Analysis of pressure-driven membrane preconcentration for point-of-care assays

As the field of point-of-care (POC) diagnostics continues to expand in applications,¹ a recurring barrier is the limited range of detection for devices such as lateral flow assays (LFAs). The limited range of detection is particularly problematic in dilute biofluids such as urine, saliva, sweat, and tears.^2–5 For example, many viral biomarkers, including those of infectious diseases, are large in molecular weight and heavily filtered in these biofluids, and, therefore, not detectable, especially in the pre-symptomatic state.^6–8 There have been numerous efforts to increase the sensitivity of biosensors,^9–11 but this involves tedious development of probes with very strong binding affinity for the analytes. This probe development approach is not broadly generalizable as it is always specific to each analyte of interest.

Other groups have developed methods which are designed to preconcentrate the sample to effectively increase the sensitivity of the sensor, rather than modify the sensor itself. For example, one group used nanotrap particles, hydrogel particles coupled to a chemical dye which can bind to various proteins and viruses, to increase the signal which can be seen from the sample.¹² This technique is unique in the sense that the nanotrap particles can work in various pH and temperature conditions. However, the process takes upwards of 30 min, requires various external pieces of equipment such as a centrifuge, and only increases the observed signal by about 10× when coupled with plaque assays and qRT-PCR (quantitative reverse transcription-polymerase chain reaction). Another group developed a cartridge which preconcentrates saliva for the detection of a molecule which indicates periodontal disease.¹³ This process only requires very little sample (20 μl), but requires specialized laboratory techniques for device fabrication and, although not explicitly quantified in the study, only gives a 1.5–2 fold increase in the signal gain.

In this article, we report on a more generalizable approach of reducing the limit of detection for POC applications by using compact devices with membrane ultra-filtration to broadly preconcentrate analytes in the sample. Pressure-driven membrane preconcentration is analyzed by using vacuum as well as high positive pressure to move water and small interferents (salt, pH) through a membrane filter, thereby concentrating the analyte of interest in the remaining sample fluid. This method utilizes ubiquitous, inexpensive materials, and has the potential for much more rapid, increased preconcentration factors compared to many other sample concentration methods, such as those mentioned previously. In addition, this method can be used to create devices which do not require any specialized external equipment, such that this method can potentially be used outside of the laboratory or physician's office for rapid at-home diagnoses via telemedicine. Because the preconcentration method presented here is generalizable to various diagnostic molecules, it is likely compatible with most existing diagnostic assays and tests. Preconcentration up to 33× is demonstrated for influenza A nucleoprotein with a 5 kDa pore polyethersulfone membrane filter. Demonstration devices include a highly automated vacuum driven device and a more manually operated positive pressure-driven device. Positive pressure can far exceed the one atmosphere limit of vacuum, and, therefore, the sample can be preconcentrated in mere minutes. Compared to other preconcentration techniques, operation of the disposable devices demonstrated here is arguably simpler, further suggesting that membrane preconcentration may be preferable to more complex and slow conventional laboratory preconcentration techniques.^12–18 

In this article, proof-of-concept demonstrations for both influenza-spiked and human chorionic gonadotropin (HCG, pregnancy hormone) spiked fluid samples are described. Influenza A was chosen because in the early stages of infection, there is an insufficient concentration of virus to be detected by a conventional influenza LFA test.^2,7 Antivirals can decrease or even prevent flu symptoms but are only effective if taken within 2–3 days of contracting the virus.¹⁹ Therefore, influenza is an appropriate target for a sample preconcentration study. Coronavirus was not tested, but based on the results for influenza, and that both are respiratory diseases with viral loading found in saliva, future work could include using this technology for coronavirus detection. HCG was also chosen for demonstration, because like influenza, it is measured in a dilute biofluid (urine). Earlier detection at lower concentrations is also highly desirable from a user-perspective (earlier detection of pregnancy).

At the onset of this study, the initial goal was a fully integrated, compact, and auto-staging device with an integrated LFA. A user would simply add fluid, and using only vacuum and capillary forces, the device would automatically preconcentrate the sample and introduce it to an assay such as an LFA. A basic representation of this mode of operation is shown in Fig. 1(a), where the sensor is fully integrated. A vacuum driven device was shown to preconcentrate 2 ml of influenza nucleoprotein-spiked solution by 10× and preconcentrate 1 ml of HCG-spiked solution by 16×. As will be detailed in the paper, the speed at which this vacuum driven device operates is very slow (∼30 min). Therefore, a second alternate device design was pursued and presented here. For this second device, positive pressure was used along with manual sample transfer to the assay [Fig. 1(b)]. Positive pressure can far exceed vacuum limits of one atmosphere (14.7 psi). This “desktop preconcentrator” was shown to preconcentrate 7 ml of fluid down to about 200 μl, resulting in ≥33× analyte concentration. This device operated in the range of tens of seconds to several minutes depending on the biofluid, and, therefore, has the potential for significant applied value bench-top diagnostics. With an understanding now provided on why this paper reviews two distinct devices, the paper next discusses their fabrication and results in detail.

In order to determine which membranes would be used throughout the duration of the study, basic flux membrane characterization was first conducted. Membrane materials can interact with the biofluid and analyte in unique and complex ways.^20,21 The purpose here was simply measurement of the membrane water flux at the pressures relevant to the devices developed in this work. The ideal membrane will retain the analyte but also pass through as many small-solutes as possible such that they are not preconcentrated and, therefore, do not impact the assay results (change in salt, pH, urea, etc.). The ideal membrane will also have the highest water flux possible to reduce the amount of time needed for preconcentration, because our intended application is for rapid point-of-care or benchtop testing. A simple test setup was, therefore, created which would measure the amount of water flux through each membrane over a given amount of time [Fig. 2(a)].

First, a 25 mm disk of each membrane was inserted into a Whatman Stainless Steel membrane filter holder (Part No. 1980-002). A steel mesh (provided with the Whatman product) was placed inside the membrane filter holder adjacent to the membrane disk; this ensured that the integrity and form of the membrane was maintained throughout the duration of the test. This membrane filter holder assembly was then attached to a standard plastic 10 ml syringe. Before the plunger of the syringe was inserted into the barrel, the barrel was filled with water (around 7 ml). The syringe plunger was modified so that the end of the plunger was joined to nylon tubing, which was, in turn, connected to a Dymax SD-100 syringe dispenser. A positive pressure (10–30 psi) was applied by the Dymax syringe dispenser, which then pushed the fluid through the membrane with a flux measurable by 15 s interval camera photos of the syringe position (flux measured in μl/min/cm²/psi).

As a reminder, initial sample volumes in the ml range are needed to allow for 10's–100's μl volume of preconcentrated samples for assays such as LFAs. Therefore, high-fluxes are desired (10's μl/min/cm²/psi). Results of the characterization for various membranes can be seen in Fig. 2(b). The amount of flux through each membrane is linearly proportional to the amount of pressure which is applied by the syringe dispenser. Therefore, flux values are presented in units of μl/min/cm² per psi to normalize the data which were obtained using different pressures, ranging from 10 to 30 psi. The flux through the polyamide membranes was extremely low, around 1–2 μl/min/cm²/psi, which was expected due to the small pore sizes of 175 and 700 Da. The polyethersulfone membranes had significantly higher and satisfactory fluxes, around 8.5 and 48.9 μl/min/cm²/psi for 5 kDa and 30 nm pores, respectively. In addition, large-pore size polytetrafluoroethylene and polycarbonate track-etched membranes data are included in Fig. 2(b) for completeness, although it should be noted that they were not considered as choices for the specific analytes in this study because the pore sizes are significantly larger than our test analytes of choice, HCG (37 kDa) and influenza A nucleoprotein (56 kDa).

After the characterization results were analyzed, the polyethersulfone membranes with pore sizes of 5 kDa and 30 nm (approximately 50 kDa) were chosen as the preferred membranes for this study. The pores are smaller than the test analytes, influenza A nucleoprotein and HCG, while maintaining higher flux values. It is further important to note that the nucleoprotein often is found as a dimer or trimer in biofluid, and, therefore, can be greater than 56 kDa in agglomerated sizes²² (otherwise much of the nucleoprotein could simply pass through the membrane).

The vacuum driven auto-staging preconcentration device of Fig. 3(a) was fabricated as follows. To create this device, a series of layers were produced by laser cutting of acrylic. The various acrylic layers were adhered with SCIGRIP 16 acrylic cement. For this device, the 30 nm polyethersulfone membrane was chosen because its ∼50 kDa pores are slightly smaller than the 56 kDa influenza A nucleoprotein and significantly smaller than the nucleoprotein dimer and trimer. The membrane, in contrast to the rest of the acrylic pieces, was attached using UV epoxy (Loctite 3355) in order to prevent solvent-induced damage of the membrane. A flow restricting sponge was also added at the inlet in order to allow a controlled and even advancement of the fluid front into the device. To maintain higher vacuum pressure during preconcentration, a second chamber was included which lies on top of the waste chamber, where a hydrophobic vent membrane separates the waste chamber from the second chamber above it. If this second chamber was not present, the vacuum pressure would too quickly decrease as the waste chamber filled with fluid, significantly slowing the concentration process. A rubber check valve was used to maintain the applied vacuum. To load the device with vacuum, the entire device was placed into a vacuum chamber for 30 min prior to testing. In real world applications, such a device would be further sealed within a conventional vacuum pouch, thus enabling long-term shelf storage. For step-by-step instructions of device construction as well as a list of materials, see the manual included in the supplementary material. Photos of a completed device can be seen in Fig. 3(c).

The operation of the vacuum-driven auto-staging pre-concentration device, in principle, is fairly simple, and follows the sequential diagrams shown in Figs. 3(a1)–3(a5). First, a fluid sample containing the analyte of interest is placed into the inlet cup of the device. Next, the foil at the bottom of the inlet cup is manually punctured, and the fluid is forcefully drawn into the sample channel of the device due to a negative pressure gradient of vacuum pressure. After entering the channel, water and small solutes are drawn through the membrane and into the waste chamber. The analyte of interest, since it is larger than the size of the membrane pores, remains trapped in the sample channel, and, therefore, becomes more concentrated with time. Once concentration is complete, an LFA is manually inserted into the outlet, breaking the foil seal. Alternately, fluid can be extracted with a syringe and transferred to the LFA or another type of assay.

It is also important to discuss the mechanism by which the device of Fig. 3 produces repeatable preconcentration of analyte. If the user knows both the volume of the device waste chamber as well as the original sample volume, the theoretical amount of concentration that will occur can be easily determined. It is important to note, however, that significant amounts of analyte are routinely lost to membrane fouling.²³ The amount of fouling that occurs during preconcentration is heavily dependent on the analyte of interest, membrane choice, amount of pressure which is applied, and the sample fluid type (urine, saliva, or water). Therefore, it is imperative that a fouling analysis be conducted for each analyte–membrane–fluid combination.

Once the amount of fouling has been determined, one can use the following Eq. (1) to determine the amount of preconcentration that will occur using the device:

Therefore, for example, if one wishes to concentrate their sample 5× and the waste chamber of the device holds 0.9 ml, the user must have 1 ml of sample to achieve this concentration for a case of 50% analyte loss to fouling.

For initial testing, 1× phosphate buffered saline (PBS), pH 7.4, was spiked with influenza A nucleoprotein (purchased from Sino Biological, Inc.). Influenza A nucleoprotein LFAs (purchased from Meridian Bioscience, Inc.) were used to create a calibration curve with various concentrations of nucleoprotein in PBS [see Fig. 3(b1) for the calibration curve]. For device testing, 25 ng/ml of influenza A nucleoprotein concentration was used.

For the initial device tests, the waste chambers in each device held 1.9 ml. Based on fouling analysis for this specific analyte–membrane combination, ∼50% analyte loss to fouling was expected. Therefore, 2 ml of fluid was used for each test in order to achieve a final sample volume of ∼0.1 ml and a resulting preconcentration of ∼10× [Eq. (1)]. To begin the test, the inlet foil seal was punctured and the test fluid was immediately drawn into the device due to the vacuum pressure. Preconcentration began as the fluid was pulled through the membrane and into the waste chamber. When the waste chamber was full, preconcentration was complete. For simplicity in these initial tests with buffer solution, the check valve was replaced with an external vacuum source. The fully integrated/automated function with the check-valve was validated for real biofluid testing (see Sec IV D).

After completion of preconcentration, ∼70 μl of the preconcentrated sample was transferred to the influenza LFA. For comparative testing, 70 μl of the original sample fluid was also pipetted onto an LFA as well, along with 70 μl of the liquid contained in the waste chamber. After waiting 10 min (development period of the LFA), these three LFAs were compared. Comparison of the preconcentrated fluid to the original sample fluid provides the magnitude of preconcentration, whereas testing the waste fluid provides confirmation that none of the analyte is passing through the membrane pores.

The results of the tests can be seen in Fig. 3(b2). Both tests show a ∼10× increase in the concentration of influenza A nucleoprotein (∼250 ng/ml). The control LFAs, when compared to the calibration, show a concentration of 25 ng/ml, as expected. It can also be seen that the waste LFAs both show zero concentration, indicating that there was no significant amount of analyte leaking through the membrane pores.

With initial success for influenza A nucleoprotein, a second series of test were performed with HCG LFA's (urine pregnancy tests) and real biofluid. To compensate for the smaller analyte size of HCG, the 5 kDa pore polyethersulfone membrane was used in place of the previously used 30 nm membrane. For all tests, pooled human urine obtained from Innovative Research, Inc. was used. A calibration curve of HCG spiked urine was prepared, and the test solution consisted of urine spiked with HCG to a concentration of 19.5 mIU/ml (the equivalent of 1.95 μg/ml). The HCG LFAs require less fluid than the influenza A LFAs, and, therefore, only 1 ml of spiked urine was used for each test with the waste-chamber volume set to 950 μl. After each device test was allowed to run to completion, approximately 35 μl of fluid was transferred to the HCG LFA and developed for 10 min. In addition, 35 μl of original sample solution was also tested. These LFAs were then compared to those from the calibration curve, and the final HCG concentration of the urine was determined. Results are shown in Fig. 3(b3). A concentration of 312.5 mIU/ml was observed (the equivalent of about 31.25 μg/ml), which corresponds to a concentration increase of 16×. Theoretically, a decrease in 1 ml to about 35 μl would indicate a 28.6× increase in concentration. However, after considering analyte loss to fouling as previously mentioned, the experimental concentration increase of 16× is very reasonable as predicted by Eq. (1).

These results indicated that the vacuum driven auto-staging device was indeed capable of concentrating a fluid sample; however, it was observed that the amount of time needed to complete the concentration process was very slow. On average, each device took ∼30 min to complete the concentration process, which does not even include the additional LFA development time. Furthermore, if we had tested influenza in relevant biofluid of saliva, the high viscosity of saliva would have made the results even slower yet. Therefore, after discussion with commercial experts in the field of medical diagnostics and LFAs, it was determined that this vacuum driven auto-staging method of preconcentration was not nearly fast enough to constitute a valuable product. If a negative pressure is utilized, one is always limited to atmospheric pressure or less (<14.7 psi). Furthermore, increasing the membrane area could complicate sample recovery and increase loss to membrane fouling. We, therefore, generally conclude and report here a negative result from an applied perspective: the vacuum driven auto-staging device is not practically useful within our present understanding of its construction and operation. Therefore, additional testing was not performed.

Returning to our initial hypothesis that pressure-driven membrane preconcentration could be of significant generalized value to diagnostics and biosensors, we then turned to an alternate device utilizing a much higher positive pressure.

To significantly decrease the amount of time for concentration to occur, a much higher pressure gradient than vacuum was clearly needed. To initially validate a controllable high pressure testing setup, a nitrogen gas tank and a regulator were used to provide a positive pressure up to 100 psi to the device of Fig. 1(b). This device, although much larger and capable of holding around 15 ml of fluid rather than 1 ml, acted similarly to the membrane filter holder mentioned in the membrane characterization portion of the study (Fig. 2). It contained both the membrane itself and a steel mesh backing (120 mesh, 304 stainless steel) to prevent membrane deformity under high pressures. Below the metal housing, a graduated cylinder was placed to catch and measure the waste fluid which had passed through the membrane.

The analyte of interest in this portion of the study was again influenza A nucleoprotein. However, despite the fact that a larger analyte was once again being used, the smaller-pore size 5 kDa polyethersulfone membrane was utilized. This membrane was chosen for the following reasons: (1) the larger pressure gradients being used should make up for the lower flux seen with the smaller membrane pore size and (2) in the interest of keeping these tests as realistic and applicable to as many analytes as possible, the smaller pore membrane was chosen since it had the potential to filter many more analytes of various sizes as long as the analytes were larger than 5 kDa.

Flux tests were conducted first to attain a clear picture of the flux response to increasing pressures. The metal housing, while detached, was filled with 10 ml of PBS and then reattached to the setup. The resulting flux data are shown in Fig. 4(a) from 20 to 100 psi. At these pressures, PBS has extremely high flow rates through even the 5 kDa polyethersulfone membrane, especially at 90 and 100 psi. At 100 psi, PBS has a flux of about 1500 μl/min/cm², which means that the device would take only 2 min to 15× preconcentrate a 3 ml sample of water at 100 psi to 100 μl using a membrane area of only 1 cm², assuming a case of 50% analyte loss to fouling. The relationship between applied pressure and flux is fairly linear, as was expected.

Next, real biofluids of saliva and urine were tested for flux with the device of Fig. 1(b). Saliva was important to test in addition to urine as it is highly relevant for influenza diagnostics. Many studies are demonstrating that saliva may be a viable fluid for diagnosing upper respiratory infections in a comparably accurate, more comfortable, and less invasive manner than using a nasopharyngeal swab.^7,24–26 Before flux testing, the saliva was filtered of viscosity-increasing mucins (using commercial saliva filters, PureSAL, Oasis Diagnostics Corp.).²⁷ 

As can be seen in Fig. 4(b), the flux values seen for urine, while still linear, were ∼7× lower than those for PBS, likely due to fouling and caking of urine solutes against the membrane. Flux values of about 205 μl/min/cm² were seen using a pressure of 100 psi, which would increase the theoretical device concentration time for 3 ml to around 15 min. Saliva, since it is much more viscous, has a low flux of 41 μl/min/cm² at 90 psi. Faster times are certainly possible with application of higher pressures (100's psi) or use of a looser membrane such as the 30 nm membrane used in the previous vacuum devices (Fig. 3).

Preconcentration with the device of Fig. 1(b) was also validated and the experiment setup such that a maximum of 35× preconcentration was possible (7 ml starting sample, 0.2 ml preconcentrated sample). Urine was spiked at 15 ng/ml with influenza A nucleoprotein and tested for preconcentration. Although urine is obviously not relevant for influenza, it was tested simply to generally confirm the potential for viral detection in urine (e.g., measles, Zika virus, etc.^28,29). After the sample fluid had run through the membrane and preconcentration was complete, the setup was disassembled and a micropipette was used to transfer 70 μl of fluid to an LFA for concentration quantification. In addition, 70 μl of the original sample fluid was placed onto a separate LFA for before/after comparison. The LFA test results for various applied pressures can be seen superimposed over the flux data in Fig. 4(b). The test results were extremely consistent using this device, and each LFA before/after pair showed a concentration result of 33×. This indicates that with a pressure of 100 psi, for a low viscosity biofluid such as urine, 10× preconcentration of 1 ml sample could be achieved in ∼5 min, or even faster with higher applied pressures. We, therefore, consider the results for high positive pressure-driven preconcentration to constitute a positive scientific result from an applications perspective. Finally, we note that the analyte loss with the 5 kDa membrane and urine was much less than observed before, since concentration factors of 33× were achieved with a maximum possible concentration factor of 35×. Therefore, the analyte loss for these high positive pressure device experiments was only around 6%, rather than 50% as with the negative pressure devices. This suggests, that in practice, each analyte, biofluid, and membrane combination must be carefully tested to determine the magnitude of analyte loss.

This study revealed both negative and positive results from an application perspective. It is clear that vacuum driven auto-staging preconcentration devices, at least with the materials and methods available for this study, would arguably be too slow for many if not most point-of-care testing applications. It is also clear that a simple bench-top high positive pressure-driven preconcentration system could have applied value. Based on these observations, a proposed picture of benchtop preconcentration device has been developed (Fig. 5). The device relies on a simple disposable sample cup and in theory would be programmable to any amount of the desired preconcentration. The membrane placed in the device would be chosen with regard to the analyte being concentrated, and, therefore, would be application-specific. With higher pressures (100′s of psi), such a device could potentially complete preconcentration within one minute for fluids such as urine. Such a device would be likely be most relevant for large analytes (1's to 10's kDa or larger) because smaller analytes will require membranes with smaller pore-sizes which result in low water fluxes and very slow preconcentration times.

See the supplementary material for a manual containing the necessary materials as well as a detailed step-by-step fabrication process for the vacuum-driven autostaging device.

The University of Cincinnati would like to acknowledge support from the Air Force Research Labs (USAF Contract No. FA8650-16-C-6760) and the Ohio Federal Research Network (Nos. PO FY16-049 and WSARC-1077-700). The 711th Human Performance Wing authors acknowledge the Air Force Research Laboratory for funding.

The data that support the findings of this study are available from the corresponding author upon reasonable request.