Molecular mechanisms of bifunctional vitamin D receptor agonist-histone deacetylase inhibitor hybrid molecules in triple-negative breast cancer

The active form of vitamin D, 1,25-dihydroxyvitamin D (1,25D), and its analogues signal through the nuclear vitamin D receptor (VDR), a ligand-regulated transcription factor, and have been extensively investigated as anticancer agents. 1,25D and its analogs have potential in combination therapies because they exhibit synergistic activities with other anticancer agents such as histone deacetylase inhibitors (HDACi). We have developed a series of hybrid molecules that combine HDACi within the backbone of a VDR agonist and thus represent fully integrated bifunctional molecules. They exhibit anti-tumor efficacy in reducing tumor growth and metastases in an aggressive model of triple-negative breast cancer. However, their solubility is limited by their hydrophobic diarylpentane cores. Our goals here were two-fold: (1) to improve the solubility of hybrids by introducing nitrogen into diarylpentane cores, and (2) to investigate the molecular mechanisms underlying their anti-tumor efficacy by performing comparative gene expression profiling studies with 1,25D and the potent HDACi suberoylanilide hydroxamic acid (SAHA). We found that substituting aryl with pyrydyl rings did not sacrifice bifunctionality and modestly improved solubility. Notably, one compound, AM-193, displayed enhanced potency as a VDR agonist and in cellular assays of cytotoxicity. RNAseq studies in triple negative breast cancer cells revealed that gene expression profiles of hybrids were very similar to that of 1,25D, as was that observed with 1,25D and SAHA combined. The effects of SAHA alone on gene expression were limited and distinct from those 1,25D or hybrids. The combined results suggest that efficacy of hybrids arises from targeting HDACs that do not have a direct role in gene regulation. Moreover, pathways analysis revealed that hybrids regulate numerous genes controlling immune cell infiltration into tumors and suppress the expression of several secreted molecules that promote breast cancer growth and metastasis.

Vitamin D is a secosteroid obtained naturally from limited dietary sources and from exposure of skin to sufficient solar ultraviolet B (UVB) irradiation via the photochemical and thermal conversion of 7-dehydrocholesterol. The active form of vitamin D, 1,25-dihydroxyvitamin D (1,25D, 1, Fig. 1) is produced via sequential hydroxylations and signals through the vitamin D receptor (VDR), a ligand-regulated transcription factor 1,2 . 1,25D has attracted extensive interest because of its "non-classical actions", including its potential as an anticancer agent 1 . This has spurred the development of numerous secosteroidal and non-secosteroidal analogues of 1,25D, and mechanistic studies have been greatly facilitated by the fact that the VDR is a soluble, well-defined and experimentally tractable target 3 . Several preclinical studies in multiple cancer models have provided evidence for the anti-tumor activities of 1,25D or its analogues [4][5][6] . In spite of these successes, 1,25D and its analogues have failed as monotherapies in cancer because of acquired tumor resistance, even though vitamin D signaling often remains intact in resistant cells. In addition, results of cancer prevention trials with vitamin D are mixed, although these www.nature.com/scientificreports/ 1,25D and SAHA together. We found that potent bifunctional hybrids can be produced by nitrogen substitutions into the diarylpentane core, with an accompanying improvement in solubility. We also found that expression profiles in 1,25D-or hybrid-treated cells are remarkably similar, consistent with gene expression is driven largely by VDR agonism, which suggests that HDACi activity of hybrids may target enzyme(s) not implicated in gene regulation. Moreover, hybrids inhibit the expression of genes encoding multiple classes of secreted proteins, including several chemokines implicated in promoting breast tumor growth and metastasis.

Results and discussion
Design and synthesis of heterocyclic VDR agonist/HDACi hybrids. Binding of 1,25D or its analogues to the VDR is strongly dependent on interactions of the terminal 1-, 3-and 25-hydroxyl groups or their surrogates to key residues in the receptor ligand binding pocket 21 . The compatibility of the hydroxamic acid as a hydroxyl replacement in 1,25D and its analogs provided a route to introduction of HDAC inhibition into the structures of VDR agonists. This led to the development of the bifunctional analogue DK-406 15,20 . However, due to its hydrophobic 3,3-diarylpentane core, DK-406 possesses limited water solubility, a feature not uncommon among many commercial drugs, including SAHA 22 . A solution of 12.5% ethanol, 37.5% PEG-400 and 50% saline was used to solubilize DK-406 and related hybrids for previous in vivo studies 20 , but it would be desirable to produce compounds with improved water solubility without sacrificing bifunctionality. As a first step, we designed new heterocyclic analogues of DK-406, namely AM-155, -171 and -172 (17a-c, Fig. 2), and AM-191-193 (26a-c, Fig. 3), where either aromatic ring was replaced with pyridine derivatives, which would be expected to improve solubility in water while not significantly disrupting VDR agonism or HDACi activity. The computed partition coefficient (logP) of heterocyclic hybrids improved from ~ 6.5 for hybrids such as DK-367 to ~ 5.0 with the nitrogen substitution ortho to the 3,3-pentane linker being the more favourable (Fig. S1) bringing them into the normal logP range for drug-like molecules 23 . The pyridine hybrids were synthesized by two similar routes based on a key aza-Achmatowicz rearrangement to access the core 3-pyridyl-3-arylpentane from a furfurylamine precursor. The synthesis of the A-ring hybrid series began with the mild para-bromination of o-cresol (7) using a system of DMSO and 48% aqueous HBr to www.nature.com/scientificreports/ afford 4-bromophenol 8 in 72% yield 24 . 4-Bromophenol 8 was subjected to O-alkylation with 1-chloropinacolone to form an α-alkoxyketone (9) which was then reduced to the corresponding secondary alcohol 10 with NaBH 4 in 71% over two steps. Metal halogen exchange of 10 with n-BuLi followed by addition to 3-pentanone afforded tertiary alcohol 11 in 51% yield. Next, Friedel-Crafts alkylation with furfuryl phthalimide 12 in the presence of BF 3 •OEt 2 afforded 3-furyl-3-arylpentane 13 in 74% yield. Deprotection using hydrazine liberated free amine 14 in 73% yield. With 14, the precursor for the aza-Achmatowicz rearrangement in hand we examined the use of m-CPBA in dichloromethane, conditions first reported by Lefebvre 25,26 . While these produced the desired product 15, the yield was disappointing (27%) due to over oxidation leading to oxime and pyridine N-oxide by-products. Fortunately, classical conditions for the regular Achmatowicz rearrangement 27 , using bromine in 1:2 H 2 O/MeOH, provided 15 in an acceptable 62% yield. Finally, O-alkylation of 3-hydroxypyridine 15 with C2, C4, and C6 alkyl bromide esters followed by treatment with hydroxylamine and KOH afforded A-ring hybrids AM-155 (n = 1, 17a), AM-171 (n = 3, 17b), and AM-172 (n = 5, 17c) in 25-53% yields over two steps. The synthesis of the B-ring series mirrored that of the A-ring series producing B-ring hybrids AM-191 (n = 1, 27a), AM-192 (n = 3, 27b), and AM-193 (n = 5, 27c) in 9 steps (Fig. 3). As will be seen below, hybrid AM-193 proved to be the most promising of all the hybrids and thus we examined its aqueous solubility. Compared to its benzenoid analog DK-367 (6b), AM-193 possessed 27% higher solubility in water (2.43 µg/mL vs. 1.91 µg/ mL) after sonication for 30 min.
Biochemical characterization of heterocyclic VDR agonist/HDACi hybrids. The hybrids were assessed to determine their HDACi and VDR agonist activity as well as their antiproliferative properties relative to DK-406. HDACi activity was assessed in vitro using an acetylated lysine substrate that releases a fluorescent reporter upon HDACi deacetylation and subsequent trypsin digestion 28 . Increasing sidechain length of pyrimidine hybrids enhanced the potency of HDAC6 inhibition, with several compounds inhibiting HDAC6 with a sub-micromolar IC 50 , similar to that of DK-406 (Fig. 4a). In contrast, pyridyl hybrids were somewhat less potent inhibitors of HDAC2 compared to HDAC6 and the normal dependence on chain length was disrupted. The hydroxamic acid sidechains of AM-193 and DK-406 are comparable in length. Previous studies showed that the relatively long side chain of DK-406 correlated with optimized HDACi activity but came at the cost of a modest loss of potency of VDR agonism 15 . Our current studies with heterocyclic hybrids showed that potency of VDR agonism improved with increasing side chain length. Next, we analyzed the effects of compounds on levels of H3K9 and H3K27 acetylation in 4TO7 cells, which are controlled by class I HDACs 29,30 . Effects of hybrids on H3 acetylation were compared to that of 100 nM SAHA, which produces almost complete inhibition of class I HDAC activity in vitro 31 . DK-406 induced H3 acetylation at K9 and K27 at 10 -5 M (Fig. 4b,c) that was comparable to that of SAHA. As AM-193 was a less potent inhibitor than DK-406 of class I enzyme HDAC2 (Fig. 4a  www.nature.com/scientificreports/ www.nature.com/scientificreports/ we tested its activity at 10 -5 M. Under these conditions, AM-193 enhanced H3 acetylation, although to a lesser degree than DK-406 (Fig. 4b,c). We also tested the capacity of analogues to induce tubulin acetylation, controlled by class II enzyme HDAC6 32 , in 4TO7 cells. Based on their relative efficacy in cytotoxicity assays (see below, As previously 20 , we were interested in studying the efficacy of hybrids in models of TNBC. We used the 4T1 cell line 20 as well as the related TNBC cell line 4TO7, which was derived from the same mouse tumor as 4T1 cells in Balb/c mice 33 . 4TO7 cells readily form primary tumors. However, unlike 4T1 cells, which are aggressively metastatic, 4TO7 cells seed other organs but do not successfully establish metastases 33 . These lines also have the advantage that anti-tumor efficacy studies can be performed in Balb/c mice, which have intact immune systems. Compounds were initially tested for VDR agonist activity in 4TO7 cells by screening induction of the VDR target gene Cyp24a1. Compounds AM-155 and -171 (17a and 17b) were weakly potent VDR agonists (Fig. S3a,b) and were not studied further. Subsequent experiments confirmed that potency of VDR agonism increased with the length of the hydroxamic acid sidechain and that AM-193 (27c) was approximately tenfold more potent in the Cyp24a1 induction assay than DK-406 in 4TO7 cells (Fig. 5a). The potency of AM-193 was confirmed in related 4T1 cells (Fig. 5b). Notably, AM-193 and DK-406 appeared to be partial agonists in 4TO7 cells, whereas they functioned as full agonists in 4T1 cells (Fig. 5a,b). The pyridyl hybrids were then tested for www.nature.com/scientificreports/ efficacy in vitro in 4TO7 and 4T1 cells using an MT-Glo assay in which only viable cells can convert a precursor into a substrate for a luciferase bioluminescence assay (Fig. 5c,d; see Methods for details). AM-193 was the most potent compound tested and appeared to be at least tenfold more potent than DK-406 (Fig. 5c,d). In the same assay, the combination of a pharmacological concentration of 1,25D (1 µM) and 100 nM of the potent HDACi SAHA was not efficacious (Fig. 5c,d). The potency of AM-193 as a cytotoxic agent was also confirmed in the human TNBC cell line MB-MDA231 (Fig. 5e). cells were chosen for further study because they are less aggressively metastatic in vivo than 4T1 cells, and we are interested in future analyses of the efficacy of hybrids in preventing metastatic seeding the 4TO7 model. Cells were treated for 6 h to focus more on genes directly regulated by the VDR, or for 24 or 48 h to gain insight into longer-term effects on gene expression. An initial pilot study was performed in 4TO7 cells treated for 6 or 48 h with 1,25D and SAHA, alone or in combination, or with DK-406 (Supplemental Table 1). Of the 150 genes regulated by 1,25D after 6 h using the relatively stringent twofold cut-off, 91 overlapped with the 290 genes regulated by DK-406, with a similar degree of overlap observed between genes regulated by 1,25D + SAHA and DK-406 (Fig. 6a). There was a substantially lower degree of overlap between genes regulated by SAHA alone and DK-406 after 6 h (Fig. 6a). Remarkably, the overlap in gene regulatory profiles of cells treated with DK-406 and those exposed to 1,25D alone or with 1,25D + SAHA was substantially greater after 48 h than at 6 h, whereas that between DK-406 and SAHA alone was minimal (Fig. 6a). A similar trend in convergence of gene regulatory profiles at 48 h was observed when comparing the common genes regulated by DK-406 and 1,25D + SAHA with those regulated by 1,25D alone, whereas the overlap between the common genes and those regulated by SAHA was minimal (Fig. 6b,c). This data suggests that largely VDR-controlled gene regulation is driving the long-term gene expression profiles in cells treated with DK-406, as well as with 1,25D + SAHA.

Gene expression profiling by
To substantiate this result, a similar RNAseq study was conducted in 4TO7 cells treated for 6 or 24 h, which included AM-193 as well as DK-406, 1,25D and/or SAHA (Supplemental Table 3). Because of their differences in potency as cytostatic and cytotoxic agents (Fig. 6 above), cells were treated with 1 µM AM-193 and 10 µM DK-406. Principal component analysis of results from 6 h time points showed a remarkable clustering of results from cell treated with 1,25D, 1,25D + SAHA, DK-406 or AM-193 (Fig. 7a), whereas profiles from vehicle or SAHA-treated cells were similar (with the exception of one outlier of the SAHA-treated samples). This result is consistent with the relatively limited effects of SAHA alone on gene expression in 4TO7 cells, above. A similar clustering of results was observed in cells treated with vehicle or SAHA for 24 h (Fig. 7a). In contrast, PCA analysis clustered results from cells treated for 24 h with 1,25D, 1,25D + SAHA, DK-406 or AM-193 in a group distinct from vehicle-treated cells and distinct from that generated from the profiles obtained at 6 h (Fig. 7a). Venn diagrams of these results illustrate the substantial overlap in genes regulated by hybrids, 125D and 1,25D + SAHA (Fig. 7b).
Previous microarray studies in human myelomonocytic cells with the HDACi trichostatin A (TSA) and 1,25D showed that short-term treatment (90 min) with TSA had widespread effects on gene expression and substantially alter 1,25D-regulated gene expression 34 . Based on these findings, the relatively modest effects we observed in our two studies on gene expression in 4TO7 cells exposed to SAHA were unexpected, given that SAHA and TSA are broad-spectrum HDACi. However, our studies and the previous one differ in the cell lines used, and the duration of the treatments (90 min vs 6, 24 or 48 h). Although there was only partial overlap between the genes regulated by DK-406 and 1,25D alone or with SAHA after 6 h, there was substantial convergence in their gene expression profiles after 48 h. This suggests that long-term effects on gene expression by hybrids are very similar to those driven by 1,25D. The results are also consistent with our observations that, although AM-193 is more potent than DK-406 as a cytostatic and cytotoxic agent, notably acting at 10 -6 M, it was less efficacious than DK-406 at inducing H3 acetylation at 10 -5 M (Fig. 5c,d, above). This raises the possibility that any therapeutic effect of the HDACi component of hybrids may be through inhibition of HDACs that do not directly control gene expression. There are several potential targets of HDACi that bind to zinc metalloenzymes (HDACs 1-11; class I, II and IV enzymes), and many HDACs, particularly class II enzymes are partially or wholly cytoplasmic 12 . Consistent with this notion, after 48 h SAHA alone only modestly altered gene expression (84 genes regulated at east twofold), of which only 19 overlapped with the 660 genes whose expression was induced at least twofold by DK-406.
Most of the genes in these pathways encode secreted proteins. The expression of several matrix metalloproteinases was elevated in cells treated with 1,25D or hybrids, enzymes that have been implicated in cancer progression 35 . In contrast, the expression of numerous cytokines and chemokines was suppressed. These include www.nature.com/scientificreports/ www.nature.com/scientificreports/ Ccl2, Ccl5, Ccl20, Cxcl10, and Cxcl11, all of which have been implicated in promoting breast tumor growth or metastasis [36][37][38][39][40][41][42] . Conversely, expression of the gene encoding Il12b was induced. Elevated Il-12 levels in the breast tumor microenvironment promote antitumor immunity and correlate with reduced metastasis 43,44 . Notably, Ccl2, whose expression is repressed, promotes recruitment of M2 type macrophages, which facilitate tumor progression 37 . Similarly, expression of the genes encoding complement factor H (Cfh) and Siglec10 (Cd24), both of which can promote tumor immune escape in breast cancer 45,46 , were inhibited. Collectively, these results suggest that hybrids would alter the immune infiltrate composition and phenotype, suppressing tumor progression, and possibly promoting anti-tumor immunity.
In addition, several genes that are downregulated encode secreted factors whose expression correlates with breast cancer metastasis, including adrenomedullin (Adm) 47,48 , urokinase-type plasminogen activator (Plaur, uPA) 49 and the secreted protease Prtn3 50 . Finally, we note that treatment with hybrids suppressed expression of Slc2A4, which encodes the glucose transporter Glut 4, whose loss impairs the viability of triple-negative breast tumors 51 . To validate these results, we selected ten target genes with differing fold up and down regulations for further study by RT/qPCR using RNA isolated from 4TO7 cells (Fig. 8a,b). The results broadly confirm the gene www.nature.com/scientificreports/ regulatory events seen in RNAseq experiments. Taken together, the results are consistent with gene expression profiles generated in AM-193 or DK-406-treated cells being driven largely by their capacity to function as VDR agonists. They also suggest that exposure to hybrids will lead to profound changes in the cellular composition of the tumor microenvironment.

Methods
Reagents. 1α,25-Dihydroxyvitamin D3 (BML-DM200) was purchased from Enzo Life Sciences and was used at a final concentration of 100 nM. Suberoylanilide hydroxamic acid (SAHA, #10009929)) was purchased from Cayman Chemical and was used at a final concentration of 100 nM.  RNA extraction, reverse transcription and qPCR. RNA extraction was performed with the FavorPrep tissue total RNA mini kit (FAVORGEN Biotech Corporation-FATRK 001) as per manufacturer's instructions. cDNA was obtained from 1 µg of RNA using 5X All-in-One RT Mastermix (Applied Biological Materials (abm) Inc.-G485) and diluted 5 times. qPCR was performed with BrightGreen 2X q-PCR MasterMix (abm-Master-Mix-LR-XL) on a Roche LightCycler 96 system. Expression of targeted genes was normalized to 18S RNA. All primers are listed in Table S3.

Western Blotting and protein analysis. After treatments, 4T07 cells were solubilized in Lysis Buffer
(20 mM Tris, pH 8, 150 mM NaCl, 1% Triton X-100, 3,5 mM SDS, 13 mM deoxycholic acid) and proteins extracted were separated on a 4-15% Tris/Glycine/SDS gel (BIO RAD). A standard protocol was used for transfer and blotting. Anti-acetylated tubulin antibody was purchased from Sigma-Aldrich (#T7451). Anti-H3 antibody was clone96C10 (Cell Signaling Technology (CST); cat # 3638). Anti-acetyl-histone H3(Lys9) was 07-352 (Merck Millipore), and anti-acetyl-histone H3 (Lys27) was #8173 (CST). Anti-GAPDH antibody was purchased from Abcam (#ab8245) and the anti-mouse IgG HRP-linked secondary antibody was purchased from Cell Signaling Technology and were used at recommended concentrations. Signal was detected using Clarity ECL chemiluminescent substrates (Bio-Rad) and ChemiDoc Imaging System from the same company was used to quantify the intensity of bands. Changes in protein levels were quantified relative to control using Image Lab software (Version 6.0.1) after normalization to GAPDH. Western-blot images presented are three biological replicates.

Ingenuity pathway analysis.
Lists of differentially expressed transcripts with a p value of ≤ 0.05 and a twofold cut-off for each comparison were analyzed with Ingenuity Pathway Analysis or IPA (QIAGEN Inc., https:// www. qiage nbioi nform atics. com/ produ cts/ ingen uityp athway-analy sis) to identify enriched diseases and biological functions. IPA uses Fisher's exact test to determine a probability value to display the association between each gene in the list and IPA-curated pathways and biological functions. A log p-value cut-off of 1 www.nature.com/scientificreports/ considered statistically significant overrepresentation of genes in a disease or biological function. Furthermore, IPA employs a z-score of greater than 2 and less than − 2 to indicate significant predicted activation and inhibition states of functions, respectively.
Statistics. Student's t-test were performed using GraphPad software (Version 8.0.1).