The triple negative metastatic mouse cell line, 4T1, was obtained from the American Type Culture Collection. All cell lines were grown in full DMEM medium, which was supplemented with 10% (v/v) foetal bovine serum (ThermoFisher Scientific) and GlutaMAX-1 (2mM) and maintained at 37 °C and 5% CO₂ in a humidified incubator. Penicillin (100 U/mL) and streptomycin sulphate (100 μg/mL) were added to ex- vivo cell cultures only. Full DMEM was further supplemented with uridine (50 μg/μL) and pyruvate (1 mM) for ρ⁰ cells. The ρ⁰ cells were derived previously by long-term culture (10-12 weeks) of 4T1 cells with 50 ng/mL ethidium bromide (11). Loss of mtDNA was monitored by the absence of the mitochondrial Cytb gene by PCR, and by pyruvate/uridine auxotrophy.

Because of the insensitivity of 4T1 cells to the cytostatic antimetabolite, 6-thioguanine (6TG), this drug is used to enumerate tumour cells from tissue. The ρ⁰D5 cell line was established previously from cells isolated in the presence of 60 µg (6TG), 5 days after subcutaneous injection of 10⁶ 4T1ρ⁰ cells into the flank of female Balb/C mice (12). The ρ⁰LM cell line was obtained previously from metastatic lung lesions in the presence of 60 µg 6TG after subcutaneous injection with 10⁶ 4T1ρ⁰ cells (11). The ρ⁰D5-LM cell line was generated from metastatic lung lesions in the presence of 60 µg 6TG after orthotopic injection of 10⁶ ρ⁰D5 cells. The ATP5B-KO3.1 cells were derived through serial dilution from ATP5B-KO3 cells, that were generated previously (12), using the CRISPR/CAS system (14).

The Incucyte S3 Live-Cell Analysis System (Sartorius) was used to measure confluency to evaluate cell proliferation rates and growth patterns of the 4T1, ρ⁰ and ρ⁰D5 cell lines. Cells were seeded in 6 well plates (2000 cells/well in full DMEM, with uridine and pyruvate supplementation for ρ⁰ cells only) and incubated at 37°C, 5% CO₂ for 14 days, with media refreshed on day seven. Phase contrast scans (at 10x) were taken every 8 hours across 40 fields of view, analysed using Incucyte software and exported to excel spreadsheets to generate growth curves. Bright field images of individual fields of view were used to visualise growth patterns.

The Seahorse XF96 extracellular flux analyser (Agilent) was used to measure cellular oxygen consumption rates (OCR) as per manufacturer’s instructions. Cells were seeded (30,000 per well) in Seahorse XF96 cell culture microplates pre-coated with poly-D-lysine solution (Agilent) the day before the experiment. After 15-18 h, the medium was replaced with Seahorse XF base medium supplemented with 5 mM HEPES, 10 mM glucose and 1 mM pyruvate with pH adjusted to 7.4, and plates were incubated without CO₂ for 60 min. The Mito-stress assay protocol consists of four consecutive injection steps of 1 μM oligomycin, 2 μM FCCP, 0.5 μM rotenone/0.5 μM antimycin A and lastly 5 μg/ml Hoechst 33342 to determine cellular abundance. Data were normalized to Hoechst staining intensity (absorption at A490, Tecan Plate reader). Raw datasets were exported to excel to generate oxygen consumption profiles.

Balb/c mice were housed in the Biomedical Research Unit at the Malaghan Institute of Medical Research. Female mice between six and ten weeks of age were used for intravenous (IV) or orthotopic (mammary fat pad) injection of tumour cells. All animal experimentation was carried out under animal ethics approval from Victoria University of Wellington (#29071). Lungs were perfused with 20mL PBS before removal; the left lobe was embedded and stained, the right lobes were used to generate metastases and to derive cell lines. For the orthotopic metastasis model, 10⁶ tumour cells in 50 μL PBS were injected into the fourth mammary fat pad of female Balb/c mice using a 25 mL gauge needle. Mice were culled via CO₂ to avoid damaging the arteries needed for perfusion, when a tumour reached 0.5cm³ or ulcerated, or if mice lost 10% bodyweight in 24h or were otherwise unwell. Lungs were excised and used in metastases assays. Mice that did not develop primary tumours were culled at 60 days and tissue around the injection site was macerated and incubated in the presence of 60 µg 6TG in T75 flasks to check for the presence of latent tumour cells. For the IV injection experimental metastasis model, 10⁶ tumour cells in 50 μL PBS were injected into the lateral tail vein using a 25 mL gauge needle. Animals were culled using a CO₂ chamber 3 weeks later.

Lung sections were macerated in a petri dish, type 4 collagenase (1 mg/mL in DMEM) was added, and samples were incubated in a shaking incubator (88 rpm, 37°C for 3 h). Digested tissue was strained and centrifuged. Cell pellets were resuspended in DMEM in the presence of 60 µg 6TG in T75 flasks. Medium was replaced after 7 days and flasks were scored for the presence or absence of colonies 7 days later.

Lung sections for use in tumour histology were perfused with 20mL PBS before being incubated in 4% PFA for 24 hours at 4°C. Samples were dehydrated with a series of EtOH incubations, cleared with xylene, embedded in paraffin wax and sectioned on a Microm HM 325 Rotary Microtome at a thickness of 5 μm, and mounted on slides. FFPE samples were dewaxed, dehydrated and stained with haematoxylin (2.5 min) and eosin (1 min), washed, dehydrated, cleared and cover-slipped. H&E sections were imaged on the VS200 Slide Scanner (Olympus) using brightfield settings. OIR files were visualised using QuPath.

Cells were plated in an 8-well chamber slide at 3 x 10⁴ cells/well and incubated at 37°C overnight. Cells were washed (3x, PBS), fixed (4% PFA, 15 min at RT), washed (3x, PBS) and permeabilised (0.1% Triton X-100, PBS, 15 min on ice), washed, blocked (5% BSA in PBS; 1 h on ice) and incubated with anti-ATP5B-AF488, clone 4.3E8.D10 (Invitrogen MA1-930-A488, LOT WF330749) (1.25 µg/mL) in 0.1% Triton X-100 + 1% BSA in PBS (1 h at RT) and washed and stained with 10 ng/mL DAPI (in PBS) or nucGreen (10 min at RT). Images were acquired as z-stacks across the focus depth, in an Olympus inverted microscope IX83 equipped with Laser Scanning Confocal Microscope head (FV3000) with a 405nm (50 mW), 514 nm (40 mW), 516 nm excitation lasers and sensitive detectors with emission bands of 430-470, 530-580 and 610-respectively. Confocal image files were pre-processed using a FIJI macro to compress z-stacks (15). CellProfiler pipelines were generated to quantify cell ATP5B staining (16).

Cells were plated in an 8-well chamber slide as above. Cells were washed (3x PBS) and incubated (45 min at RT) in 300 mL staining mixture, consisting of SYBR gold (1:40,000), MitoTracker CMXRos (1: 20,000; 50nM) and Hoechst 33342 (ThermoFisher Scientific) (1:10,000). Images were acquired as above. Maximum Intensity projection images were segmented by nuclei (Hoechst), cytoplasm (MitoTracker CMX-ROS) and mitochondrial DNA fluorescent signals (SYBR Gold). Each mtDNA nucleoid was counted and allocated to its respective cytoplasm segmentation excluding the nuclear area (see Supplementary Figure 2 ). Images were processed and analysed using FIJI software (15) and CellProfiler (16). A three-dimensional reconstruction was made using IMARIS (Bitplane, v 9.8.0).

RNA from the different 4T1 cell lines was extracted (Qiagen RNAeasy Mini Kit) and quantified using the QuantiFluor^® RNA System (Promega). RNA (500 ng) was used for library preparation for Oxford Nanopore Technologies (ONT) long-read cDNA sequencing. Only cDNA greater than 20 ng/uL was used in sequencing libraries. Briefly, this process involves reverse transcription of extracted RNA using a custom poly TVN primer that binds to polyA RNA sequences; second-strand synthesis using ONT-provided strand-switch primer, followed by PCR to incorporate barcoded ONT rapid attachment primers, then binding of rapid adapters. Samples preparation varied between early R9.4.1 runs and later R10.4.1 runs (in August and September 2023). Early samples from R9.4.1 runs were processed according to the latest ONT cDNA protocol available at the time of sequencing (e.g SQK-PCS108 with SQK-PBK004, SQK-PCB109). Up to 6 samples were multiplexed together in equimolar quantities for each run (with technical duplicate PCR reactions for some samples), then run on an ONT MinION using an R9.4.1 MinION flow cell. Later samples from the R10.4.1 runs were processed according to an in-house developed protocol that combines the ONT SQK-PCB111 PCR-cDNA barcoding kit with Kit14 rapid adapter, to allow the sequencing to be carried out on R10.4.1 flow cells. Up to 14 samples were multiplexed together in equimolar quantities for each run (with technical duplicate PCR reactions for some samples), then run on an ONT P2 Solo using an R10.4.1 PromethION flow cell. Our current protocol for this is attached.

The program LAST was used to identify ONT barcodes present in sequenced reads, followed by a customised program (17), designed to use barcode assignments to demultiplex reads into files based on their incorporated barcodes. Demultiplexed reads were then mapped to the ONT strand switch primer sequence in order to identify the direction of transcription. Reads were mapped to the mouse transcriptome using LAST, grouped by mapped transcript, and counted producing a table that was further processed using DESeq2 (18, 19) to determine differential expression (log₂ fold change) of ρ⁰SC vs ρ⁰LM and ρ⁰D5 vs ρ⁰D5LM. We used FDR-adjusted p values reported by DESeq2 (20) of 0.1 as a threshold for statistical significance. Whole-transcriptome processing differed between early R9.4.1and later R10.4.1 runs. For all R9.4.1. runs, reads were base-called with guppy v5.1.15-v6.4.8 using the most recent version of the generic DNA model at the time of calling in super-accuracy mode (dna_r9.4.1_450bps_sup). Older runs were recalled using this newest model to maintain calling consistency. For all R10.4.1 runs, reads were base-called with Dorado v0.4.0 + 0aaf16d using the most recent version of the generic DNA model at the time of calling in super-accuracy mode (dna_r10.4.1_e8.2_400bps_sup@v4.2.0). Reads were then demultiplexed using LAST [https://dx.doi.org/10.17504/protocols.io.14egnxw4zl5d/v8], oriented so that the sequence direction matched the transcription direction [https://dx.doi.org/10.17504/protocols.io.5qpvon2zzl4o/v7], then mapped to an A/T homopolymer-masked version of the Gencode M28 mouse transcriptome using LAST and converted into transcript count tables [https://dx.doi.org/10.17504/protocols.io.5qpvonn2bl4o/v17].

Count tables were processed using DESeq2 v1.40.2, using a statistical model that incorporated cell lines and sequencing runs as independent factors (i.e. “design = ~ line + run”). Differentially-expressed log fold-change values were shrunk using the apeglm method [10.1093/bioinformatics/bty895], including the shrunk adjusted p-value calculation. Variance-stabilised values (used for expression plots) were generated using the ‘vst’ function of DESeq2, and linearly scaled to set the minimum log expression level to 0 while maintaining the 99th percentile expression level.

Shrunk log fold change values were used as input in the ‘fgsea’ function of fgsea v1.26.0, splitting and separately calculating enrichment for upregulated and downregulated genes, matching against mouse gene sets H, C1, C2, and C3 from the ‘msigdb’ package. All hallmark gene sets were summarised, showing normalised enrichment scores and unadjusted p-values produced in the fgsea results.

4T1 cells were fixed and processed according to methods published in protocol 4.12 of the Association of Genetic Technologists (AGT) cytogenetics manual (21). Fixed 4T1 cytogenetic cell suspensions were fixed and slides were dehydrated in a 70%, 80%, and 95% ethanol series (3 min each). Following air drying, 1 μL of mouse anti- ATP5B/Cen10 FISH probe (Empire Genomics, USA) was resuspended in 4 μL hybridization buffer (Empire Genomics) and pipetted onto a 22x22 mm glass coverslip and co-denatured at 75°C for 4 min, incubated overnight at 37°C in a humidified chamber to permit hybridization of the probe to the target DNA. Following post-hybridization washes, ProLong™ Diamond Antifade mountant with DAPI (Thermo Fisher Scientific) was added. Slides were analysed using an Olympus VS200 slide scanner for fluorescent image analysis using a 60x oil objective. Nuclei on each FISH slide were analysed for both ATP5B and Cen10 probe signals.

A comparison of growth rates ( Figure 1A ) and growth patterns ( Figure 1B i-iii) of the ρ⁰D5 cells with ρ⁰ and WT cells showed that ρ⁰D5 cells grew slower and in a clumpy growth pattern compared to WT cells; ρ⁰ cells grew very slowly in tight clusters and only reached 20% confluency. When injected into the mammary fat pads of female Balb/c mice, ρ⁰D5 cells generated both primary tumours and lung metastases ( Figures 1C, D , Table 1 ).

We hypothesized that ρ⁰D5 cells would improve their respiratory capacity by obtaining additional healthy mitochondria from the mouse, similar to ρ⁰ cells. We used the Mitostress protocol of the Seahorse extracellular flux analyser to evaluate oxygen consumption rates (OCR) before and after exposing cells to the respiratory complex V inhibitor, oligomycin ( Figure 1E ). ATP-linked OCR can be considered a measure of OXPHOS. Comparing Seahorse profiles, basal OCR and ATP-linked OCR of ρ⁰ and ρ⁰D5 cells with those of the corresponding metastatic lung cells, we confirmed previous findings that basal OCR and ATP-linked OCR of ρ⁰LM cells closely resembled that of WT cells (p=0.340 and 0.235 respectively; t-test) ( Figure 1E ). However, contrary to expectation, the basal and ATP-linked OCR of ρ⁰D5LM cells isolated from lung lesions remained similar to those of ρ⁰D5 cells (p=0.363 and 0.300 respectively; t-test) ( Figure 1E ).

In summary, WT and ρ⁰LM had robust OXPHOS levels but both ρ⁰D5 and ρ⁰D5LM cells had very poor OXPHOS levels.

Previous research with ATP5B KO3 and KO7 cells showed that these cells had no OXPHOS, yet generated primary tumours (12). We initially confirmed these findings for ATP5B-KO3 cells. However, we found that this cell line sporadically expressed ATP5B protein in 1-2% of cells ( Supplementary Figure 1A ). This small number of ATP5B-expressing cells expanded rapidly over time ( Supplementary Figure 1B ). ATP5B expression correlated to increased ATP-linked OCR ( Supplementary Figure 1C ). Although ATP synthase activity was completely inhibited by oligomycin, this activity was re-established soon after oligomycin was removed from the medium ( Supplementary Figure 1D ) and therefore this approach was unsuitable for in vivo experiments.

We removed ATP5B-expressing cells from ATP5B KO3 cells by means of serial dilution of an early culture and found that ATP5B KO3.1 cells did not express the ATP5B protein over a 12-month period. Representative images of the presence/absence of the ATP5B protein in WT cells and ATP5B KO3.1 cells are shown in Figures 2A, B , respectively. Figure 2C shows low basal OCR of ATP5B KO3.1 cells (5% of 4T1 cells), similar to that of ρ⁰D5 cells (p= 0.721; t-test) and ρ⁰D5LM cells (p=0.173, t-test). In addition, ATP5B KO3.1 cells had no ATP-linked OCR ( Figure 2C ). The clonal status of ATP5B KO cell lines was explored by nuclear in situ hybridization using ATP5B/Cen10 FISH probes, showing that WT, ATP5-KO3 and ATP5B-KO7 cells contained both diploid and triploid cells, but ATP5B-KO3.1 cells were exclusively triploid ( Supplementary Figures 1D, E ).

In summary, we showed that ATP5B-KO3.1 cells could not perform OXPHOS.

Differences between basal OCR of ρ⁰D5/ρ⁰D5LM, ρ⁰D5/ATP5B-KO3.1 and ρ⁰D5LM/ATP5B-KO3.1 cells were not significant (p=0.363, 0.721 and 0.173, respectively; t-test) ( Figure 3A ). Energy profiles of the 4T1 cell lines shown in Figure 3B were generated by plotting basal OCR (mito-OCR) against extracellular acidification rates (ECAR) under unstressed and stressed conditions. The increase in OCR after exposure to the ionophore FCCP, which dissipates the proton gradient, is often referred to as the respiratory reserve, and is an indication of maximum OCR. The increase in ECAR after oligomycin addition can be referred to as glycolytic reserve, which occurs because blocking OCR will result in attaining maximum levels of glycolytic rates.

With respect to metastatic capacity, ATP5B-KO3.1 cells were unable to generate primary tumours when injected orthotopically in female Balb/c mice (n=17/18) ( Table 1 ). Tumour cells of the one mouse that did develop a primary tumour, expressed the ATP5B protein ( Supplementary Figure 1F ). The ρ⁰D5 cell line generated primary tumours and lung metastases in 15 out of 19 mice whereas ρ⁰D5LM cells from 4 separate clones formed primary tumours in 13 out of 14 mice in 4 separate experiments, but only 2 out of the 14 mice developed lung metastases ( Table 1 ). In order to further explore the ability of cell lines generated from lung lesions to form lung metastasis themselves, we orthotopically injected a small number of mice (n=5) with fully respiratory-competent ρ⁰LM cells (11). These cells were previously shown to form primary tumours (in 10/10 mice), and in our experiments, 4 out of 5 mice developed primary tumours. However, none of these 4 mice developed lung metastases. Using a different model of lung seeding, whereby tumour cells are directly injected into the tail vein, we found that ρ⁰D5LM cells did form lung nodules (n=7/7) in this model, unlike ATP5B-KO3.1 cells (n=0/5).

In summary, metastatic ability does not appear to be linked to either OCR, OXPHOS or respiratory reserves in our cell lines. Both WT and ρ⁰SC cells with robust OCR, OXPHOS and stress responses as well as ρ⁰D5 cells with very poor OCR, OXPHOS and stress responses all consistently generated metastases. In contrast, poorly performing ρ⁰D5LM cells and highly performing ρ⁰LM cells only generated metastases in very few of the mice. ATP5B-KO3.1 cells with no OXPHOS did not generate primary tumours.

Because of the failure of ρ⁰D5LM cells to improve their respiratory capacity, we explored whether or not these cells were unable to acquire respiration-competent mitochondria, in contrast to ρ⁰ cells (11). We assessed the number of mtDNA nucleoids in individual cells using the mtDNA-specific dye SYBR Gold (22) to stain mtDNA nucleoids, MitoTracker CMXROS (23) to stain mitochondria and Hoechst 33342 to stain nuclei. Representative three dimensional renderings of WT, r0, r0D5 and r0D5LM are shown in Figures 4A–D . An Image analysis pipeline analysed cells based on their mitochondrial content (MitoTracker CMXROS-positive area). The SYBR Gold signal located underneath or on top of the nucleus was subtracted and the number of mtDNA nucleoids per cell calculated and tabulated ( Supplementary Figures 2A-F ). The results are presented in Figure 4E and validate previous findings that only 10-15% of ρ⁰D5 cells contain mtDNA (12).

In summary, low respiratory capacity was not linked to low mitochondrial DNA (mtDNA) abundance.

Next, we explored whether or not low mtDNA transcription levels could explain the inability of ρ⁰D5LM cells to increase OCR. Figure 5 shows mtDNA transcription levels, using MinION nanopore RNA sequencing, for the mitochondrially-encoded genes m-Nd1 (RCI), mt-Cytb (RCIII), mt-Atp6 (RCV) and mt-Rnr2 (16S-RNA) (top 4 squares in Figure 5 ) as well as nuclear-encoded genes of the same respiratory complexes and mitochondrial ribosomal protein 27 (bottom 4 squares in Figure 5 ) for the different cell lines. mtDNA transcription was similar between WT, ρ⁰D5 and ρ⁰D5LM cells, ρ⁰SC and ρ⁰LM cells with mtDNA expression close to zero for ρ⁰ cells. There was also little difference in transcription of nuclear-encoded respiratory subunits between any of the cell types, consistent with previous research (24).

In summary, low respiratory capacity was not linked to poor mtDNA transcription.

Because both cell lines derived from lung metastases could not generate lung metastases themselves, we performed a gene set enrichment analysis (GSEA) of LM cell lines against their parental cell lines for the 50 hallmark gene sets ( Supplementary File 1 ) (25). Figures 6A, B show the GSEA analysis for both cell sets; Figure 6C displays the pathways with a normalised enrichment score (NES) > 1.3 and at least 20% of genes of the set enriched. Only inflammation was enriched in both primary cell lines, suggesting that different factors may affect metastatic progression in the two primary cell lines.