R esearch A rticle S pecial Focus I ssue: M atrix
effects in
LBA s
For reprint orders, please contact [email protected]
Evaluating the impact of matrix effects on biomarker assay sensitivity Background: The accuracy of highly sensitive biomarker methods is often confounded by the presence of various circulating endogenous factors in samples causing matrix effects. Method: This article outlines two different biomarker methods: hepcidin enzyme-linked immunosorbent assay (ELISA) for which an orthogonal assessment of ELISA to liquid chromatography–tandem mass spectrometry was performed to examine the potential matrix effect, and sclerostin ELISA to evaluate the matrix effect. Results: Although the potential interfering effects of the endogenous hepcidin variants (prohepcidin and clipped) showed that these proteins had >30% immunoreactivity in ELISA, the hepcidin ELISA preferentially measures full-length hepcidin when the molar ratios of full-length to variants remain >1. The correlation of ELISA to liquid chromatography–tandem mass spectrometry results showed full-length hepcidin as the major form in diseased populations. Conclusion: A fit-for-for-purpose assessment of matrix effect/selectivity was also performed for each method. This article demonstrates the utility of a fit-for-purpose approach to assess the validity of biomarker methods in evaluating the interconnected parameters of matrix effects, sensitivity and selectivity.
Methods to measure soluble ligands as diseaseor drug-related biomarkers are commonly used in both predictive/diagnostic and drug development applications [1]. The use of relevant biomarkers could provide proof of mechanism for target effect, determine proof of concept showing downstream effect, and facilitate patient stratifications or dose escalation [2–5]. Development and validation of biomarker methods by using a ligand-binding assay platform impose distinct challenges relative to methods intended for PK assessment and a comparative analysis of which has been laid out thoroughly in the review articles by Lee [5,6]. Traditionally cited challenges in developing biomarker methods compared with that of PK are the general lack of gold standards or universal references making the selection of a most appropriate reference material an important process and the presence of matrix effects. One of the distinct challenges for the method development and validation for biomarkers is the presence of endogenous interferences. Finally, development of biomarker methods requires an in depth understanding of the biology, thus additional expertise and time are critical for success. Additional significant challenges in biomarker method development are posed by the presence of endogenous biomarkers circulating as propeptides or proteolytically clipped forms. The presence of these forms in disease states
that may or may not be active or biologically relevant yet these may cross-react with the assay reagents. Furthermore, propeptides and clipped forms may circulate at different levels in diseased states than in healthy population and may also have varying degrees of cross-reactivity with the assay reagents. Differential cross-reactivity of these variants may interfere with the accurate quantification of the intended analyte and result in matrix effects in the method. Biomarker methods used during drug development are commonly designed and validated with a specific fit-for-purpose approach [5,7]. To overcome the unwanted matrix effect caused by the presence of endogenous proteins, alternative strategies are typically employed (Figure 1). Since different levels of endogenous biomarkers may be present in the matrices, it is necessary to determine baseline concentrations in both healthy and diseased states, and to identify an alternative species matrix (a surrogate matrix) in which the endogenous proteins do not crossreact with the assay reagents to prepare the standard calibrator. If a particular biomarker protein has a high degree of homology across different species, the selection of a surrogate matrix may not be possible, and in this case, a protein buffer matrix (a substituted matrix) may be used to prepare the standard calibrator. However, this may result in a greater matrix effect in the sera of diseased subjects [5,7]. Therefore, it is
10.4155/BIO.14.55 © 2014 Future Science Ltd
Bioanalysis (2014) 6(8), 1081–1091
Theingi Thway* & Hossein Salimi-Moosavi Department of Pharmacokinetics & Drug Metabolism, Amgen Inc., One Amgen Center Drive, Thousand Oaks, CA 91320, USA *Author for correspondence: Tel.:+1 805 447 1000 Fax: +1 805 499 9027 [email protected]
ISSN 1757-6180
1081
R esearch A rticle |
Thway & Salimi-Moosavi
Is there a measurable endogenous ligand/target level in intended matrices of normal and disease population?
Yes Yes
Yes No
Intended matrices can be used to make standard calibrator
Is it feasible to use surrogate matrix?
Yes Yes
Yes No
Scenario 1: Surrogate matrix that do not cross react with intended species can be used to make standard calibrator while at least serum/plasma controls are made in intended matrix
Scenario 2: Substituted matrix or appropriate buffer can be used to make standard calibrator while at least controls in intended matrix are included to monitor the long-term assay performance
Figure 1. The rationale for the selection of appropriate matrix for standard calibrator in developing a biomarker method.
necessary to determine the accuracy or recoveries of spiked samples in the intended serum at the LLOQ and 3 × LLOQ to ensure relative accurate measurement of the biomarker protein while a substituted matrix is used to prepare the standard calibrator. This article illustrates fit-for-purpose assessment of matrix effects and selectivity in two different biomarker method case studies: hepcidin and sclerostin (Scl). A logical approach to determine method sensitivity along with an orthogonal assessment to reduce matrix effects during method development is described. The antibodies generated against human hepcidin do not cross-react to rabbit hepcidin, so rabbit serum was used as the surrogate matrix. In the case of Scl, the Scl sequence is highly conserved across different species and antibodies raised against human Scl cross-react with multiple species. Therefore, a substituted matrix was the only feasible option for the Scl biomarker method. The matrix effect testing for each method during prestudy method validation was conducted 1082
Bioanalysis (2014) 6(8)
differently for each biomarker method based on data from orthogonal assessments of each biomarker method during method development. Reagents & methods Serum/reagents Normal and various diseased state individual sera and pooled human serum samples (healthy and different diseases states) and rabbit serum were obtained from Bioreclamation Inc., Hicksville, NY, USA, and were aliquoted and stored at -70 ± 10°C once they were received. Serum aliquots underwent no more than three freezing and thawing cycles. To identify the serum controls prior to pooling, at least 20 individual serum lots with volumes greater than 100 ml were measured for their endogenous hepcidin or Scl concentrations. The serum lots with biomarker protein concentrations (described below) of either Scl or hepcidin were pooled, aliquoted and stored at -70 ± 10°C. Additional reagents used are described in the applicable method subsections. future science group
Matrix effects, sensitivity & selectivity in case studies for hepcidin & sclerostin methods Serum
prescreening during method development Once the preliminary methods were developed, a minimum of 20 individual serum lots from healthy and diseased subjects/donors were analyzed to determine the serum hepcidin or Scl concentrations in normal and various applicable diseased populations to determine the required assay/method sensitivity. These would ultimately become LLOQ. Bioanalytical
method for measurement of hepcidin in human serum LC–MS/MS The sample separation and the method of measuring hepcidin in human serum using LC–MS/MS was previously described [8]. Briefly, the standards calibrator (STD) and QCs were made by spiking synthetic hepcidin peptide into 100% rabbit serum. As previously described, human hepcidin and the IS, human [13C9,15N1,Phe4]-hepcidin, were chemically synthesized using an ABI433 synthesizer (Applied Biosystems, Foster City, CA, USA) [8]. The calibration range of validated LC–MS/MS method was 2.5–500 ng/ml [8]. Bioanalytical
method for measurement of hepcidin in human serum using ELISA Standard streptavidin precoated 96-well Mesoscale Discovery (MSD®) plates were purchsed from MSD (Gaithersburg, MD, USA) and coated with biotin-conjugated human monoclonal antihuman hepcidin antibody (Clone 19D12, Amgen Inc., CA, USA), and were used to capture recombinant and endogenous hepcidin. The STD and QC were made by spiking recombinant human hepcidin, rhHep (Lot: 1913712#2, Amgen Inc.) into 100% rabbit serum (as surrogate matrix and various pooled serum lots were used). The low serum control (LSC) and high serum control (HSC) were prepared by pooling individual serum lots with endogenous hepcidin concentrations of either approximately 2 or 20 ng/ml, respectively. 10 µl each of blank serum, STD, QC, LSC and HSC were diluted 1:20 in assay buffer (1X phosphate buffer saline with 1 M NaCl, 0.5% Tween20 and 1% bovine serum albumin) and 50 µl were added to the well. Captured hepcidin (both endogenous and recombinant) was detected using a ruthenium-labeled human monoclonal antibody against hepcidin (Clone 15E1, Amgen Inc.). After a wash step, 4X a tripropylamine read buffer (MSD) was diluted 1:4 in water and added to the plate. The MSD plate was then future science group
| Research Article
read using the Sector® Imager 6000 Instrument (MSD) equipped with Discovery Workbench software (v3.0.18). The conversion of instrument responses for QC and study samples to concentration was achieved through a computer software mediated comparison to a standard curve assayed on the same plate, which was regressed according to five parameter logistic (autoestimate) regression model using the Watson (Thermo Scientific, MA, USA) data reduction package. The range of this analytical method was 0.20–100 ng/ml.
Key Terms
Bioanalytical
Prepared by spiking different concentrations of a reference standard to generate a concentration–response calibrator curve in an assay.
method for measurement of Scl in human serum using ELISA Standard 96-well MSD plates purchased from Mesoscale Discovery were coated with murine monoclonal antihuman Scl antibody (Clone A1, Amgen Inc.) and were used to capture the recombinant and endogenous Scl. The STD and QC were made by spiking recombinant Scl (various lots, Amgen Inc.), into protein buffer. The LSC was prepared by pooling the individual serum lots with the measureable Scl concentrations ranging from 150–300 pg/ml. The HSC was prepared by spiking 10 ng/ml of recombinant Scl into LSC. A volume of 20 µl each of blank serum, STD, QC, LSC and HSC were diluted 1:10 in I-Block buffer (Tropix®, Bedford, MA, USA) containing a rutheniumlabeled murine monoclonal antibody against Scl (Clone B1, Amgen Inc.) and mouse immunoglobulin. 50 µl of this mixture was added to the coated 96-well. Following a wash step, 4X a tripropylamine read buffer (MSD) was diluted 1:4 in water and added to the plate. The plate was then read using the Sector Imager 6000 Instrument (MSD) equipped with Discovery Workbench software (v3.0.18). The conversion of instrument responses for QC and study samples to concentration was achieved through a computer software mediated comparison to a standard curve assayed on the same plate, which was regressed according to five parameter logistic (auto-estimate) regression model using the Watson data reduction package (Thermo Scientific). The range of this analytical method was 0.05–50 ng/ml.
Universal references:
Reference standards that are representative of the desired form of the analyte in the unknown sample; are purified and well characterized.
Matrix effects: Direct or
indirect alteration/interference in response to the presence of unintended analytes or other interfering substances in the sample.
Standard calibrator:
Sensitivity: Ability of an
analytical method to differentiate and quantify the intended analyte in the presence of other potentially interfering components in the sample. As an industrial term, sensitivity is defined as the lowest concentration of the intended analyte that the method can reliably measure and is indicated by the LLOQ or the LOD for biomarker method.
Simulation
to determine relative inaccuracy An in silico simulation was performed to understand the relative inaccuracy in spiking low (400 pg/ml) or medium (40,000 pg/ml) levels of hepcidin into samples that already had various www.future-science.com
1083
R esearch A rticle |
Thway & Salimi-Moosavi
levels of the endogenous protein. Based on the assumption that the total error of the method was 10%, the relative inaccuracy in hypothetical endogenous concentrations of 200, 600, 1200, 3200, 8200, 20,200, 60,200 and 120,200 pg/ml was calculated. The compounding errors at each level were calculated using the following formula: e
6Spiked concentration # 0.1 @ + 6Hypothetical endogenous concentration # 0.1 @
Hypothetical endogenous concentration
o # 100
Equation 1
Results & discussion There are significant differences between the sequences of human and rat hepcidin (48% sequence homology based on BLAST analysis) and, therefore, the antibodies generated against human hepcidin do not cross-react to that of rats. For the human hepcidin ELISA method, the reagent pair (capture and detection antibodies) do not cross-react to rabbit serum, thus the standard calibrator for the assay was prepared in 100% rabbit serum (surrogate serum) due to availability of larger serum volume. In the case of Scl, the sequence homology of Scl is highly conserved among different species (>90% based on BLAST analysis). Thus, the reagent pair crossreacted with other species (such as rat or rabbit) and the protein buffer (substituted matrix) was used to make the standard calibrator (Figure 1). Setting
fit-for-purpose sensitivity requirements for biomarker methods were determined by situational needs Desired assay sensitivity for each biomarker method depends on a few critical factors: reactivity of reagent pair to both endogenous and recombinant biomarker proteins, and biological requirements to support drug development. Determining the sensitivity of a method required to support the study or program is a challenging task a bioanalytical scientist must address during
the early phase of method development. Often, the levels of the endogenous protein of interest are unknown, thus, the following practical approach is taken during the initial method development stage. For hepcidin and Scl methods, more than 10 individual lots of serum from healthy and/or diseased subjects were analyzed for their endogenous biomarker protein concentrations to determine the required assay sensitivity needed to support the drug development programs once the preliminary Scl and hepcidin ELISAs were developed. Table 1 shows the total number of serum lots that were analyzed, ranges of measured concentrations, mean serum concentrations and the percentage of samples expected to be less than the LLOQ based on mean concentrations versus the target LLOQ. For the total hepcidin levels in normal serum, if the target LLOQ is set near the mean concentrations, >40% of the samples will not have a measurable concentration. For Scl levels in normal and diseased sera, if the target LLOQ is set near the mean concentrations, >50% of the samples will not have a measurable concentration. Theoretically, it is desired to have a method that can quantify 100% of the samples, however, this may not be practical. In these cases, either the LOD or a new LLOQ can be targeted so that most samples have reportable values. Both methods must be able to measure low levels of their respective biomarker concentrations including most predose study samples, as well as postdose samples when the therapeutic has had an inhibitory effect on ligand levels. Thus, target LLOQ for each biomarker method was selected based on three major factors: the percentage of serum lots below mean concentration, whether the methods measure total or free concentration of intended biomarkers, and the inhibitory or stimulatory effect expected after the dosing of the therapeutic protein. Preferably, assay sensitivity in biomarker methods intended for primary endpoints could
Table 1. Summary information used in setting fit-for-purpose assay sensitivity of each ligand-binding assay method. Biomarker Serum type
Number Concentration of serum range/units lots
Hepcidin
49 88
0.107–63.6 ng/ml 14.5 ± 16.1 ng/ml 44 0.600–358 ng/ml 84.1 ± 73.4 ng/ml 14
200
4 0
10 12
17.3–363 pg/ml 33.2–322 pg/ml
50
10 25
Sclerostin
Normal CKD, AoC and sepsis Normal PMO
Mean concentration ± SD
140 ± 78 pg/ml 147 ± 99 pg/ml
Percentage of Intended Percentage sample below mean assay LLOQ below intended concentration (pg/ml) assay LLOQ
60 55
AoC: Anemia of cancer; CKD: Chronic kidney disease; PMO: Postmenopausal osteoporosis.
1084
Bioanalysis (2014) 6(8)
future science group
be targeted so that around 95% of the samples are measurable. However, assay sensitivity in exploratory biomarker methods could be more flexible. It is preferably that at least 80% of samples have measurable concentrations whether the methods measure total or free ligand concentrations. Based on these factors, the LLOQ of hepcidin and Scl methods was developed at 200 and 50 pg/ml, respectively, with intent to measure higher percentage of samples with measurable or reportable concentrations.
% of spiked pro-hepcidin
Matrix effects, sensitivity & selectivity in case studies for hepcidin & sclerostin methods
Orthogonal
B
future science group
A
% of spiked Hep 22
100 80 60 40 Without hHepC 25 With hHepC 25
20
0 0.125 0.250 0.500 1.000 2.000 4.000 8.000 Molar ratio of pro-hepcidin to HepC 25 40 30 20 Without hHepC 25 With hHepC 25
10
0 0.125 0.250 0.500 1.000 2.000 4.000 8.000 Molar ratio of HepC 22 to HepC 25
C 80 % of spiked Hep 20
assessment of ELISA to LC–MS/MS method during method development can provide insights on biotransformation of ligands that can affect the matrix effect Full length human hepcidin (hHepC 25) is a 25 amino acid (aa) peptide that is the central mediator of iron metabolism, and is found in the serum of both healthy and diseased individuals. Prohepcidin and other clipped variants, hHepC 22 (22 aa peptide) or hHepC 20 (20 aa peptide), have also been reported to be present in serum [5]. Initially, an LC–MS/MS method to measure hHepC 25 was developed and validated in the absence of reagent pair [8]. The LLOQ (2.5 ng/ml) of the hepcidin LC–MS/MS assay was considered to be insufficient for physiological concentrations, prompting the development of a sensitive ELISA method. Once the reagent pair was identified, the differential immunoreactivity of synthetic versus recombinant peptides and the ability of the method to measure and differentiate hHepC 25 in the absence and presence of other potentially interfering proteins, such as prohepcidin and/or clipped peptides (hHepC 20, and hHepC 22), were examined [9]. While the validated LC–MS/MS measures only hHepC 25 [5], the reagent pair used in the ELISA may have immunoreactivity to prohepcidin, hHepC 22 and hHepC 20 [9]. Therefore, the potential interfering effect of propeptide and clipped peptides in serum was examined. Spiked samples containing different molar ratios (0.25, 0.5, 1, 2 and 4) of these peptides to hHepC 25 (5.5 ng/ml) were prepared and tested in the hepcidin assay. In the absence of hHepC 25, the measured concentration of prohepcidin, hHepC 20 and hHepC 22 was assessed. Figure 2 shows the percentage recovery of prohepcidin, hHepC 22 and hHepC 20 in the ELISA method in the absence and presence of endogenous hHepC 25. Using the standard calibrator prepared with rHepC 25,
| Research Article
60 40 Without hHepC 25 With hHepC 25
20
0 0.125 0.250 0.500 1.000 2.000 4.000 8.000 Molar ratio of HepC 20 to HepC 25
Figure 2. Cross-reactivity and spiking recovery of recombinant propeptide or clipped peptides. (A) Prohepcidin; (B) hHepC 22; and (C) hHepC 20 in the absence or presence of approximately 5.5 ng/ml of endogenous hHepC 25. Spiked propeptide samples (at 0, 3, 6, 12, 24 and 47 ng/ml), spiked hHepC 22 or hHepC 20 samples (0, 1, 2, 3.5, 7 and 14.5 ng/ml each), which represent the different molar ratios (0, 0.25, 0.5, 1, 2 and 4) of these peptides to primary hHepC 25 (0 or 5.5 ng/ml) were prepared and tested in the hepcidin assay. The interfering effect of propeptide or clipped peptides in measuring hHepC 25 is tested in the ELISA method. The percentage spiked is calculated as follows: (concentration of measured/concentration of spiked peptides) for the samples without primary hHepC 25 and ([concentration of measured - 5.5]/ concentration of spiked peptides) for the samples with primary hHepC 25.
www.future-science.com
1085
R esearch A rticle |
B 500
300 250 200 150
y = 1.3× - 0.41 R2 = 0.91 n = 91
100 50 0
Hepcidin (ng/ml) – MS
Hepcidin (ng/ml) – MS
A 350
Thway & Salimi-Moosavi
400 300 200
y = 1.3× - 14 R2 = 0.95 n = 28
100 0
0
50 100 150 200 250 300 350 Hepcidin (ng/ml) – Sandwich ELISA
0
100 200 300 400 Hepcidin (ng/ml) – Sandwich ELISA
500
Figure 3. Comparison of serum hepcidin concentrations determined using the sandwich ELISA and LC–MS/MS. Samples were from patients with (A) chemotherapy-induced anemia and (B) anemia of cancer.
the ELISA method detected 69% of prohepcidin (Figure 2A), 28% for HepC 22 (Figure 2B) and 55% for HepC 20 (Figure 2C) in the absence of endogenous hHepC 25, indicating the levels of cross-reactivity of prohepcidin and clipped variants. On average, 74% of prohepcidin, 27% for HepC 22 and 41% for HepC 20 were recognized in the ELISA in the presence of endogenous hHepC 25 (Figure 2A–C). These results indicate that the ELISA reagent pair has approximately 70, 30 and 50% immunoreactivity to prohepcidin, hHepC 22 and hHepC 20, respectively, compared with hHepC 25. At lower molar ratios (less than 1), recovery of hHepC 22 and hHepC 20 was reduced, indicating the possibility of concentration-dependent
Relative inaccuracy (%)
10,000
Low (400 pg/ml) Medium (40 ng/ml)
1000
100
10
1
100
1000 10,000 100,000 Hypothetical endogenous concentration (pg/ml)
1,000,000
Figure 4. Hypothetical extrapolation of relative inaccuracy on a measurement of ligand concentration in the presence of various endogenous ligand concentrations using a method with 10% total error. The relative inaccuracy of spiking low and medium concentrations (400 pg/ml and 40 ng/ml) into various hypothetical endogenous concentrations is demonstrated.
1086
Bioanalysis (2014) 6(8)
interference from the clipped peptides. This suggests that the ELISA method will primarily measure hHepC 25 when it is the predominant species. Nonetheless, cross-reactivity of prohepcidin, hHepC 22 and hHepC 20 can lead to potential matrix effects. Since the approximately tenfold higher levels of endogenous hHepC 25 are in serum from diseased patients compared with serum of healthy individuals [9], cross-immunoreactivity of prohepcidin, hHepC 22 and hHepC 20 in this assay may likely contribute to the variable matrix effects during prestudy method validation. Therefore, it is important to predetermine the level of pro- and clipped peptides in healthy and diseased patient populations by an orthogonal method to evaluate whether there may be an impact on total hepcidin measurement. Based on preliminary data obtained by LC–MS/MS methods, there may be 20–50% of HepC 20 and HepC 22 in patient serum samples (data not shown), which could also be a source of inaccurate recovery. In addition, there could be significant levels of prohepcidin (~100 ng/ml) present in patient samples that may potentially interfere with recovery of hepcidin [10]. To understand the collective effect of prohepcidin and clipped peptides, the correlation between the ELISA and LC–MS/MS method specific for hHepC 25 was determined. The results depicted in Figure 3 show a linear relationship between the hepcidin levels measured by ELISA and LC–MS/MS methods in chemotherapy-induced anemia (Figure 3A) and anemia of cancer (Figure 3B) patient samples. Based on these results, it indicates that the majority of hepcidin species present in diseased sera is hHepC 25. The slope of the linear correlation curve indicates there was a 30% negative bias of ELISA as compared with the LC–MS/MS method. This could be caused by the presence of high levels of clipped forms in these samples. Since the hHepC 22 and hHepC 20 forms are 50% less immnuoreactive than hHepC 25, this would result in a negative bias by the ELISA method, while the LC–MS/MS method was not sensitive to levels of the prohepcidin and clipped forms in the samples. However, one could not decipher the contributions of prohepcidin and clipped forms in the inaccuracy of spike recovery without measuring their relative levels. Since the ELISA method is not specific to primary hepcidin species and cross-reacts with all forms of hepcidin, there is a need to use multiple methods, such as ELISA and LC–MS/MS, to perform orthogonal assessment. future science group
Matrix effects, sensitivity & selectivity in case studies for hepcidin & sclerostin methods Matrix
effect testing in the presence of high endogenous levels: case study 1 Once the accuracy and precision were conducted during the prestudy method validation (data not
| Research Article
shown), matrix effects and selectivity parameters were tested. Although the mean baseline serum levels of mature hepcidin in normal healthy subjects was
effects in
LBA s
For reprint orders, please contact [email protected]
Evaluating the impact of matrix effects on biomarker assay sensitivity Background: The accuracy of highly sensitive biomarker methods is often confounded by the presence of various circulating endogenous factors in samples causing matrix effects. Method: This article outlines two different biomarker methods: hepcidin enzyme-linked immunosorbent assay (ELISA) for which an orthogonal assessment of ELISA to liquid chromatography–tandem mass spectrometry was performed to examine the potential matrix effect, and sclerostin ELISA to evaluate the matrix effect. Results: Although the potential interfering effects of the endogenous hepcidin variants (prohepcidin and clipped) showed that these proteins had >30% immunoreactivity in ELISA, the hepcidin ELISA preferentially measures full-length hepcidin when the molar ratios of full-length to variants remain >1. The correlation of ELISA to liquid chromatography–tandem mass spectrometry results showed full-length hepcidin as the major form in diseased populations. Conclusion: A fit-for-for-purpose assessment of matrix effect/selectivity was also performed for each method. This article demonstrates the utility of a fit-for-purpose approach to assess the validity of biomarker methods in evaluating the interconnected parameters of matrix effects, sensitivity and selectivity.
Methods to measure soluble ligands as diseaseor drug-related biomarkers are commonly used in both predictive/diagnostic and drug development applications [1]. The use of relevant biomarkers could provide proof of mechanism for target effect, determine proof of concept showing downstream effect, and facilitate patient stratifications or dose escalation [2–5]. Development and validation of biomarker methods by using a ligand-binding assay platform impose distinct challenges relative to methods intended for PK assessment and a comparative analysis of which has been laid out thoroughly in the review articles by Lee [5,6]. Traditionally cited challenges in developing biomarker methods compared with that of PK are the general lack of gold standards or universal references making the selection of a most appropriate reference material an important process and the presence of matrix effects. One of the distinct challenges for the method development and validation for biomarkers is the presence of endogenous interferences. Finally, development of biomarker methods requires an in depth understanding of the biology, thus additional expertise and time are critical for success. Additional significant challenges in biomarker method development are posed by the presence of endogenous biomarkers circulating as propeptides or proteolytically clipped forms. The presence of these forms in disease states
that may or may not be active or biologically relevant yet these may cross-react with the assay reagents. Furthermore, propeptides and clipped forms may circulate at different levels in diseased states than in healthy population and may also have varying degrees of cross-reactivity with the assay reagents. Differential cross-reactivity of these variants may interfere with the accurate quantification of the intended analyte and result in matrix effects in the method. Biomarker methods used during drug development are commonly designed and validated with a specific fit-for-purpose approach [5,7]. To overcome the unwanted matrix effect caused by the presence of endogenous proteins, alternative strategies are typically employed (Figure 1). Since different levels of endogenous biomarkers may be present in the matrices, it is necessary to determine baseline concentrations in both healthy and diseased states, and to identify an alternative species matrix (a surrogate matrix) in which the endogenous proteins do not crossreact with the assay reagents to prepare the standard calibrator. If a particular biomarker protein has a high degree of homology across different species, the selection of a surrogate matrix may not be possible, and in this case, a protein buffer matrix (a substituted matrix) may be used to prepare the standard calibrator. However, this may result in a greater matrix effect in the sera of diseased subjects [5,7]. Therefore, it is
10.4155/BIO.14.55 © 2014 Future Science Ltd
Bioanalysis (2014) 6(8), 1081–1091
Theingi Thway* & Hossein Salimi-Moosavi Department of Pharmacokinetics & Drug Metabolism, Amgen Inc., One Amgen Center Drive, Thousand Oaks, CA 91320, USA *Author for correspondence: Tel.:+1 805 447 1000 Fax: +1 805 499 9027 [email protected]
ISSN 1757-6180
1081
R esearch A rticle |
Thway & Salimi-Moosavi
Is there a measurable endogenous ligand/target level in intended matrices of normal and disease population?
Yes Yes
Yes No
Intended matrices can be used to make standard calibrator
Is it feasible to use surrogate matrix?
Yes Yes
Yes No
Scenario 1: Surrogate matrix that do not cross react with intended species can be used to make standard calibrator while at least serum/plasma controls are made in intended matrix
Scenario 2: Substituted matrix or appropriate buffer can be used to make standard calibrator while at least controls in intended matrix are included to monitor the long-term assay performance
Figure 1. The rationale for the selection of appropriate matrix for standard calibrator in developing a biomarker method.
necessary to determine the accuracy or recoveries of spiked samples in the intended serum at the LLOQ and 3 × LLOQ to ensure relative accurate measurement of the biomarker protein while a substituted matrix is used to prepare the standard calibrator. This article illustrates fit-for-purpose assessment of matrix effects and selectivity in two different biomarker method case studies: hepcidin and sclerostin (Scl). A logical approach to determine method sensitivity along with an orthogonal assessment to reduce matrix effects during method development is described. The antibodies generated against human hepcidin do not cross-react to rabbit hepcidin, so rabbit serum was used as the surrogate matrix. In the case of Scl, the Scl sequence is highly conserved across different species and antibodies raised against human Scl cross-react with multiple species. Therefore, a substituted matrix was the only feasible option for the Scl biomarker method. The matrix effect testing for each method during prestudy method validation was conducted 1082
Bioanalysis (2014) 6(8)
differently for each biomarker method based on data from orthogonal assessments of each biomarker method during method development. Reagents & methods Serum/reagents Normal and various diseased state individual sera and pooled human serum samples (healthy and different diseases states) and rabbit serum were obtained from Bioreclamation Inc., Hicksville, NY, USA, and were aliquoted and stored at -70 ± 10°C once they were received. Serum aliquots underwent no more than three freezing and thawing cycles. To identify the serum controls prior to pooling, at least 20 individual serum lots with volumes greater than 100 ml were measured for their endogenous hepcidin or Scl concentrations. The serum lots with biomarker protein concentrations (described below) of either Scl or hepcidin were pooled, aliquoted and stored at -70 ± 10°C. Additional reagents used are described in the applicable method subsections. future science group
Matrix effects, sensitivity & selectivity in case studies for hepcidin & sclerostin methods Serum
prescreening during method development Once the preliminary methods were developed, a minimum of 20 individual serum lots from healthy and diseased subjects/donors were analyzed to determine the serum hepcidin or Scl concentrations in normal and various applicable diseased populations to determine the required assay/method sensitivity. These would ultimately become LLOQ. Bioanalytical
method for measurement of hepcidin in human serum LC–MS/MS The sample separation and the method of measuring hepcidin in human serum using LC–MS/MS was previously described [8]. Briefly, the standards calibrator (STD) and QCs were made by spiking synthetic hepcidin peptide into 100% rabbit serum. As previously described, human hepcidin and the IS, human [13C9,15N1,Phe4]-hepcidin, were chemically synthesized using an ABI433 synthesizer (Applied Biosystems, Foster City, CA, USA) [8]. The calibration range of validated LC–MS/MS method was 2.5–500 ng/ml [8]. Bioanalytical
method for measurement of hepcidin in human serum using ELISA Standard streptavidin precoated 96-well Mesoscale Discovery (MSD®) plates were purchsed from MSD (Gaithersburg, MD, USA) and coated with biotin-conjugated human monoclonal antihuman hepcidin antibody (Clone 19D12, Amgen Inc., CA, USA), and were used to capture recombinant and endogenous hepcidin. The STD and QC were made by spiking recombinant human hepcidin, rhHep (Lot: 1913712#2, Amgen Inc.) into 100% rabbit serum (as surrogate matrix and various pooled serum lots were used). The low serum control (LSC) and high serum control (HSC) were prepared by pooling individual serum lots with endogenous hepcidin concentrations of either approximately 2 or 20 ng/ml, respectively. 10 µl each of blank serum, STD, QC, LSC and HSC were diluted 1:20 in assay buffer (1X phosphate buffer saline with 1 M NaCl, 0.5% Tween20 and 1% bovine serum albumin) and 50 µl were added to the well. Captured hepcidin (both endogenous and recombinant) was detected using a ruthenium-labeled human monoclonal antibody against hepcidin (Clone 15E1, Amgen Inc.). After a wash step, 4X a tripropylamine read buffer (MSD) was diluted 1:4 in water and added to the plate. The MSD plate was then future science group
| Research Article
read using the Sector® Imager 6000 Instrument (MSD) equipped with Discovery Workbench software (v3.0.18). The conversion of instrument responses for QC and study samples to concentration was achieved through a computer software mediated comparison to a standard curve assayed on the same plate, which was regressed according to five parameter logistic (autoestimate) regression model using the Watson (Thermo Scientific, MA, USA) data reduction package. The range of this analytical method was 0.20–100 ng/ml.
Key Terms
Bioanalytical
Prepared by spiking different concentrations of a reference standard to generate a concentration–response calibrator curve in an assay.
method for measurement of Scl in human serum using ELISA Standard 96-well MSD plates purchased from Mesoscale Discovery were coated with murine monoclonal antihuman Scl antibody (Clone A1, Amgen Inc.) and were used to capture the recombinant and endogenous Scl. The STD and QC were made by spiking recombinant Scl (various lots, Amgen Inc.), into protein buffer. The LSC was prepared by pooling the individual serum lots with the measureable Scl concentrations ranging from 150–300 pg/ml. The HSC was prepared by spiking 10 ng/ml of recombinant Scl into LSC. A volume of 20 µl each of blank serum, STD, QC, LSC and HSC were diluted 1:10 in I-Block buffer (Tropix®, Bedford, MA, USA) containing a rutheniumlabeled murine monoclonal antibody against Scl (Clone B1, Amgen Inc.) and mouse immunoglobulin. 50 µl of this mixture was added to the coated 96-well. Following a wash step, 4X a tripropylamine read buffer (MSD) was diluted 1:4 in water and added to the plate. The plate was then read using the Sector Imager 6000 Instrument (MSD) equipped with Discovery Workbench software (v3.0.18). The conversion of instrument responses for QC and study samples to concentration was achieved through a computer software mediated comparison to a standard curve assayed on the same plate, which was regressed according to five parameter logistic (auto-estimate) regression model using the Watson data reduction package (Thermo Scientific). The range of this analytical method was 0.05–50 ng/ml.
Universal references:
Reference standards that are representative of the desired form of the analyte in the unknown sample; are purified and well characterized.
Matrix effects: Direct or
indirect alteration/interference in response to the presence of unintended analytes or other interfering substances in the sample.
Standard calibrator:
Sensitivity: Ability of an
analytical method to differentiate and quantify the intended analyte in the presence of other potentially interfering components in the sample. As an industrial term, sensitivity is defined as the lowest concentration of the intended analyte that the method can reliably measure and is indicated by the LLOQ or the LOD for biomarker method.
Simulation
to determine relative inaccuracy An in silico simulation was performed to understand the relative inaccuracy in spiking low (400 pg/ml) or medium (40,000 pg/ml) levels of hepcidin into samples that already had various www.future-science.com
1083
R esearch A rticle |
Thway & Salimi-Moosavi
levels of the endogenous protein. Based on the assumption that the total error of the method was 10%, the relative inaccuracy in hypothetical endogenous concentrations of 200, 600, 1200, 3200, 8200, 20,200, 60,200 and 120,200 pg/ml was calculated. The compounding errors at each level were calculated using the following formula: e
6Spiked concentration # 0.1 @ + 6Hypothetical endogenous concentration # 0.1 @
Hypothetical endogenous concentration
o # 100
Equation 1
Results & discussion There are significant differences between the sequences of human and rat hepcidin (48% sequence homology based on BLAST analysis) and, therefore, the antibodies generated against human hepcidin do not cross-react to that of rats. For the human hepcidin ELISA method, the reagent pair (capture and detection antibodies) do not cross-react to rabbit serum, thus the standard calibrator for the assay was prepared in 100% rabbit serum (surrogate serum) due to availability of larger serum volume. In the case of Scl, the sequence homology of Scl is highly conserved among different species (>90% based on BLAST analysis). Thus, the reagent pair crossreacted with other species (such as rat or rabbit) and the protein buffer (substituted matrix) was used to make the standard calibrator (Figure 1). Setting
fit-for-purpose sensitivity requirements for biomarker methods were determined by situational needs Desired assay sensitivity for each biomarker method depends on a few critical factors: reactivity of reagent pair to both endogenous and recombinant biomarker proteins, and biological requirements to support drug development. Determining the sensitivity of a method required to support the study or program is a challenging task a bioanalytical scientist must address during
the early phase of method development. Often, the levels of the endogenous protein of interest are unknown, thus, the following practical approach is taken during the initial method development stage. For hepcidin and Scl methods, more than 10 individual lots of serum from healthy and/or diseased subjects were analyzed for their endogenous biomarker protein concentrations to determine the required assay sensitivity needed to support the drug development programs once the preliminary Scl and hepcidin ELISAs were developed. Table 1 shows the total number of serum lots that were analyzed, ranges of measured concentrations, mean serum concentrations and the percentage of samples expected to be less than the LLOQ based on mean concentrations versus the target LLOQ. For the total hepcidin levels in normal serum, if the target LLOQ is set near the mean concentrations, >40% of the samples will not have a measurable concentration. For Scl levels in normal and diseased sera, if the target LLOQ is set near the mean concentrations, >50% of the samples will not have a measurable concentration. Theoretically, it is desired to have a method that can quantify 100% of the samples, however, this may not be practical. In these cases, either the LOD or a new LLOQ can be targeted so that most samples have reportable values. Both methods must be able to measure low levels of their respective biomarker concentrations including most predose study samples, as well as postdose samples when the therapeutic has had an inhibitory effect on ligand levels. Thus, target LLOQ for each biomarker method was selected based on three major factors: the percentage of serum lots below mean concentration, whether the methods measure total or free concentration of intended biomarkers, and the inhibitory or stimulatory effect expected after the dosing of the therapeutic protein. Preferably, assay sensitivity in biomarker methods intended for primary endpoints could
Table 1. Summary information used in setting fit-for-purpose assay sensitivity of each ligand-binding assay method. Biomarker Serum type
Number Concentration of serum range/units lots
Hepcidin
49 88
0.107–63.6 ng/ml 14.5 ± 16.1 ng/ml 44 0.600–358 ng/ml 84.1 ± 73.4 ng/ml 14
200
4 0
10 12
17.3–363 pg/ml 33.2–322 pg/ml
50
10 25
Sclerostin
Normal CKD, AoC and sepsis Normal PMO
Mean concentration ± SD
140 ± 78 pg/ml 147 ± 99 pg/ml
Percentage of Intended Percentage sample below mean assay LLOQ below intended concentration (pg/ml) assay LLOQ
60 55
AoC: Anemia of cancer; CKD: Chronic kidney disease; PMO: Postmenopausal osteoporosis.
1084
Bioanalysis (2014) 6(8)
future science group
be targeted so that around 95% of the samples are measurable. However, assay sensitivity in exploratory biomarker methods could be more flexible. It is preferably that at least 80% of samples have measurable concentrations whether the methods measure total or free ligand concentrations. Based on these factors, the LLOQ of hepcidin and Scl methods was developed at 200 and 50 pg/ml, respectively, with intent to measure higher percentage of samples with measurable or reportable concentrations.
% of spiked pro-hepcidin
Matrix effects, sensitivity & selectivity in case studies for hepcidin & sclerostin methods
Orthogonal
B
future science group
A
% of spiked Hep 22
100 80 60 40 Without hHepC 25 With hHepC 25
20
0 0.125 0.250 0.500 1.000 2.000 4.000 8.000 Molar ratio of pro-hepcidin to HepC 25 40 30 20 Without hHepC 25 With hHepC 25
10
0 0.125 0.250 0.500 1.000 2.000 4.000 8.000 Molar ratio of HepC 22 to HepC 25
C 80 % of spiked Hep 20
assessment of ELISA to LC–MS/MS method during method development can provide insights on biotransformation of ligands that can affect the matrix effect Full length human hepcidin (hHepC 25) is a 25 amino acid (aa) peptide that is the central mediator of iron metabolism, and is found in the serum of both healthy and diseased individuals. Prohepcidin and other clipped variants, hHepC 22 (22 aa peptide) or hHepC 20 (20 aa peptide), have also been reported to be present in serum [5]. Initially, an LC–MS/MS method to measure hHepC 25 was developed and validated in the absence of reagent pair [8]. The LLOQ (2.5 ng/ml) of the hepcidin LC–MS/MS assay was considered to be insufficient for physiological concentrations, prompting the development of a sensitive ELISA method. Once the reagent pair was identified, the differential immunoreactivity of synthetic versus recombinant peptides and the ability of the method to measure and differentiate hHepC 25 in the absence and presence of other potentially interfering proteins, such as prohepcidin and/or clipped peptides (hHepC 20, and hHepC 22), were examined [9]. While the validated LC–MS/MS measures only hHepC 25 [5], the reagent pair used in the ELISA may have immunoreactivity to prohepcidin, hHepC 22 and hHepC 20 [9]. Therefore, the potential interfering effect of propeptide and clipped peptides in serum was examined. Spiked samples containing different molar ratios (0.25, 0.5, 1, 2 and 4) of these peptides to hHepC 25 (5.5 ng/ml) were prepared and tested in the hepcidin assay. In the absence of hHepC 25, the measured concentration of prohepcidin, hHepC 20 and hHepC 22 was assessed. Figure 2 shows the percentage recovery of prohepcidin, hHepC 22 and hHepC 20 in the ELISA method in the absence and presence of endogenous hHepC 25. Using the standard calibrator prepared with rHepC 25,
| Research Article
60 40 Without hHepC 25 With hHepC 25
20
0 0.125 0.250 0.500 1.000 2.000 4.000 8.000 Molar ratio of HepC 20 to HepC 25
Figure 2. Cross-reactivity and spiking recovery of recombinant propeptide or clipped peptides. (A) Prohepcidin; (B) hHepC 22; and (C) hHepC 20 in the absence or presence of approximately 5.5 ng/ml of endogenous hHepC 25. Spiked propeptide samples (at 0, 3, 6, 12, 24 and 47 ng/ml), spiked hHepC 22 or hHepC 20 samples (0, 1, 2, 3.5, 7 and 14.5 ng/ml each), which represent the different molar ratios (0, 0.25, 0.5, 1, 2 and 4) of these peptides to primary hHepC 25 (0 or 5.5 ng/ml) were prepared and tested in the hepcidin assay. The interfering effect of propeptide or clipped peptides in measuring hHepC 25 is tested in the ELISA method. The percentage spiked is calculated as follows: (concentration of measured/concentration of spiked peptides) for the samples without primary hHepC 25 and ([concentration of measured - 5.5]/ concentration of spiked peptides) for the samples with primary hHepC 25.
www.future-science.com
1085
R esearch A rticle |
B 500
300 250 200 150
y = 1.3× - 0.41 R2 = 0.91 n = 91
100 50 0
Hepcidin (ng/ml) – MS
Hepcidin (ng/ml) – MS
A 350
Thway & Salimi-Moosavi
400 300 200
y = 1.3× - 14 R2 = 0.95 n = 28
100 0
0
50 100 150 200 250 300 350 Hepcidin (ng/ml) – Sandwich ELISA
0
100 200 300 400 Hepcidin (ng/ml) – Sandwich ELISA
500
Figure 3. Comparison of serum hepcidin concentrations determined using the sandwich ELISA and LC–MS/MS. Samples were from patients with (A) chemotherapy-induced anemia and (B) anemia of cancer.
the ELISA method detected 69% of prohepcidin (Figure 2A), 28% for HepC 22 (Figure 2B) and 55% for HepC 20 (Figure 2C) in the absence of endogenous hHepC 25, indicating the levels of cross-reactivity of prohepcidin and clipped variants. On average, 74% of prohepcidin, 27% for HepC 22 and 41% for HepC 20 were recognized in the ELISA in the presence of endogenous hHepC 25 (Figure 2A–C). These results indicate that the ELISA reagent pair has approximately 70, 30 and 50% immunoreactivity to prohepcidin, hHepC 22 and hHepC 20, respectively, compared with hHepC 25. At lower molar ratios (less than 1), recovery of hHepC 22 and hHepC 20 was reduced, indicating the possibility of concentration-dependent
Relative inaccuracy (%)
10,000
Low (400 pg/ml) Medium (40 ng/ml)
1000
100
10
1
100
1000 10,000 100,000 Hypothetical endogenous concentration (pg/ml)
1,000,000
Figure 4. Hypothetical extrapolation of relative inaccuracy on a measurement of ligand concentration in the presence of various endogenous ligand concentrations using a method with 10% total error. The relative inaccuracy of spiking low and medium concentrations (400 pg/ml and 40 ng/ml) into various hypothetical endogenous concentrations is demonstrated.
1086
Bioanalysis (2014) 6(8)
interference from the clipped peptides. This suggests that the ELISA method will primarily measure hHepC 25 when it is the predominant species. Nonetheless, cross-reactivity of prohepcidin, hHepC 22 and hHepC 20 can lead to potential matrix effects. Since the approximately tenfold higher levels of endogenous hHepC 25 are in serum from diseased patients compared with serum of healthy individuals [9], cross-immunoreactivity of prohepcidin, hHepC 22 and hHepC 20 in this assay may likely contribute to the variable matrix effects during prestudy method validation. Therefore, it is important to predetermine the level of pro- and clipped peptides in healthy and diseased patient populations by an orthogonal method to evaluate whether there may be an impact on total hepcidin measurement. Based on preliminary data obtained by LC–MS/MS methods, there may be 20–50% of HepC 20 and HepC 22 in patient serum samples (data not shown), which could also be a source of inaccurate recovery. In addition, there could be significant levels of prohepcidin (~100 ng/ml) present in patient samples that may potentially interfere with recovery of hepcidin [10]. To understand the collective effect of prohepcidin and clipped peptides, the correlation between the ELISA and LC–MS/MS method specific for hHepC 25 was determined. The results depicted in Figure 3 show a linear relationship between the hepcidin levels measured by ELISA and LC–MS/MS methods in chemotherapy-induced anemia (Figure 3A) and anemia of cancer (Figure 3B) patient samples. Based on these results, it indicates that the majority of hepcidin species present in diseased sera is hHepC 25. The slope of the linear correlation curve indicates there was a 30% negative bias of ELISA as compared with the LC–MS/MS method. This could be caused by the presence of high levels of clipped forms in these samples. Since the hHepC 22 and hHepC 20 forms are 50% less immnuoreactive than hHepC 25, this would result in a negative bias by the ELISA method, while the LC–MS/MS method was not sensitive to levels of the prohepcidin and clipped forms in the samples. However, one could not decipher the contributions of prohepcidin and clipped forms in the inaccuracy of spike recovery without measuring their relative levels. Since the ELISA method is not specific to primary hepcidin species and cross-reacts with all forms of hepcidin, there is a need to use multiple methods, such as ELISA and LC–MS/MS, to perform orthogonal assessment. future science group
Matrix effects, sensitivity & selectivity in case studies for hepcidin & sclerostin methods Matrix
effect testing in the presence of high endogenous levels: case study 1 Once the accuracy and precision were conducted during the prestudy method validation (data not
| Research Article
shown), matrix effects and selectivity parameters were tested. Although the mean baseline serum levels of mature hepcidin in normal healthy subjects was
