IntroductionParacetamol (acetaminophen) is one of the most widely used analgesics globally. When taken at the recommended therapeutic dose, it is believed to have an excellent safety profile1,2. However, in overdose, paracetamol is the commonest cause of acute liver failure (ALF) in the USA and Europe. It is estimated that there are at least 1000 paracetamol overdose (POD)-ALF cases treated in Liver Transplantation Units in the US and EU (mostly UK) each year. This group has a mortality rate of ~30–35%, with around 30% receiving a liver transplant. Most transplant recipients survive, but will require continuous lifelong treatment and frequently experience complications related to the immunosuppressive medications necessary for transplantation3. POD is common, with around 100,000 people presenting to emergency departments following a POD and ~50,000 patients requiring emergency antidote treatment to prevent drug-induced liver injury (DILI), and subsequent ALF, every year in the UK alone4. In 2021, US poison centres received more than 80,000 cases involving a paracetamol product5. Treatment with the antidote, N-acetylcysteine (NAC), is highly effective when administered within 8 h of overdose6. However, NAC offers minimal benefits if treatment is delayed for more than around 20 h.There are no specific symptoms or clinical signs of early liver damage and, therefore, clinicians rely on blood ‘liver functions tests’ to pick up liver injury after POD. These tests are performed in central hospital laboratories, which results in a time delay between blood sampling and therapeutic decision making. Furthermore, the standard serum biomarker for DILI diagnosis, alanine aminotransferase (ALT) activity, increases too slowly post-POD to accurately diagnose DILI within the NAC optimal therapeutic window. Therefore, a rapid point-of-care (POC) assay is needed to identify high-risk patients with fit-for-purpose sensitivity and specificity to enhance early targeted NAC treatment.Cytokeratin 18 (K18) is a mechanistic biomarker of liver injury that provides information about the type of cell death. The caspase-cleaved form of K18 (cc-K18) is released early during cellular structural rearrangement and apoptosis. The full-length form of K18 is passively released upon cell necrosis7. Multiple studies have demonstrated that K18 is a sensitive and specific biomarker that can accurately distinguish patients with and without DILI at an earlier time point than the gold standard, ALT8,9. K18 has regulatory support for use as a biomarker in clinical trials from the US Food and Drug Administration. K18 has potential utility for predicting DILI and prognostic assessment of outcome, further emphasising its potential as a promising DILI biomarker10,11.The gold standard method for K18 quantification is an enzyme-linked immunosorbent assay (ELISA). Although the ELISA produces accurate and quantitative results, the process takes several hours to perform by trained staff using specialist equipment and costly materials12. In clinical practice, the results would take too long, delaying NAC treatment beyond the optimal therapeutic window of 8 h. To maximise the benefit that K18 offers, a rapid and quantitative test must be developed. The test should be capable of being used in the hospital Emergency Department at the POC, with minimal training required, and be relatively cheap to produce. A suitable tool to meet this product profile is a lateral flow immunoassay (LFIA). LFIAs are paper-based tests performed in a plastic cassette that are designed to detect a biomarker of interest. Capture antibodies labelled with gold nanoparticles form sandwich immunoassays with detection antibodies on the test line when the biomarker is present in a sample, immobilising them and producing a red line. Conventionally, they provide a binary visual result based on the presence or absence of a test line, akin to the SARS-CoV-2 LFIA test. However, when quantification of the biomarker concentration is required for diagnosis, monitoring treatment, improving patient stratification, or if the concentration is very low, the LFIA can be combined with surface-enhanced Raman scattering (SERS).SERS is a vibrational spectroscopy technique that enhances the Raman scattering of molecules that are bound to a roughened metal surface via the excitation of the surface plasmons. Typically, gold and silver nanoparticles have been used to provide enhancement factors of up to 101013. To apply SERS as a read-out technique for an LFIA, nanoparticles are labelled with a Raman reporter molecule, as well as an antibody14. When the LFIA is run, and the sandwich immunoassay has formed on the test line, the line can be analysed using a Raman spectrometer. The resulting SERS spectrum will be that of the Raman reporter bound to the immobilised nanoparticle. As SERS is quantitative and increases in relation to the number of nanoparticles and Raman reporters present, the intensity of the SERS spectrum can be related to the biomarker concentration15.The coupling of SERS and LFIAs has been used to detect low concentrations of different biomarkers. Most of the previous research has focused on developing nanoparticles that produce strong SERS signals. By modifying the shape, metals and Raman reporters, limits of detection in the pg/mL range have been achieved for a variety of clinical biomarkers, including pneumolysin, interleukin-6 (IL6) and HIV-1 DNA16. However, despite the high levels of sensitivity, the SERS of the test lines were analysed on large benchtop Raman readers attached to microscopes, and Raman mapping the test lines can take up to 20 min per sample17. This methodology is not feasible in a POC setting, and, therefore, the development of small and portable Raman readers is required to allow SERS-LFIA to be utilised in a clinical environment18.In this work, we create, evaluate and refine a SERS-LFIA coupled to a bespoke handheld Raman reader (HRR), specifically designed for LFIA measurements. This allows us to quantify a biomarker, K18, to identify POD patients at an increased risk of developing DILI.ResultsCreation of a bespoke diagnostic assay and readerTo detect and quantify both full-length and cc-K18 in patient samples, a SERS-LFIA strip was created and coupled with a bespoke HRR to produce a quantitative output. The accuracy, sensitivity and specificity of the test for DILI detection were assessed in two diagnostic performance tests (study A and B). The SERS-LFIA strip contained conjugates, consisting of gold nanoparticles (AuNP) coated in a Raman reporter and antibodies (Ab). The AuNPs were synthesised via a citrate reduction method, functionalised with the Raman reporter 4,4′-dipyridyl (DIPY) and encapsulated in a silica shell (SiO2). The silica shell was then functionalised with antibodies specific to K18, as well as bovine serum albumin (BSA), which increases protection and reduces non-specific binding. The resulting conjugate is called ‘Au-DIPY-SiO2-Ab NP’. Characterisation data are presented in Supplementary Figs. 1 and 2 and Supplementary Table 1. The LFIA strip consisted of treated sample pads, the Au-DIPY-SiO2-Ab NP conjugate pad, test and control lines and a collection pad. The strip was then encased in a 3D-printed cassette (Fig. 1A). To run a sample on the SERS-LFIA, serum was diluted in running buffer and dispensed onto the sample port of the cassette. It was left to run for 30 min then analysed on a bespoke HRR.Fig. 1: Pre-clinical SERS-LFIA development.A A schematic illustrating SERS-LFIA strip architecture showing the location of the test (red) and control (blue) antibody lines, the strip housed in a 3D-printed cassette, and the immunoassays that form on the test and control line when K18 is present. B The SERS-LFIA procedure. Serum collected from a patient was diluted in running buffer and pipetted onto the sample port of the cassette. The test is allowed to run for 20 min and analysed with the HRR. In study A, the cassette was inserted into the 3D-printed accessory and manually lined up with the laser (HRR 3A, top). In study B, the cassette was slotted into the cassette sleeve and inserted into a built-in slot within the HRR (HRR 4A, bottom). The sleeve aided alignment of the line with the laser. The HRR units were connected to a laptop, and the intensity of the SERS signal of the test and control line determined whether the patient had developed DILI. Technical illustrations were generated using SolidWorks CAD software and Adobe Illustrator. Figure contains a modified illustration from Servier Medical Art, licensed under a Creative Commons Attribution 4.0 International License (CC BY 4.0). HRR handheld Raman reader, K18 cytokeratin 18, LFIA lateral flow immunoassay, SERS surface-enhanced Raman scattering.Full size imageTo use the HRR in a hospital Emergency Department, it should be operated as a Class 3R for safety, meaning that the laser must have a maximum output power of 5 mW19,20. This was considered, and the HRR was designed to capture as many scattered photons from the test and control line as possible to optimise the sensitivity at a low laser power. To achieve this, a large numerical aperture (f/1.1) was used, which allowed more photons to be collected. The miniaturised optical bench of the HRR was based around a custom volume phase holographic transmission grating optimised for high efficiency and minimal scatter (Wasatch Photonics LTD21), matched with an uncooled, commercial line-array sensor to achieve ultra-high sensitivity in a handheld diagnostic instrument that could be cost-effectively manufactured in volume. To optimise the interface with the LFIA, the laser was projected as a line onto the test line using a Powell lens. This increased the area of the test line that could be interrogated and reduced the number of measurements per sample. Therefore, the value of this HRR lies in three elements: (1) use of a highly efficient f/1.1 transmissive design with a low-cost Complementary Metal-Oxide Semiconductor detector for enhanced sensitivity, (2) inclusion of a Powell lens to maximise overlap between the excitation laser and LFIA strip lines, with consequent direct imaging onto the detector, and (3) use of a Class 3R laser to effectively and repeatably perform SERS measurements on a lateral flow strip, thus reducing laser exposure risk to facilitate implementation in clinical settings. A detailed optical schema of the HRR is shown in Supplementary Fig. 3.Studies A and B used different iterations of the HRR, consisting of two slightly different coupling methods to the SERS-LFIA strip (Fig. 1B). In study A, the HRR was coupled to the SERS-LFIA using a 3D-printed accessory that attached via magnets in front of the laser aperture (HRR 3A). The accessory was designed to hold the SERS-LFIA at the correct focal distance and allowed the user to move the SERS-LFIA strip back and forth across the laser line, to achieve the optimum overlap, and thus maximise the SERS signal from the strip. In study B, the coupling of the SERS-LFIA to the HRR was modified and enclosed to increase the usability and safety of the HRR (HRR 4A). Cassette sleeves were designed to ensure that when the SERS-LFIA cassette was slotted inside, only the test line was visible through the circular window. The sleave was then inserted into a slot built into the instrument, allowing the test line to be at the correct focal distance and position in front of the laser (Fig. 1B). In practice, the interface between the laser and sample are fully enclosed in HRR 3A and 4A, effectively rendering the devices as Class 1, as it meets the definition ‘Class 1 lasers have low radiated power or are enclosed to prevent radiation from escaping’. However, as we are without certification, we have classed them both as 3R. Each HRR was connected to a laptop, and the intensity of the test line relative to the control line, based on the intensity of the DIPY Raman reporter peak at 1612 cm−1 was used to infer DILI status.To determine the concentration of K18 in an unknown patient sample, a calibration curve was produced to include clinically relevant concentrations of K18 in serum, based on reference ranges described by Church et al.22. The K18 concentration upper limit of normal (ULN) for healthy volunteers was between 7.3 and 9.1 ng/mL. The geometric mean for DILI was 81.5 ng/mL for patients who survived/did not require a transplant 6 months post-DILI onset and 628 ng/mL for those who died/required a transplant. We used a range from 0 to 750 ng/mL to reflect these values. Representative images of the SERS-LFIA tests and resulting SERS measurements used as part of Study A are presented in Supplementary Figs. 4–6, and the results of the calibration curve developed as part of Study B are presented in Fig. 2. Healthy serum spiked at a range of K18 concentrations was applied to the SERS-LFIA and measured with the HRR. The visual output is presented in Fig. 2A, where only the control line is present for serum that is not spiked with K18. A gradual increase in the visual intensity of the lower test line is observed with increasing K18 concentrations. When the SERS-LFIA was coupled with the HRR, and the SERS signals from the lines were measured, there was greater SERS intensity measured at higher K18 concentrations (Fig. 2B). To reduce variation, the control line was also measured, and pixel-by-pixel linear regression of the test spectrum versus the control spectrum was used to standardise the SERS output. Additional information detailing the calibration curve created using the SERS intensity obtained from the test line alone versus linear regression is detailed in Supplementary Figs. 4–6. The linear regression analysis produced a SERS slope output which demonstrated a linear relationship with the spiked K18 concentrations in the samples (R2 = 0.98, Fig. 2C). There was also a correlation between the SERS slope measured compared to the gold standard ELISA measurement for K18, demonstrating a similar linear gradient for both calibration graphs (Fig. 2D).Fig. 2: Development of LFIA in combination with a handheld Raman reader (HRR) for use with clinical samples.A Representative images of the LFIA calibration curve produced using spiked serum samples at clinically relevant concentrations. B Representative SERS spectra for the calibration curve. SERS spectra collected using HRR 4A with 785 nm laser excitation, 3.5 mW laser power, 1 s acquisition and 5 scanning averages. C SERS slope for calibration curve using 3 independent donors. D K18 concentration measured via M65 ELISA and LFIA with SERS presented as mean values. ELISA enzyme-linked immunosorbent assay, K18 cytokeratin 18, LFIA lateral flow immunoassay, SERS surface-enhanced Raman scattering.Full size imageEvaluation of the diagnostic assayWe carried out two diagnostic performance evaluation studies using serum samples from the Markers and Paracetamol Poisoning Study 2 (MAPP2) trial. The multiple operators of the SERS-LFIA test were blinded to the status of the samples (non-DILI or DILI). Results were analysed by an independent statistician as per a pre-defined statistical analysis plan. Full details of the clinical studies are reported as per the Standards for Reporting of Diagnostic Accuracy Studies guidelines in the supplemental information (Supplementary Table 4).The total combined number of patients included in the retrospective case-control studies was 199 (n = 99 for the initial study A and n = 100 for study B). The demographics and clinical characteristics for the patients included in the studies are presented in Table 1. There were no substantial differences between the patients included in study A and study B. In study A, each sample was analysed in triplicate by three independent operators to assess the reproducibility of the SERS-LFIA and usability of the HRR. In study B, each sample was analysed once by a single operator as modifications to the HRR addressed variation in measurements. The K18 concentration was measured with the SERS-LFIA test in serum samples from patients presenting to the Emergency Department following a POD. This included patients who did not develop DILI (non-DILI) and those who did develop DILI. DILI was defined as an ALT elevation of ≥5 times the ULN23.Table 1 Demographics and clinical results for patients included in studies A and BFull size tableFor study A, the pre-defined primary statistical output was the K18 concentration, determined by a randomly selected first SERS measurement, from each of the 3 independent operators (Fig. 3A, B, see ‘Methods’ for details). Using the linear regression of the SERS measurements from the HRR and the calibration curve (Supplementary Fig. 6), the median concentration (IQR) of K18 was determined to be higher in the DILI cohort than the non-DILI cohort (DILI 161.0 ng/mL, 103.5–226.5 ng/mL V, non-DILI 41.0 ng/mL, 27.8–62.8 ng/mL, ROC-AUC = 0.93, 95% CI 0.88–0.98). The secondary statistical output was the geometric mean of all the K18 measurements obtained from each of the operators (Fig. 3C, D). The geometric mean (95% CI) K18 concentration for the analysers was higher in the DILI cohort than the non-DILI cohort (DILI 146.2 ng/mL (125.0–171.0 ng/mL), V non-DILI 41.3 ng/mL, 35.3–48.3 ng/mL, ROC-AUC = 0.95, 95% CI 0.91–0.99). Comparison of the concentrations of ALT, ELISA K18 and SERS-LFIA K18 from a selection of samples is presented in Supplementary Fig. 7. The data for the individual analysers are detailed in Supplementary Table 2. Disaggregated gender data are presented for studies A and B (Supplementary Fig. 8). Based on the data obtained in study A and feedback from the operators, the coupling between the SERS-LFIA strip and HRR was adapted to improve the user experience. The SERS-LFIA test, with integrated coupling between the LFIA strip and the HRR (HRR 4A) was then evaluated with serum samples in a second cohort of POD patients, study B (Fig. 3E, F), using the slope of the SERS measurements from the HRR and the calibration curve (Fig. 2C). With the refined device, used in study B, the median concentration of K18 was higher in the DILI cohort than the non-DILI cohort (DILI 213.2 ng/mL, 149.2–261.7 ng/mL V non-DILI 55.6 ng/mL, 37.2–81.4 ng/mL, ROC-AUC = 0.97, 95% CI 0.94–1.0).Fig. 3: Data for clinical studies A and B with serum samples from patients following a POD.A Study A—Primary statistical output with randomly selected first values from the three analysers, with B corresponding ROC curve. C Study A—geometric mean data points for the three analysers with D corresponding ROC curve. E Refined version of the diagnostic evaluated in study B—mean data points for analyser with F corresponding ROC curve. Bars represent the median values ± interquartile range. P values are based on two-sided Mann–Whitney tests, P