PD-Seq Identifies Candidate Cellular Targets for Apigenin.

For the comprehensive identification of apigenin cellular targets, we coupled apigenin to amino polyethyleneglycol-polyacrylamide copolymer beads (PEGA beads) after activation with 4-nitrophenyl bromoacetate (Fig. 1A). This method resulted in apigenin being coupled to the beads through either one of two –OH groups in the A-ring or through the single –OH group in the B-ring (Fig. 1A and Fig. S1A), exposing different faces of this flavone to proteins. The apigenin-loaded beads (A-beads) or the unloaded control beads (C-beads) were used to screen, in parallel, a commercially available human breast tumor phage display cDNA library (Fig. 1B). Three rounds of selection with C- and A-beads were performed in parallel, and phage DNA from each of the fractions was collected (Fig. 1C). Indexed libraries for sequencing by Illumina GAII were generated by amplifying inserts with primers in the phage arms (Table S1) and attaching indexed adapters to allow multiplexed sequencing (Fig. 1D).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 1.

Synthesis of apigenin beads and PD-Seq strategy outline. (A) Scheme for the chemical synthesis of A- (orange) and C-beads (blue). The coupling of apigenin to the beads occurred at the end of a polyethylene glycol linker (PEGA beads). Depending on the apigenin –OH group participating in the coupling to the phenyl bromoacetate group, A-beads consist of a combination of three products. (B) Schematic representation of the biopanning steps in the screening of a phage display cDNA library generated from human breast tumor cells mRNA. Three rounds of biopanning (3×), each including binding to the beads, washing, elution, and amplification, were performed in parallel using A- or C-beads. (C) Schematic representation of the fractions used to make the libraries for Illumina GAII sequencing. The preclearing and washing steps were skipped here for simplicity. The original library was an aliquot of a single amplified library purchased from Novagen. Original library (Ori-lib) and input and elution fractions (referred as E) obtained from the first and second rounds of biopanning using A- and C-beads (A-E1 and A-E2 and C-E1 and C-E2, respectively) were used to generate libraries for sequencing. (D) Schematic representation of Illumina GAII libraries preparation. PCR primers (indicated by arrows) at the cDNA insert and vector boundaries were used to amplify the cDNA-containing region, and they were subsequently ligated to Illumina adapters (gray areas) and indexed sequences (red area). (E) Heat map of nICPGs for each biopanning step was generated based on hierarchical clustering (Fig. S2). Cluster I, shown here, consisting of 160 genes is significantly enriched in the A-E2 fraction.

From a total of ∼8.3 million indexed 35-bp-long reads corresponding to the seven different libraries obtained (original library, Input1, C-E1, A-E1, Input2, C-E2, and A-E2) (Fig. 1C), 64% aligned with ORFs in the human genome (Table S2). To discover putative apigenin targets, the following criteria to the identified sequences were applied. (i) Inserts had to be in frame with the phage capsid, (ii) sequences had to be enriched in A- compared with C-beads (Fig. S2 A and B), and (iii) reads had to match one or multiple sequences in the human genome. In those cases where reads aligned to multiple coding sequences, weighted counts were used to obtain normalized in frame-aligned counts per gene models (nICPGs) (Fig. S1B). From a minimum of 15,568 genes represented in the original phage display cDNA library, which was established from combining sequence information from the original and selected fractions, 160 genes meeting these three criteria were identified (Fig. 1E), considered here the candidate apigenin targets (Table 1).

Simultaneously to PD-Seq, we characterized clones recovered after the third round of panning by the traditional phage display approach by performing bacterial infections and individual plaque analysis (Fig. S3 A and B). As is usually the case (25, 26), this approach yielded a very small number of clones. We analyzed a total of 16 plaques by PCR. Three of them showed a small fragment, which on sequencing, proved to be identical to each other. They corresponded to what we have named the MKET clone, which we found to be highly enriched in all fractions; however, it may possibly be an artifact of library preparation (Fig. S3 B–F). The remaining 13 clones had similarly sized inserts, and sequencing all of them determined that they corresponded to the C-terminal region (residues 264–341) (Fig. S3C) of hnRNPA2, also identified as the top candidate by PD-Seq (Table 1).

HnRNPA2 and its splice variant hnRNPB1 play fundamental roles in the progression of tumorigenesis by regulating splicing, mRNA stability, and mRNA transport (27). Consistent with its presence in the phage display library used here, hnRNPA2/B1 higher expression has been reported in several human cancers, including breast (28), and hnRNPA2 expression is recognized as a marker of glioblastoma and lung cancer (29⇓–31).

The hnRNPA2 C-terminal region (hnRNPA2^C) contains three of six YGGG repeats that characterize the protein–protein interaction glycine-rich domain (hnRNPA2^GRD) (Fig. 2B) (32). As a first step to determine the specificity of hnRNPA2^C for different flavonoids, we performed A-bead pull-down assays of hnRNPA2^C phage (ϕ-hnRNPA2^C) suspensions in the presence of apigenin or naringenin. Apigenin, but not naringenin, significantly competed the binding of ϕ-hnRNPA2^C to A-beads (Fig. S4A).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 2.

hnRNPA2 directly binds apigenin through the GRD. (A) Lysates from HeLa cells expressing full-length hnRNPA2-GFP or GFP alone were pulled down with A- or C-beads. Pull-down assays were resolved by SDS/PAGE and analyzed by Western blot using anti-GFP antibodies. (B) Schematic representation of GST-hnRNPA2 clones used in the pull-down assays with C- or A-beads. RRM, RNA recognition motifs of hnRNPA2. hnRNPA2^C corresponds to the C-terminal 78-aa fragment present in ϕ-hnRNPA2^C identified by conventional phage display screening. hnRNPA2^∆C corresponds to a clone in which this C-terminal 78-aa fragment was deleted. hnRNPA2^∆GRD corresponds to a clone in which the GRD domain was deleted. GST-hnRNPA2^∆C corresponds to hnRNPA2 lacking the C-terminal region. All forms of hnRNPA2 were GST-tagged, E. coli-expressed, and affinity-purified forms. (C) Different versions of recombinant affinity-purified GST-hnRNPA2 proteins were pulled down with A- or C-beads (indicated as A or C, respectively). Pull-down assays (bound) and supernatants fractions were resolved by SDS/PAGE and analyzed by Western blot using anti-GST antibodies. Arrows indicate the correctly sized products; smaller bands present in some of the lanes correspond to degradation products.

The identification of just hnRNPA2 by conventional phage display screening underscores one of its main limitations, namely the retrieval of very few candidates, likely because many bona fide targets fail to be properly amplified or selected through multiple biopanning rounds (33). The results here presented show that PD-Seq overcomes this shortcoming, providing an opportunity for the high-throughput identification of small-molecule targets.

Apigenin Targets Are Enriched in Three Main Categories.

A comparison of the sequence of 160 putative apigenin targets failed to reveal any obvious common protein domains or stretches of conserved amino acids. However, analyses based on known functional annotations and gene ontology (GO) showed that three main categories (GTPase activation, membrane transport, and mRNA metabolism/alternative splicing) are highly significantly (P < 0.01) overrepresented among the identified 160 apigenin targets (Table 1 and Fig. S2). The GTPase activation functional category contains proteins such as Rho-guanine nucleotide exchange factor 1 (ARHGEF1) involved in the activation of Rho-GTPases (34), a family of proteins regulating cell polarity and cell migration (processes previously shown to be modulated by apigenin) (35, 36). The membrane transport category includes sodium, zinc, and calcium ion transporters as well as mitochondrial membrane transport proteins, consistent with the mitochondrial envelope being a major apigenin accumulation site in sensitive cells (16). The third very significantly enriched category corresponds to factors involved in alternative splicing and mRNA metabolism, with hnRNPA2 and up-frame shift suppressor 3 homolog B (UPF3B) showing the highest fold enrichment from all identified apigenin targets (Table 1). Together, these three GO functional classes correspond to 121 of 160 candidate apigenin targets identified.

To evaluate the reliability of PD-Seq for the identification of small-molecule binding proteins, we combined three different validation approaches. The first approach consisted of searching the literature for described flavonoid binding activities for these proteins. Four putative apigenin targets [hnRNPA2, UDP-glucose dehydrogenase (UGDH), lactase (LCT), and Mucin 1 (MUC1)] (Table 1) had been previously shown to interact with different flavonoids (37⇓⇓–40). HnRNPA2/B1 was recently identified as a proanthocyanidin-interacting protein responsible for the antiviral activity of blueberry leaf extracts against hepatitis C virus (37). UGDH activity is inhibited by the flavonol quercetin, resulting in decreased proliferation of breast cancer cells (38). LCT binds and hydrolyzes a range of flavonol and isoflavone glycosides (39). Apigenin interacts with the cytoplasmic domain of MUC1, inhibiting its dimerization in breast cancer cells (40).

The second approach consisted of testing identified candidate apigenin targets that displayed enzymatic activity for apigenin effects. As control, we used naringenin, which displays none of the biological activities of apigenin (16). Two candidate apigenin targets were tested in human MDA-MB-231 breast cancer cells treated with 50 μM apigenin, naringenin, or DMSO diluent control. Isocitrate dehydrogenases 3 (IDH3) corresponds to one of three human isocitrate dehydrogenases, which catalyses the NAD⁺-dependent oxidative decarboxylation of isocitrate to α-ketoglutarate (41), whereas UGDH catalyzes the NAD⁺-dependent oxidation of UDP-glucose to UDP-glucoronate (42). In enzymatic assays conducted to investigate the effects of apigenin on these enzymes, IDH3 activity was inhibited by more than 60% in isolated mitochondria from MDA-MB-231 breast cancer cells treated with apigenin, whereas naringenin had no effect (Fig. S4B). Similarly, apigenin, but not naringenin, significantly (P < 0.05) decreased the UGDH enzymatic activity in breast cancer cell extracts (Fig. S4C). Our results show that the inhibition of UGDH by flavonols can be extended to flavones.

The third approach consisted of evaluating the interaction of the identified targets expressed as GFP-tagged with apigenin in human epithelial HeLa cells. Pull-down assays of lysates from HeLa cells expressing hnRNPA2-GFP with either A- or C-beads and immunoblotted with anti-GFP showed that hnRNPA2-GFP specifically binds the A- but not C-beads (Fig. 2A, line 2). Similar results were obtained for ARHGEF1-GFP and GFP–BCL2-associated athanogene (BAG1), an antiapoptotic protein that enhances the activity of B-cell lymphoma 2 (BCL2) (43) (Fig. S4 D, line 2 and E, line 2).

Collectively, our validation results underscore the value of PD-Seq for the identification of apigenin targets. The finding that most of the apigenin targets correspond to one of three main categories (GTPase activation, membrane transport, and mRNA metabolism/alternative splicing) has several interesting implications. First, despite being structurally similar, apigenin, but not naringenin, shows specific biochemical effects (e.g., inhibiting enzymatic activity, which is shown in Fig. S4 B and C). Second, kinases and other proteins with ATP binding proteins have been generally believed to be main targets for flavonoids (44, 45). Interestingly, our screen identified only five kinases [aarF domain containing kinase 1 (ADCK1), ephrin type-A receptor 5 (EPHA5), microtubule associated serine/threonine kinase 1 (MAST1), phosphofructokinase platelet (PFKP), and unc-51-like kinase 4 (ULK4)] (Table 1) as candidate apigenin targets, despite more than 560 kinases being represented in the library. Third, it is tempting to speculate that the three main classes of apigenin targets reflect different temporal effects of this compound, because it is well-documented that dietary components frequently elicit short- and long-term biological responses (46). If so, these targets may reveal cellular functions associated with the use of flavonoids for both preventive (long-term) and therapeutic (short-term) interventions.

Specific and High-Affinity Interaction of Apigenin and Other Flavonoids with the Dimerization Region of hnRNPA2.

The identification of hnRNPA2 as an apigenin target by both PD-Seq and conventional biopanning and the importance of this protein in cancer make the apigenin–hnRNPA2 interaction an ideal candidate for additional characterization. To determine the regions in hnRNPA2 that bind apigenin, we generated fragments that represented the full-length protein, the protein lacking the last 78 aa, the protein lacking the GRD, and just the last 78 aa fused to GST (hnRNPA2, hnRNPA2^∆C, hnRNPA2^∆GRD, and hnRNPA2^C, respectively) (Fig. 2B). Each of these proteins was then tested for its ability to interact with A- or C-beads in pull-down experiments, which were analyzed by Western blot using antibodies against GST (Fig. 2C). The interaction of hnRNPA2 with apigenin likely involves multiple sites in the GRD, because GST-hnRNPA2^∆C still binds to A-beads (Fig. 2C). Showing that the interaction with apigenin requires the GRD, the deletion of this domain (hnRNPA2^∆GRD) abolished A-bead binding (Fig. 2C). These results confirm the interaction of apigenin with hnRNPA2 and identify the GRD as the hnRNPA2 protein domain with flavones binding capacity.

To determine the affinity and specificity of the interaction between flavonoids and hnRNPA2, two strategies were pursued. First, we took advantage of the ability of apigenin and other flavones to absorb light at 310 and 370 nm (Fig. 3A). The light absorption of apigenin increases when incubated with GST-hnRNPA2 but not GST alone (Fig. 3A). Similar changes in absorption of flavonoids have been reported as a consequence of their interactions with proteins (47⇓–49). The increased absorption at 370 nm was used to determine the apparent dissociation constant (K_D) of GST-hnRNPA2 for apigenin, which was estimated at 2.66 ± 1.09 μM (Fig. 3 B and C). Comparable results were obtained if we used the absorption change at 310 nm. The strength of the interaction of hnRNPA2 for apigenin is, therefore, comparable, if not higher, than described for other protein–flavonoid interactions. For example, the K_D of quercetin for collagen was determined to be at ∼12 μM (48).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 3.

Binding affinity of the interaction of hnRNPA2 with apigenin. (A) Apigenin (10 μM) was titrated with increasing concentrations of purified GST-hnRNPA2 (0, 0.2, 0.5, 1, 2, 5, 10, and 20 μM) or GST (Inset). Changes in absorption across the UV-visible spectrum were determined over the 250- to 450-nm range. (B) Changes in absorbance of apigenin at 370 nm as determined in A. (C) Dissociation constant (K_D) of the apigenin–hnRNPA2 complex calculated using the Benesi–Hilderbrand method as described in Materials and Methods. Data represent the mean ± SEM (n = 3). (D) hnRNPA2^C (Fig. 2B) was cloned into the pFLIP2 vector in frame between the regions coding for the N-terminal CFP and C-terminal YFP. The affinity-purified FLIP2-3-hnRNPA2^C protein was incubated with increasing concentrations of apigenin (0, 1, 5, 10, 25, 50, and 100 μM) for 3 h at 37 °C. Relative fluorescence units (RFUs) were determined by spectrofluorometry (λ_ext = 405 nm; λ_emi = 460–600 nm) and represented as emission spectra. The absence of an isosbestic point in the nanosensor spectra is likely a consequence of flavonoids, particularly flavones, absorbing light at a wavelength that partially overlaps with the CFP excitation spectrum (Fig. S6). YFP fluorescence is not affected by the various flavonoids over the broad concentration range tested (Fig. S6). (E) The calculated YFP/CFP fluorescent ratios (530/480 nm) are represented over the 0- to 100-μM concentration range for each flavonoid. (F) Flavonoid-dependent changes in YFP/CFP ratios were transformed into saturation curves as described in Materials and Methods. K_D value was determined by using nonlinear regression. Data represent the mean ± SEM (n = 3).

The second strategy consisted of developing a genetically encoded flavonoid nanosensor based on FRET. To this objective, we cloned hnRNPA2^C (Fig. 3D) in a collection of vectors for expression as translational fusions between fluorescent proteins with distinct excitation/emission spectra (50). Of eight constructs tested, FRET was observed in four (Figs. S5 and S6), but in only one instance, p-fluorescent indicator protein 2-3-hnRNPA2^C (pFLIP2-3-hnRNPA2^C) (Fig. 3D), was the energy transfer increased by apigenin in a concentration-dependent manner (Fig. 3E). This construct was, therefore, used to estimate the apparent K_D of apigenin for this fragment, which was determined to be 22.99 ± 7.70 μM (Fig. 3F).

The lower affinity (higher apparent K_D) determined by FRET most likely reflects the differences between the two K_D determination methods and the absence of additional apigenin binding sites in hnRNPA2^C (Fig. 2 B and C). However, one advantage of the FRET-based method over the spectrophotometric approach is that it permits us to investigate the specificity and relative affinity of the interaction between hnRNPA2 and other flavonoids. Luteolin differs from apigenin by an additional –OH in ring B and also interacts with hnRNPA2, although with lower affinity (Table 2 and Fig. S7A). However, a methyl group in the additional B-ring –OH, which is present in chrysoeriol, a common flavone in many medicinal plants, results in the complete inhibition of the interaction (Table 2 and Fig. S7A). Apigenin is often found in the diet as C- or O-glucosides (51). Revealing the important role of the 7-O group in the interaction with hnRNPA2, apigenin 7-O-glucoside shows little, if any, biological activity (52). In contrast, highlighting the importance of the 7-O group itself in the interaction with hnRNPA2 rather than a steric hindrance by the glucosyl group, apigenin 6-C-glucoside (isovitexin) significantly binds to hnRNPA2 with lower affinity than aglycone apigenin but higher affinity than luteolin (Table 2 and Fig. S7B). Consistent with ϕ-hnRNPA2^C recognizing apigenin but not naringenin (Table 2 and Fig. S4A), neither naringenin nor the related flavanone eriodictyol affect the FRET of the nanosensor (Table 2 and Fig. S7D). Suggesting that the binding to hnRNPA2 is not limited to flavones and consistent with the fact that many biological activities of flavones being shared by flavonols (5), quercetin and kaempferol also show a significant interaction with hnRNPA2 but with lower affinity than apigenin (Table 2 and Fig. S7E). No binding of hnRNPA2 to flavopiridol or the isoflavone genistein was detected, suggesting that these biologically active compounds function through mechanisms distinct from apigenin (Fig. S7 C and F).

Apigenin Inhibits hnRNPA2 Multimerization and Splicing Activity.

In the cell, hnRNPA2 forms oligomers through the GRD, and multimerization is required for mRNA binding by the RNA recognition motif (RRM) domain (53⇓–55). To determine if the binding of apigenin to the GRD affects hnRNPA2 dimerization, the amplified luminescent proximity homogenous assay (ALPHA) was used. This assay recapitulates the reported hnRNPA2 dimerization, which is reflected in the ability of 6xHis-hnRNPA2 to form heterodimers with GST-hnRNPA2 but not GST (Fig. 4A). The addition of apigenin (100 μM) to the GST-hnRNPA2/6xHis-hnRNPA2 heterodimers resulted in luminescence being reduced to less than one-half (Fig. 4B, Api), whereas no decrease in luminescence was observed with naringenin (100 μM) (Fig. 4B, Nar). These results show that binding of apigenin to the GRD inhibits hnRNPA2 dimerization.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 4.

Apigenin affects hnRNPA2 dimerization. (A) Purified 6xHis-hnRNPA2 (125 nM) protein was incubated with 125 nM native GST-hnRNPA2, 125 nM boiled GST-hnRNPA2, or 125 nM GST for 1 h at room temperature followed by the addition of GSH-acceptor and anti–His-donor beads for 6 h. (B) GST-hnRNPA2 (125 nM) was incubated with 125 nM 6xHis-hnRNPA2 for 1 h at room temperature followed by the addition of GSH-acceptor and anti–His-donor beads for 6 h at room temperature. Apigenin (Api; 100 μM), naringenin (Nar; 100 μM), or diluent control (DMSO) was added for 15 min at room temperature. Data represent the mean ± SEM (n = 4). *P < 0.05.

To determine the biological consequences of the hnRNPA2–apigenin interaction, the effect of apigenin on alternative splicing of known hnRNPA2 substrates (29) was evaluated. The splice isoforms of c-FLIP (c-FLIP_L and c-FLIP_S) (Fig. 5A) and caspase-9 (caspase-9a and -9b) (Fig. 5B) were compared in two human cell lines with contrasting hnRNPA2 expression. MDA-MB-231 breast cancer cells express high levels of hnRNPA2 (Fig. S8), whereas the MCF-10A immortalized noncarcinogenic breast epithelial cells have very low levels of hnRNPA2 mRNA and protein (Fig. S8). In MDA-MB-231 but not MCF-10A cells, apigenin reduced the levels of c-FLIP_S (Fig. 5 A and D) and caspase-9b (Fig. 5 B and E) without affecting the splicing of baculoviral inhibitor of apoptosis repeat-containing 5 (BIRC5) (Fig. 5 C and F), an alternatively spliced non-hnRNPA2 substrate (56). In agreement with its ability to also bind hnRNPA2 (Table 2 and Fig. S7A), the flavone luteolin had a similar effect on splicing as apigenin (Fig. 5). In contrast, naringenin, a non-hnRNPA2 binding flavonoid, showed no effect on splicing (Fig. 5). Taken together, these results show that apigenin (and likely, other flavones) interacts with hnRNPA2, inhibiting its dimerization and altering the alternative splicing patterns of hnRNPA2 substrates.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 5.

Apigenin regulates alternative splicing of hnRNPA2 substrates in breast cancer cells. (A–C) MDA-MB-231 breast cancer cells were treated with 50 μM apigenin (Api), luteolin (Lut), or naringenin (Nar) or diluent DMSO for 48 h. Total RNA was isolated, and alternative splicing was analyzed by RT-PCR using specific primers for (A) c-FLIP, (B) caspase-9, and (C) BIRC5. GAPDH expression was used as the loading control. Reactions were resolved in 2% (wt/vol) agarose gels. Positions of the primers (arrows) and splicing variants (boxes) are represented schematically on the right. (D–F) Graphs represent the percent of the indicated splice isoform. Data represent the mean ± SEM (n = 3). *P < 0.05 compared with DMSO control samples.

Increased expression of hnRNPA2 has been reported in several cancers (28⇓⇓–31). Cancer cells are often resistant to apoptosis (57), which is normally triggered by the activation of signal transduction cascades that culminate in the activation of the caspases. The caspase-9a splice isoform, present in both the malignant and noncarcinogenic breast cancer cells (Fig. 5 B and E), encodes the functional apoptotic caspase-9, which is responsible for inducing cell death. In contrast, the splice isoform caspase-9b (lacking exons 3–6) (Fig. 5B), significantly increased in malignant cells and down-regulated by apigenin, encodes a caspase-9 protein that exhibits a dominant-negative activity and inhibits apoptosis (58). A similar situation happens with c-FLIP; c-FLIP_S, inhibited by apigenin, has an alternate 3′ exon (exon 7 instead of 8) (Fig. 5 A and D), which encodes a c-FLIP protein that completely prevents the activation of specific death receptors (59). Thus, the addition of apigenin to malignant cells reverts the splicing of caspase-9 and c-FLIP, two key regulators of apoptosis, to the splice variants present in noncarcinogenic cells.