Monitoring the Viral Transmission of SARS-CoV-2 in Still Waterbodies Using a Lanthanide-Doped Carbon Nanoparticle-Based Sensor Array.

The latest epidemic of extremely infectious coronavirus disease 2019 (COVID-19) has created a significant public health concern. Despite substantial efforts to contain severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) within a specific location, shortcomings in the surveillance of predominantly asymptomatic infections constrain attempts to identify the epidemiological spread of the virus. Continuous surveillance of wastewater streams, including sewage, offers opportunities to track the spread of SARS-CoV-2, which is believed to be found in fecal waste. To demonstrate the feasibility of SARS-CoV-2 detection in wastewater systems, we herein present a novel facilely constructed fluorescence sensing array based on a panel of three different lanthanide-doped carbon nanoparticles (LnCNPs). The differential fluorescence response pattern due to the counterion-ligand interactions allowed us to employ powerful pattern recognition to effectively detect SARS-CoV-2 and differentiate it from other viruses or bacteria. The sensor results were benchmarked to the gold standard RT-qPCR, and the sensor showed excellent sensitivity (1.5 copies/μL) and a short sample-to-results time of 15 min. This differential response of the sensor array was also explained from the differential mode of binding of the LnCNPs with the surface proteins of the studied bacteria and viruses. Therefore, the developed sensor array provides a cost-effective, community diagnostic tool that could be potentially used as a novel epidemiologic surveillance approach to mitigate the spread of COVID-19.

Introduction

Since the onset of the coronavirus disease 2019 (COVID-19) outbreak, stringent public health and social measures have been implemented to slow the spread of the virus. Substantial efforts have been devoted to containing the virus within a specific location but considering many uncertainties over the transmissibility and virulence of this highly contagious virus, the success of these efforts is limited. Moreover, it remains a highly complex technical task for medical professionals to track suspected cases of infection practically and efficiently from individual households. Such a massive undertaking is time-consuming and labor-intensive and, at this highly critical time, is limited by the availability of testing technologies. At this time, Centers for Disease Control and Prevention (CDC) is following a logical model for monitoring and evaluating community mitigation strategies for COVID-19. Therefore, environmental monitoring of the existence of the virus in wastewater, known as wastewater-based epidemiology (WBE), may be an early warning method and an active method for quick diagnosis of COVID-19 outbreaks irrespective of symptoms.[1,2]

COVID-19 is a respiratory disease; however, growing evidence suggests that coronavirus genomes are shed in the feces and ultimately reach wastewater.[3−5] Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) surveillance in sewage may be considered a sensitive tool to monitor the spread of the virus among the population.[6] A significant proportion of undiagnosed and asymptomatic carriers shed SARS-CoV-2 in the stool.[7−11] WBE is an early warning tool of the circulation of the virus in the population, including symptomatic and asymptomatic shedders. Studies conducted in Europe and Australia indicated that wastewater monitoring may provide a practical method to predict the potential spread of SARS-CoV-2.[7−11] The presence of SARS-CoV-2 was detected in sewage samples, a while before the public health officials reported the first case of COVID-19 in these communities. In addition, numerous studies have shown that SARS-CoV-2 can normally survive in an appropriate environment for up to a few days after leaving the human body.[12−14] Due to the limited availability of rapid diagnostic tests along with the inaccessibility of advanced instrumental techniques to all the diagnostic centers, especially the remote ones, there is potential that the analysis of SARS-CoV-2 in community wastewater-based systems could trace COVID-19 sources. WBE could therefore serve as an “early warning tool (sentinel surveillance)” of the circulation of the virus in the population, including symptomatic and asymptomatic shedders.[15−20]

The importance behind the wastewater-based surveillance of SARS-CoV-2 has thus been realized, but the rapid and early discrimination of SARS-CoV-2 from other viruses and microorganisms using the currently available techniques remains the major obstacle. Although polymerase chain reaction (PCR) is excellent in terms of sensitivity and specificity, requirements for complicated laboratory-based sample handling and a long period of data processing and review (4–6 h) are not conducive to real-time and effective onsite monitoring of samples.[21] Additionally, the wastewater contains a lot of other components apart from the genetic material. For example, the interference of other microorganisms such as Escherichia coli (E. coli), Streptococcus mutans (S. mutans), and Bacillus subtilis (B. subtilis) in wastewater may impose different stressors on the PCR assay and can affect the interpretation of the results.[22,23] Therefore, it is extremely valuable to establish an effective, transportable, and reliable analytical method to detect and differentiate low-level SARS-CoV-2 sources from other microorganisms accurately and quickly via WBE to confirm suspected cases and to screen asymptomatic infected cases without centralized laboratories.

Nanoparticles (NPs) have widely been utilized for detecting several viruses and microbes,[24−36] for example, detection of SARS-CoV-2 using suitably designed gold NPs,[37−39] influenza A virus using antibody-functionalized gold NPs,[40]E. coli using gold NPs,[41] magnetic NP system,[42] fluorescent organic NPs,[43] upconversion NPs, and fluorescent silica NPs.[44] In this context, carbon nanoparticles (CNPs) represent an emerging class of luminescent nanoprobes that offer multicolor, tunable emission based on the engineered nanoscale surface.[45−48] The other attributes that make these particles attractive for sensing applications are low photobleaching, excellent biocompatibility, and their ability to respond rapidly to the subtle changes of the nanoscale environment.[49−51] CNPs have therefore been successfully used for sensing pollutants[52] and various biological targets[53] utilizing their unique environmentally sensitive optical response.[54] To make the CNPs more amenable to change of the environment and to construct sensor array elements, lanthanide (Ln) metal doping can be explored. Lanthanide NPs have inherent advantages due to narrow emission lines and long lifetimes of lanthanide dopants in the host lattice, which could be used for detection of chemical as well as biological molecules. Here, lanthanide NPs are any NPs that are incorporated with lanthanide ions resulting in luminescent emission and large anti-Stokes or Stokes shift. The large anti-Stokes or Stokes shift changes the absorbed light into emission lines spanning from the infrared to ultraviolet region based on the choice of the lanthanide dopant. Furthermore, the lanthanide NPs exhibit excellent photostability that far exceeds that of other organic (fluorescent proteins and dyes) and inorganic fluorescent probes (quantum dots). For example, lanthanide NPs are known to exhibit stable emission for 5 h.[55−59] Therefore, utilizing lanthanide as a dopant metal can enhance the intrinsic photophysical properties of lanthanide-doped carbon nanoparticles (LnCNPs) to act as an excellent candidate of the sensor array for pathogenic virus and bacterial sensing. Lanthanides have been shown in the past to considerably expand the scope of array-based sensing, for example, detection of glyphosate and proteins and metal ions in biofluids, pH detection in human MCF-7 cells, etc.[60−63] However, to the best of our knowledge, LnCNPs acting as a recognition unit for a sensor array for detection of pathogens or bacteria is not explored yet. Therefore, LnCNPs may act as an excellent candidate for recognition as well as a transducer element for making a sensor array for pathogenic virus and bacterial sensing.

In this study, we report an easily accessible sensor array that selectively discriminates viral strains (SARS-CoV-2, SARS-CoV, and H1N1) from bacterial species (E. coli,[64−66]S. mutans,[67,68] and B. subtilis(66)) and accurately detects SARS-CoV-2 at a sensitivity level equivalent to real-time quantitative PCR (RT-qPCR). The sensor array comprises three lanthanide metal-doped CNPs (LnCNPs) and a gelatinous 3D matrix (agarose). As different pathogen species have different surface chemical and physical properties, distinct fluorescence signal patterns are generated from counterionic–ligand interactions with polycationic-caged CNPs, which can then be differentiated using a carefully designed machine learning methodology.[48,69−73] The varied response of the sensor array has also been investigated through molecular docking among the LnCNPs and the surface proteins of the studied viruses and bacteria. The theoretical docking results closely corroborate with the experimental sensor patterns. Overall, a powerful sensing platform has been demonstrated herein for the rapid and selective discrimination of COVID-19 causative virus, SARS-CoV-2, from five other different viral and bacterial contaminants present in still waterbodies with >95% accuracy. Scheme illustrates the conceptual demonstration of wastewater-based surveillance for SARS-CoV-2 transmission using our newly developed sensor array. The proposed system is straightforward and can be applied for the identification and detection of various other viral and bacterial strains in freshwater samples such as tap and lake water. With a few modifications, this method can further be introduced for the detection of other pathogens, paving the way for new techniques for alternative, low-cost, and rapid multiplexed detection systems.

Scheme 1

Schematic Representation of the Sensing Mechanism of Our LnCNP Sensor Array

The lanthanide-doped CNPs interact differentially with the viruses and bacteria generating distinct photoluminescence responses, which are then discriminated from each other by a machine learning-based pattern recognition algorithm. Clear discrimination and rapid detection of SARS-CoV-2 viral transmission in still waterbodies have been achieved using the LnCNP-based sensor array.

Experimental Section

The details of chemical reagents and solutions and NP characterization are provided in the Supporting Information.

Synthesis of LnCNPs

The synthesis of doped CNPs is based on the facile hydrothermal process. Lanthanide chloride, either PrCl3, GdCl3, or YbCl3 (300 μL, 0.2 mmol/mL), and polyethyleneimine (PEI) (1 mL,100 mg/mL) were dissolved in 20 mL of tannic acid solution (340 mg/mL), and the solution was then transferred into a 50 mL Teflon autoclave. After heating at 180 °C for 5 h, the autoclave was cooled down to room temperature naturally. The CNPs were probe-sonicated for 30 min (Q700, Qsonica Sonicator, Newtown, CT) in an ice bath (4-amp pulse; 2 s on and 1 s off). After this, the products were filtered using a 0.2 μm syringe filter to remove large entities and then lyophilized to obtain a brownish-color powder.

Quantum Yield Calculation

A quinine sulfate solution in 0.1 M H2SO2 with a concentration of 5 mg/mL was prepared and used as a reference solution to calculate the quantum yield of the three LnCNPs. A volume of 100 μL of the quinine sulfate solution was used to record the fluorescence spectra. The absorbance spectra of the three CNPs were recorded using a diluted sample (10 μL of the particle solution to 990 μL of deionized (DI) water). The quantum yield (Ø) of the three LnCNPs was calculated using the equation given below:I is the integration of the fluorescence spectrum, A is the absorbance of the sample at 365 nm, and η is the refractive index of the used solvent. The variable with the “Ref” subscript represents the values associated with the quinine sulfate. ØRef = 0.54 for the quinine sulfate dye.

Fabrication of a Sensor Platform

The as-synthesized LnCNPs (68 mg) were dispersed in 100 mL of DI water containing 800 mg of agarose at 70 °C. The solution was then deposited in 96-well plates (100 μL for each well) and allowed to solidify naturally under room temperature.

Experimental Procedure for Pathogen Selective Identification

A volume of 100 μL of virus and bacteria with various concentrations was transferred to the sensor platform deposited in a 96-well plate, and three repeated experiments were performed. For screening the viruses and real-time sensing, the wastewater was spiked with 105 genome equivalent/mL of the virus under investigation. For bacterial sensing, the wastewater was spiked with bacterial cell densities having OD600 = 0.2, that is, ∼107 of bacterial cell populations. All the fluorescence measurements were recorded using a microplate reader with λex/λem = 360/460 ± 40 nm or λex/λem = 400/500 ± 40 nm. The fluorescence of the sensor without viruses and bacteria under the same conditions was recorded as a control. The fluorescence intensities were recorded for 15 min. From these measurements, the relative fluorescence intensity was calculated and then introduced to the ML algorithm to generate the viral and bacterial-specific results using MATLAB. Each value was an average of nine independent measurements.

Results and Discussion

In this study, luminescent CNPs were initially prepared from a combined carbon source of tannic acid and a cationic macromolecule, for example, long-chain branched PEI, using a hydrothermal process. Hydrothermal-based carbonization is a convenient and rapid approach for the formation of lanthanide-doped CNPs. Tannic acid has a low carbonization temperature, while PEI can passivate the surface of CNPs due to its polyamine structures. Therefore, under hydrothermal conditions (i.e., high temperature and pressure), the tannic acid gets carbonized along with simultaneous in-situ passivation by PEI. Lanthanide dopants were further used to yield Ln-CNP-PEI (LnCNPs). The PEI macromolecule has been confirmed to be stable and undergoes no carbonization process throughout the hydrothermal treatment. PEI was found to be stable under 200 °C, and the obtained solution from the hydrothermal process possessed an unchanged color with no emission properties. Based on these observations, we propose the structure of LnCNPs as shown in Figure a for lanthanide doping, and three metals were selected, that is, Pr, Gd, and Yb. Therefore, three types of CNPs were studied in this work, that is, praseodymium-doped CNPs (PrCNPs), gadolinium-doped CNPs (GdCNPs), and ytterbium-doped CNPs (YbCNPs). The successful metal doping with Pr, Gd, and Yb was confirmed by inductively coupled plasma–optical emission spectroscopy (ICP-OES). This technique analytically quantified the content of lanthanide dopant as ∼3.792, 17.25, and 14.24 wt % for Pr, Gd, and Yb of the total NP weight, respectively. Our research group and others previously showed that tannic acid is capable of chelating metals; thus, using tannic acid (TA) as a starting material would help in the formation of a metal–phenolic complexed state.[74−77] Therefore, the lanthanide metals might form metal–phenolic coordination on the LnCNPs, and also they can be trapped in the CNP matrix during the synthesis process. Figure b shows a representative band gap model that was developed to explain the change in the photoluminescence (PL) due to the interaction of LnCNPs with viruses and bacteria. LnCNPs exhibited a smooth absorption–emission transition. However, in the presence of microorganisms, an intermediate band gap appears to play a role in altering the electron/hole recombination processes leading to a change in their PL.

Figure 1

(a) Schematic representation of the synthesis of LnCNPs. TA and branched PEI were used as carbon sources. Three LnCNPs were synthesized by the doping of CNPs with Pr(III), Gd(III), and Yb(III), known as PrCNPs, GdCNPs, and YbCNPs, respectively. (b) Conceptual band gap diagram of fluorescence manipulation due to the interaction of LnCNPs with virus or bacteria.

The as-synthesized LnCNPs were found to have positive electrophoretic potential values, which confirmed the surface abundance of positively charged amino groups (Figure a) on the sensing element. On the contrary, the viral and bacterial envelopes have varied lipid compositions with the abundance of phosphatidylethanolamine and bacterial cell surfaces were profused with anionic components, for example, long anionic polymers and teichoic acid. The analytes are therefore expected to participate in a counterionic–ligand interaction with the LnCNPs. We anticipate that the nonspecific interaction between the LnCNPs and pathogens (bacteria and viruses) will eventually lead to a differential fluorescence response. Furthermore, to reveal the size and the morphology of the synthesized LnCNPs, we used NP tracking analysis and TEM measurements. Through NP tracking analysis (NTA), we found that PrCNPs (114.1 ± 11 nm) and GdCNPs (108.9 ± 20 nm) have comparable hydrodynamic sizes, which are higher than that of YbCNPs (75.4 ± 23.8 nm) (Figure b). However, by using TEM, we found that the average diameter of the three LnCNPs was 27.42 ± 7.07, 12.81 ± 4.84, and 10.9 ± 2.82 nm for Pr-, Gd-, and Yb-CNPs, respectively (Figure c–e). It is worth mentioning that NTA measures the hydrodynamic diameter of the particles with its hydration layer; therefore, the obtained diameter is much larger than that obtained from TEM measurements, which provides the anhydrous diameter of the NPs. Additionally, UV–vis spectroscopic studies revealed one characteristic absorbance peak (Figure f) at 266 nm. This peak can be attributed to the n−π* transition of the C=O band and the π–π* transition of the conjugated C=C band. This peak was found to be common among all three LnCNPs. Furthermore, a comparison of the UV–vis spectra of the three samples revealed that the absorbance in the higher wavelength regions (>300 nm) was enhanced for PrCNPs accompanied with PL enhancement compared to GdCNPs. However, the fluorescence emission for YbCNPs was found to be the maximum among all when excited at 360 nm. A possible explanation for this observation is that the dopants get entrapped within the core of CNPs in selective quantities, disturbing the ordered ring structure of carbon and thereby creating novel emission energy traps, providing different fluorescence outcomes.[78−80] Fluorescence emission was therefore selected as an output signal as it provided higher sensitivity than absorption.[74,81,82] To further investigate the surface chemistry of the synthesized particles, FTIR spectroscopy measurements were carried out. The FTIR spectra (Figure g) show that the metal-doped CNPs possessed abundant hydrophilic groups on their surface such as O–H and N–H (3260 cm–1), which confirmed their dispersibility and stability in aqueous media. Moreover, stretching vibrations of C=C or C=O (1602 cm–1), C–N (1324 cm–1), and C–H (1477 cm–1) bands were also observed for each sample as highlighted by the dotted box, indicating the formation of polyaromatic structures in the CNPs during the reaction process. CNP-PEI shared many characteristic peaks of PEI, such as N–H at 3260 cm–1 and C–N at 1324 cm–1.[83,84] These results revealed that while TA got carbonized during pyrolysis, the PEI macromolecular chains maintained their structural integrity. This can further be substantiated from the known stability of PEI at 200 °C.[83] The LnCNPs thus showed remarkable aqueous dispersibility. No precipitation or aggregation was observed in the LnCNP suspension for up to 3 months when stored at an ambient temperature protected from light. The ζ-potential values for LnCNPs were found to be positive and greater than +20 mV for all three LnCNPs (Figure a), confirming sufficient colloidal stability when suspended in water.[46] Interestingly, our synthetic procedure resulted in LnCNPs with varying ζ-potentials. Therefore, a differential degree of counterionic–ligand interaction between the viruses/bacteria and LnCNPs is anticipated leading to distinct fluorescence fingerprints for each of the microorganisms.

Figure 2

Physicochemical characterization of LnCNPs. (a) Zeta potential measurement of the three LnCNPs. (b) Hydrodynamic size measurement of LnCNPs. (c–e) Transmission electron microscopy (TEM) images of Yb-, Gd-, and Pr-doped CNPs, respectively. Individually doped CNPs are outlined by circles. (f) Normalized absorption curves and their respective PL spectra under 360 nm excitation. (g) Fourier transform infrared (FTIR) spectra of the three LnCNPs.

For constructing the sensor array, the purified LnCNPs were then embedded in agarose to fabricate a 3D gelatinous-like platform. Agarose was chosen because of its high loading capacity and low background fluorescence (Figure S1), making it an ideal candidate for the fabrication of the sensor platform. Moreover, the large pore size of agarose facilitates the interaction between the viruses/bacteria and the nanoprobe. Figure a,b shows the representative scanning electron microscopy (SEM) images of the LnCNP–agarose sensor platform, confirming that the NPs are embedded within the agarose gel. The photographs of the LnCNPs embedded within the agarose gel confirmed the presence of the NPs within the matrix of the pristine, transparent agarose gel (Figure c,d).

Figure 3

(a) Representative SEM image of LnCNPs embedded within the agarose matrix. (b) Magnified image of panel (a) depicting the distribution of NPs within the sensor platform. (c) Photograph of the pristine agarose gel in a well plate. The inset shows the pristine agarose gel outside the plate. (d) Photograph of a representative sample of LnCNPs embedded within the agarose gel inside a well plate. The inset shows a sensor platform outside the plate. The pristine agarose gel film is transparent in the absence of LnCNPs; however, in the presence of LnCNPs, the color of the film became darker.

Fluorescence emission measurements were then conducted to assess the interaction of the embedded LnCNPs with the viruses and bacteria studied herein. We were primarily interested in monitoring the viral transmission of COVID-19 causative virus, SARS-CoV-2, in still waterbodies, which is one of the major concerns behind the community spread of this virus. We realized that mitigation of this community spread could be achieved if we can efficiently and rapidly discriminate the virus in still waterbodies through a suitably designed sensor platform. In this context, the viral discrimination of SARS-CoV-2 from SARS-CoV and influenza A (H1N1) has been realized. It was also understood that the other interferences during the sensing of SARS-CoV-2 in still waterbodies would be either from the medically relevant pathogenic bacteria and regular contaminants in hospital wastewater, including E. coli and S. mutans, or from the common nonpathogenic soil bacterium, B. subtilis. These bacteria are commonly found in the human gastrointestinal tract (GI system) and include two Gram-positive bacteria (S. mutans and B. subtilis) and one Gram-negative bacteria (E. coli). Within this set, B. subtilis and E. coli are quite distantly related eubacteria, providing a challenging testbed for bacterial sensing. The PL response of the embedded LnCNPs was then monitored for all these six variants of viruses and bacteria. It was observed that the intrinsic fluorescence of Gd-, Yb-, and Pr-doped CNPs was enhanced as a response to the SARS-CoV-2 virus, whereas it was either quenched or enhanced in the presence of the other viruses (SARS-CoV and H1N1) and different bacterial cells (Figure a, Figure S2a, and Figure S3). These observations indicate that the different species selectively affect the inherent photophysical properties of LnCNPs. The observed selectivity toward various viral and bacterial strains comes mainly from their design-contributing factors.

Figure 4

Sensor array response of three viral and three bacterial species including SARS-CoV-2. (a) Fluorescence response pattern of three viruses, SARS-CoV-2, H1N1, and SARS-CoV, and three bacteria, E. coli, B. subtilis, and S. mutans. Each value was an average of nine independent measurements; error bar shows the standard deviation of these measurements. λex = 360 ± 40 nm and λem = 460 ± 40 nm. I0 and I are the fluorescence intensity of CNPs in the absence and presence of bacteria or viruses, respectively. (b) The inset is the magnified view of the fluorescence response pattern of three viruses, SARS-CoV-2, H1N1, and SARS-CoV (c) 3D canonical score plot of the fluorescence response patterns determined by linear discriminant analysis (LDA). The plot shows one view of the 3D plot of the factors obtained from applying LDA to the fluorescence signal (the concentration of the bacteria is OD600 = 0.06 and that of the viruses is 100 genome equivalents per mL).

CNPs are characterized by their inherent PL properties. Our group and others found that the surface states are the dominant factor controlling the PL variations of CNPs.[26,45,69,85] For example, the CNP fluorescence can be tuned by varying the degree of surface oxidation or by the interaction with anionic or cationic objects. We hypothesized that different viral and bacterial strains will have a unique surface chemistry, which in turn will interact with each LnCNP differently. The electrostatic interaction between the viral or bacterial cell and the CNPs is therefore dominant by the particle surface charge. Each CNP has a distinct surface charge making its interaction with each pathogen or nonpathogen different. This difference in interaction will lead to a distinct fluorescence signature presumably due to the changes in the surface state of the CNPs. Second, the inspiration for the use of lanthanide for selective viral and bacterial sensing comes from our understanding of selective lanthanide recognition in microbial biology. A myriad of studies has enhanced our understanding of the principles underlying biological recognition and utilization of rare-earth metals such as lanthanides, which may facilitate the use of lanthanides for various bionanotechnological applications. Recent studies confirmed a strong biological correlation of lanthanide’s stimulatory properties toward bacteria, including strong biochemical evidence that lanthanides can act as inherent metals in bacterial enzymes.[86] Very recently, it has been demonstrated that the specific enzymatic reaction of methylotrophic bacteria is highly dependent on the presence of lanthanides, emphasizing the biological relevance of these trace metals. In this study, our goal is to introduce three different lanthanides as dopants in CNPs to impart their selectivity toward various viral and bacterial strains. Once the differential interaction of bacterial cells and viruses with LnCNPs was confirmed by obtaining distinctive fluorescence responses, the particles were used to sense the presence of SARS-CoV-2 and separate it from other viruses or bacteria (pathogenic or nonpathogenic).

Wastewater samples containing the SARS-CoV-2 virus have been tested using the LnCNP sensor array. The results from the sensor were benchmarked with the gold standard RT-qPCR (GoldBio commercial kit). The sensor array detected SARS-CoV-2 in samples with various viral loads as demonstrated from their Ct numbers and was also capable of detecting the virus in samples with low viral copy numbers (Ct > 35) (Table ). Thus, the environmental contamination caused by the SARS-CoV-2 virus can be easily detected using our sensor.

Table 1

Benchmarking of the LnCNP Sensor Array Results with RT-qPCR in Wastewater Samples

sample number	PrCNP sensor response, ΔI/I0	GdCNP sensor response, ΔI/I0	YbCNP sensor response, ΔI/I0	RT-PCR, (Ct)
sample #1	0.050	0.345	0.282	13.4
sample #2	0.048	0.139	0.135	28.5
sample #3	0.112	0.313	0.0713	37.89
sample #4	0.128	0.286	0.037	36.16

To evaluate the sensor response with various concentrations of bacteria/pathogen, E. coli and SARS-CoV-2 were used as model organisms. We found that the relative change in the fluorescence signal of the YbCNPs was linearly proportional (R2 = 0.903) with the SARS-CoV-2 concentration (Ct numbers) obtained from real-time PCR (RT-PCR) as shown in Figure S4. The limit of detection of the sensor array was found to be 1.52 copies/μL. Samples of E. coli with cell densities ranging from a single cell to 107 cells were also used to evaluate the sensitivity of the sensor array. Table S1 summarizes the array response to samples with cell densities ranging from 1 to 107 cells. The sensor array was found to be sensitive to bacteria up to a single cell, confirming the sensitivity of the developed system. It is worth mentioning that the current study proposes the development of a qualitative sensor array with a YES/NO answer, and it is not meant for the quantitative detection of the concentration of bacteria or viruses.

The use of the LnCNPs sensor array enables the collection and screening to be carried out in approximately 15 min at the point of collection of the sample. On the other hand a single patient test using RT-PCR involves a laboratory transport time plus more than 90 min and in some cases days in the laboratory for a full diagnosis. The availability of a rapid, inexpensive, and easy-to-use sensor array system for the identification of SARS-CoV-2 in still waterbodies could increase and prioritize access to testing in many situations and communities, easing the burden on the already stressed healthcare system.

Once the LnCNP sensor performance in detecting SARS-CoV-2 has been established, the sensor performance in differentiating SARS-CoV-2 from other viruses or bacteria (pathogenic or nonpathogenic) was evaluated. We utilized a microplate reader, an easy-to-use technique with a high throughput, to record the fluorescence intensities of LnCNPs at 460 ± 40 nm with an excitation of 360 ± 40 nm after the deposition of viral and bacterial samples. In the initial sensing study, a sample containing different concentrations of SARS-CoV-2 virus with a volume of 100 μL was deposited on the surface of the sensor array platform, whereas the fluorescence of the LnCNP sensor platform alone in the absence of any pathogen incubation was measured as a control. The relative fluorescence intensities of each LnCNP before and after adding the viral or bacterial sample, (I – I0)/I0, were used to characterize the fluorescence signature for each LnCNP against the pathogenic and nonpathogenic species of both bacteria and viruses. In addition to recording the signal output for SARS-CoV-2, the responses for two other viruses (SARS-CoV and H1N1) together with three other bacterial species were also recorded. The viral load of SARS-CoV-2 was determined using RT-PCR, whereas the bacterial cell densities were estimated using the turbidity method, a widely used measurement to quantify the cell number in a culture of growing bacteria.[87,88] As shown in Figure a, the three LnCNPs showed diverse fluorescence response signals against the samples, which presumably indicated different multivalent interactions between LnCNPs and bacteria or viruses (Figure b).

This result clearly illustrates the plausible utilization of these LnCNPs to identify the community spread of SARS-CoV-2 in still waterbodies and separate it from other similar viruses like SARS-CoV, H1N1, and other microorganisms. According to the fluorescence response patterns, the sensor array consists of three groups based on the variation of the dopant metals (Figure S2b). The color depth of the heat map was used to describe the relative fluorescence intensity. With the increase of the color depth and color change from blue to red, the relative fluorescence intensity increases. The 2D plot of the sensor response against three bacterial strains is shown in Figure S3. The autofluorescence from the bare agarose gel and the bacteria was found to be minimal compared to the LnCNP sensor response as shown in Figure S1. Once the distinct optical signature of the sensor array against different viral and bacterial strains has been established, the ability of the LnCNP sensor to separate SARS-CoV-2 from other pathogens was explored.

Distinct fluorescence patterns were found for each CNP for each sample, which motivates us to further explore the possibility to separate these responses using a computational algorithm to selectively detect the presence of SARS-CoV-2. The fluorescence response patterns of the viral and bacterial strains produced by the sensor array were analyzed using the LDA, a powerful pattern recognition algorithm.[70,89] The fluorescence pattern of these viruses and bacteria could then be transformed into a 3D canonical score plot by LDA (Figure c). LDA is an algorithm that transfers the data into a linear combination of variables or canonical functions to best explain the data by maximizing the component axis separation. The data set observation can be classified using canonical functions assigned to discrimination percentage as shown in the canonical score plot (Figure c). The percentage assigned to each canonical function indicates the total discriminatory power of the data set variables associated within each canonical function. The cumulative percentage of all canonical functions sums to 100%, where the higher percentage is associated with the maximum differentiation among an axis. The six strains, three viruses including SARS-CoV-2 and three bacteria, were well-clustered into six groups and thus can rapidly be discriminated from each other. The six classes of the selected pathogens were differentiated from each other with 95.6% accuracy, whereas SARS-CoV-2 was separated from other pathogens with 100% accuracy. By applying the LDA algorithm, we successfully grouped the bacteria and viruses into the group they belong. In Figure c, we choose the 3D view that shows the maximum differentiation. However, this does not affect the results of the LDA algorithm, which we demonstrated using statistical measures (i.e., classification accuracy). Moreover, the discrimination was achieved and confirmed by “leave-one-out” cross-validation using LDA, demonstrating that our LnCNP-based sensor array is efficient for viral and bacterial identification. We challenged the model to discriminate the samples containing SARS-CoV-2 from SARS-CoV, which have been classified and separated from each other with almost 100% accuracy. Furthermore, the model was challenged to discriminate the bacterial samples (pathogenic and nonpathogenic) from each other, which were also classified with nearly 100% accuracy. To further confirm the ability of our sensor array to predict unknown samples, we randomly selected 10 samples containing either viruses or bacteria from the six targeted species. Fluorescence signatures of these 10 samples were recorded using the sensor array and transformed to the canonical scores by the discriminant functions established from the training of the samples as shown in Figure c. In this algorithm, the Mahalanobis distances of each pathogen sample to their respective centroids of six groups were computed in a 3D space (canonical factors 1–3). The shortest Mahalanobis distance value determines the spatial arrangement of the tested sample. In this way, the 10 unknown samples were completely identified with high accuracy, demonstrating the high reliability of our sensor array. Spiked samples with known concentrations were used routinely by researchers to validate the sensor performance.[90,91] These samples are considered blind samples because the concentration of these samples was not known prior to the classification during the testing process and they were only used to evaluate the sensor's performance after obtaining the results.

Additionally, the differential response of the LnCNP sensor array toward the microbial strains made us interested to investigate the mode of interaction of LnCNPs with the macromolecular systems in question. It was hypothesized that LnCNPs will primarily interact with the surface-exposed proteins of the systems to generate differential sensing patterns.[92−96] Accordingly, we considered a transmembrane β-barrel protein and ferric citrate transporter FecA for E. coli (PDB ID: 1PO3); a transport protein and a triscatecholate siderophore binding protein FeuA for B. subtilis (PDB ID: 2WHY); a cell adhesion protein and a C-terminal domain surface protein SpaP for S. mutans (PDB ID: 3OPU); a hemagglutinin structure of an avian H1N1 influenza A virus (PDB ID: 3HTO); and a SARS-CoV spike glycoprotein (PDB ID: 5XLR) and a SARS-CoV-2 spike glycoprotein (PDB ID: 6VXX) (Figure S5).

On the other hand, the model structures of the LnCNPs were energy-minimized using the B3LYP/3-21G* method (Figure S6). These energy-minimized LnCNPs were then docked against the surface-exposed proteins of the studied viruses and bacteria and the results were closely monitored. Table S2 shows the comparative free energies of binding and clustering efficiencies of LnCNPs with the surface proteins.

It was observed that while GdCNP interacted more strongly with E. coli, B. subtilis, and SARS-CoV-2 surface proteins, PrCNP interacted more preferentially with S. mutans, H1N1 influenza A, and SARS-CoV surface proteins.

This differential and superior binding of GdCNP and PrCNP over YbCNP with the viral and bacterial proteins was also corroborated from the experimental fluorescence assay (Figure a). Thus, the theoretical docking results were found to have a one-to-one correspondence with the experimental sensing results. Further from the theoretical calculations, it was understood that LnCNPs might detect not only active pathogens but also trace of the target pathogens if the pathogenic protein remained in the sample even after inactivation. However, this needs further experimental investigation, where the interaction of LnCNPs even with the trace amount of pathogenic proteins could establish the higher sensitive and beneficial structure–activity relationship of the lanthanide center in LnCNPs.[97,98] Binding energies of the different conformations of the bacterial protein–ligand complexes are presented in Figure S7 and those of viral protein–ligand complexes are shown in Figure S8. The most stable docked geometries of these protein–ligand complexes are shown in Figure where the interacting amino acids and functional groups from the LnCNPs are listed in Table S3 for each of the complexes. It is to be noted that the current theoretical model might not be a complete representation of the studied system. The complete model would be lanthanide coordinated with hydroxyl groups tethered on ovalene and surface-coated with PEI. Simple docking studies cannot consider the PEI-coated NP where its inner core will be made up of an aromatic carbon skeleton coordinated to lanthanides and predict the differential interaction of LnCNPs with different types of pathogens. This type of elaborative model would be computationally expensive and can only be handled by molecular dynamics simulation studies. We would therefore consider performing the molecular dynamics simulation on these LnCNPs in a separate study.

Figure 5

Pictorial representation of the most stable docked geometries of (a) 1PO3 (E. coli)/GdCNP; (b) 2WHY (B. subtilis)/GdCNP; (c) 3OPU (S. mutans)/PrCNP; (d) 3HTO (H1N1 influenza A)/PrCNP; (e) 5XLR (SARS-CoV)/PrCNP, and (f) 6VXX (SARS-CoV-2)/GdCNP protein–ligand complexes. The white ball indicates Gd and the red ball indicates Pr. The dotted line indicates the hydrogen bonding interaction between the ligand and the protein.

Regarding the cross-reactivity of the LnCNP-based sensor, the sensor showed a negligible response when tested with a bovine serum albumin (BSA) protein (Figure S9). These results confirmed that the LnCNP sensor has negligible cross-reactivity toward proteins originating from nonpathogenic species. To further investigate the effect of the sample media on the sensor response (e.g., saliva, serum, and urine), S. mutans were spiked in both artificial medical-grade saliva and phosphate-buffered saline (PBS). We found that the fluorescence from the sensor array elements, that is, GdCNPs, PrCNPs, and YbCNPs, as a response to S. mutans exhibited no statistically significant difference (P > 0.05) between PBS and saliva samples (Figure S10). These data confirmed that the source of the distinct fluorescence readout is mainly the interaction of the bacteria or viruses with the CNPs, with no significant effect of the sample medium on the sensor array response. Furthermore, the data used for detection are the fluorescence signal normalized to the control, that is, in the absence of virus or bacteria, which further reduces the effect of the different media on the sensor output. Figure a shows a photograph image of the developed LnCNP-based sensor array in 96-well plates. The associated fluorescence map of each well at λEx = 360 nm and λEm = 460 nm is shown in Figure S11.

Figure 6

(a) Picture of the 96-well plate LnCNP sensor array. (b) 2D space plot of the first two canonical LDA factors. The LDA output of the SARS-CoV-2-spiked water was entirely distinguished from the other virus and bacterial species using the LnCNP-based sensor platform for real-world sample detection. Wastewater samples were tested without any further modifications and after spiking each sample with SARS-CoV-2, H1N1 (viral concentration of 105 genome equivalents per ml), E. coli, orB. subtilis (OD600 = 0.2, i.e., ∼107 bacterial cells).

Wastewater monitoring platforms can provide an effective method to predict the potential spread of SARS-CoV-2 across a community. This approach has been approved as an effective way to trace illicit drugs and obtain information on health, disease, and pathogens. In addition, numerous studies have shown that SARS-CoV-2 can normally survive in an appropriate environment for up to a few days after leaving the human body. Due to the limited availability of rapid diagnostic tests along with the inaccessibility of modern equipment to all the pathology centers, especially the remote ones, there is a potential threat of underdetermination of positive COVID-19 cases, and hence, the community wastewater could be tested for SARS-CoV-2 virus. In some cases, these wastewaters are outsourced to still waterbodies like lake water. Thus, we propose that a community spread of SARS-CoV-2 can potentially be mitigated by the efficient utilization of our sensor array. This could be accomplished by monitoring sewage pipe networks and by assessing the possible carriers of SARS-CoV-2 in some local areas and limiting the movements of that local population to reduce the spread of pathogens and the risk to public health. The developed LnCNP sensor array was successful in differentiating SARS-CoV-2 spiked in wastewater samples from various bacterium and virus species including B. subtilis, E. coli, S. mutans, and Influenza A H1N1. Figure b shows the 2D canonical plot applying the LDA analysis on the fluorescence data collected from testing SARS-CoV-2 and several bacterium and virus species spiked in wastewater samples.

The fast response time of the LnCNP-based sensor (∼15 min) outperforms the response time of the previously published approaches, which need ∼2–6 h for pathogen identification.[99−101] The total assay time including sample pretreatment and preparation is less than 20 min. Furthermore, the LnCNP sensor operates in a label-free manner by exploiting the inherent sensitivity of the CNP optical properties toward their multivalent interaction. And so, we could build an affordable and cost-effective system, leading to reducing the system complexity compared with previous microorganism sensors that need ligand conjugation such as an antibody or an aptamer to guarantee selectivity.[102−107] In comparison with the whole-cell enzyme-linked immunosorbent assay (ELISA), the LnCNP-based biosensing method provides advantages of selectivity at high sensitivity and requires no antibody in the assay. To the best of our knowledge, this is the first study to illustrate the approach of LnCNPs that can selectively detect SARS-CoV-2 and discriminate them from other types of pathogens and nonpathogens by the assistance of the pattern recognition algorithm. Taking it all together, we proposed a label-free system for selective detection of viruses and bacteria even with high sensitivity and relatively a small sample volume (100 μL). In addition, the sensor showed low cytotoxicity and long stability (>30 days). The fluorescence of the sensor array was found to be stable up to 1 month when stored at room temperature away from light. Figure S12 shows the relative change in the fluorescence signal after a month post the date of the sensor fabrication. A relative change in the fluorescence signal of less than 25% was observed after a month of fabricating the sensor array.

Furthermore, combining the power of pattern recognition tools and nanotechnology, different viral types including COVID-19 causative virus, SARS-CoV-2, and bacterial strains were discriminated with high accuracy, thus enabling the sensor to provide fast and reliable pathogen information for clinical decisions and allow continuous monitoring of infectious disease trends. The performance characteristics of the developed sensor array as a single system are superior in comparison with previously reported biosensors as it selectively discriminates pathogens and nonpathogens and accurately predicts the bacterial staining properties (Gram-positive or Gram-negative) as detailed in Table S4.

Environmental monitoring of the presence of virus in wastewater maybe an early warning method and an effective method for quick diagnosis of clusters of COVID-19 cases irrespective of symptoms. The importance behind the water-based surveillance of SARS-CoV-2 has thus been realized, but the rapid and early discrimination of SARS-CoV-2 from other viruses and microorganisms using the available techniques remains the major obstacle. Recently, several techniques such as targeted mass spectrometry assay, multiplexed amplicon-based sequencing, and RT-PCR have been deployed for detection of SARS-CoV-2 in wastewater.[108,109] However, the practicality and measurement duration of these techniques for point-of-care detection of SARS-CoV-2 in wastewater systems still remain a challenge. In this study, we developed a biosensor that allows rapid testing to detect the extent of SARS-CoV-2 presence in wastewater systems as well as distinguish SARS-CoV-2 from other microbial species in real-time, which can be potentially transformative and supplement the clinical efforts on containing the virus. Bacteriophages are a type of virus that infects bacteria, and they are more relevant in effluents and sludge coming out of the wastewater treatment plants.[110−112] The current focus of the study is to test the presence of pathogenic viruses and bacteria in the wastewater. The sensitivity of LnCNPs with effluents and sludge collected from wastewater treatment plants will be evaluated in a separate fully executed study. We intend to study the presence of MS2 and F-specific RNA (FRNA) phages as biocontrol agents in the wastewater treatment plants and monitor their effect on the sensitivity of LnCNPs toward pathogenic microorganisms. Lanthanide (Ln+3)-doped NP-based sensor arrays have been utilized herein for their luminescent properties, excellent photostability, narrow emission spectrum, lower toxicity as well as cost-effectiveness, higher resistance to photobleaching, ease of preparation, and functionalization needs. To the best of our knowledge, the adoption of Ln+3-doped CNPs for the creation of a sensor array and also their application in label-free sensing of SARS-CoV-2 and other viruses and bacteria in complex samples have not yet been explored. The LnCNP array developed here showed excellent sensitivity (1.5 copies/μL) and a short sample-to-results time (15 min). The sensor performance was evaluated using wastewater sewage samples to confirm its effectiveness in detecting SARS-CoV-2 and differentiating it from other viruses or bacteria. Interestingly, the proposed sensor’s capability is not just limited to the environmental application, and it can be further expanded to cover the clinical diagnosis of individuals without requiring RNA-extraction or nucleic acid amplification.[37−39,113]

Conclusions

We have successfully constructed a fluorescence sensor array made from three CNPs, where each one is doped with one lanthanide metal, that is, either Pr, Yb, or Gd. LnCNPs share the same organic source but are different in metal composition.

Our results indicated excellent selectivity of the LnCNP-based sensor array toward SARS-CoV-2 and various other viral and bacterial strains. By examining the fluorescence patterns of SARS-CoV-2, SARS-CoV, H1N1, E. coli, S. mutans, and B. subtilis using LDA, SARS-CoV-2 was separated from the six different strains with 100% accuracy and the six strains were identified effectively with nearly 95.6% accuracy. Moreover, the detection procedure is rapid (∼15 min) with high throughput and does not need any external requirements for washing. This method could also provide real-time and uninterrupted data to function as an early warning system to enable municipal communities and organizations to take appropriate steps to identify suspected virus carriers and mitigate the spread of an outbreak.