Report from the HarmoSter study: inter-laboratory comparison of LC-MS/MS measurements of corticosterone, 11-deoxycortisol and cortisone

Introduction

The laboratory diagnosis of adrenal diseases is primarily based on the measurement of selective steroid hormones, such as cortisol, aldosterone and 17OH-progesterone. Increasing popularity of multi-analyte panels measured by liquid chromatography-tandem mass spectrometry (LC-MS/MS) allows for the investigation of steroids of the glucocorticoid pathway as potentially relevant players in the management of endocrine disorders [1], [2], [3]. Measurements of corticosterone, 11-deoxycortisol and cortisone in conjunction with steroids with established diagnostic relevance are in particular gaining relevance for identification and characterization of subclinical and overt hypercortisolism [4], [5], [6], [7], [8], hyperaldosteronism [9], [10], [11] and adrenal cortex tumors [12], [13], [14]. Overall, recent studies highlight the preeminent utility of 11-deoxycortisol as a biomarker for detecting hypercortisolism and adrenocortical carcinoma, and of corticosterone in predicting the cardiovascular risk in patients affected by mild autonomous cortisol secretion [2], [3], [4], [5], [6], [7], [8, 12], [13], [14]. Moreover, monitoring adrenal steroids in the setting of congenital deficits in steroidogenesis facilitates identification of specific enzymes that carry mutations [15], [16], [17], [18], [19], [20]. This thereby complements genetic investigations and assists the identification of non-classical or heterozygous forms of congenital adrenal hyperplasia [19, 20].

Despite experimental evidence on the importance of glucocorticoid pathway steroids as disease-specific biomarkers or as components of multi-steroidal predictive algorithms, corticosterone, 11-deoxycortisol and cortisone are rarely measured in clinical practice. There is also a lack of tools and procedures to assess the quality of LC-MS/MS determinations. In fact, currently, there are no reference measurement procedures, nor traceable matrix-based quality controls [21], and a few quality assessment (EQA) programs are only available for 11-deoxycortisol and corticosterone. As a consequence, there is a lack of information about the reproducibility and accuracy of LC-MS/MS methods for the aforementioned corticosteroids.

To address this shortcoming, the HarmoSter consortium recently launched a study on the harmonization of the LC-MS/MS measurement of 10 circulating steroids [22]. In our first report, we showed that LC-MS/MS methods characterized by heterogeneous pre-analytical and analytical methodologies can provide a satisfactory, albeit improvable, level of harmonization for measurements of cortisol, 17OH-progesterone and aldosterone. Moreover, we highlighted advantages for the adoption of a common calibration system [22]. This second report focuses on evaluation of intra- and inter-laboratory performance of eight LC-MS/MS methods for measurements of corticosterone, 11-deoxycortisol and cortisone, and whether adoption of common calibration materials might improve harmonization.

Materials and methods

Results

Corticosterone

Concentrations in plasma and serum samples ranged from 3.7 to 97.4 nmol/L (median of all laboratories by in house calibration). The intra-laboratory duplicate measurement CVs ranged from 3.3 to 7.7% (Table 2). The CV of individual duplicates listed in Supplementary Table 2 were mostly within MAI except for Laboratory L, which exceeded the threshold in 15.4% of samples. The external calibration range was within or slightly above the in house measurement range of all laboratories. As reported in Table 2 , the two calibration sets determined significantly different results within all laboratories (p<0.001), with results from external calibration that deviated from −11.5 to 11.6% compared to in house values. Passing-Bablok comparison of results obtained by the two calibration systems within laboratories that used Chromsystems calibrators for in house calibration detected substantial consistency in Laboratory E and modest deviations in Laboratory D and L (Supplementary Figure 1).

With results from in house calibration, the median inter-laboratory CV was 10.0%, though for 38.5% the inter-laboratory CV was above MAI. Use of external calibration significantly increased the median CV% to 12.6% (p>0.001) and numbers of results >MAI to 66.7% (Figure 1).

Figure 1:

Inter-laboratory coefficients of variation as function of the calibration system. Black dots: in house calibration; white dots: external calibration. Dashed lines: maximal allowable imprecision (10.7%). The inter-laboratory CV (min–max) was overall lower when using in house compared to external calibration for corticosterone (10.0% (3.4–29.6%) vs. 12.6% (6.7–31.3%), respectively; p<0.001) and cortisone (6.2% (3.5–13.5%) vs. 8.1% (3.3–16.6%), respectively; p<0.001). No significant differences were found for 11-deoxycortisol (10.7% (4.5–32.7%) vs. 11.0% (5.1–29.7%), respectively).

Passing-Bablok analysis of laboratory results vs. the median of all laboratories is shown in Figure 2. Slopes were not statistically different from 1 in four laboratories by in house and in one laboratory by external calibration; coefficients (95CI) ranged from 0.957 (0.944–0.969) to 1.153 (1.131–1.180) with the former, and from 0.867 (0.852–0.882) to 1.239 (1.203–1.262) with the latter (Supplementary Table 5).

Figure 2:

Passing-Bablok regression of laboratory steroid measures vs. median of all laboratories when using in house calibration. Slopes were not statistically different from 1 in four laboratories for corticosterone and cortisone and in two laboratories for 11-deoxycortisol.

Figure 3 shows the plot of laboratory bias vs. the median of all laboratories. In addition, Supplementary Figure 2 shows the Bland-Altman plots of bias from individual laboratories as functions of hormone concentration. The median laboratory bias ranged from −5.6 to 12.3% with in house, and from −10.1 to 19.5% with external calibration. Laboratory L exhibited the largest median bias, while Laboratory C exhibited the largest variance, with both calibrations. Cases above TAE were detected in Laboratory C and L with in house and in Laboratory C, H and L with external calibration (Figure 3). Compared to in house, the external calibration significantly increased the median bias in five laboratories, but reduced the variance in one laboratory (Supplementary Table 6).

Figure 3:

% Bias vs. median of all laboratories as function of the calibration system. % Bias = ((laboratory value – median of all laboratories)/median of all laboratories) × 100. Segments: mean; error bars: 95% agreement limits; dashed lines: zero ± total allowable error (27.1%). Median bias were all within allowable limits. However, bias variances were variable among laboratories, on occasion exceeding the allowable limits. Unifying the calibration system did not result in improving the inter-laboratory agreement.

Discussion

Our study evaluated the harmonization status for measurements of corticosterone, 11-deoxycortisol and cortisone according to different LC-MS/MS methods and use of plasma and serum specimens. MAI and TAE were used to interpret within- and between-method performance, with the restriction that these parameters were calculated from data about biological variability available only for 11-deoxycortisol, and not corticosterone and cortisone, and obtained from just two studies that used immunoassays [28]. Unfortunately, the availability of information on within- and between-subject variability for steroids generated by techniques with adequate specificity remains limited. Nevertheless, there are some studies from which between-subject variability can be derived [32–34]. Therefore, we recommend the provision of such information within the European Federation of Clinical Chemistry and Laboratory Medicine database for biological variation [30], so that appropriate performance limits can be established.

Methods investigated here are heterogeneous according to many procedural and technical aspects. Moreover, measurement ranges and analytical sensitivity differed among methods. The reasons for these differences may reflect several factors: 1. variable reporting purposes, such as a focus on the whole pathophysiological range or only high pathological levels; 2. the MS equipment available within each laboratory; and 3. the complexity of sample processing employed before MS analyses. For example, sensitivity for 11-deoxycortisol measurement is limited for Laboratory G, according to use of a simple protein precipitation and associated need for low sample volume.

Intra-assay imprecision was overall satisfactory for LDTs, even though suboptimal performance was displayed by some laboratories for 11-deoxycortisol and cortisone. As discussed in Supplementary Discussion, peak definition plays a role in intra-assay imprecision [35]. As previously recognized, the imprecision of Laboratory L, the only commercial kit user, was larger than expected due to initial inexperience, during the harmonization study, with maintenance of the MS ionization source [22]. Therefore, we recommend that laboratories that employ LC-MS/MS should have skilled staff and robust standard operating procedures for instrument maintenance and method monitoring.

Changing in house with external calibration determined a negligible deviation within most of the methods, with a few exceptions up to 15%. For laboratories that use Chromsystems’ as in house calibrators, this may reflect variability in handling of materials or lot-to-lot variability. At variance, for laboratories using gravimetrically determined standards or CRM diluted in surrogated matrices, deviations may depend on the accuracy of the dilution and on commutability of matrices. Due to the endogenous nature of steroids, none of the calibration matrices used in this study consisted of native serum or plasma. Even though participating laboratories and Chromsystems verified the adequacy of their calibrators by in house procedures, we cannot exclude that a certain degree of intra- and inter-laboratory variability is related to suboptimal commutability of surrogate matrices.

The overall inter-laboratory performance was mostly within acceptable specifications. However, imprecision above the estimated allowable limit was found in a consistent number of samples for corticosterone and 11-deoxycortisol. The bias vs. all laboratory median exceeded the estimated allowable limit only in a few cases. Nonetheless, large variance of bias was shown by some laboratories.

Unified calibration did not substantially improve the comparability of 11-deoxycortisol, and even worsened the consistency of corticosterone and cortisone measurements among laboratories. Such a finding indicates that, in this study, calibration apparently has a minor impact on overall inter-laboratory agreement compared to other factors, such as sample preparation, matrix interference, LC and IS, which can influence not only sample measurements but also the exactitude and commutability of calibrators. This contrasts with what we previously reported for cortisol, 17OH-progesterone and, to a minor extent, aldosterone [22], but is in line with the study by Owen et al. [36] focusing on testosterone.

The reasons for good consistency in some samples and suboptimal in others may reflect the features of individual methods. Here, baseline LC separation among the isobars corticosterone, 11-deoxycortisol and 21-deoxycortisol was ensured by all methods. Nonetheless, we could speculate that the diverse nature of LC methods employed among laboratories may confer susceptibility of some to other interfering compounds, which may explain the disagreement occasionally found among measurements [37].

Though speculative, a causative relationship may be suspected for 11-deoxycortisol since this analyte showed the poorest inter-laboratory performance and is also that for which four different isotopically-labeled ISs were used among the eight laboratories; this compares to two and three different IS respectively used for corticosterone and cortisone. As covered in Supplementary Discussion, IS labeling can influence analyte quantitation by multiple mechanisms [38]. With a focus on testosterone, Owen et al. [39] described how use of D2-, D5- or ¹³C3-isotiopically-labeled IS could influence results up to 15%. However, Loh et al. [40] found that changing D2- with ¹³C3-isotope did not influence measurements of 17OH-progesterone. Such contrasting findings reinforce the concept that the ability of the IS to correct for variation in analyte detection largely depend on the complex interaction among multiple method factors.

As for corticosterone, Laboratory C displayed the largest variance of bias. Intriguingly, this laboratory used a water loss (−18 amu) Q1/Q3 transition for the MRM detection of d8-corticosterone, which may be prone to specificity issues. The same transition was used by Laboratory D, which, conversely, did not show the same problem. All laboratories used the same Q1/Q3 transition for cortisone quantitation except Laboratory H, which exhibited a large bias.

Finally, another possible source of variability is the vacuum tube used to collect blood. Indeed, gel containing tubes have been previously shown to affect steroid concentrations [41]. A strength of our project is that we obtained paired blood samples collected with three different tubes, whose impact on steroid measurement will be investigated in a dedicated study.

All the above-mentioned potential reasons for differences in measurements among laboratories are speculative. Confirmation can only be attained by studies purposely designed to test the influence of various ISs and of Q1/Q3 transitions on analyte quantitation, as well as the efficacy of novel technologies empowering LC resolution in the separation of interfering species. Such studies could help to identify and correct causes of inaccuracy and would result in improved measurement harmonization. Ideally, technical issues should be resolved before entry of assays into the clinical routine. Until analyte- or panel-specific guidance is provided to optimize LC-MS/MS performance, recommendations can only be directed toward careful method validation and EQA scheme participation. Over the time period of this study, only two laboratories participated in EQA programs for serum corticosterone and 11-deoxycortisol. To date, only a few EQA schemes are available for these two analytes, but not for serum cortisone, and materials distributed are not traceable. EQA schemes based on traceable matrix-based quality controls should be implemented for these and other steroids that enter the clinical arena [21, 42, 43]. In such a frame, our study represented aunique opportunity for laboratories to verify their performance.

All centers are ISO accredited, however, only three laboratories have the LC-MS/MS measurement of serum corticosterone, 11-deoxycortisol and cortisone specifically included in the accreditation list. Of note, subdividing the consortium according to the clinical or research function did not help elucidating reasons for the observed inter-laboratory agreement (data not shown).

As previously recognized, a limitation of our study is that no reference method was involved. Therefore, we cannot establish the exact accuracy of individual methods. Furthermore, the sample size was not entirely compliant with guidelines [42]. In addition, our study focused on basal blood collections. Future studies could be enlarged to dynamic tests to yield low and high levels of adrenal steroids.

In conclusion, our investigations of interlaboratory comparability of LC-MS/MS measurements of corticosterone, 11-deoxycortisol and cortisone indicated reasonable agreement among methods. Failure to improve harmonization with a common calibration system indicates that identification of other sources of disagreement, such as method imprecision, accuracy and specificity, may provide solutions to improve inter-laboratory performance. Our results suggest that harmonization of steroid measurements by LC-MS/MS is feasible. Such harmonization may facilitate correction of current shortcomings in assay performance that hamper the reproducibility of scientific findings, confound diagnoses or therapeutic monitoring, and prevent the establishment of consensus reference intervals.