A 13-Steroid Serum Panel Based on LC-MS/MS: Use in Detection of Adrenocortical Carcinoma

BACKGROUND : Adrenocortical carcinoma (ACC) is a rare malignancy, with an annual incidence of 1 or 2 cases per million. Biochemical diagnosis is challenging because up to two-thirds of the carcinomas are biochemically silent, resulting from de facto enzyme deficiencies in steroid hormone biosynthesis. Urine steroid profiling by GC-MS is an effective diagnostic test for ACC because of its capac-ity to detect and quantify the increased metabolites of steroid pathway synthetic intermediates. Corresponding serum assays for most steroid pathway intermediates are usually unavailable because of low demand or lack of immunoassay specificity. Serum steroid analysis by LC-MS/MS is increasingly replacing immunoassay, in partic-ular for steroids most subject to cross-reaction. METHODS : We developed an LC-MS/MS method for the measurement of serum androstenedione, corticosterone, RESULTS between steroids were increased (median the non-ACC up

aldosterone (if hypokalemic or demonstrating arterial hypertension), 17-hydroxyprogesterone, dehydroepiandrosterone sulfate (DHEAS), androstenedione, testosterone, and 17␤-estradiol (men and postmenopausal women). An alternative approach is the measurement of steroid metabolites in urine by GC-MS (20 -22 ). Two studies report clinical sensitivities of 90% and 100% and clinical specificities of 90% and 99% for GC-MS results in diagnosing ACC, respectively (20,22 ), and that the 11deoxycortisol metabolite, tetrahydro-11-deoxycortisol, provides the greatest diagnostic yield. However, 11deoxycortisol is rarely measured in serum and assays for this, and other steroid synthetic pathway intermediates such as 17-hydroxypregnenolone and pregnenolone, metabolites of which provide useful markers of malignancy in urine (20 ), are not widely available.
Here we present an LC-MS/MS method for the paneling of 13 steroids in serum, which we assessed for its ability to differentiate samples from patients with ACC from those of other adrenal lesions in the setting of a tertiary referral center for adrenal pathology.

CLINICAL SAMPLES
Samples (minimum volume 2 mL into an EDTA tube and 3 mL into a serum separator tube plus either random or 24-h urine samples) were collected by endocrine nurses from patients attending scheduled appointments in the Programed Investigation Unit at King's College Hospital NHS Foundation Trust. At this appointment, informed consent was obtained from each patient to allow biochemical testing (including steroid measurements) for clinical evaluation of their adrenal lesions in accordance with a Trust standard operating procedure for suspected adrenal cancer. All samples were collected between 9:00 and 11:30 AM. Samples were subsequently taken immediately to the laboratory for processing. EDTA and serum separator tubes were centrifuged at 2163g for 10 min, and serum and plasma were aliquoted and either analyzed immediately or stored frozen at Ϫ20°C before analysis. Urine aliquots were acidified to a pH Ͻ2 for urine metanephrine analysis. Patients were categorized in the adrenal multidisciplinary meeting using combinations of biochemistry, radiology, and histology (if available). There were 10 ACC cases (all histologically proven) and 15 with phaeochromocytoma/ paraganglioma (PCC/PGL, all histologically proven); 7 had adenoma with glucocorticoid excess; and 16 adrenal lesions demonstrated no biochemical evidence of adrenal cortical or medullary excess (NFAA group). Surgical cases were defined according to standard pathological criteria (23)(24)(25)(26). In nonsurgical cases, conventional imaging criteria were applied for stratification of benign or malignant neoplasms (27 ).

SPECIMEN PROCESSING FOR LC/MS/MS
Portions of frozen calibrators, IQC, and unknown patient/EQA samples were thawed and mixed; then 250 L was transferred into a 1.5-mL polypropylene tube. Subsequently, 250 L of IS working solution and 500 L of ice-cold acetonitrile were added, and tubes were then vortex-mixed for 30 s. Precipitated protein was pelleted by centrifugation (12000g, 10 min), and the supernatant was transferred to a 10-mL glass tube containing 300 L of deionized water (dH 2 O). Ethyl acetate (1 mL) was added, and the tube was vortex-mixed for 5 min. Following centrifugation (161g, 1 min), the top organic layer was removed to a clean 75-ϫ 10-mm, 4.5-mL glass tube. Extracts were evaporated to dryness under nitrogen at 60°C and reconstituted in 200 L of a 65 ϩ 35 (v/v) mixture of dH 2 O/ methanol and transferred to an autosampler vial.

LC-MS/MS PROCEDURE
Extracts were injected (100 L) onto the LC column at a flow rate of 0.40 mL/min. Mobile phases were (A) dH 2 O and (B) methanol, each containing 0.1% (v/v) formic acid. The LC system was controlled using Aria MX (version 1.1, ThermoFisher). The gradient elution is summarized in Table 2 of the online Data Supplement. The total analysis time was 19.7 min, including column reequilibration. Eluent flow was diverted to waste for the first 3 min. MS/MS was carried out using Xcalibur (version 2.2, ThermoFisher) in the positive mode using atmospheric pressure chemical ionization. Data were collected in high resolution (0.40 m/z full width at half maximum) in the multiple reaction monitoring mode, with 2 m/z transitions per analyte and 1 m/z per IS (see Table 3 in the online Data Supplement). Postanalysis processing used LC Quan TM (version 2.6, ThermoFisher). For assay calibration, peak area ratios (analyte quantifier to IS) were used to construct calibration graphs, with lines fitted by linear regression. The intercepts were not forced through zero, and line weighting was applied (1/concentration).

METHOD VALIDATION
To validate the developed liquid-liquid extraction (LLE) LC-MS/MS assay, the recovery, linearity, and lower and upper limits of quantification (LLoQ and ULoQ) were determined in accordance with US Food and Drug Administration Center for Drug Evaluation and Research guidance for bioanalytical method validation.
To assess LLE recovery using ethyl acetate, 2 experiments were performed using IQC material. First, absolute recovery was evaluated by directly comparing analyte peak area from protein-precipitated samples that had undergone LLE with samples undergoing protein precipitation only. Second, relative recovery was assessed using IS-corrected peak area ratios in the same samples. To test linearity, charcoal-stripped serum was spiked with steroids at concentrations covering physiological and pathological ranges and tested in triplicate. LLoQ and ULoQ were defined for each analyte as the lowest concentration at which the imprecision (%CV) was Ͻ20% (LLoQ) or Ͻ15% (ULoQ), with the measured concentration within Ϯ20% of the nominal value. Method precision was assessed using IQC material at 3 target values, either analyzed 6 times in 1 batch (intraassay precision) or in singlicate within 6 batches on different days (interassay precision). Matrix effects were assessed by the postcolumn infusion method (30 ) using an IS working solution infusion via a tee-piece during the analysis of extracted patient samples (n ϭ 5), as well as monitoring IS intraassay precision during sample analysis. Two steroid stability experiments were performed using IQC material. First, freeze-thaw stability was assessed on samples undergoing freeze-thaw cycles on 3 consecutive days. Second, postextraction stability was evaluated in samples left either refrigerated (4°C) or at room temperature for 7 days before analysis. In each case, analysis was performed against fresh calibrators.
Method comparison with cortisol, testosterone, progesterone, DHEAS, androstenedione, and 17hydroxyprogesterone UKNEQAS samples was performed (n ϭ 30 for each). Results were compared with the LC-MS/MS users group mean value for all steroids except progesterone. No UKNEQAS-registered labora-tories perform progesterone analysis by LC-MS/MS, so results were compared with the all-laboratory immunoassay mean. Anonymized excess serum samples obtained in primary care were used to determine steroid reference ranges (n ϭ 200).

STATISTICAL ANALYSIS
Statistical analysis was performed using Analyze-It ® (version 4.65.3). Good method agreement was defined by (a) Deming regression analysis demonstrating a slope of approximately 1 with 95% CIs bracketing 1 with an intercept of approximately 0 (95% CI bracketing 0) and (b) the Altman-Bland plot giving a bias with a confidence limit spanning 0. Clinical data were found not to be normally distributed using the Shapiro-Wilk test. Pairwise comparisons were performed using Mann-Whitney U-tests, with post hoc Bonferroni correction. Data are reported as median and interquartile ranges (IQR). Values of P Ͻ 0.05 were defined as statistically significant.

Results
Chromatographic resolution of 13 steroids was achieved within 14.5 min (Fig. 1). This extended time was necessary to achieve baseline separation of the targeted isobaric steroids 21-deoxycortisol, corticosterone and 11-deoxycortisol, and 11-deoxycorticosterone and 17-hydroxyprogesterone.
Recovery of steroids after protein precipitation and subsequent LLE was assessed in both absolute and relative terms. For all steroids, absolute extraction recovery was Ͼ50%, with relative extraction recoveries, evaluated after IS correction, between 90% and 110% (see Table 4 in the online Data Supplement).
The developed method was linear over several orders of magnitude (r Ն 0.99) for all steroids (Table 1). We established ULoQs for each steroid that permitted measurement at the high pathological concentrations expected in ACC (Table 1), whereas the LLoQs were sufficiently low to allow quantification of most steroids in healthy individuals. Intraassay and interassay precisions were Յ10% for all steroids (see Table 5 in the online Data Supplement). No analytically significant ion suppression/enhancement was observed, as evidenced by infusion studies (suppression Ͻ15% for all steroids studied) and IS peak area precision Ͻ20% during analysis of extracted patient samples (Table 1). All steroids were stable through 3 freeze-thaw cycles, whereas extracted samples were stable at room temperature and at 4°C for 7 days, with concentrations for all steroids within Ϯ10% of the original value measured in each stability experiment. There was a good agreement between the developed method and EQA consensus values (see Fig. 1 in the online Data Supplement). Demographic and clinical characteristics of the ACC, cortisol-producing adenoma, PPC/PGL, and NFAA groups are summarized in Table 2. Groups were well matched for age and sex, although the PCC/PGL group tended to be younger (Table 2). ACC cases presented with larger tumors. All patients with ACC and PPC/PGL underwent surgery, whereas only 57% of those with cortisol-producing adenoma and 12.5% of the NFAA group underwent more surgery.
For the ACC cases, 5 of 10 presented with clinical features of steroid hormone excess. Three females had signs of Cushing's syndrome, 1 female had androgen excess, and 1 male presented with uncontrolled hypertension. Of the remaining cases, 2 females presented with abdominal symptoms, 1 male with weight loss, and 1 male with hematuria; the initial clinical presentation was not documented for 1 male. Diagnostic workup of the ACC group using existing routine biochemical methods was standard and in line with European Network for the Study of Adrenal Tumors guidance. Urine steroid profiling was performed in 9 of 10 cases, and all profiles were consistent with ACC. Random cortisol concentration was increased in 6 of 10 cases, whereas cortisol failed to suppress to Ͻ1.8 g/dL in 3 patients undergoing overnight dexamethasone suppression testing. Serum androgens were increased in 7 patients (increased testosterone in 2 females, androstenedione in 4 of 9 cases, and DHEAS in 5 of 9 cases). 17-Hydroxyprogesterone was increased in 6 of 9 cases tested. Progesterone was detected in 2 males (usually Ͻ1.6 ng/mL). The aldosterone/renin ratio was normal in all cases tested.
The non-ACC adrenal group was divided according to biochemical and radiological criteria: Overt biochemical glucocorticoid excess was defined by the failure of  cortisol to suppress to Ͻ5 g/dL in the overnight dexamethasone suppression testing or a urine free cortisol concentration above the reference range (Ͼ71 g/dL per 24 h). In these cases, radiology demonstrated lipid-rich pathology in 6 of 7 cases, with lipid-poor adenoma in the other case. In 15 patients with radiological features of PPC/PGL, catecholamine excess was confirmed by increased plasma and/or urine metanephrines. Patients were included in the NFAA group if they (a) were proven normal on histology or the mass was shown to be stable on imaging after Ͼ12 months follow-up and (b) had no clinical evidence of hormone excess, a normal aldosterone/renin ratio (or normal blood pressure), a normal overnight dexamethasone suppression test (cortisol Ͻ1.8 g/dL), and normal plasma metanephrine and normetanephrine. In this group, radiology fell into 2 categories: Hounsfield units Ͻ10 or defined by a radiologist as a lipid-poor adenoma with no features of malignancy.
Comparison of LC-MS/MS steroid data between the ACC and non-ACC groups revealed striking differences (Table 3). Across the non-ACC adrenal lesion groups, only up to 2 steroid concentrations were increased above the reference ranges given in Table 3 in individual cases, whereas in ACC between 4 and 7 steroids were increased (median ϭ 6 steroids). 11-Deoxycortisol was increased in all ACC cases (median, 6.2 ng/mL; IQR, 2.5-9.0; normal range, Ͻ0.9 ng/mL). Other steroids increased in ACC were androstenedione and DHEAS (6 cases), cortisol, pregnenolone, and 17hydroxypregnenolone (5 cases), corticosterone (4 cases), and 17-hydroxyprogesterone, 11-deoxycorticosterone, and cortisone (3 cases). Testosterone was increased in 2 females and progesterone was detectable in 3 males with ACC (normally Ͻ0.13 ng/mL by LC-MS/MS). In 1 instance, 17-hydroxypregnenolone could not be reliably quantified because of the presence of an interfering peak. Steroid heterogeneity in ACC was demonstrated when data for each steroid in each ACC was plotted as the multiple of the median value calculated from the non-ACC adrenal lesion group (see Fig. 2 in the online Data Supplement).
Discrimination of the non-ACC adrenal lesion and ACC groups was possible using several steroids (Table 3). Whereas pregnenolone and 21-deoxycortisol could not be detected in the non-ACC adrenal lesion group, pregnenolone was measurable in 5 ACC cases and 21deoxycortisol was detectable in 2 ACC cases. All other steroids except cortisone, corticosterone, and male testosterone showed significant increases in ACC when compared with non-ACC adrenal lesion groups. 11-Deoxycortisol and 17hydroxypregnenolone provided the best discrimination between ACC and the non-ACC adrenal lesions (Table 3, Fig.  2). 17-Hydroxyprogesterone and androstenedione were increased in the ACC group, but the IQRs overlapped the normal reference ranges of these steroids (Table 3, Fig. 2).
For DHEAS, 4 cases of ACC showed concentrations in the lower half of the reference range, whereas in the other cases its concentration was dramatically increased. The DHEAS concentration was lower in the cortisol-producing adenoma group than in the PPC/PGL and NFAA groups.
Because the values in ACC were so variable, the multivariate technique of principal component analysis was applied, both to the European Network for the Study of Adrenal Tumors-recommended sex-independent steroids (cortisol, androstenedione, DHEAS, and 17hydroxyprogesterone) and to all sex-independent serum steroid panel steroids (all minus testosterone and progesterone). Both positive and negative correlations among variables after principal component analysis were observed in biplot graphs (Fig. 3). Using the European Network for the Study of Adrenal Tumors-recommended steroid measurements, all but 2 of the ACC patients were separated from the other adrenal lesions (Fig. 3A). When full panel data were included, complete separation was achieved (Fig. 3B). However, the ACC cases did not cluster, reflecting the heterogeneity of tumor steroid production, which is the hallmark of this disorder (20 ).

Discussion
In this study we showed that serum steroid paneling by LC-MS/MS is a useful tool to discriminate ACC from other non-ACC adrenal tumor lesions. Previous practice for selection of biochemical investigations has been dictated by the clinical presentation, e.g., signs of cortisol or androgen excess. This only characterizes subpopulations of ACC; Ͻ50% of cases of ACC present with clinical symptoms of hormone excess (20 ). In contrast, serum steroid paneling allows the investigation of adrenal masses more comprehensively by offering measurement of all major steroid biosynthetic intermediates. It is both the number of steroids increased and the marked increases of several synthetic intermediates without biological activity that appear particularly useful in discriminating ACC from other adrenal lesions, validating the paneling approach to adrenal mass investigation. The DHEAS concentration was lower in cortisol-producing adenoma than in other adrenal lesions, in keeping with previous observations (13,31,32 ).
The cortisol precursor 11-deoxycortisol was most discriminating for differentiating ACC from non-ACC adrenal lesions, an observation consistent with previous studies demonstrating the usefulness of measuring its urinary metabolite tetrahydro-11-deoxycortisol by GC-MS (20,22 ). In blood, 11-deoxycortisol is known to be increased in benign and malignant adrenal tumors in children, although it was not reported whether 11-deoxycortisol discriminated benign from malignant disease (33 ). That 11-deoxycortisol is such a useful marker suggests a critical change in 11␤-hydroxylase activity in ACC. 11␤-  Hydroxylase catalyzes 11-deoxycortisol conversion to cortisol within the inner mitochondrial membrane, under the control of corticotropin. Most of the other steroid pathway enzymes are located in the smooth endoplasmic reticulum. Disruption of mitochondrial oxidative phos-phorylation is common in cancer, termed the "Warburg effect" (34 ), so 11␤-hydroxylase activity may be especially impaired in ACC. Alternatively, increased concentrations of steroid precursors could interfere with corticotropin release (22 ). CYP11B1 expression has been  shown to be downregulated in ACC, along with several other steroidogenic enzymes (35 ). It may be that the heterogeneity of steroidogenesis observed in the current study is a reflection of variable loss of steroid synthetic pathway enzyme expression in each tumor. Whether this heterogeneity predicts pathological features or disease prognosis warrants further investigation. Further studies are needed to evaluate the similarities and differences in qualitative and quantitative data produced by urine steroid profiling and serum steroid paneling. Quantification of serum pregnenolone and 17hydroxypregnenolone were useful in ACC in the current study; however, the relative concentrations did not reflect the large amounts of their metabolites pregnenediol and pregnenediol often seen in urine. This discrepancy may be because these 3␤-hydroxy-5-ene steroids are largely present in serum as sulfates, analogous to DHEAS. Other than DHEAS, the sulfated 3␤-hydroxy-5-ene steroids are not measured by the LC-MS/MS method, but their sulfated metabolites are measured by GC-MS, as free compounds after enzymatic hydrolysis (29 ). Nonetheless, the current study suggests that unconjugated pregnenolone and 17-hydroxypregnenolone are still useful ACC markers.
Urine steroid metabolite measurement may offer greater clinical sensitivity over single blood measurements because 24-h collections reflect steroid production throughout the day (20 ). Nonetheless, accurate 24-h collections are often not easily obtained and may be inconvenient to patients. Serum steroid paneling by LC-MS/MS offers a viable alternative and may also be more easily interpretable for clinicians because it targets the smaller number of major circulating steroids rather than the large number of urinary steroid metabolites. In many institutions, plasma metanephrine measurement is favored for PCC/PGL exclusion in patients with large adrenal masses in which ACC is in the differential diagnosis. Combined plasma metanephrine and serum steroid panel measurements may be sufficient for the biochemical exclusion of ACC or PCC/PGL. Further work is needed to clarify the effects of diurnal variation (36 ) and age and sex (37 ) on serum steroid paneling for ACC diagnosis. Our study used agematched adrenal tumor groups with all samples collected in the morning to minimize these effects. Nonetheless, in most cases, concentrations of the most useful ACC markers exceed variations attributable to age, gender, or time of day; such increases are only otherwise encountered in forms of congenital adrenal hyperplasia (9 ).
The inherent limitations of steroid immunoassays for adrenal tumor evaluation are demonstrated again in our study. Although progesterone was detected by LC-MS/MS in 2 patients who had tested positive by immunoassay, concentrations were much smaller. Pregnenolone and 17-hydroxypregenenolone sulfates are known 17hydroxyprogesterone immunoassay interferences (38 ) and are potential progesterone immunoassay interferents. There was also evidence for interference in the androstenedione and 17-hydroxyprogesterone immunoassay results performed in the ACC cohort. Prediction of potential crossreacting steroids is difficult because of steroid secretion heterogeneity in ACC.
In summary, LC-MS/MS serum steroid paneling offers a potentially important advancement in the clinical workup of patients with adrenal lesions by combining the measurement of both common and rarely measured steroids in a single analysis. It supports the published conclusions from urine steroid profiling that it is the increased concentrations of steroid synthetic pathway intermediates that best allow discrimination of ACC from non-ACC adrenal lesions.