Inductively Coupled Plasma Optical Emission Spectroscopy as a Tool for Evaluating Lateral Flow Assays

Abstract

Lateral flow assays (LFAs) are immunochromatographic point-of-care devices that have greatly impacted disease diagnosis through their rapid, inexpensive, and easy-to-use form factor. While LFAs have been successful as field-deployable tools, they have a relatively poor limit of detection when compared to more complex methods. Moreover, most design and manufacturing optimization is achieved through time- and resource-intensive brute-force optimization. Despite increased interests in LFA manufacturing, more quantitative tools are needed to study current manufacturing protocols and therefore, optimize and streamline development of these devices further. In this work, we focus on a critical LFA component, colloidal gold conjugated to a detection antibody, one of the most commonly used reporter elements. This study utilizes inductively coupled plasma optical emission spectroscopy (ICP-OES) in conjunction with a lateral flow reader to quantitatively analyze colloidal gold distributions at the read-out test and control lines, as well as residual gold on the conjugate pad and other flow through regions. Our goals are to develop a more rigorous understanding of current LFA designs as well as a quantitative understanding of shortcomings of operational characteristics for future improvement. To our knowledge, this is the first time that ICP-OES has been used to study the initial distribution of colloidal gold on an unused LFA and its redistribution after a test is performed. Using three different brands of commercially available malaria LFAs, gold content was measured within each section of an LFA at varying parasite test concentrations. As expected, the total mass of gold remained unchanged after LFA use; however, the total mass of initial gold and its redistribution varied among manufacturers. Importantly, there are also some inherent inefficiencies that exist in these LFA commercial designs; for example, only 30% of the total gold deposited onto Brand A LFA binds to the test and control lines, sections of the test that contain interpretable signal. Using information gathered with this method, future devices could be more purposefully engineered to focus on improved binding efficiency, resulting in reduced costs, improved limit of detection, and diminished test-to-test and manufacturer-to-manufacturer variability.

Graphical Abstract

Introduction

Point-of-care (POC) diagnostic devices have become useful technologies to diagnose and manage a variety of healthcare conditions.¹ One of the most recognizable POC formats utilizes the principle of lateral flow to move fluid through porous membranes, referred to as lateral flow assays (LFAs).^2,3 LFAs are both commercially successful and easily recognizable, with familiar examples including blood glucose strips, at-home pregnancy tests, and global health infectious disease tests. These tests are easy to use, rapid, and inexpensive. Perhaps most importantly, LFAs are versatile in that they can be adapted to diagnose a wide number of disease states in a variety of settings.⁴ When rigorously developed, these devices meet the World Health Organization’s ASSURED (Affordable, Sensitive, Specific, User-friendly, Rapid and Robust, Equipment-free and Deliverable to end-users) guidelines,⁵ making LFAs an ideal choice for POC devices, particularly in resource-constrained settings.

While LFAs can utilize a range of binding chemistries, most implementations share similar components: a sample pad (SP), conjugate pad (CP), nitrocellulose membrane (NC), and a wicking pad (WP) (Figure 1A). The sample pad and conjugate pad are commonly composed of glass fiber materials while the wicking pad is typically made of cellulose fibers.^6,7 In the classic sandwich assay format,^8,9 capture reagents are immobilized onto the nitrocellulose membranes in two locations: 1) a test line with capture agents against a specific antigen (often monoclonal antibodies), and 2) a control line downstream from the test line (often a species-specific secondary antibody, i.e., goat anti-mouse IgG antibody). A detection reagent, typically a reporter element conjugated to a monoclonal antibody that is specific to the target analyte, is deposited onto the conjugate pad. A variety of reporter elements can be used to detect binding at the test and control lines,^6,10 including fluorophores, cellulose nanoparticles, dyed polystyrene microbeads, and, colloidal gold nanoparticles (AuNPs).^11,12 Colloidal gold is inexpensive, exhibits high stability, and generates signal that is detectable by visual inspection, rendering it one of the most widely used reporter elements.⁶

Figure 1.

Sandwich LFA before and after segmentation: (A) Whole LFA before division; (B) LFA divided into sections before aqua regia digestion for ICP-OES measurement. (SP = sample pad, CP = conjugate pad, NC1 = first nitrocellulose pieces, TL = test line nitrocellulose, NC2 = second nitrocellulose piece, CL = control line nitrocellulose, NC 3 = third nitrocellulose piece, WP = wicking pad)

To perform a test, a sample is added to the sample pad where it flows down the test strip by capillary action, first interacting with the reporter elements on the conjugate pad, and then with the capture agents at the test and control line. When antigen is present in the sample, it binds to the antibody-AuNP conjugate (often referred to simply as “conjugate”) and flows until the analyte-conjugate complex forms a sandwich with immobilized antibodies at the test line. Conjugate that is unbound at the test line continues to flow downstream and a fraction of it binds to the secondary antibody at the control line. In the absence of antigen, only the control line is detectable which serves as an operational control for each test by validating that the sample and conjugate have moved past the test line. Residual sample and unbound conjugate continue to flow until absorbed by the wicking pad.²

While sufficient in many use-cases, LFAs have several drawbacks, including: test-to-test variability,^11,13 limited sensitivity^9,14, and varying specificity.¹⁵ Recently, extensive experimental and modeling efforts have been undertaken to understand how to manipulate the signal of LFAs to maximize diagnostic performance.^16–19 Despite its apparent maturity, there is still much left to be discovered about the optimal LFA design, and most variables are empirically chosen on a per assay basis, including: material and reagent selection, buffer compositions, blocking conditions, and assay formats.

The ideal lateral flow design would be expected to have several key characteristics. First, the visual indicator at the test line should be proportional to the concentration of the analyte in the sample. Second, sufficient conjugate should bind to the control line to indicate that the test has performed as expected. Third, all the visual indicator initially at the conjugate pad should be entrained by the flow and none should be captured non-specifically at locations other than the test and control lines of the lateral flow strip. A test that exhibits each of these features would encompass the ideal redistribution of colloidal gold on the LFA after use and achieve the best limit of detection (LOD) with the selected reagents.

In this work, we used inductively coupled plasma optical emission spectroscopy (ICP-OES) to quantify the distribution of gold before LFA use and its redistribution after LFA use of three commercially available Plasmodium falciparum (Pf) malaria LFAs. ICP-OES is a highly sensitive analytical technique that determines the elemental composition of a sample by measuring the emission spectra when a solution is introduced to plasma. This technique demonstrates a wide linear dynamic range, experiences little chemical interference and background emission, is highly robust to matrix effects, and shows exceptional sensitivity in the parts-per-billion concentration range for most elements.²⁰ The use of ICP-OES enables spatial analysis of gold content before and after lateral flow use. In this study, ICP-OES was utilized in conjunction with a lateral flow reader (LFR) for the quantitative evaluation of the LFAs with the goal of measuring platform fundamentals of three existing commercial products and identifying features where improvement might lead to enhancements in the limit of detection of LFAs.

Experimental Methods

LFA Selection

Three brands of LFAs for the diagnosis of malaria were selected from the list of WHO-evaluated diagnostic tests for Plasmodium falciparum (Pf).²¹ These LFAs were operated according to corresponding manufacturer protocols, and the completed tests were analyzed using an LFR to obtain quantitative signals for the test and control lines prior to analysis with ICP-OES. The LFA brands are denoted as: Brand A, Brand B, and Brand C. The purpose of this work is to illustrate the use of analytical techniques to understand and improve LFAs. Therefore, each manufacturer is kept blinded so as to keep the primary focus on the methods and resulting data analysis.

Materials and Reagents

Gold standard for ICP (999 mg/L ± 2 mg/L) was purchased from MilliporeSigma (Burlington, MA, USA). Trace metal grade hydrochloric acid (HCl) and nitric acid (HNO₃) were purchased from Fisher Scientific (Hampton, NH, USA). Polyvinylidene fluoride syringe filters, 13 mm, 0.22 μm, were purchased from Tisch Scientific (North Bend, OH, USA). Deionized water used in this study was purified with a resistivity greater than or equal to 18.2 MΩ•cm. Pooled human whole blood with anticoagulant citrate phosphate dextrose was purchased from Bioreclamation IVT (Westbury, NY, USA). An in-house malaria Pf D6 strain culture was used to evaluate the LFAs.

LFA Protocol

LFAs were performed according to each manufacturer’s instructions to detect Pf Histidine-rich protein 2 (HRP2) antigen. Briefly, 5 μL of sample was added to the test, followed by 5 drops of running buffer. The sample and buffer took 20 minutes to completely flow the length of the membrane. The Pf culture aliquots used were at a parasitemia of 43,600 parasites/μL (p/μL) which, for this parasite culture, corresponds to 97.2 nM HRP2. Parasite concentrations were prepared by spiking varying amounts of P. falciparum into pooled human whole blood. This method was utilized to closely mimic positive and negative patient samples in the evaluation of commercial LFAs. While individual field-collected patient specimens will have variability that is homogenized with pooled human whole blood, the inherent biochemical and rheological complexity of the sample matrix is retained. Previous work has used malaria parasite culture spiked into human whole blood to create mock patient samples, with comparable results to field-collected samples.^22,23 Parasite concentrations of 0, 25, 50, 100, 200, 400 and 800 p/μL were studied for Brand A. Test Brands B and C were evaluated on a subset of these parasite concentrations: 0, 25, 100 and 800 p/μL.

LFA Flow

LFAs from Brands A-C were run in triplicate following manufacturer’s instructions and a video was taken using an Apple iPhone 11 Pro. 150 μL of running buffer from the corresponding manufacturer was added to the sample pad of the test and allowed to flow. Digital frame-by-frame analysis was performed in ImageJ to identify the leading edge of the fluid front on the LFAs.²⁴ The distance from the sample pad to that of the fluid front was measured in pixels and converted to millimeters using the in-frame ruler as a reference. The time for the fluid front to reach the test line was measured in ImageJ, starting from the time that the sample was added to the well. Since the Lucas-Washburn equation is only appropriate for a single porous material, data before the fluid front was visible in the viewing window of the test was removed prior to fitting with a Lucas-Washburn-like equation (Eq. 1), which was performed using the SciPy library in Python.

$$L\left(t\right)=2\sqrt{\frac{k_s\gamma \cos \theta }{\varphi \mu r_m}}\sqrt{t}$$ (1)

where $L$ is the length the fluid front has traveled, $t$ is time, $k_s$ is the superficial permeability of the porous medium, $\varphi$ is porosity, $\mu$ is viscosity, $\gamma$ is surface tension, $\theta$ is the liquid-solid contact angle, and $r_m$ is mean pore radius. Grouping the interfacial properties and the porous media properties into a single constant, $a$, results in the familiar scaling law that describes the imbibition of liquid in porous media over time (Eq. (2)).

$$L\left(t\right)~a\sqrt{t}$$ (2)

While any individual porous material will have a single characteristic value of $a$, lateral flow assays are comprised of different porous materials in series: fluid starts at the sample pad, flowing through the conjugate pad, and then onto the nitrocellulose membrane. Since we only have data for flow within the nitrocellulose membrane, where the fluid front is visible in the LFAs test window, we have modified Eq. (2) to more accurately fit our data. To obtain an estimate for a, we have included a time delay constant that accounts for the time before the fluid flow is visible in the test window, at which point it’s distance can be visually traced (Eq. (3)).

$$L\left(t\right)~a\sqrt{t-t_0}$$ (3)

LFR Operating Conditions

Upon completion, the LFAs were analyzed by a Qiagen ESEQuant LFR (Stockach, Germany) operating in reflective mode on the E1/D2 channel. In addition to the previously mentioned concentrations, unused LFAs were also evaluated. For the test and control line, signal intensity was measured in mm*mV. Each test was measured from 0 to 60 mm in the LFR, starting from the wicking pad and ending at the sample pad. The signal generated by the conjugate at the test and control lines were quantified by integrating the area under the signal curve, using a fixed baseline and including 1 mm upstream and downstream of the peak in the line scan.

Preparation of LFA Strips for Gold Digestion

Each section of the LFA was cut by hand with stainless steel razor blades (Figure 1), resulting in eight sections: sample pad (SP), conjugate pad (CP), the first section of nitrocellulose (NC1), test line (TL), the second section of nitrocellulose (NC2), control line (CL), the third section of nitrocellulose (NC3), and wicking pad (WP). Each section was placed into an individual microcentrifuge tube. Unused LFAs were also analyzed in this study, and in the absence of liquid sample, there was no test or control line on the test, resulting in the digestion and ICP-OES analysis of only four sections: SP, CP, NC, and WP.

Digestion of LFA Components for ICP-OES

Solutions of aqua regia were prepared using 3-parts HCl to 1-part HNO₃ (v/v) and 0.667 mL of the mixture was added to each tube for the dissolution of gold. Fresh solutions of aqua regia were made as necessary and remaining aqua regia was disposed of appropriately.²⁵ Each tube was vortexed and left to digest for 3 hours. Preliminary results suggested that a longer digestion time (up to 24 hours) had no effect on gold extraction from the LFA sections (data not shown). Any material that appeared pink from the gold content turned white after digestion, suggesting that gold was effectively extracted from the nitrocellulose. The digestion of some test sections resulted in a fibrous solution and required filtration through 0.22 μm PVDF filters. The samples were then diluted with 4.333 mL DI water and filtered through PVDF syringe filters. The samples were immediately analyzed by ICP-OES after acid digestion and filtration.

ICP-OES Operating Conditions

The amount of gold extracted from each section of the LFA was quantified with a Perkin Elmer Optima 7000 DV ICP-OES (Perkin Elmer, Waltham, MA, USA). Table S1 lists the instrument’s operating conditions. A sample matrix blank was comprised of 13.3% aqua regia in DI water. In order to analyze the colloidal gold on the LFAs, five ICP-OES standards of 1.0, 0.1, 0.01, 0.001 and 0.0001 ppm Au at a wavelength of 267.595 nm were utilized to generate standard curves (n=3) (Figure S1) for each individual experiment.

Calculation of Minimal Amount of Gold for Visual Detection

The minimum amount of gold necessary for visual detection was calculated using (Eq. 4), where $r$ is the radius of the gold nanoparticle, $\rho$ is the true density of the colloidal gold solution, $V$ is the sample injection volume, and $T{L}_{Au}$ is the gold concentration found on the test line at the lowest parasite concentration (Figure 7). For this calculation, the following assumptions were made: spherical gold nanoparticles were 40 nm in diameter, 19.32 g/cm³ density, and a sample injection volume of 0.5 mL.

Figure 7.

The gold concentration within each section of the LFAs as a function of brand and parasite load. A) Brand A; B) Brand B; and C) Brand C.

$$\text{Minimum}\hspace{0.17em}\text{amount}\hspace{0.17em}\text{of}\hspace{0.17em}\text{gold}=\frac{T{L}_{Au}V}{\left(\frac{4}{3}\pi r^3\right) \rho }$$ (4)

Statistical Analysis for Limit of Detection and Coefficient of Variation

The LOD for both the LFR and ICP-OES were calculated using 3σ/κ, where σ is the standard deviation of the blank and κ is the slope of the calibration curve. Each sample was performed in triplicate. The average and standard deviation for each section of the LFA for each concentration were calculated. A coefficient of variation (CV) was calculated using (σ/μ)*100, where σ is the standard deviation and μ is the average of the data set. The total gold content was calculated by adding the amount of gold found on each of the constituent sections together. The average and standard deviation were calculated for the total gold content.

Statistical Fitting of Test Line Gold Content

The gold content at the test line for each brand, as determined by ICP-OES, was fit to both linear (Eq. 5) and logistic equations (Eq. 6).

$$y=Ax+B$$ (5)

$$y=A+\frac{B-A}{1+C{x}^D}$$ (6)

In these equations, $y$ is the gold content at the test line, and $x$ is the parasite concentration. For the linear fit, the parameters $A$ and $B$ are the familiar slope and y-intercept, respectively. For the logistic expression, the parameter $A$ is the upper asymptote, $B$ is the lower asymptote, $\frac{1}{C}$ is the midpoint between asymptotes, and $D$ is the rate of increase between asymptotes. The linear and sigmoidal fits were performed using the NumPy and SciPy packages in Python, respectively.

Statistical Analysis for Distribution of Gold Content

Statistical analyses were performed in the GraphPad Prism software v. 9.0. Statistical significance was determined using two-way analysis of variance (ANOVA) with post hoc Tukey’s multiple comparisons test comparing total gold concentration at varying parasite concentrations within and between brands. One-way ANOVA with post hoc Tukey’s multiple comparisons test was used to compare gold concentration on conjugate pads of different brands.

Results and Discussion

LFA Flow Results

Anticipating that the rate of fluid flow can impact binding efficiencies, a study was performed to examine how fast gold conjugate travels each test strip for all three brands (Figure 2). Initially, visibility of the fluid front was obscured by the opaque cartridge that houses the LFA. In this region, the buffer wicks from the sample pad to the conjugate pad, where it resuspends dried gold conjugate. From there, the gold conjugate is wicked onto the nitrocellulose membrane, where it is eventually visible in the LFA’s test window. The fluid fronts on LFAs from Brand B were the first to emerge from the viewing window, followed by Brand C, and finally, Brand A. The time to reach the test line location (denoted as a dashed line in Figure 2) was analyzed and found that gold conjugate from Brand B tests reaches the test line in approximately 9 seconds, which is faster than both Brand A (14 seconds) and Brand C (17 seconds).

Figure 2.

Time study to analyze how fast (in seconds) gold conjugate flows down the membrane (in mm) to the control line and wicking pad for each brand. The test line location for each brand is denoted on the graph as a dashed line. Some error bars are smaller than width of marker.

The fluid front distance traveled through a porous media is described by the Lucas-Washburn equation^26,27 (Eq. (1)). We have fit the data in Figure 2 to a modified Lucas-Washburn equation (Eq. (3)), with the fit parameters shown in Table 1 and the fit data shown in Figure S2.

Table 1.

Parameter values and standard errors from non-linear least squares fit to Eq. (3).

	$a\left[\frac{\text{mm}}{{\text{s}}^{0.5}}\right]\pm \text{Std}.\hspace{0.17em}\text{Err}.$	$t_0\left[\text{s}\right]\pm \text{Std}.\hspace{0.17em}\text{Err}.$
Brand A	4.29 ± 0.15	36.22 ± 4.82
Brand B	3.90 ± 0.06	26.70 ± 1.69
Brand C	4.53 ± 0.14	17.99 ± 2.99

While the nitrocellulose of Brand B has the slowest wicking speed $\left(a\right)$, the fluid front of this brand reaches the test line in the shortest amount of time. The data that was fit this equation is only from when the fluid flow has reached the viewing window, at which point it is flowing on the nitrocellulose membrane. This discrepancy can therefore be attributed to the distance of the test line from the conjugate pad, as well as the time that it takes the sample to wick from sample pad to conjugate pad to nitrocellulose membrane. Variations in this time could result from additives (i.e., stabilizers, blocking reagents) added to the sample and conjugate pads, as well as the sizes and material selections.

LFR Results

In these experiments, a dilution series of Plasmodium falciparum was added to commercially available Brand A LFAs. Representative photos of these tests can be found in Figure S3. A faint test line begins to visually appear at a concentration of 25 p/μL. The test line becomes visibly darker with increased parasite density. The intensity of the test and control lines were then analyzed using an LFR (Figure 3). As expected, only a control line was observed for the blank sample. The area under the intensity linescans from the LFR for the test line signal increases as parasite concentration in the sample is increased (Figure 4). Over the range of concentrations evaluated (0 p/μL – 800 p/μL), the area for the test line signal is approximately linearly proportional to analyte concentration. A standard curve was generated in order to determine the lowest detectable signal. A LOD for this method was calculated to be 130 p/µL (denoted as a horizontal dashed line on Figure 4), which is similar to other literature reports.^21,28–30 The data demonstrates a directly proportional relationship between parasite concentration and test line intensity. Brand B and Brand C were also analyzed via LFR. As expected, parasite concentration and test line area intensity were shown to be directly proportional for these brands as well. The only observable difference was clearance of blood on the nitrocellulose membrane in Brand B that led to a decreased test line area in comparison to Brands A and C (Figure S3).

Figure 3.

Representative LFR signal profiles for individual Brand A LFAs at 0, 25, 100, and 800 p/μL. The LFR obtains signal by scanning from the wicking pad to the sample pad.

Figure 4.

LFR standard curve measuring test line signal at varying parasite concentrations for Brand A.

ICP-OES Results

After test completion and LFR analysis, LFAs were cut into their constituent sections (Figure 1B) and digested in aqua regia prior to conducting ICP-OES. The amount of gold present on the conjugate pad of an unused LFA was first analyzed for each brand. As this is the only place conjugate is deposited, this value represents the total amount of gold found on each LFA. The conjugate pad contained the most amount of gold for all brands (Figure S5), as expected. Brand B contained 72% more gold than Brand A, and 44% more gold than Brand C, highlighting the variation in proprietary formulations of the LFAs. Only 2% of gold was detected on the sample pad of Brands B and C. Finding gold dispersed throughout an LFA before use may indicate improper storage as moisture can cause migration of the gold. The relatively small amount found just outside the conjugate pad is likely a result of the physical overlap between the conjugate and sample pads, more so than a suggestion that the tests were improperly stored.

To evaluate intra- and inter-brand manufacturing variability, 15 additional conjugate pads were cut from unused tests and gold content was analyzed by ICP-OES. This data was combined with the conjugate pads from the previous unused tests to obtain a total of 18 samples for all three brands (Figure 5). For Brand A tests, the gold content on conjugate pads ranged from 0.092 ppm to 0.157 ppm Au, with an average of 0.129 ppm ± 0.017 ppm Au. There is some fluctuation in gold content among the 18 samples, with a CV of 13.5%, illustrating variability between tests. The total gold found on Brand B was almost three times higher than Brand A at an average of 0.381 ppm ± 0.053 ppm Au, where the CV was 14.0%. Brand C had a slightly lower average at 0.233 ppm ± 0.040 ppm Au with a CV of 17.2%. Comparison of CV values between brands demonstrates that Brand C has higher test-to-test variability compared to Brands A and B. There is also clear variability between manufacturers as demonstrated by the higher amount of total gold found on Brands B and C LFAs (Figure 5). The data shown demonstrates a discrepancy in the manufacturing process during gold deposition, leading to possible differences in test outcome. With initial gold content on an LFA being directly linked to the potential signal at a test line, and hence a major determinant for test sensitivity and limit of detection, along with an increased demand from test users for quantitative results,¹² there is an opportunity for improved manufacturing procedures to more uniformly deposit conjugate.

Figure 5.

Comparison of gold content on 18 same-manufacturer conjugate pads for Brand A, Brand B and Brand C. Significant differences were found between all three brands (p < 0.0001)

A mass balance of gold on the LFAs was calculated for each LFA that was run (Figure 6) by adding together the gold found on each section of the LFA. It was expected that the mass balance (total gold redistribution) would remain approximately constant, regardless of target analyte concentration, within the variations seen on the dry conjugate pads analyzed in Figure 5. This hypothesis was true for all brands of LFAs tested. Gold content for Brand A ranged from 0.090 ppm to 0.123 ppm with an average of 0.106 ± 0.011 ppm, with CVs varying from 3.13% (0 p/μL) to 15.2% (100 p/μL). It is observed that any Brand A test should have roughly 0.105 ppm total gold content, regardless of the analyte concentration. As noted before, Brands B and C contained more gold than Brand A (Figure 5). The tests from Brand B had between 0.386 ppm to 0.545 ppm total Au when comparing both used and unused tests, demonstrating a large amount of test-to-test variability within the manufacturer. On average, Brand B resulted in 0.466 ± 0.038 ppm gold per test. Moreover, an average of 0.250 ± 0.019 ppm of gold was reported for tests from Brand C, with gold content values ranging from 0.217 ppm to 0.261 ppm gold. As expected, overall gold remained constant, within the amount deposited on the conjugate pads of unused LFAs, for all brands – i.e., the total mass of gold does not change with LFA use.

Figure 6.

Total gold content as a function of parasite concentration and brand of test (n=3). In the above figure, * represents p < 0.0427, ** represents p = 0.0032. All other interactions within a single brand were found to be nonsignificant. Total gold concentration between brands was deemed statistically significant (p < 0.0001)

To identify the amount of gold that could be detected on each of the LFA components, ICP-OES standard curves were generated for each separate experiment performed, which resulted in a LOD of 0.0039 ppm Au for Brand A and LODs of 0.0023 ppm Au for Brands B-C (depicted as horizontal dotted lines in Figure 7A–C). These LODs fall just below the amount of gold found on the test line of a test run with a 25 p/μL sample. This indicates that parasitic concentrations less than 25 p/μL would likely be undetectable by this method. The use of ICP-OES to analyze LFA’s afforded almost a 5-fold improvement in sensitivity compared to the LFR. While this analysis approach is more sensitive, we are not suggesting the use of ICP-OES for point-of-care analysis, as this technique is cost-intensive and requires laboratory infrastructure and trained personnel. Rather, we have identified that even with an instrument that can measure on the order of parts-per-billion, there is a limit for how much of a performance improvement can be extracted. This five-fold increase, while substantial, suggests that the major limitation, where improvements can generate larger returns, remains the signal generated from the POC device. As a result, ICP-OES can be used to aid the manufacturing process of LFA’s to focus on increasing the sensitivity of the device, rather than improving detection instrumentation.

Additionally, our studies indicate that a minimum of 2.49 ng Au (3.85 × 10⁶ Au nanoparticles) is required (from Eq. (4)) on the test line to achieve a visible signal (at 25 p/μL) for Brand A. This calculation provides an estimate for the amount of gold nanoparticles necessary to obtain visual signal at a test line that is 5 mm wide and 1 mm thick, the area of the segment that was cut for test line digestion in these experiments. We hypothesize this calculation to be similar for every brand at the resulting limit of detection. This analysis derives from straightforward calculations, and is subject to many theoretical parameters (i.e., antibody coverage on gold nanoparticles, multiple epitopes on target biomarkers). However, it provides an approximation approach for quick feasibility calculations to determine if a target analyte is in sufficient concentration for detection.

As expected, the amount of gold conjugate bound at the test line increases with concentration regardless of brand, while the amount of gold on the control line remains relatively constant. For Brand A samples containing a visible test line (25–800 p/μL), the nitrocellulose sections closest to the wicking pad, the second section of nitrocellulose (NC 2) and the third section of nitrocellulose (NC 3), contained an amount of gold below the LOD (0.0039 ppm Au). However, the first section of nitrocellulose (NC 1) in between the conjugate pad and the test line retained, on average, almost 15% of the total gold content for Brand A (Figure 7A). Similarly, approximately 20% of gold appears to remain on the conjugate pad for Brand A, never flowing laterally down the test. Combined, this leaves one-third of the reporter element unavailable to generate signal at the test line—an obvious negative impact on test sensitivity. Furthermore, the wicking pad retained 35% of the gold content, on average. In total, nearly 70% of the total gold is either being retained by the CP and NC 1, or flowing past the test line to the WP. This quantitatively illustrates the lack of efficiency of the current LFA design, leaving only 30% of the total gold on the LFA to bind to the test and control lines.

While Brand B had more overall gold than the other brands, some similar trends observed for Brand A held for this brand as well (Figure 7B). Gold conjugate retention is observed on NC 1 (up to 15% retention for 0 p/μL) and NC 2 (up to 5% retention for 0–100 p/μL) for Brand B as well. The presence of gold on NC1 indicates non-specific binding prohibited almost 15% of the conjugate from reaching either the test or control line, which decreases the amount of signal that could be generated. On the other hand, gold present on NC2 signifies the conjugate was able to flow laterally past the test line, although it did not participate in binding. As expected, the amount of gold on the test line increased with increasing concentration. Finally, the WP contained 40% of the total gold content at both 0 p/μL and 800 p/μL, but only 30% for 25 and 100 p/μL. These tests contained overall more gold than the previous manufacturer (Figure 5), and showed an increase in non-specific binding, which can hinder sensitivity potential for these tests as approximately 60–70% of total conjugate is free to participate in binding on the test and control line.

In contrast to Brands A and B, the amount of gold on numerous sections for Brand C was below the LOD (0.0023 pm Au). Roughly 2% of the gold conjugate remained on the CP for Brand C, which is one tenth of the amount of gold found on the CP for Brands A-B, demonstrating some variability in design. In contrast to Brands A-B, Brand C had a more drastic change in gold content on the TL when moving from low to high concentration (Figure 7C). This resulted in barely visible test line for the low concentration and amount of gold very close to the LOD of the ICP-OES (0.0023 pm Au). For this brand, 48% of the total gold conjugate was contained to the WP for 0–100 p/μL, except at 800 p/μL, where only 2% was identified. This discrepancy likely correlates with the higher gold content found on the test lines for those samples. Although there was minimal non-specific binding for Brand C, a higher concentration of analyte was necessary to identify a true positive result, exemplifying a need for LFA design optimization to maximize binding potential on the test and control lines at low concentrations. Furthermore, the test line gold content (Figure 7) fits well to both sigmoidal and linear curves (Eq. (5) and (6)). In the linear fit (Table 2), each brand demonstrates similar slope and y intercept values. A positive slope and a y intercept value near zero signifies undetectable gold content on the test line at a concentration of 0 p/μL. Moreover, comparable sigmoid fit parameters are obtained for each brand. This is expected as these curves are commonly used for diagnostic assays. The test line signal intensity would be expected to plateau as parasite concentration increases further, but the point of saturation would be different for each LFA analysis instrument. Further increasing the parasite concentration would eventually lead to a decrease in signal due to the Hook effect. It appears that the concentrations evaluated in this study are well within the linear dynamic range of the instrumentation.

Table 2.

Parameter values and R² for linear and sigmoid fit to Eq. (5) and Eq. (6), respectively.

	Linear Fit Parameters			Sigmoid Fit Parameters
	A	B	R2	A	B	C	D	R2
Brand A	1.65 × 10−5	4.16 × 10−3	0.934	8.46	4.93 × 10−3	2.24 × 10−7	1.32	0.943
Brand B	9.03 × 10−5	1.41 × 10−2	0.924	4.39 × 10−1	1.07 × 10−2	5.68 × 10−4	8.87 × 10−1	0.926
Brand C	1.17 × 10−4	−1.69 × 10−3	0.982	5.92 × 10−1	1.72 × 10−3	2.43 × 10−5	1.33	0.926

Signal generation on LFAs is a complex tradeoff between the macroscale wicking of fluid, microscale diffusion of reagents through pores, and the biochemical kinetics of binding. This can be captured by Damkölher numbers, the ratio of time scales for reagent transport and reaction, at the macro- (Eq. (7)) and micro-scales (Eq. (8)).

$$D{a}_{\text{macro}}=\frac{k_{on}{C}_A}{v}$$ (7)

$$D{a}_{\text{micro}}=\frac{L^2{k}_{on}{C}_A}{D_{AB}}$$ (8)

In these equations, $L$ is the relevant length scale for transport (pore radius for diffusive transport), $v$ is the convective fluid velocity, $D_{AB}$ is the diffusion coefficient for reagent A within solution reagent B, $k_{on}$ is the biomarker on-rate binding coefficient, and $C_A$ is the concentration of biomarker. Several recent numerical approaches have sought to understand its influence on assay development.^16,31–33

In this work, we demonstrated experimental approaches to determine two key parameters of this relationship: the initial concentration of detection gold nanoparticles, and the speed at which reagents wick down the nitrocellulose membrane. Other analytical techniques, like biolayer interferometry and surface plasmon resonance, could be used to measure binding kinetics, after which Damkölher numbers could be calculated to further understand how these parameters affect test performance.

Still, commercial development of LFAs must consider other metrics beyond optimal test sensitivity and specificity, including time-to-result and cost. For instance, it is reasonable to assert that manufacturers may elect to use more gold conjugate on a faster membrane to reduce the time-to-result. In contrast, reduction of the amount of colloidal gold may not have a large effect on test signal when the target biomarkers are in abundance, which would be a reasonable approach to lower costs. While these other factors must be considered when constructing an LFA, underlying knowledge of the design selections made can improve both device performance and speed to market.

Conclusion

LFAs have been globally used as point-of-care diagnostic tools for decades, but the empirical optimization of new tests remains slow and expensive. Analytical techniques can improve the development process by providing a more fundamental understanding of current LFA design that can lead to more strategic test development. Inefficiencies were found in the design of three different commercial devices, all of which counter ideal LFA characteristics that would lead to optimal performance. In this report, we highlight the use of ICP-OES to measure the redistribution dynamics of colloidal gold within LFAs. We are not suggesting this instrument be used to analyze tests at the POC (due to size, instrument and maintenance costs, weight, and power requirement). Rather, we envision ICP-OES be used to inform manufacturing decisions in the future, prior to test deployment. As a demonstration, we use ICP-OES to measure the widely understood, but poorly quantified manufacturing variations. Comparisons of gold binding and flow speed across different test brands shows that test developers have flexibility in selection of parameters to meet their technical requirements. The use of ICP-OES allowed for a precise, comprehensive examination of the binding efficiencies of gold conjugate, and can be used in conjunction with modeling efforts to improve test development. Ultimately, this may lead to POC devices with improved LOD, less variability among tests and manufacturers, and ultimately, reduced cost and faster time to market.