PHD3 Loss in Cancer Enables Metabolic Reliance on Fatty Acid Oxidation via Deactivation of

ACC2

In Brief Fatty acid oxidation (FAO) fuels the growth of many cancers. German et al. show that
prolyl hydroxylase 3 (PHD3) inhibits nutrient-dependent FAO by activating ACC2 and blocking
mitochondrial fatty acid import. PHD3 levels are decreased in acute myeloid leukemia,
promoting FAO and a metabolic dependency, which can be targeted pharmacologically.

Highlights

Prolyl hydroxylase 3 (PHD3) hydroxylates and activates the metabolic enzyme ACC2

During nutrient abundance, the PHD3/ACC2 axis represses fatty acid oxidation (FAO)

PHD3 is low in AML, fueling a reliance on fats that can be targeted with FAO inhibitors

Re-expressing PHD3 limits FAO via ACC2 and suppresses AML in culture and in vivo

While much research has examined the use of glucose and glutamine by tumor cells, many
cancers instead prefer to metabolize fats. Despite the pervasiveness of this phenotype,
knowledge of pathways that drive fatty acid oxidation (FAO) in cancer is limited. Prolyl
hydroxylase domain proteins hydroxylate substrate proline residues and have been linked to
fuel switching. Here, we reveal that PHD3 rapidly triggers repression of FAO in response to
nutrient abundance via hydroxylation of acetyl-coA carboxylase 2 (ACC2). We find that PHD3
expression is strongly decreased in subsets of cancer including acute myeloid leukemia (AML)
and is linked to a reliance on fat catabolism regardless of external nutrient cues. overexpressing
PHD3 limits FAO via regulation of ACC2 and consequently impedes leukemia cell proliferation.
Thus, loss of PHD3 enables greater utilization of fatty acids but may also serve as a metabolic
and therapeutic liability by indicating cancer cell susceptibility to FAO inhibition.

In the past decade, a resurgence of studies has provided mechanistic insight into why tumors
upregulate glucose uptake and metabolism(Lunt and Vander Heiden, 2011).However, our
understanding of tumor metabolism is incomplete because numerous tumors are FDG-PET
negative (Long and Smith, 2011; Ono et al., 2007), suggesting many cancers utilize alternate
carbn sources. Multiple cancer types have been suggested to rely on fatty acid oxidation (FAO)
for survival (Carracedo et al., 2013), highlighting a need to identify specific lipid metabolic
programs that may go awry in cancer.

Post-translational modifying enzymes are key components of metabolic reprogramming

(German and Haigis, 2015; Hitosugi and Chen, 2013). PHDs (also called EGLN1-3) are one class
of enzymes poised to coordinate metabolism in response to changing cellular conditions. PHDs
are a conserved family of oxygen- and a-ketoglutarate-dependent enzymes that are well known
to regulate glycolytic metabolism through hydroxylation of hypoxia inducible factor (HIF)
(Gorres and Raines, 2010). Hypoxia and a number of mutations in cancer repress activity of
some PHDs, stabilizing HIFa and triggering a transcriptional program to increase glycolysis and
anabolism while limiting mitochondrial bioenergetics (Masson and Ratcliffe, 2014).
Recent reports suggest that PHDs are also responsive to cellular nutrient status (Kaelin and
Ratcliffe, 2008). This may be linked to the use of a-ketoglutarate during prolyl hydroxylation
(Dura n et al., 2013). PHD3 is notable for its particular sensitivity to a-ketoglutarate, or perhaps
more generally to the high nutrient state that may be achieved by addition of a-ketoglutarate.
Along these lines, treating mouse xenografts with cell-permeable a-ketoglutarate inhibited
growth by a PHD3-dependent mechanism (Tennant and Gottlieb, 2010). This raises the
question of whether PHD3 is responsive to fluctuations in the nutrient state. We hypothesized
that PHD3 might link nutrient status with implementation of metabolic adaptations. Therefore,
we aimed to identify metabolic pathways regulated by PHD3.

In this study, we identify acetyl-CoA carboxylase 2 (ACC2), the gatekeeper of FAO, as a PHD3
substrate. By activating ACC2, PHD3 represses oxidation of long-chain fatty acids. Fatty acid
catabolism is a dynamic cellular process that responds to metabolic imbalances and restores
homeostasis (Gerhart-Hines et al., 2007). We show that PHD3 represses FAO during nutrient
abundance, and that cells with low PHD3 have persistent FAO regardless of external nutrient
cues. In AML, PHD3 expression is dramatically decreased, contributing to a boost in fatty acid
consumption that drives AML cell proliferation and disease severity.

RESULTS

PHD3 Binds and Modifies ACC by Prolyl Hydroxylation

To probe for PHD3 substrates, we performed immunoprecipitation of PHD3 followed by liquid

chromatography tndem mass spectrometry (LC-MS2) and detected an interaction with acetyl-
CoA carboxylase (ACC). 21 ACC peptides were identified in the PHD3 immunoprecipitation,
while no ACC peptides were identified in PHD2 or negative control samples (Table S1).

IP-western blots confirmed that ACC interacted with PHD3 but not PHD1, PHD2, or anti-HA
affinity resin alone (Figure 1A).

ACC converts acetyl-CoA to malonyl-CoA, which serves as a precursor for fat synthesis and an
inhibitor of FAO (Abu-Elheiga et al., 2003). Hence, ACC is a key regulator of fatty acid
homeostasis that determines whether cells catabolize or synthesize fatty acids (Brownsey et al.,
2006).

As PHD3 belongs to a family of prolyl hydroxylases, we first examined ACC hydroxylation by

PHD3. Using immunoprecipitation of ACC followed by western blot with a pan-hydroxyproline
antibody, we observed ACC hydroxylation. PHD3 overexpression boosted ACC hydroxylation
(Figure 1B), while two catalytically inactive PHD3 mutants, H196A and R206K (Bruick and
McKnight, 2001; Rantanen et al., 2008), did not increase ACC hydroxylation (Figure 1C). ACC is
present in two spatially and functionally distinct isoforms. Cytosolic ACC1 provides malonyl-
CoA for fatty acid synthesis, while ACC2 localized to the outer mitocondrial membrane
generates malonyl-CoA to inhibit the fatty acid transport protein CPT1, suppressing
mitochondrial FAO (Brownsey et al., 2006). Several PHD3-interacting peptides found by mass
spectrometry are shared between ACC1 and ACC2 (Table S1); we therefore interrogated
whether PHD3 hydroxylates CC1 or ACC2. Immunoprecipitation of endogenous ACC1 or ACC2
by isoform-specific antibodies demonstrated hydroxylation was specific to ACC2 and amplified
by PHD3 (Figure 1D), indicating that PHD3 modulates ACC2 hydroxylation level.

We next mapped ACC2 sites of hydroxyproline, as detected by a +15.9949 molecular weight

shift (Figures 1E and 1F). Three candidate hydroxylated prolines had more than five redundant
peptides per site: prolines 343, 450, and 2131 (Figure 1G). These sites are located in the biotin
carboxylase, ATP-grasp, and carboxyltransferase domains, respectively (Figure 1H). To validate
hydroxylation of these residues, we generated proline to alanine point mutants at each site and
found that P450A mutagenesis most dramatically decreased hydroxylation (Figure 1I). Using a
reconstituted in vitro hydroxylation assay, recombinant PHD3 hydroxylated a synthetic ACC2
peptide containing P450, but not a peptide containing P2131 or a control ACC2 proline-
containing peptide (P966) (Figure 1J). These data identify P450 as a major hydroxylation site
and suggest that modification of this residue may coordinate ACC2 function.

Hydroxylation at Residue P450 Promotes ACC2 Activity and ATP Binding

At 16 Da, hydroxylation is among the smallest posttranslational modifications. Nevertheless,

the electronegativity it imparts alters proline residue pucker, favors trans-configuration of the
peptide bond, and can induce structural changes significant enough to alter protein-protein
interactions or activity (Gorres nd Raines, 2010; Loenarz and Schofield, 2011). Thus, we
investigated whether site-specific hydroxylation might regulate ACC2 function. P450 is
conserved from yeast to humans (Figure 2A) in the ATP-grasp domain, a 196-amino-acid region
within the biotin carboxylase domain that includes nucleotide-binding amino acids at residues
458463 (Abu-Elheiga et al., 1997). We mapped site P450 in the human ACC2 biotin
carboxylase domain crystal structure (PDB: 3JRW) (Cho et al., 2010) superposed with the E. coli
ATP-bound ACC biotin carboxylase domain (PDB: 1DV2) (Thoden et al., 2000). Modeling
revealed that P450 is in close proximity to the catalytic site ATP (Figure 2B).

P450 caps the adenine ring of ATP, while the phosphate groups of ATP abut the previously
described nucleotide-binding site within ACC2.

The proximity of P450 to ATP led us to hypothesize that hydroxylation modulates the ability of
ACC2 to bind ATP. We purified ATP-binding proteins from dialyzed cell lysates using ATP affinity
chromatography and determined levels of ACC bound.

ACC2 lacking the major hydroxylation site upon P450 mutation to alanine or glycine showed
decreased ATP binding versus wild-type (Figures 2C and S1A), highlighting the importance of
this residue for ATP binding. Strikingly, knockdown of PHD3 also diminished ATP binding of
ACC2 (Figure 2D), demonstrating that hydroxylation is critical for optimal ATP binding. Thus,
PHD3 may activate ACC2 by enabling greater affinity for the co-substrate ATP.

To evaluate this hypothesis, we performed in vitro ACC activity assays from 293T cells, based on
production of [14C]malonyl- CoA from [14C]bicarbonate and acetyl-CoA. ACC2 overexpression
resulted in higher activity, as did addition of the known allosteric modulator citrate (Ruderman
and Prentki, 2004), validating the specificity of this assay (Figure 2E). The P450A mutation
strongly decreased ACC activity (Figure 2E). Of note, PHD3 overexpression increased wild-type
ACC2 activity (Figure 2F) but had no effect on P450A ACC2. Conversely, PHD3 knockdown
decreased ACC2 activity (Figure 2G). These findings support the model that PHD3 boosts ACC2
activity via hydroxylation of P450.

PHD3 Represses FAO

ACC2 inhibits FAO by limiting long-chain fatty acid import into the mitochondrial matrix
(Ruderman and Prentki, 2004). Because ACC2 is activated by PHD3, we hypothesized that PHD3
represses FAO. PHD3 knockdown enhanced palmitate oxidation in 293T and HepG2 cells
(Figures 3A and S2AS2D). PHD1 and PHD2 expression were not consistently altered by PHD3
knockdown, indicating that the effect was not due to other PHDs (Figure S2B). To assess the
role of ACC2 prolyl hydroxylation on FAO, we measured palmitate oxidation in cells
overexpressing P450A ACC2. While overexpression of wildtype CC2 decreased FAO, ACC2
lacking the proline hydroxylation site did not repress FAO (Figure 3B). In contrast, P343A and
P2131A variants behaved similarly to wild-type ACC2 Figure S2E).

ACC2 knockdown increased FAO in 293T cells by 2.5-fold, while the effect of PHD3 knockdown
was of similar magnitude (1.5- to 2-fold), demonstrating that PHD3 is a major contributor to
FAO regulation (Figure S2F compared to Figures 3A and 3C; extent of ACC2 knockdown shown
in Figure S2G). Immunoblots with control ACC2 knockdown confirmed that ACC2 is 276 kDa
and often appears as two bands that may correspond to different oligomerization states
(Figures S2G and S2H), consistent with studies showing that ACC2 exists in oligomeric
complexes and that phosphorylation pushes ACC to a monomeric state (Cho et al., 2010). The
presence of ACC2 in both bands was confirmed by IP-mass spectrometry (data not shown).
Collectively, our findings demonstrate that PHD3 modulates FAO at a physiologically significant
magnitude similar to the effect of other known lipid metabolism regulators including ACC2
itself, CPT1, adiponectin, and sirtuins (Fullerton et al., 2013; Gerhart-Hines et al., 2007; Laurent
et al., 2013; Yoon et al., 2006).

ACC2 gates long-chain fatty acid import into the mitochondria, whereas short- and medium-
chain fatty acids freely diffuse and bypass ACC2 regulation (Madden et al., 1995; ODonnell et
al., 2002). To validate the role of ACC2 in PHD3-mediated regulation of FAO, we probed
whether PHD3 specifically modulates longchain FAO. Comparison of the long-chain,16-C
palmitate versus 6-C hexanoate oxidation revealed that PHD3 knockdown specifically boosted
long-chain FAO (Figure 3C). We next tested the effect of PHD3 overexpression on fatty acid
synthesis, a process mediated by ACC1. Consistent with our studies showing that ACC2, but not
ACC1, is hydroxylated by PHD3; we observed no effect of PHD3 on fatty acid synthesis (Figures
S2I and S2J). In sum, PHD3 regulates FAO at the level of ACC2.

PHD3 Repression of FAO Occurs Independently of HIF and AMPK

Because HIF is the most well-characterized target of PHD family members, we assessed
dependence on the HIF regulatory axis.

PHD3 modulated FAO regardless of the presence or absence of HIF. PHD3 overexpression
repressed FAO, while PHD3 knockdown amplified FAO, in mouse hepatoma 4 (B13NBii1) HIFb-
null cells, which lack HIF1 transcriptional activity (Wood et al., 1996) (Figures 3D3F). The
absence of HIF activity was validated by treatment with CoCl2, which stabilizes HIFa, but did
not induce HIF target genes in these cells (Figure 3F). Additionally, HIF1/ 2a protein levels were
not changed by PHD3 knockdown (Figure 3G). Furthermore, PHD3 knockdown boosted FAO in
786- O VHL-deficient renal carcinoma cells with constitutively stabilized HIF (Iliopoulos et al.,
1996) (Figures 3H and S2K). We also found that PHD3 knockdown amplified FAO both under
normoxia (21% O2) and hypoxia (1% O2) (Figure 3I), consistent with reports that PHD3 is less
sensitive to oxygen levels tan PHD1 or PHD2 (Luo et al., 2011). In total, these data demonstrate
that PHD3 represses FAO independently of HIF.

We next examined whether PHD3 activates ACC2 in concert with the major known regulator of
this metabolic node, AMPactivated protein kinase (AMPK) (Ruderman and Prentki, 2004).

Upon detecting a low cellular energy status, AMPK inhibits ACC2 by phosphorylating serine
222, disrupting the dimer-dimer interface to block formation of the more active ACC oligomer
(Cho et al., 2010). In this way, AMPK activates FAO as part of a program to restore ATP levels.
PHD3 knockdown amplified FAO in both wild-type and AMPKa-knockout mouse embryonic
fibroblasts (MEFs) (Figures 3J, S2L, and S2M). PHD3 overexpression repressed FAO in the
absence of AMPKa (Figure S2N).

Additionally, we found that AMPK phosphorylated ACC under low nutrient conditions in both
control and PHD3-knockdown MEFs, and ACC phosphorylation decreased upon nutrient
restoration regardless of PHD3 status (Figure S2O). Thus, our data suggest that PHD3 regulates
FAO through a mechanism independent of AMPK.

PHD3 Hydroxylates ACC and Represses FAO in Response to Nutrient Abundance

In many tissues, fatty acids are not a predominant fuel choice under nutrient replete conditions
but rather are reserved for times of nutrient deprivation to restore metabolic homeostasis
(Gerhart- Hines et al., 2007). During conditions of stress or low energy, cells ramp up ATP
production by activating FAO via AMPK signaling. While AMPK boosts FAO by inhibiting ACC2,
our data show PHD3 has the opposite effect of repressing FAO by activating ACC2. Because
recent reports underscore a posible link between PHDs and cellular nutrient status (Kaelin and
Ratcliffe 2008; MacKenzie et al., 2007), we wondered whether PHD3 might be a candidate for
dynamically repressing FAO in response to nutrient abundance. In control vector-treated cells,
ACC was strongly hydroxylated in the presence of high glucose medium containing serum (high)
versus cells treated 12 hr with serum-free, low glucose medium (low) (Figure 3K, vector lanes).

However, PHD3 overexpression restored ACC hydroxylation to nearly that observed in the high
nutrient state (Figure 3K). This suggests that endogenous PHD3 hydroxylates and activates ACC
particularly when nutrients are abundant. In support of this model, PHD3 knockdown
abrogated ACC hydroxylation in high-nutrient medium (Figure 3L, left). In comparison, the
effect of PHD3 knockdown was less evident in low nutrient conditions (Figure 3L, right). We
next performed a time course analysis of ACC2 hydroxylation under high- versus low-nutrient
conditions.

ACC2 hydroxylation decreased following 6 hr in low glucose, serum-free medium, and

hydroxylation increased in a PHD3- dependent manner after only 10 min of restoring high-
nutrient medium (Figures 3M and S2P). PHD3 silencing most potently represses ACC2 activity
in the time frame immediately after restoring high nutrients (Figures S2Q and S2R). Thus, our
data suggest that PHD3 is an acute metabolic toggle responsive to cellular nutrient abundance.

We reasoned that cells with low PHD3 would lack this metabolic switch, uncoupling FAO
repression in nutrient abundance.

In multiple cell lines, palmitate oxidation was enhanced in serumfree, low glucose medium but
blunted in the presence of high glucose and serum (Figures 3N and S2S). However, upon PHD3
knockdown, cells lost sensitivity to external nutrient cues and displayed elevated FAO even in
the presence of high nutrients. Similarly, supplementing low-nutrient medium with cell-
permeable versions of the TCA cycle intermediate a-ketoglutarate for 6 hr prior to FAO analysis
repressed palmitate oxidation, unless PHD3 was downregulated by shRNA (Figure 3O). This
indicates that PHD3 limits FAO in nutrient-replete conditions and that nutrient deprivation lifts
PHD3-mediated repression of FAO.

Our findings suggest PHD3 activates ACC2 to inhibit CPT1. In support, metabolomics analysis
revealed that PHD3 knockdown by short hairpin RNA (shRNA) increased long-chain
acylcarnitines FAO intermediates generated by CPT1but short- and medium-chain
acylcarnitines, which bypass the ACC2/CPT1 regulatory node, were unchanged (Figure 3P). This
profile is reminiscent of the way in which glycolytic intermediates increase upon upregulation
of glycolysis (Finley et al., 2011), provided the initial substrate is not limiting; here, too, we
observe that longchain FAO intermediates are increased when FAO is upregulated by PHD3
silencing. Our data suggest that, together, AMPK and PHD3 toggle FAO in a manner that is
sensitive to both high and low nutrient status (Figure 3Q).

Low PHD3 Expression Drives Altered Metabolism in AML

In some cancers, FAO fuels energy production, antioxidant defense via NADPH production,
nucleotide synthesis, and maintenance of the mitochondrial membrane potential to prevent
apoptosis induction (Jeon et al., 2012; Samudio et al., 2010; Schafer et al., 2009; Schoors et al.,
2015; Zaugg et al., 2011). PHD3 downregulation is a common feature of multiple cancer types
(Rawluszko et al., 2013), and we hypothesized that low PHD3 levels might predict a
dependency on FAO in cancer.

We used bioinformatics to assess the metabolic impact of low PHD3 expression in cancer.
Analysis of the Ramaswamy, Valk, and Andersson leukemia datasets from the Oncomine cncer
gene expression database (https://www.oncomine.org/resource/ login.html) showed that
PHD3 expression was lower in AML compared to a panel of other cancerous tissues (Figures
4A4C) (Andersson et al., 2007; Valk et al., 2004).AML has been previously linked to increased
fatty acid catabolism (Carracedo et al., 2013; Liu et al., 2010); thus, we assessed the impact of
PHD3 on this phenotype. Using gene expression data from patient samples in The Cancer
Genome Atlas (TCGA), we clustered AML patients into two groups (PHD3-low and PHD3-high)
using a Gaussian mixture model based on the level of PHD3 expression. Nearly 80% of patients
fell into the low PHD3 group (Figures 4D and 4E). Gene Set Enrichment Analysis revealed that
the top curated gene sets inversely correlated with high PHD3 expression in AML are largely
markers of oxidative metabolism (Figures 4F and S3AS3D). Of note, we found no significant
link between PHD3 and expression of ACC2, AMPK, or LKB1 (Figures S3ES3G) in TCGA data.
These data support a model in which high PHD3 expression may indicate AML cells are not
fueled by oxidative metabolism, while low PHD3 expression can enable a switch toward FAO.

In line with patient data, PHD3 gene expression was nearly ndetectable in a panel of AML cell
lines (MOLM14, KG1, THP1, NB4, and U937) compared to the K562 chronic myeloid leukemia
(CML) cell line (Figure 4G). Low-PHD3 AML cells show reduced ACC hydroxylation and ATP
binding (Figures 4H and 4I) and markedly increased palmitate oxidation (Figure 4J).

PHD1 and PHD2 are not repressed to the same extent as PHD3 in AML cells (Figures S3H and
S3I), indicating that PHD3 expression is specifically linked to the observed traits.

Recent studies revealed that AML frequently displays enhanced dependence on FAO (Samudio
et al., 2010). Inhibition of FAO increases sensitivity to apoptosis in cell culture and in a murine
model of human AML (Estan et al., 2014; Lee et al., 2015). Furthermore, FAO was shown to be
critical for maintenance of hematopoietic stem cells and was suggested to be involved in the
maintenance of leukemic initiation cells (Ito et al., 2012). We hypothesized that low-PHD3
leukemia cells possess a metabolic liability rooted in their dependency on FAO, so we examined
leukemia cell sensitivity to etomoxir or ranolazine, FAO inhibitors that have shown success in
treating angina and heart disease (Holubarsch et al., 2007; Nash and Nash, 2008). Etomoxir
inhibits CPT1, and ranolazine inhibits 3-ketoacylthiolase, the enzyme catalyzing the final step of
b-oxidation (Carracedo et al., 2013). Within 96 hr, FAO inhibition led to substantial cell death in
a panel of low-PHD3 leukemia cells, while viability was maintained for high-PHD3 K562 cells
(Figures 4K4N). Another high-PHD3 CML cell line, MEG01, was less sensitive to a high dose of
ranolazine compared to low-PHD3 AML cells (Figures S3J and S3K). Sensitivity to FAO inhibition
was more strongly linked to PHD3 status than to classification as AML or CML; a CML cell line
with low PHD3 expression (KU812) was found to be sensitive to treatment with etomoxir and
more closely resembled another low-PHD3 AML cell line (NB4) than a high-PHD3 CML line
(K562) (Figures S3K and S3L). Thus, blocking fatty acid catabolism has a strong cytotoxic effect
specific to low-PHD3 leukemia cells.

In high-PHD3 K562 cells, PHD3 knockdown enabled higher FAO (Figures 4O and S3M); however,
it did not create a fixed dependency on FAO or cause susceptibility to FAO inhibitors (Figure
4P). K562 cells have a preference for glycolytic metabolism

(Barger et al., 2013), and, although PHD3 knockdown allowed higher FAO, it did not force these
cells to rely on fats. In contrast, cancer cells with low levels of PHD3 displayed limited
metabolic plasticity and required sustained FAO and were particularly susceptible to
pharmacological inhibitors of FAO. These data indicate that low PHD3 expression may be a
promising candidate as a biomarker for leukemia cells that may be successfully targeted with
FAO inhibitors.

Restoring PHD3 Expression in AML Limits FAO and Cell Growth through Activation of ACC2

Our data suggest that low PHD3 expression is advantageous in AML by boosting FAO. We
hypothesized that restoring PHD3 would limit the proliferation and potency of leukemia cells.

Indeed, stable PHD3 overexpression repressed FAO by over 50% (Figures 5A and S4A), a level
similar to that achieved by etomoxir (Figure S4B). This suggests that PHD3 affects FAO at a
magnitude similar to direct CPT1 inhibition. PHD3 overexpression also diminished cell
proliferation and viability (Figures 5B, 5C, S5A, and S5B).

To assess whether PHD3 overexpression was generally toxic, we overexpressed PHD3 in K562
cells. Endogenous PHD3 levels in MOLM14 and THP1 cells are 1% of that in K562 cells (Figure
4G), and PHD3 overexpression in these cells achieved an amount roughly 10- to 60-fold greater
than that found in K562 cells (Figure S4A). PHD3 overexpression in K562 cells had Little effect
on proliferation (Figures 5D5F), suggesting PHD3 overexpression is not generally toxic.
Moreover, knocking down PHD3 did not alter proliferation in K562 cells (Figures 5G and S5C),
supporting the idea that metabolic alterations due to modulating PHD3 are not detrimental to
all cancer cells. Instead, low-PHD3 cancer cells uniquely experience severe effects when PHD3
is restored.

Our model suggests that PHD3 overexpression impairs AML cells by activating ACC2 and
repressing fatty acid catabolism.

To test directly the requirement for ACC, we assessed the effect of PHD3 overexpression when
individual ACC isozymes were silenced. We generated MOLM14 cells that overexpressed PHD3
and also expressed shRNA against ACC2, ACC1, or a non-silencing control. Importantly, shRNA
were specific for each isozyme, and knockdown of one did not lead to compensatory
upregulation of the other (Figures S4C and S4D). As anticipated, ACC2 knockdown amplified
FAO more than ACC1 (Figure 5H), and knockdown of either isozyme enhanced cell growth
(Figure 5I), fitting with previously established roles of the enzymes (Jeon et al., 2012).
Strikingly, in the absence of ACC2, PHD3 overexpression was not able to blunt FAO or impair
ellular proliferation (Figures 5H and 5I; PHD3 expression levels in Figure S4E). ACC1 was not
required for these PHD3-mediated phenotypes. Consistent results were obtained in THP1 cells
(Figures 5J and S4FS4I). These data clearly demonstrate that ACC2 is essential for the
inhibitory action of PHD3 on FAO, AML survival, and growth.

Restoring PHD3 Expression Impedes AML Potency in Cell Culture and In Vivo

After establishing that PHD3 alters AML cell metabolism and proliferation, we probed the
impact of PHD3 overexpression on leukemic potency using colony formation assays to measure
viable and functional progenitor cells. Overexpressing PHD3 dramatically decreased the
number of clonogenic MOLM14 and THP1 cells in methylcellulose assays (Figures 6A and 6B).

However in the absence of ACC2, AML colonies grew robustly regardless of PHD3
overexpression (Figures 6C and S6A). In contrast, PHD3 overexpression impaired clonogenic
capacity independently of ACC1 (Figures 6C and S6A). We observed that high-PHD3 K562 cell
colony formation is inherently less sensitive to PHD3 modulation (Figures 6D and 6E). These
results indicate that the growth advantage of low-PHD3 AML cells is indeed due at least in part
to PHD3 levels, and that reintroducing PHD3 is detrimental to this subset of leukemia.

Moreover, ACC2 but not ACC1 is required for the inhibitory effects of PHD3.

We next asked whether PHD3 overexpression inhibits proliferation in primary AML samples.
Leukemic cells from patient samples obtained from the University of Pennsylvania showed
decreased PHD3 expression compared to healthy CD34+ control bone marrow cells (Figure 6F).
Overexpressing PHD3 decreased proliferation in two of three samples, while the remaining
sample trended toward a decrease (Figure 6G). PHD3 overexpression led to similar results in
leukemic cells derived from the MLL/AF9 mouse model of AML. MLL/AF9 chromosomal
translocation is a causative factor in a substantial subset of human AML and is associated with
a 5-year survival rate of only 40% (Noordermeer et al., 2012). Compared to healthy CD11b
control cells, PHD3 was strongly decreased in leukemic cells obtained from the MLL/AF9 mouse
model of AML and decreased to a lesser extent in the Hoxa9 Meis1 mouse model of AML
(Figure 6H). In MLL-AF9 lineage-negative bone marrow cells, PHD3 overexpression decreased
AML clonogenic capacity (Figures 6I and 6J). Thus, in low-PHD3 systems, PHD3 overexpression
limits leukemic potency.

Finally, we evaluated the in vivo impact of PHD3 overexpression in low-PHD3 AML cells using
an aggressive mouse xenotransplanation model. NOD-scid IL2Rgammanull (NSG) mice were
chosen for this analysis due to their superiority in allowing engraftment of human AML cells
(Shultz et al., 2005). Cohorts of NSG mice were injected via tail vein with MOLM14 cells
overexpressing HD3 or vector. The length of survival post-injection was used as a readout of
AML severity. We observed that PHD3 overexpression in AML enhanced survival (Figure 6K).
Taken together, these data strengthen our model that low-PHD3 leukemia cells possess a
metabolic liability rooted in ACC2 activation and a dependency on FAO, and that restoring
PHD3 levels limits the proliferation and potency of AML cells.

DISCUSSION

Here, we reveal a PHD3-mediated, nutrient-dependent signaling pathway. In response to

nutrient abundance, PHD3 hydroxylates and activates ACC2 in order to repress mitocondrial
FAO. The PHD3/ACC2 axis couples cellular nutrient status with dynamic regulation of FAO via a
mechanism that is parallel to, but distinct from, AMPK. When cellular energy supplies are low,
AMPK inhibits ACC2 to activate FAO and restore bioenergetic homeostasis (Cho et al., 2010).
Here, we reveal that PHD3 contributes to regulation of this metabolic node by activating ACC2
and limiting FAO when nutrients are abundant, thus enabling fuel conservation and metabolic
efficiency. Together, AMPK and PHD3 can achieve dual modulation of FAO in response to
changing nutrient levels. From our studies, a strong correlation between ACC2 hydroxylation
and oligomerization could not be seen, but it will be interesting for future studies to explore in
detail any possible feedback between PHD3 and AMPK. The link between PHD3 and nutrient
status aligns with previous reports suggesting that PHD3 might be sensitive to a-ketoglutarate
abundance, or more generally to a high nutrient state achieved by addition of a-ketoglutarate
(Koivunen et al., 2007). We hypothesize that PHD3 may be responsive to a pool of a-
ketoglutarate near the outer mitocondrial membrane, in close proximity to ACC2. Although
there is not currently a robust and reliable technique for measuring different intracellular pools
of a-ketoglutarate, future advancements in metabolomics may make investigating this idea
possible.

Because PHD3 acts on ACC2 and not ACC1, PHD3 loss may provide a way for the cell to elevate
FAO while maintaining lipid synthesis. One traditional view of lipid metabolism considers both
ACC isozymes to be similarly regulated, such that FAO is suppressed when lipogenesis is
elevated, and vice versa.
However, our results support a mechanism by which PHD3 can specifically activate ACC2 in
order to repress FAO, without impinging upon lipid synthesis. Studies of calorie restriction
establish a precedent for this ideacollectively, three studies in C. elegans indicate that FAO
and lipogenesis can be concurrently upregulated (Amrit et al., 2016; Ratnappan et al., 2014;
Steinbaugh et al., 2015). Further studies are needed to identify how PHD3 specifically targets
ACC2, perhaps via localization to the outer mitochondrial membrane.

Our studies offer insight into mechanisms that drive fatty acid addiction in tumors. We find that
reduced PHD3 expression is a common feature of AML that enables amplification of FAO. The
regulatory node revealed here may have implications for a broad ange of cancers. In addition
to AML, PHD3 is epigenetically silenced due to promoter hypermethylation in patient samples
of glioblastoma, B cell lymphoma subtypes, invasive breast cancer, and multiple myeloma, as
well as in some prostate and colon cancer cell lines (Rawluszko et al., 2013). Several of these
cancers are linked to FAO dependency, in particular, prostate, colorectal, and breast cancers, as
well as B cell lymphoma subtypes (Carracedo et al., 2013; Liu et al., 2010). Thus, it is possible
that PHD3 loss may play a role in driving FAO in these settings as well. PHD3 suppression likely
is not the only determinant of FAO dependency in cancer, in the same way that multiple
enzymes contribute to increased glycolysis in other tumor types.

For example, PML and PPARd have been linked to the maintenance and function of HSCs (Ito et
al., 2012), and future studies may reveal the role of other enzymes in cancer cell reliance on
fats.

Our data suggest a unique opportunity in the treatment of hematological malignancies. FAO
inhibitors are promising candidates for consideration in cancer treatment (Carracedo et al.,

2013; Hermanova et al., 2016; Rodrguez-Enrquez et al., 2015). The utility of FAO inhibitors
would be greatly aided by biomarkers that identify individuals who might most likely benefit
from such a treatment. Based on the knowledge of the regulatory axis identified here, low
PHD3 expression could be considered as a marker in cancer to identify patients who may be
successfully treated with FAO inhibitors, thus moving the field toward metabolically based
treatment options in the future.

EXPERIMENTAL PROCEDURES

Mass Spectrometry

To identify hydroxyproline sites, ACC2 was transiently overexpressed in 293T cells and
immunoprecipitated with ACC2 antibody (Cell Signaling Technology).

Bound material was washed and separated by SDS-PAGE. The Coomassie stained band was
analyzed by LC-MS2 and searched against the Uniprot Human database using Sequest with
proline hydroxylation set as a variable modification (+15.9949 molecular weight [MW] shift).

ACC Activity
Reactions were performed with 50 mg 293T cell protein lysates as previously described
(Pulinilkunnil et al., 2011), with the exception of using 16.7 mM MgCl2. Further details are
provided in Supplemental Experimental Procedures.

Growth Curves

Live cells were sorted by fluorescence-activated cell sorting (FACS) on day 0 and then counted
and plated in the wells of a 24-well or 96-well plate. At indicated times, cells were counted on
the Beckman Z1 Coulter Counter.

Colony-Forming Cell Assays

After sorting for live leukemic cells, cells were resuspended in methylcellulosebased medium
(MethoCult H4434 or M3434, Stem Cell Technologies) in the presence of puromycin and also
hygromycin, where indicated. For K562, 375 cells were plated per well of a 6-well plate. For
MOLM14, THP1, and MLL-AF9 cells, 5,000 cells were plated. colony-forming units (CFUs) were
counted 821 days later.

Xenotransplantation Studies

Male NOD-scid IL2Rgammanull (NSG) mice were obtained from Jackson Laboratory and housed
in specific pathogen-free environments. Mice were 7 weeks at the time of xenotransplantation.
On day 0, mice were sublethally irradiated (2.5 Gy) and injected via tail vein with 7 3 105
Molm14 cells overexpressing vector or HA-PHD3 in 250 ml PBS. Statistical evaluation of survival
was based on the log-rank test as well as the Gehan-Breslow-Wilcoxon test for comparison of
the Kaplan-Meier curves.

Statistical Analysis

Unpaired two-tailed Students t tests were used for comparison of FAO, gene expression,
activity assays, and growth and cell viability experiments. All experiments were performed at
least two to three times.