Steric‐Engineered Thermodynamic Gating: Metastable Assemblies for Chemical‐Class Discrimination of Synthetic Cannabinoids

ABSTRACT

Synthetic cannabinoids (SCs), a rapidly evolving class of new psychoactive substances (NPS), trigger severe neurotoxicity and fatalities while evading rapid, structure‐selective on‐site screening. To enable high‐fidelity recognition of such chemically inert targets, we propose a generalizable “Steric‐Engineered Thermodynamic Gating” strategy based on disaggregation‐induced emission (DIE) mechanism that breaks the intrinsic “stability‐sensitivity” deadlock in supramolecular sensing. Distinct from trial‐and‐error optimization, this approach rationally exploits steric bulk to induce active packing frustration, creating metastable aggregates designed to selectively detect SCs via synergistic non‐covalent interactions (e.g., π–π stacking and hydrogen bonding). Functioning as a thermodynamic filter, this assembly remains inert against non‐target interferents yet selectively undergoes cooperative disassembly upon binding with specific SCs via multivalent synergy, transforming a quenched “off” state into a robust blue‐shifted “on” signal. Validating this strategy with EDMB‐PINACA, the system exhibits ultrafast response (<1 s) and high sensitivity (LOD 4.7 µM); integrated into a 3D‐printed portable chip, it enables reliable, false‐positive‐free screening in authentic samples (e.g., e‐liquids, petals) with exceptional immunity to complex matrix interference. This work establishes a methodological blueprint for engineering aggregate metastability to recognize low‐reactivity analytes, offering a theoretical foundation for designing intelligent field‐deployable optics beyond the limitations of traditional molecular recognition.

A “Steric‐Engineered Thermodynamic Gating” strategy enables selective recognition of chemically inert synthetic cannabinoids through active packing frustration. The resulting metastable H‐aggregate acts as a spring‐loaded thermodynamic filter, unlocking only upon target binding to generate fluorescence turn‐on. This approach overcomes the stability‐sensitivity trade‐off in supramolecular sensing and supports rapid on‐site screening with deep‐learning‐assisted structural discrimination.

1. Introduction

The escalating abuse of illicit drugs, particularly synthetic cannabinoids (SCs), a major class of new psychoactive substances (NPS), poses a profound threat to global public health due to their rapid evolution and high potency [1, 2, 3, 4]. Designed to circumvent regulatory frameworks, these “designer drugs” act as potent cannabinoid receptor agonists, inducing severe neurotoxicity and psychiatric sequelae that claim hundreds of lives annually [5, 6, 7]. The continuous synthesis of novel SC scaffolds perpetually outpaces legal controls, creating a “cat‐and‐mouse” dynamic that exacerbates the difficulty of timely interception [8]. Compounding this challenge is their frequent concealment within diverse matrices—ranging from herbal blends to e‐liquids—which significantly complicates analytical isolation [9]. While laboratory techniques like liquid chromatography‐tandem mass spectrometry (LC‐MS/MS) provide definitive identification [10, 11], their reliance on bulky instrumentation renders them unsuitable for urgent on‐site screening scenarios. Consequently, optical sensing methods have emerged as promising alternatives, offering high spatiotemporal precision and portability for field deployment [12]. Given that newer generation SCs are typically “chemically inert” and lack reactive handles, pioneering advancements have successfully established discrete non‐covalent recognition paradigms [13]. These state‐of‐the‐art approaches primarily harness equilibrium‐governed host–guest interactions, utilizing mechanisms like metal‐coordination or charge transfer to capture targets via weak forces [14, 15, 16, 17]. However, relying solely on discrete, one‐to‐one binding events exposes a fundamental limitation when confronting families of structurally similar analytes, such as SCs: the “Specificity‐Generalization” paradox [18]. To differentiate subtle structural analogs (e.g., varying tail or head groups), classic discrete receptors require exquisite geometric complementarity (“lock‐and‐key”); yet, this structural rigidity inherently limits their ability to adapt to emerging, chemically distinct variants [19, 20]. Conversely, broadening the receptor's tolerance to cover a wider class of SCs inevitably invites non‐specific binding from matrix interferents, compromising signal reliability [21]. Therefore, developing a sensing platform that transcends rigid structural fitting to achieve both broad‐spectrum class coverage and high‐fidelity discrimination against false positives remains an unmet challenge.

Supramolecular self‐assembly of small molecules presents a compelling alternative to resolve this paradox by shifting the recognition paradigm from discrete geometric binding toward cooperative, phase‐transition‐like behaviors [22, 23]. Unlike single‐molecule sensors restricted to linear responses, self‐assembled nano‐aggregates are driven by collective intermolecular interactions (e.g., π−π stacking) [24], enabling non‐linear signal amplification through morphological changes [25]. Such cooperative systems typically exhibit sigmoidal response curves, where the signal remains dormant until a critical perturbation threshold is reached, providing an intrinsic mechanism to suppress low‐level background noise [26]. This strategy is epitomized by DIE, where the probe is kinetically trapped in a quenched H‐aggregate state (acting as a “silent” background) and only lights up upon catastrophic disassembly triggered by the analyte [27]. Theoretically, this disassembly process requires overcoming a distinct energetic barrier, functioning as a “thermodynamic filter” that differentiates analytes based on their overall binding energetics rather than mere geometric fit [28, 29]. This energy‐governed discrimination allows for the grouping of structurally diverse but thermodynamically similar analogs into distinct response classes, thereby achieving the elusive balance between specificity and generalization [30]. However, applying this strategy to non‐reactive targets like SCs presents a delicate thermodynamic dilemma: the aggregates must be sufficiently stable to ensure silence against interferents, yet labile enough to be disassembled by the weak noncovalent interactions offered by the analyte [25]. Newer generation SCs interact principally via weak hydrophobic forces, which are often thermodynamically insufficient to disrupt robustly packed aggregates, leading to false negatives; conversely, lowering the stability to accommodate SC sensitivity inevitably erodes the energy barrier against matrix dilution or solvent effects, causing false positives [31, 32, 33]. Navigating this intrinsic trade‐off between stability (to prevent false positives) and sensitivity (to ensure response) remains a critical frontier, necessitating a rational design paradigm beyond conventional trial‐and‐error optimization.

To resolve this thermodynamic deadlock, herein, we propose a “Steric‐Engineered Thermodynamic Gating” strategy that establishes a metastable supramolecular assembly capable of functioning as a structural filter, ensuring high‐fidelity recognition of SCs even amidst complex interferences. Distinct from passive aggregation approaches, we rationally designed a series of naphthalimide derivatives and identified the tert‐butoxycarbonyl‐functionalized probe (NAP‐Boc) as the optimal molecular scaffold, utilizing its steric bulk to deliberately frustrate packing and tune aggregate metastability. When validated against the representative analyte EDMB‐PINACA, this probe exhibited exceptional performance, featuring an ultrafast response (<1 s), a low detection limit (4.7 µM), and robust immunity to more than 20 potential interferents. Mechanistically, the probe operates on a cooperative unlocking principle: the labile, quenched aggregates are disassembled only by analytes possessing specific “head/tail” structural affinities, effectively categorizing diverse SCs while rejecting non‐specific binding. To further achieve fine‐grained discrimination beyond this chemical‐class categorization, we integrated a deep learning (DL) algorithm that decodes the unique “visual fingerprints” of these fluorescent states, successfully identifying structurally similar analogs with high accuracy. Bridging the gap to practical application, we engineered a portable sensing chip capable of reliably identifying targeted SCs within challenging real‐world samples with robust immunity to complex matrix interference, proving its suitability for field‐based screening. Consequently, this work not only provides a powerful tool for forensic analysis but also establishes a methodological blueprint for engineering aggregate metastability to recognize low‐reactivity analytes.

2. Results and Discussions

2.1. Design Principles and Operating Mechanism

Seminal advancements in fluorescent sensing for SCs have predominantly relied on discrete molecular recognition paradigms (Scheme 1a). While these molecular‐scale strategies have proven highly effective for specific targets, their rigid one‐to‐one binding nature often struggles to differentiate structural analogs (generalization) while remaining susceptible to non‐specific occupation by matrix interferents (specificity). DIE‐based supramolecular sensing strategies offer a compelling alternative by shifting to a cooperative phase‐transition model that silences background noise via aggregation‐caused quenching (Scheme 1b). However, for non‐reactive targets like SCs, this approach faces an intrinsic thermodynamic dilemma: aggregates must be sufficiently stable to remain silent against interferents, yet labile enough to be disassembled by weak non‐covalent analyte interactions [31].

SCHEME 1.

(a) Conventional discrete molecular recognition paradigms based on isolated probe–analyte interactions. (b) The general disaggregation‐induced emission (DIE) strategy. (c) The proposed “Steric‐Engineered Thermodynamic Gating” strategy.

To break this “stability‐sensitivity” deadlock, we propose a “Steric‐Engineered Thermodynamic Gating” strategy (Scheme 1c). Unlike passive aggregation, we rationally install a rigid steric modulator to induce frustrated intermolecular packing that disrupts optimal cofacial organization while still permitting aggregate formation, thereby deliberately trapping the probe in a “spring‐loaded” metastable state. This assembly functions as a stringent thermodynamic filter: it maintains a high energy barrier against non‐target interferents but undergoes cooperative disassembly upon binding with specific SCs. By decoupling specificity from stability, this mechanism enables high‐fidelity, structure‐selective detection.

2.2. Steric Engineering of a Metastable H‐Aggregate as a Thermodynamic Gate

To translate this steric‐engineering strategy into a tangible molecular probe, we selected the 4‐phenylamino‐1,8‐naphthalimide (NAP) scaffold as the signaling unit because it combines a rigid π‐conjugated framework, strong microenvironment‐sensitive fluorescence, and a synthetically accessible 4‐phenylamine‐N position, which allows steric effects to be systematically tuned on a common photophysical platform without altering the emissive core [34, 35, 36, 37]. To rationally construct the requisite thermodynamic gate, we functionalized the 4‐phenylamino position with a bulky tert‐butoxycarbonyl (Boc) group (Scheme S1). This steric modification serves as a critical structural modulator: rather than preventing aggregation itself, it is intended to frustrate the formation of overly optimized, thermodynamically stable packing. As a result, the molecules are biased toward a metastable, less efficiently packed aggregate state, corresponding to a shallow local free‐energy minimum rather than the deepest thermodynamic minimum [38]. This distinction is essential for sensing: the probe must aggregate sufficiently to generate a quenched “Off” state, yet the aggregate must remain labile enough to undergo analyte‐triggered disassembly or reorganization. In this way, the Boc group creates a thermodynamic window in which aggregation is preserved, but over‐stabilization is avoided.

To validate this design and optimize the steric parameters, we synthesized a focused library of four naphthalimide derivatives by systematically varying the substituent at the 4‐phenylamine‐N position (NAP‐Ac, NAP‐MC, NAP‐Boc, NAP‐Cbz; Figure 1a and Figure S1–S15). These substituents were selected as a compact yet systematic set to modulate steric bulk, three‐dimensional shape, conformational flexibility, and auxiliary aromaticity at the same position while preserving the NAP emissive core. This modulation was intended to tune intermolecular π–π packing and balance the dual requirements of robust aggregation‐caused quenching (ACQ) and analyte‐triggered disaggregation. In this series, Ac and MC represent relatively compact substituents that were expected to permit closer intermolecular packing, whereas Boc was introduced as the key steric modulator because its bulky three‐dimensional tert‐butoxycarbonyl unit was expected to induce packing frustration and generate the targeted metastable aggregate state. Cbz was included as an important comparison because its more extended and conformationally adaptable structure, together with its additional aromatic group, allowed us to distinguish simple steric enlargement from steric topology and packing mode. As expected, all four probes exhibited strong intrinsic fluorescence in DMSO (Figure S16). In the sensing medium (DMSO/water, 1:6 v/v), the four probes exhibited differentiated photophysical behaviors, consistent with distinct aggregation states (Figures 1b and S17): NAP‐Ac, NAP‐MC, and NAP‐Boc displayed pronounced ACQ, consistent with the formation of H‐type aggregates. In contrast, NAP‐Cbz remained highly emissive, likely due to the formation of fluorescent J‐type aggregates [39], thus failing to establish a dark “Off” state.

FIGURE 1.

Rational design of a steric‐engineered thermodynamic gate for selective SC recognition. (a) Chemical structures of the naphthalimide‐based probes featuring substituents with increasing steric bulk. (b) Schematic of the three aggregation regimes in aqueous media (DMSO/H₂O, 1:6): over‐stabilized “deep sink” aggregates (NAP‐Ac/MC), the engineered metastable H‐aggregate (NAP‐Boc), and emissive J‐aggregates (NAP‐Cbz). (c) Schematic illustration of the analyte‐triggered selective disaggregation mechanism. (d) Fluorescence emission spectra of the four probes in the absence (blue lines) and presence (pink lines) of EDMB‐PINACA (sensing medium: DMSO/water, 1:6 v/v).

Based on these distinct aggregation behaviors, we hypothesized that the sensing performance would be governed by the thermodynamic stability of these aggregates: the over‐stabilized aggregates of NAP‐Ac/MC and the emissive J‐aggregates of NAP‐Cbz would remain structurally inert, whereas the metastable NAP‐Boc aggregates—sitting in a shallower energy well—would be selectively unlocked by the target (Figure 1c). Experimental validation confirmed this hypothesis (Figure 1d, sensing medium: DMSO/water, 1:6 v/v). Challenge with a representative SC (EDMB‐PINACA, 200 µg/mL in methanol) revealed stark differences. NAP‐Ac and NAP‐MC remained quenched, while NAP‐Cbz showed only marginal intensity change despite its bright baseline, accompanied by a slight blue shift in emission. Solvatochromic studies indicated that NAP‐Cbz exhibits positive solvatochromism (Figure S18), suggesting that this blue shift arises from a local decrease in microenvironment polarity upon analyte association rather than productive disaggregation from a dark state. In contrast, NAP‐Boc exhibited dramatic fluorescence turn‐on (∼57‐fold F/F ₀; Figures 1d and S19). Control experiments confirmed this enhancement was analyte‐specific, with methanol vehicle alone inducing no response (Figure S20). Quantitative signal‐to‐background ratios unequivocally identified NAP‐Boc as the only candidate achieving both a robustly quenched “Off” state and a sensitive, high‐contrast “On” response. This screening underscores that optimal performance does not arise from maximal monomeric brightness, but from deliberate positioning of the aggregate in a metastable, “spring‐loaded” shallow energy minimum precisely engineered by the bulky tert‐butoxycarbonyl substituent.

2.3. Characterization of the Quenched Metastable Off‐State

Having identified NAP‐Boc as the optimal thermodynamic gate, we characterized its quenched aggregate state. Fluorescence titration with increasing water fraction (f _w) at our standard working concentration (∼8.57 µM) revealed profound ACQ, reaching minimal background at a water fraction (f _w) near DMSO/H₂O 1:2 (Figure 2a, black curve). However, optimal sensing required balancing low background against maximal absolute turn‐on intensity. While the highest F/F ₀ occurred at 1:5, peak emissive response upon EDMB‐PINACA addition was achieved at 1:6 (Figures 2a, red curve and S21). We selected this ratio for all subsequent experiments, as it positions the aggregate in a labile state—quenched yet readily disassembled—representing the ideal shallow energy minimum in contrast to the overly stable “deep energy sink” at lower water fractions.

FIGURE 2.

Mechanistic elucidation of the metastable assembly and cooperative recognition. (a) Fluorescence intensity of NAP‐Boc (at working concentration) in the absence (black) and presence (red) of EDMB‐PINACA across varying DMSO/H₂O volume ratios, identifying 1:6 as the optimal medium. (b) UV–vis absorption spectra of NAP‐Boc in different solvent mixtures, revealing the aggregation‐caused quenching (ACQ) trend and H‐aggregate formation. (c) DLS size distribution of NAP‐Boc in pure DMSO versus the 1:6 sensing medium. (d) Schematic illustration of the NAP‐Boc/EDMB‐PINACA binding mode. (e) Temperature‐dependent UV–vis spectra of the NAP‐Boc probe aggregates (25°C–95°C). (f) Time‐resolved fluorescence decay profiles of the monomer/pre‐aggregate (DMSO, 553 nm), quenched aggregate (1:6 mix, 533 nm), and analyte‐bound complex (495 nm). (g) Hill plot analysis of the fluorescence titration data. (h) Steady‐state fluorescence spectra comparing the distinct emission states. (i) Electrostatic potential (ESP) surfaces of (i) NAP‐Boc and the control probe (ii) NAP‐Cbz. (j) Independent Gradient Model (IGM) analysis of the NAP‐Boc/EDMB‐PINACA complex.

To provide physical evidence for this aggregation‐based model, we turned to UV–vis absorption and fluorescence spectroscopy. A solvent titration experiment at a higher final concentration (∼50 µM, to ensure effective absorbance) revealed a significant hypsochromic (blue) shift of the main absorption band from ∼253 nm to 243 nm upon water addition (Figure 2b), a hallmark of H‐type excitonic coupling [28] resulting from face‐to‐face π–π stacking between the naphthalimide chromophores. Strikingly, at this same high concentration, the fluorescence behavior of the probe NAP‐Boc exhibited a more complex ACQ‐to‐AIE‐like transition (Figure S22), where the initial quenching region perfectly correlates with the formation of the H‐aggregate.

To visually corroborate the aggregation states, we characterized the probe using DLS and TEM. Unexpectedly, characterization of NAP‐Boc in DMSO revealed the presence of mesoscale assemblies with hydrodynamic diameters ranging from 400 nm to 600 nm (Figures 2c and S23a). Despite this significant physical size, the solution exhibited strong fluorescence emission at ∼553 nm. This implies that in DMSO, the highly hydrophobic probes form loose, solvent‐swollen clusters rather than tightly packed aggregates. In this state, the intermolecular packing is too disordered to facilitate the π–π coupling required for ACQ, thus preserving high emissivity characteristic of the monomeric species. Upon the addition of water (DMSO/water 1:6 v/v), the system underwent a significant structural transformation, as evidenced by both DLS (mean diameter ∼900 nm) (Figure 2c) and TEM (particles up to ∼1000 nm) (Figure S23b). The increased particle size indicates that the probe aggregates into larger, more compact structures, likely due to the increased hydrophobic interactions at higher water fractions. This large‐scale aggregation was accompanied by pronounced fluorescence quenching and a blue‐shift of the emission peak to ∼533 nm, consistent with the formation of quenched H‐aggregates. This transformation marks the establishment of the “Off” state for sensing.

2.4. Selective Disaggregation by SCs: Overcoming the Thermodynamic Threshold Via Multivalent Synergy

Having established the quenched metastable H‐aggregate as the Off state, we next investigated how the representative SC EDMB‐PINACA selectively overcomes the disaggregation barrier. This selective process relies on the formation of a stable complex, in which the highly hydrophobic SC EDMB‐PINACA (bearing ester head, indazole core, and nonpolar hydrophobic tail) engages the probe through multivalent noncovalent interactions—including π–π stacking, hydrogen bonding, and desolvation‐driven hydrophobic sequestration—thereby disrupting the labile packing and restoring the emissive state (Figure 2d). To directly visualize the structural consequence of analyte binding, DLS and TEM were performed after treatment of the NAP‐Boc aggregates with EDMB‐PINACA in the DMSO/water (1:6, v/v) sensing medium. DLS showed that the size distribution shifted markedly to smaller dimensions centered at ca. 200–250 nm, indicating substantial disruption of the large metastable assemblies in solution (Figure S24). Consistently, TEM further revealed that the large and compact aggregates observed in the absence of analyte were converted into much smaller and more dispersed nanostructures after EDMB‐PINACA addition (Figure S23c), providing direct morphological evidence for analyte‐triggered disaggregation/reorganization of the metastable H‐aggregate.

To experimentally probe the lability of this engineered metastable aggregate and validate its shallow energy minimum character, we further employed variable‐temperature (VT) spectroscopy as a non‐specific perturbation tool. VT‐UV–vis spectroscopy revealed remarkable structural plasticity. Unlike a simple monotonic dissociation, the spectra exhibited a complex, non‐linear evolution upon heating from 25°C to 95°C (Figure 2e and Figure S25a). Specifically, the absorbance underwent a distinct biphasic transition (a “V‐shaped” trend) (Figure S25b, black curve), indicating that the aggregate traverses through intermediate states rather than simply melting into monomers. This sensitive responsiveness to thermal perturbation serves as a hallmark of a labile system residing in a shallow energy minimum [22, 40]. In stark contrast, the VT‐UV–vis spectra of the underperforming, robustly aggregated NAP‐Cbz showed only a minor, monotonic decrease in absorbance (Figure S25a,b, red curve), characteristic of a thermodynamically deep, stable energy sink that is resistant to conformational reorganization.

Corresponding VT‐fluorescence measurements provided further insight into the energetics. Heating of both NAP‐Boc and NAP‐Cbz solutions resulted in progressive fluorescence quenching accompanied by a blue shift (Figure S26). This shift is likely attributed to the initial disruption of minor emissive aggregate species at lower temperatures [41]. While aggregate disassembly theoretically restores emission, this outcome is not guaranteed in amino‐/arylamino‐naphthalimide systems, because heating simultaneously promotes intramolecular motion, vibrational relaxation, and solvent‐assisted non‐radiative decay [42]. Thus, in the present case, the intrinsic thermal quenching of the excited state outweighs any fluorescence recovery arising from partial aggregate loosening/disassembly, effectively masking any potential turn‐on signal. However, a critical distinction emerges in their thermal sensitivity. A normalized intensity plot reveals that NAP‐Boc exhibits a significantly steeper fluorescence decay compared to NAP‐Cbz (Figure S26c). This heightened sensitivity corroborates its metastable, loosely packed nature, in which weaker packing constraints allow greater chromophore mobility and thereby facilitate access to thermally activated non‐radiative decay channels upon heating. Crucially, this result also emphasizes a mechanistic safeguard: non‐specific thermal energy, while sufficient to perturb the labile NAP‐Boc aggregate (as seen in UV–vis and steep fluorescence decay), cannot replicate the specific “turn‐on” triggered by the analyte. This suggests that the energy barrier for productive disassembly is selectively overcome by the specific chemical binding energy of the target, not merely by bulk environmental heating.

Time‐resolved fluorescence decay measurements provided additional dynamic evidence for the differential aggregate stability (Figure 2f). In the quenched H‐aggregate state (DMSO/H₂O 1:6), NAP‐Boc exhibited a markedly prolonged lifetime (τ _{1/ e} = 5.51 ns) compared to its monomeric/pre‐aggregated form in DMSO (2.40 ns), consistent with restricted molecular motion in rigid face‐to‐face packing. Upon addition of EDMB‐PINACA, the lifetime shortened to 4.29 ns, indicating disruption of this rigid environment and restoration of motional freedom—yet remaining longer than the free monomer, supportive of probe sequestration within a constrained hydrophobic pocket. In sharp contrast, NAP‐Cbz showed only modest lifetime elongation in the aqueous mixture (3.80 ns) with further slight increase upon analyte addition (4.17 ns), reflecting its inability to form a similarly rigid, analyte‐disruptable aggregate (Figure S27). These results independently corroborate the metastable, “spring‐loaded” nature of the NAP‐Boc H‐aggregate.

Furthermore, we confirmed that this chemically driven disassembly operates via a cooperative mechanism. Analysis of the fluorescence titration data using the Hill equation yielded a coefficient (n) of ∼1.5 (Figure 2g). A value of n > 1 signifies positive cooperativity, consistent with a cascade‐like “unzipping” process. This suggests that the initial binding of an analyte molecule induces a local conformational loosening that lowers the energy barrier for subsequent binding events. This cooperativity is a direct functional consequence of the aggregate's metastable nature, enabling the amplified, switch‐like response essential for high‐sensitivity detection.

Spectroscopic signatures of the “On” state provided crucial insight into why EDMB‐PINACA can surmount the threshold. The emission maximum underwent a unique three‐stage spectral transition: from ∼553 nm (loose clusters in polar DMSO) to a dimmed ∼533 nm (H‐aggregates in aqueous mixture), and finally to a bright, strongly blue‐shifted peak at ∼495 nm upon analyte binding (Figures 2h and S28). This significant shift to 495 nm—surpassing even the monomeric emission in DMSO—is critical. It suggests that the probe does not merely disassemble back into the bulk solvent; rather, it is encapsulated within the distinct hydrophobic pocket of the SC analyte. This hypothesis is supported by the probe's positive solvatochromism (Figure S18a), where non‐polar solvents (e.g., toluene) similarly induce a blue shift, confirming that the “On” signal arises from the specific sequestration of the probe in a non‐polar microenvironment.

Thus, representative SC EDMB‐PINACA uniquely provides the multivalent noncovalent synergy (π–π stacking, hydrogen bonding, and desolvation‐driven hydrophobic sequestration) required to compensate aggregate disruption and drive the equilibrium toward the emissive complex—interactions insufficiently matched by common interferents.

2.5. Theoretical Rationale for Metastability and Selective Unlocking

Density functional theory (DFT) calculations rationalized the steric origin of metastability. A comparative analysis of steric properties (Figure S29 and Table S1) revealed that while NAP‐Cbz possesses a larger total molecular volume, its greater conformational adaptability arises not simply from steric size, but from a broader accessible conformational space. Specifically, relaxed dihedral potential energy scans (Figure S30) showed that the carbamate‐side torsion of NAP‐Cbz [C(carbonyl)‐O‐CH₂‐Cipso] spans a broad low‐energy conformational basin, rather than a single narrowly defined minimum. In addition, compared with NAP‐Boc, NAP‐Cbz contains one additional packing‐relevant rotatable bond (Figure S31), namely the benzyl CH₂‐Ph torsion [O‐CH₂‐Cipso‐Cortho], whose shallow energy variation indicates that the pendant phenyl ring can readily reorient with only a small energetic penalty (ca. 1 kcal mol⁻¹). At the isolated‐molecule level, these features allow the Cbz substituent to access several similar‐energy conformations and thus adapt its side‐chain orientation more readily. At the aggregate level, this extra torsional degree of freedom is expected to facilitate local conformational accommodation of the benzyl group and relieve steric mismatch during intermolecular packing, thereby favoring more stable packing arrangements.

In contrast, NAP‐Boc's Boc group presents a compact, quasi‐spherical bulky 3D shield (Figure S32). The corresponding carbamate‐side scan [C(carbonyl)‐O‐Cq‐Cmethyl] displayed a regular threefold rotational profile with moderate barriers (ca. 4–5 kcal mol⁻¹), indicating that although the Boc unit is not completely rigid, its steric bulk remains concentrated in the tert‐butyl group and lacks the auxiliary low‐cost aromatic reorientation available to Cbz. This pronounced steric hindrance frustrates efficient π–π stacking [43], preventing the molecules from achieving a thermodynamically favorable, tightly packed state. Consequently, the system is forced into a structurally strained aggregate, which resides in a shallow energy minimum—a conclusion experimentally corroborated by its low thermal stability in VT experiments. This is distinct from NAP‐Ac and NAP‐MC, whose less bulky substituents foster tighter, more stable aggregates [44]. Thus, the rational introduction of the 3D Boc group is the key design element that dictates the formation of this highly responsive, labile assembly.

This steric‐driven hypothesis is further supported by electrostatic potential (ESP) maps (Figure 2i), which show that NAP‐Boc and NAP‐Cbz have similar electronic profiles. This similarity confirms that steric hindrance, rather than electronic distribution, is the dominant factor governing their differential aggregation behavior. Finally, to understand the driving force for the “turn‐on” process, an Independent Gradient Model (IGM) analysis was performed on the NAP‐Boc‐EDMB‐PINACA complex (Figures 2j and S33). The analysis visualized the extensive noncovalent interaction regions in the complex, in which the interaction region spanning the two aromatic frameworks is assigned to π–π stacking, the more localized interaction region involving the heteroatom‐containing groups is assigned to hydrogen bonding, and the interaction region associated with the nonpolar moieties is assigned to hydrophobic interaction. These robust noncovalent interactions provide the necessary enthalpic stabilization to compensate for the energetic cost of disrupting the metastable probe aggregates, thereby energetically driving the disassembly‐binding equilibrium toward the highly emissive complex.

2.6. Analytical Performance of the Thermodynamic‐Gated Probe NAP‐Boc

Having established that NAP‐Boc forms a sterically frustrated, metastable H‐aggregate that operates as a spring‐loaded thermodynamic gate, and that productive fluorescence turn‐on requires analyte‐provided binding energy to trigger cooperative disassembly, we next quantified the analytical figures of merit that this mechanism should inherently enable. EDMB‐PINACA was selected as a representative target to benchmark sensitivity, selectivity/anti‐interference, response kinetics, and operational robustness under the optimized sensing conditions (NAP‐Boc, 8.57 µM in 1:6 DMSO/water, Figure S34 and S35).

A fluorescence titration was performed by stepwise addition of increasing concentrations of EDMB‐PINACA to the probe solution. As anticipated for a DIE process, the system underwent a distinct visual transition from a non‐emissive “Off” state to a bright cyan‐green fluorescent “On” state under 365 nm UV irradiation (Figure 3a), consistent with unlocking of the quenched metastable aggregate. Spectroscopically, EDMB‐PINACA induced a progressive enhancement of the emission band centered at ∼495 nm (Figure 3b). Notably, the calibration curve (Figure 3c) revealed a segmented response characterized by two distinct linear regions, mirroring the cooperative nature of the disassembly mechanism. The initial stage (0–80 µg/mL, corresponding to 0–214 µM) exhibited a stable linear dependence (R² = 0.9951), representing the probe's response prior to the full cooperative threshold. To ensure a rigorous analytical evaluation, the limit of detection (LOD) was derived from the slope of this conservative initial region and determined to be 1.76 µg/mL (approximately 4.7 µM). As summarized in Table S2, this LOD falls within a competitive range among reported rapid optical/fluorescence methods for synthetic cannabinoid detection. This micromolar‐level sensitivity highlights the high efficiency of the analyte‐induced disassembly process, rendering the probe suitable for trace‐level detection.

FIGURE 3.

Evaluation of the sensing performance of NAP‐Boc. (a) Visual fluorescence photographs of the probe solution in the presence of varying concentrations of EDMB‐PINACA. (b) Fluorescence titration spectra recorded with the stepwise addition of the EDMB‐PINACA (0–200 µg/mL). (c) Linear correlation between the emission intensity and EDMB‐PINACA concentration. (d) Fluorescence spectra demonstrating the high selectivity of the probe against potential competing species. (e) Selectivity of the probe against potential competing species, with corresponding chemical structures inserted. The bars represent the fluorescence intensity in the presence of interferents alone. Note: Unless otherwise stated, λ _ex = 365 nm. Error bars represent standard deviation (n = 3). S01: AB‐PINACA, S02: CUMYL‐THPINACA, S03: ADB‐FUBINACA, S04: 5F‐EMB‐PICA, S05: MDMB‐4en‐PINACA butanoic acid, S06: 4OH‐MDMB‐BUTICA, S07: 5F‐MDMB‐PICA, and S08: CUMYL‐PIPETINACA.

In practical forensic scenarios, discriminating the target from complex chemical backgrounds is paramount. Mechanistically, the thermodynamic‐gate model predicts that non‐target species lacking sufficient multivalent noncovalent synergy should fail to cross the disaggregation threshold and therefore remain silent. To validate this structural stringency, we analyzed the chemical topologies of potential interferents. While EDMB‐PINACA possesses the hydrophobic ester head and extended nonpolar tail that can provide sufficient noncovalent synergy to trigger the response, other species faced critical structural barriers. SC analogs (S01–S08) with minor modifications—such as the absence of a hydrophobic tail or mismatched headgroups—incurred high desolvation penalties to exhibit low affinity. Similarly, common illicit drugs (e.g., methamphetamine, morphine) and SC precursors (indazole, indole) were excluded due to their high polarity or small, non‐planar aromatic systems, which result in weak π–π stacking interactions. Consequently, when we challenged NAP‐Boc with a broad panel of 20 potential interferents, even at a 5‐fold excess relative to EDMB‐PINACA, none of these competing agents produced a significant fluorescence response (Figures 3d,e and S36), demonstrating exceptional selectivity and a low propensity for false positives.

To further probe whether surfactant‐like or amphiphilic molecules could nonspecifically disrupt the metastable assembly, we additionally examined representative surfactants, hydrophobic aromatic molecules, and amphiphilic long‐chain compounds, including anionic, cationic, and nonionic surfactants as well as naphthalene and sodium oleate. Even when individually introduced at a 5‐fold excess relative to EDMB‐PINACA, none of these interferents produced a significant fluorescence response (Figure S37), indicating that nonspecific perturbation of the metastable assembly is insufficient to activate the probe. Importantly, anti‐interference experiments further showed that the sensing output for EDMB‐PINACA remained uncompromised in complex mixtures: in the presence of high concentrations of co‐existing interferents, including the above surfactant‐like and amphiphilic challengers, the fluorescence signal of EDMB‐PINACA maintained robust intensity (Figures S36 and S37), confirming that the disassembly driven turn‐on is resilient to co‐existing interference and matrix effects.

Accordingly, the high‐fidelity response of NAP‐Boc toward EDMB‐PINACA can be understood as a recognition process in which the cooperative binding energy of the analyte overcomes the metastable aggregate's disassembly threshold, rather than a conventional single‐site lock‐and‐key event. The Boc‐induced packing frustration positions the quenched H‐aggregate near a disassembly/reorganization threshold, while still maintaining a sufficient barrier against nonspecific hydrophobic or amphiphilic perturbations. EDMB‐PINACA can overcome this barrier because its aromatic indazole/ester‐containing head region and hydrophobic segment engage the NAP‐Boc assembly through a cooperative combination of π‐association, localized hydrogen‐bonding/polar interactions, and hydrophobic stabilization. This multivalent head/tail matching compensates for the energetic cost of disrupting the metastable aggregate and drives the equilibrium toward the emissive complex. In contrast, molecules lacking this coordinated interaction pattern fail to efficiently unlock the thermodynamic gate. Together, these results highlight that the selectivity and anti‐interference performance of NAP‐Boc are directly rooted in the sterically frustrated metastable aggregate and its requirement for specific multivalent analyte interactions. Thus, the design of NAP‐Boc‐like probes should focus on matching aggregate metastability and steric topology with the multivalent recognition pattern of the target analyte class, rather than relying solely on steric bulk or a single high‐affinity interaction.

For on‐site applications, response speed is equally critical. Time‐dependent fluorescence measurements revealed that the signal reached equilibrium within 1 s upon analyte addition (Figure S38), indicating an almost instantaneous recognition–disassembly process, as expected for a “spring‐loaded” metastable aggregate poised for rapid unlocking. In addition, the sensing system exhibited remarkable stability across an exceptionally broad pH window (pH 1.0–14.0) (Figure S39). Although the absolute fluorescence intensity shows some pH‐dependent variation, the analyte‐induced turn‐on response is preserved throughout the tested range. Consistently, DLS measurements at pH 4, 7, and 9 showed highly similar size distributions (Figure S40), indicating that the metastable aggregates retain essentially unchanged structural characteristics under representative acidic, neutral, and basic conditions. This outstanding pH tolerance highlights a key advantage of the thermodynamic gating strategy: unlike reactive probes that can suffer hydrolysis or pH‐dependent side reactions, the present response is governed by robust non‐covalent interactions (notably hydrophobic effects and π–π interactions) that preserve the off/on gating behavior across a wide pH range.

We further assessed the operational stability of the sensing system. While the premixed DMSO/H₂O working solution is best used freshly prepared and within several hours (Figure S41), the NAP‐Boc stock solution in pure DMSO can be stored for at least 15 days and still provides a clear fluorescence response after on‐demand dilution with water (Figure S42). Collectively, the combination of high sensitivity, stringent selectivity with strong anti‐interference capability, ultrafast response, wide pH robustness, and good storage stability establishes NAP‐Boc as a practical and reliable platform for real‐world forensic screening.

2.7. Structure‐Driven Hierarchical Recognition and Deep Learning‐Assisted Discrimination

Having validated the analytical performance of the thermodynamic gate—demonstrating that only analytes capable of providing sufficient multivalent noncovalent synergy can overcome the disaggregation barrier of the metastable H‐aggregate—we next explored whether structural variations among SCs systematically modulate the unlocked “On”‐state microenvironment, generating class‐ and molecule‐specific optical fingerprints. This extension elevates the platform from binary detection of a single representative analyte to hierarchical recognition of an entire structurally diverse class, directly reflecting the precision of the steric‐engineered gating mechanism.

When challenged with a diverse library of SCs, NAP‐Boc exhibited a clear tiered response governed by strict structural compatibility with the probe's labile lattice (Figure 4a–c). Class A analytes possessing the canonical dual‐feature architecture—an accepted head group (ketone or ester) paired with a highly hydrophobic tail (e.g., EDMB‐PINACA, JWH‐018)—efficiently triggered cooperative disassembly of the metastable aggregates, producing robust fluorescence turn‐on even at low concentrations (200 µg/mL, Figure 4d). In contrast, analogs with reduced hydrophobicity (Class B; ester heads with polar tails) exhibited significantly attenuated affinities with variable activation thresholds. While representative Class B members can force assembly disruption at elevated concentrations (e.g., 5F‐EDMB‐PICA), it is noteworthy that this category encompasses a spectrum of binding energies. For the specific Class B analogs (e.g., 4OH‐MDMB‐BUTICA, 5F‐MDMB‐PICA) tested in our specificity panel (Figure 3e), the energetic penalty for insertion remained insurmountable even at a 5‐fold excess, explaining their silence under operational screening conditions. Meanwhile, those bearing H‐bond donating groups (Class C; acid/amide heads, e.g., metabolites) were categorically rejected (Figure 4e).

FIGURE 4.

Structural classification and response profiling of synthetic cannabinoids. (a–c) Chemical structures of SCs categorized into (a) high‐affinity Class A, (b) low‐affinity Class B, and (c) non‐responsive Class C. (d) Fluorescence emission spectra and corresponding photographs of the probe with the 15 representative SCs. (e) Schematic classification quadrant distinguishing Class A, B, and C based on headgroup functionality and tail hydrophobicity. (f and g) Physicochemical property analysis showing the distribution of (f) Log P and (g) –Log S values across the three identified classes. Note: A01: MDMB‐CHMINACA, A02: MDMB‐CHMICA, A03: JWH‐018, A04: EDMB‐CHMICA, A05: MDMB‐CHMCZCA, B01: 5F‐ADB, B02: EMB‐FUBINACA, B03: 5F‐EDMB‐PICA, B04: AMB‐FUBICA, B05: 5F‐CUMYL‐PINACA, C01: 5F‐AB‐PINACA, C02: ADB‐BUTINACA, C03: ADB‐CHMINACA, C04: AB‐PINACA, and C05: ADB‐FUBINACA.

This hierarchical selectivity broadly correlates with the interplay of partition coefficient (Log P) (Figure 4f), aqueous solubility (Log S, Figure 4g), and topological polar surface area (tPSA) (Figure S43), although strict numerical thresholds define general trends rather than absolute boundaries. The categorical rejection of amide‐bearing Class C metabolites is clearly rationalized by their distinct physicochemical profile: they exhibit significantly higher tPSA and water solubility (higher Log S, lower Log P) compared to Class A/B. This substantial hydrophilicity imposes a prohibitive energetic penalty for the desolvation required to enter the probe's hydrophobic pocket. The discrimination between responsive Class A and the variable responsiveness of Class B, despite overlapping bulk hydrophobicity parameters, further underscores the gate's sensitivity to precise molecular motifs: terminal polar substituents (e.g., fluorine) in Class B likely introduce localized steric or electrostatic mismatches that hinder effective wedging into the shallow energy minimum of the NAP‐Boc assembly. Thus, the recognition mechanism relies on precise molecular packing rather than simple polarity partitioning.

While the probe excels at broad chemical categorization, a more profound scientific question arises: does the specific probe–analyte interaction generate unique optical signatures for structurally isomorphic Class A analogs? These analogs, despite sharing the similar core skeleton, emit in a general cyan‐green region that appears visually similar to the naked eye (Figure 5a). However, we hypothesized that the distinct side‐chain variations of each SC molecule would induce subtle, unique perturbations in the disassembly kinetics and the resulting micromorphology of the aggregates, creating “visual fingerprints” embedded within the fluorescence images (Figure 5b). To decode these high‐dimensional features and validate the probe's structural resolution, we coupled the sensor platform with a ResNet34 Convolutional Neural Network (CNN) (Figure S44). The objective was to transform subtle optical nuances into mathematically distinct predictions. The training dynamics (Figure 5c) provided compelling evidence for the distinctness of these features: the identification accuracy rapidly approached ∼1.0, and the cross‐entropy loss dropped to near‐zero within the first three epochs. Although minor fluctuations occurred at epoch 11, the model achieved stable convergence between epochs 12 and 15. This rapid learning curve suggests that the fluorescence patterns generated by different SCs are not ambiguous but possess highly distinguishable topological features that the CNN can easily extract.

FIGURE 5.

Deep learning‐enabled fine‐grained discrimination of structurally similar synthetic cannabinoids. (a) Chemical structures of the 10 representative SC analogs selected for the discrimination study. (b) Corresponding fluorescence photographs exhibiting the visual response patterns of the probe to these targets. (c) Training and validation accuracy/loss curves of the ResNet34 model. (d) Confusion matrix for multi‐class classification of 10 structurally similar SC analogs. Notably, AM‐2201 behaves as a responsive analyte at 200 µg/mL despite its Class B designation, as the ketone head provides the necessary binding energy to surmount the disaggregation threshold.

We subsequently evaluated the classification performance against a library of 10 structurally isomorphic SCs using the test dataset. The resulting confusion matrix (Figures 5d and S45) reveals an exceptional level of discriminability, with 100% identification accuracy for 9 out of 10 analytes. A minor misclassification was noted only for MDMB‐CHMICA (5% misidentified as PB‐22), yet the overall aggregate accuracy remained at 99.5%. These results serve as a powerful validation of our mechanism: they confirm that the metastable NAP‐Boc probe does not merely respond to a “class” of molecules, but interacts specifically with each unique chemical structure. By translating these specific molecular interactions into discernible optical fingerprints, the platform achieves a level of “chemical fidelity” that enables the precise identification of highly similar illicit substances, far surpassing the capabilities of traditional threshold‐based sensors.

2.8. Portable Device and On‐Site Detection in Complex Real Samples

With hierarchical chemical recognition and deep learning discrimination established as natural extensions of the thermodynamic gating mechanism, we finally demonstrated the platform's robustness in complex real‐world matrices through portable devices and on‐site testing. To enable practical on‐site screening, we designed and 3D‐printed a compact, portable detection device with a compartmentalized architecture (Figure 6a). It includes a lower compartment for probe solution, cleaning solvent, and waste‐liquid reservoir, and an upper mesh platform holding two silicon wafer substrates (Figure S46). A flip‐top lid with integrated 365 nm UV LED strip ensures uniform excitation and light shielding. The design facilitates easy sample loading, simple rinsing, and safe waste containment, making it highly practical for on‐site detection (Figure 6b,c). In practical operation, the device (Figure 6d) was used in an on‐demand mode rather than as a pre‐coated long‐term storage format: a freshly prepared NAP‐Boc working solution was deposited onto the silicon wafer immediately before measurement. After each assay, the wafer could be regenerated by ethanol rinsing for repeated use.

FIGURE 6.

Portable platform for on‐site detection. (a) Schematic illustration of the 3D‐printed attachment integrating a UV excitation source (365 nm) and a silicon wafer substrate with a probe solution placeholder. (b) Schematic diagram of the functionalization of the probe on the silicon substrate. (c) Operation process of the portable device. (d) Real picture of the portable device. (e) Representative fluorescence images of the probe‐coated silicon wafers upon exposure to increasing concentrations of EDMB‐PINACA (0–200 µg/mL). (f) Rapid visual detection of EDMB‐PINACA spiked into diverse suspected carriers, including (i) E‐liquid, (ii) petals, and (iii) Green tea. (g) Selectivity performance in complex matrices, comparing the specific response of EDMB‐PINACA against negligible signals from potential interferents extracted from the same matrices (E‐liquid, petals, and green tea). S01: AB‐PINACA, S02: CUMYL‐THPINACA, S03: ADB‐FUBINACA, S04: 5F‐EMB‐PICA, S05: MDMB‐4en‐PINACA butanoic acid, S06: 5F‐MDMB‐PICA, and S07: CUMYL‐PIPETINACA.

Performance of the integrated platform was first benchmarked quantitatively using EDMB‐PINACA. Gradient concentrations were applied to probe‐loaded silicon wafers, yielding a clear, concentration‐dependent visual transition from dark to bright cyan‐green fluorescence (Figures 6e and S47). The naked‐eye limit of detection was estimated at ∼40 µg/mL, a sensitivity that remains well‐suited for practical forensic screening thresholds. The ultimate test of a field‐deployable sensor is its performance in complex, real‐world scenarios (Figure S48). To simulate this, we challenged the platform with various common SC carriers. Samples of EDMB‐PINACA were spiked into diverse matrices including e‐liquid, flower petals, and green tea. For detection, solid samples were mechanically processed (cut or crushed), followed by simple extraction with a minimal volume of methanol (∼30 µL). The resulting extract was then applied to the silicon wafer pre‐loaded with the probe solution, enabling rapid disassembly upon analyte binding. The device successfully identified the SC in all tested matrices, exhibiting a bright and unambiguous fluorescence turn‐on (Figure 6f).

Crucially, we further evaluated whether the probe could maintain its stringent selectivity within these chemically complex environments. We compared the fluorescence response of EDMB‐PINACA against 18 representative interferents extracted from the same matrices (Figure 6g). While the target SC consistently induced robust signal enhancement, extracts containing non‐target compounds or the blank matrix alone did not elicit significant responses. In stark contrast to the target analyte, the background fluorescence remained negligible, confirming that co‐extracted matrix components do not trigger false positives. Moreover, the silicon‐wafer sensing interface maintained clear fluorescence contrast between the analyte‐treated and blank spots over repeated use cycles after ethanol washing, supporting the practical reusability of the portable platform (Figure S49). These results conclusively establish that the steric‐engineered thermodynamic gate retains its structural specificity and operational robustness in highly complex real‐world samples with only simple pretreatment. By integrating the metastable aggregate probe into a low‐cost, portable device, the platform provides a practical, rapid, and reliable solution for on‐site forensic screening of synthetic cannabinoids in challenging environments.

3. Conclusion

In summary, to resolve the intrinsic trade‐off between sensitivity and stability in supramolecular sensing, we established a “Steric‐Engineered Thermodynamic Gating” strategy using synthetic cannabinoids as a model system for low‐reactivity analytes. We rationally modulated intermolecular packing by installing a rigid Boc group at the 4‑phenylamine‑N position of a naphthalimide scaffold, deliberately introducing packing frustration to trap the probe in a quenched, metastable H‑aggregate “Off” state that is “spring‑loaded” for analyte‑driven unlocking. Functioning as a stringent thermodynamic filter, this assembly enforces a specific “Head/Tail Check”: it remains inert against structurally distinct interferents (high stability) while selectively responding only to SCs possessing the requisite multivalent noncovalent synergy (high sensitivity). This decoupling of specificity from background noise endows the system with exceptional anti‐interference capability, maintaining robust signal fidelity even in authentic complex matrices (e.g., e‐liquids). Beyond binary detection, we integrated a CNN to decode the subtle fluorescence “fingerprints” of the responsive targets, achieving 99.5% accuracy in distinguishing structurally similar SC analogs. The practical utility of this strategy was demonstrated via a 3D‐printed portable device, enabling rapid, on‐site visual screening. Looking forward, this metastable assembly and thermodynamic gating paradigm provides a versatile blueprint for designing responsive probes against non‑reactive targets, with clear potential to expand toward broader NPS panels, metabolite‑aware screening, and multiplexed point‑of‑care visual forensics.

Author Contributions

Fengbin Tao: data curation, methodology, investigation, validation, visualization, formal analysis. Nuermaimaiti Yisimayili: writing – review and editing, writing – original draft, conceptualization, validation. Chuanfang Zhao: visualization, formal analysis. Chengzhi Gu: writing – review and editing, supervision, funding acquisition. Yuwan Du: software. Yating Hu: visualization, investigation. Xincun Dou: writing – review and editing, conceptualization, resources, project administration, supervision, funding acquisition. Baiyi Zu: writing – review and editing, supervision, funding acquisition.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

The authors have cited additional references within the Supporting Information [14, 16, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58]. Supporting File: anie72780‐sup‐0001‐SuppMat.docx.

Data Availability Statement

The data that support the findings of this study are available in the Supporting Information of this article.