Main

Experimental study of protein transient states1 remains a major challenge because high-structural-resolution techniques, including nuclear magnetic resonance and X-ray crystallography, often cannot be directly applied to study short-lived protein states. In contrast, high-temporal-resolution fluorescence spectroscopy is better suited to detect transient states2,3, in addition to being convenient, highly sensitive and widely available. While large conformational changes can often be detected using intrinsic tryptophan fluorescence4, those in the ~3–9 nm range are typically detected using Förster resonance energy transfer (FRET). FRET, however, requires the complicated chemistry of site-specific insertion of fluorophores5,6. Thus, despite much progress in fluorescence techniques7,8,9,10, there remains an unmet need for sensitive strategies that are able to detect the small conformational changes11 experienced by some proteins during their function, such as enzyme catalysis12, as well as for simpler and more efficient labeling strategies.

Here, we report the design, engineering and application of a versatile tool called fluorescent nanoantennas, which sense and report protein conformational change, and in turn protein function, in real time. Since widely available fluorescent dyes display low affinity for proteins13,14, these nanoantennas are designed to drive noncovalent dye-protein interactions, making them highly sensitive to conformational changes. Each dye ought to have an affinity for a different region of a protein, depending on their structural complementarity and chemical properties. Thus, via highly programmable phosphoramidite chemistry, we synthesized nucleic acid (DNA) and polyethylene glycol (PEG) nanoantennas containing dyes and other functional modifications. We also leveraged the convenience of biotin-streptavidin (SA) noncovalent interaction, which enables quick and easy connection of biotin-labeled nanoantennas to biotin-labeled proteins.

As a first model protein to test whether nanoantennas can detect protein activity, we selected calf intestinal alkaline phosphatase (AP; EC 3.1.3.1)15. The study of intestinal AP is an active area of research16 due to its important roles in preventing inflammation17, promoting growth of the commensal microbiota18, regulating pH19, activating prodrugs20,21 and studies of fundamental biophysics22. APs have been implicated in breast, prostate, colorectal and gastric cancers23,24,25,26, metabolic syndrome27, hypophosphatasia28,29, myocardial infarction30, chronic intestinal inflammation31 and even SARS-CoV-2 infection32. This enzyme has been characterized by crystallography33,34,35, computational simulations36, unfolding37, inhibitors38,39, mutations40 and hydrolysis of substrates40,41. Classic42,43 and newer strategies44,45,46 to characterize AP-mediated hydrolysis in real time involve monitoring the rate of product generation (Extended Data Fig. 1). Unfortunately, these assays require synthetic substrates to provide a signal, whereas biomolecular substrates of AP are spectroscopically silent (for example, nucleotide triphosphates)18,19. For biomolecules, the standard malachite green assay does not permit real-time analysis47, while alternative biomolecular assays are not universal48,49. Isothermal titration calorimetry50,51 can characterize AP activity for biological substrates52, but it is not amenable to high-throughput screening. We are not aware of any FRET studies involving labeling of AP, presumably due to the small conformational changes experienced by this protein34,53. Here, we designed fluorescent nanoantennas, investigated their signaling mechanism and applied them to study AP function as well as a second protein system, Protein G interacting with various antibodies54.

Results

Mechanism of fluorescent nanoantennas

We summarize the general idea of our strategy in Fig. 1a. The DNA- or PEG-based fluorescent nanoantennas contain a fluorescent dye at one end, such as fluorescein (FAM), and biotin at the other to facilitate attachment (Start). Using biotin, we attached the nanoantenna to wild-type tetrameric SA from Streptomyces avidinii, which has four biotin-binding sites, and observed a decrease (or quenching) in FAM fluorescence (Step 1). Docking simulations (Fig. 1b), experimental evidence (Extended Data Fig. 2) and previous studies55,56 suggest that FAM binds near the unoccupied biotin-binding sites of SA. Next, we added the model protein, biotinylated calf intestinal alkaline phosphatase (bAP). Unlike specific dye labeling required for FRET experiments, our method employed nonspecific biotinylation, which can be performed conveniently on many proteins without affecting their function57. Binding of bAP to the nanoantenna-SA platform results in an increase in the fluorescence signal (Step 2), suggesting that FAM is released from SA. Upon addition of a substrate of AP, the nanoantenna generates a transient fluorescence ‘spike’ (Step 3), enabling real-time monitoring of the enzyme’s transient substrate-bound state. This result, combined with docking simulations (Fig. 1c), suggests that FAM binds near one of the two equivalent active sites.

Fig. 1: Overview of the fluorescent nanoantenna strategy to probe different regions on a protein.
figure 1

a, Cartoon and example data of fluorescent nanoantennas. For simplicity, the cartoon shows only one of each component. a.u., arbitrary units. b, The docking simulation accurately predicts the binding sites of biotin on SA and the substrate pNPP on AP. c, Docking prediction of the dyes FAM, CAL and Cy3 on SA and AP. di, Optimization of the length (d) and composition (e) of the linker for Step 1 and similarly for Step 2 (length, f; composition, g). See Supplementary Fig. 1 for corresponding fluorescence spectra. Similar results (length, h; composition, i) were observed for the fluorescence spike during hydrolysis of pNPP for Step 3. For di, n = 1 biologically independent enzyme samples were examined over three independent experiments. Data are presented as mean values ±s.e.m. jl, Kinetic signatures of ssDNA nanoantennas (NA) containing the dye FAM (λex = 498 nm, λem = 520 nm) (j), CAL (λex = 540 nm, λem = 561 nm) (k) or Cy3 (λex = 546 nm, λem = 563 nm) (l) for SA and bAP binding events, as well as pNPP hydrolysis. fluor., fluorescence. m,n, Double-dye competition kinetic signatures for FAM (m) and CAL (n). In m, the data at the top show the monitoring of FAM fluorescence of a single-dye dsDNA nanoantenna with FAM, and the data at the bottom show the monitoring of FAM fluorescence of a double-dye dsDNA nanoantenna with FAM and CAL. In n, the data at the top show the monitoring of the CAL fluorescence of a single-dye dsDNA nanoantenna with CAL, and the data at the bottom show the monitoring of the CAL fluorescence of a double-dye dsDNA nanoantenna with FAM and CAL.

We explored how the nanoantenna linker length (LX, where X is the number of nucleotides) (Fig. 1d) and composition (Fig. 1e) impact dye-protein interactions. As a ‘no linker’ L0 nanoantenna, we selected a biotin-fluorescein conjugate. Upon binding to SA, its fluorescein moiety is located just outside the biotin-binding site that it occupies58. This short nanoantenna displayed substantial fluorescence quenching. Using single-stranded DNA (ssDNA), we increased the linker length to L6 or L12, thereby enabling FAM to interact with more of the SA surface. These nanoantennas displayed moderate quenching, with their FAMs likely binding near unoccupied biotin-binding sites (Fig. 1b)55,56. The longer L24 and L48 nanoantennas displayed reduced quenching, consistent with fewer dyes being bound to the protein due to the lower effective concentration of the dye near SA. We also tried a more flexible, hydrophilic and less charged PEG-based nanoantenna (approximately L21, Supplementary Fig. 2), which displayed increased quenching. In contrast, a less flexible double-stranded DNA (dsDNA) L24 nanoantenna prevented the FAM-SA interaction.

Linker length (Fig. 1f) and composition (Fig. 1g) likewise affected the monitoring of protein binding to SA. As expected, due to its short length, L0 did not detect bAP. The longer L6 and especially L12 nanoantennas enabled FAM to detect bAP attachment, but L24 and especially L48 were too long to result in a high local concentration of FAM near bAP. Also as expected, we observed that a PEG linker enabled good FAM-bAP interaction, while a less flexible dsDNA linker did not. A molecular dynamics (MD) simulation with the optimal L12 nanoantenna revealed that its FAM could plausibly reach the bound bAP, supporting our hypothesis regarding the dye-enzyme interaction (Extended Data Fig. 3 and Supplementary Video 1). We have additionally explored other factors, such as pH variation and the ratio of components, and found that PEG linkers were less sensitive to pH variation (Supplementary Fig. 3), while using too many nanoantennas per streptavidin diminishes signaling by preventing enzyme attachment (Supplementary Figs. 46).

We next investigated the mechanism by which the nanoantennas generated a transient fluorescence spike during hydrolysis of p-nitrophenylphosphate (pNPP; step 3 in Fig. 1a). As expected, the most sensitive nanoantennas for probing bAP attachment were also the most sensitive for probing its catalytic activity (Fig. 1h,i and Supplementary Fig. 7). The stoichiometry of the added components is important; we found that adding three nanoantennas per SA maximizes signal without potentially limiting the ability of the biotinylated enzyme to bind to the remaining unoccupied biotin-binding site(s). Nanoantenna size, and therefore steric hindrance59, is also a factor (Supplementary Fig. 7d). As controls, we noted that no spike occurred when there was no hydrolysis reaction, for example, upon addition of the reaction products (Extended Data Fig. 4a–c) or upon addition of pNPP when unattached nonbiotinylated nanoantennas were employed (Extended Data Fig. 4d). We also did not observe a signal change due to intrinsic tryptophan fluorescence (Supplementary Fig. 8)4, and could not detect AP function with the protein-binding dye 8-anilinonaphthalene-1-sulfonic acid (Supplementary Fig. 9)60. We did, however, observe a spike with other nanoantenna attachment strategies (for example, covalent attachment to surface-exposed lysine residues of AP; Extended Data Fig. 4e–h), different buffer conditions (Supplementary Fig. 10) and various storage times (Supplementary Fig. 11). We also observed that the intensity of the nanoantenna’s fluorescence in the different states (that is, after addition of SA, bAP and pNPP) was sensitive to small chemical modifications that could subtly perturb the FAM-protein interaction, such as a nearby hydrophobic C16 alkane chain (Extended Data Fig. 5) and different chemical connections of FAM to the linker (Extended Data Fig. 6). Finally, under some conditions, we observed that the fluorescence spike intensity during pNPP hydrolysis was greater than the initial fluorescence of the unbound nanoantenna (Fig. 1j and Supplementary Fig. 12). FAM simply being ejected from the active site by the incoming pNPP substrate would likely just return the fluorescence to the initial baseline before protein binding. Overall, these results reinforce our proposed mechanism that the nanoantenna-mediated dye-enzyme interaction enables monitoring of the conformational changes on the enzyme’s surface during its function.

The testing of chemically diverse dyes on ssDNA L12 nanoantennas provides information about the signaling mechanism and potential universality of the strategy. Different dyes are predicted to bind to different sites on AP (Fig. 1c and Supplementary Figs. 13 and 14). Interestingly, all nine dyes tested enabled monitoring of the SA and bAP binding events, albeit weakly in some cases. Shown here are FAM, the rhodamine-based CAL Fluor Orange 560 (CAL) and cyanine 3 (Cy3) (Fig. 1j–l; see other dyes in Supplementary Fig. 15a–i). In contrast to FAM, Cy3 shows increased fluorescence upon binding to SA and decreased fluorescence upon binding to bAP7, and it is more sensitive with a dsDNA linker. However, similar to FAM, it is affected by its chemical connection (Supplementary Fig. 15e,f). We also found that the kinetics of the signal change upon addition of bAP depend both on the concentration of bAP (Supplementary Fig. 16) and on the nanoantenna properties (Supplementary Fig. 17). This suggests that dye dissociation from SA is not rate limiting and that the dye-binding location and the nature of the linker may affect the rate at which bAP binds to SA (for example, through steric hindrance). We further observed that nanoantennas with FAM, CAL and Cy3 enabled monitoring of pNPP hydrolysis. Crucially, while the three dyes provided different sensitivities toward the functional events, they all exhibited the same kinetics for pNPP hydrolysis (Supplementary Fig. 15j), indicating that the dye-protein interactions exploited herein did not interfere with protein function. From a practical perspective, FAM remains the best dye for monitoring AP function. However, from a mechanistic perspective, the other dyes provide strong evidence that nanoantennas employing chemically diverse dyes can be used to probe conformational changes at different locations.

We investigated the signaling mechanisms of dyes predicted to bind at different locations. For nanoantennas with FAM, we proposed three mechanisms for modulation of fluorescence during AP function: (1) binding of pNPP at the active site directly ejects FAM from its nearby binding site; (2) small conformational changes alter FAM’s affinity for AP and release it from the binding site; or (3) small conformational changes perturb the emission of the bound FAM. For CAL and Cy3, which are predicted to bind at locations distal to the active site, we proposed only the latter two mechanisms. Thus we used MD simulations and examined the trajectories of the bound dyes in the presence or absence of bound pNPP (Extended Data Fig. 7). FAM remained bound near the active site both in the absence and in the presence of pNPP, suggesting a strong affinity for this site. In contrast, CAL was not stabilized in its initial binding site in either the absence or presence of pNPP, suggesting low affinity. Interestingly, Cy3 remained bound in the absence of pNPP but dissociated in its presence. These simulations, therefore, suggest a FAM-signaling mechanism that is not based on ejection by pNPP nor a change in affinity, but instead by the sensing of small conformational changes in the local chemical environment61,62. For CAL, the mechanism remains uncertain, but for Cy3, the simulations suggest that conformational changes during pNPP hydrolysis transiently release the dye. The bound and unbound states of Cy3 could, for example, affect its cis–trans isomerism and therefore its fluorescence7,63,64.

If the sensitivity of a nanoantenna truly depends on whether the dye’s binding location experiences conformational change during protein function, one ought to observe a change in the fluorescence signature upon forcing the dye to bind at another location. To test this hypothesis, we employed a dsDNA L12 nanoantenna containing both FAM and CAL (Fig. 1m,n). Our reasoning was that the two dyes will compete for their preferred binding sites on bAP. On the basis of the above MD simulations, we expected that the higher affinity FAM will bring the lower affinity CAL along with it. Aside from the decrease of signal intensity, likely due to a contact-mediated quenching mechanism between the dyes65 (Extended Data Fig. 8), we observed that the presence of CAL does not substantially affect the FAM fluorescent signature (Fig. 1m). In contrast, the addition of FAM substantially affects the CAL fluorescent signature (Fig. 1n). Notably, it now enables CAL to efficiently detect the conformational change of the protein during pNPP hydrolysis (Fig. 1n). Given that CAL likely stacks on FAM (Extended Data Fig. 8), it is plausible that the fluorescent change of CAL is triggered by the same conformational change affecting the FAM dye. Importantly, the presence of the second dye did not affect the kinetics of the substrate hydrolysis (Supplementary Fig. 18a). We also found that employing FAM with other dyes led to similar results (Supplementary Fig. 18b,c). Of note, these changes in the fluorescence signature do not arise due to FAM emitting a signal that overlaps with the excitation wavelengths of other dyes (Supplementary Fig. 18d,e). Thus, these experiments provide strong evidence that FAM can redirect other dyes to a location proximal to its own binding site on AP. This observation is consistent with the aforementioned MD simulations, suggesting that FAM is more tightly bound than Cy3 and CAL to AP (Extended Data Fig. 7). Hypothetically, one could rationally employ other non-dye ‘molecular anchors’ to redirect dyes to specific locations on proteins, as FAM does for CAL on AP.

Characterizing enzyme kinetics

Nanoantennas can be used to characterize the kinetic mechanisms of enzymes. By employing ssDNA L12 FAM nanoantennas hereafter, we first observed that the addition of more pNPP increased the spike intensity and duration (Fig. 2a). The resulting fluorescence signature is reminiscent of the expected profile of the enzyme–substrate concentration ([ES]) during a typical enzymatic reaction (Extended Data Fig. 1b). The signal rapidly peaked and then maintained a steady state until the substrate began to run out. This hypothesis is consistent with the nanoantennas distinguishing between the enzyme and enzyme–substrate conformations. We confirmed the link between the fluorescence intensity of the spike and [ES] by showing that the former is proportional to the rate of reaction obtained by monitoring p-nitrophenol (pNP) generation via UV–visible spectroscopy (Fig. 2b). Indeed, plotting the spike intensity versus pNPP concentration generated a saturation binding curve that is reminiscent of a Michaelis–Menten plot (Fig. 2c; Supplementary Fig. 19). Fitting the data provided a K0.5 value that is similar to the Michaelis constant (KM) reported in the literature under the same conditions (K0.5 = 4.4 ± 0.2 μM; Supplementary Fig. 20)52.

Fig. 2: Fluorescent nanoantennas enable complete characterization of the enzyme’s kinetic mechanism.
figure 2

a, Increasing the substrate concentration increases the fluorescence spike intensity and duration, displaying a profile reminiscent of the enzyme–substrate concentration ([ES]) during a typical enzymatic reaction. b, Fluorescence intensity is correlated with the rate of reaction (and [ES]) determined by monitoring pNP generation via UV–visible spectroscopy. c, The saturation binding curve realized using the nanoantenna spike intensity displays a typical Michaelis–Menten-like plot. d, Nanoantenna-labeled enzyme allows sequential injection of pNPP without signal saturation. Here, subsequent pNPP injection (30 µM) results in prolonged reaction time and reduced spike intensity due to competitive inhibition via the accumulation of Pi. e, Extracting KM and kcat using the fluorescent nanoantenna signature of a single reaction (see Methods, Supplementary Fig. 25 and ‘Script for fitting kinetic data in MATLAB’ in the Supplementary Information for details of the fitting procedure). f, Extracting Ki using the fluorescent nanoantenna signature of multiple reactions. All experiments were performed with n = 1 biologically independent enzyme samples examined over three independent experiments. Data are presented as mean values ±s.e.m. In ac, we used 100 nM bAP from a commercially available sample, and in df we used 10 nM bAP prepared by us from AP and a biotinylation kit.

Nanoantennas enable complete kinetic characterization of an enzyme in a single experiment. Unlike the monitoring of AP kinetics by product generation42,43,44, one enzyme sample with nanoantennas can enable multiple measurements in one cuvette. This could be used to characterize various substrates or inhibitors (Fig. 2d and Supplementary Fig. 21). For example, upon performing consecutive pNPP injections, we observed a decrease in spike intensity and an increase in reaction time, consistent with accumulation of the product, inorganic phosphate (Pi), a competitive inhibitor of this enzyme. We show that by fitting a single fluorescence spike using Michaelis–Menten differential equations with competitive product inhibition51, one can extract the KM, the catalytic rate constant (kcat) and, from these, the catalytic efficiency (kcat/KM) (Fig. 2e). Indeed, we determined KM (5.0 ± 0.1 μM), kcat (32.1 ± 0.9 s−1) and kcat/KM (6.4 ± 0.3 μM−1 s−1) values similar to those reported in the literature52 (Supplementary Table 1). Notably, the KM values determined by plotting spike intensity and by fitting a single spike were also consistent. Crucially, similar values were also obtained by monitoring pNP product generation using UV–visible spectroscopy, supporting that neither the nanoantenna nor SA affect the kinetic parameters of the enzyme (Supplementary Figs. 2224 and Supplementary Table 1). Next, using a similar fitting procedure51, we modeled the eight spikes from the same enzyme sample and determined the inhibition constant (Ki), a measure of the inhibitory effect of the Pi product (Fig. 2f; see Supplementary Fig. 25a–d for complete fitting example). This Ki (48.4 ± 2.0 μM) was close to the previously reported values (Supplementary Table 2)48,52. Moreover, the decrease in spike intensities was consistent with the expected decrease in the reaction rate due to inhibition (Supplementary Fig. 25e).

Fluorescent nanoantennas can be used to monitor the AP-mediated hydrolysis of any substrate, including biomolecules. Indeed, all of the chemically diverse substrates tested herein exhibited a similar fluorescence spike during hydrolysis: pNPP, 4-methylumbelliferylphosphate (4MUP), pyrophosphate (PPi)29, β-glycerophosphate (BGP), phosphoenolpyruvate (PEP), l-phosphoserine (PSer), pyridoxal 5′-phosphate (PLP)29, d-glucose-6-phosphate (G6P), d-fructose-6-phosphate (F6P), adenosine 5′-monophosphate (AMP), adenosine 5′-diphosphate (ADP), adenosine 5′-triphosphate (ATP)19, guanosine 5′-triphosphate (GTP)18,19, phosphocreatine (PCr) and amifostine20 (Fig. 3). Using the ‘one-shot’ fitting strategy described above, each substrate displayed KM and kcat values consistent with previously reported values, when available, while others were characterized with calf intestinal AP for the first time (Fig. 3 and Supplementary Table 3). Furthermore, with 4MUP, which generates fluorescent 4-methylumbelliferone (4MU; also called hymecromone), we found that a single fluorescence spike provided KM and kcat values that agreed with those determined by the traditional product generation method (Extended Data Fig. 9). Simply fitting one 4MU product progress curve, however, provided markedly overestimated values66,67. In addition to the biomolecular substrates, we also tested amifostine, a prodrug used to protect normal cells during chemotherapy and radiotherapy, which is putatively hydrolyzed to its active metabolite form by intestinal AP20,21. While the KM of amifostine was comparable to that of the other tested substrates, its kcat was the slowest, likely due to AP not having evolved to process its P–S bond, unlike for the P–O and P–N bonds of biomolecular substrates. Fluorescent nanoantennas also enabled real-time monitoring of the hydrolysis of ~10-kDa lipopolysaccharides (LPS) (Supplementary Fig. 27)17,31. Deriving kinetic parameters for LPS, however, remains challenging due to uncertain sample concentration and the number of phosphates hydrolyzed per LPS molecule68.

Fig. 3: Fluorescent nanoantennas enabled real-time monitoring of any substrate hydrolyzed by AP.
figure 3

Nanoantenna fluorescence signatures during hydrolysis of pNPP, 4MUP, PPi, BGP, PEP, PSer, PLP, G6P, F6P, AMP, ADP, ATP, GTP, PCr and amifostine (all 300 μM). All experiments were performed with n = 1 biologically independent enzyme samples examined over three independent experiments. Data are presented as mean values ±s.e.m.

Fluorescent nanoantennas can also be used to screen for nonproduct inhibitors and activators by monitoring their effect on the AP-mediated hydrolysis of a substrate. For example, we characterized the Ki of five oxyanion inhibitors, as well as the effect of Mg2+, on the hydrolysis of amifostine by AP (Fig. 4 and Supplementary Table 4). The kinetic profiles of phosphate and vanadate agree with the theoretical result for relatively weak and strong competitive inhibitors, respectively (Extended Data Fig. 10).

Fig. 4: Screening inhibitors by monitoring AP hydrolysis using fluorescent nanoantennas.
figure 4

The inhibitory effects on AP of phosphate, molybdate, tungstate, arsenate and vanadate, as well as the effect of Mg2+ ion (black) with the substrate amifostine. Kinetic fitting of the data is overlaid (pink). Shown also for comparison is the hydrolysis of amifostine without inhibitor (gray). The baseline of the kinetic signature without the effector was adjusted for presentation of data (that is, due to the fluorescence quenching caused by vanadate and magnesium ion). All experiments were performed with n = 1 biologically independent enzyme samples examined over three independent experiments. Data are presented as mean values ±s.e.m. In all experiments, we used 30 μM inhibitor or 5 mM Mg2+, and otherwise the same conditions as in Fig. 3.

Characterizing protein conformational states

Other known conformational states of AP can be detected with fluorescent nanoantennas (Fig. 5a). Vanadate (Vi; following the same naming format as for Pi), for example, is a putative transition state analog (TSA) inhibitor of AP, meaning that it stabilizes a geometry of the enzyme that is reminiscent of the transition state34. Nanoantennas detect Vi binding to bAP via fluorescence quenching (Fig. 5b,c; see control in Supplementary Fig. 30). This quenching response contrasts with the increase in fluorescence observed upon substrate binding to the same site, further indicating that the nanoantennas can efficiently distinguish between highly similar conformational states. Interestingly, tungstate (Wi), also recently proposed as a TSA inhibitor of AP35, seems to induce a distinct conformational change as evidenced by the increase in fluorescence (Fig. 5c). In contrast, Pi and molybdate (Moi), which likewise bind at the active site, did not induce a conformational change detectable by the nanoantennas (Fig. 5c). As determined by fluorescence change, the observed dissociation constants (KD) of vanadate (0.54 ± 0.01 μM) and tungstate (7.6 ± 0.2 μM) are consistent with their Ki determined using the ‘one-shot’ fitting of the fluorescent spike (1.2 ± 0.3 μM and 10.1 ± 3.5 μM, respectively; Fig. 4 and Supplementary Table 4), and with the literature69.

Fig. 5: Detecting other states of AP using fluorescent nanoantennas.
figure 5

a, Five distinct states of AP detected during enzymatic reaction and denaturation. First, the enzyme (E) is present in solution, followed by the energetically favorable binding of a substrate (for example, pNPP) to form an ES complex. Next, the high-energy transition state (ES) is formed. This state cannot be observed directly but can be studied via TSA inhibitors that mimic its geometry, such as Vi. Since Vi is a competitive inhibitor, it also represents an enzyme–inhibitor complex (not shown). After the hydrolysis of pNPP, the bound Pi product (EP) is released from the enzyme (E + P). As a competitive inhibitor, Pi can rebind to the active site. AP can also be unfolded by thermal denaturation (Eunfolded). b, Unlike the spike during pNPP hydrolysis, Vi binding quenches the nanoantenna’s fluorescence. c, Binding curve of the nanoantenna-SA-bAP complex sensing Vi (blue) and Wi (orange). Other competitive inhibitor oxyanions do not exhibit a substantial fluorescence change. d, Thermal shift assay of the nanoantenna (gray), nanoantenna-SA platform (light green) and nanoantenna-SA-bAP complex (violet). e, Smoothing of the derivatives (black lines) reveals no transition for the unbound nanoantennas, but does for the nanoantenna-SA platform and nanoantenna-SA-bAP complex. In ce, all experiments were performed with n = 1 biologically independent enzyme samples examined over three independent experiments. Data are presented as mean values ±s.e.m.

Nanoantennas can also monitor large conformational changes, such as protein unfolding by thermal denaturation. When attached to SA, nanoantennas display a distinct transition at a melting temperature (TM) of 91.7 °C ± 0.7 °C (Fig. 5d,e). This likely represents dissociation of FAM from SA, rather than unfolding of SA or detachment of the whole nanoantenna, since the nanoantenna-SA platform remained stable over this temperature range (Supplementary Fig. 31). With bound bAP, the nanoantennas exhibit a distinct transition at a TM of 66.4 °C ± 0.1 °C (Fig. 5d,e), consistent with the unfolding temperature of AP (Supplementary Figs. 31 and 32)37. One potential application of TM determination by fluorescent nanoantennas could be the characterization of a specific protein in the presence of others (Supplementary Fig. 33). The nanoantennas further enabled derivation of the apparent Gibbs free energy (ΔG) for the thermal unfolding of bAP (ΔG = −7.8 ± 0.5 kcal mol−1 at T = 37 °C; Supplementary Fig. 34).

Rapid screening of nanoantennas

We explored the potential universality of the nanoantenna strategy by using a different model protein that involves protein–protein interaction. As a proof-of-concept, we employed Protein G from streptococcal bacteria, which binds goat immunoglobulin G (IgG) with high affinity54. To facilitate nanoantenna selection, we designed a 96-well-plate screening assay that leverages the convenience of the nanoantenna-SA platform (Fig. 6a,b and Supplementary Note 2). This enabled us to rapidly test 12 nanoantennas with different linker lengths, linker types, chemical connections and fluorophores. We first prepared the plate by adding the different nanoantennas, followed by the addition of SA. We then added biotinylated Protein G (bPG) and recorded the fluorescence intensity in all wells. Upon addition of goat IgG, we observed that nanoantenna 6 (5′ T 6-FAM L21 PEG) displayed the largest fluorescence quenching, while nanoantenna 10 (3′ Cy3 L12 ssDNA) displayed the largest fluorescence enhancement (Fig. 6c). After identifying these candidate nanoantennas, we confirmed their performance in cuvettes (Fig. 6d,e). This screening strategy also offers an opportunity to further optimize performance via semirational design of the fluorescent nanoantenna. For example, we observed that for this particular protein function, the flexible L21 PEG nanoantenna enabled the best sensitivity for the T 6-FAM fluorophore (Fig. 6c). Although the ssDNA nanoantennas with T 6-FAM were not as sensitive in comparison, we noticed that the longer ones were better than the shorter ones (Fig. 6c). Therefore, we subsequently tested a longer PEG nanoantenna (5′ T 6-FAM L41 PEG), not included in our initial screening, and found that it did indeed display improved sensitivity to goat IgG binding (Fig. 6e). Ultimately, however, we selected the Cy3 nanoantenna for subsequent investigations due to its signal-on fluorescence change (Fig. 6e).

Fig. 6: Rapid screening strategy for identification of functional nanoantennas for protein G binding IgG.
figure 6

a,b, Plate reader screening strategy (a) to rapidly identify a nanoantenna that reports the binding of goat IgG to biotinylated protein G (bPG) (b) using the nanoantenna-SA platform. c, Results obtained for rapid screening of 12 nanoantennas for the aforementioned system. d,e, Cuvette validation (d) of results for nanoantennas 1, 6 and 10 displays similar trends to the plate reader format. Also shown is the semirationally selected longer FAM L41 PEG (e). f, The nanoantenna-SA-bPG complex detects goat IgG and SARS-CoV-2 IgG/IgM while remaining silent to a sample without SARS-CoV-2 IgG/IgM and to the enzyme AP. As an additional control, replacing bPG with biotinylated protein A (bPA), which does not bind goat IgG, does not display a signal change. Ab, antibody. In c,f, all experiments were performed with n = 1 biologically independent protein and antibody samples examined over three independent experiments. Data are presented as mean values ±s.e.m.

In principle, this nanoantenna-SA-bPG complex could be used as a signal-on biosensor to detect the presence of specific types of antibodies (Fig. 6f; bPG + IgG)54. Indeed, a sample containing SARS-CoV-2 IgG/IgM led to a similar fluorescence increase (Fig. 6f; bPG + CoV antibody), while it did not respond to a sample negative for SARS-CoV-2 IgG/IgM (Fig. 6f; bPG + Ctrl) nor to the enzyme AP (nonbiotinylated) (Fig. 6f; bPG + AP). Furthermore, swapping bPG for biotinylated Protein A (bPA), which does not bind goat IgG54, also did not display a signal increase (Fig. 6f; bPA + IgG; Supplementary Fig. 35). This control with bPA further shows that the target goat IgG does not nonspecifically interact with the nanoantenna-SA platform. Overall, these results indicate that fluorescent nanoantennas can be rapidly screened for their ability to monitor distinct protein functions.

Discussion

Here, we have introduced the use of fluorescent nanoantennas as a strategy to monitor protein dynamics. A platform and linker mediate dye-protein interactions via a high local concentration. Protein conformational changes affecting the dye’s chemical environment generate a change in the fluorescence signal. By tuning linker length and dye, we have leveraged this strategy to monitor the functions of three proteins: streptavidin, alkaline phosphatase and Protein G. Several observations supported our proposed signaling mechanism. First, FAM nanoantennas detected all conformational changes in their surroundings: their binding to SA, subsequent binding of biotin or a biotinylated protein to SA, and the function of that protein. Most interestingly, the nanoantenna also detected five distinct conformational states of AP: its ground state, enzyme–substrate complexes with various substrates, the distinct conformational changes induced by vanadate34 and tungstate35 binding, and its unfolded state. Second, 16 structurally distinct substrates of AP, all hydrolyzed by the same mechanism33, exhibited similar fluorescence signatures. This fluorescence signature enabled easy characterization of Michaelis–Menten kinetic parameters of all tested substrates and inhibitors. Third, nanoantennas employing chemically diverse dyes that bind to different locations on AP differed in sensitivity but displayed the same kinetics during substrate hydrolysis. Fourth, the strategy was not limited to a single fluorescent dye, since FAM nanoantennas were optimal for monitoring bAP function while the Cy3 nanoantenna was best for bPG. Finally, MD simulations also suggested a signaling mechanism based on conformational change.

A main advantage of fluorescent nanoantennas is their convenience. For example, the nanoantennas can be used with accessible and straightforward fluorescence spectroscopy, as opposed to specialized techniques. Furthermore, various conjugation strategies can be developed to facilitate nanoantenna-protein preparation. For instance, here we developed and exploited the modular biotin-SA platform. This requires only nonspecific biotinylation of the protein of interest, as opposed to site-specific attachment chemistry of fluorophores. Indeed, lysine residues can be nonspecifically biotinylated with a simple commercially available kit. While it cannot be assumed to be necessarily true in all cases, biotinylation often does not affect protein function57. In comparison, the more complex site-specific labeling needed for FRET strategies has been found to perturb the function of β-lactamases3 and dihydrofolate reductase70. In the case where biotinylation would affect a protein’s function, other modular attachment strategies could also be developed; for example, one could envisage the use of N or C terminus affinity tags71. When employing a modular attachment strategy, efficient nanoantennas can also be rapidly screened using a 96-well plate, as we demonstrated with bPG.

Another important advantage of fluorescent nanoantennas is their versatility. Nanoantennas can be used to monitor distinct biomolecular mechanisms in real time, including small and large conformational changes—in principle, any event that can affect the dye’s fluorescence emission. Furthermore, since nanoantennas can distinguish between unbound and substrate-bound enzyme conformations, they can supplant nonnatural colorimetric42 and fluorogenic43,44 substrates, as well as laborious assays for spectroscopically silent substrates47. For example, nanoantennas enable real-time, ‘one-shot’ kinetic characterization of any substrate, such as ATP and amifostine18,19,20,21,29. In contrast, standard methods to determine KM and kcat for spectroscopically silent substrates require approximately ten measurements at different substrate concentrations47,72. They also compare favorably with other ‘one-shot’ strategies that require microfluidics to generate a range of substrate concentrations, in addition to a fluorescent product73. Nanoantennas, however, do have some limitations. For example, unlike other noteworthy techniques7,8,63,64,74,75,76,77, nanoantennas cannot quantify specific distance variations. Also, not all dye-protein combinations generate a signal change during protein function; some proteins might not work with any dyes. However, looking to the future, we believe that the universality of the nanoantenna strategy may be improved by screening a larger library of dyes and by further exploring the predictive potential of docking and MD simulations. We anticipate that our fluorescent nanoantennas will find exciting applications in the study of protein structure and function and in high-throughput screening.

Methods

Enzymes, substrates and other materials

AP used in this study was from calf intestinal mucosa. Unconjugated AP, bAP, streptavidin-conjugated AP (SA-AP), bPG, bPA and goat IgG (whole molecule) were purchased from Rockland Immunochemicals. SA was from New England Biolabs. VIROTROL SARS-CoV-2 (reactive for SARS-CoV-2 total IgG/IgM and IgG antibodies) and VIROCLEAR SARS-CoV-2 (nonreactive for SARS-CoV-2 total IgG/IgM and IgG antibodies) were from Bio-Rad Laboratories. See Supplementary Information for enzyme storage buffer conditions, as well as details about substrates, inhibitors and other reagents.

Oligonucleotide synthesis

Labeled and unlabeled oligonucleotides were made by standard phosphoramidite chemistry with a solid support DNA/RNA H-6 Synthesizer from K&A Laborgeräte. Purification of strands with a 5′ protecting group (4,4′-dimethoxytrity (DMT)) was performed with a P-8 oligonucleotide purifier. Strands without a protecting group (for example, 6-FAM and 5-FAM) were purified using high performance liquid chromatography (HPLC) with a 1260 Infinity HPLC instrument from Agilent. The mobile phase was 0.1 M triethylamine with increasing concentration of acetonitrile, and the stationary phase was an XBridge Oligonucleotide BEH C18 OBD Prep Column, 130 Å, 2.5 µm, 10 mm × 50 mm from Waters Corporation. Extinction coefficients at 260 nm were predicted using the OligoAnalyzer website (https://www.idtdna.com/calc/analyzer) from Integrated DNA Technologies. DNA was then quantified by UV–visible spectroscopy with a Cary 60 from Agilent or a NanoDrop 2000c Spectrophotometer from Thermo Fisher Scientific. Oligonucleotides were prepared as 200 µl, ~800 µM stock solutions, and used as 1 ml, 100 µM intermediate solutions; all stored at −20 °C.

Fluorescence

Fluorescence spectroscopy was recorded with a Cary Eclipse Fluorescence Spectrophotometer from Agilent. For measurements in quartz cuvettes, it was equipped with a Peltier Thermostatted Multicell Holder Accessory from Agilent. For the plate reader measurements, it was equipped with a Microplate Reader ACCY from Varian and used Nunc MaxiSorp 350 μl Black 96-well plates from Thermo Fisher Scientific. Fluorescence spectra were recorded with Scan software, fluorescence kinetics with Kinetics software and melting temperatures with Thermal software (Agilent). Typical settings were: excitation (ex)/emission (em) slit widths 5 nm, excitation 498 nm and emission 520 nm for FAM (wavelengths denoted hereafter in the format 498/520), CAL 540/561, carboxyrhodamine (ROX) 575/602, carboxytetramethylrhodamine (TAMRA) 565/580, Cyanine 3 (Cy3) 546/563, Quasar 570 (Q570) 550/570, Quasar 670 (Q670) 644/670, Pulsar 650 (P650) 460/650 and methylene blue (MB) 670/690. The photomultiplier tube detector voltage was typically 635 V for 150 nM fluorescent nanoantennas, but 800 V for those with MB or P650, as well as 800 V for experiments with 15 nM FAM nanoantennas. For plate reader measurements, it was 600 V. For kinetics, we typically used averaging time 3.0 s, cycle 0.04 min. For spectra, we typically used CAT mode with 10 scans at ‘medium’ speed (scan rate 600 nm min−1, averaging time 0.1 s, data interval 1 nm).

A typical study of nanoantenna fluorescence emission over time (for example, Fig. 1a) was as follows. The intermediate nanoantenna stock solution was added to buffer in a quartz cuvette (150 nM), followed by waiting 5–10 min for the fluorescence signal to equilibrate. After observing a stable signal, we performed subsequent additions of complementary DNA, proteins, substrates, etc. Final volume at addition of substrate was 1 ml. We typically mixed by rapidly pipetting ~10× using ~50 μl volume while being careful not to pipette bubbles into the solution. A waiting time of 3–10 min for each step was taken to ensure binding and equilibration. In most cases, cDNA and SA bound very quickly (several seconds), but biotinylated proteins took longer (sometimes up to several minutes). In a typical experiment, we added SA (50 nM) and then bAP (100 nM) for a nanoantenna:SA:bAP ratio of 3:1:2. For some faster lots of enzyme, we added less enzyme, as indicated. Last, we added the substrate (for example, pNPP). To make the effects of dilution negligible, most additions were aliquots of several μl.

For dual absorbance and fluorescence kinetics of the same sample, we used a SX20 Stopped Flow Spectrometer from Applied Photophysics with Pro-Data SX software and with a 495-nm cut-off filter. Nanoantenna-protein complex was prepared in one syringe and substrate in another syringe, which were then mixed during the measurement.

Buffer conditions

In our initial studies of the nanoantenna concept with AP and of various dyes (Fig. 1), buffer conditions were 200 mM Tris, 300 mM NaCl, 1 mM MgCl2, pH 7.0 and 37 °C, with 100 nM commercially available bAP and 50 nM SA. A ratio of three nanoantennas per SA, for example, used 150 nM nanoantennas. Later, for comparison with another recent study (Fig. 2)52, buffer conditions were 100 mM Tris, 10 mM NaCl, pH 8.0 and 30 °C, with either 100 nM bAP (Fig. 2a–c) or 10 nM bAP (Fig. 2d,e; nanoantennas and SA were adjusted proportionally). For characterization of substrates (Fig. 3), the same buffer was used but at 37 °C with 150 nM nanoantennas, 50 nM SA, commercially available 20 nM bAP and 300 μM substrate. Less bAP was used because this lot of bAP enzyme displayed faster substrate hydrolysis kinetics than previous lots that we had purchased. Note that we did observe some enzyme batch-to-batch variation (Supplementary Fig. 26), but the results were still in good agreement with the range of literature values for pNPP. See Supplementary Table 3 for complete kinetic parameters, and Supplementary Figs. 28 and 29 for a discussion of fluorescence baseline. For characterization of effectors (Fig. 4), the conditions were the same as in Fig. 3, with either 30 μM inhibitor or 5 mM Mg2+. For vanadate and related experiments (Fig. 5b,c), the same conditions were also used but with 100 nM of commercially available bAP. In experiments for thermal denaturation of AP (Fig. 5d,e), to reduce pH variation with temperature, we changed the buffer conditions to 100 mM NaCl, 50 mM Na2HPO4, pH 7.0, 37 °C and used 100 nM of commercially available bAP. For experiments with Protein G in the 96-well plate (Fig. 6c), buffer conditions were 200 mM Tris, 300 mM NaCl, pH 7.0, room temperature, with 500 nM nanoantennas, 167 nM SA, 167 nM bPG and ~1,000 nM goat IgG. For subsequent experiments with Protein G in cuvettes (Fig. 6e,f), the same buffer was used at 37 °C with 150 nM nanoantennas, 50 nM SA, 50 nM bPG or bPA and ~500 nM goat IgG, 20 μL SARS-CoV-2 antibodies or control (Ctrl) (unknown concentration) or 500 nM AP. Dilution of antibodies was 7.5 μl goat IgG in 292.5 μl (well plate) or 992.5 μl (cuvettes) of buffer, and 20 μl VIROTROL SARS-CoV-2 in 980 μl (cuvettes) of buffer. Supplementary figures typically used the same conditions as associated experiments in the main text, but see also their captions.

Software

Data analysis was performed in KaleidaGraph from Synergy Software, OriginPro v.9.0 from OriginLab and Microsoft Excel, with all data plotted in KaleidaGraph. The log D calculations, as a measure of hydrophobicity, were done by MarvinSketch software from ChemAxon. Molecular structure images were also generated with MarvinSketch. Density functional theory (DFT) computations to estimate PEG-based nanoantenna length were done via ChemCompute (https://chemcompute.org)78.

Molecular docking simulations

Docking was performed on the SwissDock web server (http://www.swissdock.ch)79,80 from the Swiss Institute of Bioinformatics. The ‘target’ protein structure for streptavidin was PDB 6M9B (Streptomyces avidinii)81. Since there was no crystal structure available for the AP used in this study, we instead built a homology model on the SWISS-MODEL web server (https://swissmodel.expasy.org) from the sequence of P19111 (Bos taurus intestinal alkaline phosphatase) and the structure of 1ZEF (Homo sapiens placental alkaline phosphatase) as the template82,83,84. The global model quality estimation was 0.79, the quaternary structure quality estimate was 0.93 and the identity was 75.52. ‘Ligand’ structures (for example, biotin, pNPP, dyes, etc.) were determined to be the major microspecies at pH 7.0 using MarvinSketch software, the manufacturer’s product description and available literature (further details in Supplementary Fig. 13), followed by optimization in Avogadro software85. Analysis of the docking simulation was done in UCSF Chimera software using the View Dock tool (Type Selection: Dock 4, 5 or 6)86. Note that dyes in the simulation did not include the attachment chemistry to the DNA, nor the DNA itself, and are accordingly an estimation of the binding site. Docking simulations were replicated ten times to confirm reproducibility (or lack thereof) for the binding site.

MD simulations

Structure preparation

All protein ligand complexes were prepared using the AP homology model. Two sets of complexes were generated for all three fluorophores (FAM, CAL and Cy3) in complex with or without pNPP substrate. From the docking study of FAM, we chose the best pose (‘position’) that would have the para and ortho position of the DNA linker attachment point accessible by the solvent. Since AP is a dimer, both binding locations were populated with different ligand conformations for double sampling. In the case of the substrate-bound pNPP/FAM complex, the FAM ligand was redocked to the AP active site with the substrate present. Again, the best scoring pose of the pNPP/FAM ligand complex from the docking run was chosen. The QuickPrep application of MOE2019 software87 with default parameters was used to create a fully parameterized all-atomistic model, which was then used to generate the input files for all MD simulations. Separately, the model of the nanoantenna-SA-bAP complex was built using the streptavidin/biotin complex (PDB 6M9B)81, the AP homology model and the rL12 nanoantenna sequence with 3′-FAM and 5′-T-biotin. The all-atomic model was again generated using the QuickPrep application. A lysine residue in proximity to the AP-binding site was biotinylated and the biotin moiety was placed in the streptavidin active site in a nonclashing conformation. Note that the manufacturer would not disclose the exact composition of the biotin connection to AP (bAP). The DNA linker was constructed using the MOE2019 DNA/RNA builder starting from the crystallized biotin molecule in the neighboring streptavidin active site. Finally, the FAM fluorophore was attached to the 3′-end of the DNA linker and the entire complex energy minimized using the MOE2019 built-in energy minimization application.

MD simulation

The simulation cell and Amber2088 input files were generated using MOE2019. The crystallographic water molecules were removed before solvation. Next, the protein/ligand complexes and AP apo structure were embedded in a TIP3P water box with cubic periodic boundary conditions, keeping a distance of 10 Å between the boundaries and the protein. The net charge of the protein was neutralized with 100 mM NaCl. For energy minimization and MD simulations, the Amber14:EHT force field was used and the electrostatic interactions were evaluated by the particle-mesh Ewald method. Each system was energy minimized for 5,000 steps using the Conjugate Gradient method. For equilibration, the system was subjected to a 100-ps simulation to gradually heat the system from 10 K to 300 K. Next, a 100-ps NVT ensemble was generated at 300 K, followed by an NPT ensemble for 200 ps at 300 K and 1 bar. Then, for each complex, a 100-ns production trajectory was generated for further analysis. The trajectory analysis and frame export for the video was done using scripts shared by the CCG support group.

Kinetic fitting (K M, k cat, K i)

Fitting was performed using MATLAB (v.R2019a) from MathWorks by following a method with a script obtained from the author51. For the script, see ‘Script for fitting kinetic data in MATLAB’ in the Supplementary Information. Briefly, Michaelis–Menten differential equations with competitive product inhibition (equations (1)–(5)) were integrated using Euler’s method with a time step of dt = 0.1 s, where KM is the Michaelis–Menten constant, Ki is the product inhibition constant and kcat is the catalytic rate constant. [S]t, [P]t and [ES]t are the concentration of substrate, product and enzyme–substrate complex at time t, respectively. Ratedil is the rate of dilution of the substrate from the pipette to the cuvette and is estimated to be 2 s during the dilution and 0 otherwise. [S]0 is the initial concentration of substrate in the syringe before dilution. [E]0 is the concentration of enzyme in the cuvette and is assumed to remain constant throughout the course of the kinetics. Note that substrate addition typically dilutes the enzyme by less than 1% and is therefore negligible.

$$\frac{{{\mathrm{d}}\left[ {\mathrm{P}} \right]_t}}{{{\mathrm{d}}t}} = \frac{{\left[ {\mathrm{E}} \right]_0 \times k_{\mathrm{cat}} \times \left[ {\mathrm{S}} \right]_t}}{{\left( {K_{\mathrm{M}}\left( {1 + \frac{{\left[ {\mathrm{P}} \right]_t}}{{K_{\mathrm{i}}}}} \right) + \left[ {\mathrm{S}} \right]_t} \right)}}$$
(1)
$$\frac{{{\mathrm{d}}\left[ {\mathrm{P}} \right]_{t + {\mathrm{d}}t}}}{{{\mathrm{d}}t}} = \left[ {\mathrm{P}} \right]_t + \frac{{{\mathrm{d}}\left[ {\mathrm{P}} \right]_t}}{{{\mathrm{d}}t}} \times {\mathrm{d}}t$$
(2)
$$\frac{{{\mathrm{d}}\left[ {\mathrm{S}} \right]_t}}{{{\mathrm{d}}t}} = - \frac{{{\mathrm{d}}\left[ {\mathrm{P}} \right]_t}}{{{\mathrm{d}}t}} + \left[ {\mathrm{S}} \right]_0 \times{\mathrm{Rate}}_{\mathrm{dil}}$$
(3)
$$\frac{{{\mathrm{d}}\left[ {\mathrm{S}} \right]_{t + {\mathrm{d}}t}}}{{{\mathrm{d}}t}} = \left[ {\mathrm{S}} \right]_t + \frac{{{\mathrm{d}}\left[ {\mathrm{S}} \right]_t}}{{{\mathrm{d}}t}} \times {\mathrm{d}}t$$
(4)
$$\frac{{{\mathrm{d}}\left[ {\mathrm{ES}} \right]_t}}{{{\mathrm{d}}t}} = \frac{{\left[ {\mathrm{E}} \right]_0 \times \left[ {\mathrm{S}} \right]_t}}{{\left( {K_{\mathrm{M}}\left( {1 + \frac{{\left[ {\mathrm{P}} \right]_t}}{{K_{\mathrm{i}}}}} \right) + \left[ {\mathrm{S}} \right]_t} \right)}}$$
(5)

The fluorescence signal was found to be correlated with the concentration of ES and is fit according to equation (6), where the baseline is the native signal of the nanoantenna-SA-bAP complex, Fmax is the fluorescence signal of the nanoantenna-SA-bAP complex when all the enzyme is bound with substrate (that is, high substrate concentration) and Fprod is the impact of the product concentration on the fluorescence signal of the nanoantenna-SA-bAP complex. Fitting was performed by using the nonlinear least-squares solver lsqcurvefit in MATLAB, which minimizes the sum of the squares of the residuals between the raw data and the computed data. Then, when applicable, the 95% confidence interval of each parameter is calculated using the nlparci function in MATLAB.

$$F_{\mathrm{spike}} = {\mathrm{Baseline}} + F_{\mathrm{max}} \times \left[ {\mathrm{ES}} \right]_t + F_{\mathrm{prod}} \times \left[ {\mathrm{P}} \right]_t$$
(6)

Enzymatic equations were sometimes modified to accommodate specific characteristics of some substrates. For example, PPi upon cleavage generates two phosphate products, rather than the one phosphate product as generated by pNPP or 4MUP. Therefore, d[P]t/dt was multiplied by 2. ADP, ATP and GTP can react multiple times and this was considered by multiplying [S]0 by the number of reactive groups. All experiments were done in triplicate.

Preparation of biotinylated AP

For most of this project, we used commercially prepared biotinylated AP. We also prepared our own biotinylated AP to explore lot-to-lot variation issues (used in Fig. 2d and Extended Data Figs. 5 and 6). For this, we used unconjugated AP from Rockland and a Biotin Protein Labeling Kit from Roche Diagnostics. To avoid unwanted side reactions, we removed Tris from the enzyme buffer with a Nanosep Centrifugal Device with Omega Membrane 30K from Pall Corporation by rinsing ten times. Then, we followed the manufacturer’s instructions for the biotinylation kit by following ‘Procedure 2: Polyclonal antibody’ based on the mass of the protein.

Preparation of nanoantenna-AP covalent conjugate

AP was first diluted to 40 µM using PBS buffer (pH 7). Then, we added 3 equivalents of freshly prepared SPDP reagent (20 mM) in DMSO. AP was incubated with SPDP solution at room temperature for 30 min. Next, we used a Zeba spin desalting column to exchange the SPDP-modified protein reaction buffer for 10 mM HEPES, 150 mM NaCl, pH 8, and to remove reaction by-products and excess nonreacted SPDP reagent. Separately, we incubated the DNA nanoantenna (5′ T 6-FAM, 3′ SH) with 1 M dithiothreitol (DTT) in 40 µl TE buffer for 30 min at 37 °C. Then, we extracted with ethyl acetate and combined the aqueous phases. We then added 8 equivalents of reduced thiol DNA to the SPDP-modified AP and let it react for 1 h at room temperature. Note that to avoid side reactions, we used a DNA strand that did not contain guanine.

Presentation of data

Error bars on graphs and expressed values represent mean ± s.e.m. for three distinct measurements. Typically, experiments were performed in triplicate, with the following exceptions: Extended Data Fig. 4h and Supplementary Fig. 7d for enzyme ratios not near the maximum, as well as Extended Data Fig. 4e,f for covalent attachment of nanoantenna to AP, which had three injections of pNPP to the same sample. All MD simulations were performed once and all molecular docking simulations were performed ten times.

Reporting Summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.