applied sciences and engineering copyright © 2020 virus … · draz et al., sci. dv. 2020 6 :...

Draz et al., Sci. Adv. 2020; 6 : eabd5354 16 December 2020

S C I E N C E A D V A N C E S | R E S E A R C H A R T I C L E

1 of 12

A P P L I E D S C I E N C E S A N D E N G I N E E R I N G

Virus detection using nanoparticles and deep neural network–enabled smartphone systemMohamed S. Draz1,2, Anish Vasan1, Aradana Muthupandian1, Manoj Kumar Kanakasabapathy1,2, Prudhvi Thirumalaraju1, Aparna Sreeram1, Sanchana Krishnakumar1, Vinish Yogesh1, Wenyu Lin3, Xu G. Yu4,5, Raymond T. Chung2,3, Hadi Shafiee1,2*

Emerging and reemerging infections present an ever-increasing challenge to global health. Here, we report a nanoparticle-enabled smartphone (NES) system for rapid and sensitive virus detection. The virus is captured on a microchip and labeled with specifically designed platinum nanoprobes to induce gas bubble formation in the presence of hydrogen peroxide. The formed bubbles are controlled to make distinct visual patterns, allowing simple and sensitive virus detection using a convolutional neural network (CNN)–enabled smartphone system and without using any optical hardware smartphone attachment. We evaluated the developed CNN-NES for testing viruses such as hepatitis B virus (HBV), HCV, and Zika virus (ZIKV). The CNN-NES was tested with 134 ZIKV- and HBV-spiked and ZIKV- and HCV-infected patient plasma/serum samples. The sensitivity of the system in qualita-tively detecting viral-infected samples with a clinically relevant virus concentration threshold of 250 copies/ml was 98.97% with a confidence interval of 94.39 to 99.97%.

INTRODUCTIONMobile health diagnostics are changing the face of modern medi-cine and health care (1, 2). The advances in consumer electronics and portable communication systems, particularly mobile phones, have led to rapid and cheaper approaches for developing point- of-care (POC) diagnostics (3–8). The global unique mobile subscribers in 2019 was approximately 5.18 billion and is estimated to reach more than 5.7 billion by 2025, and more than 10% of this number will be in sub-Saharan Africa alone where most of infection out-breaks occur (9). Such global access to mobile phones combined with its powerful computing ability and built-in sensors present a promising potential to develop digital diagnostics that may help in large-scale and efficient management of infectious diseases (3, 10).

Smartphone systems can also benefit from the recent unprece-dented advancements in nanotechnology to develop diagnostic ap-proaches. Catalysis can be considered as one of the popular applica-tions of nanoparticles because of their large surface-to- volume ratio and high surface energy (11–16). So far, numerous diagnostic platforms for cancer and infectious diseases have been developed by substituting enzymes, such as catalase, oxidase, and peroxidase with nanoparticle structures (17–20). Here, we adopted the intrinsic cat-alytic properties of platinum nanoparticles (PtNPs) for gas bubble formation to detect viruses on-chip using a convolutional neural net-work (CNN)–enabled smartphone system. Our unique approach of visual signal amplification through on-chip bubble formation com-bined with a CNN algorithm allows simple and rapid virus detec-tion using a smartphone camera without the need for any external smartphone optical attachment for image magnification and read-out or any target amplification. The virus is captured on the surface

of a microchip using monoclonal antibody (mAb) against the virus envelope protein and labeled with PtNPs, forming Pt-virus immuno-complexes on the surface of the microchip (Fig. 1, A and B). The addition of a catalyzer solution containing hydrogen peroxide (H2O2) to the formed Pt-virus complexes results in the formation of gas bubbles due to the catalase-like activity of PtNPs that drives the de-composition of H2O2 into water and O2 (Fig. 1B). The on-chip gas bubble formation was controlled by optimizing the density of the reaction medium to form stable bubbles that can be detected by the CNN-enabled smartphone system (Fig. 1, C and D). Although using metal nanoparticles for target labeling and detection has been previ-ously reported, unlike all other previous work in mobile health technologies for protein/virus detection, our system is simple in sample processing, sensitive, and adaptable to different smartphone models, mainly because our CNN–nanoparticle-enabled smartphone (NES) system does not need any hardware optical smartphone attach-ment for signal detection (21–30). We first studied the nanoprobe preparation, virus capture, and bubble formation using Zika virus (ZIKV) as a clinical model for target virus detection. The devel-oped CNN-NES was tested for qualitative detection of virus- infected samples including hepatitis B virus (HBV), HCV, and ZIKV samples with broad and clinically relevant virus concentrations.

RESULTSVirus capture and labeling using Pt-nanoprobesSpecifically designed Pt-nanoprobes were prepared with PtNPs and mAb against the envelope protein of ZIKV as illustrated in fig. S1. The preparation reaction involves the coupling of the antibody molecules to the surface of 4.5-nm PtNPs through the carbohydrate residues in the fragment crystallizable (FC) region of the immuno-globulin G (IgG) antibody using a short pyridyldithiol-and-hydrazide cross-linker of 3-(2-pyridyldithio)propionyl hydrazide (PDPH). The prepared PtNPs and their coupling to PDPH and antibody molecules were assessed by transmission electron microscopy (TEM), ultraviolet- visible (UV-vis) spectroscopy, dynamic light scattering (DLS), gel electrophoresis, and Fourier transform infrared (FT-IR) spectroscopy

1Division of Engineering in Medicine, Department of Medicine, Brigham and Women’s Hospital, Harvard Medical School, Boston, MA 02139, USA. 2Harvard Medical School, Boston, MA 02115, USA. 3Liver Center, Gastrointestinal Division, Massachusetts Gen-eral Hospital, Harvard Medical School, Boston, MA 02114, USA. 4Division of Infec-tious Diseases, Brigham and Women’s Hospital, Harvard Medical School, Boston, MA 02139, USA. 5The Ragon Institute of Massachusetts General Hospital, Massachusetts Institute of Technology and Harvard University, Boston, MA 02129, USA.*Corresponding author. Email: [email protected]

Copyright © 2020 The Authors, some rights reserved; exclusive licensee American Association for the Advancement of Science. No claim to original U.S. Government Works. Distributed under a Creative Commons Attribution NonCommercial License 4.0 (CC BY-NC).

on May 15, 2021

http://advances.sciencemag.org/

Dow

nloaded from




2 of 12

techniques (Fig. 2 and figs. S2 to S7). For antibody coupling to the surface of PtNPs, agarose gel electrophoresis showed that the mi-gration of PtNPs in gel is slightly retarded after conjugation reaction, suggesting the addition of antibodies to their surface, which increases the size of PtNPs and partially interferes with their electrostatic be-havior (fig. S6). FT-IR spectroscopy spectra of PtNPs and antibody- modified PtNPs are shown in fig. S7. The conjugation of mAb to the surface of PtNPs resulted into the following peaks: C─OH stretch around 1087.8 cm−1, the absorption bands of ammonium ions at 2316.5 cm−1, amide I groups at 1766.8 cm−1, and amide II groups at 1579.7 cm−1 (31). Bicinchoninic acid (BCA) protein assay showed an average antibody concentration of 348.32 ± 13.14 g/ml (n = 2) in the formed Pt-nanoprobe preparations, which confirms the con-jugation of the antibody to the surface of PtNPs (Materials and Methods).

The microchip surface is functionalized with anti-ZIKV mAb to allow efficient capture and labeling of on-chip ZIKV particles. Anti- ZIKV mAbs were conjugated to the surface of the chips using a surface chemistry protocol specifically designed to allow an efficient directional conjugation of antibodies using polyethylene glycol (PEG) molecules bifunctionalized with a terminal thiol and silane group (fig. S8). The glass surface of the microchip was initially si-

lanized with PEG, and then oxidized antibodies activated with PDPH were incubated on the surface of the PEG-modified chip to allow the interaction of pyridyldithiol group of PDPH with the free on-chip –SH groups. The functionalization of the chip with anti-ZIKV mAb was confirmed using SDS–polyacrylamide gel electrophoresis (SDS-PAGE) and FT-IR spectroscopy techniques. For SDS-PAGE, the immobilized antibody molecules on the surface of the chip were digested using an SDS-PAGE sample preparation buffer and eluted for testing on a 4 to 20% precast polyacrylamide gel. The results showed bands around 50 and 25 kDa that are characteristic to IgG heavy and light chains, respectively (fig. S9). FT-IR spectroscopy analysis spectra of antibody and antibody-modified chips were sim-ilar in displaying many bands that are characteristic to antibody. Specifically, the absorption bands of amide I groups at 1779.17 cm−1 and amide II groups around 1568.01 cm−1 confirmed the antibody immobilization on the surface of the chip. In addition, sharp peaks at 922.8, 724.03, 669.45, and 560.9 cm−1 are attributable to Si─O─Si stretching, H─Si─O vibrations, and C─H and reflect the efficient salinization and PEGylation of the surface of the chip (fig. S10) (31).

The capture and labeling of the virus on the microchip were evaluated using SDS-PAGE, enzyme-linked immunosorbent assay (ELISA), reverse transcription polymerase chain reaction (RT-PCR),

Fig. 1. Three-dimensional schematic of the CNN–nanoparticle-enabled smartphone for virus detection. The detection process comprises three main steps. (A) Virus capture and labeling using Pt-nanoprobes. The samples are loaded into microchips modified with mAbs against the virus envelope protein and incubated to allow the capture of target virus for 20 min. The captured virus particles are then labeled with Pt-nanoprobes within 20 min. Each step of on-chip virus capture and labeling is followed by a washing step using 10 mM phosphate buffer (pH 7.4) (B) Catalyzer solution containing hydrogen peroxide (H2O2) is added to the chip and incubated for 10 min. In the presence of the captured virus, Pt-nanoprobe complex bubbles (oxygen) are formed because of the catalytic activity of PtNPs in contact with H2O2. Scale bar is set on the basis of the microchip dimensions. (C) The on-chip bubble signal is detected using the CNN–nanoparticle-enabled smartphone (NES). Photo credit: Mohamed S. Draz, Brigham and Women’s Hospital. (D) Screenshots of the results shown by the CNN-NES. Photo credit: Mohamed S. Draz, Brigham and Women’s Hospital.

on May 15, 2021


Dow

nloaded from




3 of 12

and electron microscopy techniques. SDS-PAGE showed a high degree of protein similarity with samples eluted from the surface of the chip and ZIKV samples, confirming on-chip virus capture as the major virus structural proteins including envelope, precursor M protein, and capsid were clearly identified (Fig. 2C and fig. S9) (32). ELISA and RT-PCR results indicated that the virus capture efficien-cy on-chip were 89.08 ± 1.07 and 82.7 ± 5.66%, respectively (Fig. 2D). Under scanning electron microscopy (SEM), the number of cap-tured viral particles increased with increased virus concentrations of the samples loaded into the chips (fig. S11). We observed one viral particle per 3.7 m2 of the microchip surface when a sample with 106 copies/ml of ZIKV concentration was incubated on-chip. The average size of the formed virus-PtNP complex was 84.3 ± 26.48 nm, which is substantially larger than ZIKV with 40 to 50 nm in diameter (Fig. 2E and fig. S12) (33, 34).

Development and evaluation of the CNN-NES for virus detectionTo develop attachment-free smartphone-based virus detection, we integrated the nanoparticle bubble approach with a smartphone system modified with a trained CNN algorithm running on an An-droid smartphone (XT1575, Motorola) that can image and analyze bubbles on a microchip with a straight microchannel. The CNN algorithm uses the Inception v3 architecture, which was transfer learned using Google’s TensorFlow deep learning framework, with images of microchips containing bubbles analogous to virus sam-ples (figs. S13 to S15). To train our system, we performed five differ-ent trials with ZIKV-spiked phosphate-buffered saline (PBS) samples (n = 25) at different virus concentrations around a threshold value of 250 copies/ml, which is clinically relevant to the target viruses in

this study (HBV, HCV, and ZIKV) (35–42). The tested virus con-centrations were 0, 250, 103, 104, and 105 copies/ml, and the corre-sponding microchip images recorded with a smartphone were used for training the algorithm. On the basis of the data associated with virus-spiked samples, we also prepared a set of PtNP-spiked sam-ples that simulated bubbles formed with virus samples for training the CNN-NES. The training and validation dataset included 15,057 and the testing set included 1502 images from 38 videos of virus- spiked PBS samples (n = 5) and PtNP-spiked samples analogous to different virus concentrations of 0, 250, 103, 104, and 105 particles/ml (n = 33) (Materials and Methods). Thus, the developed NES can qualitatively identify samples as positive (viral load above or equal to the threshold) and negative (viral load below the threshold) on the basis of a virus concentration threshold of 250 particles/ml, which is clinically relevant to the target viruses in this study (HBV, HCV, and ZIKV) (35–42).

To evaluate the performance of the NES system for virus detec-tion, we used HBV-spiked serum samples (n = 22), ZIKV-spiked blood samples (n = 60), HCV-infected patient plasma/serum sam-ples (n = 27), and ZIKV-infected patient serum samples (n = 25) (Fig. 3, figs. S16 and S17, and tables S1 and S2). Each virus sample was tested using the single-channel microchip modified with mAbs against the target virus surface protein (i.e., anti-ZIKV envelope antibody for ZIKV, anti-HBV surface antigen antibody for HBV, and anti-HCV core antigen antibody for HCV) following the afore-mentioned bubble protocol using the catalysis solution containing 5% H2O2 and 20% glycerol and incubated for 10 min. We per-formed receiver operating characteristic (ROC) curve analysis to evaluate the performance of the CNN-NES in qualitatively detecting viral-spiked and viral-infected samples for ZIKV, HBV, and HCV

Fig. 2. Virus capture and bubble formation on-chip. Pt-nanoprobes preparation: (A) TEM image and particle size distribution histogram. (B) Schematic of Pt-nanoprobes structure and UV-vis absorption analysis showing spectra of PtNPs (citrate-capped PtNPs, black) and Pt-nanoprobes (PtNPs conjugated to mAb, red). Photo credit: Mohamed S. Draz, Brigham and Women’s Hospital. Virus capture and labeling on-chip: (C) SDS-PAGE analysis of ZIKV released from the surface of the chip. Lane M, protein marker; lane ZIKV on chip, ZIKV released from the chip; lane ZIKV, ZIKV standard solution (106 particles/ml, virus not added to the chip). Rectangles in red and blue edges mark bands from antibodies and ZIKV, respectively. (D) ELISA and RT-PCR assay results to detect and quantify the captured ZIKV on-chip. Error bars are SDs from two independent experiments. (E) SEM images of ZIKV captured on chip and labeled with Pt-nanoprobes: (i) ZIKV on chip and (ii) ZIKV captured and labeled with Pt-nanoprobes. The concentration of the virus sample was 106 particles/ml. The results are expressed as the average values of at least three independent experiments.

on May 15, 2021


Dow

nloaded from




4 of 12

samples with a virus concentration threshold of 250 copies/ml (Fig. 3, A and B). The areas under the ROC curve (AUCs) for ZIKV-spiked and HBV-spiked PBS samples, ZIKV-infected patient serum, and HCV-infected patient serum/plasma samples were 0.984 [con-fidence interval (CI), 0.911 to 1.000], 0.992 (CI, 0.831 to 1.000), 1

(CI, 0.863 to 1.000), and 1 (CI, 0.872 to 1.000), respectively. The AUC when all spiked and patient samples were included was 0.979 (CI, 0.939 to 0.996). Vertical scatterplots between CNN-NES prob-ability value results and PCR qualitative-testing results for spiked samples (Fig. 3C, i) and patient samples (Fig. 3C, ii) showed one

Fig. 3. Virus detection using the CNN-NES. (A) Digital images of the developed CNN-NES (i) and the actual microchips (ii) used to qualitatively test a virus-spiked sample. The system can measure the probability value of sample for qualitative testing (positive or negative) without the need for any optical smartphone attachment. Photo credit: Mohamed S. Draz, Brigham and Women’s Hospital. (B) ROC analysis of the developed CNN-NES compared to the standard RT-PCR using (i) ZIKV-spiked samples (n = 60), (ii) HBV-spiked samples (n = 22), (iii) ZIKV-infected patient serum samples (n = 25), and (iv) HCV-infected patient serum samples (n = 27). (C) Vertical scatterplots representing the CNN-NES probability values of the samples versus qualitative PCR results: (i) ZIKV-spiked samples (n = 60), HBV-spiked samples (n = 22), (ii) ZIKV-infected patient serum samples (n = 25), and HCV-infected patient serum samples (n = 27). (D) Vertical scatterplots representing the CNN-NES probability values of the samples versus quantitative PCR results: (i) ZIKV-spiked samples (n = 60), HBV-spiked samples (n = 22), (ii) ZIKV-infected patient serum samples (n = 25), and HCV-infected patient serum samples (n = 27).

on May 15, 2021


Dow

nloaded from




5 of 12

false positive and one false negative for ZIKV-spiked samples and two false positives for HBV-spiked samples. Vertical scatterplots between CNN-NES probability value results and PCR quantitative- testing results for spiked and patient samples are also shown in Fig. 3D (i and ii, respectively). These results showed that the system was confused for one ZIKV-spiked sample with a viral load of 250 copies/ml, two HBV-spiked samples with viral loads of 200 copies/ml, and one virus-free control sample [Fig. 3, C (i) and D (i)]. The system was able to accurately identify HCV-infected patient serum/plasma samples without any error [Fig. 3, C (ii) and D (ii)].

Using a panel of 25 ZIKV-infected patient serum samples and 27 HCV-infected serum/plasma samples, the performance of the CNN-NES in qualitatively (virus concentration threshold of 250 particles/ml) identifying the infected serum samples was evaluated and compared to the results obtained by the nucleic acid and im-munoassay techniques currently approved by the U.S. Food and Drug Administration and recommended by the U.S. Centers for Disease Control and Prevention (CDC) for virus detection [Fig. 3, B (iii and iv), C (ii), and D (ii); figs. S16 and S17; and table S2]. The CNN-NES showed a concordance of 100% in identifying HCV- infected patient samples compared to PCR and 100% in identifying ZIKV-infected patient samples compared to the nucleic acid assay of Aptima ZIKV Assay and 80% with both the tested immuno-assays of ZIKV Detect IgM Capture ELISA (ZIKV IgM detection) and CDC Zika IgM antibody capture enzyme-linked immunosorbent assay (MAC-ELISA), which can be attributed to the false-positive ELISA results usually caused by cross-reactivity with the immune response generated by previous flavivirus infections [Fig. 3, B (iii and iv), C (ii), and D (ii), and table S2] (1, 4). We also evaluated the specificity of the system by testing nontarget viral-infected patient samples (HCV- and HIV-infected patient serum samples) on target microchips (microchips functionalized with anti-HBV antibodies) (table S3). These qualitative test results show that the CNN-NES was able to qualitatively identify target images of target microchips tested with target virus (HBV), even when target HBV-infected pa-tient serum samples with viral loads much lower than the virus concentration of nontarget HCV- and HIV-infected patient serum samples were used. Although our system does not calculate bubble count for its analysis, we have also reported the bubble counts asso-ciated with the target and nontarget patient serum samples, as this is the simplest way of quantitatively presenting the data.

Evaluation of the CNN-NES using a sample-processing cartridgeTo further simplify sample testing using the CNN-NES system, we developed a custom-built microfluidic cartridge for sample pro-cessing (Fig. 4, A and B, and fig. S18). The developed cartridge is preloaded with all the reagents and materials required for sample testing (i.e., washing buffer, Pt-nanoprobe, and H2O2 solution), eliminating manual sample preparation and pipetting of multiple reagents and reducing potential user error (fig. S18). The microflu-idic cartridge comprises four layers of poly(methyl methacrylate) (PMMA) assembled together using double-sided adhesive (DSA) film that includes a single microfluidic channel with a total volume of 240 l (fig. S19). The sample is loaded into the sample microchip, which is then inserted into the microchip slot on the cartridge for sample processing. The microchannel on the cartridge is terminally connected to the sample and enabled with a rubber bulb to allow easy and efficient loading and removing of on-chip reagents. The

cartridge is also modified with a cellulose paper pad (7 cm in length and 1.5 cm in width) placed in a waste reservoir to absorb the reagents loaded in the sample channel during testing.

To test the performance of the developed microfluidic cartridge in sample testing, we used 1× PBS samples spiked with HBV (n = 42) and ZIKV (n = 36) (Fig. 4 and table S4). ROC analysis showed that the sensitivity and specificity of the CNN-NES when the cartridge was used for sample processing were 100.0% (CI, 93.5 to 100.0%) and 100.0% (CI, 85.2 to 100.0%), respectively. The AUC was 1.0 with a binomial exact CI ranging from 0.954 to 1.0 (Fig. 4C and figs. S20 and S21). Vertical scatterplot showed that the 78 tested virus- spiked PBS samples were correctly classified as positive and nega-tive, indicating a 100.0% accuracy compared with PCR (Fig. 4D). We also evaluated the performance of the CNN-NES in terms of detection reliability and repeatability when different smartphones, sample microchips, and cartridges were used for testing. For device to device, we tested 25 HBV-spiked PBS samples with a virus con-centration of 105 copies/ml and virus-free control PBS samples using different sets of cartridges (n = 17). The results showed no difference in identifying samples of the same group (i.e., virus-free control group and HBV-spiked sample group) tested with different cartridges (Fig. 4E). In this part of the study, we reused a set of car-tridges (n = 7) for sample testing (cartridge nos. 10, 18, 21, 22, 36, 37, and 38; Fig. 4E). These results showed no difference in the qual-itative detection results when the CNN-NES was used to test the same HBV-spiked sample (105 copies/ml) on different microchips (Fig. 4E). To evaluate the user error in sample processing with the developed cartridge, a set of 36 1× PBS samples spiked with ZIKV (n = 18) and HBV (n = 18) were tested by different trained and untrained users (n = 8). The trained and untrained users recruited for the system evaluation had different backgrounds and included a physician and non-PhD graduate and undergraduate students (Fig. 4F and table S5). The untrained users performed the tests using a sim-ple instruction sheet presented in fig. S22. The results indicated 100% agreement in correctly classifying the samples that were tested by both the trained and untrained users. To evaluate the effect of smartphone model on the performance of the CNN-NES, we used Moto X, Redmi Note 4, and Google Pixel 2 smartphones for bubble on-chip detection. The different smartphone systems were used to test the ZIKV-spiked PBS samples (n = 36). We did not observe any significant difference [P > 0.05, one-way analysis of variance (ANOVA) with Kruskal-Wallis test] in qualitatively classifying the tested sam-ples at the same virus concentration in a range of 0 to 107 particles/ml (Fig. 4G and table S6).

DISCUSSIONRapid and sensitive POC diagnostics with the ability to be seamlessly integrated with appropriate and effective surveillance mechanisms can shift the paradigm in outbreak control. The growing advances in consumer electronics and portable communication systems, par-ticularly smartphones, have led to a significant growth in smartphone subscribers worldwide, particularly in the developing countries, and faster and cheaper approaches of data acquisition. Previous mobile health technologies for target virus/protein detection have lacked broad technical applicability, including adaptability to different smartphone models, because of their dependency on smartphone- specific hardware optical attachments (21–30). We overcome this challenge by integrating our unique target labeling and signal

on May 15, 2021


Dow

nloaded from




6 of 12

amplification using catalytic properties of nanoparticles to generate bubbles on-chip with an on-phone CNN-based signal detection al-gorithm, which allowed us to develop a smartphone attachment-free optical sensor for virus detection. In addition, we used a microfluidic cartridge to enable simple sample handling that can be performed by a

lay user. The catalytic property of PtNPs allows the formation of gas bubbles creating distinct patterns on-chip, which are easy to be rec-ognized using a smartphone camera. The developed amplification- free and sensitive (250 particles/ml) CNN-NES can qualitatively detect the target virus in 50 min and is not dependent on the model

Fig. 4. Virus detection using the CNN-NES and a sample-processing microfluidic cartridge. (A) Custom-built microfluidic cartridge used for simple sample process-ing. (B) Exploded three-dimensional schematic of the developed cartridge indicating the detailed layer structure and the configuration of main components of the car-tridge. (C) ROC curve analysis of CNN-NES enabled with the developed cartridge compared to the standard RT-PCR using PBS samples spiked with viruses, including HBV (n = 42) and ZIKV (n = 36). The samples with viral loads are classified as positive (≥250 particles/ml) and negative (<250 particles/ml). (D) Vertical scatterplot 1× PBS-spiked samples. The threshold for the probability values measured was 0.5. Samples with probability values above and below 0.5 were classified as positive (1) and negative (0), respectively, compared to the RT-PCR at a threshold of 250 particles/ml. (E) Evaluation of the repeatability of qualitative detection of samples when different cartridges were used for testing the same HBV-spiked PBS samples (n = 25). The results are expressed as the average values of at least three measurements from one sample. (F) Evaluation of the performance of the system when operated by different users for sample testing using PBS samples spiked with ZIKV (n = 18) and HBV (n = 18). The results are expressed as the average values of three experiments from one sample. (G) Evaluation of the performance of the system when different smartphones were used for sample testing. The results are expressed as the average values of three independent experiments. The probability value is presented for each sample being positive. Error bars represent SDs. ns, not significant.

on May 15, 2021


Dow

nloaded from




7 of 12

of the smartphone used for detection, which is a major advance-ment in the development of mobile health technologies for virus detection. The developed CNN-NES is a powerful universal modal-ity that can be adapted to a wide number of smartphone models without the need for extensive modifications and can be used for the detection of wide range of infections. In this system, we relied on the enhanced catalytic properties and stability of PtNPs over tra-ditional peer enzymes to allow stable and rapid gas bubble for-mation (43, 44). The catalytic activity of PtNPs for gas formation is directly related to the size of the PtNPs, and larger nanoparticles are expected to have higher activity (45). However, relatively small size PtNPs (4.5 nm in diameter) were used in this study to avoid surface particle load limitation during virus labeling and to minimize the effect of hydrodynamic shear forces on the formed Pt-virus com-plexes during washing and labeling steps that may release the captured viruses on-chip (46). In addition, high catalytic activity of PtNPs in high concentration of H2O2 leads to rapid merging of the generated bubbles and forming irregular bubble shapes, which can make accurate signal detection difficult (fig. S23). To avoid rapid bubble merging and to control the stability of the visual pat-terns on-chip after virus capture and signal amplification, we used glycerol (20% of the solution) to increase the density of the catalyzer solution to ~0.73 g/cm3. The formed bubbles reached to macro-scopic sizes (up to 90 m) after 300 s of incubation at 10% H2O2 solution without visible merging, which is even higher than the 5% H2O2 concentration used in the study (figs. S24 and S25). One addi-tional parameter that played an important role to overcome the lim-itation on controlling the gas bubble formation process was the well-distributed virus capture on the surface of the chips. Our sur-face chemistry scheme that included the addition of a bifunction-al PEG cross-linker (2 kDa, 10 times smaller than antibody) allowed a controlled indirect linking of antibody to the microchip surface. PEGylation increases the conformational stability of proteins and resistance to degradation. Therefore, we used PEG on the surface of the microchips to help in stabilizing the antibody activity and avoiding nonspecific interactions, for easy washing, and to promote the stability of conjugated biomolecules (47–50). PEG acts as a flex-ible arm that provides maximum accessibility of antibody and a higher chance for avid interaction with virus. The directional con-jugation of antibody through the carbohydrate residue in the FC region of the antibody, other than amine or carboxyl groups that are usually involved in interaction with the target antigen, helps in pre-serving the full activity and specificity of the conjugated antibodies (51, 52). All these factors together with using highly specific mAb (table S7) led to a detection sensitivity of 98.97% and specificity of 91.89% (n = 134 of ZIKV, HCV, and HBV) for the reported CNN-NES virus detection approach (fig. S26). The developed bubble ap-proach was stable under different environmental conditions includ-ing different temperatures and relative humidity. The change in the incubation temperature and relative humidity did not have any significant effect (P > 0.05, unpaired t test) on the number of bub-bles generated on-chip (figs. S27 to S30). Likewise, the change in the H2O2 concentration showed insignificant effect (P > 0.05, unpaired t test) on bubble formation using a single-channel microchip (figs. S31 and S32). We further tested the shelf life time of the anti-body-functionalized chips and the results indicated the stability of the prepared chips when lyophilized and stored at room tempera-ture on shelf for 45 days (figs. S33 and S34). The current prototype of the developed bubble technology has a great potential for low-

cost POC diagnostic applications and has several advantages over the standard detection techniques currently used in virus detec-tions such as RT-PCR and ELISA techniques (tables S8 and S9). However, its integration in a fully automated quantitative system is important and can be addressed in future work using an ad-vanced microfluidic platform (8). As a nanoparticle-based tech-nology, the stability and long-term storage can be also improved by the use of specific surface chemistries in the preparation process and the addition of stabilizing agents to the nanoparticle solution (8, 53). In summary, we reported the development and evaluation of a diagnostic assay with the potential for POC viral load testing and demonstrated a unique integration of advancements in nano-technologies, microfluidics, and consumer electronics to solve an unmet clinical barrier in the management of infectious diseases. The developed platform is simple and cost effective and can be adapted to a wide variety of other pathogens and has the potential for various clinical applications.

MATERIALS AND METHODSBiosafety and human participant statementThe research work reported was approved and performed in adher-ence to guidelines and procedures approved by the Institutional Biosafety Committee of Partners HealthCare (parent organization of Massachusetts General Hospital and Brigham and Women’s Hospital) and under appropriate institutional review boards (IRB nos. 2019P002209 and 2019P001489). All the experiments with ZIKV, HBV, and HCV were performed in BSL2. Deidentified ZIKV-infected patient serum samples were obtained from certified biorepositories of consented patients’ samples at Boca Biolistics LLC. Deidentified HCV-infected and HBV-infected patient plasma/serum samples were obtained from Discovery Life Sciences Inc.Virus stock samples and viral load measurementsZIKV, virus strain PRVABC59 isolated by the CDC from a ZIKV- infected patient who travelled to Puerto Rico in 2015 (National Center for Biotechnology Information accession no. KU501215) was provided by X. G. Yu at the Ragon Institute. HBV-cultured samples were provided by R. T. Chung at Massachusetts General Hospital. HBV stock virus was obtained from HepG2 AD38 cells (54, 55) as described previously. HIV stock–cultured samples were received from our colleagues at the division of engineering in medicine at Brigham and Women’s Hospital. HCV-infected and HBV-infected patient serum/plasma samples were purchased from Discovery Life Sciences Inc. The viral loads of HBV-infected and HCV-infected patient serum samples were measured using a standard RT-PCR system and reported by the vendor (Discovery Life Sciences Inc.). The viral load of ZIKV-infected patient samples received from Boca Biolistics LLC was not known, as the PCR test results of those sam-ples were reported qualitatively by the vendor. The viral loads of all ZIKV-spiked samples were measured through RT-PCR by our col-laborator at the Ragon Institute. The viral loads of all other samples were either reported by the provider or measured with a RT-PCR system by Viracor Eurofins Inc. The spiked samples that were used in system training or testing throughout the study were prepared by serial dilutions of stock virus samples with known viral loads mea-sured by standard PCR approaches. In these cases, the control sam-ples were virus-free samples that were processed for the detection assay with the same protocol as all other viral samples, with the exception of the addition of viral-stock dilutions.

on May 15, 2021


Dow

nloaded from




8 of 12

Microchip design and fabricationSingle-channel microchips functionalized with antibodies against the target viruses were used for virus capturing. The microchip was designed using the vector graphics editor CorelDRAW X7 software (Autodesk Inc., San Rafael, CA, USA). The microchips were fabri-cated from PMMA (3.1 mm thick; McMaster-Carr, Atlanta, GA, USA), DSA film (80 m thick; iTapeStore, Scotch Plains, NJ, USA), and glass slides (25 mm by 75 mm; Globe Scientific, NJ, USA). The fabrication of the chip starts by cutting PMMA and DSA film using a laser cutter (Versa Laser, Scottsdale, AZ, USA). The PMMA was prepared to contain the inlets and outlets, while DSA film included the main testing channels. All materials were precleaned with 5% ethanol, 30% H2O2, and deionized water using lint-free tissue. The surface of the cleaned glass slides was activated using PE-25 oxygen plasma (100 mW, 15% oxygen; Plasma Etch Inc., Carson City, NV, USA) for 2 to 3 min and modified using silane- functionalized PEG thiol (Nanocs Inc., New York, NY, USA) and mAb for capture of the target virus: anti-ZIKV envelope antibody (catalog no. HM325, EastCoast Bio Inc., North Berwick, ME, USA) for ZIKV, anti-HBV surface antigen (Ad/Ay) antibody (catalog no. ab54247, Abcam, Cambridge, MA, USA) for HBV, and anti-HCV core antigen anti-body (catalog no. ab2582, Abcam, Cambridge, MA, USA) for HCV. Then, PMMA and DSA film were assembled on the modified glass slides to form the complete microchip. The entire process of clean-ing and assembling was done in a clean hood to prevent contami-nation due to stray dust and airborne particles.Nanoprobe preparation and characterizationDifferent sets of platinum nanoprobes were prepared of spherical PtNPs modified with mAb against the target virus: anti-ZIKV (ZIKV-Env) antibody (EastCoast Bio Inc. North Berwick, ME, USA) for ZIKV, anti-HBV surface antigen (Ad/Ay) antibody (HB24, catalog no. ab54247, Abcam, Cambridge, MA, USA) for HBV, and anti- HCV core antigen antibody (catalog no. ab2582, Abcam, Cambridge, MA, USA) for HCV. The synthesis protocol starts by modifying the surface of PtNPs with PDPH freshly reduced by 20 mM tris(2- carboxyethyl)phosphine (TCEP). The prepared PtNP-PDPH was then coupled to antibody oxidized using 10 mM sodium metaperi-odate for 1 hour at room temperature. The prepared PtNPs, PDPH-PtNPs, and Pt-nanoprobes were characterized using TEM, UV-vis spectroscopy, FT-IR spectroscopy, and DLS techniques.PtNP synthesis and characterizationPtNPs were synthesized using a modified method from literature (56). All glassware used was cleaned with aqua regia and ultrapure water. Thirty-six milliliters of a 0.2% solution of chloroplatinic acid hexahydrate was mixed with 464 ml of boiling deionized water. Eleven milliliters of a solution containing 1% sodium citrate and 0.05% citric acid was added followed by a quick injection of 5.5 ml of a freshly prepared 0.08% sodium borohydrate solution, contain-ing 1% sodium citrate and 0.05% citric acid. The reaction continued for 10 min, and the formed nanoparticle solution was gradually cooled down to room temperature. The formed PtNPs were modi-fied with PDPH freshly reduced by 20 mM TCEP. The prepared PtNPs were characterized using TEM, UV-vis spectroscopy, potential, and DLS.Antibody oxidation and activation with PDPH cross-linkerThe antibody was oxidized using the standard protocol in which antibody was mixed with 10 mM sodium metaperiodate and 0.1 M sodium acetate (pH 5.5) and incubated at 4°C in the dark for 20 min. Then, PDPH-activated PtNPs are allowed to react with the

oxidized antibody for 1 hour at room temperature. The prepared PDPH-modified antibodies are then washed three times using 1× PBS to remove the excess PDPH using Amicon Ultra Centrifugal filter units Ultra-4; molecular weight cutoff, 100 kDa (Millipore, MA, USA).Quantification of the number of anti-ZIKV mAb molecules per Pt-nanoprobeThe concentration of antibody was estimated using the bicin-choninic acid (BCA) protein assay according to the manufacturer’s protocol (Pierce). Absorbance of aliquots of Pt-nanoprobe solution (PtNP + antibody) and PtNPs solution (no antibody) values was measured at 562 nm using a BioTek microplate reader, and the con-centration of the antibodies was estimated after subtracting the background absorbance of nanoparticles. The number of antibody molecules on each PtNP was calculated by dividing the antibody concentration by the concentration of the PtNPs in the Pt-nanoprobe solution as follows:

The concentration of Pt-nanoprobes was adjusted to 2.39 M, and the concentration of protein was estimated to be 348.32 ± 13.14 g/ml. Considering that the molecular weight of an antibody molecule is 150 KDa, the molar concentration of antibody was calculated to be 2.32 ± 0.085 M, which implies the presence of 0.971 ± 0.0355 anti-body molecule per Pt-nanoprobe.Quantification of virus captured on-chipAliquots of ZIKV were loaded on a single-channel chip modified with anti-ZIKV (ZIKV-Env) antibody, incubated for 20 min at room temperature for virus capture, and then washed using phos-phate buffer (PB) three times. The captured virus was digested and eluted for testing using ELISA and RT-PCR techniques. The con-centration of the eluted virus was calculated from a standard curve prepared from known concentrations of virus.Bubble on-chip assayThe bubble assay relies on capture of target virus on the surface of microchips and labeled with PtNPs and the formation of gas bub-bles in the presence of H2O2. The bubble assay working protocol includes three main steps. (i) Virus capture on the surface of micro-chip. Around 20 l of the target virus sample was loaded on the microchip and incubated for 20 min at room temperature followed by a washing step using 10 mM PB (pH 7.4). (ii) Virus labeling with Pt-nanoprobes. Twenty microliters of Pt-nanoprobe solution against the target virus was loaded on the surface of the microchip with the captured virus particles. The microchip was incubated for 20 min at room temperature, and the excess Pt-nanoprobes were washed us-ing 10 mM PB (pH 7.4) three times. (iii) Bubble formation. A 5% hydrogen peroxide solution containing 20% glycerol was added to the microchip and incubated for 10 min. This protocol can be per-formed by manually adding and removing the reagents to the chip or using a microfluidic cartridge preloaded with all the reagents needed for sample testing including PB for washing, PtNP solution, and the 5% H2O2 solution. The formed bubbles were then imaged and detected using optical microscopy and the NES system. The capture of virus particles and formation of virus-PtNP complexes on the surface of microchip was confirmed using ELISA, RT-PCR, SDS gel electrophoresis, and SEM techniques.

Cartridge assembly and operationThe cartridge is a core-shell microfluidic unit that is 175 mm in length and 75.5 mm in width and prepared of a total four layers of PMMA and three layers of DSA film. The cartridge core comprises

on May 15, 2021


Dow

nloaded from




9 of 12

a set of two PMMA layers (1.5 mm in thickness) assembled together with a DSA film layer to form a single multilane microfluidic chan-nel (260 cm in length, 80 m in depth, and 1 mm in width). The inside surface of each PMMA layer forming the cartridge core is pretreated with Rain-X 620036 Plastic Water Repellent to be more hydrophobic, allowing easy flow through the microfluidic channel. The PMMA shell set comprises two PMMA layers with 3.2-mm thickness. The top PMMA layer contains a sample housing cavity for sample insertion, while the lower PMMA layer is engraved and modified with a cellulose paper absorbing pad (1 mm in thickness, 71 mm in length, and 22 mm in width). The core set is first loaded with the sample testing reagents through a loading well on the top layer in the following order: 40 l of PB solution (first washing step), 35 l of nanoprobe solution (for virus labeling), 60 l of PB solution (second washing step), 35 l of 5% H2O2 solution, and 5 l of indicator dye, separated by 18-l equivalent space of air. After-ward, the PMMA layers are added, and a rubber bulb (14.4 mm in diameter and 7.2 mm in height) is sealed under the top PMMA layer to allow easy and controlled manipulation of reagents preloaded on the cartridge. The virus sample was loaded on a single-channel sample microchip using 20-l Non-Sterile Exact Volume Transfer Pipettes (Thermo Fisher Scientific Samco 783NL). The sample microchip was then inserted in the sample housing cavity and connected to the core microfluidic channel with two plastic tips in the core set. The sample processing was started by compressing the rubber bulb until the dye indicator solution reached to position 1 (arrowhead 1 at the left side of the cartridge) to complete the virus sample washing and to add the labeling Pt-nanoprobe solution on the microchip. The cartridge and the loaded microchip were then incubated for 20 min at room temperature, following which the bulb reagent loading was continued until the dye indicator reached position 2 (arrowhead 2 at the left side of the cartridge). The microchip was then detached from the device and incubated for 10 min for bub-ble formation, and the bubbles were imaged and evaluated using the CNN-NES.Smartphone system and CNN algorithmWe used a CNN model performing supervised learning to automat-ically recognize differences between two classes of positive (infected) and negative (noninfected) samples. We used Google’s Inception v3 CNN architecture pretrained to 93.33% top-five accuracy on a data-set of 1000 object classes containing 1.28 million images of the 2014 ImageNet challenge. We then performed transfer learning by re-moving the final classification layer from the network and retrained it with our dataset. Our dataset is composed of prelabeled images organized in two different classes generated from 38 videos of ZIKV-spiked PBS samples and bubble-simulated samples prepared with different dilutions of PtNPs on a single-channel microchip system and recorded using Moto G4, Moto X, and iPhone 6 smart-phones. The videos were recorded under different orientations (0o to 360o), backgrounds (black, white, brown, and random out-door backgrounds), and illuminations (150 to 700 lux). The images were created from video files using a small Python script that is run on Linux. Frames were extracted from these videos and organized into two different classes (negative and positive) for training. We used 15,057 images for training and validation and 1502 images for testing in the development of our algorithm. We did not use any of the patient samples or spiked samples used during the prospective testing of the system for training of the network. In other words, none of the test samples were ever seen by the network during the

training phase of the neural network development. In this dataset, each image was resized to 299 × 299 pixels to be compatible with the original dimensions of the Inception v3 network architecture. This procedure, known as transfer learning, leverages the natural image features learned by the ImageNet pretrained network. The CNN algorithm was trained using back propagation. All layers of the net-work were fine-tuned using the same global learning rate of 0.001. In addition, Google’s TensorFlow deep learning framework was modified with the prepared dataset to develop an android application for testing of different virus samples. The smartphone application reports the probability value of the tested sample as being positive or negative.Sample preparation and testing using the smartphone systemVirus-spiked samples were prepared by diluting stock virus sample (Materials and Methods) in aliquots of 1× PBS (pH 7.4). A panel of ZIKV-infected patient serum samples (n = 25) was purchased from Boca Biolistics LLC (Pompano Beach, FL, USA). HCV- and HBV- infected patient plasma/serum samples were purchased from Discov-ery Life Sciences Inc. Around 20 l of each sample was loaded into the single-channel microchip and tested following bubble assay pro-tocol either using manual pipetting or the reported sample-processing microfluidic cartridge. After bubble formation, the microchips were imaged using the smartphone system. The performance of the CNN-NES system in testing virus-spiked samples was evaluated compared to the standard RT-PCR technique. For patient samples, the CNN-NES system was evaluated against commercially available assay of Aptima ZIKV Assay (Hologic, Marlborough, MA, USA), ZIKV Detect IgM Capture ELISA (InBios International Inc., Seattle, WA, USA), and Zika MAC-ELISA (CDC, USA) for nucleic acid and immunological testing of ZIKV.

Effect of temperature and humidity on bubble assayThe effect of temperature and humidity conditions on the devel-oped bubble on-chip approach was tested using ZIKV-spiked sam-ples in PBS (pH 7.2) at concentration of 105 particles/ml. Before sample testing, virus sample, functionalized microchip, and all the testing reagents including Pt-nanoprobe solution, washing buffer, and H2O2 solution were equilibrated to specific temperature and humidity. For temperature testing, virus sample was loaded on a sample microchip, and the bubble protocol including virus capture, labeling, and bubble formation was performed at different tempera-tures of 25°, 35°, and 55°C at a relative humidity of 45%. Similarly, virus samples were tested at different relative humidity values of 45, 60, 75, and 95% and a temperature of 25°C for testing the effect of humidity on bubble assay. The tested microchips and the generated bubbles were imaged and evaluated using the CNN-NES.

Shelf life testing of antibody-functionalized microchipsThe stability of antibody-functionalized microchips was evaluated using ZIKV-spiked samples in PBS (pH 7.2) at a virus concentra-tion of 105 particles/ml. Microchips functionalized with anti-ZIKV antibody were treated with 1.5 mM trehalose solution and then freeze-dried using Eyela FD-1000 Freeze Dryer for 12 hours. The dried microchips were packed using a FoodSaver Vacuum Storage unit, divided into four groups (n = 8 chips), and stored on a shelf at room temperature. Each group of stored microchips was tested us-ing the bubble assay for virus detection at different time points of 0, 2, 4, and 6 weeks, and the microchips and the generated bubbles were imaged and evaluated using the CNN-NES.

on May 15, 2021


Dow

nloaded from




10 of 12

Characterization techniques and analysesUV-vis spectroscopyAbsorption spectra were measured on Beckman Coulter DU 800 UV-Vis spectrophotometer. The samples were loaded in 10-mm path length quartz cuvettes and scanned at room temperature.Transmission electron microscopyTEM images were obtained using a JEOL 2100 TEM microscope at an acceleration voltage of 300 kV. The specimens were prepared by dropping 2 l of the sample onto ultrathin Formvar-coated 200-mesh copper grids and then dried in air. The mean diameter and size dis-tribution histogram of particles were obtained by averaging more than 100 particles from the TEM images using ImageJ software.Dynamic light scatteringDLS experiments were performed using Malvern Zetasizer (Malvern Instruments, Malvern, UK). The samples were initially filtered through a 0.22-m filter membrane and allowed to settle overnight for DLS measurements. Ultrapure water (>18 megohms) from a PURELAB Ultra water system (ELGA Ltd) was used as a diluent, and three measurements were recorded for each sample at room temperature. The average values were calculated from the reported results for three different batches.FT-IR spectroscopyFT-IR spectroscopy spectra in the region of 2000 to 500 cm−1 were collected in absorbance mode with a FTS-135 Bio-Rad FT-IR spec-troscopy spectrometer.Enzyme-linked immunosorbent assayWe used ELISA to quantify the captured ZIKV on-chip using a sandwich immunoassay kit (RV-403001-ENV-48 ELISA Kit) that was specifically developed to detect ZIKV envelope protein. ZIKV captured on-chip was eluted from the surface of the chip and digest-ed using the sample preparation buffer provided in the kit, and the virus concentration was estimated using a standard curve prepared from known virus concentrations. An ELISA 96-well plate coated with Anti-Zika envelope Ab was used to capture the target antigen in the samples for 2 hours at 37°C. After incubation, the plate was washed four times using washing buffer, and the plate was dried using paper towels. Diluted biotin detection antibody (100 l) was added and incubated for 60 min. Then, streptavidin–horseradish peroxidase conjugate solution (100 l) was added to each well and incubated for 30 min. The plate was washed four times with wash-ing solution after the incubation. Trimethylboron substrate (100 l) was added and incubated for 15 min. To stop the reaction, 100 l of stop solution was added, tapped, and mixed gently. When the solu-tion in the wells begins to turn yellow, absorbance of the plate was read at 450 nm using a BioTek plate reader.Agarose gel electrophoresisThe electrophoretic mobility pattern of the prepared Pt-nanoprobes compared with nonmodified PtNPs was evaluated with a horizontal submerged gel electrophoresis apparatus (Mini-Sub Cell GT, Bio-Rad) using a 0.7% (w/v) agarose gel in tris-acetate-EDTA buffer (pH 8.5). Ten microliters of sample aliquots were loaded into wells. The gel was subjected to a typical voltage of 100 V for 30 min and imaged with a digital camera.SDS–polyacrylamide gel electrophoresisProtein testing was performed using a Mini-PROTEAN Tetra System and 4 to 20% Mini-PROTEAN TGX Precast Protein Gels (Bio-Rad, Hercules, CA). The samples were digested in a sample preparation buffer [2.5 ml of 1 M tris-HCl (pH 6.8), 0.5 ml of double-distilled water (ddH2O), 1.0 g SDS, 0.8 ml of 0.1% bromophenol blue, 4 ml of

100% glycerol, and 2 ml of 14.3 M -mercaptoethanol completed to 10 ml with ddH2O] and heated for 5 min at 95°C on a heat block. Then, 15 l of a protein standard and 20 l of the samples were loaded on the gel. The voltage was turned up to 90 V, and the elec-trophoresis was continued for 50 min. After electrophoresis was done, the gel was rinsed in water for 3 min and stained in Bio-Safe Coomassie blue stain for about 1 hour. Last, the gel was destained and photographed using a gel documentation system.Statistical analysesStatistical analyses were performed using GraphPad Prism software version 5.01 (GraphPad Software Inc. La Jolla, CA, USA) and Med-Calc 14.8.1. (MedCalc Software bvba, Ostend, Belgium). All viral load values were converted into log10 copies/ml. The mean and SDs were calculated for each data point from at least a total of two to three independent experiments unless indicted in the text. The sta-tistical significance of differences between readings under control and experimental conditions for each bubble assay was assessed using a two-tailed unpaired t test or one-way ANOVA with Kruskal- Wallis test. The level of significance was set at P > 0.05. The strength of the linear relationship was evaluated using Pearson’s correlation, and the regression coefficient was calculated. The degree of agree-ment was assessed by Bland-Altman method. ROC and scattering plots analysis were used to define performance of bubble assay to the standard analytical technique.

SUPPLEMENTARY MATERIALSSupplementary material for this article is available at http://advances.sciencemag.org/cgi/content/full/6/51/eabd5354/DC1

View/request a protocol for this paper from Bio-protocol.

REFERENCES AND NOTES 1. J. E. B. Halliday, K. Hampson, N. Hanley, T. Lembo, J. P. Sharp, D. T. Haydon, S. Cleaveland,

Driving improvements in emerging disease surveillance through locally relevant capacity strengthening. Science 357, 146–148 (2017).

2. H. Heesterbeek, R. M. Anderson, V. Andreasen, S. Bansal, D. de Angelis, C. Dye, K. T. D. Eames, W. J. Edmunds, S. D. W. Frost, S. Funk, T. D. Hollingsworth, T. House, V. Isham, P. Klepac, J. Lessler, J. O. Lloyd-Smith, C. J. E. Metcalf, D. Mollison, L. Pellis, J. R. C. Pulliam, M. G. Roberts, C. Viboud; Isaac Newton Institute IDD Collaboration, Modeling infectious disease dynamics in the complex landscape of global health. Science 347, aaa4339 (2015).

3. C. S. Wood, M. R. Thomas, J. Budd, T. P. Mashamba-Thompson, K. Herbst, D. Pillay, R. W. Peeling, A. M. Johnson, R. A. McKendry, M. M. Stevens, Taking connected mobile-health diagnostics of infectious diseases to the field. Nature 566, 467–474 (2019).

4. J. M. Perkel, Pocket laboratories. Nature 545, 119–121 (2017). 5. M. K. Kanakasabapathy, M. Sadasivam, A. Singh, C. Preston, P. Thirumalaraju,

M. Venkataraman, C. L. Bormann, M. S. Draz, J. C. Petrozza, H. Shafiee, An automated smartphone-based diagnostic assay for point-of-care semen analysis. Sci. Transl. Med. 9, eaai7863 (2017).

6. M. S. Draz, K. M. Kochehbyoki, A. Vasan, D. Battalapalli, A. Sreeram, M. K. Kanakasabapathy, S. Kallakuri, A. Tsibris, D. R. Kuritzkes, H. Shafiee, DNA engineered micromotors powered by metal nanoparticles for motion based cellphone diagnostics. Nat. Commun. 9, 4282 (2018).

7. M. S. Draz, N. K. Lakshminaraasimulu, S. Krishnakumar, D. Battalapalli, A. Vasan, M. K. Kanakasabapathy, A. Sreeram, S. Kallakuri, P. Thirumalaraju, Y. Li, S. Hua, X. G. Yu, D. R. Kuritzkes, H. Shafiee, Motion-based immunological detection of Zika virus using Pt-nanomotors and a cellphone. ACS Nano 12, 5709–5718 (2018).

8. C. D. Chin, T. Laksanasopin, Y. K. Cheung, D. Steinmiller, V. Linder, H. Parsa, J. Wang, H. Moore, R. Rouse, G. Umviligihozo, E. Karita, L. Mwambarangwe, S. L. Braunstein, J. van de Wijgert, R. Sahabo, J. E. Justman, W. El-Sadr, S. K. Sia, Microfluidics-based diagnostics of infectious diseases in the developing world. Nat. Med. 17, 1015–1019 (2011).

9. GSMA, The Mobile Economy (2020); www.gsma.com/mobileeconomy/ [accessed on 5 March 2020].

10. E. E. Ali, L. Chew, K. Y.-L. Yap, Evolution and current status of mhealth research: A systematic review. BMJ Innovations 2, 33–40 (2016).

on May 15, 2021


Dow

nloaded from

http://advances.sciencemag.org/cgi/content/full/6/51/eabd5354/DC1

http://advances.sciencemag.org/cgi/content/full/6/51/eabd5354/DC1

https://en.bio-protocol.org/cjrap.aspx?eid=10.1126/sciadv.abd5354

https://www.gsma.com/mobileeconomy/




11 of 12

11. A. Z. Moshfegh, Nanoparticle catalysts. J. Phys. D Appl. Phys. 42, 233001 (2009). 12. H. Wei, E. Wang, Nanomaterials with enzyme-like characteristics (nanozymes):

Next-generation artificial enzymes. Chem. Soc. Rev. 42, 6060–6093 (2013). 13. I. Willner, B. Willner, Biomolecule-based nanomaterials and nanostructures. Nano Lett. 10,

3805–3815 (2010). 14. A. T. Bell, The impact of nanoscience on heterogeneous catalysis. Science 299, 1688–1691

(2003). 15. A. Roucoux, J. Schulz, H. Patin, Reduced transition metal colloids: A novel family

of reusable catalysts? Chem. Rev. 102, 3757–3778 (2002). 16. N. R. Shiju, V. V. Guliants, Recent developments in catalysis using nanostructured

materials. Appl. Catal. Gen. 356, 1–17 (2009). 17. B. Lim, M. Jiang, P. H. C. Camargo, E. C. Cho, J. Tao, X. Lu, Y. Zhu, Y. Xia, Pd-Pt bimetallic

nanodendrites with high activity for oxygen reduction. Science 324, 1302–1305 (2009). 18. D. Astruc, F. Lu, J. R. Aranzaes, Nanoparticles as recyclable catalysts: The frontier between

homogeneous and heterogeneous catalysis. Angew. Chem. Int. Ed. 44, 7852–7872 (2005). 19. R. Narayanan, M. A. El-Sayed, Shape-dependent catalytic activity of platinum

nanoparticles in colloidal solution. Nano Lett. 4, 1343–1348 (2004). 20. R. Polsky, R. Gill, L. Kaganovsky, I. Willner, Nucleic acid-functionalized Pt nanoparticles:

Catalytic labels for the amplified electrochemical detection of biomolecules. Anal. Chem. 78, 2268–2271 (2006).

21. J. E. Kong, Q. Wei, D. Tseng, J. Zhang, E. Pan, M. Lewinski, O. B. Garner, A. Ozcan, D. Di Carlo, Highly stable and sensitive nucleic acid amplification and cell-phone-based readout. ACS Nano 11, 2934–2943 (2017).

22. A. Romeo, T. S. Leung, S. Sánchez, Smart biosensors for multiplexed and fully integrated point-of-care diagnostics. Lab Chip 16, 1957–1961 (2016).

23. K. E. McCracken, J.-Y. Yoon, Recent approaches for optical smartphone sensing in resource-limited settings: A brief review. Anal. Methods 8, 6591–6601 (2016).

24. J. C. Contreras-Naranjo, Q. Wei, A. Ozcan, Mobile phone-based microscopy, sensing, and diagnostics. IEEE J. Sel. Top. Quantum Electron. 22, 1–14 (2016).

25. O. Mudanyali, E. McLeod, W. Luo, A. Greenbaum, A. F. Coskun, Y. Hennequin, C. P. Allier, A. Ozcan, Wide-field optical detection of nanoparticles using on-chip microscopy and self-assembled nanolenses. Nat. Photonics 7, 247–254 (2013).

26. Q. Wei, W. Luo, S. Chiang, T. Kappel, C. Mejia, D. Tseng, R. Y. L. Chan, E. Yan, H. Qi, F. Shabbir, H. Ozkan, S. Feng, A. Ozcan, Imaging and sizing of single DNA molecules on a mobile phone. ACS Nano 8, 12725–12733 (2014).

27. T. Laksanasopin, T. W. Guo, S. Nayak, A. A. Sridhara, S. Xie, O. O. Olowookere, P. Cadinu, F. Meng, N. H. Chee, J. Kim, C. D. Chin, E. Munyazesa, P. Mugwaneza, A. J. Rai, V. Mugisha, A. R. Castro, D. Steinmiller, V. Linder, J. E. Justman, S. Nsanzimana, S. K. Sia, A smartphone dongle for diagnosis of infectious diseases at the point of care. Sci. Transl. Med. 7, 273re1 (2015).

28. Q. Wei, H. Qi, W. Luo, D. Tseng, S. J. Ki, Z. Wan, Z. Göröcs, L. A. Bentolila, T.-T. Wu, R. Sun, A. Ozcan, Fluorescent imaging of single nanoparticles and viruses on a smart phone. ACS Nano 7, 9147–9155 (2013).

29. H. Chen, Z. Li, L. Zhang, P. Sawaya, J. Shi, P. Wang, Quantitation of femtomolar-level protein biomarkers using a simple microbubbling digital assay and bright-field smartphone imaging. Angew. Chem. Int. Ed. Engl. 58, 13922–13928 (2019).

30. L. Liu, J. Liu, H. Huang, Y. Li, G. Zhao, W. Dou, A quantitative foam immunoassay for detection of Escherichia coli O157:H7 based on bimetallic nanocatalyst-gold platinum. Microchem. J. 148, 702–707 (2019).

31. S. K. Bhardwaj, N. Bhardwaj, G. C. Mohanta, P. Kumar, A. L. Sharma, K.-H. Kim, A. Deep, Immunosensing of atrazine with antibody-functionalized Cu-MOF conducting thin films. ACS Appl. Mater. Interfaces 7, 26124–26130 (2015).

32. V. A. Kostyuchenko, E. X. Y. Lim, S. Zhang, G. Fibriansah, T.-S. Ng, J. S. G. Ooi, J. Shi, S.-M. Lok, Structure of the thermally stable Zika virus. Nature 533, 425–428 (2016).

33. G. G. Daaboul, D. S. Freedman, S. M. Scherr, E. Carter, A. Rosca, D. Bernstein, C. E. Mire, K. N. Agans, T. Hoenen, T. W. Geisbert, M. S. Ünlü, J. H. Connor, Enhanced light microscopy visualization of virus particles from Zika virus to filamentous ebolaviruses. PLOS ONE 12, e0179728 (2017).

34. D. Sirohi, Z. Chen, L. Sun, T. Klose, T. C. Pierson, M. G. Rossmann, R. J. Kuhn, The 3.8 Å resolution cryo-EM structure of Zika virus. Science 352, 467–470 (2016).

35. E. Ivanova Reipold, P. Easterbrook, A. Trianni, N. Panneer, D. Krakower, S. Ongarello, T. Roberts, V. Miller, C. Denkinger, Optimising diagnosis of viraemic hepatitis C infection: The development of a target product profile. BMC Infect. Dis. 17 (Suppl. 1), 707 (2017).

36. B. Lange, T. Roberts, J. Cohn, J. Greenman, J. Camp, A. Ishizaki, L. Messac, E. Tuaillon, P. van de Perre, C. Pichler, C. M. Denkinger, P. Easterbrook, Diagnostic accuracy of detection and quantification of HBV-DNA and HCV-RNA using dried blood spot (DBS) samples–a systematic review and meta-analysis. BMC Infect. Dis. 17 (Suppl. 1), 693 (2017).

37. WHO, Guidelines for the Prevention, Care and Treatment of Persons with Chronic Hepatitis B Infection (2015); https://apps.who.int/iris/bitstream/handle/10665/154590/9789241549059_eng.pdf?sequence=1 [accessed on 10 May 2020].

38. Quest Diagnostics Infectious Disease Inc, Zika Virus RNA Qualitative Real-Time RT-PCR (2017); https://www.fda.gov/media/96701/download.

39. X. Fan, D. Fang, D. Bin, L. Xueyong, C. Jun, W. Hongshan, Serum levels of surface large envelope proteins: Prognostic markers for hepatitis B e antigen-negative patients with adefovir dipivoxil treatment. Antivir. Ther. 14, 1149–1156 (2009).

40. J. Vermehren, B. Maasoumy, R. Maan, G. Cloherty, C. Berkowski, J. J. Feld, M. Cornberg, J.-M. Pawlotsky, S. Zeuzem, M. P. Manns, C. Sarrazin, H. Wedemeyer, Applicability of hepatitis C virus RNA viral load thresholds for 8-week treatments in patients with chronic hepatitis C virus genotype 1 infection. Clin. Infect. Dis. 62, 1228–1234 (2016).

41. K. Gourlain, A. Soulier, B. Pellegrin, M. Bouvier-Alias, C. Hézode, F. Darthuy, J. Rémiré, J.-M. Pawlotsky; the DITTO Group, Dynamic range of hepatitis C virus RNA quantification with the Cobas Ampliprep-Cobas Amplicor HCV Monitor v2. 0 assay. J. Clin. Microbiol. 43, 1669–1673 (2005).

42. V. M. Corman, A. Rasche, C. Baronti, S. Aldabbagh, D. Cadar, C. B. E. M. Reusken, S. D. Pas, A. Goorhuis, J. Schinkel, R. Molenkamp, B. M. Kümmerer, T. Bleicker, S. Brünink, M. Eschbach-Bludau, A. M. Eis-Hübinger, M. P. Koopmans, J. Schmidt-Chanasit, M. P. Grobusch, X. de Lamballerie, C. Drosten, J. F. Drexler, Assay optimization for molecular detection of Zika virus. Bull. World Health Organ. 94, 880–892 (2016).

43. C.-W. Chen, M. Akashi, Synthesis, characterization, and catalytic properties of colloidal platinum nanoparticles protected by poly (N-isopropylacrylamide). Langmuir 13, 6465–6472 (1997).

44. Y. Ju, J. Kim, Dendrimer-encapsulated Pt nanoparticles with peroxidase-mimetic activity as biocatalytic labels for sensitive colorimetric analyses. Chem. Commun. (Camb) 51, 13752–13755 (2015).

45. S. Mostafa, F. Behafarid, J. R. Croy, L. K. Ono, L. Li, J. C. Yang, A. I. Frenkel, B. R. Cuenya, Shape-dependent catalytic properties of Pt nanoparticles. J. Am. Chem. Soc. 132, 15714–15719 (2010).

46. J. Tan, S. Shah, A. Thomas, H. D. Ou-Yang, Y. Liu, The influence of size, shape and vessel geometry on nanoparticle distribution. Microfluid Nanofluidics 14, 77–87 (2013).

47. K. L. Prime, G. M. Whitesides, Self-assembled organic monolayers: Model systems for studying adsorption of proteins at surfaces. Science 252, 1164–1167 (1991).

48. M. Zheng, F. Davidson, X. Huang, Ethylene glycol monolayer protected nanoparticles for eliminating nonspecific binding with biological molecules. J. Am. Chem. Soc. 125, 7790–7791 (2003).

49. F. M. Veronese, J. M. Harris, Introduction and overview of peptide and protein pegylation. Adv. Drug Deliv. Rev. 54, 453–456 (2002).

50. P. B. Lawrence, J. L. Price, How PEGylation influences protein conformational stability. Curr. Opin. Chem. Biol. 34, 88–94 (2016).

51. S. M. Chamow, T. P. Kogan, D. H. Peers, R. C. Hastings, R. A. Byrn, A. Ashkenazi, Conjugation of soluble CD4 without loss of biological activity via a novel carbohydrate-directed cross-linking reagent. J. Biol. Chem. 267, 15916–15922 (1992).

52. M. Shi, J. H. Wosnick, K. Ho, A. Keating, M. S. Shoichet, Immuno-polymeric nanoparticles by diels–alder chemistry. Angew. Chem. Int. Ed. 46, 6126–6131 (2007).

53. L. Srisombat, A. C. Jamison, T. R. Lee, Stability: A key issue for self-assembled monolayers on gold as thin-film coatings and nanoparticle protectants. Colloids Surf. A Physicochem. Eng. Asp. 390, 1–19 (2011).

54. X. Duan, S. Li, J. A. Holmes, Z. Tu, Y. Li, D. Cai, X. Liu, W. Li, C. Yang, B. Jiao, E. A. Schaefer, D. N. Fusco, S. Salloum, L. Chen, W. Lin, R. T. Chung, MicroRNA 130a regulates both hepatitis C virus and hepatitis B virus replication through a central metabolic pathway. J. Virol. 92, e02009-17 (2018).

55. T. Severi, C. Ying, J. R. Vermeesch, D. Cassiman, L. Cnops, C. Verslype, J. Fevery, L. Arckens, J. Neyts, J. F. van Pelt, Hepatitis B virus replication causes oxidative stress in HepAD38 liver cells. Mol. Cell. Biochem. 290, 79–85 (2006).

56. N. C. Bigall, T. Härtling, M. Klose, P. Simon, L. M. Eng, A. Eychmüller, Monodisperse platinum nanospheres with adjustable diameters from 10 to 100 nm: Synthesis and distinct optical properties. Nano Lett. 8, 4588–4592 (2008).

Acknowledgments: We would like to thank A. Shokr, S. Hua, X. Mou, F. Giguel, K. M. Kochbyoki, S. Kallakuri, D. Battalapalli, N. K. Lakshminaraasimulu, A. Srivaths, A. Addula, S. Swaminathan, Y. Li, and P. Ambastha. Funding: Research reported in this publication was partially supported by the National Institutes of Health under award nos. R01AI118502, R01AI138800, and R61AI140489 and the Harvard University Center for AIDS Research (CFAR) under award no. 5P30AI060354-14. Author contributions: M.S.D. and H.S. conceived the method and planned the experiments. M.S.D., A.M., A.S., A.V., V.Y., and S.K. performed the experiments. M.K.K., P.T., and H.S. developed the concept of using CNNs for on-chip bubble signal detection. M.K.K. and P.T. developed the algorithm for CNN-NES. P.T. aided in the CNN algorithm testing. M.S.D. and A.V. designed and fabricated the microfluidic cartridge. M.S.D., S.K., A.M., A.V., P.T., and M.K.K. analyzed the data. R.T.C. provided the HBV samples. W.L. helped in preparing the HBV-cultured stock samples. X.G.Y. provided the ZIKV samples. M.S.D. and H.S. wrote the manuscript. All authors have given approval to the final version of the manuscript. Competing interests: H.S. and M.S.D. are inventors on a provisional patent related to this work filed by Brigham and Women’s Hospital (no. BWH 2020-664, filed 10 September 2020). The authors declare that they

on May 15, 2021


Dow

nloaded from

https://apps.who.int/iris/bitstream/handle/10665/154590/9789241549059_eng.pdf?sequence=1

https://apps.who.int/iris/bitstream/handle/10665/154590/9789241549059_eng.pdf?sequence=1

https://www.fda.gov/media/96701/download




12 of 12

have no other competing interests. Data materials and availability: All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Materials. Deidentified patient data used in this study may be available through the authors upon request. All reasonable requests for materials sharing will be considered. The patient samples were purchased from commercial vendors, and the exact same patient sample materials may not be available for sharing by the authors or by the commercial vendors because of the limited sample volume available from each participant. The codes and algorithms developed for this study may be obtained by data transfer agreement with Partners HealthCare and Brigham and Women’s Hospital upon request.

Submitted 26 June 2020Accepted 2 November 2020Published 16 December 202010.1126/sciadv.abd5354

Citation: M. S. Draz, A. Vasan, A. Muthupandian, M. K. Kanakasabapathy, P. Thirumalaraju, A. Sreeram, S. Krishnakumar, V. Yogesh, W. Lin, X. G. Yu, R. T. Chung, H. Shafiee, Virus detection using nanoparticles and deep neural network–enabled smartphone system. Sci. Adv. 6, eabd5354 (2020).

on May 15, 2021


Dow

nloaded from


enabled smartphone system−Virus detection using nanoparticles and deep neural network

Sreeram, Sanchana Krishnakumar, Vinish Yogesh, Wenyu Lin, Xu G. Yu, Raymond T. Chung and Hadi ShafieeMohamed S. Draz, Anish Vasan, Aradana Muthupandian, Manoj Kumar Kanakasabapathy, Prudhvi Thirumalaraju, Aparna

DOI: 10.1126/sciadv.abd5354 (51), eabd5354.6Sci Adv

ARTICLE TOOLS http://advances.sciencemag.org/content/6/51/eabd5354

MATERIALSSUPPLEMENTARY http://advances.sciencemag.org/content/suppl/2020/12/14/6.51.eabd5354.DC1

REFERENCES

http://advances.sciencemag.org/content/6/51/eabd5354#BIBLThis article cites 53 articles, 12 of which you can access for free

PERMISSIONS http://www.sciencemag.org/help/reprints-and-permissions

Terms of ServiceUse of this article is subject to the

is a registered trademark of AAAS.Science AdvancesYork Avenue NW, Washington, DC 20005. The title (ISSN 2375-2548) is published by the American Association for the Advancement of Science, 1200 NewScience Advances

License 4.0 (CC BY-NC).Science. No claim to original U.S. Government Works. Distributed under a Creative Commons Attribution NonCommercial Copyright © 2020 The Authors, some rights reserved; exclusive licensee American Association for the Advancement of

on May 15, 2021


Dow

nloaded from

http://advances.sciencemag.org/content/6/51/eabd5354

http://advances.sciencemag.org/content/suppl/2020/12/14/6.51.eabd5354.DC1

http://advances.sciencemag.org/content/6/51/eabd5354#BIBL

http://www.sciencemag.org/help/reprints-and-permissions

http://www.sciencemag.org/about/terms-service


applied sciences and engineering copyright © 2020 virus … · draz et al., sci. dv. 2020 6 :...

Documents