Integrated Analytical and Statistical Two-Dimensional Spectroscopy

Feb 27, 2017 - As part of the strategy, we introduce a new probabilistic statistical spectroscopy tool, RED-STORM (Resolution EnhanceD SubseT Optimiza...
5 downloads 3 Views 1MB Size
Subscriber access provided by UB + Fachbibliothek Chemie | (FU-Bibliothekssystem)

Article

An Integrated Analytical and Statistical Two-Dimensional Spectroscopy Strategy for Metabolite Identification: Application to Dietary Biomarkers Joram M. Posma, Isabel Garcia-Perez, James Christopher Heaton, Paula Burdisso, John C. Mathers, John Draper, Matthew R. Lewis, John C Lindon, Gary S. Frost, Elaine Holmes, and Jeremy Kirk Nicholson Anal. Chem., Just Accepted Manuscript • DOI: 10.1021/acs.analchem.6b03324 • Publication Date (Web): 27 Feb 2017 Downloaded from http://pubs.acs.org on March 1, 2017

Just Accepted “Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted online prior to technical editing, formatting for publication and author proofing. The American Chemical Society provides “Just Accepted” as a free service to the research community to expedite the dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been fully peer reviewed, but should not be considered the official version of record. They are accessible to all readers and citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors or consequences arising from the use of information contained in these “Just Accepted” manuscripts.

Analytical Chemistry is published by the American Chemical Society. 1155 Sixteenth Street N.W., Washington, DC 20036 Published by American Chemical Society. Copyright © American Chemical Society. However, no copyright claim is made to original U.S. Government works, or works produced by employees of any Commonwealth realm Crown government in the course of their duties.

Page 1 of 28

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 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

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

Analytical Chemistry

JOURNAL Analytical Chemistry (ac-2016-03324s.R2) TITLE An Integrated Analytical and Statistical Two-Dimensional Spectroscopy Strategy for Metabolite Identification: Application to Dietary Biomarkers AUTHOR NAMES Joram M. Posma,†,⊥ Isabel Garcia-Perez,†,‡,⊥ James C. Heaton,†,∆ Paula Burdisso,†,# John C. Mathers,§ John Draper,ǁ Matt Lewis,†,∩ John C. Lindon,† Gary Frost,‡ Elaine Holmes,†,∩,* and Jeremy K. Nicholson†,∩,* AUTHOR ADDRESS † Section of Biomolecular Medicine, Division of Computational and Systems Medicine, Department of Surgery and Cancer, Faculty of Medicine, South Kensington Campus, Imperial College London, London, SW7 2AZ, United Kingdom; ‡ Nutrition and Dietetic Research Group, Division of Diabetes, Endocrinology and Metabolism, Department of Medicine, Faculty of Medicine, Hammersmith Campus, Imperial College London, London, W12 ONN, United Kingdom; § Human Nutrition Research Centre, Institute of Cellular Medicine, Newcastle University, United Kingdom; ǁ Institute of Biological, Environmental and Rural Sciences, Aberystwyth University, United Kingdom; ∩ MRC-NIHR National Phenome Centre, Department of Surgery and Cancer, Faculty of Medicine, Hammersmith Campus, Imperial College London, London, W12 0NN, United Kingdom KEYWORDS Analytical Framework, Bayesian Analysis, Controlled Clinical Trial, Dietary Biomarkers, Food Challenge, Gaussian Mixture Model, Markov-Chain Monte-Carlo, Statistical Spectroscopy

1 ACS Paragon Plus Environment

Analytical Chemistry

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 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

26 27

ABSTRACT A major purpose of exploratory metabolic profiling is for the identification of molecular

28

species that are statistically associated with specific biological or medical outcomes;

29

unfortunately the structure elucidation process of unknowns is often a major bottleneck in

30

this process. We present here new holistic strategies that combine different statistical

31

spectroscopic and analytical techniques to improve and simplify the process of metabolite

32

identification. We exemplify these strategies using study data collected as part of a dietary

33

intervention to improve health and which elicits a relatively subtle suite of changes from

34

complex molecular profiles. We identify three new dietary biomarkers related to the

35

consumption of peas (N-methyl nicotinic acid), apples (rhamnitol) and onions (N-acetyl-S-

36

(1Z)-propenyl-cysteine-sulfoxide) that can be used to enhance dietary assessment and assess

37

adherence to diet. As part of the strategy, we introduce a new probabilistic statistical

38

spectroscopy tool, RED-STORM (Resolution EnhanceD SubseT Optimization by Reference

39

Matching), that uses 2D J-resolved 1H-NMR spectra for enhanced information recovery using

40

the Bayesian paradigm to extract a subset of spectra with similar spectral signatures to a

41

reference. RED-STORM provided new information for subsequent experiments (e.g. 2D-

42

NMR spectroscopy, Solid-Phase Extraction, Liquid Chromatography prefaced Mass

43

Spectrometry) used to ultimately identify an unknown compound. In summary, we illustrate

44

the benefit of acquiring J-resolved experiments alongside conventional 1D 1H-NMR as part

45

of routine metabolic profiling in large datasets and show that application of complementary

46

statistical and analytical techniques for the identification of unknown metabolites can be used

47

to save valuable time and resource.

48

2 ACS Paragon Plus Environment

Page 2 of 28

Page 3 of 28

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 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Analytical Chemistry

49

INTRODUCTION

50

Dietary interventions (DIs) are a cornerstone in the management of reducing the risk of non-

51

communicable diseases1-3 and promoting healthy aging4. However, understanding the

52

response to dietary change is compromised by poor compliance to dietary recommendations

53

and the inherently inaccurate self-reporting dietary recording tools available, with prevalence

54

of misreporting estimated at 30–88%5,6, lowering the value of such studies and data

55

It has been demonstrated that dietary biomarkers can reflect consumption of specific

56

foods and enhance dietary intake assessment at individual and population levels7-17. Dietary

57

biomarkers are based on the concept that food intake is highly correlated with excretion

58

levels of food-related compounds over a given period of time. These ‘biomarkers’ can be

59

compounds that are excreted unchanged10,17 or that have undergone metabolic conversion for

60

example by gut bacteria8,11,13,14. Metabolic profiling of biofluids using spectroscopic

61

technologies18 can detect thousands of compounds simultaneously, generating a profile that

62

can be related to specific states of health or disease. Some of the compounds in these spectral

63

profiles are potential biomarkers of dietary intake. However, finding associations between

64

self-reported dietary intakes and excreted metabolites19,20 in order to discover potential food

65

biomarkers is plagued by inaccuracies of self-reporting. Thus, confirmation of a candidate

66

biomarker is best achieved by using a controlled food challenge with subsequent validation in

67

a larger population or study cohort.17 The complementarity of the main workhorses of metabolic profiling, one-dimensional

68

Nuclear

Magnetic

Resonance

(1H-NMR)

69

proton

spectroscopy

70

chromatograpic-mass spectrometry (MS) techniques, has been extensively demonstrated in

71

the last decade21-23. However, chemical characterization of molecular species associated with

72

an outcome still is a limiting factor of exploratory metabolic profiling. NMR spectroscopy

73

provides an atom-centered spectroscopic tool for structure elucidation that can be enhanced 3 ACS Paragon Plus Environment

and

hyphenated

Analytical Chemistry

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 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

74

by statistical spectroscopic methods24,25 or by physical hyphenation with chromatographic

75

methods such as Solid-Phase Extraction (SPE)26, Liquid Chromatography (LC)21 or by LC-

76

NMR-MS21,27 to achieve a better chemical characterization of endogenous and exogenous

77

metabolites.

78

Metabolite identification in 1H-NMR spectroscopy is aided by the intrinsic correlation

79

of peaks from the same metabolite. Statistical TOtal Correlation SpectroscopY (STOCSY)28

80

makes use of this property by calculating the correlation between one spectral variable

81

(driver) and all other variables to uncover structural associations. In cases where there are

82

sufficient spectra, identification of 1H-NMR peaks using statistical methods is an efficient

83

strategy that utilizes existing spectral data without requiring additional spectroscopic

84

experiments a priori, which has obvious advantages in usage of volume-limited samples and

85

is cost-efficient.23 Since the STOCSY method was published, many derivations have aimed to

86

improve specific properties such as differentiation between structural and pathway

87

correlations by clustering, subset selection or stoichiometric relationships.24

88

Statistical correlation can be undermined by overlapped signals unrelated to the

89

metabolite of interest in a 1D-NMR spectrum and 2D-NMR experiments are still required for

90

unambiguous structure elucidation29. In addition, the structural information obtained using

91

statistical algorithms is dependent on criteria such as correlation thresholds28,30 or correlation-

92

distance cut-offs31. SubseT Optimization by Reference Matching (STORM)32 is a derivation

93

of STOCSY that aims to separate out confounding spectra that do not match a supplied

94

reference spectrum of a potential biomarker signal, thereby showing clearer spectral

95

correlations between variables for both low and high intensity signals. The reference

96

spectrum is a single spectrum that contains the signal of interest. The peak segment is

97

correlated with the same region of all samples and a high correlation indicates the samples

98

contain the same signal and are likely ‘informative’, whereas samples with a low correlation 4 ACS Paragon Plus Environment

Page 4 of 28

Page 5 of 28

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 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Analytical Chemistry

99

do not have this signal and are uninformative. Subsets of spectra and variable correlations are

100

found by carefully correcting for multiple testing in both phases and using statistical

101

shrinkage. Here, we describe a holistic strategy for the identification of unknown metabolites

102

that combines the strengths of statistical spectroscopy, NMR, multiple separation techniques

103

and MS, and apply the strategy to identify three novel dietary biomarkers. In addition, we

104

demonstrate an extension of STORM to 2D-NMR spectra to uncover the identity of unknown

105

metabolites.

106 107

EXPERIMENTAL SECTION

108

Food challenges (FCs) – To discover urinary biomarkers of peas, apples and onions three

109

FCs were designed. A total of nine healthy participants (4 women, 5 men; non-smokers; age:

110

22–32y; BMI: 21.2–25.3 kg/m2) were recruited and assigned to one of the three FCs.

111

Participants were provided with the assigned foods as part of a standardized dinner (incl.

112

125g of chicken breast as a protein source). Incremental amounts of the designated food were

113

consumed over three consecutive days: 60/120/180g for (boiled) peas, 40/80/160g for (raw)

114

apples and 20/40/60g for (fried) onions. For 24hr preceding the FC, and throughout the FC,

115

participants were asked to consume their habitual diet and avoid consumption of

116

coffee/tea/cocoa and any additional amounts of assigned foods. Cumulative urine samples

117

were collected into sterilized single-use urine containers (International Scientific Supplies

118

Ltd., Bradford, U.K.) from dinner up to and including the first morning void. Urine samples

119

were stored at −80°C until analysis.

120

Controlled clinical trial (CCT) – 19 healthy participants (9 women, 10 men; non-smokers;

121

age: 25–60y; BMI: 21.1-33.3 kg/m2) attended the NIHR/Wellcome Trust Imperial Clinical

122

Research Facility for four 3-day inpatient periods, separated by a period of >4 days, with

123

food and drink intake tightly controlled (alcohol/coffee/tea/cocoa were not provided). In 5 ACS Paragon Plus Environment

Analytical Chemistry

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 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 6 of 28

124

random order, participants followed all four DIs representing 100% (Diet 1), 75% (Diet 2),

125

50% (Diet 3) and 25% (Diet 4) of WHO healthy eating guidelines1 with respect to

126

carbohydrates, fats, fiber, fruits, salt sugar and vegetables. Full details of the clinical trial

127

design have been described previously33. Foods consumed relevant to the present study are

128

tabulated in Supporting Information.

129

Each participant collected cumulative urine samples (CS) on each day of each DI

130

from after breakfast to before lunch (CS1), after lunch to before dinner (CS2) and after dinner

131

to before breakfast the following day (CS3). 24hr urine samples were obtained by pooling the

132

cumulative samples. Aliquots of urine were transferred into Eppendorf tubes and stored at

133

−80°C until analysis. All participants provided informed, written consent prior to the CCT

134

(Registration No: ISRCTN-43087333), which was approved by the London Brent Research

135

Ethics Committee (13/LO/0078). All studies were carried out in accordance with the

136

Declaration of Helsinki.

137

1

138

for 5min. All available samples (nFC=27, nCCT=906, for missing CCT data see Supporting

139

Information) were prepared for 1H-NMR spectroscopy following the protocol described in34,

140

mixing 540µL of supernatant with 60µL of pH7.4 phosphate buffer containing trimethylsilyl-

141

[2,2,3,3,-2H4]-propionate as an internal reference standard (‘NMR-buffer’). Water-suppressed

142

1

143

Biospin,

144

(RD−gz,1−90°−t−90°−tm−gz,2−90°−ACQ) with saturation of the water resonance7. RD is the

145

relaxation delay, t is a short delay (4µs), 90° represents a radio frequency (RF) pulse that tips

146

the magnetization by 90°, tm is the mixing time (10ms), gz,1 and gz,2 are magnetic field z-

147

gradients both applied for 1ms, and ACQ is the data acquisition period of 2.73s. 1H-NMR

148

spectra were acquired using 4 dummy scans and 32 scans, 64K time domain points and with a

H-NMR analysis – Aliquots of 600µL of urine samples were centrifuged at 16,000xg at 4°C

H-NMR spectroscopy was performed at 300K on a Bruker 600MHz spectrometer (Bruker Karlsruhe,

Germany)

using

a

standard

6 ACS Paragon Plus Environment

1D

pulse

sequence

Page 7 of 28

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 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Analytical Chemistry

149

spectral window of 20ppm. Prior to Fourier transformation, free induction decays were

150

multiplied by an exponential function corresponding to a line broadening of 0.3Hz. 1H-NMR

151

spectra were normalized to the total urine volume to correct for differences in dilution.

152

1

H–1H 2D J-resolved experiments7 were acquired using a pulse sequence to detect the

153

J-couplings in the second dimension, with suppression of the water resonance (RD–90°–t1–

154

180°–t1–ACQ), where t1 is an incremented time period, RD is 2s and 180° represents a 180°

155

RF pulse, ACQ is 0.41s. J-resolved spectra were acquired using 16 dummy scans and 2 scans,

156

8K points with spectral window of 16.7ppm for f2 and 40 increments with spectral window

157

of 78Hz for f1. Continuous wave irradiation was applied at the water resonance frequency

158

using a 25Hz RF during the RD. A sine-bell apodization function was applied to f2 and a

159

squared sine-bell to f1 of the J-resolved data, followed by Fourier transformation, tilting by

160

45° and symmetrization along f1 before data analysis.

161

A suite of 2D-NMR experiments including 1H–1H TOtal Correlation SpectroscopY 1

H–1H COrrelation SpectroscopY (COSY),

1

H–13C Hetero-nuclear Single

162

(TOCSY),

163

Quantum Coherence (HSQC) and

164

(HMBC) spectroscopy were used for identification purposes25,29.

165

SPE-NMR – Apple extracts were homogenized using a Kenwood KMix Blender for 5min.

166

The puree obtained was filtered using a stainless steel filter and centrifuged for 10min at

167

16,000xg. 2mL of each sample (urine/apple) were lyophilized overnight. Freeze-dried (FD)

168

urine/apple samples were dissolved in 1mL of 50mM sodium phosphate pH8.5 and briefly

169

sonicated prior to being loaded onto a 100 mg/ml Bond Elut™ phenylboronic acid (PBA)

170

SPE-cartridge (Agilent Technologies, Stockport, U.K.). The SPE-cartridge was conditioned

171

with 1mL of 70:30v/v H2O:ACN, 0.1M HCl followed by 1mL of 50mM sodium phosphate

172

pH10. After conditioning, the sample was loaded onto the SPE-cartridge. The loaded sample

173

was washed with 2mL 50:50v/v ACN:sodium phosphate (10mM) pH8.5. The sample was

1

H–13C Hetero-nuclear Multiple-Bond Correlation

7 ACS Paragon Plus Environment

Analytical Chemistry

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 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

174

eluted using 1mL of acidified solutions (0.1M HCl) of 100% H2O, 90:10v/v H2O:ACN and

175

70:30v/v H2O:ACN and subsequently lyophilized until dry. The dried eluent fractions were

176

reconstituted in 540µL H2O and 60µL of NMR-buffer, vortexed and centrifuged prior to

177

NMR analysis.

178

LC-NMR-MS27 – 5mL of urine collected overnight after consumption of onion was

179

lyophilized overnight, reconstituted in 500µL of the original urine sample and vortexed,

180

sonicated and centrifuged (20min at 16,000xg). The supernatant was repeatedly injected

181

(7×2µL) onto a reversed-phase HPLC column (Waters Atlantis-T3, 3µm, 4.6×150mm at

182

30°C) in a Waters Acquity UPLC comprising a binary solvent manager and photodiode array

183

detector with a Waters CTC auto-sampler with 100µL sampling needle, and eluted at

184

0.8mL/min using the following gradient: 0.0–60.0min (99.9:0.1% H2O/formic acid), 60.01–

185

65.0min (99.9:0.1% methanol/formic acid), 65.01–127.5min (99.9:0.1% H2O/formic acid).

186

The chromatographic separation of the sample was fractionated using a Waters Fraction

187

Collector III. A total of 120 fractions were collected, one every 29s (starting at t=5min,

188

finishing at t=63min), and dried under a stream of nitrogen. Each fraction was re-dissolved in

189

540µL of H2O and 60µL of NMR-buffer and analyzed by 1H-NMR. A volume of 50µL of the

190

fraction containing the unknown metabolite was analyzed by reversed-phase LC-MS, Waters

191

Acquity Ultra Performance LC system coupled to Xevo G2 Q-TOF mass spectrometer

192

(Waters, Milford, MA, U.S.A.), following an established metabolic profiling method35. The

193

optimized capillary voltage, cone voltage and collision energy were 3kV/20V/4V for ESI+

194

and 1.5kV/30V/6V for ESI-, using a source temperature of 120°C and desolvation

195

temperature of 600°C. Desolvation was 1000L/h for both and cone gas flows 50L/h (ESI+)

196

and 150L/h (ESI-). Leucine enkephalin was used as the reference lock mass at 556.2771

197

([M+H]+) and 554.2615 ([M-H]-).

8 ACS Paragon Plus Environment

Page 8 of 28

Page 9 of 28

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 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Analytical Chemistry

198

RED-STORM algorithm – Here, STORM was extended in a probabilistic framework and

199

modified for applicability to 2D data to provide a clearer signature of structural correlations

200

in the data and the extended algorithm was named Resolution EnhanceD SubseT

201

Optimization by Reference Matching (RED-STORM).

202

High correlations in the data are of interest between samples and a reference spectrum

203

of a signal of interest (for subset optimization) and between a driver and all other variables

204

(for assessing variable importance). However, high correlations are not normally distributed

205

because of the upper bound on the correlation ([-1,1]), this results in their distributions being

206

negatively skewed. Therefore, the correlation () is transformed to a Fisher z-score, which

207

results in approximately normally distributed data that can be analyzed using parametric

208

methods:

209

1.

210

Next, the distribution of all z-scores is modelled using a Gaussian Mixture Model (GMM). A

211

GMM is a weighted sum of k Gaussian clusters and is defined as

212

2.

213

Where  has  data points,  are the  means,   the  variances,  the mixture weights for

214





 =  ln 

|,   ,  = ∑  ,  ,  

each of  Gaussians ( ∑  = 1 ) and , a normalized Gaussian distribution with

215

specified  and . In order to obtain clusters of variables from the Fisher z-transformed

216

correlations a parameter-free GMM is used36. The model automatically learns the optimal

217

number of clusters from the data. For completeness a description of the method in brief

218

follows, for mathematical proofs see36.

219

Each cluster mean (  ), and precision (   ) from equation 2 are given Gaussian

220 221

(  |", # ∼ , ", #   ) and Gamma (   |%, & ∼ Γ%, &   ) priors, respectively. Here, " and # are hyperparameters for the means of all clusters. Similarly, % (shape) and & 9

ACS Paragon Plus Environment

Analytical Chemistry

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 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 10 of 28

222

(scale) are hyper-parameters for the cluster precisions. The hyper-parameters themselves are

223

initialized from the observation mean (( ) and variance (( ) as vague priors ( " ∼

224

, ( , (  ; # ∼ Γ1, (   ; %   ∼ Γ1,1 ; & ∼ Γ1, (  ). The posterior

225

distributions of the cluster means are obtained from

226

3.

229

(,- .-/ 01



,

2- .-/ 1 2- .-/ 1

3

̃ = ∑87 456, 7

with

and

* =

∑87 456 ,

227

228

 |̃ , * ,  , ", # ∼ , +

where ̃ is the sum of observations from cluster 9 , * the number of observations with highest probability for cluster 9, :7 the cluster with highest probability for observation ; and

230

456, a Kronecker delta product. Similarly, the posterior distributions of the cluster precisions

231

are obtained from

232

4.

233

The posteriors for " ( "|, # ∼ , +

234

1,

235

the priors, whereas posterior for % takes a different form and makes use of the fact that

236



>? @



 |* , %, & ∼ Γ 2 - 3 A with B = ∑87 456 , 7 −   -

DE .EF/ 1 ∑G -HI D-

 

.E/ ∑G -HID- 0

/

.EF/ 1

A) and & (&|  , % ∼ Γ +% + 1,



, .F/ 13 ), # ( #|, " ∼ Γ 

F/ .EF/ > ∑G -HI .-

3) are obtained similarly to

J%|  , &  is log-concave36: %|



> 

, &  ∝ Γ 



>

LM > 

GNFO /

N

>.-F/ ?

∏