Raloxifene

New users of denosumab were defined as patients who never taken raloxifene within one year before treatment initiation, and the earliest date of denosumab prescribed was defined as the index date. The same criteria were applied for new users of raloxifene without prior denosumab treatment. PSM was applied to balance differences in baseline demographic and clinical characteristics between the denosumab and raloxifene groups. The propensity score of initiating denosumab or raloxifene was estimated by logistic regression with above mentioned baseline characteristics, with a 1:1 ratio.

We carried out our VS with AutoDock Vina [54 (link)] in the PyRx open-source software package [55 (link)]. In brief, SDF files of our 1604-compound FDA-approved library were downloaded from the ZINC15 database [46 (link),47 (link),48 (link)] and loaded into PyRx with the Open Babel chemical toolbox [64 (link)]. The SDF files were then energy-minimized and appropriately protonated to generate the required PDBQT files. Once all ligands were prepared, FL ε R3 (PDB 6var) [30 (link)] was loaded and prepared as the receptor molecule. The docking grid was prepared in a manner to ensure an unbiased dock (i.e., the grid encompasses the entire receptor molecule), and therefore, dimensions of 64.9 × 57.0 × 39.3 were used. Finally, we enabled nine possible docking poses per ligand. The intention of our VS was to rapidly screen our compound library and rank-order our lead compounds by predicted affinity. We therefore carried out selection criteria on the basis of affinity, commercial availability and drug-like properties, and dock site (Figure 2a) to identify the lead compounds from our 1604-compound library. For our first selection step, we took the 122 compounds whose top-ranked docking pose had a predicted affinity higher than that of the already known [30 (link)] ε-targeting ligand Raloxifene (−9.5 kcal·mol⁻¹) (Figure 2b). For our second selection step, we excluded all compounds that were not commercially available and/or had potential adverse effects and proceeded with 66 compounds (Figure 2c). This step required the manual curation of the compound library. We considered any anticancer drug or any compound with a mode of action that included the inhibition of fundamental cellular processes (e.g., DNA replication) as having a potential adverse effect. Given the functional importance of the PL [22 (link),23 (link),27 (link),29 (link),33 (link),34 (link)] (Figure 1a) and our previous computational [30 (link),31 (link)] (Figure 1c) and experimental [30 (link)] (Figure 1d) data, our final selection step chose compounds that targeted the ε PL. To increase our confidence that we proceeded with authentic ε PL-targeting compounds, we repeated our docking three additional times. Then, we classified confident PL docking on the basis of two criteria: (1) top-rank pose localized to the PL in >50% of the repeated runs and/or (2) >50% of all poses localized to the PL (Figure S1). In both instances, PL localization was loosely defined as having more than one contact within 5 Å of a PL nucleotide (i.e., C14–C19, including the adjacent A13, A20, U48, and U49). In total, our VS identified 12 initial lead compounds (Figure 2e).