT cells directly into femoral arteries of mice to selectively produce bone metastases. Computed tomography (CT) of tibias demonstrated less extensive osteolytic lesions in mice injected with MDA-MB-231-PISD cells (Fig. 1b). Together, these data establish that increased levels of PISD reduce the ability of breast cancer cells to metastasize.

PISD converts phosphatidylserine (PS) to phosphatidylethanolamine (PE) in mitochondria, so we expected stable overexpression of this enzyme to elevate amounts of PE in breast cancer cells. MALDI mass spectrometry showed that cultured human SUM159-PISD breast cancer cells contained ~ 3-fold higher levels of PE (770.55 m/z, Fig. 1c) than WT cells (Fig. 1d, e). To verify that stable expression of PISD elevates amounts of PE in vivo, we analyzed orthotopic tumor xenografts of PISD and WT cells by MALDI imaging mass spectrometry. Both tumors formed by SUM159- or MDA-MB-231-PISD cells contained ~ 2-fold higher levels of PE than tumors formed from matched WT cells (Fig. 1f and Figure S1C). Elevated levels of PE promote fission of membrane-bound organelles by inducing a large, negative membrane curvature [42], accounting for our prior observation that PISD expression promotes mitochondrial fission [15].

Our previous work suggests that PISD inhibits tumor growth and metastasis by inducing mitochondrial fission. To further explore this hypothesis, we next used compounds to enforce mitochondrial fission and fusion in cells. Since PISD expression increases levels of PE in cells, we treated cells with l-α-lysophosphatidylethanolamine (LPE), a cell permeant analog of the PISD enzymatic product that cells efficiently convert to PE within mitochondria [43, 44]. Using fluorescence microscopy, we found that treatment of WT cells with LPE drove mitochondrial fission with a similar phenotype to PISD cells (Fig. 2a–d). Conversely, we treated cells with leflunomide (Lef), an FDA-approved drug recently validated to stimulate expression of the mitochondrial fusion genes, mitofusins 1 and 2 (Mfn1/2) [45, 46], to promote mitochondrial fusion. Consistent with previous literature [45], leflunomide promoted mitochondrial fusion in PISD cells (Fig. 2a–d).

We next identified effects of LPE and leflunomide on functions of mitochondria. Mitochondrial fission induced by LPE slightly reduced mitochondrial mass, while fusion driven by treatment with leflunomide increased mitochondrial mass as assayed by MitoTracker Green (Figure S2A). WT cells treated with LPE (mitochondrial fission) exhibited lower mitochondrial membrane potential (Figure S2B), while leflunomide increased mitochondrial membrane potential in PISD cells (Figure S2B). We did not detect significant changes in cell size between WT, PISD, and cells treated with LPE or leflunomide (Figure S2C). Although mitochondrial morphology and membrane potential regulate oxidative phosphorylation, levels of ATP did not differ between WT and PISD cells for both SUM159 (Figure S2D) and MDA-MB-231 (Figure S2E) cells.

In addition to promoting mitochondrial fission, stable expression of PISD increased levels of PE in cells and tumors. Since phosphatidylethanolamine typically reduces ordered domains in cell membranes, we used a validated fluorescent probe of membrane polarity, laurdan [31], to quantify overall effects on degree of lipid packing, thickness of the lipid bilayer, and the rotational freedom of lipids within the bilayer. We found that expression of PISD significantly reduced plasma membrane order in both SUM159 (Fig. 2e, f) and MDA-MB-231 cells (Figure S3A-B). Consistent with the expression of PISD, we found that treating cells with LPE reduced membrane order and promoted membrane disorder in SUM159 (Fig. 2e, f) and MDA-MB-231 (Figure S3A-B) cells. Treatment of PISD cells with leflunomide did not change membrane order in either cell line. As a complimentary approach to drive mitochondrial fission, we stably expressed Drp1 in cells (Figure S4A-B) [19]. Stable expression of Drp1 promoted mitochondrial fission (Figure S4C-F). Consistent with LPE and PISD, stably expressing Drp1 reduced membrane order and promoted membrane disorder in SUM159 (Fig. 2e, f) and MDA-MB-231 (Figure S3A-B) cells. Together, these results demonstrate that lipid changes and mitochondrial phenotype directly impact membrane order in cells. More specifically, mitochondrial fission promotes membrane disorder.

Previous work demonstrates that dynamic changes in signaling inputs to mitochondria regulate mitochondrial morphology [47‐49]. Only limited information exists about reciprocal regulation between morphological heterogeneity of mitochondria and effects on cancer cell signaling. Furthermore, areas of increased membrane order, called lipid rafts, facilitate receptor activation and downstream signaling [50]. Since cells with greater mitochondrial fission show reduced order of membranes, we investigated to what extent enforcing changes in mitochondrial morphology regulated activities of Akt and ERK, two kinases in signaling pathways mutated in over 30% of all cancers [51]. Due to mutations in PI3K in SUM159 cells and KRas in MDA-MB-231 cells causing constitutive activation of Akt and ERK, respectively, we focused most assays on kinase activities of ERK in SUM159 and Akt in MDA-MB-231 cells, respectively. First, to identify effects of PISD on levels of Akt and ERK, we performed immunofluorescence staining for phosphorylated (activated) and total ERK in SUM159 and Akt in MDA-MB-231 cells, respectively, following treatment with serum. Immunostaining showed reduced activation of ERK (Fig. 3a, S5A-B) and Akt (Figure S5C-E) in cells stably expressing PISD compared to WT cells.

To identify more dynamic effects on signaling, we used breast cancer cells stably expressing fluorescent kinase translocation reporters (KTRs) for Akt and ERK. The KTR reporters utilize known downstream substrates specific for Akt or ERK. Phosphorylation of the substrate drives a reversible translocation of the reporter into and out of the nucleus (Figure S6A) [27, 28]. KTRs permit quantitative imaging studies of Akt and ERK kinase activities in single cells or populations. Since our KTR construct expresses H2B-mCherry to denote the nucleus, we performed remaining experiments with cells stably expressing a non-fluorescent PISD. qRT-PCR analysis revealed higher expression of PISD in stably transduced SUM159- and MDA-MB-231-PISD cells compared to WT cells (Figure S1D).

Using cells stably expressing KTRs for Akt and ERK, we found that PISD reduced both amplitude and duration of activation of both Akt and ERK in response to serum, as quantified by changes in localization of imaging reporters from nucleus (kinase “off”) to cytoplasm (kinase “on”) (Fig. 3b, c and S6B-C). While we designate “off” and “on” locations for KTRs, we emphasize these reporters give analog, not digital, measurements of kinase activity.

To further analyze effects of mitochondrial fission on cell signaling, we next subdivided cells into high, medium, and non-signalers based upon total area-under-the-curve for signaling. We found that the expression of PISD shifted cells to the non-signaling group in SUM159 (Figure S7A) and MDA-MB-231 (Figure S6D) cells, further demonstrating that mitochondrial fission inhibited cell signaling. Comparing average signaling curves over time for ERK (Figure S7A) and Akt (Figure S6D) in high, medium, and non-signalers, we found reduced peak amplitude of signaling in most groups in PISD cells.

To directly correlate mitochondrial morphology with signaling output, we analyzed mitochondrial morphology prior to addition of serum and kinase activities stimulated by serum in single WT cells. WT cells with the highest fraction of fused mitochondria showed significantly greater activation of ERK (Fig. 3d) and Akt (Figure S6E). Conversely, cells with greater fractions of fissioned mitochondria exhibited reduced activities of both kinases (Figure S7B-C). Together, these data show that mitochondrial fission correlates with reduced activation of Akt and ERK, while mitochondrial fusion promotes signaling by these kinases.

To further analyze effects of mitochondrial morphology on signaling through Akt and ERK, we enforced mitochondrial fission and fusion with LPE and leflunomide, respectively, and then measured signaling in response to serum. Consistent with results comparing PISD and WT cells, we found that increasing mitochondrial fission with LPE blunted activation of Akt and ERK in WT cells, while driving fusion with leflunomide potentiated signaling (Fig. 4a–d). We observed similar results in cells stimulated with epidermal growth factor (EGF). Again, mitochondrial fission driven by expression of PISD or treatment with LPE blunted ERK and Akt signaling in response to EGF (Fig. 4e–h). Cells with a greater extent of fused mitochondria responded to EGF with enhanced activation of Akt and ERK. We also tested effects of stably expressed Drp1 in both SUM159 and MDA-MB-231 cells. Consistent with PISD and WT cells treated with LPE, stable expression of Drp1 also reduced signaling to Akt and ERK in response to serum (Figure S8A-D) and EGF (Figure S8E-H). Collectively, these data show that mitochondrial fragmentation inhibits ERK and Akt signaling in response to various stimuli.

Because mitochondrial fission results in reduced signaling, we hypothesized that mitochondrial fission versus fusion phenotypes function as general regulators of processes driving metastasis, such as migration and invasion. We treated breast cancer cells with LPE or leflunomide and then quantified two-dimensional migration toward serum in a microfluidics device [15]. Consistent with our previous results, cells stably expressing PISD migrated less than corresponding WT cells (Figure S9A-B). We also found that WT cells treated with LPE migrated less than control WT cells, while treating PISD cells with leflunomide increased migration (Figure S9A-B). To identify effects of mitochondrial morphology on migration and invasion in a more complex 3D environment, we next embedded co-culture spheroids of WT or PISD breast cancer cells with human mammary fibroblasts (HMFs) into fibrin gels and treated cells with LPE, leflunomide, or vehicle control. Consistent with migration in our microfluidic device, we found that spheroids with PISD cells displayed a smaller diameter (Figure S9C-D), suggesting PISD reduces cell migration and invasion. Protrusions from the core of spheroids with PISD cells did not extend as far as spheroids from WT cells (Figure S9C and S9E). Furthermore, treating spheroids from WT cells with LPE reduced sphere diameter and invasion into the surrounding extracellular matrix (ECM). Enforcing mitochondrial fusion with leflunomide in spheroids with PISD cells rescued inhibitory effects on spheroid size and invasion (Figure S9C-E). We also observed that stable expression of PISD reduced overall movement of breast cancer cells on tissue-engineered constructs with aligned fibrils of fibronectin (Figure S9F-G). Overall, these data show that mitochondrial morphology dictates cell migration and invasion.

To directly assess effects of mitochondrial morphology on metastasis, we injected SUM159-WT, WT treated with LPE, PISD, PISD treated with leflunomide, and Drp1 cells expressing CBG luciferase directly into the left ventricle of female NSG mice. We only treated cells with LPE or leflunomide ex vivo and did not administer these agents to mice after injection of breast cancer cells. One day after injection, whole-body bioluminescence imaging showed that mitochondrial fission induced by PISD or treatment with LPE significantly reduced signal in lungs (Fig. 5a, b). Furthermore, ex vivo treatment of cancer cells with leflunomide modestly increased signal in the lung for PISD cells 1 day after injection (Fig. 5a, b). As measured by area-under-the-curve analysis of bioluminescence over 20 days post-injection, interventions that produced mitochondrial fission prior to injection (LPE treatment of WT cells) and stably throughout the experiment (PISD and Drp1) reduced overall bioluminescence signal compared to WT cells (Fig. 5c). Although WT and Drp1 cells did not differ 1 day after injection (Fig. 5a, b), stable expression of Drp1 reduced subsequent growth of breast cancer cells (Fig. 5c). Additionally, PISD cells treated with leflunomide ex vivo to drive mitochondrial fusion produced significantly higher metastases than matched PISD cells (Fig. 5c). Quantification of organ-specific metastases per group revealed that mice injected with cells with a fissioned morphology showed reduced metastases in the lung and liver (Fig. 5d). Using leflunomide to drive mitochondrial fusion in PISD cells overcame inhibitory effects of PISD on metastases to the lung and liver (Fig. 5d).

We measured Akt and ERK signaling in single cells by two-photon microscopy of Akt and ERK KTRs in SUM159 lung metastases. Shifting mitochondrial morphology through pharmacologic approaches prior to injection (WT cells treated with LPE) or stable gene expression (PISD, Drp1) significantly reduced ERK signaling (Fig. 5e, f). PISD cells and WT cells treated with LPE also showed reduced activity of Akt in lung metastases. As compared with signaling in PISD cells, ex vivo treatment with leflunomide resulted in greater activities of Akt and ERK. Imaging studies of the bone marrow also revealed lower Akt and ERK signaling in metastatic PISD cells in the femur and tibia following femoral artery injection (Figure S10A-C). Overall, these results demonstrate that mitochondrial fission inhibits metastatic potential by limiting Akt and ERK signaling.

Since mitochondrial fission inhibits metastasis and kinase signaling by Akt and ERK, we hypothesized that the expression levels of mitochondrial morphology genes correlate with patient prognosis. We first analyzed the expression of PISD in primary circulating tumor cells (CTCs) from patients with advanced breast cancer. Compared to the expression of PAI1, a gene known to correlate with worse outcome in patients [17], or EPCAM, a gene that is commonly overexpressed in breast cancer, breast CTCs expressed low levels of PISD (Fig. 6a). Survival analysis revealed that low PISD expression correlated significantly with poorer metastasis-free survival (MFS, Fig. 6b). We also found that high expression of mitofusins (Mfn1/2) correlated with poorer MFS to the bone and brain in patients with breast cancer (Fig. 6c). High levels of both Mfn1 and Mfn2 synergistically reduced MFS, suggesting interactions among these drivers of mitochondrial fusion in disease progression. Data from the TCGA for breast cancer demonstrated higher levels of PISD (Fig. 6d) and Drp1 (Fig. 6e) in patients alive 5 years after initial diagnosis. Patients alive 5 years after presentation also express lower levels of mitochondrial fusion genes Mfn1 (Fig. 6f) and Mfn2 (Fig. 6g). In summary, high levels of mitochondrial fission genes correlate with better prognosis for patients with breast cancer.