Dual-Site Acetylcholinesterase Inhibition and Multiscale Stability of Fused Quinoline Sulfonamides: A Chemoinformatic GA-MLR and Molecular Dynamics Study

1. Introduction

The global demographic transition toward an increasingly elderly population has precipitated an unprecedented neuropharmacological crisis, with Alzheimer’s disease (AD) emerging as the most significant driver of dementia worldwide. Currently accounting for approximately 60–70% of the 55 million global cases of dementia, AD is projected to affect nearly 139 million individuals by 2050 (Figure 1A–B) [1,2,3,4]. This escalation is not merely a clinical statistic; it represents a looming economic catastrophe. The global cost of dementia care was valued at USD 1.3 trillion in 2020 and is anticipated to double to USD 2.8 trillion by 2030, potentially destabilizing health systems across both high-income and developing nations (Figure 1B) [3,4]. Despite decades of intense investigation, the therapeutic landscape remains dominated by symptomatic treatments that fail to alter the underlying neurodegenerative trajectory, leaving a profound void in disease-modifying interventions [5].

The recent clinical disillusionment following the controversial approval and subsequent efficacy concerns of amyloid-targeted monoclonal antibodies, such as aducanumab and lecanemab, has highlighted the inherent limitations of mono-target paradigms [6]. These high-cost biological therapies have been plagued by significant safety issues, specifically Amyloid-Related Imaging Abnormalities (ARIA), characterized by brain edema (ARIA-E) and micro-hemorrhages (ARIA-H) [7]. Beyond primary neurodegenerative triggers, the pharmacological landscape is further complicated by recent pharmacovigilance data from the FDA Adverse Event Reporting System, which established concerning associations between common medications, such as proton pump inhibitors, and increased dementia events. This multifactorial complexity underscores the urgent necessity for novel therapeutics that possess superior safety profiles and high-occupancy target engagement to mitigate the cumulative burden on the aging brain [8]. Consequently, this shift in perspective has reinvigorated scientific interest in the “cholinergic hypothesis,” specifically repositioning the enzyme acetylcholinesterase (AChE) as a multifunctional structural target. While AChE typically facilitates the rapid hydrolysis of the neurotransmitter acetylcholine (ACh) within the synaptic cleft, the AD brain suffers from a profound ACh deficit that is catastrophically compounded by sustained enzymatic activity [9,10].

The structural anatomy of the human AChE enzyme provides a sophisticated, albeit challenging, blueprint for drug design (Figure 1C). The enzyme contains a narrow, 20-Å-deep active-site gorge with two distinct binding domains: the Catalytic Active Site (CAS) at the base and the Peripheral Anionic Site (PAS) at the entrance [11,12]. Traditional inhibitors, such as the first-generation agent Tacrine, primarily target the CAS to provide transient symptomatic relief by preserving available acetylcholine pools (Figure 1D) [13]. However, the therapeutic landscape has evolved toward identifying multi-target scaffolds, as seen with experimental leads like Clioquinol and PBT2, which aim to address metal-mediated neurotoxicity alongside enzymatic inhibition (Figure 1D) [14]. Recent biochemical evidence has established that the PAS serves as more than a simple gatekeeper; it acts as a pro-aggregating modulator of β-amyloid (Aβ) peptide aggregation [15,16]. The interaction between the PAS and Aβ peptides accelerates the formation of neurotoxic fibrils, creating a pathological bridge between cholinergic signaling failure and the amyloidogenic cascade [16,17] This dual functionality makes AChE a pivotal target for “dual-site” inhibitors, molecules designed to simultaneously anchor at the CAS to preserve ACh and blockade the PAS to halt amyloid aggregation [18,19].

Despite the conceptual elegance of dual-site inhibition, a critical research gap persists in the rational design of novel ligands that possess optimized Absorption, Distribution, Metabolism, Excretion, and Toxicity (ADMET) profiles [20]. Existing drugs like Tacrine were withdrawn from clinical use primarily due to severe, dose-dependent hepatotoxicity, a liability that remains a significant hurdle for fused quinoline scaffolds [21]. Furthermore, the Blood-Brain Barrier (BBB) represents a formidable pharmacological obstacle; many ligands that exhibit potent in vitro activity fail to achieve therapeutically relevant concentrations in the Central Nervous System (CNS) [20,21]. This necessitates a transition toward specialized neuroprotective scaffolds, such as the fused quinoline sulfonamide system (Figure 1E). By utilizing the aromatic quinoline core to anchor the CAS (Trp86) and an engineered sulfonamide linker to achieve the “longitudinal reach” required to bridge the 20-Å gap to the PAS (Asp74), we can optimize both potency and safety.

The convergence of chemoinformatics and biomimetic chromatography represents a powerful new trend in drug discovery, offering high-throughput alternatives to unethical animal testing [22,23]. Biomimetic chromatography, utilizing stationary phases like Immobilized Artificial Membranes (IAM) and Human Serum Albumin (HSA), provides experimentally grounded predictors of drug disposition by mimicking the polar-apolar interfaces of cell membranes and protein-binding environments [24]. By integrating these experimental parameters with sophisticated Quantitative Structure-Activity Relationship (QSAR) modeling, researchers can establish reliable correlations between molecular topology and biological fate [25]. In this context, the use of advanced descriptors, such as constitutional indices for conjugation assessment and GETAWAY descriptors for longitudinal reach enables a level of structural mapping previously unattainable. However, static QSAR models often fail to account for the induced-fit motions and conformational rearrangements inherent in protein-ligand interactions, necessitating the inclusion of physics-based molecular dynamics (MD) to validate the temporal stability of the predicted binding modes [26].

The development of specialized neuroprotective scaffolds that align with these standards is essential for fostering a self-sustaining scientific ecosystem. Current literature reveals a lack of comprehensive pipelines that combine large-scale descriptor pools, GA-MLR-based feature selection and long-range dynamic simulations to optimize fused ring systems for dual-site AChE engagement [27,28,29]. Bridging this gap is critical for the discovery of next-generation, disease-modifying AD therapeutics that are both clinically effective and toxicologically safe.

In this work, we address these multifaceted challenges through a rigorous computational optimization pipeline focused on novel fused quinoline sulfonamides (Figure 1F). By curating a diverse dataset of 115 analogues and generating 11,135 molecular descriptors, we aim to develop a robust GA-MLR QSAR model that satisfies the stringent OECD standards for statistical validity and interpretability. The focus is directed toward identifying the primary structural drivers, such as π-conjugation length and spatial mass distribution that allow a ligand to effectively bridge the distance between the CAS and PAS within the human AChE gorge. We prioritize the design of 16 novel derivatives guided by the resulting QSAR equation, subjecting the most promising leads to ADMET screening and blind molecular docking against the human AChE crystal structure (7XN1). The ultimate goal is to evaluate the dynamic stability and structural integrity of these lead-enzyme complexes through 100-ns molecular dynamics simulations and Dynamic Cross-Correlation Matrix (DCCM) analysis. By integrating quantitative modeling with physics-based dynamic analysis, this study provides a validated framework for the discovery of anti-Alzheimer’s agents that satisfy the requirements for scientific sovereignty and frontline neurodegenerative research.

2. Results and Discussion

2.1. QSAR Model Development and Statistical Robustness

A quantitative structure-activity relationship (QSAR) model was developed to predict the acetylcholinesterase (AChE) inhibitory activity of fused-ring quinoline analogues using multiple linear regression (MLR). Genetic algorithm-based feature selection was employed to prioritize five molecular descriptors from an initial pool of 11,135 variables, identifying max_conj_path, MATS3s, R8m, C-N-C=O, and the MNA substructure -H(-C(C-H-H-C)) as critical drivers of potency. The resulting GA-MLR model, constructed from a training set of 81 compounds, is defined by the following equation:

$${\mathrm{p}\mathrm{I}\mathrm{C}}_{50}=4.9509+0.0554·max\_conj\_path+5.3355·MAT53s+2.0399·R8m+1.186·(C-N-C=O)+1.5020·MNA$$

The model demonstrated robust statistical validity, explaining approximately 76% of the variance in the training set (R² = 0.7569; $R_{adj}^2$ = 0.7407). The predictive accuracy was supported by low error metrics (RMSE = 0.5752; MAE = 0.4367) and a high concordance correlation coefficient (CCC = 0.8616). Internal validation using leave-one-out (LOO) and leave-many-out (LMO) cross-validation yielded Q² values of 0.7244 and 0.7178, respectively. The negligible difference between R² and $Q_{LOO}^2$ (ΔR² = 0.0325) indicates high model stability and the absence of significant overfitting (Table 1). The model’s robustness was extensively evaluated through the visual diagnostics presented in Figure 2. The Experimental vs. Predicted activity plots (Figure 2A–B) demonstrate a tight linear distribution along the ideal diagonal, confirming high predictive fidelity across the potency range. To further scrutinize the error distribution, residual analysis was performed (Figure 2C–D). The random dispersion of residuals around the zero-horizontal axis indicates an unbiased model, proving that the errors are independent of the predicted potency values and confirming the suitability of the linear regression approach.

The Applicability Domain (AD) was delineated via a Williams plot (Figure 2G) to verify the model’s reliability for prospective lead design. The leverage analysis identified a threshold of h* = 0.222. Over 98% of the compounds remained within this threshold, with standardized residuals consistently localized within the ± 3σ range. Two compounds exhibited leverage values exceeding h*, yet they maintained low residuals and were thus categorized as influential points that extend the model’s chemical space rather than outliers. This confirms that the model is well-equipped to predict the potency of the novel quinoline sulfonamides described later in this study.

Finally, the possibility of chance correlation was ruled out via 2,000 iterations of Y-randomization (Figure 2I). The sharp decline in statistical performance ($R_{scr}^2$ = 0.0622) and the resulting negative cross-validation value ($Q_{scr}^2$ = -0.0961) serve as definitive proof that the model is structurally anchored . The LMO stability plot (Figure 2H) further confirms that the model’s performance is not dependent on a specific subset of data but is a consistent representation of the structure-activity relationship across the entire fused quinoline family.

2.2. Mechanistic Interpretation of Molecular Descriptors

The interpretability of the GA-MLR model is derived from the mechanistic relevance of its five constituent descriptors, which characterize the electronic, topological, and fragmental requirements for potent AChE inhibition. Each variable provides a quantitative bridge between the molecular architecture of the fused quinoline scaffolds and the specific sub-sites of the human AChE active-site gorge.

• max_conj_path (alvaDesc): This constitutional descriptor quantifies the length of the longest continuous π-conjugation path within the molecule. The positive coefficient identifies extended π-systems as primary potency drivers. Mechanistically, this facilitates high-occupancy π-π stacking interactions with the aromatic residues lining the 20-Å-deep enzyme gorge, specifically Trp86 in the CAS and Trp286/Tyr341 in the PAS.
• MATS3s (Dragon7): This 2D-autocorrelation index represents Moran autocorrelation of lag 3 weighted by the intrinsic state (I-state). It describes the electronic distribution and polarization across the molecular graph at a specific topological distance. In the context of AChE, electronic density distributed at a lag of 3 is critical for facilitating favorable electrostatic and π-cation interactions within the narrow gorge, particularly with Phe338.
• R8m (Dragon7): A GETAWAY (Geometry, Topology, and Atom Weights AssemblY) descriptor, R8m denotes R autocorrelation of lag 8 weighted by mass. This topological index captures the longitudinal reach and mass distribution of the scaffold. Its selection highlights the requirement for an inhibitor to effectively bridge the ~14–18 Å distance between the catalytic and peripheral binding sites within the human enzyme crystal structure (7XN1).
• C-N-C=O (Fragmentor): This fragment-based descriptor identifies the presence of carbamoyl or urea-like functional groups. Moieties identified by this descriptor act as essential hydrogen-bond donors or acceptors, facilitating polar contacts with the residues of the catalytic triad, such as Glu202 and Ser203, which are vital for competitive inhibition of acetylcholine hydrolysis.
• MNA (Multilevel Neighborhoods of Atoms): Multilevel Neighborhoods of Atoms (MNA) are 2D-substructural notations that describe the local environment of an atom. This specific MNA string accounts for subtle steric and hydrophobic effects in methyl-bearing local environments. Its inclusion suggests that the van der Waals fit within the narrow hydrophobic mid-gorge is highly sensitive to substitution patterns that modulate the local atomic neighborhood.

These five descriptors demonstrate that inhibitory potency is not a simple function of lipophilicity but is precisely governed by the spatial arrangement of aromatic density, electronic distribution, and fragmental reach. This mechanistic foundation provides the rational basis for the structural modifications used to design the high-potency leads described in the subsequent sections.

2.3. External Validation and Structural Generalizability

The definitive metric for the utility of a QSAR model in prospective drug design is its predictive fidelity when applied to “unseen” chemical entities. To confirm the generalizability of the five-descriptor GA-MLR model, an external validation set of eight compounds (n = 8) was utilized. These compounds, which were excluded from the initial training and feature selection phases, provide a rigorous structural “checkpoint” to verify that the model captures genuine biological interactions rather than local statistical artifacts [30,31,32,33].

The external validation yielded a coefficient of determination ($R_{ext}^2$) of 0.8620, indicating that the model successfully accounts for approximately 86% of the variance in the biological activity of the independent set. As illustrated in the regression profile (Figure 2), the predicted pIC₅₀ values demonstrate a strong linear correlation with experimental activities. The distribution of these compounds is particularly significant; they span a wide activity spectrum from low-micromolar (pIC₅₀ ≈ 4.8) to sub-nanomolar (pIC₅₀ ≈ 10.7) potencies, proving that the model maintains its accuracy even at high-potency extremes where the 20 Å AChE gorge becomes increasingly sensitive to structural nuances.

Table 2. Experimental versus calculated pIC₅₀ values for the external validation set (n = 8).

Compound	1	2	3	4	5	6	7	8
Calculated pIC50	4.782	5.288	8.429	7.620	6.272	6.024	6.306	7.314
Experimental pIC50	4.849	5.070	10.700	7.469	6.066	6.143	6.857	7.221
Reference	[30]	[31]	[31]	[30]	[30]	[32]	[32]	[33]

pIC₅₀ = -log₁₀(IC₅₀ [M]). Experimental activity values obtained from standardized literature datasets.

Table 2. Experimental versus calculated pIC₅₀ values for the external validation set (n = 8).

Compound	1	2	3	4	5	6	7	8
Calculated pIC50	4.782	5.288	8.429	7.620	6.272	6.024	6.306	7.314
Experimental pIC50	4.849	5.070	10.700	7.469	6.066	6.143	6.857	7.221
Reference	[30]	[31]	[31]	[30]	[30]	[32]	[32]	[33]

pIC₅₀ = -log₁₀(IC₅₀ [M]). Experimental activity values obtained from standardized literature datasets.

The structural architecture of the validation set (Figure 3, left panel) encompasses diverse fused-ring scaffolds, including substituted tetrahydronaphthyridines, pyrazolo-pyrano-pyridines, and indole-quinoline hybrids. The successful prediction of activity across these chemically distinct frameworks validates the robustness of the selected descriptors, specifically max_conj_path and R8m. This confirms that the model’s identification of extended π-conjugation and longitudinal reach as potency drivers is a universal requirement for human AChE inhibition within the quinoline chemical family. While the regression slope (y = 0.616x + 2.320) indicates a slight compression of the predicted range, the absence of pronounced systematic deviations or outliers confirms the model’s suitability for the prospective lead prioritization discussed in the following section.

2.4. Design and Predictive Profiling of Novel Fused Quinolines

The successful validation of the GA-MLR model established a quantitative blueprint for the rational optimization of the fused quinoline scaffold. In this phase, we applied the model equation to a prospective library of eighteen analogues (n = 18) designed to explore the structural requirements for enhanced dual-site inhibition. To ensure predictive calibration and maintain scientific integrity, the library was structured to include sixteen novel derivatives and two internal benchmarks (Compounds 2 and 4) whose experimental activities were utilized in the previous validation phase.

The design strategy focused on the synergistic modulation of the three primary potency drivers identified by the model: π-conjugation length (max_conj_path), electronic density distribution (MATS3s), and longitudinal mass distribution (R8m). Structural modifications primarily involved the introduction of sulfonamide and extended amine spacers to bridge the 20-Å enzyme gorge while maintaining favorable drug-like attributes. The predicted pIC₅₀ values for the designed series, summarized in Table 3, fall within a chemically meaningful range relative to the training set, with several novel leads exhibiting activities comparable to the high-potency benchmarks.

A critical observation from this screening was that while several novel leads exhibited predicted potencies similar to the benchmarks, Lead Compounds 19 and 20 were prioritized for subsequent physics-based evaluation. The rationale for this selection was not raw activity alone, but the achievement of superior drug-like attributes. The sulfonamide moiety in 19 and 20 provides a critical dipole for PAS interaction while maintaining synthetic accessibility (Synth ≈ 2.2) and a safety profile that distinguishes them from the literature benchmarks. As shown in Figure 4, the entire design series resides within the model’s applicability domain, ensuring that the prospective predictions are statistically sound.

2.5. ADMET Screening and Toxicity Observations

The prioritization of Lead Compounds 19 and 20 from the designed series was predicated on a multi-parameter evaluation of their pharmacokinetic (ADME) profiles and toxicological (T) risks. In neurodegenerative drug discovery, high inhibitory potency is therapeutically irrelevant if the scaffold lacks the physicochemical attributes necessary to traverse the Blood-Brain Barrier (BBB) or exhibit prohibitive systemic toxicity. The comparative ADMET parameters for Leads 19, 20, and the clinical reference drug Tacrine are summarized in Table 4.

Prioritized Lead 19 and 20 demonstrate exceptional Blood-Brain Barrier (BBB) permeability scores (0.647 and 0.768, respectively), which successfully meet the threshold for CNS-targeted therapeutics. While the reference drug Tacrine exhibits a higher BBB score of 0.977 and superior aqueous solubility (logS -2.875), it possesses a significantly lower lipophilicity (logP 2.432) compared to the novel leads (logP 4.38-4.55). The increased lipophilic character of the novel leads is a strategic requirement to optimize π-stacking interactions within the hydrophobic 20-Å AChE gorge. Furthermore, the Synthetic Accessibility (Synth) scores of ≈2.2 for the novel leads are comparable to Tacrine (2.015), indicating that these scaffolds are easily accessible for large-scale production despite their increased structural complexity. Notably, Lead 19 exhibited a superior QED score (0.520) compared to 20 (0.463), suggesting a more balanced drug-like architecture for the n-butyl sulfonamide linker.

Toxicological profiling using ProTox radar analysis revealed that while the quinoline core maintains a CYP2E1-mediated hepatotoxicity risk similar to Tacrine, the novel leads offer a significantly improved safety window. Compounds 19 and 20 exhibited markedly lower probabilities for neurotoxicity, immunotoxicity, and cardiotoxicity compared to the reference drug. This reduction in off-target interaction risk justifies the selection of these sulfonamide-linked fused quinolines as superior candidates for the treatment of Alzheimer’s disease. These results set a new trend for the development of “safe-by-design” dual-site inhibitors that mitigate the systemic liabilities associated with previous-generation AChE inhibitors.

2.6. Molecular Docking and Active-Site Gorge Interaction

Molecular docking against the human AChE crystal structure (7XN1) validated the dual-site binding hypothesis for the prioritized sulfonamides. Both prioritized Leads 19 (ΔG = -11.1 kcal/mol) and 20 (ΔG = -10.6 kcal/mol) achieved significantly higher binding affinities than the clinical reference drug Tacrine (ΔG = -9.0 kcal/mol). To delineate the structural basis for this superiority, the binding mode of Tacrine was analyzed as a baseline reference. As illustrated in the interaction fingerprint, Tacrine functions strictly as a single-site inhibitor, anchoring the Catalytic Active Site (CAS) via π-π stacking with Trp86 and a conventional hydrogen bond with Ser125. Crucially, Tacrine lacks the molecular length required to engage the Peripheral Anionic Site (PAS), leaving the Asp74 residue unoccupied.

Lead 19 demonstrated an optimized interaction fingerprint that effectively bridges the 20 Å enzyme gorge. As illustrated in the 2D interaction map (Figure 5), the tricyclic aromatic core anchors the Catalytic Active Site (CAS) via robust π-π stacking interactions with Trp86 and Tyr337, the primary residues responsible for acetylcholine recognition. Unlike Tacrine, the sulfonamide-linked n-butyl tail of Lead 19 extends toward the PAS, where it forms a critical, high-occupancy hydrogen bond with Asp74. This dual anchoring is further stabilized by π-π interactions with the PAS gatekeepers Trp286 and Tyr341, along with hydrophobic π-alkyl contacts with Phe338 and Tyr124. This binding mode confirms that Lead 19 strikes an ideal balance between rigidity and longitudinal reach (R8m), occupying the hydrophobic mid-gorge without inducing steric strain.

In contrast, Compound 20 exhibited a distinct binding penalty that explains its marginally lower affinity. While it maintains the Asp74 H-bond in the PAS, the extra methylene unit in its n-pentyl chain introduces excessive conformational freedom and steric bulk. As highlighted in the docking snapshots (Figure 5, Row 4), Compound 20 suffers from an Unfavorable Acceptor-Acceptor interaction with Ser125 (indicated by the red bubble). Mechanistically, this electronic clash, combined with the loss of the critical π-π stacking contact with Trp86 in the CAS, suggests that the longer tail of 20 forces the aromatic core into a sub-optimal orientation. These findings corroborate the QSAR model’s prediction that structural complexity must be precisely tuned to the enzyme’s physical dimensions, positioning Lead 19 as the superior candidate for inhibiting both neurotransmitter hydrolysis and PAS-mediated Aβ peptide aggregation.

2.7. Molecular Dynamics (MD) Simulation and Dynamic Stability

To validate the thermodynamic stability of the docked poses and characterize the temporal nature of the ligand-enzyme interactions, 100-ns molecular dynamics (MD) simulations were executed. The protein-ligand root mean square deviation (PL-RMSD) served as the primary indicator of complex stability. For Lead Compound 19, the protein backbone equilibrated within a range of 1.5-2.4 Å, signifying a stable global fold throughout the trajectory (Figure 6A). The ligand RMSD for 19 remained stabilized at ≈ 1.8 Å until 64 ns, followed by a transition to ≈ 4.2 Å. This shift corresponds to a productive conformational adaptation where the aromatic core curls and flattens to maximize surface complementarity with the catalytic active site (CAS) residues.

In contrast, Compound 20 exhibited a more erratic RMSD profile, with protein fluctuations reaching ≈ 2.7 Å and a less stabilized ligand trajectory (Figure 6B). This relative loss of conformational stability is likely attributed to the entropic penalty and steric hindrance of the longer n-pentyl chain, which prevents the tricyclic core from achieving the high-occupancy anchoring observed for 19.

The mechanistic basis for Lead 19’s superior stability is elucidated by the protein-ligand interaction fraction histograms (Figure 6, bottom row). Compound 19 maintained high-occupancy contacts with critical active-site residues, specifically Tyr124, Ser125, and Trp86. The interaction fractions for Tyr124 and Ser125 exceeded 1.0, indicating the formation of multiple simultaneous contact types, including conventional hydrogen bonds and stable water-mediated bridges. Persistent hydrophobic interactions with Trp86, Phe338, and Tyr341 further anchored the molecule within the 20 Å gorge. For Compound 20, while Asp74 and Trp286 interactions were maintained in the Peripheral Anionic Site (PAS), the occupancy of catalytic triad contacts was significantly reduced, corroborating the lower binding affinity reported in the docking studies.

The MD results systematically validate the GA-MLR model’s prioritization of Lead 19. The ability of this sulfonamide-linked scaffold to maintain consistent engagement with both the CAS (Trp86) and PAS (Asp74, Tyr124) while undergoing adaptive fitment changes distinguishes it as a high-confidence lead for Alzheimer’s therapy. This data provide the essential dynamic proof that structural precision in the R8m (longitudinal reach) and max_conj_path (aromatic stacking) descriptors translates directly to target engagement under physiological conditions.

2.8. Post-MD Integrity and Energy Landscapes

The structural integrity of the human AChE enzyme following 100-ns ligand binding was rigorously validated via stereochemical analysis of the final trajectory frames. For both complexes, over 89.6% of residues were located within the favored regions of the Ramachandran plot, with a cumulative total of > 97.9% in allowed regions. The minor population of outliers was strictly confined to intrinsically disordered surface loop regions, such as Ser67 and Ala141, confirming that the dual-site occupancy of the 20 Å gorge by our sulfonamide leads does not precipitate deleterious distortions in the protein backbone or catalytic machinery (Figure 7A and 7D).

The thermodynamic convergence of the binding states was evaluated by constructing the Free Energy Landscape (FEL) across the coordinates of global complex compactness (Rg) and RMSD. Lead Compound 19 achieved a highly stable binding state, reaching its global minimum at RMSD = 1.680 Å and Rg = 22.736 Å, with a relative conformational free energy (ΔG) of 5.859 kJ/mol (Figure 8B–C). The presence of a singular, deep energy funnel signifies that the n-butyl sulfonamide scaffold achieves a well-equilibrated, low-entropy binding mode.

In contrast, Compound 20 exhibited a more scattered energetic ensemble (Figure 8E–F). Its global minimum was identified at RMSD = 1.446 Å and Rg = 22.668 Å with a lower relative energy of 4.794 kJ/mol, yet the surrounding landscape is broader and less localized. This thermodynamic dispersion is a direct consequence of the increased conformational freedom of the n-pentyl chain, which prevents the tricyclic core from maintaining the rigid, high-occupancy orientation required for effective Aβ anti-aggregation through PAS blockade. These results systematically support the selection of Lead 19 as the superior candidate for clinical advancement.

2.9. Functional Residue Synchrony and Catalytic Integrity

The influence of ligand binding on the intrinsic global dynamics of human AChE was quantified using Dynamic Cross-Correlation Matrices (DCCM) for the Cα atoms (Figure 9). To establish a ground-truth for protein “breathing,” a ligand-free (Apo) simulation was utilized as a baseline reference (Figure 8C). In the Apo state, critical residue pairs such as Gly121-His447 (essential for the catalytic cycle) and Trp86-Tyr337 (responsible for substrate recognition) demonstrate pronounced anti-correlated motions (indicated by deep blue regions with C_ij < -0.25).

Lead Compound 19 effectively preserves these native anti-correlations (Figure 8A), suggesting that its dual-site anchoring at the PAS and CAS does not perturb the essential flexibility required for catalytic throughput. Conversely, Compound 20 (Figure 8B) induces a deleterious transition toward correlated motions in the Trp86-Tyr337 pair and increased rigidity in the Tyr124-Glu334 loop. This disruption of native synchrony, combined with the scattered Free Energy Landscape (Figure 7), provides the final mechanistic proof of 19’s functional superiority as a disease-modifying lead.

3. Materials and Methods

3.1. Dataset Preparation and Computational Framework

A primary dataset consisting of 115 fused quinoline analogues with experimentally determined AChE inhibitory activities was curated from peer-reviewed medicinal chemistry literature. To ensure consistency and mitigate heteroscedasticity, biological activity values (IC₅₀) were converted to the negative decadic logarithm (pIC₅₀ = -log₁₀ IC₅₀ [M]). Two-dimensional chemical structures were sketched using MarvinSketch (v24.3.1, ChemAxon) and subjected to 3D optimization utilizing the MMFF94 force field to reach the local energetic minimum prior to descriptor generation.

A comprehensive pool of 11,135 molecular descriptors, spanning constitutional, topological, geometrical, electronic, and fragmental properties was generated via the Online Chemical Database (OCHEM) server, integrating variables from the PaDEL, alvaDesc, Dragon7, MERA, and MORDRED libraries [34]. To prevent the “curse of dimensionality” and rule out uninformative variables, a rigorous pre-filtering protocol was applied: (1) descriptors with a pairwise correlation coefficient >0.95 were excluded; (2) descriptors maintaining constant or near-constant values in >80% of the dataset were removed. This reduction yielded a pool of 2,140 informative descriptors for subsequent model construction.

3.2. QSAR Model Construction and Regulatory Validation

The QSAR was established using the Genetic Algorithm-Multiple Linear Regression (GA-MLR) technique within the QSARINS (v2.2.4) platform [35,36]. The curated dataset was partitioned into a training set (n = 81, ≈ 90%) for feature selection and internal validation, and an independent external validation set (n = 8, ≈ 10%) to verify the model’s generalizability across diverse chemical scaffolds.

The statistical robustness of the resulting model was evaluated against the guidelines established by the Organization for Economic Co-operation and Development (OECD). Predictive accuracy and internal stability were quantified using the coefficient of determination (R²), cross-validated Q² (Leave-one-out and Leave-many-out), and the concordance correlation coefficient (CCC). The applicability domain (AD) was delineated through leverage analysis (Williams plot), with the leverage threshold (h*) calculated as 3(p’+1)/n (where p’ is the number of model descriptors and n is the training set size). Structural anchoring was confirmed by performing 2,000 iterations of Y-randomization, ensuring that the biological endpoints were not derived from chance correlations.

3.3. Rational Design and ADMET Evaluation

The validated GA-MLR model served as the quantitative engine for the rational design of eighteen fused quinoline derivatives. Structural modifications were systematically guided by the prioritized descriptors max_conj_path (extended π-conjugation), R8m (longitudinal reach), and MATS3s (electronic distribution). The prospective library was structured to include sixteen novel scaffolds and two internal benchmarks, Compounds 2 and 4 from the external validation set, to serve as predictive calibration anchors. Pharmacokinetic and toxicological profiles were predicted using the ADMETlab 3.0 server to evaluate the clinical viability of the designed leads [37]. Critical parameters included aqueous solubility (logS) measured in log mol/L, Caco-2 permeability quantified as log cm/s to assess intestinal absorption, and a Blood-Brain Barrier (BBB) score threshold of > 0.3 to confirm central nervous system accessibility. Synthetic accessibility was delineated on a 1–10 scale where 1 represents the highest ease of synthesis. Toxicological liabilities were quantified via the ProTox-II server to evaluate organ-specific hepatotoxicity and systemic probabilities for neurotoxicity, immunotoxicity, and cardiotoxicity relative to the reference drug Tacrine [38].

3.4. Molecular Docking and Molecular Dynamics Protocols

Molecular docking elucidated the binding orientations of prioritized Leads 19 and 20 within the human AChE crystal structure (PDB ID: 7XN1) at 2.85 Å resolution [39]. The protein was prepared by removing co-crystallized solvent and heteroatoms followed by the addition of polar hydrogens. The protein structure was processed using the CB-Dock2 cavity detection algorithm to automate the identification of optimal binding sites within the 20-Å enzyme gorge [40,41]. Scientific transparency was ensured by validating the docking protocol via self-docking of the co-crystallized ligand Tacrine; successful recovery of the experimental pose was defined by a RMSD of 0.326 Å, well within the 2.0 Å threshold for high-confidence binding mode prediction. Visualization of binding interfaces was conducted using ChimeraX [42] and ProteinPlus [43,44,45].

Long-range 100-ns MD simulations were conducted using the Desmond module in the Schrödinger Suite 2024-2 [46,47,48]. The lead-enzyme complexes were solvated in an orthorhombic box using the TIP3P water model with a 10 Å buffer. Systems were neutralized with Na⁺/Cl^- ions and subjected to the default Desmond relaxation protocol prior to production. Production runs were executed at 300 K and 1.01325 bar under the NPT ensemble utilizing the OPLS4 force field. Post-simulation analyses included Protein-Ligand RMSD, residue-level Root Mean Square Fluctuation (RMSF), and Radius of Gyration (rGyr) to evaluate temporal stability.

3.5. Post-Simulation Integrity and Functional Dynamics Analysis

The structural and thermodynamic consequences of ligand binding were evaluated using the MD-equilibrated trajectories. The stereochemical quality of the final 100-ns snapshots was verified through Ramachandran plot analysis using the MolProbity server, which quantified the distribution of (ϕ, Ψ) dihedral angles to ensure that binding within the narrow active-site gorge did not induce non-physical backbone strain [49]. Residues categorized as outliers were mapped back to the 3D structure to confirm their localization within flexible surface loops far from the binding interface.

The thermodynamic stability of the Lead 19 and Lead 20 complexes was further characterized by constructing the Free Energy Landscape (FEL). The FEL was generated by performing Principal Component Analysis (PCA) on the Cα coordinates to extract the primary modes of motion. The probability distribution of the trajectory across the first two principal components (PC1 and PC2) was used to calculate the relative Gibbs free energy (G) through the Boltzmann relationship, allowing for the identification of localized energy minima and the assessment of entropic instability in the ligand-bound states [50,51,52].

Functional synchrony between protein residues was quantified using the Dynamic Cross-Correlation Matrix (DCCM) [53]. Cross-correlation coefficients (C_ij) were calculated for all Cα pairs throughout the 100-ns trajectory to evaluate how the inhibitors influenced the native “breathing” motions of the enzyme. Positive correlations ( C_ij > 0) denoted synchronized residue movement, while negative values (C_ij < 0) identified anti-correlated motions. This analysis was critical for validating that Lead 19 preserves the essential anti-correlation of the Gly121-His447 pair required for the enzyme’s catalytic cycle.

5. Conclusions

This investigation successfully established a robust, statistically validated computational pipeline for the rational design of novel fused quinoline sulfonamides as dual-site acetylcholinesterase (AChE) inhibitors. By leveraging a five-descriptor GA-MLR QSAR model (R² = 0.7569, $Q_{LOO}^2$ = 0.7244), we demonstrated that inhibitory potency is fundamentally governed by the extension of π-conjugation (max_conj_path) and the precise spatial mass distribution (R8m) required to bridge the ~20 Å enzyme gorge. Among the eighteen analogues prioritized, Lead Compound 19 emerged as a superior candidate, achieving high-occupancy dual-site anchoring at both the Catalytic (CAS) and Peripheral (PAS) sites while maintaining an optimized pharmacokinetic profile suitable for central nervous system penetration. In contrast, Compound 20 exhibited inferior thermodynamic stability and reduced binding efficiency, a direct consequence of an unfavorable stereoelectronic clash with Ser125 and the entropic instability introduced by its longer n-pentyl chain.

The clinical potential of Lead 19 is further underscored by its ability to preserve the native functional synchrony of the enzyme, specifically the essential anti-correlated motions between the Gly121-His447 catalytic pair, while significantly reducing off-target risks for neurotoxicity and cardiotoxicity compared to the reference drug Tacrine. These findings provide a high-value structural scaffold to address the escalating global dementia crisis, particularly in high-burden demographic regions such as India. In alignment with the requirements of this Special Issue, this study provides the mechanistic and in silico foundation for the next phase of experimental validation. Future work will employ biomimetic chromatography, specifically retention analysis on Immobilized Artificial Membrane (IAM) and Human Serum Albumin (HSA) stationary phases to correlate these topological descriptors with experimental membrane permeability and protein-binding kinetics, advancing the high-throughput, animal-free screening of disease-modifying neuroprotective agents.