CD44 regulates epigenetic plasticity by mediating iron endocytosis

CD44 mediates hyaluronate-dependent iron endocytosis

We first employed nuclear magnetic resonance (NMR) spectroscopy to study interactions between Hyal and iron. Addition of Fe(III) to a low-molecular-mass (LMM) Hyal in water led to line broadening of proton signals of LMM-Hyal, indicating that this organic substrate interacts with iron at neutral pH (Fig. 1a). Acidifying the sample to protonate the carboxylates of LMM-Hyal and to disrupt organometallic complexes released the metal, thereby rescuing the proton signals of free LMM-Hyal (Fig. 1a). These data demonstrate the dynamic and reversible nature of metal coordination by LMM-Hyal in near-physiological conditions of solvent, temperature and pH.

To evaluate the potential of high-molecular-mass (HMM) Hyal-based organometallic complexes to be internalized by CD44, we knocked out (ko) CD44 from mesenchymal human mammary HMLER cells stably expressing high levels of CD44 (HMLER CD44^high) ¹⁹ (Fig. 1b). For comparison, we independently knocked out TFRC, the gene coding for transferrin receptor protein 1 (TfR1) (Fig. 1b), which mediates endocytosis of transferrin (TF) and plays a central role in the regulation of cellular iron homeostasis.

We then evaluated the capacity of parental cells and ko clones to take up fluorescently labeled HMM-Hyal (0.8 MDa). Internalization of Hyal was reduced in CD44 ko and increased in TFRC ko clones compared to parental cells (Fig. 1c), validating the functional role of CD44 in mediating this process. Hyal and TF, which interacts with CD44 and TfR1, respectively, predominantly localized in distinct vesicles in parental cells (Fig. 1d), allowing for a comparative analysis of metal uptake mediated either by CD44 or TfR1. Thus, we evaluated whether endocytic vesicles contained Fe(II) and Hyal, independent of TF. To this end, we used the Fe(II)-specific turn-on fluorescent probe RhoNox-M, which can detect the presence of Fe(II) in lysosomes ²⁰ . Parental HMLER cells exhibited RhoNox-M/Hyal-positive vesicles free of TF as observed by fluorescence microscopy. CD44 ko cells exhibited a lower proportion of RhoNox-M/Hyal-positive vesicles with an increased number of RhoNox-M/TF-positive vesicles compared to parental cells, while TFRC ko cells showed the opposite trend (Fig. 1d). Consistently, complementing CD44 ko cells with CD44 restored a higher proportion of RhoNox-M/Hyal-positive vesicles (Fig. 1e). RhoNox-M/Hyal-positive vesicles were also observed in other cancer cell lines, primary cancer cells including circulating tumor cells (CTC) of lung and colon cancers and in primary human T-cells (Extended Data Fig. 1a,b). Consistently, LNCaP prostate cancer cells, which do not show detectable levels of CD44 protein, exhibited a low proportion of RhoNox-M/Hyal-positive vesicles (Extended Data Fig. 1a).

In line with CD44 mediating iron endocytosis, supplementing cells with HMM-Hyal (0.6-1 MDa) increased levels of cellular iron in parental CD44^high and TFRC ko but not in CD44 ko HMLER and MCF7 cells as defined by the fluorescence of RhoNox-M quantified by flow cytometry (Extended Data Fig. 2a). This was observed across cell types (Extended Data Fig. 2a,b), indicating this to be a general feature of cells expressing CD44 protein. In CD44-expressing MDA-MB-468 breast cancer cells, iron uptake was more pronounced when cells were treated with Hyal of higher molecular mass (Extended Data Fig. 2c). Conversely, treating CD44-expressing cells with hyaluronidase or blocking antibodies against CD44 or TfR1 reduced iron uptake (Extended Data Fig. 2d,e). Additionally, supplementing HMLER CD44^high cells with Hyal increased cellular iron as detected by secondary ion mass spectrometry (SIMS) imaging ²¹ and quantified by inductively coupled plasma mass spectrometry (ICP-MS) (Fig. 1f,g). Supplementing HMLER CD44^high cells with Hyal also led to increased levels of the iron storage protein ferritin in a CD44-dependent manner (Fig. 1h), further demonstrating increased iron uptake in these conditions.

To evaluate the physiological significance of this pathway, we sorted CD44^high/low cell populations from two patient-derived xenografts (PDX) of breast tumors refractory to conventional chemotherapy, prone to form metastasis and associated with poor outcome. We found that CD44^high cells contained higher levels of ferritin with reduced levels of TfR1 compared to the CD44^low cell population (Fig. 1i). Collectively, these data provide evidence that CD44 mediates iron endocytosis in a Hyal-dependent manner.

Prevalence of CD44-mediated iron endocytosis in the mesenchymal state of cells

The mesenchymal state of cells is characterized by elevated levels of CD44 in several lineages ²² . To investigate the cellular changes reliant on CD44 levels, we triggered EMT in triple-negative breast cancer MDA-MB-468, luminal breast cancer MCF7 and HMLER CD44^low cell lines, primary breast cancer cells and primary lung CTC using epidermal growth factor (EGF), oncostatin M (OSM) or transforming growth factor beta 1 (TGF-β). Upon treatment, cells shifted from an epithelial to a mesenchymal phenotype according to cell morphology, reduced levels of the epithelial marker E-cadherin and increased levels of CD44 (Fig. 2a). Consistent with EMT induction, levels of the transcription factor snail increased along with that of the mesenchymal markers vimentin and fibronectin (Fig. 2b). Remarkably, CD44 levels increased together with ferritin, whereas levels of TfR1 decreased at a later time point (Fig. 2b). In line with this, CD44 loading at the plasma membrane was enhanced during EMT, which was in contrast to a reduced loading of TfR1 (Fig. 2c and Extended Data Fig. 3a), and the number of RhoNox-M/Hyal-positive vesicles increased together with cellular iron (Fig. 2d,e and Extended Data Fig. 3b). Importantly, knocking down CD44 antagonized the production of ferritin in cells treated with EGF (Fig. 2f).

Furthermore, RT-qPCR showed that mRNA levels of CD44, and FTH1 coding for ferritin, increased during EMT, whereas that of TFRC decreased (Fig. 2g). This data was in agreement with the well-established negative feedback loop regulating the biosynthesis of TfR1 at the translational level by excess cytosolic iron ^23,24 . It also indicated that in contrast to TFRC, CD44 is not subjected to this regulatory loop. Together, these results advocate for an alternative glycan-mediated iron endocytosis pathway reliant on CD44 that prevails in the mesenchymal state of cells.

EMT is characterized by a redox signature implicating iron

To identify iron-dependent cellular functions required during EMT or for the maintenance of mesenchymal states of cells, we employed a combination of quantitative proteomics and metabolomics. Out of 5,574 proteins that were detected by mass spectrometry, 255 proteins were downregulated, while 276 were upregulated in EGF-treated MDA-MB-468 cells (Fig. 3a, and Supplementary Table 1). Gene ontology (GO) revealed a bias towards proteins with oxidoreductase activity having an increased expression in the mesenchymal state (Extended Data Fig. 4a). The response to EGF was further characterized by changes in EMT markers, proteins linked to metastasis, proteins involved in the regulation of endocytosis and iron homeostasis as well as metabolic enzymes and an iron-dependent demethylase. For example, EMT was characterized by reduced levels of the epithelial proteins BCAM and CADM4, whereas vimentin was upregulated (Fig. 3a). The pro-metastatic protein CD109 was upregulated ²⁵ , which was in agreement with the notion that mesenchymal cancer cells are capable of dissemination. Consistent with increased iron uptake during EMT, levels of ferritin increased together with sorting nexin 9 (SNX9), a protein that regulates endocytosis of CD44, which has also been linked to metastasis ²⁶ . Conversely, levels of TF and TfR1 were reduced, further supporting the prevalent functional role of CD44 in mediating iron endocytosis during EMT. Importantly, we detected an increase of the iron-dependent histone demethylase PHF8, which is involved in active transcription, brain development, cell cycle progression as well as the regulation of EMT in vivo and in breast tumor growth ^27–31 . In contrast, levels of other iron-dependent demethylases that could be detected by mass spectrometry did not change significantly. Western blotting further supported protein changes identified by mass spectrometry (Extended Data Fig. 4b).

Additionally, metabolic enzymes involved in the conversion of glutamate and pyruvate to α-ketoglutarate (αKG), including glutamate dehydrogenase 1 (GLUD1), isocitrate dehydrogenase 2 (IDH2) and the iron-sulfur cluster-containing aconitase 2 (ACO2), were upregulated along with other enzymes of the Krebs cycle (Fig. 3a,b). αKG is a co-substrate required for the iron-catalyzed oxidative demethylation of histone and nucleic acid methyl marks, and is implicated in the maintenance of pluripotency ³² . Consistent with the upregulation of these mitochondrial enzymes, quantitative metabolomics indicated a marked increase of αKG in various cell lines and primary cells undergoing EMT (Fig. 3c,d, Extended Data Fig. 4c and Supplementary Table 2). We then investigated whether reducing iron uptake could potentially alter mitochondrial metabolism in cells treated with EGF. Knocking down CD44 antagonized the upregulation of αKG in cells treated with EGF (Fig. 3e), which was consistent with previous findings showing that iron restriction inhibits the activity of ACO2 ³³ , supporting a role of iron in mitochondria during EMT.

These biochemical changes, including enhanced cellular uptake of iron together with the upregulation of PHF8 and its co-substrate αKG, further hinted toward a specific regulation of EMT occurring at the chromatin level ³⁴ .

Nuclear iron is a rate-limiting regulator of epigenetic plasticity

Metals are ubiquitous in cell biology, acting as protein cofactors. Due to its unique electron configuration, iron distinguishes itself by its capacity to catalyze oxidative demethylation of protein residues, including those of histones, and methylated nucleobases. Thus, iron is a rate-limiting factor of these processes (Fig. 4a). In line with a role of iron in the nucleus of cells undergoing EMT, subcellular fractionation indicated higher levels of ferritin in the nucleus of MDA-MB-468 cells undergoing EMT (Fig. 4b) ³⁵ .

The iron- and αKG-dependent demethylase PHF8 has been shown to promote demethylation of H3K9me2 ³⁶ , a repressive histone mark whose reduction has been observed in solid tumors ^37–39 . Increase of PHF8 in MDA-MB-468 cells undergoing EMT correlated with the preferential reduction of H3K9me2 compared to the other repressive histone marks H3K9me3 and H3K27me3, as defined by quantitative mass spectrometry (Fig. 4c and Supplementary Table 3). Additionally, knocking down PHF8 prevented demethylation of H3K9me2 in MDA-MB-468 cells treated with EGF (Fig. 4d), validating a functional role of this demethylase during EMT.

Next, we examined the genome-wide distribution of H3K9me2 by chromatin immunoprecipitation sequencing (ChIP-seq) to identify PHF8-regulated genes whose expressions are iron-dependent. ChIP-seq revealed large organized heterochromatin K9 modifications, previously termed LOCKs ^37,39 . These data showed a reduction of H3K9me2 ChIP-seq reads count in the body of specific genes in cells treated with EGF, including CD44, CD109, VIM and FN1 coding for vimentin and fibronectin, respectively (Fig. 4e, Extended Data Fig. 5a and Supplementary Table 4), which was further validated by ChIP-qPCR (Extended Data Fig. 5b). These data indicate that CD44 positively regulates its own expression at the transcriptional level by mediating iron endocytosis, unlike TfR1 that is negatively regulated by excess iron at the translational level via iron responsive elements (IRE) ^23,24 . Furthermore, analysis of the transcriptome by RNA-seq indicated that the reduction of H3K9me2 reads count correlated with increased gene expression in agreement with the repressive nature of H3K9me2 (Fig. 4F and Supplementary Table 4). For instance, out of 284 genes where reduction of H3K9me2 was observed, 258 genes were transcriptionally upregulated. In addition to CD44, we identified genes involved in cancer, regulation of EMT, development, immune responses, inflammation, and wound healing, which are biological processes CD44 is linked to (Supplementary Table 5). In comparison, we did not observe a significant reduction of H3K9me2 in genes coding for EMT transcription factors. However, we identified long non-coding (lnc) RNA genes previously reported to be involved in cancer (Supplementary Table 5). In agreement with PHF8 playing a crucial role in regulating these genes through oxidative demethylation of H3K9me2, knocking down this demethylase partly blocked EGF-induced expression of iron-regulated genes, including CD44 (Fig. 4g). Consistently, ChIP-seq indicated that H3K4me3 was enriched at active promoters, which is required for the recruitment and selective activity of PHF8 at these genomic locations (Extended Data Fig. 5c). Furthermore, ChIP-seq also revealed a reduction of the repressive histone mark H3K27me3 in a more restricted set of genes (Extended Data Fig. 5d), which was consistent with increased mRNA levels of the H3K27-specific iron-dependent demethylases KDM6A and KDM6B in cells undergoing EMT as monitored by RNA-seq (Supplementary Table 4). Conversely, we did not observe changes of the heterochromatin mark H3K9me3 (Figures 4c and Extended Data Fig. 5e). While the reduction of H3K27me3 was less pronounced compared to that of H3K9me2 according to mass spectrometry and ChIP-seq reads count (Fig. 4c and Extended Data Fig. 5d), these data support the central role of iron in the regulation of gene expression, reflecting a complex scenario occurring at the chromatin level during EMT, potentially involving other iron-dependent demethylases and histone marks. In line with this, knocking down CD44 partly blocked H3K9me2 demethylation and the expression of genes regulated by PHF8 in MDA-MB-468 cells (Fig. 4h,i). Consistently, CD44^high cells from tumors exhibited higher levels of PHF8 and the iron-regulated gene fibronectin with reduced H3K9me2 compared to the CD44^low population, supporting the physiological relevance of this mechanism (Fig. 4j). Taken together, these data establish a direct functional link between CD44-mediated iron endocytosis and the regulation of gene expression involving iron-catalyzed histone demethylation.

Targeting iron homeostasis interferes with the maintenance of a mesenchymal state of cells

Small molecules provide a means to manipulate cellular processes with valuable spatial and temporal resolution ⁴⁰ . To directly address the implication of the nuclear iron pool, we set out to identify small molecule chelators of iron that could specifically target the nucleus. To this end, we functionalized known iron chelators with alkyne functional groups that can be chemically labeled in cells by means of click chemistry (Fig. 5a) ⁴¹ . Fluorescent labeling of a clickable surrogate of deferoxamine (cDFO) in cells revealed a predominant nuclear localization of this small molecule (Fig. 5a). Thus, deferoxamine (DFO) represents a suitable tool to directly interrogate functional roles of nuclear iron. For instance, DFO prevented the demethylation of H3K9me2 in cancer cell lines and primary cancer cells that were treated either with EGF, OSM or TGF-β (Fig. 5b and Extended Data Fig. 6a). Consistently, DFO antagonized the upregulation of proteins whose genes were found to be regulated by PHF8 in these conditions (Fig. 5c and Extended Data Fig. 6b). These data provide direct evidence of the functional role of nuclear iron in the regulation of the expression of these genes. Moreover, the iron chelator deferasirox (DFX), whose scaffold exhibits a preference for the mitochondrial compartment as defined by the staining of the clickable derivative cDFX (Extended Data Fig. 6c), blocked αKG production in MDA-MB-468 cells treated with EGF (Extended Data Fig. 6d) and interfered with H3K9me2 demethylation and CD44 expression (Extended Data Fig. 6e). Similarly, metformin, which has been shown to target metals in the mitochondria ⁴² and to inhibit the Krebs cycle ⁴³ , antagonized the effect of EGF (Fig. 5d,e). These data, consistent with αKG being required for iron-dependent demethylation of H3K9me2, validate a functional role of mitochondrial iron during EMT and show that targeting mitochondrial metabolism provides another means to control epigenetic plasticity.

Finally, the lysosomal iron-sequestering small molecule ironomycin has previously been shown to effectively kill mesenchymal and tumor-initiating cells in vitro and in vivo due to a higher iron content in these cells ¹⁵ . Consistently, EGF potentiated the cytotoxic effect of ironomycin against MDA-MB-468 cells (Fig. 5f) with increased lipid peroxidation as defined by C11-BODIPY fluorescence that was prevented by co-treatment with liproxtatin-1 (Fig. 5g) ⁴⁴ . This data is in line with the observed increase of iron endocytosis in these conditions, providing a powerful strategy to eradicate tumorigenic cells undergoing EMT through induction of ferroptosis ⁴⁵ . Together, these data support a mechanism whereby iron is a rate-limiting regulator of epigenetic plasticity ⁴⁶ , a general pathway prone to small molecule intervention with topological resolution.