A Flexible, Robust Microbead-based Assay for Quantification and Normalization of Target Protein Concentrations

Abstract

Although there are many methods for quantifying the concentration of specific proteins in samples, current techniques are technically challenging or do not easily lend themselves to normalization. Here, we describe a microbead-based assay for quantifying specific protein concentration(s) that is high-throughput, inexpensive, simple-to-use, and intrinsically incorporates normalization against the sample total protein content. This assay, termed the FRANC assay, exploits high affinity biotin-streptavidin binding to couple sample proteins to streptavidin-labeled magnetic microbeads. Proteins are then antibody-probed, followed by labeling of proteins on the microbead with fluorescent dye, and flow cytometry-based analysis. The FRANC assay demonstrates detection limits for target proteins in the femtogram range, with a linear range up to as much as 10 nanograms. Normalization of target protein concentrations resulted in an 80% reduction in variability as compared to non-normalized measurements. We conclude that the FRANC assay offers attractive advantages over current methods for quantifying specific protein(s) in samples.

Graphical Abstract

1. Introduction

Quantifying the concentration of specific (target) proteins in a complex sample is an essential analytical technique for many biomedical applications. Samples typically consist of lysed tissue or other complex mixtures of multiple proteins¹, and current quantification methods employ antibody-based labelling and fluorescent, luminescent, or colorimetric determination of protein amounts^2-3. For example, western blotting is commonly used for protein quantification^4-5; it has the advantage of size separation by molecular weight, but the associated protein transfer can be technically challenging and slow ^{2, 6-7}. Alternatively, enzyme-linked immunosorbent assays (ELISA), fluorescent microbead arrays, and related multiplex assays are now commonly used, as they allow for higher throughput ^8-10. The disadvantage of such assays, however, is limited flexibility, since kits containing validated antibody pairs for each protein of interest are often required.

An important consideration in all analytical techniques is sample normalization, i.e. the reporting of the concentration of the specific protein of interest in relation to levels of one or more reference proteins. Normalization accounts for user errors such as pipetting, as well as sample-to-sample variability, and is routinely performed in Western blot analysis by normalizing to a single reference protein (e.g. GAPDH, β-actin)^11-13. While single reference proteins are sufficient for normalization in certain applications, reference protein amounts often vary between samples, making a single reference protein inadequate for sample normalization in many situations^14-15. As a result, total protein-based normalization has been incorporated into Western blot analysis using different fluorescent or colorimetric total protein stains^16-19. ELISAs and multiplex assays, however, are not well suited to total protein-based normalization.

Here, we describe a microbead-based platform to quantify the concentration of one or more specific (target) proteins within a sample. It combines the beneficial features of multiplex assays with robust normalization against the total concentration of protein within each assayed sample. We refer to this as the FRANC assay (Flexible, Robust Assay for quantification and Normalization of target protein Concentration). Very briefly, the assay proceeds in four steps. First, a protein sample is attached to microbeads in an unbiased manner using biotin-streptavidin conjugation. Next, the target protein(s) of interest are detected using standard fluorophore-conjugated antibodies, after which all proteins on the microbeads are fluorescently labelled using amine-containing fluorescent dyes attaching to carboxylic acid groups (“total protein labeling”). Finally, bead-based flow cytometry reads fluorescent signals from: (1) target protein(s), and (2) total protein, allowing for direct normalization of signal (1) by signal (2). Below we describe the technology in detail and then characterize the performance of the FRANC assay.

2. Methods

2.1. Sample preparation and sourcing

A variety of protein samples, representing a range of situations encountered in laboratory work, was used for testing the FRANC assay. Goat (LAMPIRE Biological Laboratories, Pipersville, PA, USA), rabbit (Gibco, Waltham, MA, USA) and mouse (Invitrogen, Waltham, MA, USA) sera were obtained commercially. Primary adipose-derived mesenchymal stem cells (Lonza, Basel, Switzerland) were grown to confluence and were then lysed using RIPA buffer (Thermo-Fisher, Waltham, MA, USA) supplemented with Halt protease inhibitors (Thermo-Fisher, Waltham, MA, USA). Similarly, immortalized HT-1080 fibrosarcoma cells (American Type Culture Collection, Manassas, VA, USA) were lysed without treatment or after 24 hours of treatment with 25 ng/mL phorbol 12-myristate 13-acetate (PMA, MilliporeSigma, St. Louis, MO, USA)³⁰. PMA is known to activate protein kinase C and alter downstream protein expression, such as matrix metalloproteinases^30-31. Recombinant TNF-alpha and IL-1β in 0.1% (w/v) bovine serum albumin carrier protein (R&D Systems, Minneapolis, MN, USA) were reconstituted in phosphate buffered saline prior to use. The total amount of protein in each sample was quantified by the bicinchoninic acid assay (BCA, Thermo-Fisher, Waltham, MA, USA) per the manufacturer’s protocol using a bovine serum albumin standard curve, and a BCA reagent incubation time of 30 minutes at 37°C.

2.2. FRANC Assay Method

We first conjugated protein to microbeads. Sulfo-NHS-LC-Biotin (sulfosuccinimidyl-6-[biotin-amido]hexanoate, Thermo-Fisher, Waltham, MA, USA) was incubated with 10 μg of protein sample at a mass ratio of 32:1 sample protein:Sulfo-NHS-LC-Biotin for 1 hour at room temperature. These biotinylated samples were diluted as desired and loaded into 96 well plates (50μL sample per well). Streptavidin-coated hydrophobic magnetic microbeads (Dynabeads MyOne Streptavidin T1 [10mg/mL], Thermo-Fisher, Waltham, MA, USA) of 1.05 μm diameter were washed twice with 0.5% (v/v) Tween™-20 (Fisher Scientific, Waltham, MA, USA) phosphate buffered saline (referred to as T-PBS) by adding 1mL of T-PBS to 12μL of microbeads and subsequent microbead removal from solution using a neodymium bar magnet (0.75” length × 0.25” width × 0.25” height, 10.5 lbs pull force, K&J Magnetics, Pipersville, PA, USA). Microbeads were then resuspended in T-PBS at 20μg/mL (500 dilution factor) and 50 μL of bead suspension was added to each protein sample. Protein samples and microbeads were then allowed to incubate for 30 minutes at room temperature.

After incubation, microbeads were magnetically removed from suspension using a custom-made 96 well magnetic plate separator (96 ¼” diameter × ½” thick cylindrical magnets assembled in a 96-well plate, 4.8 lbs. pull strength for each magnet, K&J Magnetics, Pipersville, PA, USA). Specifically, the 96-well plate was placed on top of the magnetic separator for 30 seconds, after which the solution was removed from each well by inverting the plate over a waste reservoir. Microbeads were washed twice with 100μL of T-PBS and incubated with 5μL of the appropriate antibody solution (1μ/mL antibody concentration in T-PBS) for 30 minutes at room temperature, followed by washing with T-PBS.

To detect all proteins in the sample (“total protein”) by carbodiimide crosslinking, 50μL of 1 mM 1-ethyl-3-(3-dimethylaminopropyl) carbodiimide (EDC, MilliporeSigma, St. Louis, MO, USA) and 2.5 mM N-hydroxysulfosuccinimide (sulfo-NHS, MilliporeSigma, St. Louis, MO, USA) in T-PBS were added to microbead suspensions and immediately agitated and incubated for 15 minutes at room temperature. Following two T-PBS washes, 50 μL of 50 μg/mL 4’-(aminomethyl) fluorescein (AMF, Thermo-Fisher, Waltham, MA, USA) dye in T-PBS was added to the microbead suspension and incubated for 2 hours at room temperature under agitation. After 2 additional T-PBS washes, microbeads were resuspended in 200 μL T-PBS and analyzed by flow cytometry (Attune NxT, 12.5ul/min, 25μl volume, Thermo-Fisher, Waltham, MA, USA).

2.3. FRANC Assay Limit of Detection

To determine the limit of detection of the FRANC assay, recombinant proteins of known concentrations were used. Human recombinant TNF-α was conjugated to microbeads over a 24-step protein dilution curve, with 2-fold dilutions at each step, and microbeads at each dilution were exposed to an anti-TNF-α fluorescently-conjugated (PE-Cy7) primary antibody (eBioscience, Waltham, MA, USA). Similarly, we tested human recombinant IL-1β, with the exception that a non-fluorescent anti-IL-1β primary antibody (eBioscience, Waltham, MA, USA) was used followed by three T-PBS wash steps and fluorescent (PE) secondary antibody labeling (Abcam, Cambridge, United Kingdom).

The linear range was determined from plots of logarithmically transformed antibody-specific labeling (y-axis) vs. logarithmically transformed amount of sample protein (x-axis), and was defined as the largest possible sample protein region (x-axis) over which: (1) y-values were greater than vehicle-only controls, and (2) coefficient of determination values (R²) for linear regression of y on x were greater than 0.9. Log-log transformation was utilized to allow analysis of faint antibody signals and small protein loading amounts that are often under-weighted in traditional linear analysis. The smallest protein amount within the linear region for which the fluorescent signal was greater than three times the standard deviation of the signal from vehicle-only controls was identified as the FRANC assay limit of detection. Three technical replicates were carried out for each recombinant protein type.

2.4. Comparing total protein labeling for different protein samples

A key aspect of the FRANC assay is labeling of total protein in the sample. We compared total protein labeling data from the FRANC assay to those from the widely-used bicinchoninic acid (BCA) assay. Goat, rabbit, and mouse sera, as well as cell lysates (mesenchymal stem cells, HT-1080 cells treated with PMA, and HT-1080 untreated cells) were biotinylated, conjugated to microbeads, and then diluted into 96 well plates (2-fold serial dilution over 11 steps). Total protein labeling was performed, followed by flow cytometric analysis. The linear regions (R² > 0.90) from plots of total protein fluorescent labeling vs. loaded protein amount were determined for each sample type, as described above. Ideally, if all proteins samples were measured identically by the total protein labeling, the resulting slopes for each sample’s linear range would be equal, whereas any differences in slope values would suggest that the amino acid composition of the assayed proteins influenced the total protein labeling intensity.

For comparison, the same methodology was utilized with the BCA assay. Each protein sample was prepared for the BCA assay and sample absorbance was measured using a microplate reader (Bio-Tek Synergy H4, Winooski, VT, USA). Results were analyzed similar to total protein labeling, as described above, for each sample. However, the traditional methodology for determining the BCA assay linear range uses linear regression without the log-log transformations used when analyzing the FRANC assay. To ensure a balanced comparison, BCA and FRANC assay results were compared using both methods. The BCA reagent absorbance was plotted vs. standardized sample concentration, and the slope of the linear region was determined. The variability of slope values for the BCA assay was compared to the variability of slope values from the FRANC assay’s total protein labeling to determine the FRANC assay’s performance.

2.5. FRANC Assay Normalization Capabilities

Finally, the normalization capabilities of the FRANC assay were assessed. For simplicity, rabbit serum was initially used as the protein sample. Biotinylated rabbit serum over a range of concentrations was conjugated to microbeads, and antibody labeling was performed for each microbead sample, followed by total protein labeling. For antibody labeling, rabbit immunoglobulin G (IgG) in the sample was detected using a commercial fluorescent antibody (Donkey F[ab']2 Anti-Rabbit IgG – Phycoerythrin [PE], Abcam, Cambridge, United Kingdom)^{9, 32-33}. Antibody labelling and total protein labelling values over the concentration range were obtained by flow cytometry, and the linear range was determined for antibody signal vs. loaded sample protein amount. We then normalized the antibody-specific protein signal by the total protein signal across the linear range. The slope of the log-log transformed antibody signal vs. loaded sample protein amount was calculated with and without total protein normalization to determine whether normalization reduced the experimental variability, since a slope closer to 0 implies that the antibody signal is less sensitive to the loaded sample amount. Further, antibody signal across the linear range was normalized relative to actual loaded protein amount, as determined separately by the BCA assay, to compare FRANC assay performance to BCA normalization.

2.6. Statistics

To evaluate treatment effects between 3 or more experimental groups, ANOVA was used. The normality of distributions, required for use of ANOVA, was assessed using the Shapiro-Wilk test, with a rejection threshold of p < 0.05. To assess whether variances were equal between groups, also required for ANOVA, we used the Brown-Forsythe test, with a rejection threshold of p < 0.05. When ANOVA was suitable, Tukey’s post hoc test was used to compare all groups or Dunnett’s post hoc test was used to compare each group mean to a single control mean, where p < 0.05 indicated significant differences.

3. Results and Discussion

3.1. Overview of FRANC Assay

The FRANC assay we describe here provides a platform for quantification of specific (target) proteins within a complex mixture while robustly accounting for sample variability through total protein normalization in a manner that is integrated with the overall assay. The assay proceeds in four main steps. First, all proteins in a sample are attached to the surface of a microbead (Figure 1). A key aspect of the assay is the use of biotin-streptavidin interactions for sample protein conjugation to microbeads. By conjugating biotin to primary amine groups in a protein sample using sulfo-NHS chemistry, we can exploit the strong affinity between biotin and streptavidin, ensuring that all proteins in a sample bind to the streptavidin-coated microbead surface with similar affinities, regardless of protein structure. This key step overcomes a limitation of passive surface adsorption^20-24, i.e. the differential affinity of different proteins to passively bind to microbeads.

Figure 1. Schematic Overview of the FRANC Assay Process.

(A) Protein samples are (B) biotinylated and (C) conjugated to magnetic microbeads through biotin-streptavidin interactions. (D) Microbeads are then exposed to fluorescently-labelled antibodies to tag specific proteins of interest, and (E) fluorescently labeled to determine total protein amount, which is required to normalize against protein amount. (F-H) Microbeads are analyzed by flow cytometry. (F) Forward vs. side scatter plots are used to distinguish singlet microbeads from aggregates. Both (G) antibody and (H) total protein signals are detected for each sample-conjugated microbead set.

Second, the microbeads are exposed to one or more fluorescently tagged antibodies to detect target proteins of interest. The FRANC assay is compatible with traditional direct or indirect antibody-based fluorescent labeling of specific proteins of interest. While multiplex tests have not been performed, the FRANC assay is also expected to be compatible with a wide range of fluorophores, including simultaneous multi-antibody labeling which is commonly performed with 10 or more probes in traditional flow cytometry²⁵.

Third, a fluorescent dye is used to conjugate the total protein attached to microbeads, essential for sample protein normalization. While single reference proteins are often used in western blot analysis for normalization, the relative concentrations of many reference proteins varies across experimental conditions^12-14. Total protein labeling offers a more robust normalization method^17-18. For the FRANC assay, total protein labeling was performed by conjugating primary amine-containing fluorescent dyes to carboxylic acid groups.

Finally, the microbeads are analyzed by flow cytometry to detect both the fluorescent signal from bound antibodies and the dye signal from all labelled proteins. Signal normalization can then be performed by expressing the antibody-specific signal as a ratio relative to the signal from all labelled proteins. For ease of sample handling, magnetic microbeads are used in the FRANC assay, since they can be quickly removed from solutions without sample loss using a 96-well magnetic plate.

The protocol we present here was arrived at by an extensive series of optimization steps, in which we varied all key parameters in the assay (incubation times, concentrations, label types, etc.). For brevity we simply report the optimized approach.

3.2. Limit of Detection of the Assay

To characterize the FRANC assay’s overall performance, we first characterized the assay’s limit of detection with two different recombinant proteins having different antibody labeling chemistry. Specifically, recombinant TNF-α was biotinylated and incubated with microbeads and was then detected with an anti-TNF-α antibody conjugated to a fluorophore (PE-Cy7). This resulted in a limit of detection of 60 picograms of TNF-α protein and a linear range of 60 pg - 12.5 ng (Figure 2A). Additionally, recombinant IL-1β was detected with a non-fluorescent primary antibody and a fluorescent secondary antibody in two separate steps, producing a limit of detection of 70 femtograms (fg) and a linear range of 70 fg to 34 pg (Figure 2B). The improved limit of detection when using a secondary antibody is consistent with secondary labeling typically yielding more fluorophores per epitope.

Figure 2. Determination of the limit of detection for antibody labeling.

Antibody-specific fluorescent label vs. protein amount for (A) TNF-α recombinant protein detected by PE-Cy7 fluorescently-conjugated primary antibody and (B) IL-1β recombinant protein with unlabeled primary antibody followed by PE fluorescently-conjugated secondary antibody. The linear range is shown by a regression fit (solid line) to the log-transformed data. The gray region on each plot represents 3 times the standard deviation of no protein (blank) controls. The limit of detection is defined as the upper limit of this region. Error bars denote standard deviations (n=3 technical replicates for each, in some cases, error bars cannot be seen because the error is smaller than the data point).

3.3. Variability of total protein labeling for different protein samples

It was important to determine how consistent total protein labeling (quantification) was between different samples. Quantification of total protein signal is inherently challenging due to variation in amino acid composition across different proteins. If the labeling chemistry is highly sensitive to amino acid structure, different proteins will be “weighted” differently during quantification, resulting in differences between protein sample types. To validate the FRANC assay, it was necessary to quantify total protein concentrations in samples using an independent protein assay. For this purpose, we chose the widely-used BCA assay²⁶. However, even the BCA assay is known to give results that vary between protein sample types²⁶. Thus, the goal for the FRANC assay’s total protein quantification was to demonstrate inter-sample variability similar (or less than) to the BCA assay.

For this aspect of testing, we used six samples: three different sera (rabbit, goat, and mouse), primary mesenchymal stem cell (MSC) lysates, and cell lysates from untreated and from experimentally-treated HT-1080 immortalized cells. We plotted FRANC total protein signal (or BCA assay absorbance) vs. each sample’s protein loading amount and computed the slope of each plot over the linear region. If the assays were insensitive to protein sample composition, all slopes would be identical; in reality, this was not the case and we therefore computed a coefficient of variation for the slopes over the six samples as a measure of the assay’s sensitivity to sample composition. When using a linear regression methodology (conventional for the BCA assay; see methods) the coefficient of variation was lower with the FRANC assay (0.72 FRANC vs. 0.92 BCA). Comparing the slopes for the individual samples, 5 of the 6 samples had similar slopes for both assays, suggesting the FRANC assay performed similar to, or slightly better than, the BCA assay (Figure 3A).

Figure 3. Analysis of total protein labeling for different protein samples with the FRANC and BCA assay.

Slopes of the linear region for total protein signal vs. loaded protein amount for 6 different protein samples as measured by the FRANC assay and traditional BCA assay. Error bars denote standard error (n=2 technical replicates for each) (A) Results determined by traditional linear regression, as commonly used with BCA assay, and (B) determined using logarithmic transformation of total protein signal and loaded protein amount, as used throughout with the FRANC assay.

An alternative analysis approach is to log-transform the protein concentration and assay signal data (see methods) and regress this transformed data; doing so reduced the coefficient of variation for the slopes to 0.38 for the FRANC assay and 0.19 for the BCA assay. When comparing the log-transformed analysis approach to the more conventional direct approach, the FRANC assay performed similarly to the BCA assay for each sample (Figure 3B). Interestingly, log-transformation reduced the coefficient of variation for the BCA method more than it did for the FRANC assay, likely due to the higher variability present at very low protein amounts in the FRANC assay’s linear range (less than 0.1ng protein amount). These protein levels were not detectable with the BCA assay, which showed a limit of detection of approximately ĝ. If end-users required reduced variability for small protein amounts, different functional groups could be targeted, or brighter total protein fluorescent dyes utilized. Further improvements to total protein signal are possible by simultaneously combining multiple chemistries to detect multiple functional groups. In conclusion, while variation between samples was apparent in determination of total protein levels for both the FRANC and BCA assays, the performance of these two assays was comparable, with slightly higher sample-to-sample variability in the FRANC assay being offset by a greater detection limit.

3.4. Total Protein Normalization

Finally, we evaluated the ability of the FRANC assay to robustly normalize antibody signal by the total protein signal. We detected rabbit IgG within a sample of rabbit serum using AMF dye for simultaneous total protein labeling. As expected, the antibody and total protein signals increased with sample protein amount (Figure 4A), with these two signals increasing at comparable rates, as required for normalization. When the antibody signal was normalized by the total protein signal, the ratio was relatively constant across sample protein amount, as desired (Figure 4A). This behavior was quantified by the slope of plots of antibody signal (or antibody signal normalized by total protein signal) vs. sample protein amount, where slope values closer to zero suggest less antibody-labeling signal variability over the linear range. It was observed that normalization reduced these slopes from 0.74 to 0.12 (Figure 4B). For comparison, antibody signal was normalized to loaded rabbit serum sample protein amount, as determined by the conventional BCA assay. This resulted in a slope of −0.25, which is further from the desired value of zero than the FRANC assay’s total protein normalization methodology.

Figure 4. Normalization by total protein reduces assay variability.

(A) A rabbit serum sample was probed with fluorescently-tagged anti-rabbit IgG, followed by measurement of total protein (n=4 technical replicates). Normalized antibody signals were calculated by dividing antibody signal by respective total protein signal across the linear range. Linear range for antibody, total protein, and normalized antibody are shown (solid regression lines) on a log-log plot of assay signal vs. sample protein amount. (B) Slope of regression to fluorescent signal vs. sample protein amount over the linear region for antibody signal, antibody signal normalized to total protein, and antibody signal relative to loaded sample protein amount, as determined by BCA assay (n=4 technical replicates). Error bars denote standard deviation. Statistical significance (p < 0.05) was determined by one-way ANOVA, post-hoc Tukey analysis.

This is an extreme example to demonstrate the normalization capabilities of the FRANC assay, since practical assay applications would not use normalization over such a wide range of protein amounts, but would instead utilize normalization to account for loading error and slight quantification differences between replicates and sample types. Nonetheless, the FRANC assay’s normalization approach reduced variability to about half that observed when using total protein amounts determined separately from the BCA assay. In conclusion, the FRANC assay’s total protein-based normalization was effective at accounting for loading variability and user error that cannot be accounted for when relying on protein quantification that is not fully integrated into the assay.

Overall, the FRANC assay was attractive compared to conventional western blotting, ELISA, and multiplex assays, as summarized in Table 1. Limit of detection was found to be picogram protein amounts for direct antibody labeling, which is comparable to ELISAs, multiplex assays, and conventional western blotting^27-28. Indirect antibody labeling resulted in detection in femto- to picogram quantities, similar to recent advances in microfluidic western blotting that have also reported detection of femtogram quantities²⁹. The FRANC assay requires approximately 5 to 6 hours from start to finish, of which most time is “hands-off” incubations, leading to approximately 1 hour of hands-on time. The cost of the required reagents and quantities needed per 96-well sample are modest, resulting in a cost of approximately $3 per 12-step, single sample dilution scheme. However, it is recommended that end-users confirm reagent concentrations and antibody dilutions for their specific application, which will affect the assay cost. In conclusion, the normalized protein expression microbead assay presented herein is an improved approach that integrates total protein labeling, important for assay normalization, with traditional fluorescent antibody-specific protein labeling.

Table 1.

Overview of the performance of the normalized protein expression assay

Specification	Normalized Protein Expression Microbead Assay
Normalization Capability	Strong. Intrinsic to each assay sample for accounting for sample loading and other user errors
Limit of Detection	TNF-α + PE-Cy7 fluorescent primary antibody: 60 pg IL-1β + unlabeled primary antibody + PE secondary antibody: 70 fg
Assay’s Linear Range	TNF-α : ~60 pg to ~12.5 ng   IL-1β: ~70 fg to ~34 pg
High-Throughput Aspects	Applicable to 96 or 384-well plate formats, multiplex fluorescent antibodies up to flow cytometer limitations
Required Protein Sample Amount	Less than 0.5 μg
Overall Assay Procedure Time	5 to 6 hours from protein sample to flow cytometric analysis
Hands-on Time	Approximately 1 hour
Reagent Cost per Sample	Approximately $3 per 12-step, single sample dilution curve

4. Conclusions

In summary, the FRANC assay we report here beneficially combines a microbead platform, which allows for high-throughput simple protein handling in multiplex assays, with total protein-based signal normalization. Although western blotting can account for loading and sample variability via various normalization approaches, it is slow and technically challenging. Conversely, ELISA and multiplex assay-based protein quantification is much faster, but normalization is not intrinsically feasible within the assay, since only a subset of proteins in the sample (i.e. those recognized by “capture” antibodies on microbeads or well-plates) is assayed. We suggest that the FRANC assay will be of use as an improved method for quantifying specific (target) proteins within a sample.

Highlights.

• The FRANC assay we report here beneficially combines a microbead platform, which allows for high-throughput simple protein handling in multiplex assays, with total protein-based signal normalization.

• The FRANC assay demonstrates detection limits for target proteins in the femtogram range.

• The FRANC assay exhibits an operating range up to as much as 10 nanograms of loaded protein

• Normalization of target protein concentrations resulted in an 80% reduction in variability as compared to non-normalized measurements.

• We suggest that the FRANC assay will be of use as an improved method for quantifying specific proteins within a sample.