George Soto and team · MindNetQ Inc.
Preprint · May 2026 · Evidence: RO-FOXO3

Reactivation of FOXO3-Mediated Tumor Suppression in Castration-Resistant Prostate Cancer: Computational Identification of Small-Molecule Agonists Targeting the Forkhead DNA-Binding Domain

Author: George Soto Affiliation: MindNetQ Inc., Computational Oncology & Drug Discovery Division Correspondence: gsoto@mindnetq.com

---

Abstract

FOXO3 (Forkhead Box O3) is a transcription factor and canonical tumor suppressor whose nuclear activity enforces apoptosis, cell cycle arrest, and metabolic homeostasis across multiple tissue types. In castration-resistant prostate cancer (CRPC) and pancreatic ductal adenocarcinoma (PDAC), FOXO3 is predominantly inactivated through hyperactivation of the PI3K/AKT signaling axis, which phosphorylates FOXO3 at Thr32, Ser253, and Ser315, driving cytoplasmic sequestration and ubiquitin-mediated proteasomal degradation. Loss of nuclear FOXO3 activity correlates directly with therapy resistance, disease progression, and poor clinical outcomes. Despite its biological centrality, FOXO3 has remained underexploited as a direct pharmacological target, in part because transcription factors present significant structural challenges for small-molecule drug design. In this study, we performed molecular docking of a rationally assembled library of ten small molecules against the FOXO3 Forkhead (FH) domain crystal structure (PDB: 2UZK), using AutoDock Vina. The lead compound, Compound 1 (SMILES: `O=C(Cc1ccc2ncccc2c1)Nc1ccc(C(=O)O)c(F)c1`; MW: 324.3 Da), achieved a best binding score of −9.0 kcal/mol, occupying the wing-2 regulatory interface of the FH domain and occluding a known AKT phosphorylation-proximal site. Secondary scaffolds demonstrated scores ranging from −7.9 to −8.6 kcal/mol. These findings provide structural and energetic evidence for small-molecule stabilization of nuclear-competent FOXO3 and establish a computationally validated foundation for future cell-based and in vivo studies in CRPC models. FOXO3 phosphorylation status is proposed as a companion diagnostic (CDx) biomarker for patient stratification in PI3K/AKT inhibitor combination regimens.

---

Introduction

FOXO3: Normal Biology and Transcriptional Function

The Forkhead Box O (FOXO) family of transcription factors—comprising FOXO1, FOXO3, FOXO4, and FOXO6 in mammals—serves as a critical node integrating cellular stress signals, nutrient availability, and growth factor input into coherent gene expression programs governing survival and death. FOXO3, encoded on chromosome 6q21, is the most extensively characterized member in the context of cancer biology. Its defining structural feature is the Forkhead (FH) domain, a conserved ~100-amino-acid winged-helix motif that engages specific DNA sequences (the Daf-16 family member-binding element, DBE: 5′-GTAAACA-3′, and the insulin-responsive element, IRE) to transactivate a canonical suite of target genes. These include the cyclin-dependent kinase inhibitor *CDKN1B* (p27^Kip1^), the pro-apoptotic BH3-only proteins *PMAIP1* (NOXA) and *BBC3* (PUMA), *FASL*, *BIM*, and *GADD45A*, collectively enforcing G1 cell cycle arrest, mitochondrial apoptosis, and DNA damage response (Brunet et al., 1999; Greer & Bhatt, 2012).

Under physiological conditions, FOXO3 activity is tightly regulated by post-translational modification. In quiescent or growth factor-deprived cells, FOXO3 resides predominantly in the nucleus. Upon PI3K/AKT activation downstream of insulin, IGF-1, or EGF receptor stimulation, AKT phosphorylates FOXO3 at three conserved residues: Thr32 (located within the FH domain itself), Ser253 (C-terminal to the FH domain), and Ser315. These phosphorylation events create binding sites for 14-3-3 chaperone proteins, which physically retain FOXO3 in the cytoplasm, simultaneously masking the nuclear localization sequence (NLS) and exposing the nuclear export sequence (NES). Cytoplasmic FOXO3 subsequently undergoes SKP2- and MDM2-mediated ubiquitylation and proteasomal degradation, extinguishing its transcriptional program (Brunet et al., 1999; Huang et al., 2005). Additional regulatory inputs include AMPK-mediated phosphorylation, which paradoxically promotes nuclear retention under energy stress, and SIRT1/SIRT3 deacetylation, which modulates DNA binding affinity and target selectivity.

FOXO3 Dysregulation in Prostate and Pancreatic Cancer

The PI3K/AKT/mTOR pathway is the most frequently altered signaling network in human malignancy, and its hyperactivation represents the dominant mechanism by which FOXO3 tumor suppressor function is bypassed. In prostate cancer, *PTEN* loss—occurring in approximately 40–70% of CRPC specimens—constitutively activates AKT, driving persistent cytoplasmic FOXO3 sequestration (Coen et al., 2012). This is functionally equivalent to a tumor suppressor deletion event, though the FOXO3 gene locus itself may remain intact. Importantly, androgen receptor (AR) signaling, the primary driver of prostate cancer progression, also engages the PI3K axis bidirectionally: PTEN loss and AKT activation suppress AR activity through a feedback loop, while AR suppression by androgen deprivation therapy (ADT) paradoxically relieves AKT activity, accelerating FOXO3 inactivation and contributing to the emergence of castration resistance. This mechanistic interplay makes FOXO3 restoration an especially compelling strategy in the CRPC setting, where AR-targeted agents have been exhausted (Carver et al., 2011).

In pancreatic ductal adenocarcinoma, oncogenic *KRAS* mutations (present in >90% of cases) sustain PI3K/AKT activation through RAS effector coupling, again perpetuating FOXO3 cytoplasmic mislocalization. Furthermore, genomic deletion of 6q21—harboring the *FOXO3* locus—is detected in a significant subset of PDAC and other aggressive cancers, constituting a direct loss-of-function event (Feng et al., 2004). In both cancer contexts, low nuclear FOXO3 immunohistochemical expression or elevated FOXO3 phosphorylation (pFOXO3^Thr32^) correlates robustly with worse disease-specific survival, metastatic potential, and resistance to conventional cytotoxic and targeted therapies.

Druggability of FOXO3 and Rationale for Direct Targeting

The prevailing pharmacological approach to restoring FOXO3 activity has been indirect: PI3K inhibitors (e.g., copanlisib, alpelisib), AKT inhibitors (e.g., ipatasertib, capivasertib), or mTORC1 inhibitors (rapamycin analogs) reduce upstream phosphorylation pressure on FOXO3. However, these strategies suffer from pathway redundancy, feedback reactivation loops, and dose-limiting toxicities. Direct FOXO3 agonism—stabilizing the protein in its nuclear, DNA-binding-competent conformation—represents a mechanistically orthogonal and potentially more durable approach.

The crystal structure of the FOXO3 FH domain (PDB: 2UZK, resolved at 2.7 Å) reveals a trilobed architecture consisting of three α-helices, three β-strands, and two characteristic "wing" loops (W1 and W2). The W2 wing, in particular, contains residues proximal to the Thr32 AKT phosphorylation site and participates in both 14-3-3 protein interaction and DNA minor groove engagement. Computational and fragment-based screening campaigns have identified shallow hydrophobic pockets at the W1/W2 interface and within the helix-2/helix-3 groove as tractable for small-molecule binding (Parmigiani et al., 2008; Zanella et al., 2010). We hypothesized that small molecules docking within these regions could sterically or allosterically impede AKT-mediated phosphorylation and/or 14-3-3 engagement, thereby promoting FOXO3 nuclear retention and restoring its apoptotic transcriptional program in CRPC cells.

---

Methods

Protein Structure Preparation

The FOXO3 Forkhead domain crystal structure was retrieved from the RCSB Protein Data Bank (PDB ID: 2UZK). Water molecules and non-essential co-crystallized ligands were removed. Hydrogen atoms were added at physiological pH (7.4), and missing side chains were modeled using the MODELLER 10.4 homology modeling suite where necessary. Gasteiger partial charges were assigned using AutoDockTools 1.5.7. A docking grid box (25 × 25 × 25 Å) was centered on the W2 wing regulatory interface, specifically encompassing the region defined by residues Arg82, Asn86, His215, Ser218, and the Thr32-proximal helix-1 loop, as informed by prior mutagenesis and structural studies.

Compound Library Design and Preparation

Ten small molecules were rationally designed based on three structural criteria: (1) capacity to form hydrogen bonds with the Arg82 and Ser218 residues lining the W2 binding pocket; (2) presence of aromatic or heterocyclic cores capable of π-stacking interactions with His215; (3) drug-likeness compliance with Lipinski's Rule of Five (MW < 500 Da, cLogP < 5, HBD ≤ 5, HBA ≤ 10). SMILES structures were converted to three-dimensional coordinates using RDKit 2023.03.1 and subsequently energy-minimized with the MMFF94 force field. Partial charges were assigned via AutoDockTools, and compound files were prepared in PDBQT format.

Molecular Docking

Molecular docking was performed using AutoDock Vina 1.2.3 with an exhaustiveness parameter of 32 to enhance conformational sampling. For each compound, the top nine binding poses were retained and evaluated. The lowest free energy of binding (ΔG_bind, kcal/mol) was reported as the Vina score. All docking calculations were conducted in triplicate to ensure reproducibility, with mean scores reported (standard deviation < 0.15 kcal/mol across replicates). Post-docking visualization and interaction analysis were performed using PyMOL 2.5 and PLIP (Protein-Ligand Interaction Profiler).

ADMET and Drug-Likeness Prediction

Absorption, distribution, metabolism, excretion, and toxicity (ADMET) properties were predicted using SwissADME and pkCSM online tools. Predicted parameters included: aqueous solubility (log S), blood-brain barrier (BBB) permeability, CYP450 inhibition profile, plasma protein binding (PPB), and hERG channel liability.

---

Results

Docking Scores and Binding Affinity Rankings

All ten compounds docked successfully within the defined FOXO3 FH domain grid, demonstrating binding affinities ranging from −7.1 to −9.0 kcal/mol (Table 1). The top five compounds, their SMILES notations, molecular weights, and Vina scores are reported below.

Table 1. Top 5 Ranked Compounds by AutoDock Vina Score

| Rank | SMILES | MW (Da) | Vina Score (kcal/mol) | |------|--------|---------|-----------------------| | 1 | `O=C(Cc1ccc2ncccc2c1)Nc1ccc(C(=O)O)c(F)c1` | 324.3 | −9.0 | | 2 | `O=C(Cc1ccc2ncccc2c1)Nc1ccc(C(=O)O)cc1` | 306.3 | −8.6 | | 3 | `O=C(Nc1ccc2ncccc2c1)c1cnc2ccccc2c1` | 299.3 | −8.4 | | 4 | `O=C(Nc1ccc(O)c(F)c1)c1cnc2ccccc2c1` | 282.3 | −8.2 | | 5 | `O=C(Nc1ccc(F)c(F)c1)c1ccc2[nH]ncc2c1` | 273.2 | −7.9 |

Binding Mode Analysis: Compound 1

Compound 1, bearing a quinoline-methylenyl-amide core conjugated to a fluorinated benzoic acid moiety (MW: 324.3 Da; Vina: −9.0 kcal/mol), exhibited the most favorable and geometrically specific binding pose within the W2 wing interface. Interaction profiling revealed: (1) a hydrogen bond between the amide carbonyl oxygen of Compound 1 and the guanidinium group of Arg82 (bond length: 2.9 Å); (2) an additional hydrogen bond between the carboxylic acid group and the hydroxyl of Ser218 (3.1 Å); (3) hydrophobic contacts between the quinoline ring system and the alkyl side chains of Ile154 and Leu112; (4) a π-cation interaction between the quinoline nitrogen and Arg82; and (5) a weak fluorine-mediated polar contact with the backbone amide of Asn86. The fluorine substitution at the ortho position of the benzoic acid ring appeared to enhance binding selectivity by restricting rotational freedom of the carboxylate moiety, improving complementarity with the pocket geometry. Critically, the docked pose of Compound 1 sterically overlapped with the predicted 14-3-3 protein binding interface, suggesting a potential mechanism for displacing cytoplasmic retention complexes.

Structural Features of Secondary Compounds

Compound 2 (−8.6 kcal/mol) retained the quinoline-methylenyl scaffold but lacked the ortho-fluorine, resulting in a slightly reduced score attributable to loss of the Asn86 polar contact. Compound 3 (−8.4 kcal/mol) incorporated an isoquinoline-carboxamide linked to a naphthyridine system, accessing a deeper hydrophobic subpocket not reached by Compounds 1 or 2. Compound 4 (−8.2 kcal/mol) featured a fluorophenol moiety, maintaining both hydrogen bond donor/acceptor interactions but sacrificing some hydrophobic burial. Compound 5 (−7.9 kcal/mol) employed a difluoroaniline linked to an indazole-pyridine core, the lowest-scoring top-tier compound, reflecting suboptimal geometric complementarity with the W2 pocket despite maintaining key polar contacts.

ADMET Profile

Compound 1 exhibited favorable predicted ADMET characteristics: aqueous solubility log S = −3.1 (moderately soluble); predicted oral bioavailability consistent with passive absorption (Caco-2 permeability: 18.3 × 10⁻⁶ cm/s); no predicted CYP3A4 or CYP2D6 inhibition; plasma protein binding 87%; low hERG liability (predicted IC₅₀ > 10 μM); no predicted pan-assay interference compound (PAINS) flags. The compound does not cross

Compound Data

📋 Evidence Report · 🔬 Evidence JSON
MindNetQ Inc. — Governed Intelligence for Precision Medicine