Abstract
Photonic ring resonance is a property of light where in certain circumstances specific wavelengths are trapped in a ring resonator. Sensors based on silicon photonic ring resonators function by detecting the interaction between light circulating inside the sensor and matter deposited on the sensor surface. Binding of biological material results in a localized change in refractive index on the sensor surface, which affects the circulating optical field extending beyond the sensor boundary. That is, the resonant wavelength will change when the refractive index of the medium around the ring resonator changes. Ring resonators can be fabricated onto small silicon chips, allowing development of a miniature multiplex array of ring based biosensors. This paper describes the properties of such a system when responding to the refractive index changed in a simple and precise way by changing the ionic strength of the surrounding media, and in a more useful way by the binding of macromolecules to the surface above the resonators. Specifically, a capture immunoassay is described that measures the change of resonant wavelength as a patient serum sample with anti-SS-A autoantibodies is flowed over a chip spotted with SS-A antigen and amplified with anti-IgG. The technology has been miniaturized and etched into a 4 × 6 mm silicon chip that can measure 32 different reactions in quadruplicate simultaneously. The variability between 128 rings on a chip as measured by 2 M salt assays averaged 0.6% CV. The output of the assays is the average shift per cluster of 4 rings, and the assays averaged 0.5% CV between clusters. The variability between chips averaged 1.8%. Running the same array on multiple instruments showed that after some improvements to the wavelength referencing system, the upper boundary of variation was 3% between 13 different instruments. The immunoassay displayed about 2% higher variability than the salt assays. There are several outstanding features of this system. The amount of antigen used on the chip for each test is around 200 picograms, only a few microliters of sample is necessary, and the assays take < 10 min.
1. Introduction
1.1. Optical ring resonance
Optical ring resonance is a property of light that yields the removal of specific wavelengths when light enters a circular waveguide, called a ring resonator. Specifically, wavelengths of light that are exactly equal to the circumference of the ring divided by an integer, times the refractive index of the surrounding media, will become trapped and resonate within the ring, while all other wavelengths of light can leave the resonator (Iqbal et al., 2010, Ksendzov and Lin, 2005, De Vos et al., 2007, Kwon and Steier, 2008, Luchansky et al., 2010) (Fig. 1a). The resonant wavelengths that are trapped in the ring leave a negative peak in the spectrum of light leaving the ring. See Appendix A for a more detailed technical description of ring resonance and the principles of operation of the Maverick instrument.
The waveguide can be made in such a way that a portion of the light energy extends beyond the surface of the waveguide in the form of an evanescent tail that interacts with the material in the immediate proximity of the waveguide. Any matter that changes the index of refraction will change the resonant wavelengths in the ring resonator. It follows that when the refractive index of the surrounding media changes, the wavelengths of light that remain trapped in the ring resonator will change accordingly. The resonant wavelengths will shift proportionately higher as more matter is deposited above the ring (Fig. 1b). Thus, binding of material including protein and DNA can be detected directly since they have higher refractive indices than water (Washburn et al., 2010, Qavi et al., 2011). To enhance and amplify the signal, polystyrene beads (Luchansky et al., 2011, Iqbal et al., 2015) or enzymatic deposition of an insoluble precipitate (Kindt et al. 2013) above the rings can be used.
Several proof of concept studies have been published on the M1 instrument, an earlier version of the Maverick™ detection system used in this paper. These include refractive-index detection of molecules eluted in liquid chromatography (Wade and Bailey 2014), measurement of interactions between proteins and lipid bilayer nanodiscs (Sloan et al. 2013), detection of bacterial tmRNA (Scheler et al. 2012), and kinetic analysis of thrombin binding molecules (Byeon and Bailey, 2011), among others. An assay for detection of monocyte chemotactic protein 1 was validated (Valera et al. 2015). The system has been recently reviewed (Kindt and Bailey 2013). Kinetic analysis of CD200 binding to CD200-R and preliminary results of some immunoassays were performed on the Maverick instrument (Iqbal et al. 2015). The main difference between the M1 instrument and the Maverick is that the M1 ran a single chip at a time while the Maverick automatically runs up to 12 chips one after the other.
1.2. The silicon chip
A 4 × 6 mm silicon chip is manufactured in the same fabrication plant that manufactures computer chips, but in this case the chip is designed to manipulate light rather than electricity, that is photons are directed through the chip instead of electrons. There are 136 rings etched onto the chip. Each ring is 30 μm in diameter. Eight are used as controls and remained covered by a Teflon-like per-fluoropolymer that coats most of the chip. The other 128 rings are organized into 32 clusters of 4 rings each, which are not covered and so are exposed to the material that flows above them. These rings are functionalized by spotting test-specific reagents on to them, which act as analyte specific capture probes. Thus, each test is run in quadruplicate (Fig. 2). The clusters are organized in 2 rows with 16 clusters in each row. The gasket covering each chip has a flow channel for each row of clusters. Thus, 2 samples can be tested for 16 assays each, or 1 sample can be tested on 32 assays in the current version of the chip.
When the assay is being performed, light enters the chip through grating couplers, travels through waveguides to the ring resonators, and then returns to the grating couplers where the intensity at each wavelength of the returning light is measured. Negative peaks in the intensity of light indicate the resonant wavelengths, and the shift in the wavelengths of the negative peaks indicate a change in the refractive index above the ring cluster, which in turn is proportional to the mass that has bound to the reagent above the cluster. A much more detailed technical description of the Maverick system is in Appendix A.
The measured signal is the number of picometers (pm) the resonance peak shifts during the course of the assay. The shift in each ring is measured separately, and the average of the shifts from the 4 rings in the cluster is output as Genalyte Response Units (GRU). An outlier within a cluster can be removed automatically by the software by applying Chauvenet’s criterion if it is > 3 standard deviations from the mean.
1.3. Spotting the chip with reagents
Placing analytes onto Genalyte’s silicon chips is done by using the sciFLEXARRAYER S5 (Scienion, AG, Berlin). The Scienion spotter uses a glass piezo-electric nozzle (PDC-70), which dispenses 280–360 pl per drop. The on-board camera ensures that each drop is precisely placed on a cluster by using real time image recognition of fiducial marks on each chip and drop trajectory after each dispense. The spotter is enclosed in a glass chamber with humidity control to minimize static charge and to keep the spotted analytes from drying too quickly.
Other techniques for activating the silicon surface of these chips to allow binding of biological macromolecules have been described (Kirk et al., 2013).
1.4. Gaskets and reagents for arrays
In order to be used in the Maverick instrument, up to 12 functionalized chips are placed into a gasket that holds the chips in place (Fig. 2) so that they can be exposed to the scanning instrumentation of the Maverick, see description below. Aspiration tubes in the gasket allow liquid to be drawn up from the sample wells, and microfluidic channels allow the reagents to flow over the chips and eventually go into a waste container. The gaskets are paired with a 96 well reagent plate that contains diluted samples, wash buffers and amplification reagents. Since there are 12 chips per gasket, 2 channels per chip and up to 16 tests per channel, 384 tests can be performed on the chips in one array. These tests could be organized so that different samples are run on each channel, allowing 24 samples to be tested on 16 assays each. In the opposite extreme, different reagents could be spotted on every cluster on every chip, allowing 1 sample to be tested on 384 different assays.
1.5. The Maverick instrument
The Maverick is designed to automate the entire procedure of running assays once the array with the functionalized chips and the reagent plate with the diluted samples and necessary buffers are placed into the instrument. A USB stick containing the information needed to run the paired array and reagent plate is placed into the instrument, giving the software the recipe for the assay and the specific reagent that is spotted on each cluster. The Maverick contains a main controller board that runs the processes, a continuously variable laser centered around 1550 nm, a beam splitter so that light can go into each of the 136 waveguides, an etalon that is used as a reference so the instrument does not need to be tuned, multiple photo detectors to capture ring and etalon signals, pumps to move the fluid from the reagent plate over the chip, and motors to move the reagent plate so the appropriate buffer is available to the aspiration tubes.
1.6. This study
This study describes the intrinsic properties of the chip and instrument using tests that change the refractive index with different molarities of salt. Arrays using salt steps can be run multiple times, allowing direct measurement of chip to chip and instrument to instrument variability. Those results are compared to an immunoassay to determine the variability that is added to the system by the steps needed to perform a standard sandwich immunoassay.
2. Materials and methods
All chemicals were purchased from Fisher Scientific (Hampton, NH) unless otherwise noted. SS-A antigen (Arotec Diagnostics, Wellington, New Zealand) was prepared at 200 μg/ml in Drycoat Assay Stabilizer without added protein (Virusys Corporation, Taneytown, MD, USA) plus 1% sucrose and spotted onto the chips as described below. Running buffer is 0.15 M NaCl, 10 mM sodium phosphate, 0.09% sodium azide, 0.05% Tween 20, 1% bovine serum albumin and 0.1% goat IgG, pH 7.3. Running buffer is used as the sample diluent and for the 100 μg/ml goat anti-human IgG (Jackson ImmunoResearch Laboratories, Inc., Westgrove, PA, USA) amplification reagent used in the immunoassay to detect IgG anti-SS-A autoantibodies. The high salt buffers for the salt assays were made by adding 0.5 M NaCl or 2.0 M NaCl to the running buffer.
2.1. Spotting the chip with SS-A
Prior to spotting, the chips are placed into an anodized chip rack for processing. The chips go through a series of washes for 2 min each: acetone to remove the photoresist that protects the chip surface, amino silane in acetone,(3-aminopropyltriethoxysilane, Thermo Scientific, Waltham, MA) to cover the chip surface with a uniform layer of functional amino groups, and isopropanol followed by water to wash off excess reagents prior to spotting. The chips are then dried using nitrogen gas and transferred onto the spotting deck.
Three drops of about 300 pl of Bis-sulfosuccinimidyl suberate (BS3) (Thermo Scientific, Waltham, MA), an amino-amino homobifunctional cross linker is placed on the desired rings and allowed to completely dry prior to continuing. A further 3 drops of about 300 pl of SS-A at 200 μg/ml in Drycoat plus 1% sucrose is then spotted onto the clusters that were activated with BS3. This is < 0.2 ng/spot of antigen. In this study all 32 clusters on the chip were spotted with SS-A.
After all analytes are spotted, the chips are allowed to dry for 1 h in the humidity chamber. The chips are stored in a sealed, desiccated, nitrogen filled foil bag. Chips are stored at 2–8 °C before and after assembly into a Maverick consumable which consists of a gasket and aspiration tubes.
The blank chips used in the salt array were processed up to the step where they were rinsed and then dried with nitrogen gas. Then they were put into arrays and used in the salt assays.
2.2. Running the salt assays
For the salt assays, all steps flow at 30 μl/min. The equilibrium and baseline are 1 min and the salt step lasts for 1.5 min. First there is an equilibrium step with running buffer to obtain the base resonance for each ring, then a high salt step, then a baseline step with running buffer. Typical sensograms for 0.5 M NaCl and 2.0 M NaCl are shown in Fig. 3a and b. The signal for the salt assays is determined by subtracting the average of the last 30 s of the baseline step from the average of the last 30 s of the preceding salt step.
The salt arrays can be run multiple times each. In this study 4 arrays were used for each salt concentration, and each array was run 4 times. Specifically, each array was run one time each on 3 different instruments and one time on the base instrument used as a control for the instrument to instrument variability study. That yields 16 runs on 13 instruments total.
2.3. Calculating signals for the immunoassay
For the anti-SS-A assay, the serum sample was diluted 1:50 in running buffer. The protocol for performing the autoantibody test is as follows. The chip is equilibrated by flowing running buffer over the chip for 1.5 min at 40 μl/min in order to get a base resonance frequency for each ring, followed by diluted sample for 3 min at 40 μl/min, then a wash step with running buffer for 2 min at 40 μl/min, followed by the amplification step for 3 min at 30 μl/min. The 2 baselines are determined by averaging the wavelength of ring resonance during the last 30 s of the wash and the last 30 s of the amplification step. Thus, the entire assay takes < 10 min per chip. Since the chip is in the reading head of the instrument for the entire assay, an array of 12 chips takes about 2 h to run. The signal is the shift from the first baseline to the second baseline. A typical sensogram showing all 16 clusters from a chip is shown in Fig. 4.
2.4. Amplification with anti-IgG
Amplification with anti-IgG is done for 2 reasons. In complex matrices such as serum and whole blood, molecules other than IgG may bind to the antigen, so the initial shift is not specific. Besides increasing specificity, the anti-IgG amplifies the initial signal from specific IgG binding to the antigen.
2.5. Immunoassay design
Two arrays each were run on 3 instruments. These arrays can only be run once, so no instrument to instrument comparison was made with the SS-A chips. It should be noted that the study of the immunoassay was purposefully performed to minimize any variability except that intrinsic to the Maverick system. A bulk dilution of a high titer serum sample was made, and all clusters on all chips were spotted with the same antigen. Thus, typical sources of variation that would occur in standard testing, such as variability in sample dilution, between users, between days, between lots of reagents, and the variability in low positive samples are not included in this study.
2.6. Calculations of variability
Several levels of precision were measured in this study. Level 1 is the ring to ring variability within a chip measured in pm, yielding 1 value per chip, which is the %CV value for all 128 rings in a chip. Because the output of the assay is the average of the 4 rings in a cluster reported as GRU, level 2 is the within cluster variability, which yields 32%CV values per chip. Level 3 is the cluster to cluster variability in a chip. This is a key evaluation because the value of a cluster is the output of the instrument. Level 4 is the chip to chip variability in an array of 12 chips, yielding one %CV value per array. Level 5 is the total variability in the study based on the GRU of the clusters. The calculation is the mean value of all 32 clusters per chip for the 12 chips in each of the 4 arrays of the salt study and each of the 6 arrays in the immunoassay.
2.7. Instrument to instrument variability
Instrument to instrument variability is calculated by running the same salt array on 4 different instruments. The variability between the same chips run on the 4 different instruments was calculated directly. Each set of 4 instruments could be normalized with the other 3 sets because 1 instrument was common to all sets. Thus, all 13 instruments could be compared to the same base instrument. This was not done with the SS-A arrays because they can only be run once each.
3. Results
Fig. 3a and b shows typical sensograms for all 16 clusters in a channel for the 0.5 M and 2.0 M NaCl assays, respectively. The signal is the difference between the baseline taken in running buffer and the signal in the higher salt buffer. For 0.5 M NaCl and 2.0 M NaCl the average shifts are 295 and 1146 GRU, respectively, for all 4 arrays that were tested on 13 different instruments.
The ring-to ring variability within a chip is calculated by taking the average picometer shift of the 128 individual rings on each chip and calculating the standard deviation and % CV. In 0.5 M NaCl the 192 chips tested in this study averaged 0.87% CV with variability between 0.5% to 1.8% CV. In 2.0 M NaCl the average was 0.65% CV with variability ranging from 0.4% to 1.1% CV. There was a near normal distribution of the % CV around the average, with a small tail at higher values (not shown).
The output of standard Genalyte assays is GRU, which are calculated by measuring the picometer shift in each of the 4 rings in a cluster, removing an outlier if present, and then averaging all remaining rings. This is done for each of the 32 clusters in a chip. Thus, the variability in the picometer shifts within the 4 rings in a cluster is a relevant measure of the robustness of the assays. For the salt arrays, the within cluster % CV was between 0.02% to 4.9% for 0.5 M salt and 0.0% to 2.3% for 2 M salt. (Fig. 5a). Because of the large number of samples in these calculations, the ones to the far right of the distribution cannot be seen at this scale.
Cluster to cluster variability within a chip is the most representative value of a regular assay. Precision in GRU from each cluster is slightly better than when measuring all rings independently. For the 0.5 M tests, the variability ranged from 0.3% to 1.7% CV, with an average of 0.60% CV with a tail of the distribution towards the higher side. For the tests with 2.0 M NaCl, the variability ranged from 0.2% to 1.1% CV, with an average of 0.5% CV (Fig. 5b).
Chip to chip variability was determined by taking the sum of the GRU for the 32 clusters of each chip and comparing that to the sum of the other 11 chips in that array. Each array was tested 4 times, and 4 arrays were tested, yielding 16 sets of chip to chip variability data. The chip to chip variability was dependent on the array. For the 0.5 M NaCl study, the chip to chip variability in array 3 ranged from 0.4% to 0.7% CV in the 4 replicate runs. In arrays 1 and 4 the variability was between 1.1% and 1.4% CV. Array 2 showed the highest variability ranging from 1.3% to 1.8% CV (Table 1).
Table 1. Chip to chip variability in 0.5 M NaCl.
| Array # | Instrument # | Chip to Chip Variability (%CV) | |||
|---|---|---|---|---|---|
| 1 | 22, 16, 15, 21 | 1.4% | 1.1% | 1.4% | 1.4% |
| 2 | 28, 22, 24, 34 | 1.3% | 1.5% | 1.8% | 1.5% |
| 3 | 30, 25, 22, 03 | 0.4% | 0.5% | 0.7% | 0.4% |
| 4 | E1, 33, 53, 22 | 1.3% | 1.3% | 1.2% | 1.1% |
In the set of arrays that were used for the 2.0 M NaCl study the chip to chip variability in array 6 ranged from 0.5% to 0.7% CV in the 4 replicate runs on different instruments. In array 8 the variability was between 1.0% and 1.2% CV. In array 7 it was between 1.0% and 1.5% CV. Array 5 showed the highest variability ranging from 1.3% to 1.8% CV (Table 2).
Table 2. Chip to chip variability in 2 M NaCl.
| Array # | Instrument # | Chip to Chip Variability (%CV) | |||
|---|---|---|---|---|---|
| 5 | 22, 16, 15, 21 | 1.6% | 1.3% | 1.8% | 1.7% |
| 6 | 28, 22, 24, 34 | 0.6% | 0.5% | 0.7% | 0.6% |
| 7 | 30, 25, 22, 03 | 1.5% | 1.0% | 1.2% | 1.2% |
| 8 | E1, 33, 53, 22 | 1.1% | 1.2% | 1.1% | 1.0% |
Instrument to instrument variability was determined by comparing identical chips run on different instruments. These comparisons could be made in sets of 3 instruments compared to the base instrument. The differences between the 12 instruments and the base instrument when running the 0.5 M NaCl arrays were between − 1.7% and + 7.0% (Table 3). With the 2.0 M NaCl arrays the differences ranged from − 1.0% to + 6.0%. The rank order of differences between test instruments and the base instrument was nearly identical in the 2 studies.
Table 3. Instrument to instrument variability.
| Difference from instrument 22 (control) comparing %CV from mean of all rings | ||
|---|---|---|
| Instrument # | 0.5 M NaCl | 2 M NaCl |
| 15 | 7.0% | 6.0% |
| 16 | 5.2% | 4.2% |
| 30 | 3.6% | 3.8% |
| 52 | 1.6% | 2.3% |
| 25 | 1.9% | 1.3% |
| 24 | − 0.1% | 1.0% |
| 28 | − 0.5% | 0.9% |
| 22 | 0.0% | 0.0% |
| E1 | -0.3% | − 0.3% |
| 33 | − 0.3% | − 0.8% |
| 53 | − 0.8% | − 0.9% |
| 21 | − 1.2% | − 0.9% |
| 03 | − 1.7% | − 1.0% |
Optimization of instrument to instrument variability was performed by adjusting the etalon in the 3 instruments with the highest variability. The improvements brought the differences of the test instruments to the base instrument from 6.0%, 4.2% and 3.8% to 1.4%, 3.0% and − 0.1%, respectively.
Precision studies with anti-SS-A antibodies were analyzed the same way as the salt data, except no instrument to instrument comparison was made because the immunoassays could only be run one time. Six arrays were tested, 2 each on 3 different instruments. A typical sensogram showing the shifts at each step of the assay for all 16 clusters in a channel is shown in Fig. 4.
The ring to ring variability within a chip was not much higher than with the salt assays. On average it was 2.4%, ranging from 0.6% to 6.4% for individual chips. For all chips in the arrays the shifts ranged from 563 to 631 pm.
The ring to ring variability within a cluster is very small with 99.6% of the clusters < 5% CV. The range was from 0.04% to 16.0% (Fig. 6a). Because of the distribution of the value of the rings in the sample with 16.0% CV, outlier removal could not be performed. This sample is not visible on the graph due to the scale.
The cluster to cluster variability within a chip was slightly better than the ring to ring variability, averaging 2.2% CV with a range of 0.3% to 6.1% CV (Fig. 6b).
Chip to chip variability within an array ranged from 2.3% to 3.7% CV. When all 72 chips were compared there was 5.0% CV (Fig. 7).
3.1. Outlier removal
Only a small number of outliers were removed. For the 0.5 M NaCl study, the 2.0 M NaCl study and the SS-A study, just 0.35%, 0.17%, and 0.15% of the rings were removed, respectively. In one instrument, some clusters in the salt arrays did not register and were thus not measured; 0.21% for 0.5 M NaCl and 0.05% for 2.0 M NaCl. All clusters registered in the immunoassay but one cluster was removed as statistical outlier because it was > 4 standard deviation from the mean of the chip.
4. Discussion and conclusion
The results of this study show that the precision inherent to the instrument and the chips is sufficient for use in immunoassays. Previous studies on the M1 have detailed some of the physical properties of ring resonance on a chip, such as the range of sensitivity above the surface, 1/e is 63 nm. Also, several proof of concept assays and a validated assay for the Maverick have been described. Here, we have studied the intrinsic precision of the individual rings fabricated onto the chip in order to determine how suited this novel technology is for performing immunoassays in a clinical setting. The low variability in the signals between the clusters in a chip, between chips and between instruments show that the system is robust. This good precision is expected, since silicon chips are manufactured with high precision, and etalons can be used to precisely calibrate wavelengths of light.
Before optimization of the instruments, the greatest variability between any instrument and the base instrument was < 9%. After optimization, the largest difference was 3%. With further work on the procedure, we expect to reduce instrument to instrument variability even further.
Just as the baseline of variability in the system was determined by running the salt arrays, a baseline for variability in an immunoassay was determined by running a high titer sample over chips where every cluster was spotted with the same antigen, SS-A. The format of this immunoassay was chosen to yield the best possible precision. In this system the immunoassays showed only a few percent higher CV than the salt system. The difference between this system and a typical precision study on a multiplexed immunoassay is that usually there would only be one cluster per antigen in a channel, and other sources of variability such as dilution, chip to chip, lot to lot and instrument to instrument would be measured. We expect there will be higher variability in that situation. A full standard validation of a multiplex immunoassay on whole blood in addition to serum samples will be the focus of another publication.
The Maverick has performance characteristics that will allow it to be used for immunoassays in an outpatient clinic. Preliminary results show that whole blood yields the same results as serum and plasma in immunoassays. This allows the test to be run in a near-patient setting since the sample does not need to be sent to a lab for processing into serum or plasma. Because results can be delivered in < 15 min from the time the blood is drawn to the time the results are reported out from the Maverick, the next studies will focus on clinical trials using validated multiplex immunoassays.
The Maverick has several advantages in a clinical setting compared to other recent multiplexed technologies such as Meso Scale Discovery (MSD). The Maverick has a smaller foot print and can thus better fit into a place with limited space such as a physician’s office. Perhaps the most important differences between the instruments are the ease of use and time to results. The MSD requires processing of a blood sample into serum followed by two 2 h incubations, with wash steps in between. The MSD instrument itself does not do these steps automatically. In contrast, the Maverick can use whole blood and is completely automated after the array and the reagent tray with diluted patient sample are placed into the instrument, including running the assay steps, analyzing the data and reporting the results. In addition, results from 1 chip can be reported in < 15 min from the time the blood is drawn, allowing doctors and patients to have results at the time of an exam.
Read More: https://www.sciencedirect.com/science/article/pii/S002217591630388X?via%3Dihub