key: cord-0775105-gs4zfk8m
authors: Moitra, Parikshit; Chaichi, Ardalan; Abid Hasan, Syed Mohammad; Dighe, Ketan; Alafeef, Maha; Prasad, Alisha; Gartia, Manas Ranjan; Pan, Dipanjan
title: Probing the mutation independent interaction of DNA probes with SARS-CoV-2 variants through a combination of surface-enhanced Raman scattering and machine learning
date: 2022-03-22
journal: Biosens Bioelectron
DOI: 10.1016/j.bios.2022.114200
sha: ba1c0927ba43faece526bc802d1595686eea1a8b
doc_id: 775105
cord_uid: gs4zfk8m

Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) evolution has been characterized by the emergence of sets of mutations impacting the virus characteristics, such as transmissibility and antigenicity, presumably in response to the changing immune profile of the human population. The presence of mutations in the SARS-CoV-2 virus can potentially impact therapeutic and diagnostic test performances. We design and develop here a unique set of DNA probes i.e., antisense oligonucleotides (ASOs) which can interact with genetic sequences of the virus irrespective of its ongoing mutations. The probes, developed herein, target a specific segment of the nucleocapsid phosphoprotein (N) gene of SARS-CoV-2 with high binding efficiency which do not mutate among the known variants. Further probing into the interaction profile of the ASOs reveals that the ASO-RNA hybridization remains unaltered even for a hypothetical single point mutation at the target RNA site and diminished only in case of the hypothetical double or triple point mutations. The mechanism of interaction among the ASOs and SARS-CoV-2 RNA is then explored with a combination of surface-enhanced Raman scattering (SERS) and machine learning techniques. It has been observed that the technique, described herein, could efficiently discriminate between clinically positive and negative samples with ∼100% sensitivity and ∼90% specificity up to 63 copies/mL of SARS-CoV-2 RNA concentration. Thus, this study establishes N gene targeted ASOs as the fundamental machinery to efficiently detect all the current SARS-CoV-2 variants regardless of their mutations.

sensitivity and ~90% specificity up to 63 copies/mL of SARS-CoV-2 RNA concentration.

Thus, this study establishes N gene targeted ASOs as the fundamental machinery to efficiently detect all the current SARS-CoV-2 variants regardless of their mutations.

The decline in coronavirus disease -2019 cases in much of the world in late spring and early summer signaled a new phase in the fight against the disease (Telenti et al., 2021) . This progress was made possible by the rapid deployment of vaccines (Mallapaty et al., 2021) . However, due to emerging new variants such as Beta, Delta, Lambda, and most recently Omicron, as well as the vaccine reluctance, nations across the globe have witnessed an alarming increase in COVID-19 cases in recent months Karim and Karim, 2021) . As a result, it continues to spread rapidly throughout the world, causing unprecedented disruption in modern society. While vaccines continue to be extremely effective at preventing serious illness caused by COVID-19, new data from the United Kingdom (U.K.), Israel, and the U.S. have raised new concerns about their ability to prevent infection from the delta as well as more transmissible and potentially concerning Omicron variant (Janik et al., 2021; Mercatelli and Giorgi, 2020; L. Wang et al., 2021) . Furthermore, while the Omicron variant fuels a fresh wave of infections, decline in extensive testing and monitoring together with the unavailability of an effective diagnostic test to selectively detect these emerging variants without any false negativity, makes it impossible to determine how much the virus is circulating in communities and who remains most vulnerable (Daria et al., 2021; Mohiuddin and Kasahara, 2022) .

These SARS-CoV-2 genetic variants represent permanent changes in the DNA sequence and may affect three or more nucleotides in a gene (Pachetti et al., 2020) .

These alterations may result into failure of diagnostics, reduction in vaccine effectiveness, low vaccine-induced protection against severe disease, more severe clinical disease, and significantly diminished susceptibility to diagnostics or approved therapeutics (Barton et al., 2021; Dighe et al., 2021; Liu et al., 2021; . As new variants continue to emerge, a deeper understanding of the phenotypes of these variants in terms of infectivity, transmissibility, virulence, and antigenicity must be gained (Copin et al., 2021; Liu et al., 2021) . As a result, there is still an unmet requirement for routine largescale comprehensive testing to prevent COVID-19 spread and provide safe environments for socio-economic activities. And we believe that this can only be done by developing a probe that will interact with all the variants of SARS-CoV-2 independent of their mutations. Unpredicted combinations of mutations will continue to emerge, such perceptions will allow predictions of virus phenotype (Jungreis et al., 2021; Petrova and Russell, 2018) and in turn, help to develop newer probes for mutation independent interaction with SARS-CoV-2 genes.

In this regard, here we have designed and developed several sets of novel DNA Probes, i.e., antisense oligonucleotides, targeted towards nucleocapsid phosphoprotein (N), envelope (E) and RNA-dependent RNA polymerase (RdRp) gene segments of SARS-CoV-2. We hypothesized that a combination of computational and vibrational spectroscopic techniques along with machine learning (ML) algorithm can be applied to study the mutation-independent interaction of these DNA probes with SARS-CoV-2 RNA.

Following the successful design and synthesis, these ASOs were connected to gold J o u r n a l P r e -p r o o f nanoparticles (AuNPs) and their target binding energies and agglomeration patterns were then investigated in presence of SARS-CoV-2 RNA. We found that N-gene targeted ASOs were the most efficient ones in selectively hybridizing SARS-CoV-2 target RNA with optimum efficiency. The target RNA sites for N-gene ASOs were then examined against the current SARS-CoV-2 variants where it was found that these ASOs should interact with all the known mutated forms of SARS-CoV-2 in a similar manner without any loss in sensitivity and specificity. The mechanism of interaction between these N genetargeted ASOs and SARS-CoV-2 RNA was then evaluated with surface-enhanced Raman scattering (SERS) and machine learning (ML) techniques (Abid Hasan et al., 2019; Chaichi et al., 2021; Chang et al., 2015; Gartia et al., 2010; Prasad et al., 2020; Xu et al., 2012 Xu et al., , 2011 . This combinatorial effort discriminated SARS-CoV-2 positive and negative clinical nasal swab samples with ~100% sensitivity and ~90% specificity with an analytical detection limit of 63 copies/mL. However, the characterization of different SARS-CoV-2 N-gene mutated forms will continue to evolve and provide extremely useful information on specific mutations or their combinations those may not have been identified yet in circulating viruses . Towards this goal, we computationally evaluated next hypothetical single, double, and triple point mutations on the RNA site. Further close monitoring of the ASO-RNA hybridization profile revealed stable interaction between the two, even after a hypothetical single point mutation was introduced at the target RNA site. Thus, this study examined the efficiency of universal DNA probes and their fundamental interaction with the entire known mutated forms of SARS-CoV-2.

J o u r n a l P r e -p r o o f There are many attempts to extend or replace nucleic acid-based methods to detect SARS-CoV-2 virus or differentiate its variants (Hu et al., 2021; Kevadiya et al., 2021a) .

However, numerous challenges still remain with these techniques, such as, extended operation time, requirement of laboratory-based hospital with the accessibility of large stationary equipment and reagents (Carter et al., 2020a; Vandenberg et al., 2021) . The major bottleneck that lies with the widely used qPCR technique is the requirement of completely different set of primers and multiple other reagents for the detection of the emerging variants of SARS-CoV-2 ( Barreto et al., 2020; Esbin et al., 2020; Vega-Magaña et al., 2021) . But the current study relies on the change in Raman spectral signature of the designed N-ASO capped AuNP based probe. This technique does not require any nucleic acid purification or amplification steps to detect its target RNA, rather it involves unique antisense oligonucleotides that can bind to SARS-CoV-2 RNA with similar efficiency irrespective of its current mutations. We showed that the same two N gene targeted ASOs can successfully be used for a POC, rapid, cost-effective, and selective diagnosis of COVID-19. Further, because of the dependency of the studied technology on Raman spectra, high-throughput screening is quite possible in limited duration. This assay is also found to be quite sensitive (LOD = 63 copies/mL) and better than most of the currently known qPCR-based techniques (Afzal, 2020) . Thus, the current study differentiates its signature from the literature known nucleic acid-based studies. We envisage that this study will lead to the real-time multispectral imaging and rapid optical processing of clinical samples towards the selective and sensitive diagnosis of COVID-19 variants.

J o u r n a l P r e -p r o o f 

We designed and developed novel ASOs targeted towards different genetic segments of SARS-CoV-2, e.g., N, E and RdRp gene for the selective and sensitive binding of RNA from the wild type of SARS-CoV-2. More importantly, herein, we investigated the mechanism of interaction of selected ASOs with the various mutated forms of SARS-CoV-2 and studied the ASO-RNA hybridization phenomena through a unique combination of SERS and ML-based techniques.

Accordingly, particular gene sequences (RdRp: 13, 236; E: 26, 472 and N: 28, 533 ) from the whole genome sequence of SARS-CoV-2 (wild type isolate SARS-CoV-2/human/USA/WA-CDC-WA1-A12/2020, MT020880) was chosen and multiple ASO sequences of 20 nucleotides in length were developed (Fig. 1a) . The choice of ASOs were primarily based on the optimum GC content and theoretically calculated target binding and disruption energies at 37 °C in 1 M NaCl aqueous solution. We propose to select four ASOs, two for the front (ASO1 and ASO2), and two targeting the end region (ASO3 and ASO4) of the gene. The ASOs will also target closely following sequences at each location Dighe et al., 2021; Moitra et al., 2021b Moitra et al., , 2021a Moitra et al., , 2020 . Accordingly, four ASOs were selected both for N and E genes, whereas only two were selected for the RdRp gene as ASOs with high enough binding energies are not available for this target J o u r n a l P r e -p r o o f (Table 1) . Soligo software were used to predict the ASOs (Ding et al., 2004) . The detailed methodology is discussed in the method section of supporting information. It was noticed that the mean target binding energy was the highest for N gene-targeted ASOs (N-ASO1-4) and decreased gradually for E gene-targeted ASOs (E-ASO1-4) and was the minimum for RdRp targeted ones (O-ASO1-2). Based on this theoretical observation, it can also be presumed that the effective ASO-RNA hybridization will follow the same trend and will be the highest for N-ASO1 and N-ASO2. We anticipated that once these ASOs were connected to a plasmonic nanoparticle, such as AuNPs, the agglomeration among ASO capped AuNPs would exhibit a strong SERS response (Khan et al., 2018; Misra et al., 2018; Moitra et al., 2020; Pan et al., 2011 Pan et al., , 2010 and should also exhibit similar phenomena where a mixture of N-ASO1 and N-ASO2 may be the optimum one (Fig. 1a) . To prove this theoretical assumption, we differentially modified the ASOs, one at the 5' end and the other at the 3' end by thiol groups ( Table 1) . The thiol modified ASOs were then used to cap the citrate stabilized AuNP Zhu et al., 2021) . The agglomeration patterns of these ASOs conjugated to AuNPs (Au-ASO NPs) were then investigated in presence of RNA extracted from clinically positive and negative SARS-CoV-2 samples. We employed UV-Visible absorbance spectroscopy and dynamic light scattering (DLS) to initially investigate the aggregation phenomena. SARS-CoV-2 positive nasal swab samples (N=10) having varying cycle threshold (Ct) numbers from 13 to 28 (Table S1 ) were considered and RNA was extracted from these samples by standard protocol (Rump et al., 2010; Wozniak et al., 2020) . Representative two samples, one having low Ct number (P1), i.e., high viral copy number and the other having high Ct value (P2), i.e., low viral copy number, was considered for this study. It was observed that the absorbance of plasmonic Au-ASO NPs at 520 nm increased with broadening of full-width half maximum (FWHM) when added with SARS-CoV-2 positive RNA. P1 having a high viral copy number increased the absorbance more than P2 which had a low viral copy number. An insignificant increase in absorbance with no change in FWHM was observed when Au-ASO NPs were added with SARS-CoV-2 negative RNAs. This change in surface plasmon band of AuNPs indicated the successful aggregation of Au-ASO NPs only in presence of their target RNAs. Ostadhossein et al., 2020; Rump et al., 2010; Srivastava et al., 2020) It was also observed that the mixture of N-ASO1 and N-ASO2 gave a better aggregation response than the mixture having N-ASO3 and N-ASO4 ( Fig.   2a) . Similar is the case for E-ASO1 and E-ASO2 compared to the combination of E-ASO3 J o u r n a l P r e -p r o o f and E-ASO4 (Fig. 2b) . Further, the change in absorbance was found to be the highest for N-ASO1 and N-ASO2, followed by E-ASO1 and E-ASO2 and the minimum for O-ASO1 and O-ASO2 (Fig. S1 ). This improved agglomeration of N-ASO1 and N-ASO2 mixture in presence of SARS-CoV-2 RNA was further supported by DLS experiments.

The average hydrodynamic diameter was found to be the highest for N-ASO1 and N-ASO2 mixture in presence of RNA extracted from P1 compared to the other mixture of Au-ASO NPs (Fig. 2d) . Thus, these experiments corroborated with our theoretical observation supporting better binding followed by improved ASO-RNA hybridization in case of N-ASO1 and N-ASO2 mixture.

ASOs. This interesting observation, where we found that N-gene targeted ASOs were more efficient than E gene and RdRp gene-targeted ASOs in hybridizing SARS-CoV-2 RNA, led to the curiosity to investigate these N-ASOs more closely. We asked the question whether these ASOs will still be effective against the current SARS-CoV-2 mutations or not. We realized that the viability of the assay can be improved significantly if a mostly conserved region of N-gene sequence can be targeted by the ASO probes.

Accordingly, the design of N-ASOs was revisited and the target RNA sites for these ASOs were examined against the current SARS-CoV-2 variants. Recently, the genetic surveillance of SARS-CoV-2 strains circulating around the world has revealed several variants with one or more mutations that may affect detection by nucleic acid-based testing methods (Carter et al., 2020b; Yu et al., 2021) .

It was observed that the currently known mutations associated with SARS-CoV-2 variants mostly occur at 606-617, 702-704 and 1131-1133 of N-gene sequence ( Table   J o CoV-2, will start to agglomerate because of their proximity in the gene sequences Moitra et al., 2020) . This will cause a change in surface plasmon resonance and introduce a change in the SERS response of the AuNPs (Gitman et al., 2021; Kevadiya et al., 2021b; Shrivastav et al., 2021) . The N-ASO1+2 capped AuNPs were synthesized and analytically characterized by UV-Visible spectroscopy (Fig. S2 ) and

transmission electron microscopy (Fig. S3a) . The nanoparticles were then admixed with SARS-CoV-2 RNA and the corresponding bright field and Raman microscopic images for N-ASO1+2 capped AuNPs are shown in Fig. 3a The presented Raman spectra (Fig. 3c) is therefore an average of all the acquired Raman spectra. To clarify, we have now represented the spectrum collected from different spots of the Raman map (Fig. S4) .

The representative Raman spectra of N-ASO1+2 capped AuNPs at varying concentrations of SARS-CoV-2 RNA ranging from 6.3 × 10 2 copies/mL to 6.3 × 10 5 copies/mL (Fig. 3c ) and from 6.3 × 10 6 to 6.3 × 10 10 copies/mL (Fig. 3d ) are shown where the SERS response of the nanoparticles increased with increase in RNA concentration.

The spectral data are consecutively stacked and independently labeled for better understanding. This confirmed the enhanced agglomeration among the N-ASO1+2 capped

AuNPs in presence of increasing concentration of SARS-CoV-2 RNA. It was observed that the Raman peaks specific for oligonucleotide backbone, ribose and deoxyribose sugars were visible both at low viral RNA concentration, i.e., 63 copies/mL, and at high viral RNA concentration, i.e., 6.3 × 10 7 copies/mL (Fig. 3e) indicating efficient hybridization among the ASOs and their target RNA strands (Abid Hasan et al., 2019; Gartia et al., 2010) . To better discern the SERS spectral features of N-ASO1+2 AuNPs with low concentrations of SARS-CoV-2 RNA (63 copies/mL), the spectrum was first denoised using Savitzky-Golay method and a higher magnification of the figure was provided (Fig. S5) . Raman peaks were accordingly assigned for both concentrations and enlisted in Table S3 and S4.

With increase in SARS-CoV-2 RNA concentration, there will be increase in agglomeration among N-ASO1+2 AuNPs leading to change in surface plasmon resonance of the gold nanoparticles Dighe et al., 2021; J o u r n a l P r e -p r o o f al., 2021b, 2020). Further, when the AuNP-RNA conjugate of different concentrations was added onto the surface, the molecule will take certain orientation at the nanostructure surface. It is known that the SERS effect is highly localized, i.e., the electrical field normal to the surface, and therefore, only a signal from part of the molecule, close to the substrate, will be enhanced. Thus, when the molecule is adsorbed on the surface, its symmetry will change, and so do the selection rules which will be followed by the selective enhancement of certain vibrational modes (Pérez-Jiménez et al., 2020) . This describes the corresponding symmetry properties of the modified Raman dipole and changes in the relative intensities of the Raman peaks caused by the EM field polarization at the metal surface. Also, because the electromagnetic field enhancement is wavelength dependent, the different spectral regions of the spectrum may be enhanced differently (Kim et al., 2019) . Thus. the spatial variation of the SERS intensities can be explained by several processes: (i) differences in local absorption of light by the AuNP-ASO due to the differences in the local concentration of particles; (ii) a difference in local concentration of SARS-CoV-2 RNA molecules, and (iii) local variations of the Raman enhancement effect due to specific microenvironments (orientation of the molecule) (Zeisel et al., 1998) .

Hence, we propose that there will be change in SERS spectral signature with increase in SARS-CoV-2 RNA concentration. Further, for our analysis, we rely on the area under the peaks over a wavenumber range, small shift in the peak position will not affect the conclusion.

We have further acquired the SERS spectra for SARS-CoV-2 RNA alone without the addition of N-ASO1+2 AuNPs (Fig. S6) . This spectrum showed a 1270 cm -1 Raman peak which may be assigned to the in-plane bending of C8-H, N9-H, str N7-C8 of adenine J o u r n a l P r e -p r o o f and stretching mode of N3-C4, C4-C5, C6-N1, bending mode of N1-H, C5/6-H of uracil (Madzharova et al., 2016) . Hence, we used area under the Raman peak at 1270 cm -1 to calculate the limit of detection. The limit of detection for N-ASO1+2 capped AuNPs towards SARS-CoV-2 RNA was also evaluated and found to be 63 copies/mL which is quite significant and highly comparable to the currently available RT-PCR techniques (Fig. 3f) . 

RNA. In presence of SARS-CoV-2 RNA, agglomeration among N-ASO1+2 AuNPs were observed through transmission electron microscopy ( Fig. S3a, b) . It was observed that the absorbance of AuNPs at 526 nm in absence of the target RNA shifted towards 704 nm in presence of the target SARS-CoV-2 RNA (Fig. S7a) . Different sizes of AuNPs with water solvation was then simulated and their corresponding surface plasmon resonance (SPR) band was predicted. It was found that the absorbance peak at 526 nm matched with the AuNP size of 40 nm, while the SPR band at 704 nm corresponded with AuNP size of 170 nm (Fig. S7b) . The effect of this difference in AuNP size was then evaluated for change in electric field and extinction properties. According to Mie theory, as the AuNP-size changes, the electromagnetic field of particles will change, and the multipole effect can be observed. It was observed that under 785 nm laser excitation, the electric field intensity varied with increase in sample size for a single spherical AuNP (Cheng et J o u r n a l P r e -p r o o f al., 2020). Hextapolar and quadrupolar electromagnetic field was found for the AuNPs with 170 nm and 40 nm respectively when the incident laser beam was propagated along z-direction and focused at the origin (Fig. S8a, b) . It can be proposed that increase in size of AuNP, i.e., in presence of SARS-CoV-2 RNA, made it easier to couple with the incident light (laser excitation of 785 nm). Because of this red shift towards higher wavelength, a multipolar distribution has been observed for higher size of AuNPs (Fig.   S8b ). Next, false color maps, simulated using Mieplot, showing the scattered intensity as a function of scattering angle and the wavelength of light has been shown in Fig. S8c , d

for AuNP size of 40 and 170 nm respectively (Laven, 2003) . It can be seen that the plots are quite symmetric for smaller size of AuNPs (in absence of SARS-CoV-2 RNA), while they lost their symmetry with increase in size (in presence of SARS-CoV-2 RNA).

The high sensitivity of N-ASO1+2 capped AuNPs towards the binding of SARS-CoV-2 RNA grew our confidence to study the interaction in clinical nasal swab samples either as COVID-19 positive or negative. We presumed that because of this strong and efficient hybridization among the ASOs and RNA, the binding interaction might be evident even 4a ) and negative, N1 (Fig. 4b) clinical samples were shown. The respective Raman spectra of N-ASO1+2 capped AuNPs admixed with direct SARS-CoV-2 positive (Fig. 4d) and negative (Fig. 4e) clinical samples were also shown without the extraction of RNA but with just the addition of lysis buffer. The range of all the acquired spectra are shown as shaded color for each sample while the dark color is representing the mean spectra for each of the sample. The Raman peaks were assigned for the spectra in Fig. 4a, 4b and enlisted in Table S5 . Raman peaks were also assigned for Fig. 4d , 4e and noted in Table S6 .

Now, in order to extract useful information from the Raman spectra, multivariate statistical analysis method is generally followed. Principal component analysis (PCA) is the widely applied conventional multivariate statistical analysis technique which by orthogonal transformation converts high dimensional Raman spectra into multiple unrelated independent variables, called principal components (PCs), avoiding any loss of valuable information (Crow et al., 2005; Gautam et al., 2015) . Here, most of the important information are distributed over the first few PCs and the contributions of the rest of PCs are negligible. Fig. S9 shows the contribution of eigenvalues of each PCs to the total variance of all SERS spectra for clinical samples. It was observed that the eigenvalues reduced rapidly with increasing PC numbers and only the first few PCs hold the maximum variance of the obtained Raman data. It can be seen that the first five principal components described more than 99% of the variation of the corresponding dataset.

Based on the obtained dataset, we performed the PCA and PCA score plot showing the separation between negative and positive RNA samples (Fig. 4c) and direct clinical J o u r n a l P r e -p r o o f samples (Fig. 4f) . From these figures, it can be noticed that due to possible interference from sample matrix (background signal), PCA was not able to accurately distinguish negative samples from the positive ones. Though, there were some clusters of points on two sides, but a good amount of overlap was also present. Here, each point is representing the Raman spectra of each sample. However, we calculated several numbers of principal components (PCs) for these analyses where we found that PC1

shifted to negative values for SARS-CoV-2 positive RNA samples. On the other hand, PC3 shifted to positive values for SARS-CoV-2 positive clinical samples (Fig. 4c, f) . Fig.   S10 and S11 show the loading plots for PCA of clinical samples for PC3 and PC1

respectively. Here, we highlighted the characteristic peaks of our dataset which indicated the differences in the various samples. It was noticeable that in both of these figures the location of the characteristic peaks was very similar. This indicated the variation of all the datasets was not significant enough and thus it led to the overlap in the score plot as observed in Fig. 4c, 4f. Fig. S12 shows the box plot for the distribution of spectral data for positive and negative groups of clinical samples. The spectral data of the negative samples from the dataset significantly (P < 0.001) shifted towards the negative PC3 range compared to positive samples. Although, PC1 and PC3 could be used to distinguish positive and negative RNA as well as clinical samples, respectively, we further explored the diagnostic capability of these binary classification (positive vs negative) systems. The score plot shown in Fig. 4c and 4f pointed to a less accurate classification system using the PCA scores alone.

The unsupervised PCA method has thus been used to demonstrate the outlier from the Raman dataset and extract the most useful information to be followed by the 

among the clinical samples. Here, we realized that only the strong binding affinity among the N-ASO1+2 AuNPs with their target RNA sequences obtained through SERS was not enough in selectively differentiating SARS-CoV-2 positive samples from the negative ones with high confidence value (Carlomagno et al., 2021; Huang et al., 2021; Sanchez et al., 2021; Yin et al., 2021) . Hence, we used the Support Vector Machine (SVM) algorithm to improve the predictive performance of our assay. At first, we developed a model based on the training dataset and based on that model we predicted the rest of our dataset. Fig.s 4g and 4i are showing the receiver operating characteristic (ROC) curve of our developed model based on the raw data for RNA and clinical samples respectively. However, when we applied the standard normal variate (SNV) algorithm to perform preprocessing of our dataset followed by training the model, we have observed significant improvement of our model. The performance of the algorithm can be measured based on the specificity and sensitivity of the results as tabulated in Table S7 . Overall, by J o u r n a l P r e -p r o o f applying the SNV operation, we have successfully increased the sensitivity and specificity of the probes for the clinical samples, tested directly without the extraction of RNA, from 30% to ~100% and 50% to 90%, respectively. On the other hand, for isolated RNA samples, the specificity was increased from 58% to 95% but, the sensitivity did not improve (went from 67% to 65%). The respective ROC curves for the RNA and direct clinical samples fitted after SNV operation are shown in Fig.s 4h and 4j respectively.

Thus, we have successfully discriminated SARS-CoV-2 positive and negative samples directly from the clinical nasal swabs with the addition of lysis buffer and N-ASO1+2 AuNPs through a combinatorial effort from SERS and ML-based techniques. Representative Raman microscopic images of N-ASO1+2 capped AuNPs admixed with SARS-CoV-2 positive and negative RNA (Fig. 5a, b) and direct clinical samples (Fig. 5c, d) respectively. Distribution for whole Raman spectrum is from 200-3200 cm -1 . For RNA samples, map has around 1500 pixels (that is 1500 Raman spectra) and for clinical samples, map has around 2500 pixels (that is 2500 Raman spectra). All the chemicals were procured from reputable commercial vendors and used without any further purification steps. The custom designed and thiol-modified ASOs were procured from Sigma Aldrich and stored at −20 °C until further use. All the experiments were carried out at constant room temperature of 25 °C unless otherwise specified. RdRp: 13, 236; E: 26, 472 and N: 28, 533 ) from the whole genome sequence of SARS-CoV-2 (wild type isolate SARS-CoV-2/human/USA/WA-CDC-WA1-A12/2020, MT020880) was chosen and multiple ASO sequences of 20 nucleotides in length were selected from Soligo software output (Ding et al., 2004) . The choice of ASOs were primarily based on the optimum GC content and theoretically calculated target binding and disruption energies at 37 °C in 1 M NaCl aqueous solution. We propose to select four ASOs, two for the front (ASO1 and ASO2), and two targeting the end region (ASO3 and ASO4) of the gene. The ASOs will also target closely following sequences at each location. Accordingly, four ASOs were selected both for N and E gene, whereas only two were selected for RdRp gene ( Table 1) . We were not able to select more than two ASOs J o u r n a l P r e -p r o o f for RdRp gene target as ASOs with high enough binding energies are not available for this target.

The gold nanoparticles were prepared from a solution of chloroauric acid and sodium citrate following a previously published literature protocol (Kumar et al., 2020; Moitra et al., 2020; Schwartz-Duval et al., 2020) . 4.9. Calculation of Limit of Detection. Further to calculate the limit of detection from the Raman calibration curve, N-ASO1+2 AuNPs (concentration of 2*10 11 particles/mL) were added with SARS-CoV-2 RNA samples having the concentration in the range from 1 fg/mL (63 copies/mL) to 1 µg/mL (63 x 10 9 copies/mL). Five spectra were collected for each of the samples and the average spectra were reported. Limit of detection (LoD) and the limit of blank (LoB) was then calculated from these spectra. LoB was estimated by measuring replicates of a blank sample and calculating the mean and standard deviation (SD). For each concentration, as mentioned before, five data points were collected. The

LoB was calculated as follows (Armbruster and Pry, 2008) : LoB = mean blank + 1.645*(SD blank)

Assuming a Gaussian distribution of the raw analytical signals from blank samples, the LoB represents 95% of the observed values. Thus, with mean intensity of 325 from blank sample (no molecule present) and SD of 142, the LoB was calculated to be 558.6.

On the other hand, LoD is the lowest analyte concentration likely to be reliably distinguished from the LoB and at which detection is feasible. It is therefore greater than LoB. It is generally calculated using the following equation (Armbruster and Pry, 2008) :

The lowest concentration used was 1 fg/mL and the SD of the 1 fg/mL sample was found to be 212. Hence, the sample at LoD must have intensity > or = 907.3. From our experiment, we found that the lowest concentration whose corresponding Raman intensity > 907.3, was 1 fg/mL (or 63 copies/mL). Hence, the LoD of the current sensor was noted as 63 copies/mL.

The extinction spectra were simulated using nanoHUB (Juluri et al., 2016) . The electromagnetic field were calculated using nanoDDSCAT program in nanoHUB (Jain et al., 2019) . The dispersion curves (wavelength vs Scattering angle) were generated using MiePlot program 3.4

(http://www.philiplaven.com/mieplot.htm). previously. This multivariate analysis was completed with the "Principal Component Analysis for Spectroscopy" toolbox of Origin Lab 2020 (Origin Lab, Northampton, MA)).

The spectral differences were distinguished with Principal components (PC). The PCs were calculated based on the covariance matrix and 8 components were extracted. These parameters were set before performing the analysis and based on the variation of the dataset we received several PCs. Though, mean centering is a common step for calculating principal components, but we didn't have to perform it separately as while computing the covariance matrix the software performed mean centering implicitly. To visualize the results, we plotted score plot where each point is representing a Raman spectrum separately while selecting two or more PCs. The clustering of each Raman spectra from the score plot shows the separation of the data. Here, we used 95% confidence level to show the clustering of the Raman spectra. In order to find the vibrational fingerprint, we plotted the loading plot calculated from PCA. From the loading plot, we also found the peak locations which dominated the PCA. Besides performing these analyses on raw data, we also used min-max normalization through Origin where the normalization was performed in [0, 1] range. Later, a smoothing operation was also performed to reduce the noise of the dataset with the help of Savitzky-Golay method.

During this analysis, window size was selected as 5 and polynomial order was set to 2.

Based on these posts processed dataset, PCA was performed in a similar fashion.

Besides performing PCA, we also developed an algorithm based on Support vector machine (SVM) analysis.

This supervised analysis helps to classify groups. Here based on the Raman spectra we J o u r n a l P r e -p r o o f tried to separate our dataset between positive and negative group of samples. Support vector machine classification model by MATLAB was used in our method. Here at first, we trained a model and then predict rest of our dataset Based on the classification results we calculated the specificity and sensitivity of our approach. Later, we used similar approach on a selected range of dataset. Here, the selected range of the wavelength was 400-1800 nm. The reasoning for selecting this range was, we noticed most of the peaks of the dataset located within this range and thus contribution for separation was much higher. Besides this selected range, we performed the Standard Normal Variate (SNV) analysis to improve the results we gathered from previous stage. To get the SNV results we calculated mean and standard deviation of each of the dataset. The mean value was subtracted from the raw spectra and then it was divided by the standard deviation of that particular dataset. These operations were performed through MATLAB and each dataset was considered as a matrix while performing the calculation.

The single stranded models for N-ASO1 and RNA strands (nonmutated and mutated ones) were made in Web 3DNA 2.0 software (Li et al., 2019) and then docked on a HNADOCK server (He et al., 2019) to obtain the docking score and conformationally stable models.

Supporting Information. Figures S1-S12, Tables S1-S7 and materials and methods section have been provided in the supporting information.

J o u r n a l P r e -p r o o f

The manuscript was written through the contributions of all authors. All authors have approved the final version of the manuscript.

The authors declare the following competing financial interest(s): Prof Pan is the founder or co-founder of three university-based start-ups. None of these entities, however, supported this work. All the other authors declare no competing financial interest.

Rapid, Ultrasensitive, and Quantitative Detection of SARS-CoV-2 Using Antisense Oligonucleotides Directed Electrochemical Biosensor Chip

Hyperspectral Mapping for the Detection of SARS-CoV-2 Using Nanomolecular Probes with Yoctomole Sensitivity

RNA-extraction-free nano-amplified colorimetric test for point-of-care clinical diagnosis of COVID-19

Limit of blank, limit of detection and limit of quantitation

Diagnosing the novel SARS-CoV-2 by quantitative RT-PCR: variations and opportunities

Effects of common mutations in the SARS-CoV-2 Spike RBD and its ligand, the human ACE2 receptor on binding affinity and kinetics

How bad is Omicron? What scientists know so far

COVID-19 salivary Raman fingerprint: innovative approach for the detection of current and past SARS-CoV-2 infections

Assay Techniques and Test Development for COVID-19 Diagnosis

Assay Techniques and Test Development for COVID-19 Diagnosis

Label-free lipidome study of paraventricular thalamic J o u r n a l P r e -p r o o f nucleus (PVT) of rat brain with post-traumatic stress injury by Raman imaging

Bifunctional Nano Lycurgus Cup Array Plasmonic Sensor for Colorimetric Sensing and Surface-Enhanced Raman Spectroscopy

FDTD simulation of the optical properties for gold nanoparticles

The monoclonal antibody combination REGEN-COV protects against SARS-CoV-2 mutational escape in preclinical and human studies

The use of Raman spectroscopy to differentiate between different prostatic adenocarcinoma cell lines

Detection of highly muted coronavirus variant Omicron (B.1.1.529) is triggering the alarm for South Asian countries: Associated risk factors and preventive actions

A rapid RNA extractionfree lateral flow assay for molecular point-of-care detection of SARS-CoV-2 augmented by chemical probes

Sfold web server for statistical folding and rational design of nucleic acids

Overcoming the bottleneck to widespread testing: a rapid review of nucleic acid testing approaches for COVID-19 detection

Rigorous surface enhanced Raman spectral characterization of large-area high-uniformity silver-coated tapered silica nanopillar arrays

Review of multidimensional data processing approaches for Raman and infrared spectroscopy

Laboratory Diagnosis of SARS-CoV-2 Pneumonia

SARS-CoV-2 variants, spike mutations and immune escape

HNADOCK: a nucleic acid docking server for modeling RNA/DNA-RNA/DNA 3D complex structures

Characteristics of SARS-CoV-2 and COVID-19

On-Site Detection of SARS-CoV-2 Antigen by Deep Learning-Based Surface-Enhanced Raman Spectroscopy and Its Biochemical Foundations

The Emerging Concern and Interest SARS-CoV-2 Variants

Extinction, Scattering and Absorption Efficiencies of Single and Multilayer Nanoparticles

SARS-CoV-2 gene content and COVID-19 mutation impact by comparing 44 Sarbecovirus genomes

Omicron SARS-CoV-2 variant: a new chapter in the COVID-19 pandemic

Diagnostics for SARS-CoV-2 infections

Diagnostics for SARS-CoV-2 infections

Electrically-receptive and thermally-responsive paper-based sensor chip for rapid detection of bacterial cells

Study of Chemical Enhancement Mechanism in Non-plasmonic

Natural tripeptide capped pH-sensitive gold nanoparticles for efficacious doxorubicin delivery both: In vitro and in vivo

Simulation of rainbows, coronas, and glories by use of Mie theory

The emergence, genomic diversity and global spread of SARS-CoV-2

Web 3DNA 2.0 for the analysis, visualization, and modeling of 3D nucleic acid structures

Study of support vector machine and serum surfacespectroscopy for noninvasive esophageal cancer detection

Reduced neutralization of SARS-CoV-2 B.1.617 by vaccine and convalescent serum

Surface-Enhanced Hyper-Raman Spectra of Adenine

How COVID vaccines shaped 2021 in eight powerful charts

Geographic and Genomic Distribution of SARS-CoV-2 Mutations

In situ plasmonic generation in functional ionic-gold-nanogel scaffold for rapid quantitative bio-sensing

Investigating the aggressiveness of the COVID-19 Omicron variant and suggestions for possible treatment options

Selective Naked-Eye Detection of SARS-CoV-2 Mediated by N Gene Targeted Antisense Oligonucleotide Capped Plasmonic Nanoparticles

Rapid and low-cost sampling for detection of airborne SARS-CoV-2 in dehumidifier condensate

Synthesis and characterisation of N-gene targeted NIR-II fluorescent probe for selective localisation of SARS-CoV-2

Oligodots: Structurally Defined Fluorescent Nanoprobes for Multiscale Dual-Color Imaging in Vitro and in Vivo

Emerging SARS-CoV-2 mutation hot spots include a novel RNA-dependent-RNA polymerase variant

A Facile Synthesis of Novel Self-Assembled Gold Nanorods Designed for Near-Infrared Imaging

Recent advances in colloidal gold nanobeacons for molecular photoacoustic imaging

Surface-enhanced Raman spectroscopy: benefits, trade-offs and future developments

The evolution of seasonal influenza viruses

Ripple mediated surface enhanced Raman spectroscopy on graphene

Comparison of commercial RNA extraction kits for preparation of DNA-free total RNA from Salmonella cells

Detection of SARS-CoV-2 and its S and N proteins using surface enhanced Raman spectroscopy

Intratumoral generation of photothermal gold nanoparticles through a vectorized biomineralization of ionic gold

A comprehensive review on plasmonic-based biosensors used in viral diagnostics

Complementary Oligonucleotide Conjugated Multicolor Carbon Dots for Intracellular Recognition of Biological Events

After the pandemic: perspectives on the future trajectory of COVID-19

Considerations for diagnostic COVID-19 tests

RT-qPCR Assays for Rapid Detection of the N501Y, 69-70del, K417N, and E484K SARS-CoV-2 Mutations: A Screening Strategy to Identify Variants With Clinical Impact

Raman spectroscopy and PCA-SVM as a non-invasive diagnostic tool to identify and classify qualitatively glycated hemoglobin levels in vivo

Ultrapotent antibodies against diverse and highly transmissible SARS-CoV-2 variants

Antibody resistance of SARS-CoV-2 variants B.1.351 and B.1.1.7

A simple RNA preparation method for SARS-CoV-2 detection by RT-qPCR

Quick detection of contaminants leaching from polypropylene centrifuge tubes with surface-enhanced Raman spectroscopy and ultraviolet absorption spectroscopy

Monolithic Integrations of Slanted Silicon Nanostructures on 3D Microstructures and Their Application to Surface-Enhanced Raman Spectroscopy

An efficient primary screening of COVID-19 by serum Raman spectroscopy

Nucleic Acid-Based Diagnostic Tests for the Detection SARS-CoV-2: An Update

Near-field surface-enhanced Raman spectroscopy of dye molecules adsorbed on silver island films

Rapid, Ultrasensitive, and Highly Specific Diagnosis of COVID-19 by CRISPR-Based Detection

The authors gratefully acknowledge the receipt of funding from the National Institute of