Systematic integration of molecular profiles identifies miR-22 as a regulator of lipid and

folate metabolism in breast cancer cells

Costas Koufaris1,3, Gabriel N Valbuena1, Yotsawat Pomyen1,5, Gregory D Tredwell1,

Ekaterina Nevedomskaya4, Chung-Ho E Lau1, Tianlai Yang1, Adrian Benito1, James K Ellis1,

Hector C Keun1,2,*.

1Division of Cancer, Department of Surgery & Cancer & 2Centre for Systems Oncology and

Cancer Innovation, Imperial College London, Hammersmith Hospital, London W12 0NN,

UK

3Department of Cytogenetics and Genomics, Cyprus Institute of Neurology and Genetics,

P.O. Box 23462, 1683 Nicosia, Cyprus.

4Center for Proteomics and Metabolomics, Leiden University Medical Center, The

Netherlands

5Translational Research Unit, Chulabhorn Research Institute, Bangkok, Thailand

Corresponding author: Dr. Hector Keun, Division of Cancer, Department of Surgery &

Cancer, Imperial College London, Institute of Reproductive and Developmental Biology,

Hammersmith Hospital, London W12 0NN, UK. Tel: +44 (0)20 7594 3161. email:

h.keun@imperial.ac.uk

1

Abstract:

Dysregulated microRNA (miRNA) mediate malignant phenotypes, including metabolic

reprogramming. By performing an integrative analysis of miRNA and metabolome data for

the NCI-60 cell line panel we identified a miRNA cluster strongly associated with both MYC

expression and global metabolic variation. Within this cluster the cancer-associated and

cardioprotective miR-22 was shown to repress fatty acid synthesis and elongation in tumour

cells by targeting ATP-citrate lyase (ACLY) and fatty acid elongase 6 (ELOVL6), as well as

impairing mitochondrial one-carbon metabolism by suppression of methylenetetrahydrofolate

dehydrogenase/cyclohydrolase (MTHFD2). Across several datasets expression of these target

genes were associated with poorer outcomes in breast cancer patients. Importantly, a

beneficial effect of miR-22 on clinical outcomes in breast cancer was shown to depend on the

expression levels of the identified target genes, demonstrating the relevance of

miRNA/mRNA interactions to disease progression in vivo. Our systematic analysis

establishes miR-22 as a novel regulator of tumour cell metabolism, a function that could

contribute to the role of this miRNA in cellular differentiation and cancer development.

Moreover, we provide a paradigmatic example of effect modification in outcome analysis as a

consequence of miRNA-directed gene targeting, a phenomenon that could be exploited to

improve patient prognosis and treatment.

2

Introduction

Cancer cells exhibit altered metabolic behaviour compared to non-cancerous cells, enabling

them to survive in nutrient-deprived, poorly vascularized tumour regions and generate the

macromolecules required for rapid division (1, 2). These metabolic alterations include

enhanced synthesis of nucleotides and lipids from glucose and glutamine, increased nutrient

uptake and acid export, and amplified choline metabolism. The metabolic reprogramming of

cancer cells is a highly complex process affected by the identity of genetic lesions,

interactions with the microenvironment, and cell context. Substantial research now focuses on

elucidating the molecular mechanisms underlying the cancer metabolic phenotype. Improved

knowledge of these mechanisms could facilitate patient diagnosis/prognosis and the

identification of novel targeted and synthetically lethal metabolic interventions (3-5).

The function of mature miRNAs is to fine-tune the expression of their target genes by causing

mRNA degradation and/or translational repression, with each miRNA potentially regulating

hundreds of proteins (6). It is now recognized that miRNAs have considerable potential as

stratification markers and therapeutic targets in cancer. Several miRNAs have been attributed

oncogenic or tumour suppressive properties (7-9) and miRNA signatures are prognostic in

human cancers (10, 11). Some studies have also identified specific miRNAs involved in

regulating key aspects of cancer metabolism such as glycolysis (12-14), glutaminolysis (15),

and proline metabolism (16), although possibly many more miRNA metabolic regulators

remain to be discovered. Consequently, identification of the key miRNAs targeting metabolic

enzymes or their effectors in tumours could enhance understanding of the mechanisms

driving metabolic reprogramming of cancer.

The functional linking of miRNAs to metabolic phenotypes is hindered by the high false-

positive and false-negative rates of in silico target predictions and the laborious/time-

consuming process of experimentally verifying miRNA:mRNA interactions. Here we

explored the systematic analysis of the global relationship between miRNA variation and

3

metabolomic profiles as a method to prioritize miRNAs for further experimental

investigation. The bidirectional orthogonal projections to latent structures (O2-PLS)

algorithm is a multivariate regression model for integrating and defining predictive co-

variation between “omic” datasets (17). Several applications of the O2-PLS methodology

have been successfully carried out for combining transcriptomic and metabolomic data (18),

proteomic and metabolomic data (19) and data on different classes of lipid metabolites (20).

In this study we performed O2-PLS analysis on metabolomics (21) and miRNA expression

(22) profiles that had been generated previously for the NCI-60 cell line panel, a collection of

lines originating from nine different tumour types. Our systematic analysis led us to the

identification of miR-22 as an important regulator of cancer cell metabolism.

RESULTS

MYC-regulated miRNAs are associated with altered metabolism in the NCI-60 panel of

cell lines.

As a starting point for our analysis, an O2-PLS model was constructed to capture the co-

variation between global intracellular metabolite levels across the NCI-60 panel and

expression levels of miRNAs predicted to target metabolic genes. By randomly permuting the

metabolomic and miRNA profiles associated to each cell line and generating “null” models

(Fig.S1A-D) we were able to identify from the model parameters (individual variable Q2) a

subset of 13 miRNAs (23 probesets out of 316) and 18 metabolites (out of 154) that were

significantly correlated (Fig.1A-B). These associations were confirmed by direct inspection

of the univariate Pearson correlation coefficients within this subset of molecules (p<0.05 in

84% of the correlations displayed) (Fig.1C). The metabolite set identified by this model

included intermediates in glycolysis (3-phosphoglycerate), TCA cycle (citrate), one-carbon

metabolism (S-adenosylmethionine, SAM), lipid (palmitate, cholesterol) and nucleotide

metabolism (xanthosine), all pathways known to be significantly perturbed in cancer. The

4

miRNAs identified by our model in the NCI-60 dataset included miR-23a/b and let-7a,

miRNAs that have previously been identified as important metabolic regulators (13, 15, 16),

verifying the validity of our approach. All of the miRNAs identified by our analysis (miR-17-

92 cluster; miR-106-363 cluster; miR-23a/23b; miR-142a-3p; miR-22) have been reported

previously to be regulated by the oncogenic transcription factor c-Myc (MYC) (15, 23-25),

consistent with the hypothesis that the transcription factor fine-tunes metabolism through

miRNAs (15, 16). We confirmed a high degree of correlation between MYC expression and

the metabolome-associated miRNAs in the NCI-60 panel (Fig.1D). As the panel is very

heterogeneous and includes cell lines from hematological malignancies likely to be

specifically driven by MYC expression we repeated the multivariate analysis including only

cell lines derived from solid tumours and obtained a highly coincident set of associated

metabolites/microRNAs (Fig.S1E-F), albeit with lower statistical significance. These

observations suggested that a MYC-correlated microRNA module is associated with the

metabolic phenotype of cancer cells within the NCI-60 panel, irrespective of tumour site of

origin.

The miRNA miR-22 represses de novo fatty acid synthesis and elongation by targeting

ATP-citrate lyase (ACLY) and long chain fatty acid elongase (ELOVL6).

In our multivariate model, miR-22 showed the most robust association to metabolism (i.e. the

highest individual Q2 value) and high individual correlations to metabolites. This miRNA has

been proposed to affect cellular proliferation, differentiation and apoptosis via such targets as

ER, p21, TET2, SIRT1, HDAC4, and MYCBP (25-33), but direct regulation of metabolism

by miR-22 has not been previously reported. Consequently we focused on investigating the

regulatory functions of this miRNA on cancer metabolism. As a strategy to identify

metabolic targets of miR-22 we defined the intersection of genes (i) repressed by >1.5-fold

following over-expression of miR-22 in both breast MCF-7 (GSE17508) (33) and ovarian ES-

2 (GSE16568) (34) cell lines (ii) predicted to be targets of miR-22 by three or more target

prediction algorithms (Fig.2A &Suppl. File 1). Predicted direct targets of miR-22 were

5

enriched among genes repressed in both cell lines (7 out of 18; Fisher’s exact test, p<0.01).

Among the seven selected genes were MYCBP, a verified miR-22 target (33), three as yet

uncharacterized genes (C17orf58, PRICKLE4, and PRPF38A) and the nuclear receptor

coactivator NCOA1. Significantly, two metabolic enzymes, ACLY and ELOVL6, were

identified as putative miR-22 targets. ACLY is key regulator of tumour cell metabolism

catalyzing the conversion of citrate into acetyl-CoA for fatty acid or cholesterol synthesis

(35). Depletion of ACLY in cancer cells has also recently been shown to cause accumulation

of triglycerides and coincident depletion of ELOVL6 (36). ELOVL6 is a member of a family

of enzymes that are rate-limiting for the generation of fatty acids beyond 16 carbons and

preferentially converts palmitate to stearate (37). Consequently, miR-22 would be expected to

inhibit the pathway of de novo lipid synthesis by repressing ACLY and the elongation of fatty

acids by repressing both ACLY and ELOVL6 (Fig.2B). Genes repressed following miR-22

transfection were significantly enriched for pathways involved in fatty acid synthesis in both

MCF-7 and ES-2 cell lines (Suppl.File 2), consistent with this hypothesis.

To study the effect of miR-22 on ACLY and ELOVL6 we transiently over-expressed the

miRNA in MCF-7 cells (Fig.S2A), a tumour line with low basal levels of this miRNA, and

observed repression of mRNA (Fig.2C) and protein (Fig.2D) for the two genes.

Overexpression of the miRNA also suppressed the activity of luciferase reporters containing

the wild-type 3’-UTR of ACLY and ELOVL6 but had no effect when the putative miR-22

interaction site were mutated, implying direct and specific inhibition of these genes (Fig 2E).

To examine the metabolic consequences of the coordinated repression of these two key

metabolic enzymes by miR-22 we used isotopomer spectral analysis (ISA) (Fig. 2F) and GC-

MS measurements of transesterified palmitate and stearate in MCF7 cells cultured in U-13C6-

glucose (Fig. 2G-H) to model the incorporation of glucose-derived carbon into lipids (38,

39). MiR-22 over-expression caused a consistent reduction in the fraction of de novo

synthesized palmitate and stearate incorporated into lipids (g(t)) after 5 and 48 hours

incubation with labeled glucose (Fig. 2G, Fig. S3A-C) and reduced the fraction of lipogenic

6

acetyl-CoA derived from glucose after 48 hours incubation (Fig. 2G), supporting a link

between miR-22 and de novo lipogenesis. We also observed a decrease specifically in the

proportion of stearate synthesized by elongation of endogenous palmitate (Fig. 2G-H),

consistent with downregulation of ELOVL6. Direct suppression of ACLY and ELOVL6

using siRNA phenocopied the effects of miR-22 on stearate synthesis, providing further

evidence that the activity of these enzymes was relevant to the functional effects of this

miRNA (Fig. S4),

The miRNA miR-22 influences one-carbon metabolism by repressing bifunctional

methylene tetrahydrofolatedehydrogenase/cyclohydrolase (MTHFD2).

Beyond the role of miR-22 in the regulation of lipid metabolism, we had observed that in the

NCI-60 panel the miRNA was strongly anti-correlated to levels of SAM (Spearman’s rho=-

0.63, p=0.0002 significant after Bonferroni correction). The synthesis and recycling of SAM

is coupled to folate-dependent one-carbon metabolism; hence we hypothesized that this

miRNA regulated members of this metabolic pathway. Three enzymes of one-carbon

metabolism were predicted by at least three prediction algorithms to be targets of miR-22:

MTHFD2, MAT2A, and MTHFR (Suppl. File 1). To investigate the potential regulatory

effect of miR-22 on these three enzymes we examined MCF-7 cells over-expressing the

miRNA (Fig.S2A). Over-expression in MCF-7 cells of miR-22 resulted in a strong repression

of MTHFD2 mRNA and protein levels (Fig.3A-B), but not of MAT2A or MTHFR (Fig.S2B-

C). Similar repression of MTHFD2 protein levels was also observed upon miR-22 expression

in another breast tumour cell line, T47D (Fig.S2D). Additionally, miR-22 suppressed

luciferase activity specifically for wild-type 3’UTR of MTHFD2, while having no effect on a

reporter with a mutated miR-22 binding site, demonstrating direct targeting (Fig.3C).

MTHFD2 is crucial for the regeneration of mitochondrial tetrahydrofolate (THF) that is

needed for purine synthesis, the SAM cycle, and mitochondrial glycine synthesis (MGS) (40)

(Fig.3D). Hence miR-22 mediated suppression of MTHFD2 activity is likely to alter

7

mitochondrial one-carbon metabolism in cancer cells and to modulate sensitivity to anti-folate

chemotherapy.

Interactions of miR-22 with its metabolic targets affect clinical outcomes in breast

cancer.

To evaluate the relevance of our findings to human cancer, we examined gene expression

profiles across several publicly available datasets of breast tumours, the cancer with currently

the strongest evidence implicating miR-22 deregulation to the development of the disease (28,

30, 33, 41, 42). In two distinct breast cancer datasets miR-22 was shown to be significantly

negatively correlated to MYC (from The Cancer Genome Atlas: rho=-0.16, p=1x10-8; from

Farazi et al. (2011): rho=-0.25, p=0.001), similar to previous observations in the NCI-60 cell

line panel (Fig.1D). We next examined whether miR-22 was associated with expression

levels of the metabolic targets identified here (MTHFD2, ACLY, ELOVL6) in breast cancer

tissue. In two independent tumour series (43, 44), levels of the mature miR-22 as determined

by miRNAseq were significantly correlated to mRNA expression of MTHFD2 (Fig.4A)

consistent with regulation of this target gene by miR-22 in vivo. For the other two identified

targets significant correlation was observed in at least one of these datasets (Fig. 4B). Since

miR-22 is intragenic we also considered using the expression levels of the miR-22 host gene

(MIR22-HG) as a surrogate marker for the mature miRNA given the much wider availability

of transcriptomic data from breast cancer tissue containing probes for this transcript. MIR22-

HG and miR-22 were highly positively correlated in TGCA dataset (rho=0.51, p<1x10-16).

Consistent with the regulatory effect of miR-22 on ACLY, ELOVL6, and MTHFD2 we

observed negative correlations between in vivo expression of MIR22-HG and mRNA levels

of all three target genes in several independent datasets (Fig.4B). These observations support

our hypothesis of miR-22 suppressing these metabolic target genes in human tumours

(Fig.4C)

We next investigated the association between the expression of MIR22-HG, ACLY, ELOVL6

and MTHFD2 and relapse-free survival in breast cancer patients. We found that a combined

8

analysis of relapse-free survival data using a random effects model for these patient cohorts

indicated a significant inverse (protective) association between MIR22-HG and relapse rate

(Fig. 5A, upper vs. lower expression tertiles HR 0.68, p=0.001) as well as a positive

association between tumour expression of miR-22 target genes and patient outcomes (Fig.5B-

D, MTHFD2 HR 1.58, p=0.019; ACLY HR 1.34, p=0.06; ELOVL6 HR 1.54, p=0.004).

Kaplan-Meier analyses also indicated a sequential increase in progression rates across tertiles

of expression for MTHFD2 and MIR22-HG, while in the case of ELOVL6 the exposure-

response relationship observed was that the upper tertile was distinct from the middle and

lower tertile groups which were equivalent (Fig. S5). No clear trend across tertiles was

observed for ACLY. Finally, we examined whether miR-22 expression could interact with

target gene expression to modify patient outcomes. Our strategy was first to stratify patients

with respect to target gene expression and then to test for a significant trend in the effect of

miR-22 on relapse-free survival. This analysis demonstrated that the protective effect of

elevated MIR22-HG expression exhibited a systematic progression from undetectable in the

high (upper tertile) ELOVL6 expression group to being a major effect (HR 0.58, 95% CI

0.39-0.88) in the low (lowest tertile) group (Fig 6A, trend test p=0.0001) with the best

outcomes observed for the combination of the lowest tertile of ELOVL6 expression and high

(above median) MIR22-HG expression. This is consistent with a competitive model of

miRNA-mRNA interaction where the degree of functional suppression is dependent on the

relative concentrations of the mRNA and the miRNA. Conversely, we also observed that the

hazard ratio for patients with elevated levels of ELOVL6 was significantly inversely

dependent on MIR22-HG expression (Fig. 6B, trend test p<10-4) with rates of progression

only separable on the basis of MIR22-HG levels within the low (below median) ELOVL6

subset (Fig. 6B, blue curves) but identical for the high ELOVL6 subset (Fig. 6B, red curves) .

These observations suggest that miR-22/ELOVL6 interaction could be a paradigmatic

example of effect modification between cancer-associated miRNAs and their mRNA targets

on disease progression in vivo; it also indicates that a combined biomarker analysis

considering miRNA/mRNA interactions could give additional prognostic information by

9

reporting non-additive/synergistic effects. Similar modifying effects on outcomes could be

detected for miR-22/MTHFD2 (Fig. S6) but not miR-22/ACLY (Fig. S7) perhaps indicating

that the latter interaction, although detectable in vitro, was not relevant in vivo. Collectively

our in vitro and bioinformatic analyses provide compelling evidence that miR-22 and its

downstream metabolic network may influence malignant properties and patient outcomes in

human breast cancer.

DISCUSSION

Studies investigating the role of miR-22 in cancer development have reported contradictory

findings. Initial studies supported the hypothesis that miR-22 acts as a tumour-suppressor in

various tissues, including in breast (33, 41), liver (31), colon (45) and ovary (34). A tumour-

suppressive role for miR-22 is in agreement with its reported repression of cell proliferation

(25), promotion of cell senescence (30), and activation of apoptosis (27). In breast cancer

specifically miR-22 has been suggested to block cancer progression by repressing the

expression of ERα (33), CDK6 and SIRT1 (30), EV-1 (41), CD147 (42). In contrast, a recent

report links miR-22 upregulation with epithelial-mesenchymal transition and metastasis in

mouse models of breast cancer via suppression of TET family of methyl cytosine

dioxygenases, leading to hypermethylation of the miR-200 promoter (28). In a related study,

repression of TET was proposed to lead to increased hematopoietic stem cell self-renewal,

defective differentiation and ultimately myelodysplastic syndrome and leukaemia (29).

Additionally miR-22 has been reported to regulate the PTEN tumour-suppressor gene (46).

Consequently, it is possible that miR-22 can act as both an oncogene or as a tumour-

suppressor depending on the specific cellular context, including the metabolic

microenvironment.

Importantly, the data presented here suggest that another important effect of miR-22 is the

suppression of anabolic pathways that contribute to cancer development. MTHFD2 is one of

the most highly overexpressed metabolic genes across a wide range of tumour types (47), and

is a crucial component of mitochondrial one-carbon metabolism that sustains nucleotide

10

synthesis and high proliferation rates (48). It may also contribute to antioxidant defence

through NADPH generation (49, 50). Inhibition of ACLY leads to decreased proliferation in

cells reliant on de novo lipogenesis (35), and acetyl-coA generated by ACLY also supports

histone acetylation associated with regulating gene expression (51). Upregulation of ELOVL6

has been implicated in the development of breast (52) and liver cancer (53).

Beyond its role in cancer, miR-22 is a highly conserved gene throughout animal evolution

(54), and analysis of miR-22 expression across multiple tissues indicates that this gene is one

of the most abundantly expressed miRNAs in diverse adult tissues in vivo (Fig.S11). miR-22

has recently been characterized as a cardiac and skeletal muscle-enriched miRNA that is

upregulated during myocyte differentiation and cardiomyocyte hypertrophy (32) and its

deletion in vivo promotes stress-induced cardiac dilation and contractile dysfunction (55). In

this model, miR-22 may also target other metabolic genes including PGC-1α, PPARα, and

SIRT1 (56).

Our finding that a miRNA can act as an effect modifier of a prognostic gene that it targets is

potentially highly significant. It suggests a role for the rational investigation of

miRNA/mRNA interaction pairs as combination biomarkers for prognostication or

personalized medicine in oncology. Several lines of evidence point to MTHFD2 as a possible

target for the development of selective anti-cancer agents (47, 48), and MTHFD2 expression

might be a predictive marker for susceptibility to anti-folate chemotherapy (57, 58). Thus

miR-22 levels may provide an additional, mechanistically-justified means to refine MTHFD2-

based patient stratification and enrich further the subgroup likely to respond to specific

treatments, such as methotrexate. Conversely it may be possible to avoid unnecessary

morbidity in patients unlikely to benefit from a treatment regime associated a high degree of

toxicity.

In conclusion, using an integrative biology approach to the study of cancer cell molecular

profiles (‘systems oncology’) we have identified a novel mammalian pathway of metabolic

regulation involving a ubiquitously expressed miRNA that appears to be relevant to

progression in human breast cancer. Our correlative analyses also suggest that the pathway

11

may act downstream of key transcription factors such as MYC, however we did not test this

hypothesis explicitly and further work will be required to resolve the contrasting evidence in

the literature with regard to MYC-regulation of miR-22 (24, 33, 59). Since miRNAs appear to

‘fine tune’ activity of multiple targets rather than ‘switch off’ specific mRNAs, it is likely that

miR-22 will not uniquely control phenotypic response but rather act as part of a wider, as yet

uncharacterised RNA interaction network which integrates signals from several pathways.

Given the conservation and high abundance of miR-22, the interactions we have proposed

may have potential significance in normal physiology. While the metabolic pathway we have

reported may contribute to the proposed functions of miR-22 in cardiac protection,

embryogenesis, growth and tumourigenesis, further work will be required to elucidate the true

significance of miR-22 in human disease.

Materials &Methods

O2-PLS modeling

MiRNA normalized log2 expression across the NCI-60 cell panel was downloaded from the

NCBI GEO database (accession number GSE26375) (22). Metabolomic data was downloaded

from http://dtp.nci.nih.gov/mtargets/download.html on Mar 28, 2012. From the initial 1151

miRNA probes, 316 were selected corresponding to miRNAs targeting metabolic genes as

defined by Recon1 metabolic reconstruction (60). Gene-targeting miRNAs were predicted by

TargetScan (61). Levels of 154 identified metabolites were log-transformed and used in the

analysis. O2PLS (17, 62) is an extension of a supervised multivariate analysis method -

Partial Least Squares (PLS) – with an integrated Orthogonal Signal Correction filter (OSC)

that allows prediction in both directions between multivariate matrices X and Y. The miRNA

expression and metabolite data were centered and univariate-scaled. 7-fold cross-validation

was performed to calculate Q2 values. Sample permutation was performed to estimate the

significance of the obtained Q2 levels. Analysis was performed in the R statistical

environment using in-house developed scripts and open-source packages.

12

Cell culture

MCF-7 and T47D cells were obtained from the National Cancer Institute and maintained in

MEM supplemented with 10% FBS, 1x non-essential amino acids, and 1x

penicillin/streptomycin at 37oC and 5% CO2.

Quantitative RT-PCR (qRT-PCR)

Total RNA was extracted from cells using Trizol (Invitrogen). For mRNA and miRNA

analysis reverse transcription was performed using the High capacity cDNA reverse

transcription kit (Applied Biosystems) and the miRNA reverse transcription kit (Applied

Biosystems) respectively. The qRT-PCR was performed using the TaqManFast advanced

master mix and pre-designed TaqMan probes (Applied Biosystems) in the ABI 7500 Real-

Time PCR system. Relative miRNA levels were calculated using the ΔΔCT method with U6

RNA used for normalization.

Immunoblotting

Protein was isolated from cells using RIPA buffer (Sigma-Aldrich) and quantified using a

BCA assay (Pierce). Monoclonal anti-MTHFD2 (ab56772), anti-MAT2A (ab86424), anti-

MTHFR (ab113637), anti-ACLY (ab61762), and anti-ELOVL6 (ab69857) antibodies were

purchased from Abcam. Monoclonal anti-beta-Actin (AC-15) antibody was purchased from

Sigma-Aldrich. Protein lysates were resolved in 10% Acrylamide gels by SDS-PAGE

electrophoresis and the relative intensities were determined using the ImageJ software.

Transient transfections of miR-22

The miR-22 mirVana miRNA mimic and mirVana mimic negative control #1 (referred to in

figures as the scramble control) were obtained from Applied Biosystems. Transfections were

performed using SiPortNeoFx (Applied Biosystems) according to the manufacturer’s

instructions, with a final concentration of 50nM for the miRNA mimic/inhibitors and the

scramble control. Unless otherwise stated summary data shown are for n=3 independent

experiments.

13

RNA silencing of ACLY and ELOVL6

The Silencer Select siACLY, siELOVL6, and negative control #1 (referred to in figures as

control) were obtained from Applied Biosystems. Transfections were performed using

SiPortNeoFx (Applied Biosystems) according to the manufacturer’s instructions, with a final

concentration of 50nM for the miRNA mimic/inhibitors and the scramble control.

Luciferase reporter assay

The ACLY 3’-UTR reporter clone pMirTarget-ACLY (SC211546) and the MTHFD2 3’-UTR

reporter clone pMirTarget-MTHFD2 (SC212261) and corresponding reporter clones with

base substitutions (pMirTarget-ACLY-MUT and pMirTarget-MTHFD2-MUT) were obtained

from Origene Technologies. The ELOVL6 3’-UTR reporter clone GoClone-ELOVL6 and the

corresponding reporter clone with base substitutions (GoClone-ELOVL6-MUT) was obtained

from SwitchGear Genomics. Sequences for the wild-type and mutant seed regions in the 3’

UTR are as follows:

ACLY seed region

WT 5’TCCACAAAGATTCTGGGCAGCTGCCACCTCAGTCTC 3’ | || |

MUT 5’TCCACAAAGATTCTGAGAAGATGCCACCTCAGTCTC 3’

ELOVL6 seed region

WT 5’AAACACAAAACCCAAGGCAGCTTAGGGATAATTAGGT 3’ | || |

MUT 5’AAACACAAAACCCAATGTAGATTAGGGATAATTAGGT 3’

MTHFD2 seed region

WT 5’CTTGATAATCATTTGGGCAGCTTGGGTAAGTACGCAA 3’ | | |

MUT 5’CTTGATAATCATTTGGTCGGACTGGGTAAGTACGCAA 3’

pTK-CLuc (New England Biolabs) containing the Cypridina luciferase gene was

cotransfected for the normalization of transfection efficiencies. MCF7 cells were first co-

transfected with appropriate plasmids and microRNA mimics in 24-well plates, and then

harvested and lysed for luciferase assays 24h after transfection. The culture medium was

14

collected for Cypridina luciferase assays. Cells transfected with pMirTarget plasmids were

assayed for firefly luciferase using ONE-Glo Luciferase Assay Reagent (Promega), while

cells transfected with GoClone plasmids were assayed for Renilla luciferase using the

LightSwitch Assay Reagent (SwitchGear Genomics). Control Cypridina luciferase expression

from the pTK-CLuc plasmid was assayed using the BioLux Cypridina Luciferase Assay Kit

(New England Biolabs). Luminescence was quantified using a Plate Luminometer (Synergy

H1, Biotek), and the target reporter luciferase activity was normalized by luminescence from

Cypridina luciferase. Summary data shown are for n=3 independent experiments.

13C-glucose labeling experiments

MCF-7 cells that had been transiently transfected with miR-22 miRNA mimic or the negative

control mimic were cultured with U-13C6-glucose for 5 or 48 hours in 6 well plates (n=4).

MCF-7 cells transfected with siACLY, siELOVL6, or the negative control siRNA were also

cultured with U-13C6-glucose for 48 hours in 6 well plates (n=5). After trypsinisation, MCF-7

cells were resuspended in MEM (with 10% FBS, 5% non-essential amino acids and 5%

penicillin/streptomycin) and 200k cells were seeded per well and allowed to adhere overnight.

Cells were then incubated for 1 hr with glucose and glutamine-free DMEM with 5.6 mM 12C6-

glucose, 2 mM glutamine, 10% dialysed-FBS (BioSera) and 5% penicillin/streptomycin to

equilibrate intracellular metabolites with the media in preparation for addition of the 13C-

labeled carbon source. After 1 hr, the media was replaced with glucose- and glutamine-free

DMEM supplemented with 5.6 mM 13C6-glucose, 2 mM glutamine, 10% dialysed-FBS

(BioSera) and 5% penicillin/streptomycin and incubated for 5 or 48 hrs.

At 5 or 48 hours the media was aspirated and the cell monolayer washed with cold (4 C)

Ringer’s buffer, which was aspirated before addition of 750 L of cold (approx. -70C)

methanol. The methanol-quenched cells were then scraped from the well and the sample was

transferred to a clean tube. To increase metabolite recovery, each well was washed with a

further 750 L of cold methanol and pooled with the first sample. The methanol-quenched

samples were then dried in a rotary evaporator under reduced pressure. Metabolites were

15

extracted from the methanol-quenched cell sample using a dual phase aqueous

methanol/chloroform method. The upper aqueous fraction and lower organic fraction were

transferred to separate silanized GC-MS vials. Fatty acid esters present in the organic fraction

were transesterified with the strong base sodium methoxide, leaving the free fatty acids to be

silylated with MTBSTFA in a subsequent step. Myristic acid d27 (10 l, 1.5 mg/ml) was

added as an internal standard. The dried samples were reconstituted with a solution of

methanol/toluene (333 l, 1:1 v/v), treated with 0.5 M sodium methoxide (167 l) and

incubated at room temperature for 1 hour. The reaction was stopped by the addition of 1 M

NaCl (500 l), and conc. HCl (25 l). The fatty acids were extracted with two volumes of

hexane (500 l), and the combined organic layers were dried under N2. Samples were then

silylated by reconstituting with 40 l acetonitrile and treating with 40 l MTBSTFA (with 1%

TBDMCS) (Thermo), and incubating at 70C for 60 min. Following derivatization, 2-

fluorobiphenyl in anhydrous pyridine (10 l, 1 mM) was added to the samples as an injection

standard and the samples were transferred to deactivated glass vial inserts. The order of

sample preparation and analysis on GC-MS was randomized.

GC-MS Analysis

GC-MS analysis was performed on an Agilent 7890 GC equipped with a 30m DB-5MS

capillary column with a 10m Duraguard column connected to an Agilent 5975 MSD operating

under electron impact (EI) ionization (Agilent Technologies). Samples were injected with an

Agilent 7693 autosampler injector into deactivated splitless liners using helium as the carrier

gas. Metabolites were assigned using an in-house generated library with the deconvolution

program AMDIS (63). Individual isotopomer peaks were integrated using MatLab scripts

based on the program GAVIN (64), to obtain a mass isotopomer distribution vector (MID) for

each metabolite. MIDs were normalized such that the sum of the metabolite isotopomer

abundances was equal to one, and corrected for naturally occurring elemental isotopes based

on the method described by Millard et al. (65)

16

Isotopomer spectral analysis (ISA)

ISA was performed with MATLAB scripts developed in-house and provided estimates for

three parameters that describe the biosynthesis of fatty acids; the relative enrichment in the

lipogenic AcCoA pool from a given 13C labeled tracer (D), the fraction of fatty acids that are

newly synthesized (g(t)) (38) ,and the elongation of endogenous fatty acids (Elongation)

through the addition of a single AcCoA from the synthesis pool (39). These parameters were

estimated for replicate fatty acid MIDs, and parameter errors were verified via parameter

sensitivity analysis using a Monte Carlo approach (66).

Statistical analyses

All summary experimental data are presented as means with error bars indicating s.e.m. A

two-sided Student’s t-test was used for significance testing where appropriate unless

otherwise stated.

Survival analysis and in vivo correlations

Survival analyses on possible target genes of miR-22 were conducted on breast cancer

datasets housed in the Kaplan Meier Plotter database (67) and The Cancer Genome Atlas

(TCGA) (43). All gene expression data, apart from the TCGA dataset, was downloaded from

the Gene Expression Omnibus (GEO) database. All gene expression from GEOdb was pre-

processed by the GCRMA procedure, which takes into account gene specific binding before

quantile normalisation. TCGA gene expression data (tier 3 data – log2 of lowess normalized

data) was downloaded from the TCGA data portal. miRNA-seq data from TCGA and Faraziet

al. were normalised by Trimmed Mean of M-values (TMM) normalisation (68). Survival data

were extracted directly from GEOdb where available. Survival data of the datasets that were

not available on GEOdb were extracted from the Kaplan Meier Plotter database (67). The

datasets used are GSE1456, GSE16391, GSE2034, GSE2990, GSE3494, GSE6532, GSE7390

and GSE9195. All correlation analyses used Spearman rank correlation. Hazard ratios were

17

calculated using a Cox proportional hazard function dichotomized on upper and lower tertliles

of gene and microRNA expression. All data analysis was done in the R statistical

environment. Gene expression pre-processing of datasets from GEOdb was done using the

‘simpleaffy’ package (69). miRNA-seq data normalisation was conducted using the ‘edgeR’

package (70). Survival analysis was conducted using the ‘survival’ package (71).

MiRNA Target Prediction

MiRNA target prediction algorithms are characterized by high false-positive and false-

negative rates. By combining the results of multiple algorithms, it is possible to reduce the

inaccurate predictions of these algorithms. Here we used miRSystem (72), which combines

information from Diana-microT, miRanda, miRBridge, rna22, PITA, Pictar, and TargetScan

miRNA target prediction algorithms. We considered as putative targets of miR-22 to be those

predicted by three or more algorithms. The predicted targets for miR-22 can be seen in

Supplementary File 1. To identify putative miR-22 metabolic direct targets we then

determined the overlap between genes predicted to be targets of miR-22 by three or more

target prediction software and to be repressed by more than 1.5-fold in MCF-7 (33) and ES-2

(34) transfected with the miRNA.

Expression patterns of miR-22

Expression of miR-22 across various adult tissues was determined from independent

microarray (73) and next-generation sequencing (74) datasets. Expression data associated

with differentiation of cells in vitro also were obtained from published studies as indicated in

Supplementary Table 1.

Code availability

R scripts used for the analysis of tumour gene expression profiles are publicly available and

can be sourced from http://CRAN.R-project.org. The R script used for O2-PLS analysis is

available for non-commercial purposes only on request from e.nevedomskaya@nki.nl.

MATLAB code for ISA is available on request from the authors.

18

Acknowledgements

HK, CK, GT & JE acknowledge support by the European Community's Seventh Framework

Programme - Health (FP7/2007-2013) project DETECTIVE (grant agreement no.

266838).  HK & JE are also supported by Cancer Research UK programme grant A15115.

HK & GV are supported by the EC FP7/2007-2013 project Euro-MOTOR (grant agreement

no. 259867).  CHL is supported by a UK Biotechnology and Biological Sciences Research

Council (BBSRC) PhD studentship (grant no. BB/F529270/1 for the Institute of Chemical

Biology (Imperial College London) Doctoral Training Centre).  YP is supported by a Royal

Thai Government Scholarship. TY is supported by a UK MRC PhD studentship (Imperial

College London Faculty of Medicine Doctoral Training Award). AB is supported by the EC

FP7/2013-2018 project HeCaTos (grant agreement no. 602156). We also acknowledge

valuable discussions with Dr Charlotte Bevan, Prof Charles Coombes, Dr Jake Bundy, Prof

Nigel Gooderham and Dr Tim Ebbels.

Author Contributions

CK & HK conceived the project.  CK, GV & HK prepared and wrote the final manuscript and

figures with support from other authors.  CK, GV, and JE conducted cell experiments to

confirm miR-22 regulation of target genes.  GV established the luciferase reporter assays.

GT, JE, & GV conducted 13C labelling experiments.  GT established all GC-MS protocols and

conducted the modelling of isotopomer distributions.  CHL conducted supporting

metabolomic analysis.  EN conducted the PLS modelling, TY and AB carried out

confirmatory protein analyses. YP conducted the bioinformatic analysis of all patient datasets

and CK conducted all other bioinformatic analyses.  HK managed the project.  All authors

made a significant practical and intellectual contribution to the manuscript.

19

Conflict of Interest

The authors declare that they have no conflict of interest.

References

1. Schulze A, Harris AL. How cancer metabolism is tuned for proliferation and vulnerable to disruption. Nature. 2012;491(7424):364-73. Epub 2012/11/16.2. Vander Heiden MG, Cantley LC, Thompson CB. Understanding the Warburg effect: the metabolic requirements of cell proliferation. Science. 2009;324(5930):1029-33. Epub 2009/05/23.3. Anastasiou D, Yu Y, Israelsen WJ, Jiang JK, Boxer MB, Hong BS, et al. Pyruvate kinase M2 activators promote tetramer formation and suppress tumorigenesis. Nature chemical biology. 2012;8(10):839-47. Epub 2012/08/28.4. Rohle D, Popovici-Muller J, Palaskas N, Turcan S, Grommes C, Campos C, et al. An inhibitor of mutant IDH1 delays growth and promotes differentiation of glioma cells. Science. 2013;340(6132):626-30. Epub 2013/04/06.5. Frezza C, Zheng L, Folger O, Rajagopalan KN, MacKenzie ED, Jerby L, et al. Haem oxygenase is synthetically lethal with the tumour suppressor fumarate hydratase. Nature. 2011;477(7363):225-8. Epub 2011/08/19.6. Bartel DP. MicroRNAs: target recognition and regulatory functions. Cell. 2009;136(2):215-33. Epub 2009/01/27.7. He L, Thomson JM, Hemann MT, Hernando-Monge E, Mu D, Goodson S, et al. A microRNA polycistron as a potential human oncogene. Nature. 2005;435(7043):828-33. Epub 2005/06/10.8. Shimono Y, Zabala M, Cho RW, Lobo N, Dalerba P, Qian D, et al. Downregulation of miRNA-200c links breast cancer stem cells with normal stem cells. Cell. 2009;138(3):592-603. Epub 2009/08/12.9. Medina PP, Nolde M, Slack FJ. OncomiR addiction in an in vivo model of microRNA-21-induced pre-B-cell lymphoma. Nature. 2010;467(7311):86-90. Epub 2010/08/10.10. Volinia S, Croce CM. Prognostic microRNA/mRNA signature from the integrated analysis of patients with invasive breast cancer. Proceedings of the National Academy of Sciences of the United States of America. 2013;110(18):7413-7. Epub 2013/04/17.11. Vecchione A, Belletti B, Lovat F, Volinia S, Chiappetta G, Giglio S, et al. A microRNA signature defines chemoresistance in ovarian cancer through modulation of angiogenesis. Proceedings of the National Academy of Sciences of the United States of America. 2013;110(24):9845-50. Epub 2013/05/24.12. Eichner LJ, Perry MC, Dufour CR, Bertos N, Park M, St-Pierre J, et al. miR-378( *) mediates metabolic shift in breast cancer cells via the PGC-1beta/ERRgamma transcriptional pathway. Cell metabolism. 2010;12(4):352-61. Epub 2010/10/05.13. Zhu H, Shyh-Chang N, Segre AV, Shinoda G, Shah SP, Einhorn WS, et al. The Lin28/let-7 axis regulates glucose metabolism. Cell. 2011;147(1):81-94. Epub 2011/10/04.14. Jiang S, Zhang LF, Zhang HW, Hu S, Lu MH, Liang S, et al. A novel miR-155/miR-143 cascade controls glycolysis by regulating hexokinase 2 in breast cancer cells. The EMBO journal. 2012;31(8):1985-98. Epub 2012/02/23.

20

15. Gao P, Tchernyshyov I, Chang TC, Lee YS, Kita K, Ochi T, et al. c-Myc suppression of miR-23a/b enhances mitochondrial glutaminase expression and glutamine metabolism. Nature. 2009;458(7239):762-5. Epub 2009/02/17.16. Liu W, Le A, Hancock C, Lane AN, Dang CV, Fan TW, et al. Reprogramming of proline and glutamine metabolism contributes to the proliferative and metabolic responses regulated by oncogenic transcription factor c-MYC. Proceedings of the National Academy of Sciences of the United States of America. 2012;109(23):8983-8. Epub 2012/05/23.17. Trygg J, Wold S. O2-PLS, a two-block (X–Y) latent variable regression (LVR) method with an integral OSC filter. J Chemometr. 2003;17(1):53-64.18. Bylesjo M, Eriksson D, Kusano M, Moritz T, Trygg J. Data integration in plant biology: the O2PLS method for combined modeling of transcript and metabolite data. The Plant journal : for cell and molecular biology. 2007;52(6):1181-91. Epub 2007/10/13.19. Rantalainen M, Cloarec O, Beckonert O, Wilson ID, Jackson D, Tonge R, et al. Statistically integrated metabonomic-proteomic studies on a human prostate cancer xenograft model in mice. Journal of proteome research. 2006;5(10):2642-55. Epub 2006/10/07.20. Kirwan GM, Johansson E, Kleemann R, Verheij ER, Wheelock AM, Goto S, et al. Building multivariate systems biology models. Analytical chemistry. 2012;84(16):7064-71. Epub 2012/08/04.21. National Cancer Institute. Molecular Target Data - NCI/NIH Developmental Therapeutics Program Data. 2013; Available from: https://wiki.nci.nih.gov/display/NCIDTPdata/Molecular+Target+Data.22. Sokilde R, Kaczkowski B, Podolska A, Cirera S, Gorodkin J, Moller S, et al. Global microRNA analysis of the NCI-60 cancer cell panel. Molecular cancer therapeutics. 2011;10(3):375-84. Epub 2011/01/22.23. O'Donnell KA, Wentzel EA, Zeller KI, Dang CV, Mendell JT. c-Myc-regulated microRNAs modulate E2F1 expression. Nature. 2005;435(7043):839-43. Epub 2005/06/10.24. Chang TC, Yu D, Lee YS, Wentzel EA, Arking DE, West KM, et al. Widespread microRNA repression by Myc contributes to tumorigenesis. Nature genetics. 2008;40(1):43-50. Epub 2007/12/11.25. Marzi MJ, Puggioni EM, Dall'Olio V, Bucci G, Bernard L, Bianchi F, et al. Differentiation-associated microRNAs antagonize the Rb-E2F pathway to restrict proliferation. The Journal of cell biology. 2012;199(1):77-95. Epub 2012/10/03.26. Pandey DP, Picard D. miR-22 inhibits estrogen signaling by directly targeting the estrogen receptor alpha mRNA. Molecular and cellular biology. 2009;29(13):3783-90. Epub 2009/05/06.27. Tsuchiya N, Izumiya M, Ogata-Kawata H, Okamoto K, Fujiwara Y, Nakai M, et al. Tumor suppressor miR-22 determines p53-dependent cellular fate through post-transcriptional regulation of p21. Cancer research. 2011;71(13):4628-39. Epub 2011/05/14.28. Song SJ, Poliseno L, Song MS, Ala U, Webster K, Ng C, et al. MicroRNA-antagonism regulates breast cancer stemness and metastasis via TET-family-dependent chromatin remodeling. Cell. 2013;154(2):311-24. Epub 2013/07/09.29. Song SJ, Ito K, Ala U, Kats L, Webster K, Sun SM, et al. The oncogenic microRNA miR-22 targets the TET2 tumor suppressor to promote hematopoietic stem cell self-renewal and transformation. Cell stem cell. 2013;13(1):87-101. Epub 2013/07/06.

21

30. Xu D, Takeshita F, Hino Y, Fukunaga S, Kudo Y, Tamaki A, et al. miR-22 represses cancer progression by inducing cellular senescence. The Journal of cell biology. 2011;193(2):409-24. Epub 2011/04/20.31. Zhang J, Yang Y, Yang T, Liu Y, Li A, Fu S, et al. microRNA-22, downregulated in hepatocellular carcinoma and correlated with prognosis, suppresses cell proliferation and tumourigenicity. British journal of cancer. 2010;103(8):1215-20. Epub 2010/09/16.32. Huang ZP, Chen J, Seok HY, Zhang Z, Kataoka M, Hu X, et al. MicroRNA-22 regulates cardiac hypertrophy and remodeling in response to stress. Circulation research. 2013;112(9):1234-43. Epub 2013/03/26.33. Xiong J, Du Q, Liang Z. Tumor-suppressive microRNA-22 inhibits the transcription of E-box-containing c-Myc target genes by silencing c-Myc binding protein. Oncogene. 2010;29(35):4980-8. Epub 2010/06/22.34. Nagaraja AK, Creighton CJ, Yu Z, Zhu H, Gunaratne PH, Reid JG, et al. A link between mir-100 and FRAP1/mTOR in clear cell ovarian cancer. Mol Endocrinol. 2010;24(2):447-63. Epub 2010/01/19.35. Zaidi N, Swinnen JV, Smans K. ATP-citrate lyase: a key player in cancer metabolism. Cancer research. 2012;72(15):3709-14. Epub 2012/07/13.36. Migita T, Okabe S, Ikeda K, Igarashi S, Sugawara S, Tomida A, et al. Inhibition of ATP citrate lyase induces triglyceride accumulation with altered fatty acid composition in cancer cells. International journal of cancer Journal international du cancer. 2013.37. Guillou H, Zadravec D, Martin PG, Jacobsson A. The key roles of elongases and desaturases in mammalian fatty acid metabolism: Insights from transgenic mice. Progress in lipid research. 2010;49(2):186-99. Epub 2009/12/19.38. Kelleher JK, Masterson TM. Model equations for condensation biosynthesis using stable isotopes and radioisotopes. The American journal of physiology. 1992;262(1 Pt 1):E118-25. Epub 1992/01/01.39. Lligona-Trulla L, Arduini A, Aldaghlas TA, Calvani M, Kelleher JK. Acetyl-L-carnitine flux to lipids in cells estimated using isotopomer spectral analysis. Journal of lipid research. 1997;38(7):1454-62. Epub 1997/07/01.40. Pike ST, Rajendra R, Artzt K, Appling DR. Mitochondrial C1-tetrahydrofolate synthase (MTHFD1L) supports the flow of mitochondrial one-carbon units into the methyl cycle in embryos. The Journal of biological chemistry. 2010;285(7):4612-20. Epub 2009/12/02.41. Patel JB, Appaiah HN, Burnett RM, Bhat-Nakshatri P, Wang G, Mehta R, et al. Control of EVI-1 oncogene expression in metastatic breast cancer cells through microRNA miR-22. Oncogene. 2011;30(11):1290-301. Epub 2010/11/09.42. Kong LM, Liao CG, Zhang Y, Xu J, Li Y, Huang W, et al. A Regulatory Loop Involving miR-22, Sp1, and c-Myc Modulates CD147 Expression in Breast Cancer Invasion and Metastasis. Cancer research. 2014;74(14):3764-78. Epub 2014/06/08.43. The Cancer Genome Atlas Network. Comprehensive molecular portraits of human breast tumours. Nature. 2012;490(7418):61-70.44. Farazi TA, Horlings HM, Ten Hoeve JJ, Mihailovic A, Halfwerk H, Morozov P, et al. MicroRNA sequence and expression analysis in breast tumors by deep sequencing. Cancer research. 2011;71(13):4443-53. Epub 2011/05/19.45. Alvarez-Diaz S, Valle N, Ferrer-Mayorga G, Lombardia L, Herrera M, Dominguez O, et al. MicroRNA-22 is induced by vitamin D and contributes to its antiproliferative, antimigratory and gene regulatory effects in colon cancer cells. Human molecular genetics. 2012;21(10):2157-65. Epub 2012/02/14.

22

46. Xu XD, Song XW, Li Q, Wang GK, Jing Q, Qin YW. Attenuation of microRNA-22 derepressed PTEN to effectively protect rat cardiomyocytes from hypertrophy. Journal of cellular physiology. 2012;227(4):1391-8. Epub 2011/05/28.47. Nilsson R, Jain M, Madhusudhan N, Sheppard NG, Strittmatter L, Kampf C, et al. Metabolic enzyme expression highlights a key role for MTHFD2 and the mitochondrial folate pathway in cancer. Nature communications. 2014;5:3128. Epub 2014/01/24.48. Jain M, Nilsson R, Sharma S, Madhusudhan N, Kitami T, Souza AL, et al. Metabolite profiling identifies a key role for glycine in rapid cancer cell proliferation. Science. 2012;336(6084):1040-4. Epub 2012/05/26.49. Fan J, Ye J, Kamphorst JJ, Shlomi T, Thompson CB, Rabinowitz JD. Quantitative flux analysis reveals folate-dependent NADPH production. Nature. 2014;510(7504):298-302. Epub 2014/05/09.50. Shin M, Bryant JD, Momb J, Appling DR. Mitochondrial MTHFD2L Is a Dual Redox Cofactor-specific Methylenetetrahydrofolate Dehydrogenase/Methenyltetrahydrofolate Cyclohydrolase Expressed in Both Adult and Embryonic Tissues. The Journal of biological chemistry. 2014;289(22):15507-17. Epub 2014/04/16.51. Wellen KE, Hatzivassiliou G, Sachdeva UM, Bui TV, Cross JR, Thompson CB. ATP-citrate lyase links cellular metabolism to histone acetylation. Science. 2009;324(5930):1076-80. Epub 2009/05/23.52. Doria ML, Ribeiro AS, Wang J, Cotrim CZ, Domingues P, Williams C, et al. Fatty acid and phospholipid biosynthetic pathways are regulated throughout mammary epithelial cell differentiation and correlate to breast cancer survival. FASEB journal : official publication of the Federation of American Societies for Experimental Biology. 2014. Epub 2014/06/28.53. Muir K, Hazim A, He Y, Peyressatre M, Kim DY, Song X, et al. Proteomic and lipidomic signatures of lipid metabolism in NASH-associated hepatocellular carcinoma. Cancer research. 2013;73(15):4722-31. Epub 2013/06/12.54. Christodoulou F, Raible F, Tomer R, Simakov O, Trachana K, Klaus S, et al. Ancient animal microRNAs and the evolution of tissue identity. Nature. 2010;463(7284):1084-8. Epub 2010/02/02.55. Gurha P, Abreu-Goodger C, Wang T, Ramirez MO, Drumond AL, van Dongen S, et al. Targeted deletion of microRNA-22 promotes stress-induced cardiac dilation and contractile dysfunction. Circulation. 2012;125(22):2751-61. Epub 2012/05/10.56. Gurha P, Wang T, Larimore AH, Sassi Y, Abreu-Goodger C, Ramirez MO, et al. microRNA-22 promotes heart failure through coordinate suppression of PPAR/ERR-nuclear hormone receptor transcription. PloS one. 2013;8(9):e75882. Epub 2013/10/03.57. Tedeschi PM, Markert EK, Gounder M, Lin H, Dvorzhinski D, Dolfi SC, et al. Contribution of serine, folate and glycine metabolism to the ATP, NADPH and purine requirements of cancer cells. Cell death & disease. 2013;4:e877. Epub 2013/10/26.58. Lehtinen L, Ketola K, Makela R, Mpindi JP, Viitala M, Kallioniemi O, et al. High-throughput RNAi screening for novel modulators of vimentin expression identifies MTHFD2 as a regulator of breast cancer cell migration and invasion. Oncotarget. 2013;4(1):48-63. Epub 2013/01/09.59. Polioudakis D, Bhinge AA, Killion PJ, Lee BK, Abell NS, Iyer VR. A Myc-microRNA network promotes exit from quiescence by suppressing the interferon

23

response and cell-cycle arrest genes. Nucleic acids research. 2013;41(4):2239-54. Epub 2013/01/11.60. Duarte NC, Becker SA, Jamshidi N, Thiele I, Mo ML, Vo TD, et al. Global reconstruction of the human metabolic network based on genomic and bibliomic data. Proceedings of the National Academy of Sciences of the United States of America. 2007;104(6):1777-82. Epub 2007/02/03.61. Friedman RC, Farh KK, Burge CB, Bartel DP. Most mammalian mRNAs are conserved targets of microRNAs. Genome research. 2009;19(1):92-105. Epub 2008/10/29.62. Trygg J. O2-PLS for qualitative and quantitative analysis in multivariate calibration. J Chemometr. 2002;16(6):283-93.63. Stein SE. An integrated method for spectrum extraction and compound identification from gas chromatography/mass spectrometry data. Journal of the American Society for Mass Spectrometry. 1999;10(8):770-81.64. Behrends V, Tredwell GD, Bundy JG. A software complement to AMDIS for processing GC-MS metabolomic data. Analytical biochemistry. 2011;415(2):206-8. Epub 2011/05/18.65. Millard P, Letisse F, Sokol S, Portais JC. IsoCor: correcting MS data in isotope labeling experiments. Bioinformatics. 2012;28(9):1294-6. Epub 2012/03/16.66. Helton JC, Johnson JD, Sallaberry CJ, Storlie CB. Survey of sampling-based methods for uncertainty and sensitivity analysis. Reliability Engineering & System Safety. 2006;91(10–11):1175-209.67. Gyorffy B, Lanczky A, Eklund AC, Denkert C, Budczies J, Li Q, et al. An online survival analysis tool to rapidly assess the effect of 22,277 genes on breast cancer prognosis using microarray data of 1,809 patients. Breast cancer research and treatment. 2010;123(3):725-31. Epub 2009/12/19.68. Robinson MD, Oshlack A. A scaling normalization method for differential expression analysis of RNA-seq data. Genome biology. 2010;11(3):R25. Epub 2010/03/04.69. Wilson CL, Miller CJ. Simpleaffy: a BioConductor package for Affymetrix Quality Control and data analysis. Bioinformatics. 2005;21(18):3683-5. Epub 2005/08/04.70. Robinson MD, McCarthy DJ, Smyth GK. edgeR: a Bioconductor package for differential expression analysis of digital gene expression data. Bioinformatics. 2010;26(1):139-40. Epub 2009/11/17.71. Therneau TM. A Package for Survival Analysis in S. 2013; Available from: http://CRAN.R-project.org/package=survival.72. Lu TP, Lee CY, Tsai MH, Chiu YC, Hsiao CK, Lai LC, et al. miRSystem: an integrated system for characterizing enriched functions and pathways of microRNA targets. PloS one. 2012;7(8):e42390. Epub 2012/08/08.73. Meiri E, Levy A, Benjamin H, Ben-David M, Cohen L, Dov A, et al. Discovery of microRNAs and other small RNAs in solid tumors. Nucleic acids research. 2010;38(18):6234-46. Epub 2010/05/21.74. Kuchen S, Resch W, Yamane A, Kuo N, Li Z, Chakraborty T, et al. Regulation of microRNA expression and abundance during lymphopoiesis. Immunity. 2010;32(6):828-39. Epub 2010/07/08.

24

FIGURE LEGENDS

Fig.1. O2-PLS modeling of metabolite-miRNA associations in the NCI-60 panel reveals a

MYC-associated miRNA network influencing the tumour cell metabolome.

Modelcontains2 predictive, 4 X-orthogonal and 2 Y-orthogonal components; R2X = 0.09, R2Y

= 0.14, Q2X(cum) = 0.06, Q2Y(cum) = 0.12) (a) O2-PLS model correlated component

‘loadings’ and individual variable cross validated predicted explained variance (Q2) for

miRNAs; (b) O2-PLS model correlated component ‘loadings’ and Q2 for metabolites; (c)

Heatmap showing univariate Pearson correlation of selected miRNAs with metabolites in the

NCI-60 panel (based on permutation test, p<4e-05 for individual Q2 values); (d)

Pearsoncorrelations of miRNAs identified from multivariate model with MYC mRNA

expression across the NCI-60 panel.

Fig. 2. miR-22 represses fatty acid synthesis and elongation by direct targeting of ACLY

and ELOVL6 (a) Identification of putative miR-22 direct targets, identifying common

targets among genes with > 1.5-fold downregulation after miR-22 transfection of ES-2 cells

(34), genes with > 1.5-fold downregulation after miR-22 transfection of MCF-7 cells (33),

and genes predicted to be miR-22 targets by three or more target prediction algorithms

(Supplementary Data file 1); (b) Scheme of de novo synthesis and elongation of fatty acids

and the involvement of ACLY and ELOVL6 in these pathways; (c) qRT-PCR detection of

25

ACLY and ELOVL6 mRNA in MCF-7 cells transfected with either miR-22 mimic, or a

negative mimic control (scramble) after 48 hours. Data are normalized to β-actin and shown

as mean ± s.e.m., n=3. *p<0.05; (d) Western blot for ACLY and ELOVL6 in MCF-7

transfected with miR-22 mimic or scramble control. A representative immunoblot and a bar

chart of relative protein levels are shown (mean ± s.e.m., n=3) ; (e) Validation of direct

targeting of ACLY and ELOVL6 by miR-22 using a luciferase reporter assay. Vector

containing the luciferase reporter harboring the wild-type 3’UTR of genes (WT) or the 3’UTR

with specific mutation of miR-22 binding sites (MUT; see methods) were co-transfected with

miR-22 or negative control oligonucleotide into MCF-7 cells. Data are shown as mean ±

s.e.m., n=3; (f) Isotopomer spectral analysis (ISA) model of fatty acid biosynthesis; (g)

Estimated ISA parameters for stearate transesterified from lipids in MCF-7 cells transfected

with either miR-22 mimic, or a negative control, and incubated with U- 13C6-glucose for 48

hours. Data are shown as mean ± s.e.m., n=4; Asterisks note p-values after a Student’s t-test.

** p<0.01, *** p<0.001 (h) Mass isotopomer distribution (MID) of measured stearate

transesterified from lipids in MCF-7 cells transfected with either miR-22 miRNA mimic, or a

negative mimic control, and incubated with U-13C6-glucose for 48 hours. Data are shown as

mean ± s.e.m., n=3.

Fig. 3.The mitochondrial enzyme MTHFD2 is repressed by miR-22 and affects one-

carbon metabolism in cancer cells (a) qRT-PCR for MTHFD2. Data were normalized to β-

actin; (b) Western blot for MTHFD2 in MCF-7 transfected with miR-22 mimic or scramble

control. A representative immunoblot and a bar chart of relative protein levels are shown

(mean ± s.e.m., n=3) ; (c) Validation of direct targeting of MTHFD2 by miR-22 using a

luciferase reporter assay. Vector containing the luciferase reporter harboring the wildtype

3’UTR of gene (WT) or the 3’UTR with specific mutation of miR-22 binding sites (MUT)

were co-transfected with miR-22 or negative mimic control oligonucleotide into MCF-7 cells;

(d) Scheme of mammalian mitochondrial one-carbon metabolism and the metabolic role of

MTHFD2. All data are shown as mean ± s.e.m., n=3. **p<0.01; ***p<0.001.

26

Fig.4. The association of target gene expression with outcomes in cancer and

expression of miR-22 (a) Distributions of log2 normalised MTHFD2 mRNA

expression for each sample quartile of mature miR-22 expression in human tumour

samples. n= 1126 (43) and 161 (44); (b): Correlations between target gene expression

and miR-22 levels in multiple datasets. (c) Schematic of probable relationship of

metabolic gene regulation by miRNA.

Fig.5 Associations between miR-22, target gene expression and patient outcomes.

Meta-analysis of association between relapse-free survival and gene expression for (a)

miR-22 host gene (MIR22-HG), (b) MTHFD2, (c) ACLY, and (d) ELOVL6. Patients

were stratified by lower and upper tertiles of mRNA expression. Cox proportional

hazard regression was used to obtain the hazard ratio for having high expression from

each dataset. Cox regression coefficients were then combined using a random effect

model, resulting in a combined hazard ratio. + denotes datasets with ER-positive

samples, and ++ denotes datasets that all samples are ER-positive. Where multiple

probesets were available for a given gene, analyses resulting in the highest overall HR

are shown.

Figure 6. Effect modification on patient outcomes between miR-22 and

ELOVL6. (a) Kaplan-Meier analysis of relapse-free survival stratified by expression

tertiles of ELOVL6 as the effect modifier, (U, M and L are upper, middle and lower

tertiles, respectively), and upper and lower expression of MIR22-HG as the effect

driver (H – above median expression L – below median). (b) Kaplan-Meier analysis

of relapse-free survival stratified by expression tertiles of MIR22-HG as the effect

27

modifier (U, M and L are upper, middle and lower tertiles, respectively), and upper

and lower expression of ELOVL6 as the effect driver (H – above median expression L

– below median).. The extent of effect modification was determined by the Cochran-

Armitage trend test.

28