Urine Steroid Metabolomics as a Biomarker Tool for Detecting Malignancy in Adrenal Tumors
Wiebke Arlt, Michael Biehl, Angela E. Taylor, Stefanie Hahner, Rossella Libé, Beverly A. Hughes, Petra Schneider, David J. Smith, Han Stiekema, Nils Krone, Emilio Porfiri, Giuseppe Opocher, Jerôme Bertherat, Franco Mantero, Bruno Allolio, Massimo Terzolo, Peter Nightingale, Cedric H. L. Shackleton, Xavier Bertagna, Martin Fassnacht, and Paul M. Stewart
Centre for Endocrinology, Diabetes, and Metabolism (W.A., A.E.T., B.A.H., P.S., N.K., C.H.L.S., P.M.S.), School of Clinical and Experimental Medicine, School of Mathematics (D.J.S.), and School of Cancer Sciences (E.P.), University of Birmingham, Birmingham B15 2TT, United Kingdom; Johann Bernoulli Institute for Mathematics and Computer Science (M.B., H.S.), University of Groningen, 9700 AK Groningen, The Netherlands; Endocrine and Diabetes Unit (S.H., B.A., M.F.), Department of Medicine I, University Hospital, University of Würzburg, 97080 Würzburg, Germany; Department of Endocrinology (R.L., J.B., X.B.), INCa- COMETE, Cochin Hospital, Institut Cochin, Institut National de la Santé et de la Recherche Médicale Unité 1016, René Descartes University, 75006 Paris, France; Familial Cancer Clinic (G.O.) and Division of Endocrinology (F.M.), Veneto Institute of Oncology Instituto di Ricovero e Cura a Carattere Scientifico and Department of Medical and Surgical Sciences, University of Padua, 35100 Padua, Italy; Department of Clinical and Biological Sciences (M.T.), Internal Medicine I, University of Turin, 10124 Turin, Italy; and Wellcome Trust Clinical Research Facility (P.N.), University Hospital Birmingham National Health Service Foundation Trust, Birmingham B15 2TH, United Kingdom
Context: Adrenal tumors have a prevalence of around 2% in the general population. Adrenocor- tical carcinoma (ACC) is rare but accounts for 2-11% of incidentally discovered adrenal masses. Differentiating ACC from adrenocortical adenoma (ACA) represents a diagnostic challenge in patients with adrenal incidentalomas, with tumor size, imaging, and even histology all providing unsatisfactory predictive values.
Objective: Here we developed a novel steroid metabolomic approach, mass spectrometry-based steroid profiling followed by machine learning analysis, and examined its diagnostic value for the detection of adrenal malignancy.
Design: Quantification of 32 distinct adrenal derived steroids was carried out by gas chromatog- raphy/mass spectrometry in 24-h urine samples from 102 ACA patients (age range 19-84 yr) and 45 ACC patients (20-80 yr). Underlying diagnosis was ascertained by histology and metastasis in ACC and by clinical follow-up [median duration 52 (range 26-201) months] without evidence of metastasis in ACA. Steroid excretion data were subjected to generalized matrix learning vector quantization (GMLVQ) to identify the most discriminative steroids.
Results: Steroid profiling revealed a pattern of predominantly immature, early-stage steroidogenesis in ACC. GMLVQ analysis identified a subset of nine steroids that performed best in differentiating ACA from ACC. Receiver-operating characteristics analysis of GMLVQ results demonstrated sensitivity = specificity = 90% (area under the curve = 0.97) employing all 32 steroids and sensitivity = specificity = 88% (area under the curve = 0.96) when using only the nine most differentiating markers.
Conclusions: Urine steroid metabolomics is a novel, highly sensitive, and specific biomarker tool for discriminating benign from malignant adrenal tumors, with obvious promise for the diagnostic work-up of patients with adrenal incidentalomas. (J Clin Endocrinol Metab 96: 3775-3784, 2011)
Abbreviations: ACA, Adrenocortical adenoma; ACC, adrenocortical carcinoma; CT, com- puted tomography; DHEA, dehydroepiandrosterone; DHEAS, DHEA sulfate; GC/MS, gas chromatography/mass spectrometry; GMLVQ, generalized matrix relevance LVQ; LDA, linear discriminant analysis; LVQ, learning vector quantization; 17-HP, 17-hydroxyproges- terone; 5-PD, pregnenediol; 5-PT, pregnenetriol; ROC, receiver-operating characteristics; THDOC, tetrahydro-11-deoxycorticosterone; THS, tetrahydro-11-deoxycortisol.
A drenal tumors are common, with a reported preva- lence of around 2% in the general population, based on computed tomography (CT) and autopsy series (1-3). Prevalence increases with age, with 1% of 40-yr-olds and 7% of 70-yr-olds harboring an adrenal tumor (1-3). In an aging society, coupled with the widespread use of abdom- inal imaging, healthcare professionals are confronted with an increasingly large number of patients with incidentally discovered adrenal masses, i.e. adrenal incidentalomas, that require diagnostic work-up but with an optimal ap- proach yet to be determined (1, 2, 4-7). Two key questions that need to be addressed in such a patient are, first, whether the tumor autonomously secretes hormones that could have a detrimental effect on health and, second, whether the adrenal mass is benign or malignant.
Results of previous studies indicate that up to 25% of adrenal nodules are hormonally active (1, 2, 8), which may require surgical or medical intervention. Exclusion of cat- echolamine excess caused by a pheochromocytoma aris- ing from the adrenal medulla is a mandatory part of the diagnostic work-up (1,2). Overt and autonomous cortisol and aldosterone secretion result in Cushing’s syndrome and primary hyperaldosteronism, respectively; both may lead to premature cardiovascular death if left untreated. Overproduction of adrenal androgen precursors, dehy- droepiandrosterone (DHEA) and androstenedione, is rarer but may pose diagnostic problems, in particular in women with a relatively small adrenal mass and a pheno- type resembling polycystic ovary syndrome. Additionally, adrenal tumors are frequently associated with lower levels of autonomous cortisol secretion in patients lacking overt clinical features of Cushing’s syndrome. There remains ongoing debate around the optimal treatment for these patients with mild or subclinical Cushing’s syndrome (9-11).
Adrenocortical carcinoma (ACC) is a rare tumor but accounts for 2-11% of incidentally discovered adrenal masses undergoing diagnostic work-up in specialist refer- ral centers (1,2, 8). ACC carries a poor prognosis, and cure can be achieved only by complete surgical removal with- out capsule violation (12, 13). Even when basing the his- topathological assessment on the entire tumor specimen, the differentiation between benign and malignant lesions represents a major diagnostic challenge (14, 15); molec- ular markers that may reliably indicate malignancy are being developed but require surgical tumor removal (16- 18). The risk of malignancy increases with the diameter of the adrenal mass (1, 2, 8). However, size alone is a poor predictor, with, for example, sensitivity and specificity rates of 81 and 61%, respectively, for a 4-cm cutoff (19). Imaging provides some guidance, in particular tumor den- sity assessed by CT and magnetic resonance imaging-
based chemical shift analysis (12, 19). However, inhomo- geneous tumors are difficult to assess, and although sensitivities of up to 100% have been reported in some imaging studies, specificity is generally poor (1, 2, 7, 19). [18F]Fluorodeoxyglucose positron emission tomography has been reported to be both highly sensitive (97%) and specific (91%) in differentiating benign from malignant adrenal tumors (20) but is expensive and not always avail- able, thus does not lend itself for routine screening.
Although 60-70% of ACC are biochemically found to overproduce hormones, this is not clinically apparent in many cases. This may be explained by relatively inefficient steroid production in ACC, manifesting with increased steroid precursors, due to a dedifferentiated and thus in- complete pattern of steroidogenic enzyme expression. This appears to be supported by some case reports and a small series analyzing steroid excretion in ACC (21-24).
Here we undertook steroid metabolite excretion anal- ysis by mass spectrometry followed by computational analysis and tested the performance of this novel bio- marker tool in detecting malignancy and hormone excess in a large cohort of patients with adrenal tumors.
Patients and Methods
Study population
The 24-h urine samples from adrenal tumor patients were collected between 2006 and 2009 in six clinical specialist referral centers participating in the European Network for the Study of Adrenal Tumors (ENS@T; www.ensat.org), with approval of local ethical review boards and informed consent. Only 24-h urines from patients with adrenal tumors identified as either benign [adrenocortical adenoma (ACA)] or malignant (ACC) as conclusively as possible were included in the analysis. Underlying diagnosis had been ascertained by histology and evidence of me- tastasis in ACC and by imaging, biochemical, and clinical fol- low-up showing no evidence of metastasis in ACA patients. Ex- clusion criteria were pregnancy and treatment with steroids or drugs known to induce expression and activity of hepatic cyto- chrome P450 enzymes (e.g. mitotane) or to alter steroid secretion in any other way (e.g. cytotoxic chemotherapy, ketoconazole, metyrapone, mifepristone, spironolactone, or eplerenone).
Biochemical analysis
Patients underwent a comprehensive adrenal tumor work-up comprising clinical assessment including standardized biochem- ical testing to detect glucocorticoid excess (24-h urinary free cortisol), mineralocorticoid excess (paired plasma renin and se- rum aldosterone measurements), and adrenal androgen excess [serum DHEA sulfate (DHEAS)] with the additional measure- ment of serum 17-hydroxyprogesterone (17-HP) and 17ß-estra- diol. Autonomous glucocorticoid excess was defined as in- creased urinary free cortisol excretion. Aldosterone excess was ascertained by a serum aldosterone/plasma renin activity ratio higher than 750 pmol/liter/ng/ml·h where the absolute serum aldosterone was higher than 450 pmol/liter. Adrenal androgen
excess was defined as serum DHEAS above the normal reference range. Increased serum 17ß-estradiol in men or postmenopausal women was accepted as evidence of tumor-related estrogen ex- cess. In all patients, pheochromocytoma had been excluded by analysis of 24-h urinary catecholamine or metanephrine excre- tion or measurement of plasma metanephrines.
Measurement of 24-h urinary steroid metabolite excretion in ACA and ACC patients and in a healthy control cohort (26 men, 62 women, age range 18-60 yr) was carried out by a long- established method using gas chromatography/mass spectrom- etry (GC/MS). Urine samples had been stored at -20 C before analysis, which was carried out within 2 yr of collection. A de- tailed description of this methodology has been published pre- viously (25, 26). In summary, free and conjugated steroids were extracted from 1 ml urine by solid-phase extraction. Steroid con- jugates were enzymatically hydrolyzed, reextracted, and chem- ically derivatized to form methyloxime trimethyl silyl ethers. GC/MS was carried out on an Agilent 5973 instrument operating in selected-ion-monitoring (SIM) mode to achieve sensitive and specific detection and quantification of 32 selected steroid metabolites. These were chosen to include important repre- sentatives of steroid groups such as androgen metabolites, glucocorticoid metabolites, mineralocorticoid metabolites, and 3ß-hydroxy-45 steroid precursors as shown in Fig. 1 (Sup- plemental Table 1, published on The Endocrine Society’s Jour- nals Online web site at http://jcem.endojournals.org, gives a de- tailed description of quantified steroids).
Computational analysis of steroid metabolite excretion
All numerical steroid excretion values were log-transformed and subsequently normalized by subtracting the respective mean values obtained in healthy controls with a similar age and sex distribution (n = 88) and dividing by the corresponding SD, which yielded 32 log-transformed steroid excretion values for each patient, expressed on a scale set by the control group.
Further data analysis was carried out by learning vector quan- tization (LVQ), which identifies typical representatives of the different classes from a set of example data (27). Matrix rele- vance LVQ incorporates the data-driven identification of a suit- able distance measure distinguishing the steroid excretion profile of ACA from ACC patients. The specific technique used was generalized matrix relevance LVQ (GMLVQ), a variant that op- timizes the discriminative power of the method (28).
Data classification by LVQ is based on prototypes, i.e. typical representatives of the classes, here ACA and ACC, obtained from labeled example data in a computerized training process. The training aims to efficiently discriminate the classes by assigning data points according to their distances from the prototypes. GMLVQ additionally incorporates the data-driven identifica- tion of a suitable distance measure in the learning process. Math- ematically, the data points and prototypes are vectors of the 32 steroid excretion values; the dissimilarity or distance measure is parameterized by a symmetric 32 × 32 matrix À = 2TQ, where Q is also 32 × 32. Diagonal values of the relevance matrix add up to 1 and correspond to the importance of individual steroids in separating the classes; off-diagonal values give the importance of pairs of steroids in the classification scheme. For additional details of the application of GMLVQ on the steroid metabolite data and preceding data handling and transformation, see Sup- plemental Methods.
For comparison with GMLVQ analysis, we also employed statistical modeling techniques, linear discriminant analysis (LDA) and a standard implementation of logistic regression (29) that was performed employing a standard algorithm (G02GB; Numerical Algorithm Group, Oxford, UK).
Statistical analysis
Steroid excretion data were represented as median and inter- quartile ranges, the latter being calculated using the Cleveland algorithm implemented in SigmaPlot (Systat Software Inc., Chi- cago, IL). Total excretion by individual steroid and steroid sub- group (androgens, mineralocorticoids, glucocorticoids, and re- spective precursor steroids) were compared between control, ACA, and ACC groups using the Kruskal-Wallis nonparametric test; for significance values of P < 0.05, pair-wise comparisons were carried out using Dunn’s post hoc test.
Results
Patient characteristics, clinical presentation, and routine biochemistry findings
Demographic and clinical characteristics of 102 pa- tients with ACA and 45 patients with ACC. are summa- rized in Table 1.
According to routine biochemistry results, 33 of the 102 ACA patients (32%) showed evidence of hormone excess. Of those, 14 patients had isolated glucocorticoid excess and 13 patients isolated mineralocorticoid excess. Four additional patients had combined gluco- and min- eralocorticoid excess. Two patients had mildly elevated DHEAS levels. Routine biochemistry revealed no evidence of hormone excess in the remaining 69 ACA patients (68%). Serum 17-HP was measured in 87 ACA patients and increased in five (6%), in two with glucocorticoid excess and in three as an isolated finding.
Among the 45 ACC patients, routine biochemistry showed evidence of hormone excess in 33 patients (73%). Isolated glucocorticoid and adrenal androgen excess was documented in 11 and seven patients, respectively. Twelve ACC patients had combined glucocorticoid and adrenal androgen excess, with additional aldosterone excess in two of them. Three additional patients had 170-estradiol excess, in combination with glucocorticoid and adrenal androgen excess in two and one, respectively. Twelve ACC patients (27%) showed no evidence of glucocorti- coid, mineralocorticoid, or adrenal androgen excess; how- ever, three of them had a raised serum 17-HP. Overall, serum 17-HP was increased above the reference range in 22 of 39 ACC patients (56%).
Urinary steroid metabolite profiling
Urinary steroid profiling revealed distinct differences between ACA and ACC patients (Fig. 2). ACC patients had a significantly higher excretion of androgen precursor
A
HJC
-
Adrenal Steroidogenesis
H
CH
CH3
P
HJC
H
H
HO
Cholesterol
Glucocorticoid precursors
Mineralocorticoid precursors
Mineralocorticoids
CYP11A1
Q
O
Q
Q
HO
Q
Q S
Q
H3
CH3
H
CH3
CH
13
OH
HO
CH
OH
HO
OH
HO
OH
CH
5
HSD3B2
H
-
CYP21A2
H
CYP11B2
CH
CYP11B2
CH
CYP11B2
CI
H
H
H
H
H
H
H
CYP11B1
M
H
H
$ H
H
¥ H
HO
Pregnenolone
o
Progesterone
o
Deoxycorticosterone
o
Corticosterone
18OH-Corticosterone
o
o
Aldosterone
CYP17A1
CYP17A1
Q
Q
Q
Q
H3
CH3
Q
H
OH
I
CH3
CH3
CH
“H”OH
OH
HO
CH3
“OH
OH
O
CH
HOH
OH
CH
9
HSD3B2
CH
CYP21A2
CH
9
CYP11B1
CH
H
HSD11B2
CH
H
H
H
H
H
2
H
H
M
HSD11B1
H
H
HO
17-Pregnenolone
o
17-OHP
0
11-Deoxycortisol
0
Cortisol
O
Cortisone
CYP17A1
CYP17A1 I
Glucocorticoids
H3
9
CH
9
CH
OH
7
H3
OH
CH
HSD3B2
CH
4
HSD17B3
CH
SRDA2
CH
-
4 H
H
H
2.
H
H
H
H
H
HO
o
o
o
DHEA
Androstene dione
Testosterone
DHT
Androgen precursors
Androgens
B
Healthy Controls (n=88)
4500
4000-
Androgens and
3500-
Urinary Steroid Excretion (ug/24hr)
androgen precursors
Gluco- corticoids
3000-
2500-
2000-
1500-
Mineralo- corticoids and MC precursors
Glucocorticoid precursors
1000-
500-
0
E
An 1
Etio 2
DHEA 3
16xDHEA 4
5-PT 5
5-PD 6
THA 7
5a-THA 8
THB 9
5a-THB 10
5x5ß-THAIdo 11
THDOC 12
5a-THDOC 13
PD 14
3x5a-17HP 15
17-HP 16
PT 17
P’TONE 18
THS 19
F 20
6ß-OH-cortisol 21
THF 22
5a-THF 23
a-cortol 24
B-cortol 25
11ß-OH-Andro 26
11ß-OH-Etio 27
E 28
THE 29
a-cortolone 30
ß-cortolone 31
11-Oxo-Etio 32
metabolites [pregnenediol (5-PD), pregnenetriol (5-PT), DHEA, 16a-OH-DHEA; see Supplemental Table 1] and also of metabolites of active androgens (androsterone and etiocholanolone; Fig. 2A and Supplemental Table 2). Al-
though the deoxycorticosterone metabolites tetrahydro- 11-deoxycorticosterone (THDOC) and 5a-THDOC were significantly increased in ACC (Fig. 2B and Supplemental Table 2), other mineralocorticoid precursor metabolites,
| ACA group (n = 102) | ACC group (n = 45) | |
|---|---|---|
| Median age (range) at time of urine collection (yr) | 60 (19-84) | 55 (20-80) |
| Sex (male, female) | 39, 63 | 24, 21 |
| Tumor load at time of urine collection | Adrenal tumor (n = 102) | Adrenal tumor, no metastasis (n = 9) Adrenal tumor plus metastasis (n = 26) ACC metastasis after removal of primary tumor (n = 10) |
| Maximum diameter of adrenal tumor at time of urine collection (median and range) | 26 (9-78) mm | 90 (14-230) mm |
| Surgical removal of adrenal tumor | 24/102 (24%) | 30/45 (67%) |
| Median Weiss scoreª | 1 (0-2) (n = 15) | 6 (3-9) (n = 21) |
| Duration of follow-up (median and range) | 52 (26-201) months since diagnosis (all patients alive) (n = 102) | 14 (1-187) months from diagnosis to death due to metastatic ACC in deceased patients (n = 35) 45 (30-100) months since diagnosis in alive patients (n = 10)b |
a The Weiss system scores the presence or absence of nine histopathological features (Weiss score range 0-9); scores under 3 are indicative of a benign adrenal tumor, scores of 3 are borderline, and scores of 4 and above are indicative of malignancy (14).
b Seven of the 10 surviving patients suffer from metastatic disease. The three remaining patients have not shown evidence of recurrence yet (all three initially presented with early-stage disease [ENS@T II (13); primary tumor diameters 80, 89, and 160 mm; histology indicative of ACC with Weiss scores of 5, 7, and 4, respectively; current survival times 45, 51, and 42 months, respectively].
such as those of corticosterone, did not differ between ACA and ACC. Glucocorticoid precursor metabolites PD, PT, 17-HP, and tetrahydro-11-deoxycortisol (THS) were significantly higher in ACC, as were free cortisol, 6BOH- cortisol, tetrahydrocortisol, and a-cortol (Fig. 2, C and D, and Supplemental Table 2). Representation of all steroid excretion data in the form of a heat map suggested in-
creased excretion of THS (steroid 19) in many ACA and ACC patients and high excretion of the androgen precur- sor metabolites 5-PD and 5-PT (steroids 5 and 6) in most of the ACC patients (Supplemental Fig. 1).
Statistical comparison of total excretion per steroid subclass revealed that ACC patients exhibited signifi- cantly increased excretion of metabolites derived from the
A
Androgen and androgen precursor metabolites
C
Glucocorticoid precursor metabolites
Urinary Steroid Excretion (µg/24hr)
Urinary Steroid Excretion (ug/24hr)
Urinary steroid excretion (µg/24hr)
Urinary steroid excretion (ug/24hr)
80000
75000
70000
65000
60000
5500
5000
4500
15000
10000
.5000
5
10000
15000
20000
25000
30000
35000
40000
45000
$0000
75000
80000
22000
-20000
18000
16000
- 14000
12000
- 10000
8000
6000
4000
-2000
2000-
4000
1000
12 2000-
14000
18000
20000
22000-
9
e
6000-
8000
An
PD
E
Etio
H
3x5x-17HP
**
DHEA
17-HP
16aDHEA
PT
5-PT
P’TONE
5-PD
THS
ACA (n=102)
ACC (n=45)
ACA (n=102)
ACC (n=45)
B
Mineralocorticoid and mineralocorticoid precursor metabolites
D
Glucocorticoid metabolites
Urinary steroid excretion (µg/24hr)
Urinary steroid excretion (ug/24hr)
100
Urinary Steroid Excretion (ug/24hr)
Urinary Steroid Excretion (ug/24hr)
8
22000
20000
18000
16000
14000
12000
10000
8000
0009
4000
2000
2000
4000
6000
8000
10000
12000
14000
16000
18000
20000-
22000
9
0
8
₹
g
8
&
&
10
£
900
0
0
THA
1
-
F
6a-OH cortisol
4
5a-THA
4
THF
Y
*
THB
1
P
5a-THF
F
5a-THB
a-cortol
**
B-cortol
5x5@-THAIdo
HI
E
H
THDOC
-1
-
THE
4
a-cortolone
-
”
5a-THDOC
1
B-cortolone
IL
ACA (n=102)
ACC (n=45)
ACA (n=102)
ACC (n=45)
| Metabolite subclass | Urinary excretion of steroid metabolite subclasses (µg/24 h) | ||
|---|---|---|---|
| Controls (n = 88) | ACA (n = 102) | ACC (n = 45) | |
| Androgen metabolites (steroids 1-2) | |||
| Median | 2,787 | 1,426 | 4,679 |
| IQR | 1,808-4,305 | 724-2,292 P= 4 × 10-9ª | 1,936-8,807 P = 0.31ª |
| P = 1 × 10-10b | |||
| Androgen precursor metabolites (steroids 3-6) | |||
| Median | 493 | 717 | 8,417 |
| IQR | 320-944 | 389-1,115 | 2,543-57,453 |
| P = 0.24 | P = 6 × 10-16 | ||
| P= 8 × 10-12 | |||
| Mineralocorticoid metabolites (steroids 7-11) | |||
| Median | 598 | 568 | 667 |
| IQR | 355-941 | 381-824 P = 0.84 | 301-1,149 P = 0.84 |
| P = 0.84 | |||
| Mineralocorticoid precursor metabolites (steroids 12-13) | |||
| Median | 25 | 20 | 122 |
| IQR | 13-44 | 14-36 | 52-333 |
| P = 1.00 | P= 3 × 10-4 | ||
| P = 5 × 10-11 | |||
| Glucocorticoid precursor metabolites (steroids 14-19) | |||
| Median | 973 | 879 | 7,646 |
| IQR | 570-1,317 | 545-1521 | 3,704-19,103 |
| P = 1.0 | P = 6 × 10-15 | ||
| P = 3 × 10-17 | |||
| Glucocorticoid metabolites (steroids 20-32) | |||
| Median | 7,763 | 11,655 | 14,526 |
| IQR | 5,639-11,382 | 8,454-15,906 P = 2 × 10-5 | 8,587-37,802 P= 2 × 10-7 P = 0.18 |
Steroid subclasses comprise the sum of individual steroids as specified, with numbering of individual steroids referring to Fig. 1. Statistical analysis was performed employing Kruskal-Wallis nonparametric testing and Dunn’s post hoc test. IQR, Interquartile range.
a Comparison of controls vs. ACA and ACC, respectively.
b Comparison of ACA vs. ACC.
precursors of androgen, mineralocorticoid, and glucocor- ticoid synthesis (all P < 0.001 vs. controls; Table 2). Even when analyzing the 12 ACC patients without evidence of hormone excess on routine biochemistry, urinary excre- tion of androgen and glucocorticoid precursor metabo- lites remained significantly increased (P = 0.015 and P = 0.002, respectively; Supplemental Table 3). By contrast, steroid precursor metabolite excretion in ACA did not differ from healthy controls (Table 2).
Urinary excretion of androgen metabolites was signif- icantly decreased in ACA (P < 0.001; Table 2), as also suggested by the ACA heat map (steroids 1-6 in Supple- mental Fig. 1). However, the subgroup of ACA patients with evidence of hormone excess on routine biochemistry showed normal androgen metabolite excretion (P = 0.13; Supplemental Table 3). Compared with controls, urinary excretion of active glucocorticoid metabolites was signif- icantly increased in both ACA and ACC (both P< 0.001;
Table 2). Among ACC patients, only those with abnormal routine biochemistry had increased glucocorticoid metab- olite excretion (P < 0.001; Supplemental Table 3). In con- trast, increased glucocorticoid metabolite excretion was similarly present in ACA patients with and without ab- normal routine biochemistry (both P < 0.001 vs. controls; Supplemental Table 3).
When analyzed by sum of steroid subclass metabolites, 82% of ACC showed increased excretion of at least one steroid subclass, compared with only 29% in ACA and 16% in controls. The majority of ACC patients (69%) showed androgen precursor excess, either combined with increased excretion of glucocorticoid precursors (36%) or both glucocorticoid and mineralocorticoid precursors (33%). Isolated glucocorticoid precursor excess was found in 9% of ACC and 1% of ACA patients. Combined mineralocorticoid and glucocorticoid precursor excess was found in 4% of ACA and 2% of ACC. None of the
patients had isolated androgen metabolite excess; isolated androgen precursor metabolite excess was found in 2% of both ACA and ACC patients.
We did not detect any effect of age, sex, tumor size, or presence of metastasis on the above described character- istics of the urine steroid metabolomes in ACC and ACA patients.
Machine learning classification performance and selection of differentiating steroid markers
GMLVQ analysis was applied to the steroid data to achieve classification of the samples by determining a dis- criminative distance measure that characterizes the differ- ences between ACA and ACC. Figure 3A depicts the cor- responding relevance matrix, achieved by averaging 1000 analysis runs with random splits of the steroid excretion data into training set (90%) and test set (10%). Analysis of the diagonal elements of the relevance matrix revealed that three steroid metabolites, THS, 5-PT, and 5-PD, were most informative in discriminating ACC from ACA (Fig. 3B). In addition, we identified the six next most discrim- inative markers (Fig. 3C). The selection of fewer than 10 steroid markers was arbitrarily decided to facilitate future transfer of our biomarker tool to a high-throughput mea- surement platform, i.e. tandem mass spectrometry, which is limited by the number of steroids that can be analyzed in a single run. The excretion values of the identified nine most discriminative steroid metabolites in ACA and ACC patients are depicted in Fig. 3D.
Receiver-operating characteristics (ROC) analysis re- vealed an excellent performance of GMLVQ (Fig. 3E). The use of three and nine markers, selected individually in each training run, was only slightly inferior to the ROC curve generated from the entire data set of all 32 steroids, with similar sensitivities and specificities all at or slightly below sensitivity = specificity = 90% (Fig. 3E).
Compared with GMLVQ, LDA performed relatively poorly when applied to the total set of 32 steroid markers, due to strong overfitting effects. However, overfitting was not present anymore when LDA was applied to the subsets of the three and nine most discriminative markers, respec- tively, after those had been selected by GMLVQ. Simi- larly, logistic regression fitting failed to converge when applied to the full data set, but when applied to the two steroid marker subsets, logistic regression produced results that were slightly inferior but comparable to GMLVQ. Thus, both LDA and logistic regression confirmed the validity of the steroid marker selection by GMLVQ.
Discussion
Urinary steroid profiling with subsequent machine learn- ing analysis of steroid excretion data differentiates benign
and malignant adrenocortical tumors with high sensitivity and specificity. Our sensitivity and specificity of 90% is superior to the diagnostic value of imaging techniques such as CT, magnetic resonance imaging, and positron emission tomography, currently employed in the differ- ential diagnosis of adrenal incidentalomas (19, 30-34). In particular, the high specificity of our method renders it invaluable in the detection of adrenocortical malignancy, a diagnosis that is virtually impossible to achieve with any single imaging technique.
In contrast to a random blood sample, the analysis of urinary steroid metabolite excretion also comes with the added advantage of noninvasive assessment of the 24-h net steroid production and thus a higher sensitivity in de- tecting abnormal steroid secretion. Our results indicate that combined androgen and glucocorticoid excess is a characteristic feature in ACC, identified by urinary steroid profiling in 69% of cases but in only 27% of cases by routine biochemistry. Routine serum biochemistry found that the glucocorticoid precursor 17-HP was elevated in 56% of ACC and 6% of ACA patients. However, mass spectrometry-based steroid excretion profiling much more strikingly demonstrated a characteristic accumula- tion of precursor steroids rather than end products of ad- renal steroidogenesis in 85% of ACC patients, across all steroid classes, thereby indicating early-stage, immature steroidogenesis as a characteristic of malignant adrenal tumors. The high incidence of steroid precursor excess supports the concept of dedifferentiation of ACC cells, resulting in a shift from full steroidogenic potential toward proliferative capacity. Of note, steroid profiling also de- tected significantly increased steroid precursor excretion in those ACC (27%) that had been classified as endocrine inactive based on the results of routine biochemical work-up.
We identified the 11-deoxycortisol metabolite THS as the most discriminative steroid in differentiating ACC from ACA. THS was significantly increased in both ACA and ACC patients compared with controls, but excretion levels were significantly higher in ACC than in ACA, sug- gesting inhibition or lack of expression of 11ß-hydroxy- lase, the enzyme that converts the glucocorticoid precur- sor 11-deoxycortisol to active cortisol. Increased secretion of 11-deoxycortisol by adrenal tumors has been previ- ously described (22) but has not been used for differenti- ating benign from malignant adrenal tumors.
Adrenal-derived androgen excess was a characteristic feature of ACC and detected with a much higher sensitivity by our biomarker tool that demonstrated urinary andro- gen precursor excess in 71% of ACC and 2% of ACA, whereas serum DHEAS was increased in only 42% and 2%, respectively. By contrast, low urinary androgen me-
35000
A
0.06
D
35000
Urinary steroid excretion (µg/24hr)
Urinary steroid excretion (µg/24hr)
30000
25000
20000
15000
10000
5000
0009
0000L
15000
0,000
25000
0000€
0.05
5
0
0
THS
0.04
5-PT
10
0.03
5-PD
PT
0.02
15
THDOC
0.01
5xTHA
20
Etio
0
5x THF
0.01
25
PD
ACA (n=102)
ACC (n=45)
-0.02
30
E
1
-0.03
5
10
15
20
25
30
B
0.8
0.1
Sensitivity
0.08
0.6
32
0.06
0.9
9
0.4
3
0.04
0.8
0.02
0.2
0
0.7
0
0.1
0.2
0.3
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
0
C
0
0.2
0.4
0.6
0.8
1
100
1-Specificity
80
60
40
20
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31 3 32
| 32 | 9 | 3 | |
|---|---|---|---|
| AUC (±SD) | 0.965± | 0.960± | 0.942± |
| 0.054 | 0.049 | 0.072 | |
| spec=sens at | 90.0% | 87.7% | 87.2% |
| sens >99% at spec= | 29.2% | 29.0% | 12.1% |
| sens >95% at spec= | 71.2% | 70.1% | 56.9% |
| spec >99% at sens= | 82.8% | 80.8% | 77.0% |
| spec >95% at sens= | 85.8% | 83.2% | 83.5% |
FIG. 3. Results of GMLVQ analysis. A, Relevance matrix as obtained by GMLVQ as an average of 1000 randomized training runs. The panel provides a gray-scale representation of the off-diagonal elements in the average relevance matrix. Both the x- and the y-axes correspond to the numbering of individual steroids (n = 32) in Fig. 1 and Supplemental Table 1; each square corresponds to the combination of two steroids. Large positive and negative values as represented by bright and dark squares, respectively, indicate that the corresponding pair of steroid markers is highly relevant for the discrimination of ACC from ACA. The gray scale on the right defines the numerical values of the matrix elements. For clarity, the diagonal elements of the relevance matrix have been omitted, as indicated by the white line. B, All 32 diagonal elements of the relevance matrix, quantifying the significance of each single steroid marker for the discrimination of ACC from ACA (color code as in Fig. 1B), with all significances adding up to the sum of 1. Error bars correspond to the observed sD over the 1000 randomized training runs. C, Respective percentages of the 1000 randomized training runs in which single steroid features were identified as being among the nine most relevant features for the differentiation of ACC from ACA. D, Steroid biomarkers selected after GMLVQ analysis as the nine most relevant markers for differentiating ACC from ACA. Box plots represent median and interquartile ranges; the whiskers represent 5th and 95th percentile, respectively (color code as in Fig. 1B). E, ROC curve for all steroid metabolites (n = 32) and the three and nine steroid markers identified as most discriminating. The inset represents a magnification of the upper left-hand corner of the ROC curves, provided for visual clarity. Numerical characteristics are shown of the threshold-average ROC curves for all 32 steroids and the subsets of three steroids (THS, 5-PT, and 5-PD) and nine steroids [THS, 5-PT, 5-PD, PT, THDOC, 5aTHA Etio, 5@THF, and PD; for explanation of steriod metabolite abbreviations see Supplemental Table 1] identified as most discriminative after GMLVQ analysis. ROC curves and all values for areas under the curve (AUC) and sensitivity (sens) and specificity (spec) correspond to average test set performances over 1000 random splits of the data set into 90% training data and 10% test data.
tabolite excretion was a characteristic feature in ACA, in particular in those classified as endocrine inactive by rou- tine biochemistry. Interestingly, steroid profiling revealed significantly elevated glucocorticoid metabolite excretion in all ACA patients, i.e. also in those with no evidence of hormone excess in the routine biochemical work-up. This
would support the concept of a continuum of hormone secretion in benign adrenal adenomas from increased uri- nary metabolite excretion to clinically overt disease. Uri- nary steroid profiling may represent an exquisitely sensi- tive tool for the quantification of steroid excess in the follow-up of patients with subclinical Cushing’s syndrome
(9, 10) and also borderline primary hyperaldosteronism. Similarly, steroid profiling could be employed to detect the tumor-specific steroid fingerprint when monitoring for re- currence or treatment response in ACC.
Importantly, our approach combining mass spectrom- etry-based steroid profiling and GMLVQ analysis identi- fied nine steroids selected as the most discriminatory in differentiating ACC from ACA, with similar classification performance of this subset in comparison to all quantified steroids. These results will facilitate the transfer of this profiling approach from the time-consuming GC/MS plat- form to high-throughput liquid chromatography/tandem mass spectrometry. Rapid profiling of up to 10 steroids by liquid chromatography/tandem mass spectrometry is en- tirely feasible (35-37) and will allow the development of a diagnostic screening test that is sensitive, specific, fast, and cost effective. It would be an exciting prospect for the clinical management of patients with ACC if urine steroid metabolomics would also be proven as a sensitive tool for the early detection of tumor recurrence, thus potentially obviating the need for frequent follow-up imaging if sim- ilarly sensitive as imaging. Furthermore, future studies will need to determine whether the steroid metabolome fin- gerprint of a tumor correlates with clinical behavior and prognosis.
Our study has the limitation of being retrospective. However, through collaboration with many centers across Europe, here we have already studied a large number of patients with ACC and ACA in whom tumor dignity, i.e. the definition of benign vs. malignant, had been ascer- tained as conclusively as possible, including a long period of follow-up that ensured detection of recurrence/metas- tasis (ACC) or lack of recurrence (ACA). It is possible that a small minority of the 102 ACA represent early-stage ACC that have been successfully removed but did not re- cur. However, this is unlikely and to a degree unavoidable given the inherent difficulties in the histopathological as- sessment of adrenal tumors (15). The availability of these unique and well characterized gold standard reference co- horts of ACA and ACC patients, facilitated by European collaboration, were a mandatory precondition for provid- ing proof of concept for our method.
In conclusion, our data provide strong evidence for the discriminative power of the selected steroid markers, clearly indicating that urine steroid metabolomics, i.e. the combination of mass spectrometry-based steroid profiling and subsequent computational data analysis, represents a novel and highly promising biomarker approach to the differential diagnosis of adrenal tumors. Before imple- mentation of our novel approach as a diagnostic test in routine clinical practice, prospective validation in large
cohorts of patients with adrenal tumors is warranted, which we are currently embarking on.
Acknowledgments
We are grateful to the European Network for the Study of Ad- renal Tumors (ENS@T; www.ensat.org) for invaluable help with the collection of samples for this study. We thank Jon Deeks (Professor of Biostatistics, University of Birmingham, Birming- ham, UK) and Richard Riley (Senior Lecturer in Medical Statis- tics, University of Birmingham) for their critical review of the manuscript and helpful discussions.
Address all correspondence and requests for reprints to: Professor Wiebke Arlt M.D., D.Sc., F.R.C.P., F.Med.Sci., Centre for Endocrinology, Diabetes, and Metabolism (CEDAM), School of Clinical and Experimental Medicine, University of Birmingham, Birmingham B15 2TT, United Kingdom. E-mail: w.arlt@bham.ac.uk.
This work was supported by the Medical Research Council UK (MRC Strategic Grant G0801473 to W.A. and P.M.S.) and the ENS@T. ENS@T gratefully acknowledges support by the European Science Foundation (Research Networking Program ESF-ENS@T), the European Commission (FP7 Collaborative Research Project ENS@T-CANCER 259735), and the Claire Khan Adrenal Trust Fund.
Disclosure Summary: W.A. and P.M.S. are inventors and M.B. is a contributor to a patent application on the use of steroid profiling as a biomarker tool in the differential diagnosis of ste- roid-producing and steroid-dependent tumors (PCT/GB2010/ 000274). All other authors did not declare a conflict of interest.
References
1. Grumbach MM, Biller BM, Braunstein GD, Campbell KK, Carney JA, Godley PA, Harris EL, Lee JK, Oertel YC, Posner MC, Schlechte JA, Wieand HS 2003 Management of the clinically inapparent ad- renal mass (“incidentaloma”). Ann Intern Med 138:424-429
2. Mansmann G, Lau J, Balk E, Rothberg M, Miyachi Y, Bornstein SR 2004 The clinically inapparent adrenal mass: update in diagnosis and management. Endocr Rev 25:309-340
3. Barzon L, Sonino N, Fallo F, Palu G, Boscaro M 2003 Prevalence and natural history of adrenal incidentalomas. Eur J Endocrinol 149:273-285
4. Terzolo M, Reimondo G, Angeli A 2009 Definition of an optimal strategy to evaluate and follow-up adrenal incidentalomas: time for further research. Eur J Endocrinol 161:529-532
5. Young Jr WF 2007 Clinical practice. The incidentally discovered adrenal mass. N Engl J Med 356:601-610
6. Cawood TJ, Hunt PJ, O’Shea D, Cole D, Soule S 2009 Recom- mended evaluation of adrenal incidentalomas is costly, has high false-positive rates and confers a risk of fatal cancer that is similar to the risk of the adrenal lesion becoming malignant: time for a rethink? Eur J Endocrinol 161:513-527
7. Nieman LK 2010 Approach to the patient with an adrenal inciden- taloma. J Clin Endocrinol Metab 95:4106-4113
8. Mantero F, Terzolo M, Arnaldi G, Osella G, Masini AM, Alì A, Giovagnetti M, Opocher G, Angeli A 2000 A survey on adrenal incidentaloma in Italy. Study Group on Adrenal Tumors of the Ital- ian Society of Endocrinology. J Clin Endocrinol Metab 85:637-644
9. Reincke M, Nieke J, Krestin GP, Saeger W, Allolio B, Winkelmann W 1992 Preclinical Cushing’s syndrome in adrenal “incidentalo- mas”: comparison with adrenal Cushing’s syndrome. J Clin Endo- crinol Metab 75:826-832
10. Terzolo M, Bovio S, Reimondo G, Pia A, Osella G, Borretta G, Angeli A 2005 Subclinical Cushing’s syndrome in adrenal inciden- talomas. Endocrinol Metab Clin North Am 34:423-439, x
11. Stewart PM 2010 Is subclinical Cushing’s syndrome an entity or a statistical fallout from diagnostic testing? Consensus surrounding the diagnosis is required before optimal treatment can be defined. J Clin Endocrinol Metab 95:2618-2620
12. Allolio B, Fassnacht M 2006 Adrenocortical carcinoma: clinical update. J Clin Endocrinol Metab 91:2027-2037
13. Fassnacht M, Johanssen S, Quinkler M, Bucsky P, Willenberg HS, Beuschlein F, Terzolo M, Mueller HH, Hahner S, Allolio B German Adrenocortical Carcinoma Registry Group; European Network for the Study of Adrenal Tumors 2009 Limited prognostic value of the 2004 International Union Against Cancer staging classification for adrenocortical carcinoma: proposal for a Revised TNM Classifica- tion. Cancer 115:243-250
14. Weiss LM, Medeiros LJ, Vickery Jr AL 1989 Pathologic features of prognostic significance in adrenocortical carcinoma. Am J Surg Pathol 13:202-206
15. Lau SK, Weiss LM 2009 The Weiss system for evaluating adreno- cortical neoplasms: 25 years later. Hum Pathol 40:757-768
16. de Fraipont F, El Atifi M, Cherradi N, Le Moigne G, Defaye G, Houlgatte R, Bertherat J, Bertagna X, Plouin PF, Baudin E, Berger F, Gicquel C, Chabre O, Feige JJ 2005 Gene expression profiling of human adrenocortical tumors using complementary deoxyribonu- cleic Acid microarrays identifies several candidate genes as markers of malignancy. J Clin Endocrinol Metab 90:1819-1829
17. de Reyniès A, Assié G, Rickman DS, Tissier F, Groussin L, René- Corail F, Dousset B, Bertagna X, Clauser E, Bertherat J 2009 Gene expression profiling reveals a new classification of adrenocortical tumors and identifies molecular predictors of malignancy and sur- vival. J Clin Oncol 27:1108-1115
18. Giordano TJ, Kuick R, Else T, Gauger PG, Vinco M, Bauersfeld J, Sanders D, Thomas DG, Doherty G, Hammer G 2009 Molecular classification and prognostication of adrenocortical tumors by tran- scriptome profiling. Clin Cancer Res 15:668-676
19. Hamrahian AH, Ioachimescu AG, Remer EM, Motta-Ramirez G, Bogabathina H, Levin HS, Reddy S, Gill IS, Siperstein A, Bravo EL 2005 Clinical utility of noncontrast computed tomography attenu- ation value (Hounsfield units) to differentiate adrenal adenomas/ hyperplasias from nonadenomas: Cleveland Clinic experience. J Clin Endocrinol Metab 90:871-877
20. Boland GW, Dwamena BA, Jagtiani Sangwaiya M, Goehler AG, Blake MA, Hahn PF, Scott JA, Kalra MK 2011 Characterization of adrenal masses by using FDG PET: a systematic review and meta- analysis of diagnostic test performance. Radiology 259:117-126
21. Minowada S, Kinoshita K, Hara M, Isurugi K, Uchikawa T, Niijima T 1985 Measurement of urinary steroid profile in patients with adrenal tumor as a screening method for carcinoma. Endocrinol Jpn 32:29-37
22. Doerr HG, Sippell WG, Drop SL, Bidlingmaier F, Knorr D 1987
Evidence of 11ß-hydroxylase deficiency in childhood adrenocorti- cal tumors. The plasma corticosterone/11-deoxycorticosterone ra- tio as a possible marker for malignancy. Cancer 60:1625-1629
23. Grondal S, Eriksson B, Hagenas L, Werner S, Curstedt T 1990 Steroid profile in urine: a useful tool in the diagnosis and follow up of adrenocortical carcinoma. Acta Endocrinol (Copenh) 122:656- 663
24. Małunowicz EM, Ginalska-Malinowska M, Romer TE, Ruszc- zynska-Wolska A, Dura M 1995 Heterogeneity of urinary steroid profiles in children with adrenocortical tumors. Horm Res 44:182- 188
25. Shackleton CH, Marcos J 2011 GC/MS steroid profiling: diagnosis of disorders affecting steroid synthesis and metabolism. In: Gross M, Caprioli R, eds. The encyclopedia of mass spectrometry. Vol 8. Am- sterdam: Elsevier; 789-813
26. Krone N, Hughes BA, Lavery GG, Stewart PM, Arlt W, Shackleton CH 2010 Gas chromatography/mass spectrometry (GC/MS) re- mains a pre-eminent discovery tool in clinical steroid investigations even in the era of fast liquid chromatography tandem mass spec- trometry (LC/MS/MS). J Steroid Biochem Mol Biol 121:496-504
27. Kohonen T 1997 Self-organizing maps. Berlin: Springer
28. Schneider P, Biehl M, Hammer B 2009 Adaptive relevance matrices in learning vector quantization. Neural Comput 21:3532-3561
29. McCullagh P, Nelder JA 1989 Generalized linear models. London: Chapman and Hall
30. Ilias I, Sahdev A, Reznek RH, Grossman AB, Pacak K 2007 The optimal imaging of adrenal tumours: a comparison of different methods. Endocr Relat Cancer 14:587-599
31. Groussin L, Bonardel G, Silvéra S, Tissier F, Coste J, Abiven G, Libé R, Bienvenu M, Alberini JL, Salenave S, Bouchard P, Bertherat J, Dousset B, Legmann P, Richard B, Foehrenbach H, Bertagna X, Tenenbaum F 2009 18F-Fluorodeoxyglucose positron emission to- mography for the diagnosis of adrenocortical tumors: a prospective study in 77 operated patients. J Clin Endocrinol Metab 94:1713- 1722
32. Gratz S, Kemke B, Kaiser W, Heinis J, Behr TM, Höffken H 2010 Incidental non-secreting adrenal masses in cancer patients: intra- individual comparison of 18F-fluorodeoxyglucose positron emission tomography/computed tomography with computed tomography and shift magnetic resonance imaging. J Int Med Res 38:633-644
33. Blake MA, Cronin CG, Boland GW 2010 Adrenal imaging. AJR Am J Roentgenol 194:1450-1460
34. Korobkin M, Brodeur FJ, Francis IR, Quint LE, Dunnick NR, Londy F 1998 CT time-attenuation washout curves of adrenal adenomas and nonadenomas. AJR Am J Roentgenol 170:747-752
35. Soldin SJ, Soldin OP 2009 Steroid hormone analysis by tandem mass spectrometry. Clin Chem 55:1061-1066
36. Janzen N, Sander S, Terhardt M, Peter M, Sander J 2008 Fast and direct quantification of adrenal steroids by tandem mass spectrom- etry in serum and dried blood spots. J Chromatogr B Analyt Technol Biomed Life Sci 861:117-122
37. Rossi C, Calton L, Hammond G, Brown HA, Wallace AM, Sac- chetta P, Morris M 2010 Serum steroid profiling for congenital adrenal hyperplasia using liquid chromatography-tandem mass spectrometry. Clin Chim Acta 411:222-228