Abstract

Simple Summary

   Maintaining the energy stability is critical for cell surviving and
   adapting in a vagary environment. Glycolysis and mitochondrial
   oxidative phosphorylation (OXPHOS) are two major energy production
   pathways in cells. Here we demonstrated that menin regulates the
   expression of OXPHOS and glycolytic genes, and this regulation can
   further be modified by a group of menin-associated proteins (MAPs)
   including KMT2A, MED12, WAPL, and GATA3. Downregulation of menin and
   MAP genes altered the proportion of glycolysis and OXPHOS for energy
   generation, and we found a counteracting function of menin and MAPs
   when the shRNA knockdown cells are exposed to metabolic stress. Menin
   and MAPs may serve as transcriptional sensors for balancing the
   preference between glycolysis and OXPHOS. This coordinated regulation
   is crucial for cell adaption to stressful microenvironments using
   different pathways for energy production.

Abstract

   The interplay between glycolysis and mitochondrial oxidative
   phosphorylation (OXPHOS) is central to maintain energy homeostasis. It
   remains to be determined whether there is a mechanism governing
   metabolic fluxes based on substrate availability in microenvironments.
   Here we show that menin is a key transcription factor regulating the
   expression of OXPHOS and glycolytic genes in cancer cells and primary
   tumors with poor prognosis. A group of menin-associated proteins
   (MAPs), including KMT2A, MED12, WAPL, and GATA3, is found to restrain
   menin’s full function in this transcription regulation. shRNA
   knockdowns of menin and MAPs result in reduced ATP production with
   proportional alterations of cellular energy generated through
   glycolysis and OXPHOS. When shRNA knockdown cells are exposed to
   metabolic stress, the dual functionality can clearly be distinguished
   among these metabolic regulators. A MAP can negatively counteract the
   regulatory mode of menin for OXPHOS while the same protein positively
   influences glycolysis. A close-proximity interaction between menin and
   MAPs allows transcriptional regulation for metabolic adjustment. This
   coordinate regulation by menin and MAPs is necessary for cells to
   rapidly adapt to fluctuating microenvironments and to maintain
   essential metabolic functions.

   Keywords: menin, menin-associated proteins, oxidative phosphorylation,
   glycolysis, circulating tumor cells

1. Introduction

   Menin encoded by the MEN1 gene is a multifaceted protein participating
   in a variety of cellular functions [[46]1]. Earlier studies revealed
   that germline mutations accompanied by somatic alterations of MEN1 lead
   to the development of multiple endocrine neoplasia type 1 in endocrine
   glands [[47]2]. Through a two-hit mechanism, this tumor-suppressor
   inactivation has also been observed in lung, breast, and
   gastrointestinal cancers [[48]3,[49]4,[50]5]. Although MEN1 is well
   recognized as a classical tumor-suppressor gene, subsequent studies
   found aberrant upregulation of the gene in solid tumors and leukemias
   [[51]6,[52]7]. The gene product, menin, appears to interact with
   oncoproteins to promote transcription programs for cellular
   transformation and proliferation [[53]8,[54]9]. In particular, menin
   forms a complex with KMT2A (formerly MLL1) or other family proteins to
   activate gene transcription via histone methylation and deacetylation
   in solid tumors [[55]10,[56]11]. Recent findings indicated that menin
   collaborates with hormone signaling involved in the progression of
   breast and prostate cancers [[57]10,[58]12]. Indeed, the initial
   evidence points to complex regulatory roles of menin in response to
   different physiological conditions [[59]13].

   A growing body of studies has now implicated that menin acts as a
   transcriptional driver, engaging in two regulatory events for
   tumor-promoting and tumor-suppressing activities, respectively
   [[60]13,[61]14]. To do so, menin must interact with different factors
   to specify the active or suppressive mode for regulating gene
   transcription. Biochemical and crystal structural analyses revealed
   that menin is a scaffold protein providing multiple docking sites for
   transcriptional machinery components [[62]1,[63]14,[64]15]. Further
   studies found that menin individually cooperates with oncogenic
   signaling components, including TGF-β/SMDAs, NFκB, β-catenin, estrogen
   receptor alpha (ERα), and an androgen receptor, for transcriptional
   modulation of downstream targets responsible for cellular
   transformation and proliferation, DNA replication and repair, chromatin
   modification, and hormone resistance
   [[65]9,[66]12,[67]16,[68]17,[69]18]. While these studies give insights
   into a multifaceted role of menin in oncogenesis, it is still a
   challenge to know how menin interacts with diverse protein partners in
   cancer cells responding to different microenvironments.

   To this end, we combined RNA-seq transcriptomic profiles, BioID
   protein–protein interactions [[70]19], and in silico correlation
   comparisons to investigate crucial functions of menin in breast cancer
   cells. The rationale was that menin had been implicated as a
   proliferative determinant during breast tumorigenesis [[71]10]. In
   addition to confirming a role of menin in ERα-modulated signaling
   [[72]10,[73]18], our integrated analysis identified a group of
   menin-associated proteins (MAPs) for transcriptional regulation of
   genes involved in metabolic energy pathways, including glycolysis and
   mitochondrial oxidative phosphorylation (OXPHOS) [[74]20]. As a
   positive modulator, menin interacts with a MAP for metabolic regulation
   in a context-dependent manner. Whereas a MAP negatively influences
   OXPHOS, the same protein may positively regulate glycolysis.
   Furthermore, we set out to demonstrate that close-proximity
   interactions between these proteins are essential for the coregulation
   of metabolic gene transcription in response to microenvironmental
   changes.

2. Results

2.1. Gene Transcription of Energy Metabolism Pathways Modulated by MEN1

   Increased MEN1 expression was found in primary breast tumors relative
   to normal controls in The Cancer Genome Atlas (TCGA) cohort (p < 0.001,
   [75]Figure S1A). The upregulation was positively associated with
   elevated MEN1 copy-numbers (p < 0.0001, [76]Figure S1B). In addition,
   patients with ERα-positive tumors that displayed high expression levels
   or elevated copy-numbers of MEN1 appeared to have poor prognosis
   ([77]Figure S1C,D). To further investigate how this aberrant expression
   contributes to tumorigenesis, we performed an RNA-seq analysis of MEN1
   shRNA knockdown in two ERα-positive breast cancer cell lines T47D and
   MCF-7, respectively ([78]Figure 1A,B). Differentially expressed genes
   (i.e., fold change ≥1.5 and ≤0.66 for upregulated and downregulated
   genes, respectively, from two biological repeats) were identified in
   both shRNA knockdown cells relative to vehicle controls. Although the
   knockdown efficiencies (i.e., 88–95% at the RNA level and 40–60% at the
   protein level) were comparable in both cell lines, more differentially
   expressed genes were observed in T47D knockdown cells than in MCF-7
   knockdown cells ([79]Figure 1C). The gene set enrichment analysis also
   reflected more notable impacts of menin on signaling transduction
   pathways in T47D cells relative to MCF-7 cells ([80]Figure 1D).
   Consistent with a previous finding [[81]10], the pathway analysis
   confirmed a role of menin in regulating ERα signaling in breast cancer
   cells. Surprisingly, we found that metabolic energy pathways were the
   most affected by MEN1 knockdown ([82]Figure 1D,E). In particular, the
   MEN1 knockdown had a greater influence on genes encoding structural
   components and regulatory functions for glycolysis and OXPHOS
   ([83]Figure 1F). The RT-qPCR analysis of 16 representative genes
   involved in glycolysis and in two mitochondrial electron transport
   chains (i.e., OXPHOS complexes I and II) further confirmed the
   robustness of RNA-seq data ([84]Figure 1G). Collectively, the results
   find a previously unknown role of menin in regulating metabolic energy
   pathways.

Figure 1.

   [85]Figure 1
   [86]Open in a new tab

   Identification of MEN1-modulated genes in breast cancer cells. (A)
   RT-qPCR of MEN1 in T47D or MCF-7 cells treated with vehicle or MEN1
   shRNA lentivirus (n = 3). (B) Quantitative Western immunoassays (WES)
   of menin expression in T47D or MCF-7 cells treated with vehicle or MEN1
   shRNA lentivirus (n = 3). (C) Venn diagrams of differentially expressed
   genes (fold change ≥1.5 or ≤0.66) in T47D or MCF-7 cells after shMEN1
   knockdown compared with vehicle controls (n = 2). (D) Pathway
   annotation analysis of MEN1-upregulated and MEN1-downregulated genes in
   T47D or MCF-7 cells using DAVID including cancer hallmark pathways. (E)
   Schematic illustration of five major metabolic pathways. (F) Expression
   heat maps of oxidative phosphorylation (OXPHOS) and glycolytic genes in
   both MEN1 knockdown T47D and MCF-7 cells (fold changes relative to
   vehicle controls). (G) Bar charts of the expression levels of
   representative OXPHOS and glycolytic genes affected by MEN1 knockdown
   in T47D or MCF-7 cells using RT-qPCR. Data are presented as mean ± S.D.
   Unpaired two-tailed Student’s t-test was used for statistics. * p <
   0.05, ** p < 0.01, and *** p < 0.001.

2.2. Energy Metabolism Genes Coregulated by Menin and Menin-Associated
Proteins in Primary Tumors

   As menin is a putative metabolic regulator, we postulated that this
   protein acts as a crucial component of a transcription complex to bind
   or interact with other factors for gene regulation. Therefore, we used
   a proximity-based biotin labeling technique or BioID to identify
   menin-associated proteins (MAPs) in T47D and MCF-7 cells ([87]Figure
   S2A; details in Methods). Proteins within a 10 nm radial proximity of
   ligase-fused menin were biotinylated in breast cancer cells transduced
   with doxycycline-inducible BirA-menin in the presence of biotin and
   doxycycline, and the labeled proteins were pulled down by
   streptavidin-beads ([88]Figure 2A,B) [[89]19,[90]21]. Analysis of
   liquid chromatography–mass spectroscopy (LC–MS/MS) on biotinylated
   proteins identified 248 and 132 putative nuclear MAPs in T47D and MCF-7
   cells, respectively ([91]Figure 2C). Based on known functions, MAPs
   were categorized into eight groups involved in basal transcription, DNA
   repair, mRNA modeling, nuclear scaffolding, site-specific
   transcription, cohesin complexes, and histone modification ([92]Table
   S1). Of the 35 MAPs shared by T47D and MCF-7 cells, ten were detectable
   even in control cells without biotin added, suggesting the presence of
   low endogenous biotin and probably their strong interactions with menin
   ([93]Figure 2D). The spatial nearness of individual MAPs to menin was
   estimated according to biotinylated protein quantities ([94]Figure 2E;
   [95]Figure S2B). To check the robustness of BioID findings, capillary
   Western immunoassays (WES) were conducted on nuclear and cytoplasmic
   lysates of T47D and MCF-7 cells, respectively, using specific
   antibodies against menin and four representative MAPs—KMT2A, MED12,
   WAPL, and GATA3 ([96]Figure 2F). The assays confirmed the association
   of these MAPs with menin was predominant in the nuclei, implicating
   their partnership with menin in gene regulation.

Figure 2.

   [97]Figure 2
   [98]Open in a new tab

   Identification of menin-associated proteins (MAPs) in breast cancer
   cells. (A) and (B) WES of BirA-Menin fusion proteins (A) and
   biotin-labeled proteins (B) in total lysates of BirA-MEN1 BioID
   engineered T47D or MCF-7 cells after incubating with or without
   doxycycline and biotin. (C) Schematic purification and proteomic
   identification of MAPs using LC–MS/MS. (D) Heatmap of the
   quantification of 35 MAPs commonly shared in T47D and MCF-7 cells. MAPs
   further verified by WES immunoassays were indicated by arrows. (E)
   Network analysis of 35 MAPs in MCF-7 cells. The distance between menin
   and MAPs represented the quantitative ratio of each MAP and menin. MAPs
   marked in blue were further assayed by WES. (F) Nuclear or cytoplasmic
   lysates of BirA-MEN1 BioID engineered T47D or MCF-7 cells after
   streptavidin beads pull-down were detected by WES with antibodies
   against menin, KMT2A, MED12, WAPL, GATA3, LaminA/C, or GAPDH. FL, full
   length; SP, spliced form.

   We further performed an in silico analysis of chromatin
   immunoprecipitation sequencing (ChIP-seq) data using anti-menin and
   anti-KMT2A [[99]10]. The menin and KMT2A binding sites are localized in
   the proximity of promoters and gene bodies of representative glycolysis
   and OXPHOS genes, suggesting a transcription regulatory role of menin
   and KMTs on these genes ([100]Figure S2C,D). Some of the menin and
   KMT2A binding sites are colocalized in some regions of gene loci. These
   confirm that menin and KMT2A are closely associated and likely
   coregulate these energy metabolism genes.

   Since the role of MAPs in controlling energy metabolism remains to be
   explored, we sought to determine the coexpression relationship between
   menin/MAP genes and OXPHOS/glycolytic genes. In silico analysis
   generated pairwise expression correlation heatmaps in 1031 primary
   breast tumors and 98 normal controls of the TCGA cohort ([101]Figure
   3A,B; [102]Figure S3A,B). We observed that expression of MEN1 was
   moderately negatively correlated (r = −0.22) with the mean OXPHOS gene
   expression in normal controls but both turned to being positively
   correlated (r = 0.35) as they progressed into malignant tumors
   ([103]Figure 3C). In contrast, increased expression of three MAP
   genes—KMT2A, MED12, and WAPL—showed strong negative correlation with
   OXPHOS expression in both tumors and controls while GATA3 exhibited no
   correlation. It is notable that the expression of MEN1 was also
   positively correlated with that of glycolytic genes, although to a
   lesser degree than OXPHOS genes ([104]Figure 3D). There was no
   expression correlation between glycolytic genes and KMT2A, MED12, or
   WAPL, except for GATA3 showing a weak negative association. Next, we
   determined the coexpression relationship between MAP genes and MEN1 in
   primary tumors that were divided into four groups based on median
   expression levels: 1: high (above the median value) MAP and low (below
   the median value) MEN1; 2: high MAP and high MEN1; 3: low MAP and low
   MEN1; and 4: low MAP and high MEN1 ([105]Figure 3E and [106]Figure
   S3C). A trend of increased mean expression of OXPHOS genes (complex
   I-V) could be observed from Group 1 to 4. For example, high KMT2A and
   low MEN1 expressed tumors (Group 1) displayed lowest OXPHOS expression.
   In contrast, tumors with a reverse expression pattern of KMT2A and MEN1
   (Group 4) had the highest OXPHOS expression ([107]Figure 3—left panel).
   Based on the data here and the forementioned menin1 as a positive
   regulator for OXPHOS gene expression, KMT2A seems to suppress MEN1
   function. This negative regulation was also seen between MEN1 and two
   other MAP genes (MED12 and WAPL, [108]Figure 3E—middle two panels), but
   was not evident for GATA3. However, the negative coexpression
   relationship of MEN1 and three MAPs was not observed for glycolytic
   gene expression in the Group 1–4 tumors. Overall, the in silico
   analysis suggests an essential role of menin in upregulating cancer
   energy metabolism and the three MAPs may counteract menin, particularly
   in the transcription of OXPHOS genes.

Figure 3.

   [109]Figure 3
   [110]Open in a new tab

   Expression correlation relationship of menin/MAPs genes and
   OXPHOS/glycolytic genes. (A) Workflow of the in silico correlation
   analysis of gene expression in The Cancer Genome Atlas (TCGA) breast
   cancer cohort. (B) Heatmaps of the expression correlation between
   MEN1/selected 4 MAP genes and OXPHOS genes (upper) or glycolytic genes
   (lower) in normal (N) and tumor (T) samples. The genes are arranged
   from the highest to the lowest according to gene expression correlation
   coefficients of MEN1–OXPHOS genes or MEN1–glycolytic genes in breast
   tumors. (C) and (D) Scatter plots and linear regression analyses of
   MEN1/selected MAPs expression and mean expression of OXPHOS genes (C)
   or glycolytic genes (D) in normal and tumor samples. (E) Violin plots
   (lower panel) shows the average expressions of the genes of OXPHOS
   complexes I-V and glycolysis in the samples of each of the
   corresponding 4 groups are shown as violin plots. Based on the median
   values (where ≥median is “high” and <median is “low”) of the expression
   of the corresponding individual genes (KMT2A, MEN12, WAPL, and GATA3)
   and MEN1, the TCGA breast tumor samples were divided into 4 groups—1:
   high-low, 2: high-high, 3: low-low, and 4: low-high (upper panel).
   Letters on top of the violin plot denote statistical significance,
   where two groups with different letters are significantly different (p
   < 0.05) and those with the same letter are not.

2.3. Counteracting Roles of MAPs in Menin-Modulated OXPHOS and Glycolysis

   To mechanistically validate menin/MAPs-coregulated energy metabolism
   outcomes, we first carried out shRNA knockdowns of MEN1 and four MAP
   genes (KMT2A, MED12, WAPL, and GATA3), which led to expression
   alterations of glycolytic and OXPHOS genes analyzed in T47D and MCF-7
   cells (see examples in [111]Figure 1G and [112]Figure S4A–C). Then, we
   determined whether these knockdowns affect ATP profiles in breast
   cancer cells based on the oxygen consumption rate (OCR) and
   extracellular acidification rate (ECAR) assays. The MEN1 knockdown
   resulted in reduced ATP production from both OXPHOS and glycolysis in
   T47D and MCF-7 cells ([113]Figure 4A,B). Individual knockdowns for
   KMT2A, MED12, WAPL, or GATA3 suppressed the overall ATP generation to a
   lesser degree than the MEN1 knockdown ([114]Figure 4A,B). The finding
   demonstrates that MAPs appeared to be regulators of ATP synthesis
   derived from the two metabolic pathways.

Figure 4.

   [115]Figure 4
   [116]Open in a new tab

   Bioenergetic dynamics are regulated by menin and MAPs in T47D and MCF-7
   cells. (A) and (B) Glycolytic and OXPHOS ATP productions in T47D (A) or
   MCF-7 (B) cells infected with vehicle, shMEN1, shKMT2A, shMED12,
   shWAPL, or shGATA3 lentivirus. Statistics represented the difference of
   glycolytic or OXPHOS ATP production between shRNA knockdown and vehicle
   controls. (C) and (D) Bar charts representing mitochondrial functions
   in the single knockdown of MEN1, KMT2A, MED12, WAPL, or GATA3 and their
   vehicle control in T47D (C) or MCF-7 (D) cells. (E) Schematic summary
   of mitochondrial dynamics affected by the knockdown of MEN1 or MAPs.
   (F,G) Bar charts represented the glycolytic functions in T47D (F) or
   MCF-7 (G) cells subject to gene knockdown by shMEN1, shKMT2A, shMED12,
   shWAPL, or shGATA3 lentivirus. (H) Schematic summary of glycolytic
   functions affected by the knockdown of MEN1 or MAPs. Data are presented
   as mean ± S.D. (n = 15–20 technical-replicate wells). Statistical
   significance was performed by an unpaired two-tailed Student’s t-test
   between treated groups and corresponding controls. * p < 0.05, ** p <
   0.01, and *** p < 0.001.

   To explore how menin or a MAP affects different OXPHOS processes, we
   added inhibitors to induce metabolic stress in shRNA knockdown cells or
   vehicle controls ([117]Figure S4D–F). Prior to the inhibition, MEN1
   knockdown caused low basal OCRs, consistent with attenuated ATP
   production from OXPHOS in T47D and MCF-7 cells (p < 0.001, [118]Figure
   4C,D—upper panels). Basal respiration showed a slight increase or
   decrease in KMT2A, MED12, WAPL, or GATA3 knockdown cells compared to
   vehicle controls. When the mitochondrial membrane potential was
   additionally disrupted by carbonyl cyanide-4 (trifluoromethoxy)
   phenylhydrazone (FCCP) [[119]22], MEN1 knockdown cells exhibited
   defects in both maximal respiration potential and spare respiratory
   capacity for energy demand ([120]Figure 4C,D—middle and lower panels).
   The findings indicated that menin plays a crucial role in the
   modulation of these OXPHOS functions. On the contrary, when KMT2A,
   MED12, or GATA3 was knocked down, we observed an increase in both
   maximal and spare respiration rates in these cells, suggesting that
   these four MAPs are negative modulators of OXPHOS activities that are
   mainly consistent with transcription reduction ([121]Figure 3C,E and
   [122]Figure 4E).

   We also investigated the glycolytic functions of cells subjected to
   shRNA knockdown and metabolic stress (i.e., transient glucose depletion
   for 1 h; [123]Figure S4G–I). When glucose was added back to the media,
   all five knockdown cells (MEN1, KMT2A, MED12, WAPL, and GATA3)
   displayed an attenuated ability to fully utilize glucose via glycolysis
   ([124]Figure 4F,G—upper panels). In addition, there was a general
   decrease of utilizing the glycolytic reserve in knockdown cells, except
   for T47D cells with GATA3 knockdown ([125]Figure 4F,G—middle and lower
   panels). Taken together, these metabolic stress tests found that menin
   and the four MAPs mainly act as positive regulators of glycolysis
   although primary tumor transcription data do not show the relevance
   ([126]Figure 3D,E and [127]Figure 4H).

2.4. Essential Interactions between Menin and KMT2A for OXPHOS Functions

   The above findings prompted us to investigate whether physical
   interactions between menin and MAPs are required for metabolic
   regulation. To this end, we determined how the small molecule inhibitor
   MI-503 [[128]23] disrupts the menin-KMT2A interaction that negatively
   influences downstream ATP production in breast cancer cells. After
   optimizing treatment doses based on a cell viability test ([129]Figure
   S5A), we treated T47D and MCF-7 cells with or without 1 μM MI-503 for
   72 h and then determined the status of menin–MAP interactions using
   coimmunoprecipitation and WES. MI-503 treatment caused a modest loss of
   menin and two MAP proteins—KMT2A and GATA3—likely resulting in an
   unstable menin complex that may contribute to a decline in cell
   viability ([130]Figure 5A). The presence of MI-503 greatly abolished
   the menin and KMT2A interaction but led to a 2-fold increase in the
   binding between menin and a GATA3 variant (GATA3-FL) in T47D cells
   ([131]Figure 5B). The menin-KMT2A interaction was also affected in
   MI-503-treated MCF-7 cells to a lesser degree (0.5×; [132]Figure 5B).
   However, we observed a 1.5-fold increase in the binding between menin
   and two GATA3 variants (GATA3-FL and GATA3-SP) in MCF-7 cells. In
   addition to KMT2A and GATA3, we also demonstrated that another MAP,
   WAPL, also coimmunoprecipitated with menin in T47D providing another
   line evidence of physical association between menin and MAPs
   ([133]Figure S5B).

Figure 5.

   [134]Figure 5
   [135]Open in a new tab

   Integrity of the menin–KMT2A complex is required for OXPHOS functions.
   (A) WES of T47D or MCF-7 cells treated with DMSO or 1 μM MI-503 for 3
   days (left). Relative protein expression normalized to the average of
   LaminA/C in WES (right). FL, full length; SP, spliced form. (B) Nuclear
   lysates of T47D or MCF-7 cells treated with DMSO or 1 μM of MI-503 for
   3 days were immunoprecipitated with the menin antibody or IgG, and
   assayed by WES (upper). Relative protein expression in WES (lower). The
   protein expression in DMSO treated input was normalized as 1. FL, full
   length; SP, spliced form. (C) Glycolytic or OXPHOS ATP production in
   T47D or MCF-7 cells treated with 1 μM MI-503 for 0, 1, 3, 6, and 72 h,
   or DMSO control for 72 h. (D,E) Bar charts of the Seahorse
   mitochondrial stress test (D) and glycolytic stress test (E) on T47D or
   MCF-7 cells treated with 1 μM MI-503 for 0, 1, 3, 6, and 72 h, or DMSO
   for 72 h. Data are presented as mean ± S.D. (n = 10–15
   technical-replicate wells). An unpaired two-tailed Student’s t-test was
   used to determine statistical significance for the difference between
   MI-503-treated groups and its controls. * p < 0.05, ** p < 0.01, and
   *** p < 0.001.

   Although the possibility of MI-503’s direct blocking of OXPHOS ATP
   production was not ruled out, cells at 1–6 h post-treatment by MI-503
   showed a transient decrease in OXPHOS gene expression, but could
   partially be recovered to the original levels at 72 h ([136]Figure
   S5C), indicating a transcriptional regulation by menin-MAPs. As a
   result, a drastic shift of ATP production from OXPHOS to glycolysis
   occurred after treating breast cancer cells with MI-503 ([137]Figure
   5C). The reduced ATP synthesis from OXPHOS was largely reflective of
   diminished basal, maximal, and spare respiration capacity at 1–6 h
   after the treatment ([138]Figure 5D and [139]Figure S5E). This
   respiratory inhibition was slightly reversed at 72 h post-treatment. We
   postulated that the attenuated OXPHOS transcription could be attributed
   in part to the disruption of menin-KMT2A by MI-503, leading to the
   defects in OXPHOS machineries.

   Dissociation of menin-KMT2A, however, did not reduce glycolytic ATP
   generation in MI-503-treated cells while there were temporal expression
   changes of glycolytic genes ([140]Figure S5D). Instead, we observed a
   sharp increase in the glycolytic ATP level to compensate for the loss
   of OXPHOS ATPs in treated cells. Metabolic fluxes might contribute to
   this energy compensation [[141]24], even though the expression of the
   corresponding genes was transiently affected by MI-503. When glucose
   was added back to the media after a 1 h depletion in the metabolic
   stress assay, MI-503-treated cells displayed an even greater ability
   than untreated cells to utilize glycolysis ([142]Figure 5E and
   [143]Figure S5F). While the level of glycolytic capacity remained
   stable at 1–6 h after the treatment, the glycolytic reserve became
   largely depleted in treated cells. In this case, ATP compensation came
   predominantly from the glycolytic reserve function when the OXPHOS
   pathway was blocked in MI-503-treated cells. At 72 h, we observed that
   the overall glycolytic function could be even more efficiently restored
   than OXPHOS. While dimethyl sulfoxide (DMSO), the solvent to dissolve
   MI-503, is known to negatively affect cellular glucose uptake
   [[144]25], these glycolytic functions seemed to work better in
   MI-503-treated cells than in control cells exposed to long-term DMSO.
   The finding might be compatible with an overall recovery of glycolytic
   gene expression in cancer cells after a long-term 72 h treatment with
   MI-503. Although the menin-KMT2A disruption negatively influences
   glycolytic gene expression, our finding suggests that this transient
   interference may not have an adverse effect on downstream glycolysis
   function, which is used to compensate for ATP loss from OXPHOS.

2.5. Upregulation of MEN1, OXPHOS, and Glycolytic Genes in Circulating Tumor
Cells

   We next used circulating tumor cells (CTCs) as in vivo surrogates to
   validate our findings based on cultured cell models and in silico
   analyses ([145]Figure S6A). CTCs isolated from five breast cancer
   patients were briefly cultured in a sphere-forming medium (#91130,
   Fujifilm Irvine Scientific, Santa Clara, CA, USA) for enrichment
   [[146]26]. Epithelial-like and cancer-stemness features of CTCs were
   confirmed with an immunofluorescence analysis [[147]27,[148]28]
   ([149]Figure S6B). Single-cell RT-qPCR of 93 CTCs was conducted with a
   gene panel of MEN1, KMT2A, MED12, WAPL, and GATA3 and 18 representative
   genes of OXPHOS and glycolysis. t-distributed stochastic neighbor
   embedding (t-SNE) profiling identified five clusters (#01–05;
   [150]Figure 6A), and violin plots displayed CTC clusters from high to
   low expression levels of MEN1 ([151]Figure 6B). Cluster #01 and 02 had
   high MEN1 expression, which also corresponded to increased levels of
   OXPHOS and glycolytic genes ([152]Figure 6B). Besides, the expression
   levels of the four MAP genes—KMT2A, MED12, WAPL, and GATA3—were lower
   than that of MEN1 in these two clusters.

Figure 6.

   [153]Figure 6
   [154]Open in a new tab

   MEN1 and OXPHOS expression are increased in breast circulating tumor
   cells (CTCs). (A) t-SNE profile plots and cell clustering of 93 CTCs
   from 5 breast cancer patients based on the single cell RT-qPCR
   expression profiling of 11 OXPHOS genes (NDUFA7, NDUFA11, NDUFA13,
   NDUFB7, NDUFS7, NDUFS8, NDUFV1, SDHA, SDHB, SDHC, and SDHD). (B) Violin
   plots of MEN1 or selected MAPs expression, mean expression of 7
   glycolytic genes (ALDOA, ALDOC, ENO1, PFKL, PFKP, PGK1, and TPI1) or
   mean expression of 11 OXPHOS genes (aforementioned) in the five cell
   clusters. Statistical significance among clusters was carried out using
   the Duncan multi-range test. (C) Mean expression of 7 glycolytic genes
   and 11 OXPHOS genes in these 93 breast CTCs or in the TCGA primary
   breast cancer cohort. (D) Glycolytic and OXPHOS ATP productions of T47D
   or MCF-7 cells after circulation (n = 6–10 technical replicates). (E,F)
   Mitochondrial (E) and glycolytic (F) functions of T47D or MCF-7 cells
   after circulation (n = 5–9 technical-replicate wells). Statistics
   represented the difference between no circulating control and each
   treatment. Data are presented as mean ± S.D. An unpaired two-tailed
   Student’s t-test was used for statistical significance determination. *
   p < 0.05, ** p < 0.01, and *** p < 0.001.

   The high expression profiles of OXPHOS and glycolytic genes in CTCs
   (particularly those in #01 and 02) were distinct from those observed in
   TCGA primary tumors, known to preferentially depend on glycolysis for
   the Warburg metabolism [[155]29] ([156]Figure 6C). Therefore, we
   suggest that high OXPHOS and glycolysis activities can be a metabolic
   adaptation of CTCs to survive under biophysical and physiological
   stress during the blood circulation [[157]30]. To further investigate
   the impact of this microenvironmental change on energy metabolism, T47D
   and MCF-7 cells were subjected to circulation in an in vitro system
   simulating capillary flows [[158]31] ([159]Figure S6C). After 2 h in
   the circulation, less than 50% of cells were viable for ATP
   measurements ([160]Figure S6D). These post-circulation cancer cells
   displayed slight decreases of ATP production from both glycolysis and
   OXPHOS ([161]Figure 6D). When these cells were subjected to metabolic
   stress tests, basal respiration rates were slightly lowered in
   post-circulation cells, compared to non-circulation controls
   ([162]Figure 6E—upper panels). However, maximal rates remained
   unchanged due to the contribution of spare respiratory capacity from
   these post-circulation cells ([163]Figure 6E). Although T47D and MCF-7
   cells responded differently with respect to their glycolytic functions,
   the glycolytic reserve was moderately increased in both cell lines
   ([164]Figure 6F). Taken together, we suggest that the enhancement of
   both spare/reserve energy capacities could be partially responsible for
   the survival of cancer cells in the blood circulation.

3. Discussion

   Glycolysis and OXPHOS are two major metabolic pathways to provide
   sufficient ATP for the maintenance of essential cellular functions and
   cancer growth [[165]20]. These pathways are tightly regulated,
   depending on the availability of oxygen or glucose in microenvironments
   [[166]32]. While previous reports focused on post-translational and
   post-transcriptional controls of glycolysis and OXHPOS, emerging
   evidence reveals that metabolic genes are frequently coregulated at the
   transcriptional level [[167]33,[168]34,[169]35,[170]32]. A few
   transcription factors are known to have a direct role in the regulation
   of energy metabolism [[171]34,[172]37,[173]38]. For example, HIF-1α or
   SIX1 binds to conserved DNA motifs in promoter regions of multiple
   glycolytic genes, enhancing the transcription for the biosynthesis of
   glycolytic pathway enzymes [[174]32,[175]39]. Genomic analysis also
   revealed coregulatory modules for synchronized transcription of OXPHOS
   genes [[176]33,[177]35]. Menin promotes glycolysis through MYC for
   tumor progression [[178]8]. Our present findings further implicate that
   menin acts as a key factor to concordantly regulate the transcription
   machineries of both metabolic pathways. The coordinate regulation by
   menin is necessary for cells to constantly maintain energy homeostasis
   and to rapidly adapt to fluctuating microenvironments.

   Assisting in menin-mediated transcription is a group of coregulatory
   partners or MAPs based on a protein–protein interaction assay based on
   the data from the four representative MAPs as more relation details
   remain to be determined. With limited data of the shRNA knockdown
   effect on OCR and ECAR, WAPL, GATA3, and MED12 may act like KMT2A as
   menin’s coregulators in energy metabolism as depicted in [179]Figure
   4E,H. These four and likely other MAPs may compete for association or
   binding with menin as more GATA3 isoforms are coimmunoprecipitated with
   menin when KMT2A-menin binding is interfered by MI-503. The menin–MAPs
   complex dynamics may be regulated by menin upstream signaling pathways
   and also important for the regulation of energy metabolism.

   While menin is a positive regulator of glycolysis and OXPHOS, MAPs
   contribute to transcriptional switches between the two pathways. For
   example, KMT2A may counteract the positive regulatory mode of menin for
   OXPHOS transcription although the same protein facilitates active
   transcription of many glycolytic genes. It has been suggested that
   KMT2A interacts with multiple partners and is cleaved by taspase 1 into
   two fragments inside a cell—the N-terminal domain, known to bind
   transcription repressors, and the C-terminal domain, which is a
   transcriptional activator [[180]40,[181]41]. Future investigation is
   warranted to determine if the two protein substructures independently
   regulate the transcription of OXPHOS and glycolytic genes.
   Nevertheless, this transcriptional coregulation is needed when growing
   tumor cells encounter a hypoxic microenvironment. The coupled effect
   can be less prominent when cancer cells prefer OXPHOS for ATP
   generation in nutrient-rich culture media (see [182]Figure 4A,B).
   However, the dual functionality (positive and negative coregulation)
   can clearly be distinguished when KMT2A shRNA knockdown cells are
   subjected to metabolic stress tests. This transcriptional plasticity is
   further shown in CTCs that exhibit both high OXPHOS and glycolysis
   activities ([183]Figure 6C—left panel), compared with primary tumors
   that dominantly utilize the latter ([184]Figure 6C—right panel). In
   addition, we demonstrated that CTCs exploit the reserve functions of
   both pathways for survival in an in vitro circulation experiment.

   A close-proximity interaction between a MAP and menin is essential for
   metabolic adjustment. Specifically, KMT2A catalyzes H3K4 trimethylation
   for open promoter chromatin to allow adjacent menin to recruit RNA
   polymerase II and other transcription factors to bind to the region
   [[185]42,[186]43]. Therefore, MI-503-mediated disruption of menin-KMT2A
   leads to an abolishment of OXPHOS gene transcription. As a result,
   decreased processing and synthesis of related proteins may lead to
   defects in OXPHOS complexes and their function for ATP generation. The
   tethering of menin-KMT2A is also important for transcription regulation
   of glycolytic genes. In the presence of MI-503, we also observed a
   decrease in glycolytic transcription activities, which likely lead to a
   reduced level of glycolytic enzymes. However, this transient inhibition
   is less impactful for glycolysis than OXPHOS. One possible explanation
   points to the required structural assembly of multiple OXPHOS complexes
   as a functional unit in the electron transport chain [[187]44] while
   glycolytic enzymes operate independently to catalyze their
   corresponding intermediates for energy production. Therefore, although
   the MI-503 treatment greatly reduces ATP production from OXPHOS,
   existing glycolytic enzymes continue to operate and are capable of
   compensating for the energy loss.

4. Materials and Methods

4.1. Cell Cultures, shRNA Knockdown, and MI-503 Treatment

   Human breast cancer cell lines T47D and MCF-7 were obtained from the
   ATCC and cell authentication was conducted at the ATCC using short
   tandem repeat DNA profiling. MISSION shRNA plasmids, shMEN1
   (TRCN0000040141), shKMT2A (TRCN0000005956), shMED12 (TRCN0000018578),
   shWAPL (TRCN0000167548), and shGATA3 (TRCN0000019301), were obtained
   from Sigma-Aldrich (sequences listed in [188]Table S2). Vehicle plasmid
   pLKO.1-puro (Addgene plasmid #8453) was obtained from Addgene. Plasmids
   were purified by QIAprep Spin Miniprep kit (Qiagen), and 293T cells
   were cotransfected with 3^rd Generation Packaging Mix (Applied
   Biological Materials, Richmond, BC, Canada) combined with either the
   shRNA plasmid or the vehicle plasmid for lentivirus production. The
   lentivirus-containing medium was harvested after 48 h of transfection,
   and cells were infected with lentivirus for 24 h. The infected cells
   were selected using puromycin at 4 μg/mL for 1 week. MI-503
   (Selleckchem, Huston, TX, USA) was dissolved in dimethyl sulfoxide
   (DMSO) as a 10 mM stock and diluted to the working concentrations in
   the medium.

4.2. RNA-seq and Quantitative Reverse Transcription PCR (RT-qPCR)

   Total RNA was isolated by PureLink RNA Mini Kit (Thermo Fisher
   Scientific, Waltham, MA, USA) with two biological replicates.
   Sequencing of cDNAs was performed with Illumina HiSeq3000, and the
   paired-end 100 bp FASTQ files were generated. We used STAR alignment
   software to align the sequencing data with the human reference genome
   GRCh38, and RSEM software to quantify the expression levels as
   fragments per kilobases of transcript per million (FPKM) mapped reads.
   Upregulated and downregulated genes were defined by a fold change ≥1.5
   and ≤0.66, respectively, in shMEN1 knockdown cells relative to vehicle
   controls after filtering out the low expressed genes (FPKM values <
   10). The pathway enrichment analysis was performed by the Database for
   Annotation, Visualization, and Integrated Discovery (DAVID,
   [189]https://david.ncifcrf.gov). The heat maps were generated using
   log[2] (fold change) based on normalized FPKM values. For RT-qPCR, the
   total RNA was isolated by the PureLink RNA Mini Kit (Thermo Fisher
   Scientific, Waltham, MA, USA) and the complementary DNA (cDNA) was made
   from 1 μg of total RNA by high-capacity cDNA reverse transcription kits
   (Applied Biosystems, Foster City, CA, USA). The primers of qPCR were
   listed in [190]Table S3 (F1 and R1). qPCR quantification was performed
   by LightCycler 480 SYBR Green I Master (Roche, Basel, Switzerland) with
   3 biological repeats, and relative expression levels of genes were
   calculated using the 2^−ΔΔCt method.

4.3. Proximity-Dependent Biotin Identification (BioID) Assay

   Cloning and construction of the Tet-on inducible BioID construct were
   previously described [[191]21]. Briefly, the pRetroX-mycBioID-MCS
   vector was cloned from the pcDNA3.1mycBioID plasmid (Addgene plasmid
   #35700) and pRetroX-Tight-Pur (Clontech). The full-length human MEN1
   cDNA from pBABE hygro MEN1 WT (Addgene plasmid #11024) was cloned into
   pRetroX-mycBioID-MCS at Not1 and Mlu I site by the Gibson reaction (New
   England Biolabs, Ipswich, MA, USA) to obtain pRetroX-mycBirA-MEN1.
   Engineered parental T47D and MCF7 cell lines that stably express a Tet
   repressor were infected with mycBirA-MEN1 retrovirus and stable cell
   lines were obtained under the selection of G418 (600 μg/mL for T47D and
   300 μg/mL for MCF-7) and puromycin (1 μg/mL). To induce mycBirA-menin
   protein expression, stable T47D and MCF-7 cells were incubated in the
   medium with 2 μg/mL of doxycycline for 24 h and then in the medium with
   50 mM biotin for another 24 h. For WES, the total lysate was collected
   by the RIPA buffer (Thermo Fisher Scientific, Waltham, MA, USA)
   containing the protease/phosphatase inhibitor (Thermo Fisher
   Scientific, Waltham, MA, USA); the cytoplasmic and nuclear lysates were
   isolated by NE-PER™ Nuclear and Cytoplasmic Extraction Reagents (Thermo
   Fisher Scientific, Waltham, MA, USA). For LC–MS/MS, cells were lysed by
   the RIPA buffer containing protease/phosphatase inhibitors. Free
   biotins were filtered out by 10 kDa filters (Sigma-Aldrich, St. Louis,
   MO, USA) and the biotin-labeled proteins were captured by incubation
   with 200 μL of streptavidin magnetic beads (Thermo Fisher Scientific,
   Waltham, MA, USA) for 5 h at 4 °C and collected by a magnetic stand.
   The beads were subject to 5 washes with 1 mL of RIPA buffer for 10 min
   at 4 °C and then 4 washes with PBS for 10 min at room temperature (RT).
   About 10% of the PBS-beads mixture was boiled for 10 min to release the
   proteins and the supernatant proteins were analyzed using WES. After
   removing the supernatant from the remaining mixture, beads were kept at
   −80 °C until further mass spectrometry analysis in the Proteomics Core
   at the Sanford-Burnham Prebys Medical Discovery Institute.

   The total sample peptide amount was evaluated by Pierce Quantitative
   Colorimetric Peptide Assay, 500 Assays (Thermo Fisher Scientific,
   Waltham, MA, USA) before loaded into LC–MS. To reconstitute dried
   samples, 2% acetonitrile and 0.1% formic acid were used and
   reconstituted samples were analyzed by LC–MS/MS using a Proxeon EASY
   nanoLC system (Thermo Fisher Scientific, Waltham, MA, USA) coupled to
   an Orbitrap Elite mass spectrometer (Thermo Fisher Scientific, Waltham,
   MA, USA). An analytical C18 Acclaim PepMap column (75 µm × 500 mm, 2 µm
   particles; Thermo Fisher Scientific) was used to separate peptides at a
   rate of 300 nL min^−1 for over 121 min: 1–6% B in 1 min, 6–23% B in 56
   min, 23–34% B in 37 min, 34–48% B in 26 min, and 48–98% B in 1 min (A =
   FA 0.1%; B = 80% ACN: 0.1% FA).

   Operation of the mass spectrometer was set up in data-dependent
   acquisition mode. Spectra of the first spectrometer (MS1) were measured
   by Orbitrap analyzer with a resolution of 60,000 (automatic gain
   control (AGC) target of 1e6 and a mass range from 350 to 1450 m/z).
   Collision-induced dissociation method was used to generate fragments
   for the second spectrometer (MS2) and data was acquired by the linear
   ion trap (AGC target of 1e4, isolation window of 2 m/z, and a
   normalized collision energy of 35). The duration of dynamic exclusion
   was enabled with 30 s.

   MaxQuant software (v1.5.5.1) was used to process MS raw files with most
   of the default settings, and the Andromeda search engine was used to
   search against the human UniProt database and GMP cRAP sequences (the
   common contaminants database). Variable modifications were set to
   methionine oxidation and N-terminal acetylation, while fixed
   modification was set to carbamidomethylation. The minimum ratio count
   was set to two for label-free protein quantification, and peptides for
   quantification were set to unique and razor. Proteins and peptides were
   selected based on the false discovery rate (FDR) set to 0.01, with a
   minimum length of seven amino acids.

   We analyzed the label free quantitation (LFQ) intensity and defined the
   biotinylated differential proteins (menin associated proteins, MAPs) as
   ≥3-fold in doxycycline^+/biotin^+ cells compared to the
   doxycycline^+/biotin^− cells’ counterpart. The heat map was generated
   using log(LFQ intensity+1). The network analysis was performed by R
   package igraph and the distance was defined as the ratio between the
   LFQ intensity of each MAPs and menin.

4.4. WES and Immunoprecipitation Assays

   Antibodies were obtained commercially and the dilutions were used as
   the following: menin (Bethyl Laboratories, A300-105A; 1:100), KMT2A
   (Bethyl Laboratories, A300-374A; 1:10), GATA3 (Cell Signaling
   Technology, 5852S; 1:25), WAPL (Cell Signaling Technology, 77428S;
   1:50), MED12 (Cell Signaling Technology, 4529S; 1:25), LaminA/C (Cell
   Signaling Technology, 2023S; 1:50), GAPDH (Cell Signaling Technology,
   2118S; 1:200), and β-Actin (R&D System, MAB8929; 1:100). For WES, 0.5
   μg/μL of protein lysate was used for one reaction. Menin, KMT2A, MED12,
   WAPL, GATA3, LaminA/C, and GAPDH were detected by the Anti-Rabbit HRP
   Detection module (ProteinSimple, San Jose, CA, USA) or by 1×
   Anti-Rabbit HRP conjugate (ProteinSimple, San Jose, CA, USA) in the
   Anti-Mouse HRP Detection module (ProteinSimple, San Jose, CA, USA) when
   combined with β-Actin. β-Actin was detected by the Anti-Mouse HRP
   Detection module (ProteinSimple, San Jose, CA, USA). For detection of
   biotin labeled proteins, total lysates were visualized by the
   streptavidin-HRP and chemiluminescent substrate (ProteinSimple, San
   Jose, CA, USA). The WES assays were run with Jess/Wes 12–230 kDa
   Pre-filled Plates or Jess/Wes 66–440 kDa Pre-filled Plates
   (ProteinSimple, San Jose, CA, USA) dependent on the molecular weights
   and analyzed by Compass for SW software (ProteinSimple, San Jose, CA,
   USA). The original WES images can be found at [192]Figure S8.

   For immunoprecipitation, T47D and MCF-7 cells were treated with DMSO or
   1 μM of MI-503 for 3 days. The nuclear lysates were incubated with the
   menin antibody (Bethyl Laboratories, A300-115A) or rabbit IgG (kch
   504250, Diagenode) at 4 °C for 18 h, and the antibody was captured by
   Protein A/G magnetic beads (Thermo Fisher Scientific, Waltham, MA, USA)
   at RT for 2 h. Then, the beads were washed with 1 mL of TBST (25 mM
   Tris, 0.15 M NaCl, and 0.05% Tween-20, pH 7.5), twice with 1 mL of NETN
   buffer (50 mM Tris-HCl pH 8, 150 mM NaCl, 1% NP-40, and 1 mM EDTA), and
   finally with 1 mL of ddH[2]O. For elution, the beads were resuspended
   with 30 μL of RIPA buffer containing 2% SDS and protease/phosphatase
   inhibitors at RT for 10 min, and the supernatant was separated from
   beads by a magnetic stand.

4.5. Seahorse Assays

   Cells were seeded on the cell culture microplate (Agilent, Santa Clara,
   CA, USA) with 12,000 cells per well and cultured to reach 90%
   confluent. For the Seahorse mitochondrial stress test and real-time ATP
   rate test, cells were preincubated in pH 7.4 Seahorse DMEM media
   (Agilent, Santa Clara, CA, USA) containing 10 mM glucose, 2 mM
   glutamine, and 1 mM pyruvate in a non-CO[2] incubator at 37 °C for 1 h.
   For the Seahorse glycolysis stress test, cells were preincubated in pH
   7.4 Seahorse DMEM media containing 1 mM glutamine in an incubator with
   the same settings for 1 h. For Seahorse assays after circulation, cells
   were resuspended in either Seahorse mitochondrial stress or Seahorse
   glycolysis stress media. Thereafter 1500 cells were seeded per well on
   the cell culture microplate (Agilent, Santa Clara, CA, USA) and was set
   at RT for 20 min followed by incubation at 37 °C in a non-CO[2]
   incubator for 20 min. The sensor cartridge (Agilent, Santa Clara, CA,
   USA) was rehydrated 1 day before the assay and the inhibitors were
   serially added just before the assay. The final concentrations of
   chemicals used were as the following: for the mitochondrial stress
   test, pore A: 1 μM oligomycin; pore B: 1 μM FCCP; and pore C: rotenone
   and antimycin A, 0.5 μM each. For the glycolysis stress test, pore A:
   10 mM glucose; pore B: 1 μM oligomycin; and pore C: 5 mM 2-DG. For the
   real-time ATP rate test, pore A: 1.5 μM oligomycin and pore B: rotenone
   and antimycin A, 0.5 μM each. The assays were performed by a Seahorse
   XFe96 Analyzer (Agilent, Santa Clara, CA, USA) and analyzed by Seahorse
   XFe96 Analyzer Wave 2.4 (Agilent) software. For normalization, Hoechst
   33,342 (Thermo Fisher Scientific, Waltham, MA, USA) was combined to add
   in the last injection at the final concentration of 20 μM to stain the
   cell nucleus, and the images were taken by Cytation1 (BioTek, Winooski,
   VT, USA) with Agilent Seahorse XF Imaging and Cell Counting software.
   For validating the ATP production measurement by Seahorse XFe96
   analyzer, a comparative quantitation was performed by the CellTiter-Glo
   Luminescent Cell Viability Assay Kit ([193]Figure S7).

4.6. Circulating Tumor Cells (CTCs) Isolation and Culture

   Blood samples were collected from advanced breast cancer patients
   according to the IRB protocol approved by the University of Texas
   Health Science Center at San Antonio (CTRC# 07-32). About 15 mL of
   Ficoll (GE Healthcare Life Sciences) was placed in a 50 mL conical
   tube, and 8 mL of the blood sample was carefully added on the top of
   the Ficoll. CTCs remained in the interphase peripheral blood
   mononuclear cells (PBMCs) layer after centrifugation at 400× g at RT
   for 30 min. The PBMC layer was carefully moved into a 15 mL conical
   tube and mixed with 10 mL of the MojoSort buffer (Biolegend, San Diego,
   CA, USA). After centrifugation again at 200× g for 5 min at 4 °C, CTCs
   were cleaned by the MojoSort buffer. Then, CTCs were cultured and
   expanded in the PRIME-XV tumorsphere medium (Fujifilm Irvine
   Scientific, Santa Clara, CA, USA) for 1 week for enrichment. For
   collection, cells were dispersed and washed by the MojoSort buffer and
   contaminated blood cell depletion was carried out using 10 μL of Human
   CD45-nanobeads (Biolegend, San Diego, CA, USA). CD45^− CTCs were
   harvested in the supernatant by a magnetic stand after a few washes and
   resuspensions with the MojoSort buffer.

4.7. Immunofluorescence Analysis

   CTCs were fixed with 4% paraformaldehyde (Santa Cruz Biotechnology) for
   10 min and permeabilized with PBS containing 0.5% Triton X-100 for 5
   min at RT. Nonspecific binding was blocked with 5% goat serum in PBS
   containing 0.1% Triton X-100 for 30 min at RT. Immunostaining was
   performed at 4 °C overnight with primary antibodies, followed by
   fluorescence-conjugated secondary antibodies at a concentration of 5
   μg/mL (Thermo Fisher Scientific). Images were captured with the Zeiss
   LSM710 confocal microscope equipped with 64× NA 1.4 oil DIC (Carl
   Zeiss). Antibodies used for staining were as follows: rabbit anti-ALDH1
   (bs-10162R, Bioss, 1:200), mouse anti-pan-CK (4545, Cell Signaling
   Technology, 1:3000), rabbit anti-EpCAM (ab71916, Abcam, 1:200), rabbit
   anti-NANOG (ab21624, Abcam, 1:800), mouse anti-OCT4 (ab184665, Abcam,
   1:500), rat anti-CD44 (ab40983, Abcam, 1:400), mouse anti-CD45
   (ab33533, Abcam, 1:200), goat anti-rabbit (A-11037), goat anti-rabbit
   (A-110340), and goat anti-mouse (A-11032).

4.8. Single-Cell qRT-PCR

   Single CD45^−/Calcein AM^+ CTCs were retrieved manually and lysed in 4
   μL of CellsDriect 2× reaction mix (Invitrogen) and the target genes
   were preamplified as described previously with modifications (26). The
   preamplification reaction contained 4 μL of single cell lysate, 1.8 μL
   of DNA suspension buffer (Teknova), 0.2 μL of SuperScript III
   RT/Platinum Taq mix (Invitrogen), and 1 μL of the premixed outer primer
   sets (F1 and R1, sequences were listed in [194]Table S3, 500 nM per
   primer). The thermal cycle was 50 °C, 15 min; 95 °C, 2 min; 95 °C, 15
   s; and 60 °C, 4 min, and amplicons were amplified for 20 cycles.
   Non-amplified DNA was removed by Exonuclease I (New England BioLabs) at
   37 °C for 30 min. Real-time PCR reactions were run using BioMark^TM HD
   System (Fluidigm) with the semi-nested primer pair ([195]Table S3, F1
   and R2) and analyzed by Real-Time PCR Analysis Software (Fluidigm). The
   relative expression was normalized to the ACTB level to obtain ΔCt, and
   the log(−2^ΔCt) was used as the relative gene expression for the t-SNE
   analysis and violin plots with R package ggplot2.

4.9. In Vitro Flow-Based Assay

   Cell circulation was performed using an ibidi pump system (ibidi) with
   μ-slide I Luer 0.6 mm (ibidi). About one million cells in 10 mL of the
   medium were loaded in a perfusion set held on a fluidic unit. The
   fluidic unit was incubated in a humidified incubator aired with CO[2]
   at 37 °C, and the flow rate was 16.67 mL/min with 5 s switching time
   controlled by PumpControl Software. After circulation, cells were
   collected and centrifuged at 500× g for 5 min. The cell pellet was
   suspended and washed in PBS before cell viability assays and Seahorse
   assays.

4.10. Cell Viability Assay

   Cell viability assay was performed by the CellTiter-Glo Luminescent
   Cell Viability Assay Kit (Promega). For MI-503 treatment, T47D or MCF-7
   cells were seeded in 96-well plate with 200 cells per well and treated
   with DMSO or MI-503 for 3 days. For circulated cells, 2000 cells after
   circulation were seeded in 50 μL of medium in 96-well plates. The assay
   reagent mixture was directly added into the cell medium at a 1:1 ratio
   for 15 min to lyse the cells, and the luminescent signal was detected
   using the Thermo Luminoskan Ascent Microplate Reader (Thermo Fisher
   Scientific).

4.11. In Silico Analysis

   RNA-seq data of primary tumors and normal controls in The Cancer Genome
   Atlas (TCGA) breast cancer cohort were downloaded from Broad Genome
   Data Analysis Center (GDAC) Firehose. Expression levels of MEN1, MAPs,
   and glycolytic and OXPHOS genes were extracted, and Pearson correlation
   coefficients of the expression of individual MEN1/MAP genes and
   individual OXPHOS/glycolytic genes were calculated and plotted as
   heatmaps with R package heatmap.2. Mean expression levels of OXPHOS and
   glycolytic genes were plotted against the expression of MEN1 and the 4
   MAP genes (KMT2A, MED12, WAPL, and GATA3). Violin plots, dot plots, and
   scatter plots were produced with R package ggplot2. A Kaplan–Meier
   survival analysis was performed using R package survminer on
   ERα-positive and ERα-negative patients, separately, of the METABRIC
   dataset downloaded from the cBioPortal. High expression levels of MEN1
   were defined as top 20%, and low expression levels bottom 20%.
   Segmented copy number values > 0 was considered as amplification, while
   <0 as deletion. ChIP-seq data was obtained from Gene Expression Omnibus
   database (GEO number [196]GSE85317) [[197]10] and used EaSeq to analyze
   and generate peaks.

4.12. Data Availability

   The RNA-seq data in this study is available through the Gene Expression
   Omnibus (GEO) under accession number [198]GSE145955.

5. Conclusions

   In conclusion, we demonstrated that menin has a previously
   uncharacterized role in metabolic energy arbitration, in addition to
   modulating DNA replication and repair and cellular transformation
   [[199]45,[200]46]. The mode of menin-regulated transcription can
   further be adjusted by a group of MAPs. We propose that MAPs are
   transcriptional sensors to determine a preference between glycolysis
   and OXPHOS. Upstream metabolic signaling modulators (e.g., MYC, PGC-1α,
   AMPK, HIF-1α, mTOR, and RAS) [[201]8,[202]30,[203]32,[204]47,[205]48]
   may synergize with menin and MAPs for transcription modulation. To fuel
   energy for proliferation and differentiation, this coordinate
   regulation is essential for cells to assess available sources (i.e.,
   oxygen, glucose, or free amino acid) for ATP generation.

Acknowledgments