Abstract
Host cell protein (HCP) and residual Protein A are still measured mainly by enzyme-linked immunosorbent assay (ELISA). The choice is not made because ELISA is the most accurate method, but because it is the most sensitive, has the highest throughput and costs the least. Its result, however, is an immunological equivalent, not a mass concentration: the antibodies recognize only the proteins their epitope repertoire covers, and anything outside that repertoire is not counted. Understanding this is the starting point for every interpretive limit of these methods. The article follows four lines. First, coverage is the largest source of uncertainty: measured with two antibody sets whose mass-based coverage is 60% and 95%, the same sample can give results about 1.6-fold apart (95/60), even when both are run equally well. Second, bridging must use the same matrix, the same dilution and several levels: a bridge run only in buffer, or at a single concentration, cannot reveal differences in matrix effect or slope. Third, dilution linearity is the main in-use test for matrix effects: spike recovery is done once at validation, whereas dilution linearity can be checked on every sample. Fourth, the hook effect is the most dangerous failure in residual testing: under antigen excess the signal falls as concentration rises, so an out-of-limit sample reads as compliant without tripping any system-suitability limit. The article also covers the risk of extrapolating into the 4PL plateaus, the link between residual Protein A and chromatography process parameters, and interference from rheumatoid factor and anti-drug antibodies in clinical samples.
Keywords: enzyme-linked immunosorbent assay; host cell protein; antibody coverage; kit bridging; hook effect; dilution linearity; residual Protein A
1 Introduction
Process-related impurities introduced during biologics manufacture include host cell proteins, residual Protein A (ligand leached from affinity chromatography), host cell DNA, detergents and antifoams, process enzymes and inducers. The risk posed by HCP is not simply "toxicity": HCP can act like an adjuvant and affect immunogenicity, can carry protease activity that degrades the product, and can co-purify with the product and become enriched during purification. All three risk pathways call for quantitative control of HCP, and control starts with measuring it correctly.
Current practice has three weak points.
Treating immunological equivalents as absolute concentrations. Antibodies recognize only their epitope repertoire; components outside it are not counted, so results are systematically lower than the true total. The size of this bias depends on antibody coverage, not on how well the assay is run — no amount of good practice makes up for insufficient coverage.
Missing evidence of coverage. Without a coverage assessment, "total HCP 50 ng/mg" has no definite meaning: it may be 80% of the true total, or 40%. In method validation reports this item is often left blank.
Oversimplified bridging. Commercial kits are used early in development and platform kits after scale-up, and the two have to be bridged. Bridging is often reduced to "measure one sample once on each method; pass if the results are close". With no low, medium and high levels and no real matrix, such a bridge cannot detect differences between the two methods' matrix effects.
Taking the immunological equivalent as its thread, this article explains the roles of sandwich and competitive formats, the plateau limits of 4PL fitting, a three-tier evidence framework for coverage, and what bridging and dilution linearity actually achieve in method transfer.
2 Principles
2.1 Sandwich and competitive formats: the format decides what can be measured
In the sandwich format, a capture antibody coated on the solid phase binds the antigen in the sample, and an enzyme-labelled detection antibody recognizes a second epitope, forming an antibody–antigen–antibody sandwich. After unbound material is washed away, substrate is added; the signal increases with the amount of antigen.
This format has a strict prerequisite: the antigen must carry at least two accessible epitopes that do not overlap sterically. Short peptides, small molecules and partially denatured proteins whose epitopes are masked do not meet it. HCP is a mixture, a substantial part of which consists of small proteins or fragments, and the sandwich format detects these components less efficiently — the physical reason why coverage is such a prominent issue in HCP testing.
In the competitive format, antigen in the sample competes with a fixed amount of enzyme-labelled antigen for a limited amount of antibody; the signal decreases as antigen increases. Only one epitope is needed, so the format suits small molecules and short peptides, but its sensitivity and precision are usually lower than those of the sandwich format and it tolerates sample matrix less well.
The selection logic is therefore clear: large molecules, multiple epitopes and high sensitivity — sandwich; small molecules and a single epitope — competitive. HCP and Protein A both belong to the first group.
2.2 Signal amplification and curve fitting
With horseradish peroxidase (HRP), the amount of product formed per unit time is proportional to the amount of enzyme, so the signal is enzymatically amplified. This is why ELISA reaches the pg/mL range, and also why it is sensitive to incubation time, temperature and substrate lot.
Readout is either end-point or kinetic. End-point reading stops the reaction at a fixed time; it is simple but affected by errors in reaction timing. Kinetic reading measures the rate per unit time; it is more robust to timing errors but requires a reader capable of kinetic measurement. The choice should be made during method development and fixed in the SOP.
The standard curve is fitted with a four-parameter logistic (4PL) model:
y = (A − D) / [1 + (x / C)B] + D
Each parameter has a physical meaning: A is the response at zero concentration (lower plateau), D the response at saturating concentration (upper plateau), C the half-maximal effective concentration (EC50) and B the slope factor. How these four parameters behave underlies everything said in §4.1.
2.3 Immunological equivalents: the result is an epitope equivalent, not a mass
This is the single most important statement of the article: ELISA reports an immunological equivalent.
Let the total mass of HCP in a sample be M and the mass of components the antibodies recognize be Mcovered. The method's result corresponds to Mcovered, not M; the ratio of the two is the coverage. When coverage is <100%, results are systematically lower than the true total, and the bias has one direction only (always low) rather than scattering high and low at random.
2.4 The hook effect: antigen excess
Under antigen excess the sandwich format shows a signal that falls as concentration rises — the hook effect. The mechanism: at very high antigen concentration, capture and detection antibodies bind separate antigen molecules and cannot form sandwiches; monovalent binding increases at the expense of bivalent sandwiches, and the signal drops.
Scroll sideways to see the full figure →
In residual testing the hook effect errs in one direction only: it reads high concentrations as low. An HCP sample above the limit may fall in the hook region when measured undiluted and give a passing result, while system suitability, the standard curve and the controls all look normal — because every control lies within the standard-curve range and none is above the limit.
There is only one way to recognize it: dilute, retest and see whether the signal rises. A rising signal confirms a hook effect.
3 Methods
3.1 General procedure and sample handling
Coat (carbonate buffer, 4 °C overnight or 37 °C for 2 h) → block (BSA or casein) → add samples and incubate → wash → add detection antibody → wash → develop with substrate → stop → read.
Samples must be serially diluted in the method diluent into the standard-curve range, with dilution linearity verified. In HCP testing this step is not optional: the product itself is often around 10 mg/mL, whereas HCP limits are usually expressed in ng/mg (ppm, relative to product protein mass). The two quantities have different units and cannot simply be compared; convert first. At 10 ng/mg, a 10 mg/mL product contains about 100 ng/mL HCP, roughly five orders of magnitude below the product concentration — that is the real reason the sample cannot be measured undiluted.
Dilution brings two problems:
- Matrix dilution effects: dilution changes salt concentration, pH and surfactant concentration, which can change how efficiently antibodies bind antigen. The higher the dilution, the closer the matrix is to the diluent and the closer recovery usually comes to 100% — a trend that is itself diagnostic (see §4.2).
- Loss of low-abundance components: dilution pushes low-abundance HCP below the limit of quantitation, where it is counted as not detected rather than as a low value, so results at high dilution are biased low.
The dilution should therefore not be "the largest one that is safe" but "the smallest one that lands in the middle of the standard curve".
3.2 Kit suitability assessment and bridging
A kit suitability assessment should cover five items:
| Item | What to check |
|---|---|
| Expression system of the standard and antibodies | Same host as the product (CHO / HEK293 / E. coli)? Different hosts have different HCP populations |
| Evidence of coverage | Does the supplier provide coverage data? If not, assess it in-house |
| Spike recovery in the product matrix | Must use the product matrix; buffer is not an acceptable substitute |
| Dilution linearity | At least three dilutions; CV of back-calculated values within the limit set in the method |
| Bridging to the previous method | See below |
The keys to bridging are same matrix, same dilution, several levels:
- Same matrix: bridging samples must be real product matrix and, if one platform method serves several products, should cover different products.
- Same dilution: compare the two methods each at its own optimal dilution. Forcing a common dilution compares tolerance to dilution, not agreement between methods.
- Several levels: cover at least low, medium and high HCP levels (for example near the limit, 2× and 5× the limit). A single level cannot reveal a difference in slope between the methods.
Acceptance criteria for bridging should be set before the study; a relative difference within ±30% is a common order of magnitude, adjusted to the product and method capability. If the two methods differ in coverage, a systematic offset is expected; the question is then whether the offset is stable, not whether it is zero.
3.3 Antibody coverage studies
Coverage is the proportion of HCP species in the HCP population that the antibodies recognize. It belongs to method development, not to routine release testing.
Three approaches:
2D electrophoresis with immunoblot comparison. Host proteins are separated by 2D electrophoresis, transferred and probed with the antibodies under test; recognized spots are compared spot by spot with a total-protein stain of the same sample to estimate coverage. It is visual and shows which proteins are not recognized, but it is semi-quantitative, insensitive to low-abundance proteins and labour-intensive.
Cascade immunization. The HCP population is affinity-purified with the antibodies under test, and further animals are immunized to raise complementary antibodies, widening the recognition repertoire step by step. Used when coverage is clearly insufficient and needs strengthening.
Mass-spectrometry coverage. HCP species in the sample are identified by LC-MS/MS; the material immunocaptured by the antibodies is analysed the same way, and the two identifications are compared to calculate the proportion of species the antibodies can pull down. This is currently the accepted way to quantify coverage at species level, and HCP control strategies in licence applications rely on it more and more.
Cost and information content increase in that order. At least the first approach should be completed during method development; the third is recommended for key products.
3.4 Where method development and validation stand in compliance terms
Developing a residual-impurity method is not the same as routine release testing, nor does it replace validation. Development outputs may be used for process monitoring and release only after validation. Default acceptance values in guidelines (such as 80%–120% spike recovery) are references to be adjusted per product, because product concentration, matrix complexity and HCP limits vary widely and a single recovery window may not suit every product.
4 Data processing and curve fitting
Immunoassays have no "integration" in the chromatographic sense; their data processing centres on curve fitting, dilution linearity and back-calculation bias. This section covers those three.
4.1 The four 4PL parameters and extrapolation into the plateaus
Of the four 4PL parameters, A and D (the plateaus) are the least stable to fit. The reason is simple: standard-curve points cluster in the middle, where response changes quickly with concentration, while the plateaus have few points and little change in response. Estimates of A and D are therefore very sensitive to individual points — remove the highest standard and D can shift by 5%–10%.
This instability leads to two practical requirements:
No extrapolation into the plateaus. If a sample's response lies near A or D, the back-calculated concentration is extremely sensitive to small response errors, because there d(response)/d(concentration) ≈ 0 and so d(concentration)/d(response) → ∞. A 2% difference in response can mean more than 20% in concentration. The method should require sample responses to fall within EC20–EC80 of the standard curve, with redilution and retesting otherwise.
Curve parameters must be fixed in the report. Report the standard-curve parameters for each batch of samples (at least A, D, C, B and R²); otherwise the results cannot be reproduced, because fitting software differs in default weighting, starting values and outlier handling, and the same data can yield different curves.
4.2 Dilution linearity: the in-use test for matrix effects
To test dilution linearity, dilute a sample serially 1:2, 1:4, 1:8 and 1:16, measure each dilution, multiply by the dilution factor to back-calculate the original concentration, and calculate the CV or relative difference of the back-calculated values.
Scroll sideways to see the full figure →
Three typical results lead to three conclusions:
| Change of back-calculated value with dilution | Conclusion | Action |
|---|---|---|
| Consistent across dilutions (CV within limit) | No matrix interference; any point can be used | Report the dilution that falls in the middle of the curve |
| Rises monotonically with dilution, then plateaus | Matrix interference (suppression at low dilution) | Use a dilution on the plateau and fix that minimum dilution in the method |
| Falls monotonically with dilution | Low-abundance species lost at high dilution, or the diluent itself interferes | Use a low dilution and check the diluent blank |
The second pattern is the most common. At low dilution, salts, surfactants and the product itself suppress antibody binding through competitive adsorption or steric hindrance, and results are low; once the matrix has been diluted out, the suppression lifts and results rise towards the true value. The trend itself is evidence of a matrix effect, and more convincing than a one-off spike recovery.
Spike recovery at validation therefore shows only that recovery is acceptable at that dilution, not at every dilution. Dilution linearity is the test that can be applied to every sample.
4.3 Back-calculation bias and how results are reported
The back-calculation bias of each standard (the concentration obtained by reading its response back off the curve, relative to the nominal concentration) is a direct measure of curve quality. A common requirement is that each standard back-calculates within ±20% (±25% at the lower limit of quantitation), with no systematic pattern. If low standards are generally high and high standards generally low, the curve shape does not fit the data; consider another model or removal of outliers.
A report should make three things clear:
- Nature of the result: immunological equivalent or absolute amount, and whether coverage has been assessed.
- Dilution: the minimum dilution the result corresponds to, and whether dilution linearity holds at that dilution.
- Key curve parameters: at least EC50 and R², for review.
4.4 A three-tier evidence framework for coverage
Scroll sideways to see the full figure →
Coverage is not a single number but a combination of evidence. We recommend organizing it in three tiers — total, species, coverage:
- Total tier: ELISA gives the immunological equivalent and answers "how much".
- Species tier: LC-MS/MS gives the list of HCP species and their relative abundance and answers "which ones".
- Coverage tier: mass-spectrometry results for antibody-immunocaptured material, compared with the species tier, answer "which ones the antibodies pull down and which they do not".
Only the three tiers together support a complete conclusion. With the first tier alone, the conclusion is "50 ng/mg immunological equivalent, coverage unknown"; with all three, it becomes "total HCP about X ng/mg; the five most abundant proteins are all recognized by the antibodies, and unrecognized components together account for less than Y%".
5 Results and discussion
5.1 Recognizing and handling the hook effect
As described in §2.4, the danger of the hook effect lies in its direction: it reads high concentrations as low, and because every control lies within the standard-curve range, nothing raises an alarm.
Criteria and actions:
- Criterion: the signal rises on dilution. This is the only reliable in-use criterion.
- Action: retest at higher dilution until the signal lies in the middle of the standard curve and dilution linearity holds.
- Prevention: require in the method that samples tested for the first time are run at at least two dilutions, so that detecting a hook effect becomes part of routine work rather than an investigation after the fact.
5.2 Coverage and the upper limit of interpretation
How far an HCP value can be interpreted is limited by antibody coverage. If coverage has not been assessed, the report should state that the method's coverage of this product's HCP population has not been verified, and the value should be used only for comparisons between lots measured by the same method.
This rule is often weakened in practice, because "not verified" looks like self-criticism. From a regulatory point of view, however, a value whose limits are stated honestly is more useful than an absolute value whose limits are not: the first can be used correctly, the second will be used wrongly.
5.3 Orthogonal confirmation: the roles of ELISA and mass spectrometry
ELISA gives the total; LC-MS/MS identifies the species. The two complement each other and neither can replace the other: ELISA cannot say which proteins are present, and mass spectrometry cannot give a total immunological equivalent (quantitation by MS depends on each protein's ionization efficiency, so comparability across proteins is limited).
This is exactly why HCP control strategies in licence applications increasingly include mass-spectrometry data: regulators want to know how much of the HCP population your 50 ng/mg covers, not just that you measured 50. The pharmacopoeias have responded: in 2024 USP gave LC-MS measurement of residual HCP its own chapter, 〈1132.1〉, alongside the existing 〈1132〉 (mainly immunological methods such as ELISA), and the European Pharmacopoeia has chapter 2.6.34 on host-cell protein assays. Mass spectrometry has moved from research tool to method with a compendial chapter — the factual basis for the view taken in this section.
5.4 Interference in clinical samples: three sources, three mechanisms
One point is easily confused. In an HCP sandwich assay, both capture and detection antibodies are anti-HCP antibodies, whereas anti-drug antibodies (ADA) bind the product and cannot bind both reagent antibodies at once. The common claim that "ADA bridging causes false positives" therefore does not apply here. The real interferences come from three sources with different mechanisms:
First, rheumatoid factor (RF) and human anti-animal antibodies (HAAA). These target immunoglobulins themselves and can bind the Fc regions of both capture and detection antibodies, bridging them — this is the real source of bridging false positives, and it is not rare in clinical samples, especially from subjects with an autoimmune background.
Second, anti-drug antibodies (ADA). ADA bind the product and interfere indirectly: complexes with the product mask or compete for HCP binding epitopes, or change the accessibility of HCP in the matrix. The effect is mainly false negative.
Third, soluble target or excess product. These compete with HCP for the solid-phase capture antibody and also lower the result.
No single remedy covers all three. For RF/HAAA, add an excess of non-specific animal IgG to the diluent (or use Fab-fragment antibodies) to absorb them; for ADA, increase the dilution or use acid dissociation before measurement to break up complexes; for target competition, increase the dilution and correct with blank matrix. Method documents should state which interferences may occur for the product and how each is verified, rather than a general note to "watch for interference".
5.5 Residual Protein A: a release test and a process parameter
Residual Protein A levels are directly related to elution pH, cleaning conditions (NaOH concentration and contact time) and the number of cycles the column has been used.
This relationship has two uses:
- As a release test: it confirms the clearance capability of the affinity chromatography step.
- As reverse evidence on process parameters: Protein A leaching increases with column age, making it a direct indicator for column lifetime; leaching also increases when cleaning is insufficient, which supports optimization of the cleaning process.
Protein A data should therefore not be treated only as pass or fail but included in the annual product quality review as evidence linking process and quality. It is one of the few residual tests that can feed back into process optimization.
6 Pharmacopoeial comparison
| Item | Chinese Pharmacopoeia 2025 | USP | Ph. Eur. |
|---|---|---|---|
| General guidance on validation of analytical procedures | Volume IV, General Chapter 9101, Guidelines for Validation of Analytical Procedures for Drug Quality Standards | 〈1225〉 Validation of Compendial Procedures | Based mainly on ICH Q2(R2) |
| Residual HCP | Limits set in individual monographs; method development and validation follow General Chapter 9101; the assay may refer to General Chapter 3429 | 〈1132〉 Residual Host Cell Protein Measurement in Biopharmaceuticals (mainly ELISA); 〈1132.1〉 HCP measurement by LC-MS (added in 2024) | 2.6.34 Host-cell protein assays |
| Residual Protein A | Per individual monograph | No dedicated general chapter; validate the method in-house | Per individual monograph |
| General requirements for immunoassays | General Chapter 3429, enzyme-linked immunosorbent assay | 〈1103〉 Immunological Test Methods—ELISA | Per individual monograph |
Note: USP 〈1132.1〉 (LC-MS) was approved in November 2024 and published in USP–NF 2025 Issue 1; its official date has not yet been set. Ph. Eur. 2.6.34 is a relatively recent chapter dedicated to HCP. On the Ph. Eur. side, validation of analytical procedures relies mainly on ICH Q2(R2). Always cite the current official text.
7 Conclusions
- ELISA results are immunological equivalents, not mass concentrations. With insufficient coverage, results are systematically low, and the size of the bias is set by coverage rather than by how well the assay is run. Without a coverage assessment, results can only be compared between lots measured by the same method.
- Coverage is a method-development item, not a release test. At least a 2D electrophoresis–immunoblot comparison should be completed; for key products, MS-based coverage is recommended to give species-level quantitative evidence.
- Bridging must use the same matrix, the same dilution and several levels. A bridge run only in buffer, or at one concentration, cannot reveal differences in matrix effect or slope. The question is whether the offset is stable, not whether it is zero.
- Dilution linearity is the main in-use test for matrix effects. Back-calculated values that drift monotonically with dilution and then level off prove matrix interference; single-point results are then unusable, and the method must be fixed at the lowest dilution on the plateau.
- The hook effect errs dangerously and raises no alarm. The criterion is a signal that rises on dilution; the preventive measure is to make "at least two dilutions" routine.
- Do not extrapolate into the plateaus. Sample responses must fall within EC20–EC80, and curve parameters must be fixed in the report; otherwise results cannot be reproduced.
- Residual Protein A serves both release and process control and should be included in the evidence linking process and quality, not used only as a pass/fail decision.
References
Primary (pharmacopoeias, guidelines and standards, with links)
- Chinese Pharmacopoeia Commission. (2025). Pharmacopoeia of the People's Republic of China, 2025 ed., Vol. IV, General Chapter 9101: Guidelines for Validation of Analytical Procedures for Drug Quality Standards. https://2025.chp.org.cn/
- Chinese Pharmacopoeia Commission. (2025). Pharmacopoeia of the People's Republic of China, 2025 ed., Vol. IV, General Chapter 3429: Enzyme-linked immunosorbent assay. https://2025.chp.org.cn/
- ICH. (1999). Q6B: Specifications: Test Procedures and Acceptance Criteria for Biotechnological/Biological Products. Step 4. ich.org
- ICH. (2023). Q2(R2): Validation of Analytical Procedures. Step 4. ich.org
- USP. 〈1132〉 Residual Host Cell Protein Measurement in Biopharmaceuticals. USP-NF. usp.org
- USP. 〈1132.1〉 Residual Host Cell Protein Measurement in Biopharmaceuticals by Liquid Chromatography-Mass Spectrometry. USP–NF 2025, Issue 1 (approved 1 November 2024; official date to be determined). uspnf.com
- USP. 〈1103〉 Immunological Test Methods—Enzyme-Linked Immunosorbent Assay (ELISA). USP-NF. usp.org
- USP. 〈1225〉 Validation of Compendial Procedures. USP-NF. usp.org
- EDQM. 2.6.34 Host-Cell Protein Assays. European Pharmacopoeia, 11th ed. https://www.edqm.eu/
Secondary (selected published papers)
- Zhu-Shimoni, J., Yu, C., Nishihara, J., et al. (2014). Host cell protein testing by ELISAs and the use of orthogonal methods. Biotechnology and Bioengineering, 111(12), 2367–2379.
- Bracewell, D. G., Francis, R., & Smales, C. M. (2015). The future of host cell protein (HCP) identification during process development and manufacturing linked to a risk-based management for their control. Biotechnology and Bioengineering, 112(9), 1727–1737.
- Wang, X., Hunter, A. K., & Mozier, N. M. (2009). Host cell proteins in biologics development: identification, quantitation and risk assessment. Biotechnology and Bioengineering, 103(3), 446–458.
- Shukla, A. A., & Hinckley, P. (2008). Host cell protein clearance during protein A chromatography: development of an improved column wash step. Biotechnology Progress, 24(5), 1115–1121.
- Hogwood, C. E. M., Bracewell, D. G., & Smales, C. M. (2014). Measurement and control of host cell proteins (HCPs) in CHO cell bioprocesses. Current Opinion in Biotechnology, 30, 153–160.