**10. LOD and LOQ**

Assessment of limit of detection/limit of quantitation (LOD/LOQ) is required for most analytical methods developed to monitor product quality attributes. In the latest version of ICH Q2R1, the terms Detection Limit (LD) and Quantitation Limit ( QL) are used instead of LOD and LOQ, respectively. ICH defines LOD (DL) as the minimum level of analyte which can be readily detected, while LOQ (QL) has been defined as the minimum level of analyte which can be quantified with acceptable accuracy and precision. Practical application of LOD is related to the decision about integration of chromatograms, electropherograms or spectra, while LOQ is related to the decision on whether to report the results of tests on official documents, such as the Certificate of Analysis (CoA) for the lot.

Analytical Method Validation for Biopharmaceuticals 127

Considering LOD from the perspective of the decision to include or disregard a peak for integration purposes, stresses the importance of signal-to-noise ratio as a key parameter governing peak detection. Defining LOD as 3.3 x noise creates a detection limit, which can serve as a universal property of methods applicable to all analytes and different instruments (because sensitivity factor has been disregarded in this form of the equation). LOD expressed in this format is a dynamic property due to the dependency on the type of instrument, status of the instrument, and quality of the consumables. LOD determined this

The decision about reporting a specific analyte on the CoA is typically linked to specifications. After the decision about integration has been made for all analytes resolved (defined) by the method, the results are recorded in the database (e.g. LIMS). When all analytical tests are completed, the manufacturer creates the CoA by extracting the relevant information from the database. Only a subset of the results, which are defined by specifications, will be listed on the CoA. The specifications will depend on the extent of peak characterization and the clinical significance of the various peaks (Apostol, Schofield et al. 2008). Therefore, the list will change (evolve) with the stage of drug development. In such a context, LOQ should be considered as the analyte specific value expressed in units of protein concentration, a calculation for which instrument sensitivity cannot be disregarded (in contrast to LOD estimation). This indicates that a potential exists for diverse approaches

Application of LOD/LOQ to purity methods presents specific challenges that deserve additional consideration. The reporting unit for purity methods is percent (%) purity, a unit that is not compatible with the unit in which LOD or LOQ are typically expressed (units of concentration or amount). The signal created by the analyte may vary with the load, while the relative percentage of the analyte does not change. This creates a situation where the analyte of interest can be hidden within the noise or, alternatively, can be significantly above the noise for the same sample analyzed at two different load levels within the range allowed by the method. This has been addressed by the concept of "dynamic LOQ" by combining statistical and S/N approaches (Apostol, Miller et al.


The above equation expresses LOQ as a function of signal-to-noise ratio and the observed purity of the analyte. Both parameters can change from test-to-test, due to equipment variability and sample purity variability. Therefore this equation should be viewed as the

way will be expressed in units of peak height, e.g. mV or mAU.

to the practice of determining the LOD and LOQ.

2009).

N = level of peak-to-peak noise

dynamic (live) assessment of LOQ.

S = peak height for the analyte of interest P = purity level for the analyte of interest

ICH Q2R1 suggests three different approaches: visual inspection, signal-to-noise-ratio, or variability of the slope of the calibration curve (statistical approach). Vial and Jardy, and Apostol et al., evaluated different approaches for determining LOD/LOQ and concluded that they generate similar results (Vial and Jardy 1999; Apostol, Miller et al. 2009). It is prudent to verify LOD/LOQ values obtained by different calculations. If those values are not within the same order of magnitude, then the integrity of the source data should be investigated.

The statistical approach is most commonly practiced, and is associated with the use of well known equations:

$$\text{LOD} = 3.3 \times \text{SD/S}$$

$$\text{LOD} = 10 \times \text{SD/S}$$

SD = standard deviation of response

S = slope of calibration curve (sensitivity)

The SD can be easily obtained from linear regression of the data used to create the calibration curves. The most common way to present calibration data for the purpose of linear regression is to graph the expected analyte concentration (spiked, blended) vs. the recorded response (UV, Fl, OD etc). This type of plot is characteristic of analytical methods for which the response is a linear function of the concentration (e.g. UV detection that follows the Beer-Lambert law). In cases where the measured response does not follow a linear dependency with respect to concentration (e.g., multi-parameter fit response of immunoassays), the response should be transformed to a linear format, such as semilogarithmic plots, so that the equations above can be utilized.

The slope used in these equations is equivalent to instrument sensitivity for the specific analyte, reinforcing the fact that LOD/LOQ are expressed in units of analyte concentration (e.g. mg/ml) or amount (e.g., mg). Since the LOD and LOQ are functions of instrument sensitivity, these values, when defined this way, are not universal properties of the method transferable from instrument to instrument, or from analyte to analyte.

Considering LOD from the perspective of the decision to include or disregard a peak for integration purposes, stresses the importance of signal-to-noise ratio as a key parameter governing peak detection. Defining LOD as 3.3 x noise creates a detection limit, which can serve as a universal property of methods applicable to all analytes and different instruments (because sensitivity factor has been disregarded in this form of the equation). LOD expressed in this format is a dynamic property due to the dependency on the type of instrument, status of the instrument, and quality of the consumables. LOD determined this way will be expressed in units of peak height, e.g. mV or mAU.

The decision about reporting a specific analyte on the CoA is typically linked to specifications. After the decision about integration has been made for all analytes resolved (defined) by the method, the results are recorded in the database (e.g. LIMS). When all analytical tests are completed, the manufacturer creates the CoA by extracting the relevant information from the database. Only a subset of the results, which are defined by specifications, will be listed on the CoA. The specifications will depend on the extent of peak characterization and the clinical significance of the various peaks (Apostol, Schofield et al. 2008). Therefore, the list will change (evolve) with the stage of drug development. In such a context, LOQ should be considered as the analyte specific value expressed in units of protein concentration, a calculation for which instrument sensitivity cannot be disregarded (in contrast to LOD estimation). This indicates that a potential exists for diverse approaches to the practice of determining the LOD and LOQ.

Application of LOD/LOQ to purity methods presents specific challenges that deserve additional consideration. The reporting unit for purity methods is percent (%) purity, a unit that is not compatible with the unit in which LOD or LOQ are typically expressed (units of concentration or amount). The signal created by the analyte may vary with the load, while the relative percentage of the analyte does not change. This creates a situation where the analyte of interest can be hidden within the noise or, alternatively, can be significantly above the noise for the same sample analyzed at two different load levels within the range allowed by the method. This has been addressed by the concept of "dynamic LOQ" by combining statistical and S/N approaches (Apostol, Miller et al. 2009).

$$\text{LOQ} = 10 \times \left(\text{S/N}\right)^{\text{-1}} \times \text{P}$$

N = level of peak-to-peak noise

126 Analytical Chemistry

investigated.

known equations:

SD = standard deviation of response

S = slope of calibration curve (sensitivity)

logarithmic plots, so that the equations above can be utilized.

transferable from instrument to instrument, or from analyte to analyte.

**10. LOD and LOQ** 

Assessment of limit of detection/limit of quantitation (LOD/LOQ) is required for most analytical methods developed to monitor product quality attributes. In the latest version of ICH Q2R1, the terms Detection Limit (LD) and Quantitation Limit ( QL) are used instead of LOD and LOQ, respectively. ICH defines LOD (DL) as the minimum level of analyte which can be readily detected, while LOQ (QL) has been defined as the minimum level of analyte which can be quantified with acceptable accuracy and precision. Practical application of LOD is related to the decision about integration of chromatograms, electropherograms or spectra, while LOQ is related to the decision on whether to report the results of tests on

ICH Q2R1 suggests three different approaches: visual inspection, signal-to-noise-ratio, or variability of the slope of the calibration curve (statistical approach). Vial and Jardy, and Apostol et al., evaluated different approaches for determining LOD/LOQ and concluded that they generate similar results (Vial and Jardy 1999; Apostol, Miller et al. 2009). It is prudent to verify LOD/LOQ values obtained by different calculations. If those values are not within the same order of magnitude, then the integrity of the source data should be

The statistical approach is most commonly practiced, and is associated with the use of well

LOD = 3.3 x SD/S LOQ = 10 x SD/S

The SD can be easily obtained from linear regression of the data used to create the calibration curves. The most common way to present calibration data for the purpose of linear regression is to graph the expected analyte concentration (spiked, blended) vs. the recorded response (UV, Fl, OD etc). This type of plot is characteristic of analytical methods for which the response is a linear function of the concentration (e.g. UV detection that follows the Beer-Lambert law). In cases where the measured response does not follow a linear dependency with respect to concentration (e.g., multi-parameter fit response of immunoassays), the response should be transformed to a linear format, such as semi-

The slope used in these equations is equivalent to instrument sensitivity for the specific analyte, reinforcing the fact that LOD/LOQ are expressed in units of analyte concentration (e.g. mg/ml) or amount (e.g., mg). Since the LOD and LOQ are functions of instrument sensitivity, these values, when defined this way, are not universal properties of the method

official documents, such as the Certificate of Analysis (CoA) for the lot.


The above equation expresses LOQ as a function of signal-to-noise ratio and the observed purity of the analyte. Both parameters can change from test-to-test, due to equipment variability and sample purity variability. Therefore this equation should be viewed as the dynamic (live) assessment of LOQ.

### **11. System suitability**

System suitability is intended to demonstrate that all constituents of the analytical system, including hardware, software, consumables, controls, and samples, are functioning as required to assure the integrity of the test results. System suitability testing is an integral part of any analytical method, as specified by ICH Q2R1. However, guidance is vague and reference is often made to Pharmacopeias for additional information. The USP, EP and JP contain guidance for a broad scope of HPLC assays, including assays of the active substance or related substances assays, assays quantified by standards (external or internal) or by normalization procedures, and quantitative or limit tests. While each type of assay is described in the compendia, the specific system suitability parameters to be applied for each type of assay, is not included with the description. Thus, some interpretation is required. The interpretation of how to best meet the requirements of the various compendia while still maintaining operational efficiency is a significant challenge for industry.

Analytical Method Validation for Biopharmaceuticals 129

purpose of system suitability is to ensure appropriate system performance (including standards and controls), not to try to differentiate individual sample results from historical trends (e.g., determining equivalence of results from run-to-run). In practice, setting system suitability parameters that are inappropriately stringent can result in the rejection of assay results with acceptable precision and accuracy. It is highly advisable to ensure the participation of Quality Engineers and/or other staff members with appropriate statistical

ICH Q2R1 prescribes that the evaluation of robustness should be considered during the development phase. The robustness studies should demonstrate that the output of an analytical procedure is unaffected by small but deliberate variations in method parameters. Robustness studies are key elements of the analytical method progression and are connected

Method robustness experiments cannot start before the final conditions of the method are established. It is a good practice to identify operational parameters for the method and to divide them in the order of importance into subcategories according to their relative

1. "Essential" category: includes method parameters that are critical to the method output

2. "Less important" category: includes method parameters that are not as critical as those in the "Essential" category, but may still affect the method output. These parameters

3. "Depends on Method" category: includes parameters that may affect the method output differently for different methods, such that these parameters should be treated

4. "Not useful" category: includes method parameters that are known to have no impact on the method output, such that these parameters need not be evaluated for

It is highly impractical to evaluate the impact of all possible parameters on the output of the method. Therefore, robustness studies could be limited to the demonstration that the reported assay values are not affected by small variations of "essential" operational parameters. It is a good practice to prospectively establish a general design (outline) for such studies. Typically, in these types of studies a reference standard and/or other appropriate samples are analyzed at the nominal load. The studies may be carried out using the onefactor-at-a-time approach or a Design of Experiment (DOE) approach. The selection of assay parameters can vary according to the method type and capabilities of the factorial design, if applicable. It is essential to study the impact of all essential factors, and it is important to establish prospectively "target expectations" for acceptable changes in the output, to ensure that these robustness studies do not repeat the development work. The maximum allowable

expertise when setting system suitability parameters.

**12. Method robustness** 

to the corresponding qualification studies.

importance, which are exemplified below:

and therefore require evaluation;

differently for each method;

robustness.

should be evaluated at the scientist's discretion;

Existing guidance for system suitability was developed for pharmaceutical compounds and may not be directly applicable for proteins which, due to their structural complexity and inherent heterogeneity, require additional considerations beyond those typically required for small molecules. For example, appraisal of resolution by measuring the number of theoretical plates (commonly done for small molecules), may not be the best way to assess the system readiness to resolve charge isoforms of a protein on an ion exchange column. This may be due to the relatively poor resolution of protein peaks resulting from inherent product microheterogeneity, when compared to the resolution typically seen with small molecules. However, this methodology (the number of theoretical plates) may be a very good indicator to measure the system performance for size exclusion chromatography (SEC), which does not typically resolve product isoforms resulting from microheterogeneity.

To appropriately establish system suitability, we need to consider both the parameter that will be assessed and the numerical or logical value(s), generally articulated as acceptance criteria, associated with each parameter. System suitability parameters are the operating parameters that are the critical identifiers of an analytical method's performance. System suitability should be demonstrated throughout an assay by the analysis of appropriate controls at appropriate intervals. It is a good practice to establish the system suitability parameters during method development, and to demonstrate during qualification that these parameters adequately evaluate the operational readiness of the system with regard to such factors as resolution, reproducibility, calibration and overall assay performance. Prior to validation, the system suitability parameters and acceptance criteria should be reviewed in order to verify that the previously selected parameters are still meaningful, and to establish limits of those parameters, such that meaningful system suitability for validation is firmly established.

One important issue that merits consideration is that the setting of appropriate system suitability parameters is a major contribution to operational performance in a Quality environment, as measured by metrics such as invalid assay rates. A key concept is that the purpose of system suitability is to ensure appropriate system performance (including standards and controls), not to try to differentiate individual sample results from historical trends (e.g., determining equivalence of results from run-to-run). In practice, setting system suitability parameters that are inappropriately stringent can result in the rejection of assay results with acceptable precision and accuracy. It is highly advisable to ensure the participation of Quality Engineers and/or other staff members with appropriate statistical expertise when setting system suitability parameters.
