A selective and orally bioavailable VHL-recruiting PROTAC achieves SMARCA2 degradation in vivo

Targeted protein degradation offers an alternative modality to classical inhibition and holds the promise of addressing previously undruggable targets to provide novel therapeutic options for patients. Heterobifunctional molecules co-recruit a target protein and an E3 ligase, resulting in ubiquitylation and proteosome-dependent degradation of the target. In the clinic, the oral route of administration is the option of choice but has only been achieved so far by CRBN- recruiting bifunctional degrader molecules. We aimed to achieve orally bioavailable molecules that selectively degrade the BAF Chromatin Remodelling complex ATPase SMARCA2 over its closely related paralogue SMARCA4, to allow in vivo evaluation of the synthetic lethality concept of SMARCA2 dependency in SMARCA4-deficient cancers. Here we outline structure- and property-guided approaches that led to orally bioavailable VHL-recruiting degraders. Our tool compound, ACBI2, shows selective degradation of SMARCA2 over SMARCA4 in ex vivo human whole blood assays and in vivo efficacy in SMARCA4-deficient cancer models. This study demonstrates the feasibility for broadening the E3 ligase and physicochemical space that can be utilised for achieving oral efficacy with bifunctional molecules.

Heterobifunctional degrader molecules -also known as proteolysis-targeting chimeras (PROTACs) -target 23 disease-causing proteins for destruction. They function by binding to both an E3 ligase and to the target protein. 24 The induced proximity results in subsequent ubiquitination of the target protein, earmarking it for degradation by 25 the proteasome. In cells, their mode of action enables degraders to achieve levels of target selectivity, breadth of 26 target scope and efficacy not attainable with a classical inhibitor 1,2 . Due principally to the convenience of 27 administration, oral dosing regimens dominate small molecule therapeutic delivery, however the design of orally 28 available degraders is challenging due to their larger size and related physicochemical properties. To date, the 29 clinical translation of orally active degraders has been confined to the use of a single E3 ligase -Cereblon (CRBN), 30 greatly limiting the potential therapeutic scope of PROTACs 3,4 . Here we show for the first time that orally 31 bioavailable PROTACs can be developed that utilise other E3 ligases, in this case the von Hippel-Lindau 32 (VHL)/ElonginB-ElonginC (VCB) complex, which is recruited by a ligand of larger molecular weight than the 33 immunomodulatory drugs (IMiDs) that bind CRBN. We found that design of linker composition and exit vector 34 placement could be guided and rationalised by ternary co-crystal structures, yielding molecules which exhibit high 35 potency and suitable pharmacokinetic properties to translate to oral in vivo efficacy. Furthermore, our lead 36 molecule ACBI2 demonstrates strong selectivity for the BAF (SWI/SNF) complex ATPase SMARCA2 over its highly 37 similar paralogue SMARCA4 in human whole blood and consistent preferential degradation of SMARCA2 in all cell 38 lines tested. This permits pharmacological evaluation of the synthetic lethality concept of selectively targeting 39 SMARCA2 in SMARCA4-deficient cancers in vivo and in vitro 5-7 . We qualify ACBI2 as an orally bioavailable 40 SMARCA2 degrader that will be made freely available to the community. We anticipate that our results and the 41 methodologies employed will provide a blueprint to arrive at oral efficacy of other E3-recruiting degraders. 42 43 Discovery of a novel potent SMARCA2/4/PBRM1 binder to enable targeted protein degradation . 1 PROTACs are often made by the up-cycling of existing protein of interest (POI) binders, allowing quick 2 generation of protein degrader tool compounds. However, the optimisation of physicochemical properties to turn 3 in vitro degrader tools into in vivo drugs often remains challenging and can prohibit the progression of the 4 compounds into the clinic. For the SMARCA2/4 bromodomains (BDs), two sub-micromolar binders have been 5 described in the literature. PFI-3 was reported as a bona fide SMARCA2/4 BD-binding tool compound (Figure 1a), 6 albeit with modest affinity and a latent risk for unfavourable chemical stability 8 . In addition, Genentech reported 7 a phenol-amino-pyridazine derivative (GEN-1) with attractive stability, physicochemical properties and affinity 8 (Figure 1a) 9 . We successfully used this motif to generate SMARCA2/4 degraders 10 , but the polarity of the binding 9 motif prohibited subsequent optimisation towards sufficient oral bioavailability for this VHL-based PROTAC. 10 Driven by the multiple advantages of orally bioavailable drugs, such as a high patient acceptance, the possibility 11 of non-sterile self-administration, cost-effectiveness and ease of large-scale manufacturing 11 , we set out to obtain 12 orally active SMARCA2 degraders. 13 As a first step, we elected to discover a novel SMARCA2 bromodomain binder, incorporating only the absolute 14 minimum of hydrogen bond donors, a high degree of rigidity and distinct and well-defined exit vectors. In addition, 15 a reliable synthetic route, with the possibility for late-stage functionalisation, was considered essential. Inspired 16 by Sutherell et al. 12 , who previously identified binders for the bromodomain (BD) of PBRM1 (PBRM1 BD5 ), which 17 shares a high degree of similarity to the bromodomain to SMARCA2, we first characterised the molecular 18 interactions that are mandatory for BD binding using a high-resolution crystal structure of "compound 26" 12 (PDB: 19 5FH7, Figure 1a). We found the halogen bond to the Met371 PBRM1 backbone (BB) carbonyl and the hydrogen 20 bonding interaction between Asn739 PBRM1 and the quinazolinone core to be indispensable. Incorporating insights 21 of halogen bond optimisation from an alternative benzoxazinone lead series (Figure 1b) to the quinazoline core 22 led to the design of scaffold compound 1 (Figure 1c). We also evaluated the possibility to replace the halogen 23 bonding interaction by a hydrogen bonding interaction in compound 2, however this was detrimental to binding 24 affinity (Figure 1c). Next, we hypothesised that placing a basic centre close to Glu1417 SMARCA2 could improve the 25 binding affinity and would also balance the solubility of the compound. The attachment of a piperidine was 26 superior to other linear and cyclic basic moieties, such as piperazines or amino-carbocycles, and led us to 27 compound 3 as a novel, purposefully designed SMARCA2/4 BD binding motif for PROTAC generation (Figure 1c). 28 The expected binding mode of compound 3 was confirmed by solving a co-crystal structure of its nor-methyl 29 analogue compound 4 with SMARCA2 BD (Figure 1d, PDB: tbd, 1.32 Å resolution, see Supplementary Table 1)  30 revealing that the key interactions of the quinazoline core towards Asn1464 and Leu1456 as well as Glu1417 via 31 the basic nitrogen of the piperidine are addressed in the SMARCA2 BD . Finally, SPR binding kinetics of compound 3 32 (SPR SMARCA2 BD KD = 46 nM) revealed a similar behaviour as previously reported for GEN-1 (SPR SMARCA2 BD KD = 33 206 nM), however with significantly reduced hydrogen bond donor count. (Figure 1e, Extended Data Table 1), thus  34 offering a superior starting point for the generation of orally available PROTACs. 35 36 1 Figure 1. Biophysical and structural characterisation of compound 3 and previously disclosed SMARCA 2 bromodomain binders. a. Comparison of previously disclosed SMARCA binders with sub-µM binding affinity: PFI-3 3 8 , GEN-1 9 , "compound-26" 12 . b. Optimisation of the halogen bonding interaction to Leu1456 based on an 4 alternative benzoxazinone lead series. c. SAR of novel SMARCA binding scaffold with reduced hydrogen bond 5 donor count. d. Superposition of the SMARCA2 BD : compound 4 complex (green, a close analogue of compound 3) 6 with PDB: 6HAZ (yellow, GEN-1) highlighting key interactions towards Asn1464, Leu1456 and Glu1417. e. SPR 7 sensorgrams for binding of compound 3/GEN-1 to SMARCA2 BD ; mean values reported with standard deviation 8 (n=3 independent experiments). 9 10 Identification of an in vivo active SMARCA2 degrader via exit vector hopping 11 We previously reported potent dual degraders of SMARCA2/4, utilising a phenolic exit vector from the 12 VHL ligand, and hypothesised that PROTACs based on our new ligand compound 3 would offer new opportunities 13 for in vivo degrader optimisation 10 . PROTACs with PEG-and alkyl-based linkers showed moderate target 14 Asn1464 Leu1456 degradation in RKO cells, with 27 -75% maximal degradation (Dmax) for SMARCA2 (Extended Data Table 2). 1 Notably, compound 5 demonstrated partial degradation of SMARCA2 (DC50 = 78 nM, Dmax = 46%) while sparing 2 SMARCA4 completely (Figure 2a). Kinetic experiments demonstrated that compound 5 did not show degradation 3 of SMARCA2 at 4 hours, suggesting a slow rate of degradation (Extended Data Table 2). Rapid degradation kinetics 4 may reduce the need for prolonged in vivo exposure. We have previously shown that E3 Ligase : PROTAC : POI 5 ternary complex stability can impact the rate of degradation 13 . To support our understanding of the 6 thermodynamics of ternary complex formation in this series, we established SPR and TR-FRET assays (Extended 7 Data Tables 1 and 3), and solved a high resolution co-crystal structure of the VCB : compound 5 : SMARCA2 BD 8 complex (PDB: tbd, Figure 2b, see Supplementary Table 2). Consistent with poor target degradation at 4 hours, 9 the data show moderate ternary complex binding affinities and a limited buried surface for this ternary complex 10 (1837 Å 2 ). Whilst the position of SMARCA2 BD may be influenced by crystal contacts formed in the closely packed 11 crystal lattice (Extended Data Figure 1a-e), the structure indicates a de novo protein-protein interaction (PPI) 12 between Asn67 of VHL and Gln1469 of SMARCA2. It is noteworthy that SMARCA4 features a leucine residue in 13 this position and is thus less capable of forming such a PPI, offering a possible rationale for the observed selectivity 14 of compound 5. In pursuit of more stable complexes that could result in rapid degradation based on the co-crystal 15 structure, we switched the VHL exit vector from the phenolic to the benzylic position (Figure 2c), which was 16 enabled by new synthetic methods (see SI). This resulted in compound 6, a fast and potent dual degrader of 17 SMARCA2/4 (SMARCA2 DC50 = 2 nM, Dmax = 77% ; SMARCA4 DC50 = 5 nM, Dmax = 86%, in RKO cells after 4 hours) 18 ( Figure 2c, Extended Data Table 3) that displayed the expected antiproliferative effect (IC50 = 2 nM, Extended Data 19 Table 4) in a SMARCA4 deficient lung cancer cell line, NCI-H1568. We also observed concurrent degradation of 20 PBRM1 and could rescue SMARCA2 degradation by inhibition of the VHL-HIF1-α interaction using VH298, 21 neddylation inhibition using MLN4924 or proteasome inhibition by MG132 (Extended Data Figure 2a-c). We solved 22 the ternary complex crystal structure of VCB : compound 6 : SMARCA2 BD , in which the compound adopts a 23 different binding pose to that of compound 5 with an increased buried surface area of 2050 Å 2 (Figure 2d), 24 consistent with an observed increase in ternary complex half-life and ternary binding affinity in SPR and TR-FRET 25 assays, compared with compound 5, offering an explanation for the differential cellular SAR (Figure 2a, Extended 26 Data tables 1 and 3). PK studies revealed that the improved microsomal stability of compound 6 compared with 27 compounds from the phenolic series (Extended Data Table 4) translated into a low clearance of 7 mL/min/kg in 28 mouse, and the compound was quantitatively bioavailable upon subcutaneous administration (Extended Data 29 Table 5). Single dose subcutaneous treatment of an NCI-H1568 tumour xenograft model with compound 6 reduced 30 median SMARCA2 levels in tumours by 82% relative to control-treated samples at six hours after treatment, with 31 slight recovery of the signal observed after 48 hours treatment, to a median decrease of 69% (Figure 2e). This 32 translated in significant tumour growth inhibition (TGI) in two different treatment regimens that were both well 33 tolerated (Figure 2f, Extended Data Table 5). In tumour samples collected at the end of the study, compound 6 34 treatment resulted in undetectable levels of SMARCA2 as assessed by IHC (Extended Data Figure 2d) A structure guided approach to discover selective, orally bioavailable VHL PROTACs 20 Based on our observations that the short three carbon linker used in compound 6 efficiently forms a high 21 affinity ternary complex leading to potent degradation of SMARCA2/4, we elected to focus on a small set of alkyl 22 and ether based analogues with the objective of improving selectivity and oral bioavailability. Linker elongation 23 and branching led to compounds, such as compound 7 and compound 8, which show remarkable selectivity for 24 SMARCA2 over SMARCA4 degradation (Extended Data Table 2; Figure 3a). Linker branching by installation of an 25 additional methyl group as in compound 8 further improved the permeability in a Caco-2 cell assay (Extended 26 Data Table 4). To gain a better understanding for the molecular basis of this selectivity, we again turned to crystal 27 -69% compound 6 5 mg/kg structure analysis. We were able to solve the ternary crystal structure of compound 9, a close analogue of 1 compound 8, that only differs in the VHL binder site where the fluorine is replaced by a dimethyl amino group 2 (Figure 3a, 3b). This ternary structure revealed that an extensive network of de novo electrostatic interactions 3 between SMARCA2 and VHL was formed, leading to the formation of a ternary complex significantly different from 4 the previously observed ones (Extended Data Figure 3a). Furthermore, the SMARCA2-specific residue Gln1469 5 was involved in VHL : SMARCA2 BD interactions, as previously observed for the SMARCA2-selective molecule 6 compound 5, albeit in this case in the context of a different overall arrangement of the two proteins. In SMARCA2, 7 Gln1469 positively interacts with VHL residues Phe91 and Asp92 via hydrogen bonds, an interaction that cannot 8 occur in SMARCA4 that harbours Leu1545 instead of Gln1469. In summary, linker elongation with an oxygen and 9 linker branching gave rise to compounds that were selective towards SMARCA2, a preference rationalised by 10 ternary complex structures. However, despite good microsomal stability and measurable Caco-2 permeability 11 (Extended Data Table 4), the compounds showed a high efflux ratio, preventing oral bioavailability. 12 Learning that linker elongation and branching improves selectivity and permeability, we hypothesised that this 13 should also apply to the more lipophilic all-carbon series. Linker elongation from three to five carbon atoms also 14 resulted in a slight SMARCA2/4 selectivity improvement, by two-to three-fold within the all-carbon linker series, 15 as observed by comparing compound 6 with compound 10 (Extended Data Table 2). Linker elongation led to an 16 improved permeability for compound 10, which, due to its good microsomal stability and moderate solubility 17 (Extended Data Table 4), constituted the first orally bioavailable SMARCA2 VHL PROTAC (Figure 3c, Extended Data 18 Table 5). We tested compound 10 in an NCI-H1568 xenograft study, treating mice orally with 100 mg/kg, and 19 evaluated SMARCA2 levels in viable tumour tissue 48 hours after treatment start. A median decrease of 80% 20 compared to vehicle control-treated tumours was detected (Figure 3d). In subsequent studies, mice were treated 21 orally with 30 mg/kg or 100 mg/kg compound 10 daily. A significant TGI of 64 % within the 30 mg/kg and 76% 22 within the 100 mg/kg treatment group could be reached at day 17 of the experiment (Figure 3e). The compound 23 was well tolerated in both dose groups as assessed by body weight changes (Extended Data Figure 3b). At the end 24 of the study, SMARCA2 levels were undetectable by IHC in most treated tumour samples from both groups 25 (Extended Data Figure 3c). Together, compound 10 achieved significant in vivo activity and oral bioavailability, 26 unprecedented for VHL-based PROTACs. the key PPIs between VCB and SMARCA2 BD /SMARCA4 BD , highlighting the selectivity-inducing hydrogen bonding 7 between Gln1469 of SMARCA2 BD and VCB vs. Leu1545 in SMARCA4 BD . c. Plasma profiles of compound 10 in mouse 8 after administration of 5 mg/kg i.v. or 30 mg/kg p.o. The oral bioavailability of compound 10 was 3%. d. NCI-H1568 9 tumour bearing mice were treated orally with 100 mg/kg compound 10 or vehicle control (n=4 animals per group) 10 and tumours were collected 48 h after treatment. The average tumour size at treatment start was ~290 mm 3 . 11 SMARCA2 levels in viable tumour tissue were determined using DAB-based IHC staining. Each datapoint 12 represents the background-normalised DAB optical density (OD) within the viable tumour area of one tumour 13 section, corresponding to one individual tumour. Mean OD levels and standard deviations are indicated in the 14 graph. Numbers above the data points represent the median levels of SMARCA2 signal decrease relative to vehicle 15 control-treated samples. e. NCI-H1568 tumour bearing mice were treated orally with compound 10 at 30 or 16 100 mg/kg daily. Average tumour volume at the beginning of treatment was ~220 mm 3 . At day 17 after treatment 17 start, TGI was 76% for the 100 mg/kg and 64% for the 30 mg/kg treatment group (adj. p-value = 0.0009 for either 18 regimen vs. control, U-test with Bonferroni-Holm correction). Values represent the mean of 10 animals per group, 19 error bars indicate standard deviation. 20

21
ACBI2 is an orally active degrader that selectively degrades SMARCA2 over SMARCA4 22 It has been shown that despite having properties such as molecular weight >1000 Da, 2D topological polar 23 surface area and number of rotatable bonds well beyond those deemed optimal for passive cell permeability, 24 PROTACs can adopt more compact 3D conformations that yield 3D polar surface and radius of gyration more 1 consistent with that required for permeability 14,15 . To date, this has been translated into oral bioavailability only 2 for CRBN-based PROTACs 16-18 and led to the perception that it is more challenging to achieve for VHL-based 3 degraders 19 . We hypothesised that if we could enable more compact conformations of compound 10 via changes 4 to the linker region, we could enhance absorption, reduce efflux, and ultimately enhance oral bioavailability. We 5 therefore incorporated an additional methyl group on the C5 linker of compound 10, a change which on the 6 analogous ether series (see compound 7 vs. compound 8) gave improvements in Caco-2 permeability. This 7 modification yielded ACBI2, a highly potent VHL PROTAC (IC50 = 7 nM), which degrades SMARCA2 with a >30-fold 8 window over SMARCA4 in RKO cells (SMARCA2 DC50 = 1 nM, SMARCA4 DC50 = 32 nM) (Extended Data Table 2). 9 The branched linker significantly reduced ACBI2 efflux, which directly resulted in an improved oral bioavailability 10 of 22% (Figure 4a). 11 Molecular dynamics (MD) simulations and NMR studies comparing compound 10 and ACBI2 were performed to 12 provide a molecular basis for the link between conformational restraint and reduced efflux ratio. Conformational 13 ensembles from simulations indeed showed a trend towards collapsed structures with lower radius of gyration 14 leading to lower free energies for ACBI2. Consequently, lower polar surface area values tended to be favoured 15 within the conformational ensemble of ACBI2 (Extended Data Figure 4a). NMR studies were carried out to evaluate 16 long-range nuclear Overhauser effects (NOEs) (Figure 4b). We determined long range NOEs for ACBI2 that were 17 not detectable in compound 10. Furthermore, under identical experimental conditions, the sign of the NOE 18 crosspeaks was different for the two compounds, indicating a different degree of mobility and therefore 19 compactness of the compounds, with ACBI2 having the more compact structure (see SI Supplementary Figure 1). 20 Encouraged by the improved oral bioavailability and selectivity profile of ACBI2, we characterised the 21 compound in more detail in vitro. A panel of cell lines showed varying levels of sensitivity to ACBI2, correlating 22 with genetic dependency on SMARCA2 due to mutation or lower expression of SMARCA4 (Figure 4c). Accordingly, 23 ACBI2 treatment caused rapid and complete degradation of SMARCA2 in two sensitive cell lines (A549 and NCI-24 H1568; Extended Data File Proteomics). Interestingly, we observed that the extent of selectivity of SMARCA2 degradation over SMARCA4 32 varied in cell lines expressing both proteins (HCT116 and RKO, Extended Data Figure 5a). We investigated if this 33 might be correlated with differences in half-life or re-synthesis rates of either SMARCA2 or SMARCA4, and indeed 34 observed a trend towards higher selectivity in RKO, the cell line with shorter half-lives and faster re-synthesis of 35 both SMARCA2 and SMARCA4 (Extended Data Figure 5b and 5c). We tested ACBI2 selectivity in five additional cell 36 lines and confirmed preferential degradation of SMARCA2 over SMARCA4 in all of them (Extended Data Figure  37 5d). We cannot formally rule out other differences between these cell lines as contributors to differential 38 selectivity (e.g. proliferation rate, mutations or expression levels of SMARCA2, SMARCA4 and VHL or ratios 39 thereof), but do not observe obvious trends towards either of those in this small cell line panel. 40 We went on to test ACBI2 in vivo and observed dose-dependent SMARCA2 degradation in NCI-H1568 and 41 A549 engrafted tumour bearing mice following short-term treatment (Figure 4e and 4f). Correspondingly, ACBI2 42 (administered at 80 mg/kg orally qd) significantly inhibited tumour growth in an A549 xenograft model (Figure 4g) 43 and was well tolerated (Extended Data Figure 5e). Median SMARCA2 protein levels in tumours collected at the 44 end of this study were decreased by 90% compared to control treated animals (Figure 4h). Finally, we tested ACBI2 45 ex vivo treatment of human whole blood, obtained from three different healthy donors, and observed near-46 complete degradation of SMARCA2 with clear selectivity over SMARCA4 (Figure 4i). Together, these data 47 demonstrate that oral bioavailability in combination with preferential degradation of one close paralog, 1 SMARCA2, over the other, SMARCA4, can be achieved in vitro and in vivo with our VHL based protein degrader 2 ACBI2. 3 4 5 Figure 4. ACBI2 is an orally bioavailable degrader that preferentially degrades SMARCA2 and induces lung cancer 6 tumour growth inhibition. a. Plasma profiles of ACBI2 in mouse after administration of 5 mg/kg i.v. or 30 mg/kg 7 p.o. The oral bioavailability of ACBI2 was 22%. b. Structure of ACBI2 and selected long range NOEs are highlighted 8 in the strip plot of the 2D NOESY spectra. c. The indicated cell lines were treated with ACBI2 for 144-192 h, and 9  cell viability was measured using CellTiter Glo (n= 2-7 independent experiments). Displayed are IC50 values with 1 95% confidence interval from 4-parametric logistic curve fit. Heatmap provides sensitivity to genetic depletion by 2 CRISPR, gene expression and mutation data from DepMap/CCLE, scaled for each row, i. e. across cell lines, but 3 separate for each parameter. d. Effects of ACBI2 (purple) and negative control compound 12 (gray, cis-4 hydroxyproline analogue of ACBI2 which is not capable of binding VHL) on the proteome of NCI-H1568 cells 5 treated with the compounds at 100 nM for 4 h. Data are plotted as the log2 of the normalised fold change in 6 abundance against -log10 of the p-value per protein from 3 independent experiments. All t-tests performed were 7 two-tailed assuming equal variances. e. NCI-H1568 (average tumour size at treatment start ~470 mm 3 ) or f. A549 8 (average tumour size at treatment start ~360 mm 3 ) tumour bearing mice were treated orally with 100 mg/kg, 20 9 mg/kg, 5mg/kg ACBI2 or vehicle control (n=5 animals per group) and tumours collected 24 or 48 h after treatment. 10 Tumours from one NCI-H1568 tumour-carrying animal treated with 20 mg/kg ACBI2 and collected at 24 and 48 h, 11 respectively, could not be analysed due to poor tumour quality. SMARCA2 levels in viable tumour tissue were 12 determined using DAB-based IHC staining. Each datapoint represents the background-normalised DAB optical 13 density (OD) within the viable tumour area of one tumour section, corresponding to an individual tumour. Mean 14 OD levels and standard deviations are indicated in the graphs. Numbers above the datapoints represent median 15 levels of SMARCA2 signal decrease relative to control-treated samples. g. A549 tumour bearing mice were treated 16 orally with 80 mg/kg ACBI2 once daily. Average tumour volume at the start of treatment were ~220 mm 3 . At day 17 21 of treatment, a TGI of 47% (p-value = 0.0351 vs. control) was measured. Values represent the mean of 10 18 animals per group, error bars indicate standard deviation. h. At the end of the study shown in g, tumours were 19 collected and SMARCA2 levels in viable tumour tissue determined using DAB-based IHC staining (as for e and f). 20 The median level of SMARCA2 staining in ACBI2 treated tumours was reduced by 90% compared to vehicle control-21 treated tumours. i. Human whole blood from three healthy donors was treated with the indicated concentrations 22 of ACBI2 for 18 h in the dark. Protein was extracted from PBMCs and relative SMARCA2 and SMARCA4 levels (each 23 normalised to GAPDH) measured using automated Western blotting. Each data point represents an average value 24 based on three biological replicates, shown as a percentage relative to DMSO-treated control samples. Error bars 25 indicate standard deviation. 26

27
Discussion 28 An oral route of administration for a new small molecule therapeutic is currently considered the rule 29 rather than the exception. Despite an increasing number of orally dosed bifunctional degraders in the clinic, all 30 those disclosed to date rely solely on ImiDs to recruit the CRBN E3 ligase recognition subunit 16-18 . Whilst 31 successful, this restriction greatly limits the long-term therapeutic scope and is predicated to some degree on an 32 assumption that larger E3 ligase recruiting motifs cannot yield orally available PROTACs. Here, we introduce three 33 principles for arriving at orally available VHL-based degraders that we believe to be of general utility: Firstly, the 34 de novo design of novel and potent protein of interest binders that display improved physicochemical properties 35 at the outset of bifunctional degrader design. Secondly, crystallographic knowledge of ternary complex binding 36 modes guides exploration of new exit vector space to achieve more stable complexes and consequently more 37 potent and faster degraders. Lastly, we show how small linker modifications can influence compound 38 conformations leading to more compact arrangements with reduced 3D polar surface area and radius of gyration. 39 We were also able to identify compounds that selectively degrade SMARCA2 over SMARCA4 without appreciable 40 differences in affinity for the binding ligand alone. As has been shown previously in the field of targeted protein 41 degradation 21,22 , we were also able to identify compounds that discriminate and preferentially degrade highly 42 homologous target proteins (here, the bromodomains of SMARCA2 over SMARCA4) without appreciable 43 differences in binding affinity for the target ligand alone. 44 BAF (SWI/SNF) chromatin remodelling complexes play critical roles in cancer 23,24 . For example, it has 45 recently been shown that androgen receptor (AR) and forkhead box A1 (FOXA1) expressing prostate cancer cells 46 are sensitive to simultaneous degradation of BAF complex subunits SMARCA2, SMARCA4 and PBRM1 25 . The 1 synthetic lethality between SMARCA2 and SMARCA4, resulting in sensitivity of SMARCA4-deficient cells to loss of 2 SMARCA2, has been discovered and validated by genetic methods 5-7 , but pharmacological validation and 3 exploitation of this synthetic lethal relationship has been hampered by the lack of suitably selective small 4 molecules, in particular for effective in vivo use in animal models 10,20 . Here we show that ACBI2 is capable of 5 inducing near-complete degradation of SMARCA2 in mouse lung cancer xenograft models that leads to tumour 6 growth inhibition. At the same time, ACBI2 offers a clear window of selectivity between SMARCA2 and SMARCA4 7 degradation in human whole blood and a consistent preference for the degradation of SMARCA2 over SMARCA4 8 in cell lines expressing both ATPases. Nevertheless, it is notable that despite efficient degradation of SMARCA2 in 9 vivo, only tumour stasis was observed upon compound treatment in the models studied therein. This is 10 unexpected given the strong effects observed upon SMARCA2 deletion or knockdown in functional genomic 11 studies 5-7 , suggesting a disconnect with pharmacological degradation. While it cannot entirely be excluded that 12 more efficacious degraders may cause stronger effects, it is also possible that cells can more readily adapt to loss 13 of SMARCA2 and SMARCA4 in vivo than in vitro, highlighting the need for in vivo validation of therapeutic concepts. 14 The possibility that other indications such as prostate cancer or multiple myeloma are more dependent dual loss 15 of SMARCA2 and SMARCA4 remains 25 . In either case, appropriate drug combinations could enhance in vivo efficacy 16 and warrant dedicated investigation in the future. To promote further understanding in the community, ACBI2 17 will be made available upon request via the opnMe innovation platform (https://opnme.com/). We anticipate that 18 our study will induce a significant shift in thinking around the design of orally efficacious bifunctional molecules 19 and hope that the studies described here will encourage others to explore the chemical and biological space that 20 may be utilised to discover orally active bifunctional small molecule therapeutics. SMARCA2 BD ternary complex comprising the asymmetric unit (asu), surrounded by closely packed symmetry-4 related copies (A-E, surface coloured teal, orange, blue, green, yellow respectively in panel a). Panels a(i) and b 5 show the identical orientation, however in b for symmetry copies A-E the surface is not shown, to more clearly 6 identify the asu (copy 1, both cartoon and surface in both panels; VCB shown as purple cartoon and surface; 7 compound 5 shown as yellow sticks; SMARCA2 BD , SM, shown as rainbow cartoon from blue N-terminus to red C-8 terminus and light grey surface). Panels c-e show in closeup some of the interactions of the SMARCA2 BD that 9 mediate crystal contacts, including hydrogen bond/salt bridge interactions and van der Waals interactions, in 10 particular residues Lys5, Lys36 and Lys71 of SMARCA2 BD . 11 NCI-H1568 tumour bearing mice were treated subcutaneously with 5 mg/kg compound 6 with two different 8 treatment schedules ("Treatment 1/2", see methods for details). Body weight was measured daily. Displayed are 9 mean and standard deviation of change per day for n=10 animals. f. Tumours from e were collected at the end of 10 the study and SMARCA2 levels in viable tumour tissue were determined using DAB-based IHC staining. Each 11 datapoint represents the background-normalised DAB optical density (OD) within the viable tumour area of one 12 tumour section, corresponding to an individual tumour. Mean OD levels and standard deviations are indicated in 13 the graphs. In most cases, SMARCA2 levels in tumours from treated animals were below the limit of detection. orientations for SMARCA2 BD with compound 9 (yellow), compound 5 (green) and compound 6 (blue). b. NCI-H1568 4 tumour bearing mice were treated orally with compound 10 at 30 or 100 mg/kg daily. Average tumour volume at 5 the beginning of treatment was ~220 mm 3 . Body weight was measured daily. Displayed are mean and standard 6 deviation of change per day for n=10 animals. c. Tumours from b were collected at the end of the study and 7 SMARCA2 levels in viable tumour tissue from four animals per group were determined using DAB-based IHC 8 staining. Each datapoint represents the background-normalised DAB optical density (OD) within the viable tumour 9 area of one tumour section, corresponding to an individual tumour. Mean OD levels and standard deviations are 10 indicated in the graphs. In most cases, SMARCA2 levels in tumours from treated animals were below the limit of 11 detection. 12 Displayed are means of 1-3 independent experiments, error bars indicate standard deviation. Right panel shows 1 DC50 values with 95% confidence interval from 4-parametric logistic curve fit. c. Immunoblot of SMARCA2 and 2 PBRM1 degradation in NCI-H1568 cells treated with indicated concentrations of compound 6 and ACBI2 for 4 h. 3 d. Immunoblot of SMARCA2 and PBRM1 degradation in NCI-H1568 cells treated with 100 nM compound 6 and 4 ACBI2 alone and in combination (indicated by +) with 10 μM VH298, MLN4924 and MG132 for 4 h. e. qPCR for 5 KRT80 mRNA levels (normalised to GAPDH housekeeping gene) after 18 h ACBI2 treatment at the indicated 6 concentrations in A549 and NCI-H1568 cells. Means of 2 independent experiments are displayed relative to the 7 DMSO control, error bars indicate standard deviation. f. Effects of compound 6 (purple) and negative control 8 compound 11 (gray, cis-hydroxyproline analogue of compound 6 which is not capable of binding VHL) on the 9 proteome of NCI-H1568 cells treated with the compounds at 100 nM for 4 h. Data are plotted as the log2 of the 10 normalised fold change in abundance against -log10 of the p-value per protein from 3 independent experiments. 11 All t-tests performed were two-tailed assuming equal variances.  DC50 values with 95% confidence interval from 4-parametric logistic curve fit.     Table 3: Measurement of PROTAC cooperativity (α) and ternary complex affinity (ternary IC50) via TR-FRET -competition for SMARCA2 BD , SMARCA4 BD and PBRM1 BD5 ± VCB respectively. PROTAC 1 cooperativity (α) is calculated for SMARCA2 BD , SMARCA4 BD and PBRM1 BD5 respectively ratio between their TR-FRET IC50 / TR-FRET IC50 +VCB values measured in the same run. Data represent means with errors 2 stated as ± standard deviation with repeats (n) specified in brackets. -   Table 5: Compound pharmacokinetics in pharmacokinetic and pharmacology studies in NMRI mice. All data represent mean of 3 animals.

2
Protein crystallography and protein production 3 Protein production for SMARCA2 and the VCB complex was done as previously described 10  complex per asymmetric unit. The structure was solved by molecular replacement using PHASER 27  surface activation with EDC/NHS (contact time 420 s, flow rate 10 µl/min) SMARCA2 BD and SMARCA4 BD at 0.01 -42 0.05 mg/ml in coupling buffer (10 mM Na-Acetate pH 6.5, 0.005% TWEEN 20 and 50 µM PFI-3 8 ) were immobilised 1 to a density of 100-5000 Response Units (RU). The reference surface was subsequently deactivated using 1 M 2 ethanolamine. For VCB immobilisation, streptavidin (Sigma Aldrich, prepared at 1 mg/ml in 10 mM sodium acetate 3 coupling buffer, pH 5.0) was immobilised by amine coupling to a density of 3000-5000 RU after which biotinylated 4 VCB complex (0.125 mM in running buffer) was streptavidin-coupled to a density of 100-500 RU. Biotinylated VCB 5 was prepared as previously described 21 . The reference surface was generated by deactivating the EDC/NHS-6 treated surface with 1 M ethanolamine. All interaction experiments were done at 6°C in running buffer (20 mM 7 TRIS, pH 8.3, 150 mM potassium chloride, 2 mM magnesium chloride, 2 mM TCEP, 0.005% TWEEN 20, 1% dimethyl 8 sulfoxide). For ternary complex measurements, a sensor chip surface with VCB immobilised was used (preparation 9 as described above). Experiments were run in dual-inject mode with 10 µM SMARCA present during the injection 10 and dissociation phase. Sensorgrams from reference surfaces and blank injections were subtracted from the raw 11 data prior to data analysis using Biacore Insight software. Affinity and binding kinetic parameters were determined 12 by global fitting using the 1:1 interaction model with a term for mass-transport included. per well. The next day, compounds were added using a digital dispenser and a "T0" sample was measured for 38 reference. Cells were incubated for 144-192 hours and viability (luminescence) was measured using CellTiter Glo 39 or CellTiter Glo 2.0 (Promega) according to manufacturer's instructions after equilibrating cells and reagents at 40 room temperature and 10-20 min incubation time while shaking. Values were displayed relative to negative 1 controls (DMSO) and curves were fitted using a 4-parametric logistic model. Compound solubility was determined by dilution of a 10 mmol/l compound solution in DMSO into buffer to a final 28 concentration of 125 µg/ml. Dilution into a 1:1 mixture of acetonitrile and water was used as reference. After 24 29 h, the incubations were filtrated and the filtrate was analyzed by LC-UV. 30 31

Microsomal stability 32
The degradation kinetics of 1 µmol/l compound in 0.5 mg/ml liver microsomes were inferred in 100 mM Tris-HCl 33 pH 7.5, 6.5 mM MgCl2 and 1 mM NADPH at 37°C. Reactions were terminated by addition of acetonitrile and 34 precipitates separated by centrifugation. Compound concentrations in supernatants were measured by HPLC- 35 MS/MS and clearance was calculated from compound half-lives using the well-stirred liver model. 36

Plasma protein binding 38
Binding of compound to plasma proteins was determined by equilibrium dialysis of 3 µmol/L compound in plasma 1 (or plasma dilutions in PBS) against PBS through an 8 kDa molecular-weight cut-off cellulose membrane (RED 2 device, Thermo Fisher) at 37°C for 5 h. After incubation, aliquots from donor and acceptor compartments were 3 precipitated and the concentrations in the supernatants were determined by quantitative LC-MS/MS. Calibration 4 and quality control samples were prepared using blank plasma and internal standard. The fraction unbound was 5 calculated as ratio of the compound concentration in the acceptor compartment to the concentration in the donor 6 compartment. 7

Bidirectional permeability measurement in Caco-2 cells 8
Bidirectional permeability of test compounds across a Caco-2 cell monolayer was measured as described 35  where Q is the amount of compound recovered in the receiver compartment after the incubation time t, C0 the 24 initial compound concentration given to the donor compartment, and s the surface area of the Transwell inserts. 25 Efflux ratio is calculated as the quotient of Papp,BA (mean of duplicate) to Papp,AB (mean of duplicate). The P-gp 26 substrate apafant and one low permeable compound (BI internal reference, Papp ≈ 3 × 10 −7 cm/s, no efflux) were 27 included in every assay plate. In addition, Transepithelial electrical resistance (TEER) values were measured for 28 each plate before the permeability assay. All three parameters (efflux of the reference substrates, Papp values of 29 the low permeable compound, and TEER values) were used to ensure the quality of the assays. 30 31

Animals and xenograft experiments 32
Female BomTac:NMRI-Foxn1 nu mice were obtained from Taconic Denmark at an age of 6-8 weeks. After arrival of 33 the local animal facility at Boehringer Ingelheim RCV GmbH & Co KG mice were allowed to adjust to housing 34 conditions at least for 5 days before the start of the experiment. Mice were group-housed under pathogen-free 35 and controlled environmental conditions and handled according to the institutional, governmental and European 36 Union guidelines (Austrian Animal Protection Laws, GV-SOLAS and FELASA guidelines). Animal studies were 37 approved by the internal ethics committee and the local governmental committee. Food and water were provided 38 ad libitum. To establish subcutaneous tumours mice were injected with 5 x 10 6 NCI-H1568 in PBS with 5% FCS or 39 with 1 x 10 7 A549 cells in PBS with 5% FCS. Tumour diameters were measured with a caliper three times a week. 40 The volume of each tumour (in mm 3 ) was calculated according to the formula "tumour volume = length * 1 diameter 2 * π/6." To monitor side effects of treatment, mice were inspected daily for abnormalities and body 2 weight was determined daily after the start of treatment. Animals were sacrificed when the tumours reached a 3 size of 1,500 mm 3 . Mice were randomised into treatment groups when the average tumour size reached ~210-4 220 mm 3 . All administrations were dosed with 10 ml/kg (s.c. and oral). Control mice were dosed subcutaneously 5 with 10% HP-β-CD in 50% Ringer solution and orally with 15% HP-β-CD, that means the control mouse treatment 6 was corresponding to the solvent of the compounds. Compound 6 was dosed subcutaneously either in a d1-4 q3d, 7 d6-8 q2d, d11-13 q2d treatment ("Treatment 1") or a q3 or 4d treatment ("Treatment 2"). Compound 10 was 8 dosed at 30 or 100 mg/kg and ACBI2 at 80 mg/kg, both compounds orally with a daily dosing. 9 For the evaluation of the statistical significance of tumour growth inhibition, a one-tailed nonparametric Mann-10 Whitney-Wilcoxon U-test was performed, based on the hypothesis that an effect would only be measurable in 11 one direction (i. e. expectation of tumour inhibition but not tumour stimulation). Analysis was performed on the 12 day indicated for each experiment. The p-values obtained from the U-test were adjusted using the Bonferroni-13 Holm correction. By convention, p-values ≤0.05 indicate significance of differences. 14

Immunohistochemistry and imaging analysis 15
Xenograft samples were fixed in 10% formaldehyde for 24 hours and later moved to ethanol and embedded in 16 paraffin. 2 µm-thick sections were cut using a microtome, then placed on glass slides (KLINIPATH, silan printer 17 slides PR-S:001), and subsequently dewaxed. SMARCA2 immunohistochemistry was carried out on the Leica BOND 18 RX platform (Leica Biosystems) according to manufacturer's instructions using a human-specific SMARCA2 19 antibody (Cell Signaling #11966 clone D9E8B RRID:AB_2797783 1:400) in a 20 minute incubation in EDTA-based 20 pH 9 ER2 buffer (high-pH epitope retrieval buffer or heat-induced epitope retrieval buffer). After staining, the 21 slides were cover-slipped with Shandon Consul-Mount glass covers, scanned using 3D Histotech slide scanner 22 (Leica Biosystems). All slides were reviewed and evaluated for quality by a board-certified MD specialist in 23 Anatomic Pathology (PC). Imaging analysis was performed using the digital pathology platform HALO (Indica Labs). 24 A tissue-classifying algorithm was trained to selectively recognise viable tumour tissue against stroma, necrosis, 25 and skin. The tissue classification output for each scan was reviewed and manually edited as necessary. A cell 26 detection and scoring algorithm was trained to measure DAB optical density (OD) in the nuclei of tumour cells. A 27 positivity threshold for DAB OD was determined by normalisation with respect to the DAB OD as calculated from 28 bona fide negative tissue (e. g., murine stroma as background). The average, background-normalised DAB OD of 29 tumour cell nuclei was used to quantitate SMARCA2 expression in each xenograft sample. 30