Identification of HMGA2 inhibitors by AlphaScreen-based ultra-high-throughput screening assays

The mammalian high mobility group protein AT-hook 2 (HMGA2) is a multi-functional DNA-binding protein that plays important roles in tumorigenesis and adipogenesis. Previous results showed that HMGA2 is a potential therapeutic target of anticancer and anti-obesity drugs by inhibiting its DNA-binding activities. Here we report the development of a miniaturized, automated AlphaScreen ultra-high-throughput screening assay to identify inhibitors targeting HMGA2-DNA interactions. After screening the LOPAC1280 compound library, we identified several compounds that strongly inhibit HMGA2-DNA interactions including suramin, a century-old, negatively charged antiparasitic drug. Our results show that the inhibition is likely through suramin binding to the “AT-hook” DNA-binding motifs and therefore preventing HMGA2 from binding to the minor groove of AT-rich DNA sequences. Since HMGA1 proteins also carry multiple “AT-hook” DNA-binding motifs, suramin is expected to inhibit HMGA1-DNA interactions as well. Biochemical and biophysical studies show that charge-charge interactions and hydrogen bonding between the suramin sulfonated groups and Arg/Lys residues play critical roles in the binding of suramin to the “AT-hook” DNA-binding motifs. Furthermore, our results suggest that HMGA2 may be one of suramin’s cellular targets.


Results
An automated uHTS assay to identify HMGA2 inhibitors. We previously used a PDI-ELISA assay to screen a small library containing 29 DNA-binding compounds and successfully identified several small molecules that disrupt HMGA2 binding to the minor groove of AT-rich DNA sequences 33 . Although this assay performs well in a 96 well plate format, it is not suitable for automated ultrahigh throughput screening of 100,000s of compounds due to the need for streptavidin-coated assay plates and multiple wash steps. To address these limitations, we established a new assay using AlphaScreen technology. The assay entails binding a biotin-labeled DNA oligomer FL814 and His-tagged HMGA2 to streptavidin-coated donor beads and nickel chelate (Ni-NTA) acceptor beads, respectively (Fig. 1). A series of DNA binding studies were performed to determine the optimal conditions for the AlphaScreen Primary Assay (Fig. S1a-d). After these experiments, 12.5 nM of FL814 and 62.5 nM of HMGA2 were chosen for the assay. The assay tolerated up to 1% DMSO without any significant change in signal. We have previously identified two commercially available compounds netropsin and WP631 In this study, we also developed a LANCE time-resolved fluorescence energy transfer (TR-FRET) assay as a secondary assay for our screening. The His-tagged HMGA2 and biotinylated oligomer FL814 were linked to Europium-labeled anti-6 × His antibody and APC-labeled streptavidin, respectively. A series of DNA binding studies were performed to determine the optimal conditions for the secondary Assay (Fig. S2). After these experiments, 12.5 nM of FL814 and 62.5 nM of HMGA2 were chosen for the assay, which is the same for the primary assay. The assay tolerated up to 0.25% DMSO without any significant change in signal (Fig. S2c). For a counterscreen assay to exclude those compounds that nonspecifically bind to the protein surface, an AlphaScreen assay of H4 peptide binding to bromodomain-containing protein 4 (BRD4) was adopted (Fig. S2e). Since we are looking for compounds with anti-metastasis or anti-obesity activities, compounds with high levels of cytotoxicity are undesirable. A cytotoxicity assay using the ATPlite Luminescence Assay System and the human epidermoid carcinoma cell line, A431, was developed to eliminate compounds that exhibit cytotoxicity. Compounds that potently and selectively inhibit HMGA2 binding to FL814, and display a IC 50 > 50 μM are will be prioritized for further characterization. Those that do not will be excluded. Furthermore, potential DNA-binding compounds, i.e., DNA intercalators, minor groove binders, and DNA alkylating agents, can be identified by analyzing their chemical structures and will be excluded as well.
Screen the LOPAC1280 compound library. With the establishment of the miniaturized, automated uHTS assays, we screened the Sigma LOPAC1280 collection of pharmacologically active chemical compounds at a final concentration of 5 µM (Fig. S3a).  Table S1 show our results and parameters of the AlphaScreen primary uHTS assay. The following 16 compounds showed ≥ 50% inhibition of HMGA2-FL814 interactions: cisplatin, cDPCP, carboplatin, mitoxantrone, Ro 90-7501, aurintricarboxylic acid, GW5047, indirubin-3′-monoxime, 6-hydroxy-DL-DOPA, methyl-3,4-dephostatin, tyrphostin 51, (2′Z,3′E)-6bromoindirubin-3′-oxime, reactive blue-2, JFD00244, steviol, and suramin. These 16 compounds were cherry picked and then subjected to testing in the LANCE TR-FRET secondary assay. The following seven compounds demonstrated ≥ 50% inhibition of HMGA2-DNA interactions in both uHTS assays: cisplatin, cDPCP, carboplatin, mitoxantrone, Ro 90-7501, aurintricarboxylic acid, and suramin. Dry powders of these seven compounds were purchased for additional testing in these assays. The identity of the compounds was confirmed by LC/ MS, and fresh stock solutions (10 mM) were prepared in 100% DMSO. We next performed a series of titration experiments and determined the potencies (IC 50 ) values of these seven hits in the primary AlphaScreen assay, and the secondary LANCE TR-FRET assay, as well as the counter screen assay, and cytotoxicity assay. Table S2 summarizes our results. Cisplatin, cDPCP, carboplatin, and mitoxantrone are known DNA-binding agents and likely represent compounds that inhibit HMGA2-DNA interactions by binding to the AT-rich DNA sequence of FL814. Additionally, these DNA-binding compounds may also inhibit other essential cellular functions, which prevent them from further investigation. Intriguingly, although Ro 90-7501, aurintricarboxylic acid, and suramin do not bind to DNA due to their chemical properties, these three compounds strongly inhibit HMGA2-DNA interactions. Of particular interest is suramin, a highly negatively charged antiparasitic drug 34 (Fig. S4) that potently inhibits HMGA2-FL814 interactions with an IC 50 of 2.58 μM (Fig. 3). Additionally, suramin did not inhibit H4 peptide binding to BRD4 in the counter screen assay (Fig. 3c) and is not cytotoxic to human A-431 cells (Fig. 3d, Table S2). As a final validation of suramin as a "hit" from the HTS, we confirmed its inhibitory effect on HMGA2 interactions with DNA using the PDI-ELISA assay (Table 1). Thus, suramin meets all criteria that we set for the identification of novel HMGA2-DNA inhibitors. www.nature.com/scientificreports/ Suramin and analogues strongly bind to HMGA2 and ATHP3. We next sought to determine the mechanism by which suramin inhibits HMGA2-DNA interactions. Since suramin carries 6 negative charges (Fig. S4), it should not bind to DNA due to the fact that DNA is highly negatively charged. Instead, it should bind to HMGA2 because HMGA2 is positively charged 35 . ITC studies revealed that suramin physically interacts with HMGA2 (Fig. 4a, Table 1). These studies revealed that there are two types and a total of five suramin binding sites on HMGA2. The first type of three suramin binding sites has a binding constant of 4.08 ± 0.92 × 10 6 M −1 with the following thermodynamic parameters: ΔG, − 9.02 kcal/mol; ΔH − 14.58 kcal/mol; and -TΔS, 5.56 kcal/ mol. The second type of two suramin binding sites has a binding constant of 4.34 ± 2.16 × 10 4 M −1 with the following thermodynamic parameters: ΔG, − 6.33 kcal/mol; ΔH − 6.57 kcal/mol; and − TΔS, 0.24 kcal/mol. These five suramin binding sites of HMGA2 were confirmed by our mass spectrometric experiments at high suramin to HMGA2 ratios (Fig. S5). Because HMGA2 contains three highly positively charged "AT-hook" DNA binding motifs, it is reasonable to assume that suramin strongly binds to these highly positively charged motifs through charge-charge interactions. Our ITC experiments of suramin titrating into an "AT-hook" peptide 3 (ATHP3) The couterscreen assay using the BRD4 AlphaScreen Assay. (d) The cytotoxicity assay. All assays were described in "Methods". The standard deviation was calculated according to three independent experiments. The curve represents the best fit of a four parameter logistic that was determined by nonlinear regression. Data points represent mean ± SEM .  Table 1). Likewise, all these chemical compounds except sodium 1-naphthalenesulfonate strongly bind to both HMGA2 and ATHP3. Intriguingly, although the negative charge is important, it is not the only parameter that determines their inhibition potency. For example, NF110 only carries 4 negative charges. However, it strongly inhibits HMGA2-DNA interactions with an IC50 of 0.87 ± 0.04 μM. In contrast, although NF023 has 6 negative charges, its inhibition IC50 was determined to be 10.63 ± 0.46 μM, tenfold higher than that of NF110. These results suggest that both the charge and structure are very important for the inhibition of HMGA2-DNA interactions by these related compounds.

Effects of suramin on tumor cells.
Previous studies showed that HMGA2 plays a critical role in the epithelial-mesenchymal transition (EMT) induced by transforming growth factor-β (TGF-β) 36 . Further studies showed that HMGA2 directly binds to the promoter region of the transcriptional factor SNAIL and coregulates SNAIL expression with intracellular transducers, Smads during the EMT induction 37 . Since HMGA proteins including HMGA2 are highly expressed in glioblastomas and glioblastoma-derived brain tumor stem cells (BTSCs) 38 , where they play pivotal roles in regulating self-renewal, differentiation and symmetric/asymmetric division 38,39 , BTSCs are a good cell assay system to evaluate the effect of suramin on the expression of SNAIL and several differentiation and stemness markers, such as MSI1 (Musashi RNA Binding Protein 1), ID2 (inhibitor of DNA binding 2), and OLIG2 (oligodendrocyte transcription factor) 38 . Here, we exposed two BTSC lines (BTSC#83 and BTSC#30p) to different concentrations of suramin. Consistent with the cytotoxicity studies (Table S2) Fig. S8b,d,e) and, in contrast, increased the expression of OLIG2, an oligodendrocyte differentiation marker 38 (Fig. 5c). Our previous results showed that silencing of another AT-hook protein HMGA1 in BTSC#83 and BTSC#30p downregulated MSI1 expression and induced cells to adhere and extend neurite-like structures or acquire flat epithelial morphology 38 . Fig. S8 also shows that silencing of HMGA1 induces reduction in ID2 expression in BTSC#30p but not in BTSC#83 (Fig. S8c). Similarly, treatment with 100 μM Suramin for 48 h induces reduction in ID2 expression in BTSC#30p but not in BTSC#83 (Fig. S8d,e), suggesting that HMGA silencing produces similar effects to suramin treatment. Although these results support a hypothesis by which suramin targets HMGA-DNA interactions in these two BTSC cell lines and therefore induces their differentiation, suramin may also work with other cellular components/enzymes, such as protein-tyrosine phosphatases 40 and various growth factors 41 , to achieve its biological functions.
Next, we examined whether suramin is able to dissociate HMGA2 from its target DNA sequences inside cells. Since the ID2 gene promoter region contains multiple AT-rich DNA sequences and can be regulated by HMGA2 36,37 presumably through binding to these AT-rich DNA sequences, Chromatin Immunoprecipitation Analysis (ChIP) was performed to determine whether suramin could dissociate HMGA2 from the ID2 gene promoter region. The role of HMGA1 and HMGA2 in regulating ID2 transcription was confirmed by analyzing its expression in mouse embryonic fibroblasts (MEFs) knock-out for HMGA1, HMGA2, and the A1/A2 double mutant, respectively 42 . These three knock-out MEFs showed a significant down-regulation of ID2 expression (Fig. S9), indicating that both HMGA1 and HMGA2 can regulate the ID2 expression. We identified two  www.nature.com/scientificreports/ AT-rich regions, hereafter indicated as "region A" and "region B", in the human ID2 promoter, containing the HMGA2 consensus sequence 26 (Table S2) for our ChIP experiments. Due to availability, the high proliferation rate, and the expression of HMGA proteins, the anaplastic thyroid carcinoma cell line 8505c was used. Briefly, after 8505c cells were treated with 100 and 200 µM suramin for 6 days, the ChIP experiments were performed. Fig. S10 shows our results. In the absence of suramin, the sample immunoprecipitated with the anti-HMGA2 antibody is enriched both in region A and B (Fig. S10a), indicating that HMGA2 is able to bind to both regions. Unexpectedly, we did not detect significant HMGA2 binding reduction in the presence of 100 µM of suramin (Fig. S10). Nevertheless, 200 µM of suramin caused a drastic reduction in the enrichment of the anti-HMGA2 precipitated sample with respect to the sample immunoprecipitated with IgG only, indicating the loss of the binding of HMGA2 to regions A and B. These results demonstrate that suramin can inhibit HMGA2-DNA binding in 8505c cells. Similarly, the mRNA expression level of ID2 was significantly decreased in the presence of 200 and 400 µM of suramin (Fig. S10b).

Discussion
In this article, we report the development of a miniaturized, automated AlphaScreen uHTS assay to identify small molecule inhibitors targeting HMGA2-DNA interactions. This uHTS assay has excellent screening parameters with a Z′ and S/B ratio at 0.83 and 438, respectively (Table S1). After screening the LOPAC1280 compound library, we found that suramin, a highly negatively charge anti-parasitic drug that is used to treat African sleeping sickness and river blindness 43,44 , is a potent inhibitor of HMGA2-DNA interactions. Further, our results show that suramin's inhibition of HMGA2-DNA interactions stems from its binding to HMGA2 with high affinity, particularly to ATHPs. Our molecular modeling studies confirmed this hypothesis (Fig. 6). The suramin-ATHP3 complex shows that the suramin sulfonated groups are close to the Arg and Lys residues, providing ionic/ hydrogen-bond interactions. The ring structures at the two ends of suramin are folded to embrace the central proline ring from either side of the residue. DNA-binding proteins, such as transcriptional factors are excellent targets for anticancer therapy 45 . Indeed, several clinically important anticancer drugs, such as doxorubicin and cisplatin 46,47 , bind to DNA, disrupt protein-DNA interactions, and as a result prevent DNA-binding proteins including transcriptional factors from binding to their target DNA sites 48 . However, a major disadvantage of these anticancer drugs is that they bind to DNA nonspecifically. Consequently, they kill cancer cells as well as normal cells causing serious side effects. Therapeutically active compounds that specifically block oncogenic transcriptional factors from binding to their DNA recognition sites would be superior medicines. Such medicines would be expected to have fewer, less severe www.nature.com/scientificreports/ side effects, thereby potentially increasing the dose and duration of chemotherapy and perhaps improving treatment outcomes as well as quality of life. Transcription factors are widely considered to be "undruggable" because they usually do not have enzymatic activities, and lack deep pockets to accommodate drug-like molecules 49 . Further, the lack of HTS assays to identify inhibitors from small molecule repositories also contributes to the perceived undruggable nature of transcriptional factors 33 . Our results refute these notions, at least with respect to the potential druggability of HMGA2. Previously, we developed protein-DNA or protein-RNA interaction enzyme-linked immunosorbent assays (PDI-ELISA or PRI-ELISA) to identify inhibitors targeting specific protein-nucleic acids interactions 33 . These methods are versatile and can be applied to any nucleic acid binding proteins as long as an antibody is available. In fact, one may use tagged proteins, such as His-tagged proteins so that antibodies against His-tag can be used in PDI-ELISA or PRI-ELISA. Regardless, as we pointed out above, because streptavidin-coated 1536-well plates are expensive, PDI-ELISA or PRI-ELISA cannot be configured into a miniaturized, automated uHTS assay in a 1536 well plate format. Additionally, too many washing steps were used, which make PDI-ELISA or PRI-ELISA a "lengthy" procedure and unsuitable for automated uHTS studies. Here we showed that AlphaScreen assays could be utilized to screen for inhibitors against HMGA-DNA interactions by using His-tagged HMGA2 linked to nickel chelate (Ni-NTA) acceptor beads. The use of AlphaScreen uHTS assays has several advantages. Because His-tag can be added to most of DNA-binding proteins and usually does not interfere with their DNA binding activities, AlphaScreen assays can be used to screen and identify inhibitors targeting most DNA-binding proteins. Another advantage is that AlphaScreen uHTS assays are cost-effective. For instance, in this study, only one 1536-well plate was used in the screening of the LOPAC1280 compound library. It is anticipated that AlphaScreen uHTS assays using His-tagged DNA binding proteins will be utilized to identify inhibitors for other DNA-binding proteins in the future. The most intriguing result of this uHTS study is the identification of suramin as a potent inhibitor for HMGA2-DNA interactions. The inhibition is likely through a mechanism by which suramin tightly binds to highly positively charged ATHPs. This is not surprising because suramin, a polysulphonated naphylurea, carries 6 negative charges at physiological conditions (Fig. S4). As shown in Fig. 6, charge-charge interactions and hydrogen bonding between the suramin sulfonated groups and Arg/Lys residues play critical roles in the binding of suramin to ATHPs. This inhibition mechanism is different from that of netropsin and other DNA minor groove binders that prevent HMGA2 from binding to the minor groove of AT-rich DNA sequences 15,33 . In fact, suramin is the first chemical compound that was found to tightly bind to the intrinsically disordered protein (IDP) HMGA2 and the "unstructured" DNA-binding motif ATHPs. Since HMGA1 proteins also carry multiple "AT-hook" DNA-binding motifs 50,51 , suramin is expected to strongly inhibit HMGA1-DNA interactions as well. Furthermore, it is reasonable to predict that suramin tightly binds to many positively charged motifs on protein surface that contain lysine and/or arginine residues.
Suramin is a century-old drug synthesized in the 1920s by Bayer to treat human African trypanosomiasis (HAT) 52 . However, the mechanism or mode of action is still unknown 53 . Suramin is usually administered by intravenous injection due to the poor intestinal absorption 34 and binds to serum proteins, such as albumin and low-density lipoprotein (LDL), immediately after administration 34 . It is believed that the parasite takes up the drug through receptor-mediated endocytosis of the protein-bound drug 54 . Suramin is highly effective against blood-stream forms of the parasite but not very active against procyclic trypanosomes 55 . Since the bloodstream forms of Trypanosoma brucei lacks a functional mitochondrion and entirely depend on glycolysis for their energy needs, this led to the hypothesis that the glycolytic pathway of the parasite is the target of the drug 56 . However, so far there is no direct evidence to support this hypothesis 34 . Another interesting feature of T. brucei is that its mitochondrion contains the so-called kinetoplast DNA (kDNA), comprising of ~ 73% AT base pairs 57 . Recent studies showed that certain DNA minor groove-binding compounds were able to enter the mitochondrion of T. brucei, bind to AT sequences of kDNA and, as a result, cause cell death 58 . It was suggested that these minor groove-binding compounds might disrupt the functions of kDNA binding proteins, such as the TbKAP6 protein that is essential for kDNA replication and maintenance and also for cell viability 59 . TbKAP6 contains two HMG boxes and binds to DNA minor groove 59 . Interestingly, this protein carries several highly positively charged motifs similar to ATHP 59 . One possibility is that suramin tightly binds to these positively charged motifs in TbKAP6 and prevents its binding to kDNA. Further studies are needed to support this hypothetic mechanism.
Another interesting function of suramin is its well-studied anticancer and anti-metastasis activities 41,60,61 although the mechanism is still obscure. Possible mechanisms include inhibiting important enzymes, such as cullin-RING E3 ubiquitin ligases 62 , protein-tyrosine phosphatases 40 , and blocking various growth factors binding to their receptors 41 . In fact, suramin has "so many" targets that its cellular and physiological activities are expected to stem from not just one single target but multiple targets 63 . Here we reveal another potential target, HMGA2 and/or HMGA proteins for this intriguing small molecule. It is possible that binding of suramin to HMGA2 prevents the protein from binding to the target DNA sequence of the SNAIL promoter 37 and decreases the expression of SNAIL (Fig. 5c) 36,37 . The differentiation of BTSCs induced by suramin is consistent with this hypothesis (Fig. 5a). Our ChIP analysis showing that high concentrations of suramin can dissociate HMGA2 from its target DNA sequences inside cells is also consistent with this hypothesis (Fig. S10). Nevertheless, it will be difficult to design experiments demonstrating that HMGA2 and/or HMGA proteins are the suramin's only target or main target in cells. What we learned so far is that suramin likely prefers binding to positively charge motifs on protein surface similar to the way it binds to the "AT-hook" DNA-binding motifs (Fig. 6). It is also possible to identify a specific inhibitor for HMGA2 and HMGA proteins with the established uHTS in the future. Please note that suramin has several disadvantages for further development. First, suramin binds to many proteins in vitro and in vivo. It may not be very specific for HMGA2 targeting. Secondly, suramin is not orally bioavailable and must be given intravenously. Third, it has limited bioavailability through tissue due to the fact that 98% of the drug are protein bound in blood serum. Furthermore, suramin causes a fair number of side effects and unpredictable pharmacokinetics.

Methods
Materials. The His-tagged mammalian HMGA2 was purified using a Ni-NTA agarose column followed by an SP Sepharose fast flow column as described previously 26 . An extinction coefficient of 5810 M −1 cm −1 at 280 nm was used to determine its concentration 26 . Biotin-labeled DNA oligomer FL814 carrying a specific SELEX binding site of HMGA2 26,33 was purchased from Eurofins MWG Operon, Inc. AlphaScreen histidine (nickel chelate) detection kits containing nickel chelate acceptor beads and streptavidin donor beads (#6760619M), LANCE Ultra ULight-anti-6xHIS (#TRF0105) and LANCE Eu-W1024 Streptavidin (#AD0062), as well as 1536 well Optiplates plates (#6004290) were purchased from Perkin Elmer. Netropsin was purchased from Sigma and used without further purification. Suramin and NF023 were purchased from MilliporeSigma, Inc. Sodium 1-Naphthalenesulfonate was obtained from TCI America, Inc. NF110, NF340, NF449, and NF546 were purchased from Tocris Biosciences. His-tagged BRD4-BD1 was purchased from BPS Bioscience (#31042). Pre-acetylated Biotin- HMGA2 AlphaScreen ultra HTS assay. Using a Labcyte Echo 555 acoustic dispenser 5 nL of DMSO were added to columns 1-4 and 45-48 of a white 1536 well plate, while 5 nL of 2 mM compounds in DMSO were added to columns 5-44. Using a Beckman BioRAPTR FRD bulk reagent dispenser 1 µL of assay buffer (30 mM Citrate, 300 mM NaCl, and 0.005% Tween 20) was added to columns 1 and 2. Next 1 µL of assay buffer containing 125 nM HMGA2 was added to columns 3-48 with a Beckman BioRAPTR FRD bulk reagent dispenser. Finally, 1 µL of assay buffer containing 25 nM FL814 was added to every well of the plate using a Beckman BioRAPTR FRD bulk reagent dispenser. The plate was then centrifuged at 200 × g for 1 min. After 30 min at room temperature, 2 µL of bead buffer (10 mM HEPES, 300 mM NaCl and 0.005% Tween 20) containing 20 µg/ mL anti-6xHIS acceptor beads and 20 µg/mL streptavidin donor beads were dispensed into every well using a Beckman BioRAPTR FRD bulk reagent dispenser. The plates were then centrifuged at 200 × g for 1 min. After 1 h at room temperature the plates were read on a Perkin Elmer Envision multimode plate reader in AlphaScreen mode.
HMGA2 Lance assay. Using a Labcyte Echo 555 acoustic dispenser 5 nL of DMSO were added to columns 1-4 and 45-48 of a white 1536 well plate, while 5 nL of 2 mM compounds in DMSO were added to columns 5-44. Using a Beckman BioRAPTR FRD bulk reagent dispenser 1 µL of assay buffer (10 mM Tris, 300 mM NaCl and 0.005% Tween 20) was added to columns 1 and 2. Next, 1 µL of assay buffer containing 125 nM HMGA2 was added to columns 3-48 with a Beckman BioRAPTR FRD bulk reagent dispenser. Finally, 1 µL of assay buffer containing 25 nM FL814 was added to every well of the plate using a Beckman BioRAPTR FRD bulk reagent dispenser. The plate was then centrifuged at 200 × g for 1 min. After 30 min at room temperature 1 µL of assay buffer containing 250 nM LANCE Ultra ULight-anti-6xHIS and 1 µL of assay buffer containing 12 nM LANCE Eu-W1024 Streptavidin was dispensed into every well using a Beckman BioRAPTR FRD bulk reagent dispenser. The plates were then centrifuged at 200 × g for 1 min. After 1 h at room temperature the plates were read on a Perkin Elmer Envision multimode plate reader in TR-FRET mode (excitation @340 nm, first emission at @665 nm, second emission at 615 nm).
BRD4 AlphaScreen assay. Using a Labcyte Echo 555 acoustic dispenser 5 nL of DMSO were added to columns 1-4 and 45-48 of a white 1536 well plate, while 5 nL of 2 mM compounds in DMSO were added to columns 5-44. Using a Beckman BioRAPTR FRD bulk reagent dispenser 1 µL of assay buffer (50 mM HEPES, 100 mM NaCl, 0.1% BSA and 0.0005% CHAPS) was added to columns 1 and 2. Next, 1 µL of assay buffer containing 50 nM BRD4 was added to columns 3-48 with a Beckman BioRAPTR FRD bulk reagent dispenser. Finally, µL of assay buffer containing 50 nM peptide was added to every well of the plate using a Beckman Bio-RAPTR FRD bulk reagent dispenser. The plate was then centrifuged at 200 × g for 1 min. After 60 min at room temperature 2 µL of assay buffer containing 20 µg/mL anti-6xHis acceptor beads and 20 µg/mL streptavidin donor beads was dispensed into every well using a Beckman BioRAPTR FRD bulk reagent dispenser. The plates were then centrifuged at 200 × g for 1 min. After an overnight incubation at room temperature the plates were read on a Perkin Elmer Envision multimode plate reader in AlphaScreen mode. www.nature.com/scientificreports/ bulk reagent dispenser. The plates were centrifuged at 200 × g for 1 min and then incubated at room temperature. After 10 min the plates were read using a Perkin Elmer Viewlux multimode plate reader in luminescence mode.

Cell viability assay.
Protein-DNA interaction ELISA (PDI-ELISA) assay. The PDI-ELSA assays were described previously 33 and used to determine the inhibition IC 50 against HMGA2-DNA interactions. The apparent inhibitory IC 50 values were obtained using the following equation: y = min + ( (max−min) 1+10 ((x−logIC 50 )×Hillslope ) , where x, y, max, and min represent the inhibitor's concentration, the inhibition level, the maximum inhibition value, and the minimum inhibition value, respectively.
Isothermal titration calorimetry (ITC). ITC experiments were conducted using a VP-ITC titration calorimeter (MicroCal, Inc., Northampton, MA) interfaced to a PC. Origin 7.0, supplied by the manufacturer was used for data acquisition. For a typical ITC experiment, the titration was set up so that 10 µL of 0.2 mM suramin or analogs was injected every 120 s, up to a total of 29 injections, into an HMGA2 sample (1.44 mL of 5 µM) or ATHP3 sample (1.44 mL of 15 µM) in the sample cell in 10 mM Tris-HCl pH 8.0 and 1 mM EDTA. The heat liberated or absorbed is observed as a peak corresponding to the power required keeping the sample and reference cells at identical temperatures. The peak produced during the injection is converted to heat output by integration and corrected for cell volume and sample concentration. Control experiments were also carried out to determine the contribution of the heats of dilution arising from (1) suramin or analogs into buffer and (2) buffer into HMGA2 or ATHP3. The net enthalpy for the titration reaction was determined by subtraction of the component heats of dilution.
Native mass spectrometry. A custom-built nano electrospray unit (nESI) was coupled to a Maxis Impact HD Q-TOF mass spectrometer (Bruker, Billerica, MA) for all the native mass spectrometry analysis. Quartz glass capillaries (O.D.: 1.0 mm and I.D.: 0.70 mm) were pulled utilizing a P-2000 micropipette laser puller (Sutter Instruments, Novato, CA) and loaded with 10 µL aliquot of the sample solution. Sample solutions consisted of 1-10 µM HMGA2 in 10 mM ammonium acetate solution at physiological pH (pH = 6.7). For the observation of the HMGA2-Ligand complexes, a 1:1, 1:3 and 1:10 ratio of 5 µM concentration of the HMGA2 and Ligand (suramin) were prepared in 10 mM ammonium acetate and let it rest for 10 min prior infusion. A typical nESI source voltage of +/− 600-1200 V was applied between the pulled capillary tips and the MS instrument inlet. Ions were introduced via a stainless-steel tube (1/16 × 0.020″, IDEX Health Science, Oak Harbor, WA) held at room temperature into the TIMS cell. Solvents, methanol, and ammonium acetate salts utilized in this study were analytical grade or better and purchased from Fisher Scientific (Pittsburgh, PA). A Tuning Mix calibration standard (G24221A) was obtained from Agilent Technologies (Santa Clara, CA) and used as received. Mass spectra were processed using Bruker Compass Data Analysis version 5.1 (Bruker Daltonik GmbH).
Cell cultures and growth curves. BTSC#83 and BTSC#30p and their culture conditions have been described 38 . For growth curves, suspension cultures were mechanically disaggregated and filtered through 37 µm filters (StemCell Technologies, Vancouver, Canada) in order to obtain bona fide single-cell suspension. 1 × 10 3 cells were plated in U-bottom 96-well plates and treated or not with 100, 200 and 400 µM Suramin. Cell growth was assayed at the indicated time using the CellTiter Assay System (Promega). The growth conditions of 8505c anaplastic thyroid carcinoma cells and mouse Embryonic Fibroblasts (MEFs) knock-out for HMGA1, HMGA2, and HMGA1/A2 double mutant have been described previously 42,64 . P values were calculated by using two-way ANOVA followed by Sidak's multiple comparison test.
Protein extraction and Western blot. The single-cell suspension was obtained by mechanical disaggregation of the spheres and total proteins were extracted 48 h later, as previously described 38 . After separation by SDS-polyacrylamide gel electrophoresis, proteins were blotted on nitrocellulose membranes (GE Healthcare Europe Gmb) and hybridized with the following antibodies: anti-HMGA1, anti-HMGA2 (Genetex) or affinitypurified anti-HMGA2, anti-α-actin (SantaCruz Biotechnologies).

RNA extraction and qRT-PCR analyses.
The single-cell suspension was obtained by mechanical disaggregation of the spheres and RNA was extracted 48 h later, by using Direct-zol RNA MiniPrep (Zymo Research). One μg of RNA was retrotranscribed, by using QuantiTect® Reverse Transcription Kit (Qiagen) and qRT-PCR was performed as described 38 . Primers used in the qRT-PCR experiments are listed in Table S3. The 2 -ΔΔCt formula was used to calculate the differential gene expression.
Chromatin immunoprecipitation (ChIP) assay. Chromatin immunoprecipitation was performed as described 65 . Briefly, for each sample, 5 × 10 6 cells were collected by trypsinization and cross-linked with 1% formaldehyde; after sonication, the samples were immunoprecipitated with 10 µL anti-HMGA2 antibody (Genetex) or with control normal rabbit IgG and the chromatin was extracted. qPCR amplification was performed on 6 µL of immunoprecipitated DNA, using primers amplifying regions A and B of the ID2 promoter (Table S2). C t values of samples precipitated with anti-HMGA2 and IgG antibodies were normalized with C t values of input DNA. Fold change enrichment was calculated in comparison to normalized IgG, using the 2 −ΔΔCt formula. P values were calculated by using one-way ANOVA followed by Sidak's multiple comparison test.

Molecular dynamics simulation.
Since HMGA2 is shown to be an intrinsically disordered protein, we first performed molecular dynamics (MD) simulations to generate multiple disordered conformations of Scientific Reports | (2020) 10:18850 | https://doi.org/10.1038/s41598-020-75890-0 www.nature.com/scientificreports/ the suramin-interacting segment of HMGA2 chain (-GEKRPRGRPRKW-). Using the Charmm-Gui web interface 66 , the peptide was solvated in a cubic water box with TIP3 water and the system was neutralized by adding five Clions. The final system contained ~ 16,000 atoms. NAMD 2.12 67 was used to perform all-atom molecular dynamics with CHARMM36 force field 68 . The particle mesh Ewald (PME) method 69 was used for calculating the long-range ionic interactions. The system was minimized for 10,000 steps, followed by a 100 ps equilibration at 300 K with 1 fs time step. A 100-ns production simulation with 2-fs time step was performed at a constant pressure of 1 atm. and T = 300 K. The Nose-Hoover Langevin-piston method 70 was used for pressure coupling, with a piston period of 50 fs and a decay of 25 fs, and the Langevin temperature coupling with a friction coefficient of 1 ps −1 was used for maintaining the temperature. From the 100-ns simulation trajectory, 1000 protein pdb frames were extracted using Visual Molecular Dynamics (VMD) 71 .
In silico docking studies. Suramin was docked to 1000 MD-generated conformations using AutoDock Vina 1.1.2 72 . The protein pdb files and the suramin compound structure were first converted to pdbqt format for docking. Using custom scripts, Suramin was screened against the protein conformations and the resulting scores of the complexes were sorted and ranked according to their binding affinities.