Computational modeling and synthesis of Pyridine variants of Benzoyl-Phenoxy-Acetamide with high glioblastoma cytotoxicity and brain tumor penetration

Glioblastomas are highly aggressive brain tumors for which therapeutic options are very limited. In a quest for new anti-glioblastoma drugs, we focused on specific structural modifications of benzoyl-phenoxy-acetamide (BPA) present in a common lipid-lowering drug, fenofibrate, and in our first prototype glioblastoma drug, PP1. Here, we propose extensive computational analyses to improve selection of the most effective glioblastoma drug candidates. Initially over 100 structural BPA variations were analyzed and their physicochemical properties such as water solubility (−logS), calculated partition coefficient (ClogP), probability for BBB crossing (BBB_SCORE), probability for CNS penetration (CNS-MPO) and calculated cardiotoxicity (hERG), were evaluated. This integrated approach allowed us to select pyridine variants of BPA that show improved BBB penetration, water solubility, and low cardiotoxicity. Herein the top 24 compounds were synthesized and analyzed in cell culture. Six of them demonstrated glioblastoma toxicity with IC50 ranging from 0.59 to 3.24mM. Importantly, one of the compounds, HR68, accumulated in the brain tumor tissue at 3.7+/−0.5mM, which exceeds its glioblastoma IC50 (1.17mM) by over 3-fold.


Introduction
Glioblastomas are the most aggressive brain neoplasms with a dismally low 5 year patient survival rate of below 5% 1 . According to the World Health Organization (WHO) glioblastomas are classi ed as grade I and grade II (low-grade gliomas), grade III (anaplastic), and grade IV (glioblastoma) 2 . Current standardof-care therapies include maximal surgical resection, followed by radiotherapy plus concomitant and maintenance temozolomide (TMZ) 3 . In addition, a large variety of different genetic and epigenetic modi cations have been found in glioblastomas, among which p53, EGFR, PTEN, and IDH mutations are the most common [4][5][6][7][8][9] . However, these validated molecular targets, as well as immunotherapies, including immune checkpoint inhibitors 10 , tumor vaccines 11 , and chimeric antigen receptor T cell (CAR T) therapies 12 have all been extensively studied, but failed to signi cantly improve therapeutic outcome in glioblastoma patients.
There are several reasons why it is di cult to develop more effective glioblastoma therapy: (1) Glioblastomas are characterized by many dysregulated pathways that cannot be blocked simultaneously via a single therapy 13 : (2) Glioblastomas are highly in ltrating and heterogenous tumors that are very di cult to remove by surgical resection without compromising function of surrounding brain areas 14 ; (3) It is di cult to diagnose glioblastoma in its early stages, therefore, large highly in ltrating and vascularized tumors are often present at diagnosis 15 ; (4) Use of rodent syngeneic and patient-derived models are common to optimize clinical protocols. One major problem is that these experimental tumors are typically ~ 10 3 -10 4 smaller than human tumors, therefore data from drug delivery, drug retention, and tissue penetration experiments obtained from these small animal models are di cult to extrapolate to glioblastoma patients 16 ; and nally, (5) The blood brain barrier (BBB) prevents the majority of anticancer drugs from reaching tumor sites at clinically relevant concentrations, and current methods to enhance BBB penetration are not very effective for glioblastoma patients 17 .
One drug that readily crosses the BBB is temozolomide (TMZ). Upon oral administration, TMZ maximum plasma concentration can be reached in about one hour, and the elimination half-life is approximately 1.8 hours. Importantly, penetration e ciency of TMZ into the central nervous system (CNS) is experimentally estimated to be about 20% of plasma levels. This is important because, applying this estimate to calculate logBB (Brain-Blood Distribution) 18 this equation produces a value of -0.7, which indicates a high capability of the compound to cross the BBB. In spite of these positive features, TMZ-treated glioblastoma patients develop TMZ-resistance and recurrent tumors are practically incurable 19 . In addition, TMZ has been used in combination with other drugs, which enhanced its therapeutic effects.
One example is a combination of TMZ with lipid lowering drugs, including statins 20 and brates, like for instance feno brate (FF), which has strong anti-glioblastoma activity in cell culture, and in glioblastoma intracranial mouse models 21 . However, we have also found, that FF ability to cross the BBB is low, and the compound is quickly processed by blood and tissue esterases to form PPARa agonist, feno bric acid (FFA), which is no longer effective in triggering tumor cell death 22 . We made numerous modi cations to the FF chemical structure and selected our rst drug candidate, PP1 23 , which similar to FF blocks mitochondrial respiration and triggers a severe ATP depletion. With both substances, this is followed by phosphorylation/activation of AMP-activated protein kinase (AMPK -intracellular energy sensor), blockade of p70S6K phosphorylation (marker of active protein synthesis), activation of autophagy (p62 degradation), and extensive glioblastoma cell death 24 . In spite of these promising results, and in spite of the fact that we detected PP1 in the brain at therapeutically relevant concentrations (Supplementary materials, page 248) PP1 anti-glioblastoma therapeutic effects were only marginal when the treatment was applied to large intracranial tumors 24 . These data indicate that additional adjustments to the BPA scaffold, focused on improving compound cytotoxicity, BBB penetration and retention in the brain tumor tissue, are continuously needed.

Results And Discussion
In early stages of drug design it is important to evaluate relevant physicochemical properties of prospective drug candidates. For over two decades, Lipinski's rule of ve was the gold standard in drug design 25 . For drugs associated with the CNS, their ability to cross the brain blood barrier (BBB) is the most important characteristic and is not directly addressed by the Lipinski's rule of ve. Recently, two computational scores have been introduced to evaluate the probability of new drug candidates to penetrate the CNS. One is CNS Multiparameter Optimization (CNS-MPO) algorithm 26 , which uses 6 physicochemical properties [ClogP (calculated partition coe cient -lipophilicity), ClogD (calculated distribution coe cient at physiological pH (7.4) -lipophilicity), MW (molecular weight), TPSA (topological polar surface area), HBD (hydrogen bond donor at pH = 7), and pKa (-log acid dissociation constant)] for estimating the probability of entering the CNS. The values for CNS-MPO score range between 0 and 6, and the values ≥ 4.0 have been used as cut-offs for compounds with increased ability to penetrate the CNS 26 . Another is the Blood-Brain Barrier Score (BBB_SCORE) 27 . BBB_SCORE is based on ve physicochemical including number of aromatic rings, heavy atoms, MWHBN (a value comprising molecular weight, hydrogen bond donor, and hydrogen bond acceptor), pKa and topological polar surface area. Similar to CNS-MPO, BBB_SCORE also considers 4.0 as the cut-off for the acceptable BBB penetration 27 . Consequently, the CNS-MPO and BBB_SCORE serve as complementary algorithms in which CNS-MPO provides information regarding probability of the compound to be found in the CNS, and BBB_SCORE is oriented towards speci c physiochemical properties that increase probability of the compound to cross the BBB.
In regard to toxicity, one of the most common hurdle in drug testing is cardiotoxicity triggered by inhibition of cardiac potassium channel coded by the human ether-à-go-go-related gene (hERG). This toxicity test became a mandatory requirement for drug design and development, and can be calculated by the hERG algorithm 28 29 . Based on the compound structure, hERG algorithm gives a score which is indictive of the compound's inhibitory effect towards the hERG. In other words, for the drug to be considered safe, it's hERG IC50 should be signi cantly higher than its therapeutic IC50 28 .
In order to determine glioblastoma-speci c and therapeutically relevant values for all proposed algorithms/scores, including hERG, we have selected fteen drugs (see Supplementary Materials, pages 57-79) that are either currently used as glioblastoma drugs, or are in clinical trials for glioblastoma patients 30 . Results in Figure 1B show that six of these compounds ( Figure 1A) have relatively high water solubility (LogS<0), low lipophilicity (logD<3), acceptable values for CNS-MPO and BBB_SCORE (ranging between 3 and 5), and low cardiac toxicity (hERG£5.5). We used these critical values as a guideline for designing new anti-glioblastoma drug candidates, which are based on the BPA chemical structure present in a common lipid lowering drug, feno brate 21,22,31,32 , and in our rst glioblastoma prototype drug, PP1 23,24 .
In search for more effective benzoyl-phenoxy-acetamide (BPA) derivatives, we have decided to explore pyridine variants ( Figure 2). This is because the pyridine moieties are part of a diverse group of compounds with broad pharmacological applications 33,34 . In addition, it is well documented that pyridine based compounds have a high probability to penetrate the CNS 35 . Here, we focus on developing and testing ve pyridine base BPA variants, which are grouped based on the structural relationship between pyridine and amide moieties: (I) one with directly attached unsubstituted pyridine moiety (pyridine-BPA), (II) and one with a separation unsubstituted pyridine moiety with one carbon (methylenepyridine-BPA), (III) with two carbons (ethylenepyridine), (IV) a benzo fused pyridine (benzopyridine-BPA, and (V) a hydroxy substituted pyridine (hydroxypyridine-BPA) ( Figure 2).
These structural variations of BPA-pyridines were designed and tested using calculated values from four algorithms: cardiotoxicity (hERG) 36 ; brain penetration capability (CNS-MPO and BBB_SCORE), and water solubility (-logS). The values for best 24 BPA-pyridines are presented in Figure 3 (for complete list of calculated values see Supplementary Materials, pages 77-138). According to our initial calculations, compounds in Figure 3 have low calculated cardiotoxicity (all predicted hERG values are below 5.5), and acceptable water solubility (estimated -logS below 7.5). In addition, CNS-MPO scores for almost all BPA-pyrimidines are over 3, which indicates probability of brain penetration above 50%. Furthermore, these compounds have the BBB_SCORE between 4 and 5, which suggests high probability for crossing the BBB.
After determining that the proposed 24 BPA-pyridines have acceptable brain penetration ability, water solubility and low cardiotoxicity, we developed speci c preparation methods foreach compound. From our previous studies, we have demonstrated that the carboxylic group of feno bric acid (FFA) and its derivatives have exceptionally low reactivity toward nucleophilic acyl substitution. The obvious synthetic pathway for preparation of these compounds is by coupling FFA with corresponding amines 37 . There are two major problems in performing FFA coupling with aminopyridines: (a) FFA has exceptionally low reactivity due to steric constrains generated by two methyl groups located alpha to the carbonyl group, and (b) amino groups of aminopyridines are exceptionally weak nucleophile due to amino group electron delocalization through the pyridine ring. The di culty of performing this kind of coupling reaction was recently studied by others 38 . However, FFA can be converted into corresponding feno bric chloride ( Figure 4), which is su ciently reactive to aliphatic amines and activated anilines. But, coupling reactions with aminopyridines such as 4-aminopyridine is challenging at best. This is a major drawback to drug design because there are wide applications for pyridine scaffolds in medicinal chemistry 34 . For these reasons, we have decided to explore nucleophilicity of aminopyridines through their computational data. In molecules of similar structural framework, it is possible to compare frontier molecular orbitals to determine order of nucleophilic and electrophilic reactivity 39 . Accordingly, the molecule with highest HOMO (the highest occupied molecular orbital) are the most nucleophilic while the molecule with the lowest HOMO are the least nucleophilic via molecular comparison. We have used density functional theory to compute HOMO energies. Results provided in Supplementary Materials (page 247) indicate that aminopyridines should be less reactive (low HOMO) in comparison to aniline while aminophenol is more nucleophilic. Based on our calculations, the coupling of sterically hindered FFA through the reactive feno bric chloride (FFC) with 4-aminophenol should give a product with a highly isolated yield. Our previous studies 23,40 demonstrated that FFC is a viable, reactive FFA intermediate for amine coupling. For these reasons, we have developed a new synthetic methodology, which is shown in Figure 4, below.
There are two distinctive groups of HR compounds from the viewpoint of amine reactivity and their preparation (a) one with an aromatic heterocyclic ring directly attached to the nitrogen atom (HR66-HR70, HR82, HR83, HR88-HR90), and (b) one with separation of the aromatic heterocyclic ring by one or two aliphatic carbons (HR71-HR81, HR84-HR87). Preparation of the second group of compounds (with reactive amines) is straightforward by mixing FFC with corresponding amines in presence of a base. Because these amines are more nucleophilic than water and reactions are completed in several minutes at room temperature, reactions can be performed with an environmentally friendly base such as sodium carbonate in water. Isolated yields are nearly quantitative and this method (Method A) is well suited for small (milligrams) and large (hundreds of grams) scale synthesis. Also, this synthetic procedure does not require any special precautions or preparations. However, preparation of the rst group, with heterocyclic aromatic ring directly attached to the amino group, requires certain precautions. Because water is a better nucleophile than these amines', reaction must be performed in water free (dry) conditions. In addition, less nucleophilic amines are also prone to oxidation in basic conditions 40 . Because of this, such reactions should be carried out in an oxygen free atmosphere (Method B). Under dry conditions, a pyridine solution of a corresponding aminopyridine was mixed with anhydrous sodium carbonate and kept under nitrogen atmosphere overnight. A separate dichloromethane solution of FFC was also prepared in dry conditions under nitrogen atmosphere. The dichloromethane solution was slowly added to cold (~0-5°C) pyridine suspension of a corresponding aminopyridine and sodium carbonate under nitrogen atmosphere. Nearly quantitative yields were obtained with reactions performed at 0 -5°C for 3 hours, then at room temperature overnight, and nally, at 60°C for additional 3 hours. Product was isolated after solvent evaporation and water addition by simple ltration and extensive water washing.
This method gave products in high yield, more the 97% purity, and did not require additional puri cation by either extraction or chromatography. It is also applicable to milligram and multigram preparation scales for amines with wide ranges of reactivity ( Figure 4). showing that if the compound do not interact with cell membranes and has MPA lower than 60 Å 2 , it should be able to penetrate the CNS via passive diffusion [41][42][43] , therefore, MPA is considered as a better parameter than molecular weight in discriminating compound ability for entering the CNS 44 .
In the context of comparing compounds with high structural similarity, frontier orbital energies (HOMO and LUMO) can be used in determining which of the similar compounds may have a better chance for the BBB penetration 45 . In this regard, HR67 and HR68 ( Figure 5) both have low HOMO and LUMO energies, along with acceptable lipophilicity (ClogD), and very promising glioblastoma IC50 values (0.59 and 1.17mM, respectively), indicate that these two compound could become leading candidates for the brain tumor drug development.
Next group of compounds belong to methylenepyridine-BPA ( Figure 6). All of these compounds show anti-glioblastoma activity at 25µM. Their estimated distributions (ClogD) are relatively high indicating high lipophilicity of the compounds. MPA values are below 60 suggesting that there is no obstacle to CNS penetration in regard to molecular size. Polarizability (PL) is between 40 and 50 indicating that these molecules can adapt to the binding area of a biomolecule through complementary polarization 46 . Considering computed frontier orbital energies (LUMO and HOMO), HR74 should have the best binding ability. Considering experimentally determined cell viability, all HR compounds in Figure 6 are highly cytotoxic at 25 mM except HR75. IC50 values for the most promising compounds in this group, HR73 and HR76, are 3.24 and 2.87 mM, respectively.
Next, we asked if the compound activity would change by adding ethylene linker to ethylenepyridine-BPA or by increasing the pyridine's molecular delocalization in benzopyridine-BPA ( Figure 7). As expected, by increasing the number of carbon atoms either by adding methylene group or additional aromatic ring, lipophilicity increased noticeably (for instance, ClogD for HR83 is 7.29). As size of the molecule increases, MPA values follow, suggesting that HR81, HR82, and HR83 may have a low probabilities of penetrating the CNS. Considering the energies of both frontier molecular orbitals from three similar compounds (HR78, HR79, and HR80) HR80 should be the most active. Indeed, obtained cell viability (CV) data for these compounds correlated with computational data and suggest that best drug candidate from this group is HR80. Although both benzopyridine-BPAs (HR82 and HR83) have encouraging IC50 values, 1,4 and 2.75 mM, respectively, their computed physical properties such as ClogD and MPA suggest that there is very low probability for these compounds to penetrate the CNS, further supporting HR80 as the best glioblastoma drug candidate in this group.
We have also explored the importance that chirality may have on activity of these compounds (Figure 8).
For instance, HR84 and HR85 are structural isomers of HR78 that show moderate cytotoxicity at 25 µM (CV = 26.33). There are noticeable differences between the two stereoisomers: R isomer HR84 being three times more cytotoxic than S isomer HR85 (Figure 8). On the other hand, there is no difference between racemic HR86 (both R&S) and optically pure R isomer HR87. This nding is reasonable because of the pyridine nitrogen proximity to the chiral center. One could argue that pyridine nitrogen binding is sterically diminished in HR85 in comparison with HR84. Because of the position of pyridine nitrogen regarding chiral center, steric difference is diminished in HR87, which associates with improved cytotoxicity. The hydoxypyridine-BPAs can exist in both hydroxypyridine and amide form with the amide form being preferable by 8.97 kcal/mol according to the DFT wB97X-D/6-31G* computational method. Computational studies as well as NMR spectroscopy indicate that HR88 is in hydroxypyridine form while HR89 and HR90 are in their amide form.
Of the three hydroxypyridine-BPAs, only computed data for HR89 and HR90 can be compared because they are in amide form while HR88 is in the hydroxypyridine form. All computed parameters suggest that both HR89 and HR90 should easily penetrate the CNS. However, computed frontier orbital energies for HR90 are lower indicating it should be more effective in penetrating the BBB. In addition, HR90-induced glioblastoma cytotoxicity is nearly 10-fold greater in comparison to HR89, making HR90 better drug candidate. In conclusion, HR87 and HR90 are the best drug candidates in this group with corresponding, glioblastoma-speci c IC50 values, 5.38 and 2.05 mM, respectively ( Figure 8C).
When pursuing new drug candidates, it is very important to envision possible metabolites and estimate their physicochemical properties, including toxicity 47 . We used computational methods to generate metabolites of all studied compounds then evaluated their toxicity, solubility, lipophilicity and CNS penetration using the BioTransformer method 48 . Only results for our best drug candidates, HR67 and HR68, are reported here (Figure 9). Reaction types involved in Phase I metabolism are hydrolysis, oxidation, and reduction 49 . All predicted metabolites of HR67 and HR68 have better or comparable computed water solubility, as well as computed abilities to penetrate the CNS. Importantly, all these metabolites are predicted to be relatively safe (hERG values range 5.76 to 4.95) with the exception of amide hydrolysis, which may trigger some cardiac toxicity and should be further analyzed (hERG=3.91) ( Figure 9).
Our data presented in Figures 1 -9 allowed us to perform high throughput and unbiassed selection of BPA-based compounds, which have a good chance of becoming glioblastoma drug candidates. Two pyrimidine variants of BPA, HR67 (PP23) and HR68 (PP21), were subsequently tested for their ability to penetrate arti cial BBB model membranes ( Figure 10A), and for tissue distribution (HR68 only), which was evaluated following intraperitoneal (ip) HR68 delivery into mice bearing patient-derived intracranial glioblastoma ( Figure 10B). Although CNS-MPO scores for HR67 (3.71) and HR68 (3.71) are slightly lower compared to the CNS-MPO of our prototype drug, PP1 (CNS-MPO = 3.9) 23,24 , these two compounds can cross the BBB model membrane ( Figure 10B), and importantly, we have detected HR68 in the brain tumor tissue at concentrations over 3-fold higher than glioblastoma-speci c IC50 of HR68 (1.17mM) ( Figure  10C).
In conclusion, introducing pyridine moieties to the BPA scaffold improves chemo pharmacological properties of new drug candidates. In particular, water solubility and predicted CNS penetration are higher than in previously studied alkyl and phenolic derivatives of BPA 23,51 . It was also demonstrated here that properly positioned pyridine moiety in respect to BPA increased anti-glioblastoma potency of these compounds with glioblastoma-speci c IC50 values being close to 1 µM. Importantly, these speci c modi cations, which increased molecular exibility and improve water solubility of the compounds, was achieved without compromising glioblastoma speci c cytotoxicity. In addition, stereochemistry of the chiral center close to pyridine is important because different three dimensional orientation of pyridine nitrogen can change compound interaction with targeted biomolecules. in dichloromethane (1L). Prepared FFC (0.3 mmol) was slowly added dropwise over a period of 45 minutes at ice-water bath temperature into a magnetically stirred mixture of sodium carbonate (106 g; 1 mol) in water (500 ml) and 1-(pyridine-4-yl) ethanamine (24.4 g; 0.2 mmol) in tetrahydrofuran (500 ml). After addition was complete, the resulting reaction mixture was stirred at room temperature overnight. The reaction mixture volume was reduced by 75% via solvent evaporation under air ow (produced by air pump). The resulting white suspension was mixed with water (500 ml) and the insoluble product was separated by ltration, washed with water (10x50 ml), and dried at 50°C overnight. The isolated yield was 97% (82 g) of pure product. The ltrate was acidi ed with hydrochloric acid to pH = 2. The resulting white solid precipitate was separated by ltration, washed with water (10x20 ml) and dried at 50°C overnight to give 34.2 g (95% recovery) of feno bric acid. 1  Method C. Small scale preparation applicable to all HR66-HR90. Preparation of 2-(4-(4chlorobenzoyl)phenoxy)-2-methyl-N-(2-oxo-1,2-dihydropyridin-3-yl)propanamide (HR89). Freshly prepared dichloromethane (5 ml) solution of feno bric acid chloride was made from feno bric acid (80 mg; 0.25 mmol) and oxalyl chloride (1 mmol) as described above. These were added, under a nitrogen atmosphere while stirring, to a pyridine (10 ml) -tetrahydrofuran (10 ml)-sodium carbonate (212 mg; 2 mmol) of 3aminopyridin-2(1H)-one (27.5 mg; 0.25 mmol) solution. The resulting mixture was stirred at room temperature in the nitrogen atmosphere for 3 hours, followed by stirring under the nitrogen atmosphere at 60°C for an additional 3 hours. After cooling to room temperature, the solvent was evaporated under air ow (produced by air pump) at room temperature yielding a solid residue. This solid residue was mixed with dichloromethane (30 ml) and? water (100 ml). The water layer was discarded, and the organic layer was washed with water (3x100 ml), 5% sodium carbonate (3x100 ml), and dried over anhydrous sodium carbonate. The drying material was separated by ltration. The volume of the ltrate was reduced to ~ 2 ml, then hexanes (~10 ml) were added. The resulting solution was left uncovered at room temperature, allowing the solvent to slowly evaporate. The resulting white crystalline product was separated by ltration, washed with hexane (3x3ml) and air-dried overnight. Isolated yield=93% (95 mg). 1  Cell culture and viability assays. Human glioblastoma LN-229 cells (ATCC CRL-2611) were maintained as a semi-con uent monolayer culture in DMEM with 1 g/L glucose, sodium pyruvate and L-glutamine (Corning), supplemented with 10% heat-inactivated FBS (Gibco) and P/S (50 units/mL of penicillin and 50 µg/mL of streptomycin) at 37°C in a 5% CO 2 atmosphere. Prior to treatment with HR compounds, cells were plated in 96-well plates (BD Falcon) at an initial density of 2 x 10 4 cells/cm 2 . Twenty-four hours after plating, stock solutions of HR compounds were prepared in DMSO, diluted in cell culture medium and added to previously plated cells in triplicate for every experimental condition ( nal concentration 25 µM). DMSO (0.5%) was used as vehicle control. After 72h incubation, an MTT assay was performed to measure cell metabolic activity (surrogate for cell viability). Following a 1.5 h incubation with 0.5 mg/ml MTT in serum free low glucose DMEM, the resulting formazan crystals were dissolved in 5mM HCl in isopropanol and the absorbance read at 540 nm. Data represent mean values expressed as the percentage of vehicle control ± SD. Phase contrast images of treated cells were taken 72 hours after treatment with HR compounds using a BZ-X800 uorescence microscope (Keyence) equipped with a 20x objective. The drug dose resulting in 50% inhibition of cell metabolic activity (surrogate for cell viability) was measured using MTT assay, at 72-hour time point, and half maximal inhibitory concentration (IC50) was calculated using GraphPad Prism 8.

Methods
In vitro model of the Blood Brain Barrier (BBB). The BBB was re-created in vitro using a modi ed protocol provided by Stone et al 53 . Brie y, 24-well transwell inserts (Falcon, catalog number 353096) were coated with 10mg/cm 2 of Collagen Type IV (Sigma) for 24 hrs at 4°C. Inserts were washed with sterile water and air-dried for 2 hrs. Next, the inserts were coated with 2mg/cm 2 poly-L-lysine (ScienCell) for 1 hr at 37°C, then washed twice with sterile H 2 0 and air-dried for 2 hrs. Primary human astrocytes (1.5x10 5 ) and 3x10 4 primary human pericytes (both ScienCell) were resuspended in 25ml of astrocyte medium and pericyte medium (ScienCell), respectively, then combined in a 1:1 ratio for 50ml total volume. Dried, coated inserts were turned upside down such that the basolateral surface was exposed at the top, and 50ml of the cell mixture was added to the membrane, covered with the plate lid, and incubated for 2 hrs at 37°C to allow cell adhesion. Any medium remaining on top of the membrane was carefully removed before returning inserts to their upright position with the apical surface facing upward, as they were placed in a 24-well plate containing 500ml per well of astrocyte/pericyte medium (1:1). An additional 300 mL of medium was added to the apical compartment. Four days after plating, the apical compartment medium was removed, and 3.75x10 4 of telomerase-immortalized vain endothelial cells (TIVE; provided by Dr. Rolf Renne) in 50ml of TIVE medium 54 were added and incubated for 5 hrs at 37°C to allow cell attachment, followed by the addition of an extra 250ml of TIVE medium. Half the volume of the corresponding media in the lower and upper compartment was replaced with fresh media every third day. Ten days after initial plating, trans-endothelial electric resistance (TEER) was measured using a EVOM 2 meter with a STX3 electrode (World Precision Instruments). The ability of selected HR compounds to pass through the in vitro BBB was tested using inserts with effectively reconstructed BBB as con rmed by TEER values 53,55 .
High Performance Liquid Chromatography (HPLC)-based detection of selected HR compounds: Following TEER measurement, the medium from the apical compartment (insert) of the in vitro BBB model (Fig. 10A) was replaced with 350mL of fresh TIVE medium containing corresponding compounds [HR67 (PP23), HR68 (PP21), both used at 25mM. In addition, 25mM feno brate (FF), which does not cross the BBB 22 ,was used as a negative control, and 50mM caffeine was used as a positive control 56 . Plates containing the inserts were returned to the incubator (37°C, 5% CO 2 ), and after 3 hrs of incubation, conditioned media from the well and insert (Fig. 10A) were collected. The aliquots (100μl) of the collected samples were subsequently mixed with 100μl of 100% acetonitrile, centrifuged (16.000 rpm at 4°C for 10 min) and supernatants collected for HPLC -based detection of HR67 and HR68.
HPLC analyses were performed using an UltiMate 3000 system (Thermo Scienti c) equipped with an analytical YMCbasic, 3µm, 150 x 4.6 mm column (octyl silane C8; YMC America, Inc.). Isocratic elution of the compounds was performed using a mobile phase composed of solvent A (50 mM acetic acid in dH 2 O) and solvent B (acetonitrile) mixed at predetermined ratios for each compound ( Table 1). All separations were carried out with 5μl sample volume at a ow rate of 1 ml/min, at 25°C. The concentration of each compound was calculated using serial dilutions of the known concentration of the compound separated at the same run with experimental and control samples. After separation, integrated areas under the peak were used to prepare calibration curves and to determine concentration of the compounds. kidneys, spleen, heart, brain and brain tumor were subsequently collected, solid tissues were washed from blood in PBS, and placed on ice for an immediate sample preparation for HPLC analysis (see above).
Tissues were prepared for HPLC by mixing 150 μl of sample tissue (~120 mg) that had been mixed with 3 volumes of Methanol : H 2 O mix (4:1), well-blended using TissueRuptor II (Qiagen), and centrifuged at 15,000g for 10 minutes at 4 o C. Supernatants were collected in 1.5 ml Eppendorf tubes and incubated at 95 o C for 3 min. Following ash cool on ice, samples were centrifuged again at 15,000g for 10 minutes at  The pyridine region of BPA skeleton was selected for modi cation (circle) in search of the optimal antiglioblastoma drug.      Mice were treated intraperitoneally (ip) with HR68 diluted in 20% cyclodextrin at 15 mg/kg/day and the levels of HR68 in the blood, heart, liver, kidney, spleen, lung, brain and in brain tumor (BT), were evaluated by HPLC, as we previously reported 21,24 . Data represent average values with standard deviation (n=3). Please note that average glioblastoma IC50 for HR68 is 1.17mM, and we detected 3.7+/-0.5mM of HR68 in the brain tumor tissue.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download. SupplementaryMaterialscopy.pdf