Metabolic adaptability in metastatic breast cancer by AKR1B10-dependent balancing of glycolysis and fatty acid oxidation

The different stages of the metastatic cascade present distinct metabolic challenges to tumour cells and an altered tumour metabolism associated with successful metastatic colonisation provides a therapeutic vulnerability in disseminated disease. We identify the aldo-keto reductase AKR1B10 as a metastasis enhancer that has little impact on primary tumour growth or dissemination but promotes effective tumour growth in secondary sites and, in human disease, is associated with an increased risk of distant metastatic relapse. AKR1B10High tumour cells have reduced glycolytic capacity and dependency on glucose as fuel source but increased utilisation of fatty acid oxidation. Conversely, in both 3D tumour spheroid assays and in vivo metastasis assays, inhibition of fatty acid oxidation blocks AKR1B10High-enhanced metastatic colonisation with no impact on AKR1B10Low cells. Finally, mechanistic analysis supports a model in which AKR1B10 serves to limit the toxic side effects of oxidative stress thereby sustaining fatty acid oxidation in metabolically challenging metastatic environments.

reductase, Akr1b8. Akr1b8, and its human orthologue AKR1B10 3 are NADPHdependent enzymes that can reduce a variety of carbonyl substrates 4 . These include the conversion of retinal to retinol 5,6 resulting in decreased retinoic acid signalling, conversion of the isoprenyl aldehydes farnesal and geranylgeranal to farnesol and geranylgeranol 7 generating precursors for protein prenylation and the reduction of cytotoxic aldehydes 8 . Although AKR1B10 expression is upregulated in a variety of cancers including hepatocellular 9,10 ; lung 11 , pancreatic 12 and breast 13,14 , the mechanism by which elevated levels of AKR1B10 enhances metastasis is not known.
We demonstrate that AKR1B10 High cells are characterised by a reduced glycolytic capacity and an increased utilisation of fatty acid oxidation (FAO), and that this metabolic switch is required for successful colonisation of secondary sites but not primary tumour growth or metastatic dissemination.

Akr1b8/AKR1B10 promotes breast cancer metastasis
To identify novel enhancers of breast cancer metastasis we analysed a syngeneic in vivo shRNA screen, focusing on shRNAs that were significantly under-represented in the 4T1-Luc tumour-bearing lungs of BALB/c mice compared to preinoculation 4T1-Luc mouse mammary carcinoma cells ( Fig. 1a; see Methods). 81 shRNAs were found to be significantly depleted (Z score <-2) in the metastatic lung samples (Fig. 1b) and were then filtered by removing shRNAs that (a) did not align to the predicted target gene, (b) were significantly depleted in less than 3 of the 4 biological replicates, (c) targeted genes with expression in the lowest 50 th percentile based on gene expression profiling of 4T1 cells directly isolated from tumours 15 , and (d) when comparing the preinoculation cells to the initial plasmid library (Fig. 1a) showed a significant difference in abundance (Z score >2 or <-2) indicating an effect on cell viability. Filtering resulted in a shortlist of 23 significantly depleted shRNAs targeting genes encoding putative metastasis enhancers ( Fig. 1c and Extended Data Table 1).
Of particular interest was the presence of the metabolic enzyme aldo-keto reductase 1b8 (Akr1b8) in this shortlist. The human orthologue of Akr1b8, AKR1B10, has been reported to be upregulated in a number of cancer types including breast cancer 13,14 , but the clinical and metabolic consequences of this altered expression have not been investigated. First, 4T1-Luc cells were transduced with lentiviral constructs containing empty vector (shCTRL), a non-targeting shRNA (shNTC) or two independent shRNAs targeting Akr1b8 (shAkr1b8-4 and shAkr1b8-7) (Extended Data Fig. 1a). Consistent with the screening data, where we compared shRNA representation in the starting plasmid pools with the preinoculation cells, Akr1b8 knockdown had no significant effect on cell viability when cultured in full medium in vitro (Extended Data Fig. 1b). By contrast, following intravenous inoculation, the two Akr1b8 knockdown cell lines showed a significant decrease in lung colonisation as monitored by in vivo IVIS imaging and ex vivo lung weight (Fig. 1d).
Although these data validate the in vivo shRNA screen, intravenous inoculation does not assess the full metastatic ability of tumour cells. Consequently, we next performed a spontaneous metastasis assay in which cells were inoculated orthotopically into 4 th mammary fat pad of BALB/c mice (Fig. 1e). No differences were observed in tumour take, primary tumour growth or tumour weight at the end of the experiment, however, there was a significant reduction in lung metastasis in the Akr1b8 knockdown group. Finally, we addressed whether this metastatic impairment resulted from impairment of tumour cell survival in the circulation. 4T1-Luc shNTC and shAkr1b8 cells were labelled with cell tracker dyes, mixed at a 1:1 ratio and Page 5 of 31 injected via the tail vein into BALB/c mice (Fig. 1f). Imaging of the lungs 1 hour postinjection confirmed that equal number of cells had been inoculated. Examination of lungs 16 hours post-injection revealed no significant difference between the number of control and Akr1b8-knockdown tumour cells retained in the lungs indicating that Akr1b8 expression does not impact on survival in the circulation or lodging in the vasculature but is required for efficient colonisation of tumour cells within the metastatic site.
Expression of AKR1B10 correlates with increased risk of metastatic relapse in breast cancer patients To address the clinical relevance of the data obtained with the 4T1 mouse models, expression of AKR1B10, the human orthologue of murine Akr1b8 3 , was analysed in human primary breast cancers present in the TCGA database. Within the intrinsic subtypes, AKR1B10 expression is significantly higher in the HER2-enriched and basal-like breast cancers compared to luminal A and luminal B cancers (Fig. 2a) and analysis by receptor expression revealed significantly higher AKR1B10 expression in ER-compared to ER+ breast cancers, and in HER2+ compared to HER2-breast cancers (Fig. 2a). The latter finding is consistent with a previous report that overexpression of AKR1B10 correlates with HER2 positivity in ductal carcinoma in situ (DCIS) 20 . An equivalent pattern of expression was seen in a panel of breast cancer cell lines 21 (Extended Data Fig. 2a). Similarly, AKR1B10 protein levels are variable with low levels in the ER+ ZR75.1 and MCF7 lines and high levels in the basal-like BT20, MDA-MB-468 and HCC1395 lines (Fig. 2b, upper panel). For further studies, AKR1B10 was ectopically expressed in the AKR1B10 Low MDA-MB-231 and MDA-MB-453 lines and expression was knocked down by shRNA in the AKR1B10 High HCC1395 line (Fig. 2b, lower panel). Levels of ectopically expressed protein were equivalent to that found in AKR1B10 High lines, while shRNA knockdown reduced protein levels to that observed in AKR1B10 Low lines.

Page 6 of 31
As with the 4T1-Luc cells (Extended Data Fig. 1b), the human breast cancer cell lines with altered AKR1B10 levels showed no difference in in vitro viability as monitored in a colony formation assay (Fig. 2c), yet when inoculated intravenously into BALB/c Nude mice, AKR1B10 High MDA-MB-231 cells gave rise to a significantly increased tumour burden in the lungs compared to AKR1B10 Low MDA-MB-231 cells (Fig. 2d). Again, there was no effect on the ability of the cells to survive in the circulation and lodge in the lung tissue (Fig. 2e), supporting the hypothesis that AKR1B10 functions to maintain efficient growth of tumour cells within the metastatic tissue.
Consistent with these findings, in a dataset of 1,746 unselected breast cancers 22 , high expression of AKR1B10 significantly correlated with reduced distant metastasis-free survival when considering all patients or only ER-patients. A similar trend was seen in HER2+ patients, however, the number of samples was too low to reach statistical significance (Fig. 2f). No association with outcome was seen in ER+ only patients. As AKR1B10 has been associated with chemoresistance via its ability to metabolise anti-cancer drugs 23 , we also examined the subset of untreated patients (Extended Data Fig. 2). Again high expression of AKR1B10 (upper quartile) was significantly associated with reduced distant metastasis-free survival in ER-, but not ER+, breast cancer patients. have been implicated as regulators of cellular metabolism. Aerobic glycolysis, also known as the Warburg effect, is a common feature of many cancers and characterised by increased metabolism of glucose to lactate, which is transported out of the cell resulting in local acidification. The Seahorse XF Glycolysis Stress test was used to assess glycolytic function of cells by measuring the extracellular acidification Page 7 of 31 rate (ECAR) in the media (Fig. 3a). Following addition of glucose, the glycolytic rate was significantly reduced in AKR1B10 High , compared to AKR1B10 Low , breast cancer cells, as was their glycolytic capacity and glycolytic reserve. Moreover, glucose uptake was significantly reduced in all three AKR1B10 High cell lines (Fig. 3b), indicating that AKR1B10 High cells have a reduced requirement for glucose. Consistent with this hypothesis, in 2D culture AKR1B10 High and AKR1B10 Low cells showed only a modest difference in cell growth when cultured in full DMEM (4.5 g/L D-glucose) but in low glucose (LG) DMEM (1 g/L D-glucose) AKR1B10 Low cells showed a significantly impaired growth rate (Fig. 3c). These data were recapitulated first in a 3D in vitro assay where AKR1B10 High tumour spheroids showed increased growth in LG DMEM compared to the AKR1B10 Low spheroids (Fig. 3d) and in colony forming assays where AKR1B10 High cells were significantly more tolerant to low glucose conditions ( Fig. 3e) In addition to aerobic glycolysis, tumour cells can utilise glutamine and/or fatty acids to generate sufficient ATP and metabolites to support cellular activities. As To address clinical relevance of these findings, we used a FAO 88-gene set (FAO88; see Methods) and demonstrated that AKR1B10 expression positively correlated with a high FAO88 score in triple negative (TN) and ER-breast cancer which AKR1B10 modulates these activities? AKR1B10 is distinguished from the other well-characterised AKR1B subfamily member AKR1B1 by its increased catalytic activity for retinals, isoprenyl aldehydes and, importantly, for cytotoxic aldehydes such 4-hydroxy-2-nonenal (4-HNE) 23 . The latter is a toxic lipid peroxide by-product of the elevated reactive oxygen species (ROS) levels associated with oxidative stress. The interaction between FAO and ROS is complex. It is well documented that FAO, via its ability to generate NADPH, reduces ROS levels 26 (Fig. 5d,e). However, lipid peroxidation levels did not increase when AKR1B10 High cells were cultured in LG DMEM, suggesting that the AKR1B10 functions to limit oxidative stress-associated toxicities and consequent FAO inhibition.
Second, do these cellular mechanisms operate in physiologically relevant settings? To address this, 3D tumour spheroids were treated with the FAO inhibitor etomoxir. Etomoxir had no effect on growth of AKR1B10 Low tumour spheroids but inhibited the increased growth observed in the AKR1B10 High tumour spheroids (Fig.   6a). More importantly, mice were inoculated intravenously with MDA-MB-231-Luc AKR1B10 High or AKR1B10 Low cells and, after 7 days when the tumour cells will have extravasated into the lung tissue, treated with or without etomoxir. As previously shown (Fig. 2d), MDA-MB-231 High cells gave rise to a significantly increased lung tumour burden as monitored by in vivo IVIS imaging and ex vivo measurement of lung weight (Fig. 6b) and this increased AKR1B10 High metastatic colonisation was effectively impaired by etomoxir treatment, with no effect of etomoxir on the growth of AKR1B10 Low cells.

Page 10 of 31
Together these data support a model in which AKR1B10 functions to maintain FAO in tumour cells, particularly during metastatic colonisation of the pro-oxidative lung microenvironment 2 .

Discussion
The data presented here demonstrates that AKR1B10 expression is elevated in ERand HER2+ breast cancers and that within these breast cancer subtypes, high AKR1B10 expression is associated with an increased incidence of metastatic relapse at secondary sites. In contrast to previous reports 29, 30 , we find that AKR1B10 High breast cancer cells do not display altered survival or proliferation properties when cultured in vitro in full medium (Fig. 2c, Fig. 3c,d) or when inoculated orthotopically into the fat pad of recipient mice (Extended Data Fig. 2). However, AKR1B10 High cells are more successful than AKR1B10 Low cells when cultured in nutrient poor conditions such as in low glucose (Fig. 3c,d) or when colonising the lungs (Fig. 1d, Fig. 2d and expression has been demonstrated to promote breast cancer metastasis in a variety of models systems 35,36,37 and to be associated with increased FAO and an enhanced ability of cells to survive in 3D acini assays 31 . Conversely, impairment of FAO decreases cell survival in acini assays 27 and reduces tumour burden in the lungs and livers following intravenous inoculation 38 . Here we demonstrate that AKR1B10 High cells fuel FAO by an increased uptake of exogenous fatty acids. To date, the best characterised fatty acid transporters are CD36, fatty acid translocase and low density lipoprotein receptor, and it is of particular interest is the recent identification of CD36 bright cells marking a population of metastasis-initiating cells 39 , and that these cells display an upregulated FAO signature.   Table 3) and selected in 2.5 µg/mL puromycin. The cells were cultured for an additional 3 passages in selective medium to enrich the infected cell population.

In vivo shRNA screen
As previously detailed 48   supplemented with 2.5 mM glucose, 0.5 mM carnitine and 5 mM HEPES, and the plate was incubated at 37°C for 1 hour in a non-CO 2 incubator. 30 µL of 1 mM palmitate-BSA substrate (Agilent) was loaded directly into port A of a Seahorse loading sensor cartridge and OCR measured on an XFe96 Analyzer.

Cell based assays
Colony formation assay. 0.2-5x10 4 cells were seeded per well in a 6-well plate. 7-10 days post seeding, plates were stained with crystal violet, dried and scanned at 1200 dpi using the GelCount colony counter system. Images were analysed using Fiji.
Cell viability assay. 1x10 2 cells in 100 µL medium were seeded per well in a 96-well plate and incubated at 37°C. Immediately after seeding, and every 24 hours Page 17 of 31 afterwards, cells viability was analysed by CellTiter-Glo (Promega). Fold change was calculated relative to the plate read at seeding (time 0). For 3D viability assays, 5x10 3 cells were seeded into ultra-low adherence U-bottomed 96-well plates (Corning). On day 6 tumour spheroids were lysed in CellTiter-Glo for 30 minutes and viability analysed using a Victor X5 plate reader. Where indicated, etomoxir (200 µM) or vehicle (DMSO) was added 24 hours after seeding.
2D cell growth. 1x10 3 cells were seeded per well in a 96-well plate and subject to live cell imaging (IncuCyte).
Glucose uptake assay. 1-5x10 4 cells were seeded in 100 µL culture medium containing 10% FBS into a 96-well plate and incubated for 24 hours at 37°C. Cells were washed twice with PBS before the Glucose Uptake-Glo assay (Promega) was performed according to the manufacturer's protocol. Images were analysed using basic algorithms in the CellProfiler software package (cellprofiler.org) to quantify oxidised (green) and non-oxidised (red) BODIPY probe.

Statistical analysis
Statistics were performed using GraphPad Prism 7. Unless stated otherwise, all numerical data is expressed as the mean ± standard deviation (SD) for in vitro assays and ± SEM for in vivo tests. Comparisons between 2 groups were made using the two-tailed, unpaired Student's t-test. Comparisons between multiple groups were made using one-way analysis of variance (ANOVA), and two-way ANOVA for comparisons between multiple groups with independent variables. Bonferroni posttesting with a confidence interval of 95% was used for individual comparisons.