Full Paper www.small-journal.com Rapid and Digital Detection of Inflammatory Biomarkers Enabled by a Novel Portable Nanoplasmonic Imager Alexander Belushkin, Filiz Yesilkoy, Juan Jose González-López, Juan Carlos Ruiz-Rodríguez, Ricard Ferrer, Anna Fàbrega,* and Hatice Altug* 1. Introduction New point-of-care diagnostic devices are urgently needed for rapid and accurate diagnosis, particularly in the management of life-threatening Sepsis is a life-threatening condition infections and sepsis, where immediate treatment is key. Sepsis is a critical where the body’s own response to an infec- condition caused by systemic response to infection, with chances of survival tion progressively leads to organ system drastically decreasing every hour. A novel portable biosensor based on dysfunction, failure and finally death.[1] Although the ongoing advancements in nanoparticle-enhanced digital plasmonic imaging is reported for rapid modern medicine have increased the sur- and sensitive detection of two sepsis-related inflammatory biomarkers, vival rates of organ recipients and patients procalcitonin (PCT) and C-reactive protein (CRP) directly from blood serum. with serious illnesses, such as cancer and The device achieves outstanding limit of detection of 21.3 pg mL−1 for PCT and HIV, the occurrence of sepsis has been 36 pg mL−1 for CRP, and dynamic range of at least three orders of magnitude. increasing globally, imposing a significant The portable device is deployed at Vall d’Hebron University Hospital in Spain challenge.[2] Moreover, the emergence of drug-resistant pathogens is contributing and tested with a wide range of patient samples with sepsis, noninfectious significantly to the global sepsis threat.[3] systemic inflammatory response syndrome (SIRS), and healthy subjects. Current reports indicate that sepsis causes The results are validated against ultimate clinical diagnosis and currently mortality reaching up to 50% and affects used immunoassays, and show that the device provides accurate and robust more than 30 million people every year performance equivalent to gold-standard laboratory tests. Importantly, the worldwide.[2] In the United States alone, $24 billion is spent every year for the man- plasmonic imager can enable identification of PCT levels typical of sepsis agement of sepsis, making it the most and SIRS patients in less than 15 min. The compact and low-cost device is a expensive condition to treat.[4] Sepsis pro- promising solution for assisting rapid and accurate on-site sepsis diagnosis. gresses extremely rapidly and is consid- ered a medical emergency. In fact, sepsis progresses so fast that every hour of delay A. Belushkin, Dr. F. Yesilkoy, Prof. H. Altug in diagnosis and initiating appropriate treatment increases Institute of BioEngineering École Polytechnique Fédérale de Lausanne the mortality by 7.6%.[5] Therefore, development of point-of- CH-1015 Lausanne, Switzerland care devices enabling early and accurate diagnosis of sepsis at E-mail: hatice.altug@epfl.ch different clinical settings is an urgent need as they can allow Dr. J. J. González-López, Dr. A. Fàbrega timely intervention and lower the mortality rate.[6–8] Department of Clinical Microbiology However, sepsis diagnosis is not easy, especially in the early Vall d’Hebron University Hospital Vall d’Hebron Institut de Recerca stages, or when sepsis presents with an atypical clinical picture. Universitat Autònoma de Barcelona Early stages of sepsis usually present stealth symptoms with 08035 Barcelona, Spain significant patient-to-patient heterogeneity, leading to potential E-mail: anna.fabrega@vhir.org delays from the onset of sepsis until the most appropriate treat- Dr. J. C. Ruiz-Rodríguez, Dr. R. Ferrer ment starts.[9,10] The dysregulated host response to infection, Intensive Care Department which is typically present in sepsis patients, is primarily sup- Vall d’Hebron University Hospital, Shock Organ Dysfunction and Resuscitation Research Group ported by an inflammatory process, which is termed SIRS, for Vall d’Hebron Institut de Recerca systemic inflammatory response syndrome. However, it is chal- 08035 Barcelona, Spain lenging to discriminate whether the biological insult leading to The ORCID identification number(s) for the author(s) of this article SIRS is an infectious process or not, an essential criteria for the can be found under https://doi.org/10.1002/smll.201906108. antimicrobial treatment to be worthy and effective.[1] Tradition- © 2019 The Authors. Published by WILEY-VCH Verlag GmbH & Co. ally, for the microbiological diagnosis of sepsis, blood-culture KGaA, Weinheim. This is an open access article under the terms of the based pathogen investigation is performed, which typically Creative Commons Attribution-NonCommercial-NoDerivs License, which permits use and distribution in any medium, provided the original takes 4–5 days.[9] The current recommendation is to administer work is properly cited, the use is non-commercial and no modifications broad-spectrum antibiotics as soon as possible, ideally in the or adaptations are made. first hour after sepsis recognition, before a definitive microbio- The copyright line for this article was changed on 26 December 2019 logical diagnosis is made based on the pathogen identification after original online publication. tests.[11] This approach leads to overuse of antibiotics, which DOI: 10.1002/smll.201906108 worsens the global health challenge of antibiotic resistant Small 2020, 16, 1906108 1906108 (1 of 11) © 2019 The Authors. Published by WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.advancedsciencenews.com www.small-journal.com bacteria development, because neither all SIRS patients have an media. The bioassay is performed in a single step without signal underlying infectious process, nor all infections are caused by amplification or washing procedures, and the plasmonic detec- bacterial pathogens. In contrast, a new clinical trend in sepsis tion is robust against variations in optical properties of samples. management relies on blood testing to monitor patient’s sys- Importantly, the imager is made of low-cost off-the-shelf optical temic response for rapid patient stratification. These tests, components, and its small size enables deployment and opera- which quantify circulating biomarkers, such as the two most tion in on-site clinical settings. Moreover, the sensor chips are universal sepsis-related biomarkers procalcitonin (PCT) and produced in a scalable cost-effective manufacturing and the bio- C-reactive protein (CRP), help diagnose sepsis timely and assay reagents can be stored stably over weeks. We first show reduce overprescription of antibiotics. Although these inflam- the application of DENIS for highly sensitive, quantitative and matory biomarkers alone cannot be the sole basis for accurate robust detection of PCT and CRP biomarkers. The system sepsis diagnosis, their blood levels provide valuable quantita- enables ultrahigh detection sensitivities with limit of detection tive information to triage patients based on their anomalous (LOD) down to 21 and 36 pg mL−1 for PCT and CRP, respectively, response allowing for personalized treatment.[12–18] The current and a wide dynamic range of at least three orders of magnitude. gold standard clinical tools to test blood circulating proteins are We evaluate the performance of DENIS in clinical settings by largely based on complex and lengthy immunoassays, such as testing the samples from sepsis, noninfectious SIRS (ni-SIRS), enzyme-linked immunosorbent, immunoturbidimetric (ITA), and healthy subjects provided by Vall d’Hebron Sepsis Bank. We and chemiluminescent (CLIA) immunoassays, which require benchmarked our results to the clinically validated assays and fully equipped clinical laboratories. However, having rapid the diagnoses of patients and in agreement, we observed that access to the patient’s systemic response, using portable, low- rapid and quantitative detection of PCT and CRP from blood cost, and easy-to-use devices that can rapidly detect biomarkers serum of the selected 34 subject cohort assisted in discrimina- in a simple manner from small sample volumes can drastically tion of healthy and sick groups. Moreover, PCT detection with improve sepsis care at different clinical settings such as emer- DENIS show performance equivalent to gold standard clinical gency care units, ambulances, and outpatient health centers. assays in noninfectious SIRS and sepsis patients. Finally, DENIS Currently, active ongoing efforts are being undertaken both can perform rapid detection by recording video of the plasmonic in the biomedical industry and basic research to find new diag- sensor and analyzing the signals in real time. The time-resolved nostic solutions and develop point-of-care devices to rapidly results revealed that we can identify samples with high (sepsis) diagnose sepsis on-site in a cost-effective way. For instance, a and medium (noninfectious SIRS) PCT levels in 5 and 10 min, compact device proposed by Samsung (LABGEO IB10) operates respectively, reducing the test time to under 15 min. by detecting PCT with a centrifugal microfluidic disk. Another recent technology that is being investigated by Abionic for sepsis detection combines nanofluidic channels with centrifugal disk. 2. Results Although nanosized channels can improve the sensitivity they are prone to clogging. Moreover, these technologies employ mechani- 2.1. Detection Mechanism cally moving parts that can compromise device robustness over time. So far, no portable system has received US Food and Drug The DENIS detection mechanism is based on a large-area Administration approval for measuring biomarkers that effec- Au-NHA device supporting SPR and localized surface plasmon tively predict sepsis at the point of care. Research efforts have modes.[28–31] Au-NHAs exhibit extraordinary optical transmis- been made to develop devices for sepsis biomarkers detection sion (EOT) at its resonance wavelength, controlled by the period based on optical schemes, including plasmonics and fluorescence. and geometry of the nanoholes, when illuminated in a collinear Surface plasmon resonance (SPR) and surface-enhanced Raman optical path.[32] This plasmonic resonance is locally distorted scattering sensors were reported, however they could not achieve upon the binding of single Au-NPs (100 nm diameter). These clinically relevant sensitivity.[19,20] Fluorescent schemes have been distortions result in local suppression in the EOT, which ena- implemented to detect sepsis markers, however they either rely on bles detection of single NP-labeled molecules.[27] Although the expensive and bulky optical setups, such as total internal reflec- NPs are subwavelength in size, their plasmonic interaction tion microscopy,[21] or do not meet clinical sensitivity.[22,23] Prom- with the Au-NHA allows for high-contrast imaging of surface- ising recent approaches use microfluidic extraction of cells from bound NPs in a simple bright field optical reader over large small amount of patient blood to quantify expression of CD64 field of view directly inside complex samples. The imaged data biomarker, however they employ sophisticated design and control is not affected by the bulk sample background because the reso- of microfluidic chip.[24,25] Alternatively, a sensitive system based nance transmission through Au-NHAs is strictly dependent on on magnetic beads and fluorescent enzymatic amplification has the surface localized field with ≈100 nm decay length. More- been proposed. The system provides a low limit of detection for over, the contrast generated by a single particle is significantly measuring a sepsis-related biomarker IL-3 but at the expense of above the noise level induced by the camera and the sample, multiple assay steps and 1 h long detection time.[26] therefore individual molecules binding can be robustly quanti- Here, we report a portable digital nanoparticle-enhanced plas- fied in a digital way directly in complex media (Figure S1, Sup- monic imager for rapid detection of inflammatory biomarkers porting Information). In contrast, more conventional affinity (DENIS). The unique nanoplasmonic imaging mechanism is sensors average the readout over sensor areas, and require a based on gold nanoparticle (Au-NP) binding to plasmonic gold certain density of analyte binding to reach a sufficient signal nanohole array (Au-NHA),[27] which enables quantification of to be detected over the system noise, such as background fluc- individual molecule binding on the sensor surface in complex tuations or reader noise. The digitized detection of individual Small 2020, 16, 1906108 1906108 (2 of 11) © 2019 The Authors. Published by WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.advancedsciencenews.com www.small-journal.com Figure 1. Portable digital nanoparticle-enhanced plasmonic imager for biomarkers detection. a) PCT and CRP, which are blood-circulating protein biomarkers secreted by the host body in response to systemic inflammation, are detected using DENIS. A single-step bioassay directly in human serum enables rapid molecular results, critical for the early diagnosis of sepsis, by detecting individual Au-NPs binding to Au-NHA. b) A prototype DENIS reader developed for highly sensitive and multiplexed detection of biomarkers. The device uses a CMOS camera and a narrow-band LED source to record the transmitted images from a nanoplasmonic chip. c) SEM image of a Au-NHA area after a bioassay showing the bound NPs. Inset shows a single nanoparticle bound inside a nanohole. d) Plasmonic image of a Au-NHA area with bound nanoparticles. The binding of Au-NPs on Au-NHAs causes local transmission suppression through distortion of plasmonic excitations in the Au-NHA and can be digitally detected using far-field imaging. The inset shows a normalized intensity contrast induced by a single nanoparticle trapped in a nanohole. e) To enable microarray based biosensing, capture antibodies are immobilized on the Au-NHA sensor surface using noncontact ultralow volume robotic liquid dispenser. The image shows nine sensor chips with spotted microarrays under dispensing nozzle. f) Photograph of a Au-NHA plasmonic chip. The sensors are fabricated using wafer-scale DUVL lithography. g) Image of a plasmonic microarray field of view with antibody spots, and labeled with QR codes. particles with DENIS enables highly sensitive quantification beam etching (Figure 1f). The nanofabrication process enables of low analyte concentrations at the diffusion limit, and statis- production of over 50 highly uniformly nanostructured chips tical assessment of the signal.[27] Importantly, single NPs can (1 cm × 1 cm) per 4 in. wafer and eight wafers per batch with be distinguished on plasmonic images from NPs aggregates negligible variation in optical properties.[33] This robustness and sediment by the size and the shape of the plasmonic image in the manufacturing of nanostructured sensors is critical to spikes.[27] enable the transfer of the technology to a clinical diagnostic tool. We postpatterned the fabricated chips with an array of quick response (QR) codes (Figure 1g; Figure S2, Supporting 2.2. Plasmonic Microarray Sensors Information). In addition to tagging individual chips and microarray locations on the chips, QR codes encode the infor- We use the detection technique to quantify PCT and CRP bio- mation about the fabrication batch parameters and the wafer markers from human serum samples. In the presence of the enabling a registry record of the manufactured sensors. biomarker in the sample, antibody-conjugated Au-NPs bind to the Au-NHA surface functionalized with complementary anti- bodies (Figure 1a) and can be detected using a portable reader 2.3. Portable Optical Reader (Figure 1b). Individual Au-NPs bound inside or close to the nanoholes (Figure 1c) create strong local intensity contrast The nanoplasmonic imager is made of off-the-shelf optical (Figure 1d), which corresponds to digital detection of single components and comprises a narrowband light emitting diode analyte molecules. (LED) source, spectrally matching the EOT wavelength of Au- To enable microarray based multiplexed biosensing, capture NHAs, a custom-built aluminum holder for the nanoplasmonic antibodies are bioprinted on the sensor surface using a high- chip, an objective, and a CMOS camera. The collimated LED throughput, noncontact, low-volume liquid dispenser as shown light is transmitted through the sample chamber and modu- in Figure 1e. The Au-NHA sensors are manufactured using lated by the Au-NHA sensor. The sensor images are acquired by a low-cost wafer-scale deep UV lithography (DUVL) and ion a 50× objective and CMOS sensor with 4104 × 3006 pixels and Small 2020, 16, 1906108 1906108 (3 of 11) © 2019 The Authors. Published by WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.advancedsciencenews.com www.small-journal.com measurement chamber (Figure 2). We pre- pare large batches of Ab-NPs with optical density OD = 20 and store them in phos- phate buffer saline (PBS 1×) buffer with BSA (1% w/v) and Tween20 (0.05 % w/v). The Ab-NP suspension is stable at 4 °C for at least 5 weeks, over which the tests were performed. The sensor chips are first uniformly function- alized with copoly-DMA-MAPS-NAS-fluori- nated polymer (MCP-2F), containing activated amine reactive NHS-ester groups, to ensure stable antibody immobilization. Moreover, the fluorocarbon polymer helps prevent non- specific NP binding, and fouling by serum proteins.[34] To form antibody microarrays, we spot antibodies specific to the biomarkers as well as mouse isotype immunoglobulin-G (IgG) for nonspecific negative control both at 200 µg mL−1. Each microarray spot of ≈150 µm diameter is formed by a single 400 pL droplet dispensed with 400 µm period. Importantly, the use of microarray printer not only enables functionalization of large number of chips, but also minimizes the amount of antibodies Figure 2. Detection and quantification of PCT and CRP using DENIS. A serum sample (I) is mixed with Au-NPs functionalized with antibodies against PCT or CRP. Ab-NPs bind to the used, which are one of the most expensive antigen in serum (II). Sample with Ab-NPs is injected on Au-NHA sensor chip functionalized components of the assay. The uniform anti- with polymer and spots of complementary aPCT or aCRP antibodies as well as nonspecific body immobilization on the sensor surface is mouse IgG as negative control. Ab-NPs bind on the Au-NHA chip functionalized with comple- ensured by adding trehalose (0.5% w/v) and mentary antibody spots in a sandwich assay in the presence of the antigen (III). Tween20 (0.01% w/v) in the PBS 1× spotting buffer. The sensor chips with IgG microar- 3.45 µm pixel size. Each image shows a 900 µm × 700 µm field rays are blocked with BSA (1% w/v) to minimize the nonspecific of view (FOV) with ≈1.2 µm resolution (Figure 1g). The reader binding. A capillarity-based disposable microfluidic platform is weight is below 1 kg and the dimensions are 10 × 10 × 35 cm. sealed on the sensor chips using a silicone spacer.[33] The anti- Such compactness means it is easily transportable and can be body immobilized chips can be stored at 4 °C stably for at least deployed in most on-site settings. 5 weeks (Figure S6, Supporting Information). To increase the per- The control of the camera, image acquisition, and data pro- formance of our assay, we optimized a stabilizing buffer, which cessing are performed on a laptop computer using custom consists of PBS Tween20 (PBST, 1×, 0.05% w/v) and NaOH Matlab functions. In brief, images are recorded with custom auto- (50 × 10−3 m) in 3:1 ratio. To perform the bioassay, 20 µL of exposure to ensure optimal intensity profile, corrected for source serum is mixed with 4 µL of Ab-NP suspension and 16 µL of sta- illumination, and the NPs binding is quantified over the sensing bilizing buffer, and the mixture is loaded into the measurement spots by computing the percentage of dark pixels using a cutoff chamber with a pipet. The loaded cartridge is either inserted threshold. QR tags from the images can be automatically recog- into the reader for time-resolved data acquisition or measured nized by custom Matlab function to register individual FOVs. For after an incubation period. end-point measurements, signals from at least five microarray To characterize the performance of DENIS system, calibra- areas from each sensor chip are recorder for statistical signifi- tion titration curves were obtained by spiking known amounts cance. In the case of time-resolved measurements, images of a of PCT and CRP biomarkers in reference samples (Figure 3). single FOV are recorded with 30 s intervals, and the signals are Each data point represents mean value and standard deviation extracted from at least three microarray spots to estimate mean from five measurements collected from different microarray and standard deviation. In order to control for nonspecific NPs areas. The LOD values were estimated by measuring the mean binding and background fluctuations, in each measurement the signal from blank samples and adding three times the standard signals from bovine serum albumin (BSA) blocked background deviation. Our detection method achieves LOD of 21 and are subtracted from the signals of Ab microarray spots. 36 pg mL−1 for PCT and CRP, respectively. Moreover, the PCT detection dynamic range of DENIS is three orders of magni- tude (i.e., from ≈21 to >1E4 pg mL−1). The clinical serum levels 2.4. Bioassay of PCT usually range from tens of pg mL−1 for healthy individ- uals to few hundreds pg mL−1 for SIRS patients free of bacterial Our simple and rapid bioassay is performed in a single step, infection, and above ≈1 ng mL−1 in sepsis patients.[35] Therefore, where antibody functionalized Au-NP (Ab-NP) suspension our method covers the full relevant diagnostic range. The clin- is mixed directly with blood serum and then injected into the ical serum levels for CRP range from few µg mL−1 for healthy Small 2020, 16, 1906108 1906108 (4 of 11) © 2019 The Authors. Published by WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.advancedsciencenews.com www.small-journal.com DENIS to the CLIA measurements in the clinical lab (inset Figure 4b). In every sample measured, we controlled for non- specific interactions between Ab-NPs and the Au-NHA surface by recording signals from mouse IgG spots and BSA blocked areas (Figure 5). Notably, we observed no significant signals from these control measurements on every chip tested, there- fore no washing steps were needed. To validate the performance of the DENIS platform, we mapped the PCT and CRP detection results of 34 individuals to the measurements with CLIA and ITA, based on the clinical Figure 3. Determining the detection sensitivity for PCT and CRP bio- diagnosis (see Figure 6a,b and Figure S3 in the Supporting markers with the DENIS system. The calibration curves were obtained by titrating a known concentration of a) PCT or b) CRP and measuring the Information). In the sampled group, the measured levels associated signal. of PCT and CRP using both DENIS and clinical assays were significantly elevated in noninfectious SIRS group compared individuals to hundreds of µg mL−1 for septic patients,[35] which to healthy subjects (PPCT DENIS < 1.7E−5, PCRP DENIS < 1.7E−3, are significantly higher than the values for PCT. To measure unpaired t-test), and in sepsis group compared to noninfectious them with the same assay, we diluted serum for CRP measure- SIRS (PPCT DENIS < 2.3E−7, PCRP DENIS < 4.9E−3, unpaired t-test). ments in order to comply with the dynamic range of the device, However, CRP was reported in multiple sources to be a gen- which covers from 36 to 1E5 pg mL−1. Importantly, the sensi- eral inflammatory marker that is not well discriminative of the tivity and dynamic range enabled by DENIS are similar to the stage or the source of inflammatory host response.[35,36] There- performance of gold standard lab techniques, without the need fore, in order to compare the performance of DENIS to clini- for time-consuming amplification and washing steps or bulky cally validated assays in assisting the discrimination between readout equipment. sepsis, noninfectious SIRS, and healthy groups, we constructed receiver operating characteristic (ROC) curves using PCT meas- urements with both techniques as a predictor (Figure 6c,d). 2.5. Clinical Tests of the DENIS System We also calculated the corresponding area under the curve (AUC). ROC curves are an established metric to analyze We deployed the DENIS system at the Vall d’Hebron University Hospital and tested with clinical samples previously collected from patients with sepsis, noninfectious SIRS and healthy individuals (previous informed consent agreement). Even though the new sepsis definitions prefer to avoid the term SIRS, this clinical condition is still worthy when defining the appropriate study groups with research purposes.[1] We obtained and measured on the DENIS platform samples, previ- ously anonymized, from 34 individuals, including 18 patients with sepsis, 11 with noninfectious SIRS, and 5 healthy donors (Table S1, Supporting Information). The detection of PCT and CRP levels in the patient serum was performed as described above using only 20 µL of sample for each measurement. The measurements were done in a blind manner, where nei- ther the patient status nor the biomarker concentrations were disclosed during the testing in order to eliminate bias. After the completion of the clinical tests, the results collected from DENIS assay were compared with the routine clinical determi- nations. Specifically, CRP and PCT detection were performed using ITA and CLIA, respectively, and bacterial infection was determined with blood culture. The example images of single PCT spots obtained by DENIS system directly in serum from the three patient groups are presented in Figure 4a. Samples from septic patients contain high levels of PCT or CRP, and result in dense binding of particles on antibody microspots of Figure 4. a) Different concentration of PCT in serum results in different the Au-NHA sensor, while noninfectious SIRS samples pro- density of particle binding. From left to right: representative microspots duce images with sparsely bound NPs, and healthy samples imaged from clinical samples associated to a sepsis patient, a noninfec- present negligible NP binding on the sensing areas. Figure 4b tious SIRS patient, and a healthy donor are shown. b) Quantification of PCT and CRP biomarkers using DENIS in the serum of typical representa- shows CRP and PCT values measured from nine representa- tive patients with sepsis, noninfectious SIRS and healthy controls. Error tive individuals from the three groups. Each data point repre- bars represent standard deviation between signals of five spots measured sents the mean and STD from at least five different spots. We for each sample. Inset shows correlation between DENIS and CLIA meas- observed consistent results correlating the PCT readouts from urements of PCT levels for the typical patients. Small 2020, 16, 1906108 1906108 (5 of 11) © 2019 The Authors. Published by WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.advancedsciencenews.com www.small-journal.com Figure 5. Quantification of plasmonic imaging signals for a) PCT and b) CRP, as well as from nonspecific mouse isotype control antibody spots, for every patient. The two biomarkers were measured for 34 subjects, including 18 sepsis patients, 11 noninfectious SIRS patients, and 5 healthy controls. The signal from mouse antibodies was small for all patients, indicating little nonspecific interactions between Ab-NPs and Au-NHA sensor. Signals plotted as difference between signal from Ab spots and BSA background. For each point, at least five different antibody spots were measured to estimate the mean and the standard deviation of the signal. diagnostic ability of a test to discriminate the true state of sub- with a severe disease progression within hours. We show the jects, and to compare two alternative diagnostic tests when each diagnostic use of DENIS with minimal time-to-result enabled is performed on the same subject.[37] DENIS measurements by time-resolved measurements. The system can record real- of PCT provided equivalent performance to CLIA in discrimi- time video of the sensor (Video S5, Supporting Information), nating between the three study groups. With both methods, from which NP binding signals are extracted and analyzed in we observed 100% sensitivity and specificity in classifying real time. In Figure 7, we present time-resolved PCT detec- between healthy and sick subjects, moreover PCT was a good tion signals from sera of representative sepsis patients, non- indicator of host response in sepsis compared to noninfectious infectious SIRS patients, and healthy individuals. Each plot SIRS, with AUC = 0.995, sensitivity = 0.944, and specificity = 1. shows mean signal from three PCT sensing spots and its 95% Although the patient cohort of 34 subjects is not large enough confidence interval in the shaded area. To establish the back- to reflect the general population statistics, these findings are in ground signal and its variation, we plot averaged signal from agreement with the literature, where PCT levels in blood were three BSA blocked areas and compute the running time aver- shown to be a useful indicator to assist in discrimination of aged 95% confidence interval. We consider a reliable detection noninfectious SIRS from sepsis patients.[35,36] Importantly, the when the PCT signal’s 95% lower confidence interval exceeds results of the DENIS assay showed robust performance, even the background’s upper confidence interval. We characterized though the patients serum samples presented significant visual PCT cutoff detection times for high (sepsis), medium (non- differences in the color and consistency (Figure S4, Supporting infectious SIRS) and low (healthy) PCT concentration groups Information). and observed that we detect sepsis levels in less than 5 min and noninfectious SIRS levels in around 10 min, while PCT signal from healthy individuals did not significantly exceed back- 2.6. Rapid Biomarker Detection ground over 2 h. These results show that PCT detection can be performed in under 15 min, including sample mixing, injec- Rapid turnaround time is an essential characteristic of a sepsis tion and readout, which is particularly important in intensive diagnostic device, as sepsis is a time-dependent emergency care units. Small 2020, 16, 1906108 1906108 (6 of 11) © 2019 The Authors. Published by WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.advancedsciencenews.com www.small-journal.com Figure 6. PCT measurements using DENIS to assist diagnosis. Samples from 34 clinically diagnosed individuals, including 18 sepsis patients, 11 noninfectious SIRS patients, and 5 healthy controls, were tested. Box and whiskers plot with scattered data points for sepsis, noninfectious SIRS and healthy samples are shown for a) plasmonic DENIS and b) clinical CLIA measurements. Each data point represents an individual patient. PCT serum levels were significantly higher for noninfectious SIRS group compared to healthy (P < 1.7e−5 and P < 1.1e−5 with DENIS and CLIA, respectively), and for sepsis compared to noninfectious SIRS group (P < 2.3e−7 and P < 1.1e−7 with DENIS and CLIA, respectively). Receiver−operation characteristic curves were constructed for PCT measurements with c) DENIS and d) CLIA. The performance of DENIS was equivalent to CLIA. For both methods, AUC is 1 to discriminate healthy from sick patients. PCT levels also provided good classification between noninfectious SIRS and sepsis patients, with AUC = 0.995, sensitivity = 0.944, and specificity = 1 for both methods. 3. Discussion at the level of gold-standard immunoassays. Both PCT and CRP levels were significantly elevated in the sick groups compared DENIS device proved to be particularly suitable for on-site to healthy subjects. In the sampled group DENIS measure- operation due to its portability, inexpensive off-the-shelf optical ments of PCT levels showed performance equivalent to clini- components, and because it does not involve any moving parts cally validated assay in assisting classification between sepsis or complex microfluidic elements prone to clogging. Impor- and noninfectious SIRS patients, with 94.4% sensitivity and tantly, the sensor chips and bioassay reagents can be stored 100% specificity. These results show that DENIS platform can in the fridge stably over weeks and unlike previously reported be effectively deployed to help rapidly triage patients between optical detection schemes[21–23] do not require fluorescent tags noninfectious SIRS and sepsis groups. The detection is robust that can be expensive, unstable, and difficult to produce. The against background noise, nonspecific adsorption of proteins, simple and rapid bioassay enables highly sensitive and robust and variations in optical properties of the serum, enabling detection of PCT and CRP, covering the full clinically relevant reliable detection. In combination with small sample volume ranges. The high sensitivity of the assay is enabled by the used (20 µL), this makes DENIS suitable for frequent bio- unique plasmonic mechanism that works by detecting indi- marker testing, such as required during treatment response vidual molecules on the Au-NHA surface through imaging of monitoring. single Au-NPs, without the need for complex signal amplifica- The sensor chips were fabricated in a robust wafer-scale cost- tion procedures. The system achieved 21.3 pg mL−1 LOD sen- effective process using DUVL lithography. The functionaliza- sitivity for PCT and 36 pg mL−1 LOD for CRP, and a dynamic tion of sensors was performed in a scalable way with microarray range of at least three orders of magnitude, which is well within bioprinter that uses minimal amounts of antibodies. In combi- the clinical biomarker levels. nation with easy-to-use bioassay and a compact optical reader, DENIS was validated with measurements of PCT and CRP this makes the DENIS technology suitable for the development biomarkers from sera of 34 individuals, including 18 sepsis of a clinical diagnostic device. Owing to the high sensitivity, patients, 11 noninfectious SIRS patients, and 5 healthy controls, rapid time to result, compactness and cost efficiency, DENIS is and was shown to provide consistent and robust performance expected to be employed in hospitals as well as small clinic and Small 2020, 16, 1906108 1906108 (7 of 11) © 2019 The Authors. Published by WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.advancedsciencenews.com www.small-journal.com no more than few µL of sample could provide more robust characterization of the patient’s host response and potentially predict survival.[13,25,39,40] Importantly, the detection principle is not limited to proteins and can be extended to small peptides, DNA, or RNA, as long as a pair of complementary recogni- tion elements is available. The development of the bioassay to operate with whole blood would deliver even faster time to result by eliminating serum extraction step and better posi- tion the test for the use in resource-limited settings. Further reducing the costs of the imager and scaling up the production of Au-NHA sensors would push the per-test costs even lower and open the path to potential market entry. Finally, the devel- opment of a fully integrated easy-to-use software interface will facilitate introduction of the device to clinical professionals and help validate the technology on large sets of clinical samples. 4. Conclusion Our study reports a portable digital nanoparticle-enhanced plas- monic imager that enables rapid detection of two inflammatory sepsis-related biomarkers, PCT and CRP. The unique nanoplas- monic mechanism through imaging of single Au-NPs binding on the Au-NHAs enables highly sensitive and rapid biomarker detection directly in blood serum. The compact nanoplasmonic reader, built of inexpensive components, weighs less than 1 kg. The portable device was deployed in hospital settings, where it was shown to provide equivalent performance to gold-standard laboratory assays using a wide range of samples from patients with sepsis, noninfectious SIRS, and healthy donors. Moreover, identification of PCT levels typical for sepsis, SIRS, and healthy subjects can be performed in less than 15 min, which is signifi- cantly faster than laboratory immunoassays. Therefore, DENIS platform promises to become a powerful tool in managing acute inflammatory conditions at the point-of-care in an afford- Figure 7. Time-resolved PCT detection using DENIS. Real-time measure- able way, improving the quality of health care for more people. ments of selected subjects: a) high PCT content (sepsis patient, [PCT] = 24.26 ng mL−1), b) medium PCT content (noninfectious SIRS patient, [PCT] = 0.79 ng mL−1), and c) low PCT content (healthy control subject, [PCT] < 50 pg mL−1). Error bars from the PCT signal represent 95% con- fidence interval extracted from data collected from three different aPCT 5. Experimental Section antibody microspots from a single image frame. The background signals Au-NHA Sensor Fabrication: Au-NHAs were produced using wafer- are collected from three off-the spot BSA blocked areas. The blue shaded scale low-cost DUVL process and ion beam etching. Fused silica wafers intervals indicate dynamic time averaged 95% confidence interval for with 4 in. diameter were used as substrate. The wafers were cleaned background signal. Insets show the first 30 min time period. The detec- with RCA solution (1:1:5 H2O2:NH4OH:H2O by volume) for 15 min, tion time is defined as the cutoff point when the PCT signal’s 95% lower rinsed with deionized water for 5 min, and dried under nitrogen stream. confidence interval exceeds the background’s upper confidence interval. Next, the cleaned wafers were coated with 10 nm of titanium (Ti) and The detection time is estimated to be less than 5 min for sepsis and 120 nm of gold (Au) using Alliance-Concept EVA 760 electron-gun 10 min for noninfectious SIRS. evaporator. The NHAs with 200 nm diameter and 600 nm period were patterned over whole wafers using a 248 nm deep-UV stepper (ASML PAS 5500/300 DUV). The exposed wafers were treated with ion-beam physician’s office settings, where the access to large and costly etching (Oxford Instruments PlasmaLab 300 IBE), to transfer the laboratory testing equipment is limited, therefore assisting nanohole arrays into metal films. The photoresist was stripped using timely diagnosis and reducing antibiotic prescription. oxygen plasma. We envision several ways in which the current DENIS system To label the Au-NHA sensors, QR codes were formed with 50 nm could be improved. First, the imaging readout enables scaling Ti film. The NHA wafers were coated with 1.3 µm thick lift-off positive up the test for multiplexed measurements.[38] Sepsis manifests photoresist (AZ1512) using Süss ACS200 Gen3 coater, and the QR in a complex system of pathophysiological responses, spanning microarray patterns were exposed using Heidelberg Instruments MLA150 Maskless Aligner. The resist was developed, and 50 nm Ti multiple organs. Creating a sepsis marker panel and expanding film was evaporated using an electron gun evaporator. Wafers covered DENIS to multiplexed detection of relevant biomarkers with with photoresists and Ti were diced into 1 × 1 cm2 sensor chips, comparable dynamic ranges, such as presepsin, MR-pro-adre- and the photoresist was removed from chips by immersing in resist nomedullin, pancreatic stone protein, and interleukin 6 from remover with sonication at 70 °C overnight. The final chip cleaning was Small 2020, 16, 1906108 1906108 (8 of 11) © 2019 The Authors. Published by WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.advancedsciencenews.com www.small-journal.com performed with oxygen plasma for 5 min at 500 W and 1 min immersion Hospital Biobank (PT17/0015/0047), integrated in the Spanish National in RCA solution (1:1:5 H2O2:NH4OH:H2O) to ensure uniformly clean Au Biobanks Network. Samples were processed following standard operating surface. procedures with the appropriate approval of the Clinical Research Ethics Chemicals and Biologicals: Ammonium hydroxide solution (ACS Committee (approval reference number PR(AG)11/2016). Sample reagent, 28–30%), hydrogen peroxide (H2O2 30%), BSA lyophilized, PBS, selection was done to ensure a wide range of concentrations to be Tween20, and d-(+)-trehalose dihydrate were purchased from Sigma- measured with DENIS, from low PCT and CRP values associated with Aldrich. MCP-2F polymer was bought from Lucidant. Ethanol (EtOH) healthy individuals, to medium levels usually observed in noninfectious absolute was from Thermo Chemicals. Pierce protease inhibitor (PI) SIRS patients and increasing values that are characteristic of sepsis tablets, recombinant human PCT and anti-PCT monoclonal antibodies and septic shock patients. It should be noted that sepsis is a complex QNO5 and CALC, used for spotting on Au-NHA and NP conjugation, syndrome and patient groups could have overlapping biomarker levels. respectively, were provided by ThermoFisher (BRAHMS). Antihuman For each patient the samples were collected at a single time point. C-reactive protein monoclonal IgG antibodies PCR-196 and PCR-183, For sepsis patients, samples were collected at the time sepsis was used as capture and recognition antibodies, respectively, were provided diagnosed. For noninfectious SIRS patients, samples were collected by Diesse Diagnostica Senese. Au-NP conjugation kit with lyophilized in the immediate postoperative period. The samples were stored in spherical Au-NPs of 100 nm dimeter coated with polyethylene glycol 500 µL aliquots at −80 °C in cryovials. Samples were handled on ice and (PEG, 10 kDa) and activated with 1-Ethyl-3-(3-dimethylaminopropyl) protected from light. carbodiimide-N-hydroxysuccinimide chemistry (EDC-NHS), were Serum Bioassays: The bioassays were performed by mixing patient purchased from Cytodiagnostics. serum with the Ab-NP conjugates (aPCT or aCRP) and a stabilizing Au-NHA Functionalization: Au-NHA chips were coated with MCP-2F buffer and injecting the mixture into the sensor measurement chamber. (copoly-DMA-MAPS-NAS-fluorinated) polymer (Lucidant) to yield A 20 µL aliquot of serum sample was mixed with 16 µL of buffer and stable and reproducible functionalization with Ab microarray spots 4 µL of aPCT Ab-NP stock by gentle pipetting and inserted onto the and prevent the sensors from nonspecific fouling by serum and NPs. measurement chamber. In the case of CRP measurements, the patient The polymer contains activated NHS-ester groups that enable covalent serum was first diluted in PBS 1× with bovine serum albumin 7%. immobilization of molecules through amino groups. The MCP-2F The serum dilution was done in two steps, first 100 times and then polymer stock was diluted with coating solution from Lucidant, and 500 times, to minimize the error, reaching the final dilution of 50 000 times. the Au-NHA chips were immersed in the solution for 30 min at room Then, a 20 µL aliquot of the diluted serum was mixed with 16 µL of temperature. Next, the Au-NHA sensors were washed in large volumes measurement buffer and 4 µL of anti-CRP Ab-NPs by gentle pipetting of mQ water and dried under nitrogen stream. The sensors were placed and inserted to the sensor chamber. Sensor chamber was formed by in a vacuum oven at 80 °C for 15 min (<2 mmHg) to dry the polymer placing a capillarity-based disposable microfluidic cartridge (Microtec) on and ensure stable adhesion to the Au surface. Polymer-functionalized the Au-NHA sensor chip using a silicone spacer for sealing. The loaded chips were immediately spotted with Ab microarrays because NHS-ester sensor was either immediately inserted into the portable reader for real- groups of the polymer degrade in the presence of air humidity. time measurements or incubated at room temperature, and end-point The Ab microarrays were formed on the sensors by spotting Ab readouts were taken at the end of the incubation period (2 h). solutions on polymer coated Au-NHA surface using noncontact The stabilizing buffer was optimized to ensure optimal bioassay piezoelectric ultralow volume dispensing system (sciFLEXARRAYER S3, performance and prevent NP aggregation during the measurements. Scienion). Spotting solutions containing Ab specific against PCT, or The buffer consists of 0.05% PBST 1× and 50 × 10−3 NaOH mixed in CRP, or mouse isotype control IgG, all at 200 µg mL−1 were prepared 3:1 volume ratio. in PBS 1× with trehalose (0.5% w/v) and Tween20 (0.01% w/v), Calibration Measurements: In order to perform calibration to ensure uniform Ab distribution over each spot and stabilize Abs measurements of the PCT bioassay, known concentrations of during sensor storage. Microarray spots of 400 pl were dispensed with recombinant human PCT (50 pg mL−1, 100 pg mL−1, 300 pg mL−1, a period of 400 µm inside a humid chamber. The humidity control 500 pg mL−1, 1 ng mL−1, 3 ng mL−1, 10 ng mL−1, and 50 ng mL−1) were was set at dew point (65% relative humidity) to prevent evaporation. spiked in 7% BSA PBS 1× with protease inhibitors (1 tablet per 50 mL of The sensor chips with IgG microarrays were blocked with BSA buffer). Next, the calibration samples, as well as a blank negative control (1% w/v) to inactivate nonreacted EDC-NHS groups and minimize the containing BSA 7% in PBS 1× and PI without PCT, were measured nonspecific binding. The sensor chips were stored until use at 4 °C for identically to the serum bioassays. A 20 µL aliquot of calibration sample up to 4 weeks. was mixed with 16 µL of measurement buffer and 4 µL of Ab-NP, Au-NP Functionalization: Au-NPs of 100 nm diameter coated with incubated over 2 h and measured with the portable reader to obtain the 10 kDa PEG and activated with EDC-NHS were functionalized with aPCT corresponding signals. For each calibration sample, at least five different or aCRP antibodies according to the following protocol. Antibodies antibody sensing spots were measured to compute the mean and the were diluted in protein resuspension buffer (Cytodiagnostics) to standard deviation of the signal. 500 µg mL−1 concentration. The antibody solution (40 µL) was mixed To obtain calibration curve for CRP, serial dilutions of a real human with reaction buffer (50 µL, Cytodiagnostics), and added to the vial of sample were used with a known CRP concentration characterized by an lyophilized Au-NPs. The Au-NP mixture with Abs was incubated at room immunoturbidimetric test (266.9 µg mL−1). The sample was diluted in temperature for 4 h with a mild mixing using Eppendorf ThermoMixer at BSA 7% PBS 1× at dilution ratios of 1:750 × 103, 1:500 × 103, 1:100 × 103, 700 rpm. After incubation, 10 µL of quencher solution (Cytodiagnostics) 1:50 × 103, 1:10 × 103, and 1:103 times, and the measurements were was added to stop the reaction. The Au-NPs suspension was centrifuged performed identically to PCT. Calibration curves for PCT and CRP were at 150 rcf for 30 min to wash the NPs from nonreacted Abs. The fitted to the measured points, according to the following equation supernatant was removed and NPs were resuspended in 1 mL of PBS 1× with 1% w/v BSA and 0.05% w/v Tween20. Centrifugation and a−d resuspension steps were repeated 3 times. At the final resuspension y=d+ b (1) 1+   x step, Au-NPs pellet was diluted in a 100 µL of PBS 1× with 1% w/v BSA c  and 0.05% w/v Tween20 to obtain Au-NP stock solution with optical  density OD = 20 (measured at 572 nm), or 7.7E+10 NP mL−1. The where a is a theoretical response at zero concentration, b is a slope Ab-NPs solution was stored at 4 °C for up to 4 weeks. Before performing factor, c is a concentration inflection point, and d is a theoretical bioassay, Ab-NPs were vortexed at 800 rpm for 30 s to resuspend the response at saturation. NPs. Optical Reader: The nanoplasmonic imager is made of off-the-shelf Biobank Samples: Samples and data from the patients used in this optical components and effectively comprises a portable bright-field study were provided by the Sepsis Bank of Vall d’Hebron University imaging microscope. A narrowband light emitting diode (LED) source Small 2020, 16, 1906108 1906108 (9 of 11) © 2019 The Authors. Published by WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.advancedsciencenews.com www.small-journal.com (Thorlabs M660L4 ) in combination with a bandpass filter with 660 nm C-reactive protein (CRP) and anti-CRP monoclonal IgG antibodies, and center wavelength and 10 nm full width at-half-maximum (Thorlabs) is BRAHMS ThermoFisher for providing recombinant human procalcitonin used for narrow-band illumination at the wavelength of Au-NHA EOT (PCT) and anti-PCT monoclonal antibodies. The authors are grateful peak. The light from the LED is collimated with an aspherical condenser for the support of the École Polytechnique Fédérale de Lausanne, lens (Thorlabs). A custom aluminium holder for the sensor chamber including the Technology Transfer Office’s Enable program. The was fabricated to accommodate the sensor inside the reader. The light is authors acknowledge the European Union’s Horizon 2020 Research and transmitted through the sample chamber, modulated by Au-NHA sensor, Innovation Program under grant agreement no. 644956 (RAIS project). and the imaging is performed using a 50× objective (Nikon) and a black- The VHIR-HUVH is supported by Plan Nacional de I+D+i 2013-2016 and-white CMOS camera (IDS B-UI-3200SE-M-GL). The CMOS camera and Instituto de Salud Carlos III, Subdirección General de Redes y has 4104 × 3006 pixels with 3.45 µm pixel size. The system enables Centros de Investigación Cooperativa, Ministerio de Economía, Industria y imaging of ≈900 µm × 700 µm FOV area with an ≈1.5 µm resolution Competitividad, and Spanish Network for Research in Infectious Diseases (diffraction-limited spot). The reader dimensions are 10 × 10 × 35 cm (REIPI RD16/0016/0003)—cofinanced by European Development Regional and the weight is less than 1 kg. Fund “A way to achieve Europe,” Operative program Intelligent Growth Image Processing: Image acquisition and processing was performed 2014. The authors kindly appreciate the generous donation of samples and using custom Matlab functions and a graphical user interface from a clinical data of the donors of the Sepsis Bank of HUVH Biobank. laptop connected to the CMOS camera. Images from Au-NHA sensors were recorder using custom autoexposure function to ensure similar intensity profiles across all images, and so that image histogram is always covered by the dynamic range of the CMOS. The images were Conflict of Interest normalized by the background illumination to exclude the source The authors declare no conflict of interest. variations. To extract the NP signal, image areas were analyzed using a fixed intensity threshold to quantify bound NPs. The percentage number of pixels in an area darker than the threshold was used for quantification. The intensity threshold was optimized by matching plasmonic images to Author Contributions scanning electron micrographs of identical Au-NHA areas to maximize the number of detected NPs, while minimizing false-positive signals.[27] A.B. and F.Y. contributed equally to this work. A.B., F.Y., and H.A. Importantly, Au-NP aggregates can be identified and filtered out from conceived the study. A.B. and F.Y. developed the DENIS device and the signal, as described previously.[27] A custom Matlab function was designed the experiments. A.F., J.G., J.R., and R.F. coordinated collection used to recognize QR codes from the images and extract encoded of human samples and clinical data and assisted with interpretation of information. For the end-point measurements at least five different results. A.B. and F.Y. conducted experiments and analyzed the data. A.B., FOV were acquired and analyzed from each sensor chip for statistical F.Y., and H.A. wrote the manuscript with review from all other authors. significance. In order to control for nonspecific NPs binding and background fluctuations, signal from each measured spot was corrected by subtracting the signal from adjacent BSA blocked background area. Keywords During time-resolved measurements, the images of a single FOV were recorded with 30 s interval. Each field of view had at least three antibody gold nanoparticles, imaging biosensors, nanoplasmonics, point-of-care spots, which enabled to assess the mean and standard deviation of the diagnostics, sepsis signal from each frame. In order to evaluate detection cutoff times, signal from three PCT sensing spots and its 95% confidence interval (±1.96*δ) Received: October 23, 2019 were estimated at each time point. To establish the background signal Published online: December 12, 2019 and its variation, average from three BSA blocked areas is plotted at each time point and running time averaged at 95% confidence interval is computed. The time-averaged standard deviation at every time point T was computed to incorporate variances of background signals of all [1] M. Singer, C. S. Deutschman, C. W. Seymour, M. Shankar-Hari, individual previous frames ti D. Annane, M. Bauer, R. Bellomo, G. R. Bernard, J.-D. Chiche, C. M. Coopersmith, R. S. Hotchkiss, M. M. Levy, J. C. Marshall, ∑ i = 0 δ t2 G. S. Martin, S. M. Opal, G. D. Rubenfeld, T. van der Poll, n δT = i (2) J.-L. Vincent, D. C. Angus, JAMA, J. Am. Med. Assoc. 2016, n  315, 801. A reliable detection was considered when the PCT signal’s 95% lower [2] C. Fleischmann, A. Scherag, N. K. J. Adhikari, C. S. Hartog, confidence interval exceeds the background’s upper confidence interval. T. Tsaganos, P. Schlattmann, D. C. Angus, K. Reinhart, Am. J. Respir. Software Availability: All custom plug-ins associated with this Crit. Care Med. 2016, 193, 259. manuscript are available upon request. [3] C. Llor, L. Bjerrum, Ther. Adv. Drug Saf. 2014, 5, 229. [4] C. J. Paoli, M. A. Reynolds, M. Sinha, M. Gitlin, E. Crouser, Crit. Care Med. 2018, 46, 1889. [5] A. Kumar, D. Roberts, K. E. Wood, B. Light, J. E. Parrillo, S. Sharma, Supporting Information R. Suppes, D. Feinstein, S. Zanotti, L. Taiberg, D. Gurka, A. Kumar, Supporting Information is available from the Wiley Online Library or M. Cheang, Crit. Care Med. 2006, 34, 1589. from the author. [6] C. W. Seymour, F. Gesten, H. C. Prescott, M. E. Friedrich, T. J. Iwashyna, G. S. Phillips, S. Lemeshow, T. Osborn, K. M. Terry, M. M. Levy, N. Engl. J. Med. 2017, 376, 2235. [7] E. Rivers, B. Nguyen, S. Havstad, J. Ressler, A. Muzzin, B. Knoblich, Acknowledgements E. Peterson, M. Tomlanovich, N. Engl. J. Med. 2001, 345, 1368. The authors would like to thank Center of MicroNano Technology at [8] J. A. Russell, N. Engl. J. Med. 2006, 355, 1699. École Polytechnique Fédérale de Lausanne and University of California [9] B. Reddy, U. Hassan, C. Seymour, D. C. Angus, T. S. Isbell, K. White, Santa Barbara Nanofabrication Facility for providing support with chip W. Weir, L. Yeh, A. Vincent, R. Bashir, Nat. Biomed. Eng. 2018, manufacturing, Diesse Diagnostic Senese SPA for providing human 2, 640. Small 2020, 16, 1906108 1906108 (10 of 11) © 2019 The Authors. Published by WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.advancedsciencenews.com www.small-journal.com [10] R. Ferrer, I. Martin-Loeches, G. Phillips, T. M. Osborn, S. Townsend, [25] Y. Zhou, Y. Zhang, A. Johnson, A. Venable, J. Griswold, D. Pappas, R. P. Dellinger, A. Artigas, C. Schorr, M. M. Levy, Crit. Care Med. Anal. Chim. Acta 2019, 1062, 110. 2014, 42, 1749. [26] J. Min, M. Nothing, B. Coble, H. Zheng, J. Park, H. Im, G. F. Weber, [11] M. M. Levy, L. E. Evans, A. Rhodes, Intensive Care Med. 2018, C. M. Castro, F. K. Swirski, R. Weissleder, H. Lee, ACS Nano 2018, 44, 925. 12, 3378. [12] H. H. Liu, M. W. Zhang, J. B. Guo, J. Li, L. Su, Ir. J. Med. Sci. 2017, [27] A. Belushkin, F. Yesilkoy, H. Altug, ACS Nano 2018, 12, 4453. 186, 207. [28] S. A. Maier, Plasmonics: Fundamentals and Applications, Springer [13] C. Pierrakos, J.-L. Vincent, Crit. Care 2010, 14, R15. Science & Business Media, Springer, USA 2007. [14] L. Magrini, F. Travaglino, R. Marino, E. Ferri, B. D. Berardinis, [29] L. Guo, J. A. Jackman, H.-H. Yang, P. Chen, N.-J. Cho, D.-H. Kim, P. Cardelli, G. Salerno, S. D. Somma, Eur. Rev. Med. Pharmacol. Sci. Nano Today 2015, 10, 213. 2013, 14, 133. [30] H. Guner, E. Ozgur, G. Kokturk, M. Celik, E. Esen, A. E. Topal, [15] K. L. Becker, R. Snider, E. S. Nylen, Br. J. Pharmacol. 2010, 159, 253. S. Ayas, Y. Uludag, C. Elbuken, A. Dana, Sens. Actuators, B 2017, [16] M. M. Shaikh, L. E. Hermans, J. M. van Laar, Rheumatology 2015, 239, 571. 54, 231. [31] T. Siegfried, Y. Ekinci, O. J. Martin, H. Sigg, ACS Nano 2013, 7, [17] L. Chen, C. Feng, J. Dong, Y. Zhai, X. Chen, B. Li, X. Zhou, W. Chen, 2751. T. Li, Int. J. Clin. Exp. Med. 2016, 9, 13763. [32] N. C. Lindquist, A. Lesuffleur, H. Im, S.-H. Oh, Lab Chip 2009, 9, 382. [18] C. I. Michaelidis, R. K. Zimmerman, M. P. Nowalk, M. J. Fine, [33] F. Yesilkoy, R. A. Terborg, J. Pello, A. A. Belushkin, Y. Jahani, K. J. Smith, J. Gen. Intern. Med. 2014, 29, 579. V. Pruneri, H. Altug, Light: Sci. Appl. 2018, 7, 17152. [19] G. Sener, E. Ozgur, A. Y. Rad, L. Uzun, R. Say, A. Denizli, Analyst [34] A. M. Leclair, S. S. Ferguson, F. Lagugné-Labarthet, Biomaterials 2013, 138, 6422. 2011, 32, 1351. [20] A. H. Nguyen, Y. Shin, S. J. Sim, Biosens. Bioelectron. 2016, 85, 522. [35] M. Meisner, K. Tschaikowsky, T. Palmaers, J. Schmidt, Crit. Care [21] D. Rascher, A. Geerlof, E. Kremmer, P. Krämer, M. Schmid, 1999, 3, 45. A. Hartmann, M. Rieger, Biosens. Bioelectron. 2014, 59, 251. [36] I. A. Meynaar, W. Droog, M. Batstra, R. Vreede, P. Herbrink, Crit. [22] S. Kumar, S. Tripathy, A. Jyoti, S. G. Singh, Biosens. Bioelectron. Care Res. Pract. 2011, 2011, 594645. 2019, 124–125, 205. [37] K. Hajian-Tilaki, Caspian J. Intern. Med. 2013, 4, 627. [23] F. Baldini, L. Bolzoni, A. Giannetti, M. Kess, P. M. Krämer, [38] C. Jung, J. A. Hawkins, S. K. Jones Jr., Y. Xiao, J. R. Rybarski, E. Kremmer, G. Porro, F. Senesi, C. Trono, Anal. Bioanal. Chem. K. E. Dillard, J. Hussmann, F. A. Saifuddin, C. A. Savran, 2009, 393, 1183. A. D. Ellington, Cell 2017, 170, 35. [24] U. Hassan, T. Ghonge, B. R. Jr., M. Patel, M. Rappleye, I. Taneja, [39] Q. Zou, W. Wen, X. Zhang, World J. Emerg. Med. 2014, 5, 16. A. Tanna, R. Healey, N. Manusry, Z. Price, T. Jensen, J. Berger, [40] M. J. Llewelyn, M. Berger, M. Gregory, R. Ramaiah, A. L. Taylor, A. Hasnain, E. Flaugher, S. Liu, B. Davis, J. Kumar, K. White, I. Curdt, F. Lajaunias, R. Graf, S. J. Blincko, S. Drage, J. Cohen, Crit. R. Bashir, Nat. Commun. 2017, 8, 15949. Care 2013, 17, R60. Small 2020, 16, 1906108 1906108 (11 of 11) © 2019 The Authors. Published by WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim