1.

Introduction

As an emerging class of carbon nanomaterials, carbon dots (CDs) have garnered researchers’ interests in the past decade due to their excellent biocompatibility, replete surface functional groups, water dispersibility, and unique photoluminescence.^1–5 These extraordinary properties have opened new avenues for their advanced application in cell labeling, bioimaging, drug delivery, sensors, and energy related devices.^6–13

Among their favorable properties, the photoluminescence of CD has been widely utilized. CDs exhibit relatively symmetrical emission spectra on the wavelength scale, and the emission peaks are wide with a Stokes shift comparable to organic dyes.^14–17 Interestingly, the emission spectra can be tuned by exciting the CDs at various wavelengths, a phenomenon known as wavelength dependent behavior, which can potentially be exploited for multiplexing.^18,19 However, the real mechanism underneath the photoluminescence (PL) property is a subject of wide controversy.^20–22

To design new carbon materials with improved properties, understanding the underlying photophysics at the molecular level is crucial; however, due to its complexity, many aspects of the mechanism remain unknown. CDs show differential photoluminescence depending on the chemical structures including graphitic conjugated cores, molecular fluorophores, and surface defect states.^23–30 All CDs share the same ${\mathrm{sp}}^2$ graphitic core, which may contain nitrogen as part of an aromatic ring, despite their different synthesis routes. It has also been proposed that CD surfaces contain moieties rich in carbon, nitrogen, and oxygen. The goal of this work is to introduce the same functionalities in isomeric form and understand how the distribution of electron clouds influences the photophysical properties of the resultant CDs. Herein, we investigated the role of the N position in the amine ring of the precursor for CD preparation and its effect on the fluorescence properties of the synthesized nanoparticles (NPs). CDs from diamino pyridine were derived in a facile one pot hydrothermal synthesis. Fluorescence measurements revealed the typical excitation-dependent emission of these NPs but with various multiplicities of the emission centers and different Stokes shifts.

They were subsequently investigated for their size distribution by TEM and AFM, and it was concluded that the photophysical properties were independent of the size of the NPs. Multiple chemical characterization techniques revealed that the chemical compositions of these NPs were thoroughly different and were dominated by the amine position of diaminopyridine (DAP) and the aromatic heterocycles being formed in the system. Density functional theoretical (DFT) calculations led us to conclude that the para- and meta-positions of the amines in the original precursor isomers resulted in NPs with similar properties based on their molecular electrical potential surface’s properties.³¹ We also concluded that the bulk properties were an ensemble average of the fluorescent properties on the single-particle level and can be effectively engineered by the precise choice of the amine containing precursors.

CDs are emerging candidates for fluorescence-based bioimaging due to their unique properties, e.g., good PL, easy synthesis routes, economical synthesis, inexpensive starting materials, water solubility, small sizes, prominent biocompatibility, and excellent photostability.^32–34 As an alternative to quantum dots and organic dyes, CDs are a class of fluorescent carbon nanomaterials used in bioimaging, sensing, and optoelectronics. As a result of their unique optical properties, such as tunable emission, facile synthesis, and low toxicity, CDs have applications in biology, medicine, and the environment.^35–39 These properties make them translatable and advantageous over other sensing methods for the detection of oral microbiota under biological environment. The various fluorescent behaviors of these CDs have then been employed for selective discrimination of oral microbiota via a machine learning (ML)-driven analysis. The oral microbiota harbors hundreds of species, and it is the cause of several infectious diseases such as dental caries, periodontitis, tonsillitis, and systemic diseases, e.g., infective endocarditis and diabetes.^40–42 The current gold standard method for the oral microbiota discrimination is the identification of bacterial genus and species by 16S rRNA sequencing.^43,44 This process involves the bacterial DNA extraction and sequencing, which is a laborious, time-consuming, and expensive strategy. Here for the first time, we propose an array that works on the premise of various emergent functional groups in the CDs from the different isomers responding differently to the surface electronic properties of bacteria.

An efficient pattern recognition tool known as linear discriminant analysis (LDA) has been utilized for selective oral bacteria identification based on the fluorescence response of the built array.^45,46 This method is a supervised statistical method for classification and intends to maximize the difference between within-class and between-class variations.⁴⁷ The developed array was able to rapidly and accurately identify the type of bacteria and was substantiated from DFT calculations and docking results. The current technology can even identify different types of bacteria in a blended sample of oral microbiota, which could aid dentists in making informed decisions about the potential treatment for the dental diseases.

2.

Materials and Methods

3.

Results and Discussion

We created a library of CDs based on DAP via the cocarbonization of DAP compounds with urea in a hydrothermal method. In a typical synthesis, 0.18 g of each DAP isomer was mixed with 0.1 g urea and dissolved in 28 ml of water. The mixture was then transferred to an autoclave vessel for nucleation at 180°C for 2 h. Care was taken to minimize the systematic errors using the same autoclave vessel for all five samples and with the same cooling rate. The NPs were then filtered, dialyzed against water (sink condition), and freeze dried for subsequent use.

The transmission electron microscopy (TEM) and atomic force microscopy (AFM) of these NPs were carried out as shown in Figs. 1(a)–1(e). The transmission electron microscopy indicated the presence of spherical particles. It was realized that the anhydrous size of 2,3 DAP NPs, 2,4 DAP NPs, 2,5 DAP NPs, and 2,6 DAP NPs are almost similar and measured as $9\pm 3$, $8\pm 4$, $10\pm 2$, and $3\pm 1\mathrm{nm}$, respectively, whereas for 3,4 DAP NPs, the anhydrous diameter was much bigger and measured as $40\pm 7\mathrm{nm}$ [Figs. 1(f)–1(j)]. These CDs of $<10\mathrm{nm}$ in diameter generally have lower contrast than other high density, high atomic number metallic NPs under TEM owing to the weak electron scattering cross section of carbon. Additionally, none of these particles contain heavy atoms or are stained with heavy elements. As a result, TEM images of them tend to be low contrast. On the other hand, the AFM height measurement yielded similar results for all NPs, possibly due to the flattening of the NPs on the surface of mica.

Fig. 1

Physiochemical characterizations of DAP NPs: TEM images of (a) 2,3 DAP NPs, (b) 2,4 DAP NPs, (c) 2,5 DAP NPs, (d) 2,6 DAP NPs, (e) 3,4 DAP NPs; AFM images and the height distribution for (f) 2,3 DAP NPs, (g) 2,4 DAP NPs, (h) 2,5 DAP NPs, (i) 2,6 DAP NPs, (j) 3,4 DAP NPs; (k) electrophoretic zeta potential; (l) FTIR; and (m) Raman spectra of NPs.

Subsequently, the electrophoretic $\zeta$-potential showed negative values for all prepared CDs [Fig. 1(k)]. The absolute high zeta potential values ($\sim 10\mathrm{mV}$) pertain to the reasonable colloidal stability of these NPs, which originate from the replete hydrophilic surface functional groups as was concluded from FTIR.

FTIR, as a powerful tool in determining the surface functional groups, was utilized [Fig. 1(l)]. The CD surface presented an abundance of polar functional groups with a strong band observed between 3100 and $3500{\mathrm{cm}}^{-1}$ due to the stretching vibration of the polar functional group in $\mathrm{O}─\mathrm{H}$ and $\mathrm{N}─\mathrm{H}$, with the peak centered at $1131{\mathrm{cm}}^{-1}$ due to $\mathrm{C}─\mathrm{O}$. On the other hand, $\mathrm{C}═\mathrm{C}$, which is typical of aromatic structures appeared, between 1539 and $1635{\mathrm{cm}}^{-1}$. The peak centered at $1673{\mathrm{cm}}^{-1}$ could be related to amide and/ or $\mathrm{C}═\mathrm{N}$ as result of graphitic N, whereas the strong peak centered at $1497{\mathrm{cm}}^{-1}$ could be ascribed to $\mathrm{C}─\mathrm{N}═$. Finally, the asymmetric stretching vibration of $\mathrm{C}─\mathrm{N}$ and $\mathrm{C}─\mathrm{O}$ could be detected at 1127 to $1159{\mathrm{cm}}^{-1}$.

Subsequently, Raman spectroscopy was utilized to distinguish the D and G bands, which typically appear for CDs [Fig. 1(m)]. The D-band (diamond-band) ($1355{\mathrm{cm}}^{-1}$), or the disorder band, occurs as a result of the out of plane vibration of ${\mathrm{sp}}^2$ C in a pool of ${\mathrm{sp}}^3$ molecular defect states. On the other hand, the G band ($1605{\mathrm{cm}}^{-1}$), or the graphitic band, corresponds to E2g mode from the in-plane vibration in ${\mathrm{sp}}^2$ C within the aromatic domain. The ratio of D/G band was calculated, and it turned out that the D/G ratio is 2,6 DAP NPs (0.92) < 3,4 DAP NPs (1.10) < 2,5 DAP NPs (1.14) < 2,4 DAP NPs (1.18) < 2,3 DAP NPs (1.37). The increase in the D/G ratio implies the lowering of the crystallinity, meaning that 2,6 DAP NPS possesses the largest ${\mathrm{sp}}^2$ domain.

We performed UV–Vis spectroscopy to determine the absorbance associated with these NPs [Fig. 2(a)]. For all NPs, at least two absorbance states could be identified: the core state occurring in the UV region and a surface state. The core state, which is attributed to $\pi \to \pi *$ transition in the aromatic hydrocarbon domains, of 2,4 DAP NPs, and 3,4 DAP NPs appeared in the high-energy UV band, whereas 2,3 DAP NPs, 2,6 DAP NPs, and 2,5 DAP NPs underwent a bathochromic shift. On the other hand, the surface state, which indicates a transition between $n$-orbital of heteroatom (i.e., oxygen and nitrogen) and the $\pi *$ of polyaromatic core, appeared in the following order of wavelengths: 2,4 DAP NPs< 3,4 DAP NPs < 2,3 DAP NPs = 2,6 DAP NPs < 2,5 DAP NPs. Interestingly, a second surface state is attributed to a lower extinction coefficient for 2,3 DAP NPs and 2,5 DAP NPs, which is red shifted in 2,5 DAP NPs. This exitonic band could be due to lower energy species on the surface of CDs, which resulted in narrowing of the electronic band gap in these two cases. These results match well with the simulated UV–Vis results simulated from DFT calculations of the original precursors as indicated in Fig. 2(b). For better understanding, the experimental and theoretical (from DFT) UV–Vis spectra are plotted separately in Fig. S1 in the Supplementary Material.

Fig. 2

(a) Experimental UV–Vis spectra of various NPs and the inset indicates the magnified region between 233 and 390 nm. (b) The predicted spectra of DAP molecules based on DFT calculations. (c) 2D fluorescence map for (i) 2,3 DAP NPs, (ii) 2,4 DAP NPs, (iii) 2,5 DAP NPs, (iv) 2,6 DAP NPs, (v) 3,4 DAP NPs, the emissions centers are identified; and (vi) the normalized emission fluorescent spectra of NPs excited at 360 nm.

Figure 2(c) demonstrates the 2D excitation–emission photoluminescence maps for various CDs obtained from DAP. From an asymmetric shape of excitation–emission centers, an excitation-dependent emission phenomenon usually reported in CDs that disobeys the Kasha–Vavilov’s rule can be concluded.⁵⁵ The red edge excitation shift might be due to the radiative recombination by surface traps caused by various surface functional groups. The emission centers for the NPs were determined as follows: for 3,4 DAP NPs (311 and 365 nm) with the Stokes’ shift value of $4735.45{\mathrm{cm}}^{-1}$, for 2,5 DAP NPs (257 and 413 nm) with the Stoke’s shift value of $\mathrm{14,691}{\mathrm{cm}}^{-1}$, for 2,4 DAP NPs (281 and 335 nm) with the Stoke’s shift of $5742{\mathrm{cm}}^{-1}$, for 2,6 DAP NPs (251 and 431 nm) with the Stoke’s shift of $\mathrm{16,656}{\mathrm{cm}}^{-1}$, and for 2,3 DAP NPs at (251 and 360 nm) with the Stoke’s shift of $\mathrm{12,063}{\mathrm{cm}}^{-1}$. The difference between emission maxima and excitation maxima followed this trend: 2,6 DAP NPs < 2,5 DAP NPs < 2,3 DAP-NPs < 2,4 DAP NPs < 3,4 DAP NPs. The observed behavior can be justified based on the chemical properties of these NPs as was elucidated by XPS and mass spectroscopy (mass spec), which suggest that the photoluminescence is governed by a combination of both core states and the fluorophores at the edge state.

XPS aided us in identifying surface functional bonds, and the presence of C, N, and O was detected in the survey spectra with peaks at 285, 400, and 531 eV, respectively, (Figs. S2–S11 in the Supplementary Material). The carbon adventitious peak was corrected to 284.8 eV, and the C1 and N1S are shown in Figs. 3(a) and 3(b), respectively. The atomic percentage ratio of O/C and N/C is shown in Figs. 3(e) and 3(f), and it was determined that 3,4 DAP NPs has the highest level of oxidized species while also possessing the highest N/C doping. The decomposition of C1s core state revealed the presence of $\mathrm{C}─\mathrm{C}/\mathrm{C}═\mathrm{C}$ at 284.8 eV, $\mathrm{C}─\mathrm{O}/\mathrm{C}─\mathrm{N}$ at 286.3 eV, carbonyl group $─\mathrm{C}═\mathrm{O}$ at 288.1, and $─\mathrm{COOH}$ at 289.9 eV [Fig. 3(c)]. The binding energies in N1s region [Fig. 3(d)] at 398.9, 400.5, 401, and 404.5 eV were attributed to pyridinic nitrogen, pyrrolic nitrogen, and nitro, respectively. The O1s decomposition also revealed the presence of $\mathrm{C}─\mathrm{O}$ and $\mathrm{C}═\mathrm{O}$ at 534.4 and 532.1 eV, respectively. 2,5 DAP NPs has the most ${\mathrm{sp}}^2$ C followed by 2,3 DAP NPs, whereas the rest seem to have almost the same amount. 2,5 DAP NPs has the least ${\mathrm{sp}}^3$ C followed by 2,6 DAP NPs = 2,3 DAP NPs and then 3,4 DAP NPs = 2,4 DAP NPs, whereas 2,6 DAP NPs has the highest carbonyl amount; 2,3 DAP NPs and 2,4 DAP NPs are the only species having the carboxyl group.

Fig. 3

Deconvoluted (a) C1s region and (b) N1s region; (c) the comparison of various C-containing functional groups; (d) N-containing functional groups; and (e) the O/C and (f) N/C atomic percent distribution for different NPs.

The N in the CDs can exist in various forms, and it can drastically affect the fluorescent behavior. In these cases, N substitutes for the C atom in the polyaromatic cyclic structure, leading to the $\sigma$-bond formation. On the other hand, the aromatic π bond forms and the fifth electron enters the antibonding $\pi *$ molecular orbital to form either a six- or five-member ring. This leads to the creation of pyridinic and pyrrolic moieties that affect the electronic properties of CDs. Pyrrolic N enhances fluorescence, whereas pyridinic N inhibits it due to the greater distortion that is caused by the five-membered ring. Pyridinic N causes more bathochromic-shift than pyrrolic N due to the contribution of the lone pair electron in the $\pi$-system aka delocalized electron. This freely moving electron has various wavefunctions, which can result in the narrowing of the $\pi \to \pi *$ transition and hence a red-shift in the spectra.⁵⁶ Interestingly, graphitic nitrogen, which is known to trigger the red shift, is absent in the XPS spectra, which could suggest that the fluorescent is originating from the doped fluorophores formed during the synthesis rather than the core state. More importantly, the amine N, which also highly contribute to the red-shift in the CDs, are missing as determined by XPS, which could be due to full oxidation of the surface exposed amines.⁵⁷ Because XPS is a surface characterization technique, to get more insight into the composition of these NPs and understand their photophysical properties, which is not immediately apparent by other methods, we sought structural elucidation by mass spectroscopy analysis.

The predicted molecules corresponding to the $m/z$ values detected in mass spectroscopy and the proposed mechanism for the production of these molecules are shown in Figs. 4(a)–4(e). Furthermore, we did further time-dependent DFT calculations to numerically determine the bandgap values and the distribution of the frontier orbitals in the predicted products, as is shown in Fig. 4(f). From these mechanisms combined with theoretical calculations, several points can be concluded and can be correlated with the observed PL behavior.

Fig. 4

(a)–(e) The predicted molecules based on mass-spec results and (f) the band gap energy and the HOMO–LUMO energy. Note that molecular orbital energies are in Hartree Fock (HF).

First, the steric effect occurs between two amino groups in both the 2,3 DAP and 3,4 DAP compounds due to the presence of two amino groups in the ortho position. Both compounds showed similar fluorescence properties due to the similarities of their structural arrangements. However, more red shifted emission occurs for 2,3 DAP NPs than 3,4 DAP NPs, which can be explained by the presence of the imidazole ring that is formed during the reaction between diamino pyridine and urea in the thermal condition. The bandgap value was calculated to be 2.15 and 2.24 eV for molecules 2,3 DAP NP-P1 and 2,3 DAP NP-P2, respectively, which is the smallest bandgap among all of the predicted molecules. Second, the more blue shifted emission observed for 2,6 DAP NPs among other CDs can be explained due to the presence of highest band gap energy between HOMO and LUMO (3.99 eV). On the other hand, the 2,5 DAP NPs product with a bandgap of 2.73 eV showed a red shifted emission due to extended conjugation occurring as both amine groups are in para-position. Next, 2,4 DAP NPs showed a slightly more red shifted emission than 2,5 DAP NPs. This fluorescence behavior was due to two reasons: (i) the less band gap energy between HOMO and LUMO (2.35 eV) and (ii) the presence of more conjugation because of the involvement of a greater number of DAP molecules to get the final products.

Time-resolved photoluminescence (TRPL) spectroscopy was conducted with lasers at 390 nm [Fig. 5(a)] and 510 nm (Fig. S12 in the Supplementary Material), which correspond to the excitation of both the core and surface states, respectively. The NPs indicated double exponential decay when excited at both 390 and 510 nm, suggesting the presence of multiple emission centers on the surface and hence multiple radiative recombination pathways involved in this phenomenon. The photoluminescence of CD is still a matter of debate, and it has been attributed to various mechanisms, including the quantum size effects, surface traps, and intrinsic state.^58–60 CDs are comprised of a carbongenic ${\mathrm{sp}}^2$ core, whereas the surface is replete with surface functional groups that mediate the fluorescence properties.⁶¹ Herein, the relaxation of the photoexcited electrons during $\pi \to \pi *$ transition can result in two exponential decay components: the first decay, i.e., ${\tau }_1$, arises from the molecular state ${\mathrm{sp}}^2$ relaxation, whereas ${\tau }_2$ originates from the heterogeneous surface state.⁴⁸ The amplitude weighted average decay lifetime ${\tau }_{\text{average}}$ is shown in Fig. 5(b). The difference in ${\tau }_{\text{average}}$ can be attributed to the density, nature, and orientation of the surface functional groups, which directly affect the charge carrier trapping and excitation relaxation pathways as has been concluded from the XPS results. On the other hand, the information about mass spectroscopy led us to similar conclusions. In 2,3 DAP NPs, the radiative decay mechanisms were more well-defined, leading to a slower decay time. It is also interesting to note that the average lifetime was shorter in TRPL when excited at 510 nm compared with 390 nm, suggesting that the surface state has a more fleeting relaxation time compared with the core state.

Fig. 5

(a) TRPL spectra of NPs with ${\lambda }_{\mathrm{exc}}=390\mathrm{nm}$; (b) amplitude weighted average decay lifetime; (c) %quantum yield (QY) values for the five different DAP nanoparticles synthesized herein; (d) non-radiative decay rate; and (e) radiative decay rate.

Furthermore, the PL QY was obtained by comparing the absorbance (365 nm) over fluorescence integral of the NPs (${\lambda }_{\mathrm{ex}}=365\mathrm{nm}$) versus the similar parameter for quinine sulfate dissolved in 0.1 N ${\mathrm{H}}_2{\mathrm{SO}}_4$ ($\mathrm{OD}<0.1$) with a known QY = 0.53 [Fig. 5(c)]. The calculated relative QY for 2,3 DAP NPS, 3,4 DAP NPs, 2,4 DAP NPs, 2,5 DAP NPs, and 2,6 DAP NPs was 31%, 25%, 22%, 22%, and 22%, respectively.

To understand the transfer phenomena governing the photoluminescence and investigate the emission kinetics, we correlated the QY and TRPL results to calculate the radiative and non-radiative decay rate components of the emitters via the following equations:⁶²

By solving the above equations, one can obtain $k_{\mathrm{NR}}$ and $k_R$, which stand for non-radiative recombination rate and radiative recombination rate in ${\mathrm{ns}}^{-1}$, respectively [Figs. 5(d) and 5(e)].

To further elucidate the photophysical properties, we investigated the photo-blinking state of CDs using single-particle imaging in the TIRF microscopy [Figs. 6(a)–6(e)].⁴⁸ Figure 6(f) demonstrates an example of intensity time trace of X NPs, indicating a multistep blinking that could be attributed to multiple emissive centers on the CDs. The blinking of CDs was conceived with the characteristic on and off states as identified by intensity thresholding with the $\text{mean}\pm 3x$ standard deviation ($\sigma$).^49,63 The intermittent and stochastic nature of these CDs with bright spots at certain time points are indicated in time lapse image movies (Videos 1, 2, 3, 4, and 5; see Fig. 6 caption). Two parameters are used to determine the level of the brightness of DAP NPs: instantaneous intensity and duty cycle. The instantaneous intensity was determined to be quite similar for all of the DAP NPs [Fig. 6(g)]. The average of finding a particle in the emissive state during the acquisition cycle of NPs was determined and indicated in Fig. 6(f). On the other hand, the average duty cycle seems to vary between various NPs [Fig. 6(h)]. The origin in the brightness between different species was not clear but could possibly be due to the various trap state distributions on the CDs.

Fig. 6

Photophysical properties of DAP NPs on the single-particle level, Representative single-particle emission images of (a) 2,3 DAP NPs, (b) 2,4 DAP NPs, (c) 2,5 DAP NPs, (d) 2,6 DAP NPs, and (e) 3,4 DAP NPs as captured by an EMCCD camera. (f) Representative photoluminescence intensity trajectory indicative of particle brightness, (g) the instantaneous intensity of the NPs, and (h) the average ON duty time (Video 1, MP4, 30.7 MB; [URL: https://doi.org/10.1117/1.JBO.28.8.082807.s1] Video 2, MP4, 30.7 MB [URL: https://doi.org/10.1117/1.JBO.28.8.082807.s2] Video 3, MP4, 30.7 MB; [URL: https://doi.org/10.1117/1.JBO.28.8.082807.s3] Video 4, MP4, 30.7 MB; [URL: https://doi.org/10.1117/1.JBO.28.8.082807.s4] Video 5, MP4, 30.7 MB; [URL: https://doi.org/10.1117/1.JBO.28.8.082807.s5]).

In addition, we carried out bleaching experiments to determine the photostability of the CDs. All NPs presented a single-step bleaching time that can accordingly be attributed to the presence of highly coupled chromophores previously noted in the conjugated polymers (Figs. S18–S22 in the Supplementary Material).^64–66

Thus having unraveled the underpinning PL mechanisms mediated by the precursor’s composition, five DAP NPs were utilized as a building element for the sensor arrays [Fig. 7(a)]. The fluorescence response of the five DAP NPs when interacted with each bacteria represent the fluorescence signature toward each bacterium. The signal from each DAP NPs was found to be unique to the type of bacterium [Fig. 7(b)]. In addition, the synthesized DAP NPs showed a distinguishable signal toward blended samples of bacteria [Fig. 7(c)]. To evaluate the ability of the built sensor array using the five DAP NPs, the fluorescence signatures of each bacteria [Figs. 7(b) and 7(c)] produced by the sensor array were analyzed using the LDA, a powerful statistical method extensively harnessed in pattern recognition.^67–70 Using LDA, the fluorescence pattern of the bacteria sample could be transformed to a 2D canonical score [Fig. 7(d)]. First, the model was trained using a training dataset consisting of the fluorescence response to the five oral bacteria. The five oral bacteria were well-clustered into five groups and discriminated completely from each other. Next, we challenged the model to predict the type of bacteria from a blinded testing set of fluorescence data. The five oral bacteria were classified with 100% accuracy of discrimination, and the assay proved to provide a highly effective sensor array for pathogen identification, as shown in Fig. 7(d).

Fig. 7

(a) Schematic illustration of sensor array from 5 different NPs and oral bacterial separation based on ML tools. The fluorescence output of NPs when exposed to (b) single species and (c) coculture of bacteria. (d) LDA classification of single bacterial strain and (e) a coculture of several bacterial strains achieving 100% separation.

The increase in emission intensity indicates that the bacteria work as an electron donor, whereas the DAP NPs serves as an electron acceptor.⁷¹ The unique fluorescence response of each DAP NPs to each bacterium may be attributed to the fact that each one of the five DAP NPs is varied in terms of the density, nature, and orientation of the surface functional groups, which directly affect the charge carrier trapping and excitation relaxation pathways and thus their fluorescence signal when interacted with the bacterial surface.

In clinical settings, the samples will contain mixed bacterial specimen. Thus the sensor array should be able to identify the oral bacteria in a complex sample with a bacterial blend. To this end, eight representative blended samples consisting of four mixtures of two species of microbes and four mixtures of three species of microbes were used as the targeted mixed culture of the oral bacteria. The ML tool and the LDA-based tool were found to be very valuable for properly analyzing the photoluminescence data.^72–75 LDA was used to transform the fluorescence output obtained from the sensor array as a response to bacterial mixtures to a 2D canonical score plot. The LDA algorithm was able to cluster the eight blended pathogens into eight groups with a 100% discrimination accuracy [Fig. 7(e)]. Based on the established function from the training part, 16 randomly selected blended samples were also completely identified in a blind test with a detection accuracy of 100%. As a practical application of the DAP NPs, it is proved that these particles can be used to construct a sensing array to detect bacteria. The combined use of this array along with a powerful algorithm can further expand the application of the array to identify the type of pathogen in a tested sample. We recognize that, under complex in vivo conditions, the specificity of CDs for binding to target bacterial species can be a limitation. An efficient computational algorithm will therefore play a vital role in this study to ensure that the CDs only bind to the desired bacteria and not to other non-target species that may be present in the oral environment.

Further, these experimental results were corroborated with computational studies. The differential response of DAP NPs toward different microbial strains led us to investigate their interaction with various surface proteins available on the bacteria. It may be hypothesized that, over the short exposure time of NPs to the bacterial strains, the CDs could only interact with the surface exposed proteins to generate a differential sensing response. Accordingly, we considered maltodextrin binding protein MalE1 for L. casei (PDB ID: 5 MTU); ${\mathrm{Zn}}^{2+}$-dependent intercellular adhesion protein for S. epidermidis (PDB ID: 4 FUM); V-region of antigen I/II for S. mutans (PDB ID: 1 JMM); type III-A Csm-CTR1 complex, AMPPNP bound protein of S. salivarius (PDB ID: 6 IFK) and SrpA Adhesin protein of S. sanguinis (PDB ID: 5EQ2) (Fig. S23 in the Supplementary Material).

On the other hand, the model structures of DAPCDs were built as different ovalene derivatives⁴⁸ and energy minimized by the B3LYP/3-21G(d) method⁷⁶ in GAMESS software.⁵¹ The HOMO–LUMO levels were also calculated; these indicated that HOMO was largely located over the graphene-like $\pi$-motif structures, whereas LUMO was located over the pyridine-like moieties for all of the structures (Fig. S24 in the Supplementary Material). The HOMO–LUMO gap was found to be the lowest for 2,5 DAP NPs, indicating its plausible faster response toward the analytes. These energy-minimized CDs were then docked against the surface exposed proteins of the studied bacteria, and the results were closely monitored. Table S2 in the Supplementary Material shows the comparative free energies of binding and clustering efficiencies of CDs with the surface proteins. Comparing both the average free energy of binding and clustering efficiency of the DAP derivatives, it can be concluded that 2,5 DAP NPs interacted more strongly with the surface proteins of L. casei, S. salivarius, and S. sanguinis, whereas 2,3 DAP NPs and 2,6 DAP NPs interacted more strongly with the surface proteins of S. epidermidis and S. mutans, respectively. It is to be noted that the trend might differ with the change in surface proteins under study. The most stable docked geometries of these protein–ligand complexes are presented in Fig. 8 and Fig. S25–S29 in the Supplementary Material, and the histograms representing the conformational binding energies of each of the complexes were shown in Tables S3–S7 in the Supplementary Material.

Fig. 8

Representative pictorial representation of the most stable docked geometries of V-region of antigen I/II for S. mutans (PDB ID: 1JMM) with (a) 2,3 DAP NPs, (b) 2,4 DAP NPs, (c) 2,5 DAP NPs, (d) 2,6 DAP NPs, and (e) 3,4 DAP NPs.

4.

Conclusions

The hydrothermally derived CDs prepared under the same controlled condition from DAP isomers and urea were synthesized and characterized for their photophysical properties. We indicated via a comprehensive bulk state and single-particle level investigation that the PL properties of CDs can be regulated by the precursors’ isomeric position of nitrogen. We attributed these changes to the molecular products formed during the synthesis; their photophysical responses were predicted by DFT calculations. We emancipated the difference in the PL properties of CDs in a rapid method relying on ML algorithms to segregate the bacterial species leading to dental biofilm, which has huge implications for oral health. In the future, we will explore the efficacy of these CDs in differentiating various pathogenic bacteria in dental plaque developed under in vivo conditions. However, the major limitation to this study is the stability, poor QY of the CDs under in vivo conditions. Our future approaches will utilize various coating materials to improve the in vivo efficacy of the system. Further, the biocompatibility and immunogenicity of these CDs under in vivo conditions will be explored in a detailed manner. In our follow-on studies, the in vivo photoluminescence of the CDs will also be analyzed through ML algorithms to automatically identify the presence of pathogenic bacteria.