King s Research Portal - CORE fileKing s Research Portal DOI: 10.1016/j.jalz.2016.08.003 Document...

King’s Research Portal

DOI:10.1016/j.jalz.2016.08.003

Document VersionPublisher's PDF, also known as Version of record

Link to publication record in King's Research Portal

Citation for published version (APA):Proitsi, P., Kim, M., Whiley, L., Simmons, A., Sattlecker, M., Velayudhan, L., ... Legido-Quigley, C. (2016).Association of blood lipids with Alzheimer's disease: A comprehensive lipidomics analysis. Alzheimer's &Dementia. 10.1016/j.jalz.2016.08.003

Citing this paperPlease note that where the full-text provided on King's Research Portal is the Author Accepted Manuscript or Post-Print version this maydiffer from the final Published version. If citing, it is advised that you check and use the publisher's definitive version for pagination,volume/issue, and date of publication details. And where the final published version is provided on the Research Portal, if citing you areagain advised to check the publisher's website for any subsequent corrections.

General rightsCopyright and moral rights for the publications made accessible in the Research Portal are retained by the authors and/or other copyrightowners and it is a condition of accessing publications that users recognize and abide by the legal requirements associated with these rights.

•Users may download and print one copy of any publication from the Research Portal for the purpose of private study or research.•You may not further distribute the material or use it for any profit-making activity or commercial gain•You may freely distribute the URL identifying the publication in the Research Portal

Take down policyIf you believe that this document breaches copyright please contact [email protected] providing details, and we will remove access tothe work immediately and investigate your claim.

Download date: 18. Feb. 2017

http://dx.doi.org/10.1016/j.jalz.2016.08.003

https://kclpure.kcl.ac.uk/portal/en/publications/association-of-blood-lipids-with-alzheimers-disease(839259c2-ac1a-4b3c-92c9-3a218d1f76ec).html

Featured Article

Association of blood lipids with Alzheimer’s disease:A comprehensive lipidomics analysis

Q6 P. Proitsia,*,1, M. Kimb,1, L. Whileyb, A. Simmonsa,c, M. Sattleckera,c, L. Velayudhana,d,M. K. Luptone, H. Soininenf, I. Kloszewskag, P. Mecoccih, M. Tsolakii, B. Vellasj,

S. Lovestonek, J. F. Powella,2, R. J. B. Dobsona,c,2, C. Legido-Quigleyb,2Q1

aKing’s College London, Institute of Psychiatry, Psychology & Neuroscience, London, UKQ2bKing’s College London, Institute of Pharmaceutical Science, London, UK

cNIHR Biomedical Research Centre for Mental Health and Biomedical Research Unit for Dementia at South London and Maudsley NHS Foundation TrustdOxleas NHS Foundation Trust

eQIMR Berghofer Medical Research Institute, Brisbane, AustraliafDepartment of Neurology, Kuopio University Hospital and University of Eastern Finland, Kuopio, Finland

gDepartment of Old Age Psychiatry & Psychotic Disorders, Medical University of Lodz, Lodz, PolandhSection of Gerontology and Geriatrics, Department of Medicine, University of Perugia, Perugia, Italy

iMemory and Dementia Centre, Aristotle University of Thessaloniki, Thessaloniki, GreecejDepartment of Internal and Geriatrics Medicine, INSERM U 1027, Gerontopole, Hopitaux de Toulouse, Toulouse, France

kDepartment of Psychiatry, University of Oxford, Warneford Hospital, Oxford, UK

Abstract Introduction: The aim of this study was to (1) replicate previous associations between six blood lipidsand Alzheimer’s disease (AD) (Proitsi et al 2015) and (2) identify novel associations between lipids, clin-ical AD diagnosis, disease progression and brain atrophy (left/right hippocampus/entorhinal cortex).Methods: We performed untargeted lipidomic analysis on 148 AD and 152 elderly control plasmasamples and used univariate and multivariate analysis methods.Results: We replicated our previous lipids associations and reported novel associations betweenlipids molecules and all phenotypes. A combination of 24 molecules classified AD patients with.70% accuracy in a test and a validation data set, and we identified lipid signatures that predicteddisease progression (R2 5 0.10, test data set) and brain atrophy (R2 � 0.14, all test data sets exceptleft entorhinal cortex). We putatively identified a number of metabolic features including cholesterylesters/triglycerides and phosphatidylcholines.Discussion: Blood lipids are promising AD biomarkers that may lead to new treatment strategies.� 2016TheAuthors.PublishedbyElsevier Inc. onbehalf of theAlzheimer’sAssociation.This is anopenaccess article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

Keywords: Alzheimer’s disease; Dementia; Brain atrophy; sMRI; Rate of cognitive decline; Lipidomics; Metabolomics;

Biomarkers; Machine learning; Multivariate; Classification; Random forest

1. Background

Alzheimer’s disease (AD) is a devastating illness andone of the major public health challenges of the 21stcentury. The lack of effective treatments and early diag-nosis highlights the importance of the identification of

noninvasive biomarkers, for early diagnosis and diseaseprogression. Blood metabolites have recently emergedas promising AD biomarkers [1–4]. They are smallmolecules which could theoretically cross the alreadycompromised AD blood-brain barrier [5]; they are easilyaccessible, and they represent an essential aspect of thephenotype of an organism and a molecular “fingerprint”of disease progression [6,7]. They can therefore aid earlydiagnosis, recruitment into trials and may help identifynew therapeutic targets.

1These authors contributed equally to the manuscript.2These senior authors contributed equally to the manuscript.

*CorrespondingQ3 author. Tel.: ---; Fax: ---.

E-mail address: [email protected]


1552-5260/� 2016 The Authors. Published by Elsevier Inc. on behalf of the Alzheimer’s Association. This is an open access article under the CC BY-NC-ND

license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

FLA 5.4.0 DTD � JALZ2275_proof � 28 September 2016 � 2:14 pm � ce

Alzheimer’s & Dementia - (2016) 1-12

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

61

62

63

64

65

66

67

68

69

70

71

72

73

74

75

76

77

78

79

80

81

82

83

84

85

86

87

88

89

90

91

92

93

94

95

96

97

98

99

100

101

102

103

104

105

106

107

108

109

110

111

112

113

114

115

116

117

118

119

120

121

122

123

124

125

126

127

128

129

130

131

132

133

134

Delta:1_given name

Delta:1_surname

Delta:1_given name

Delta:1_surname

Delta:1_given name

Delta:1_surname

Delta:1_given name

Delta:1_surname

Delta:1_given name

Delta:1_surname

Delta:1_given name

Delta:1_surname

Delta:1_given name

Delta:1_surname

Delta:1_given name

Delta:1_surname

http://creativecommons.org/licenses/by-nc-nd/4.0/

mailto:[email protected]


http://creativecommons.org/licenses/by-nc-nd/4.0/


A number of blood metabolomic studies have highlightedthe role of lipid compounds, such as phosphatidylcholines(PCs) in AD [1–4]. We previously identified three PCs thatwere diminished in mild cognitive impairment (MCI)individuals and AD patients [4] and were further associatedwith poorer memory performance and decreased brain func-tion during aging [8]. We further performed lipidomics anal-ysis and identified 10 metabolites that predicted AD in anunseen test data set with 79% accuracy [9]; six analyteswere putatively identified as cholesteryl esters (ChoEs),molecules related to PCs, and were reduced in MCI and AD.

Here, we performed lipidomics analysis in a sample of142 AD patients and 135 healthy controls aiming to (1) repli-cate our previous associations [9] and (2) discover new lipidsand combinations of lipids associated with clinical AD diag-nosis and AD endophenotypes, such as the rate of cognitivedecline and brain atrophy measures. This is to our knowl-edge the most comprehensive blood lipidomics study todate to identify lipid signatures associated with AD andAD endophenotypes, improving our current knowledge ofmolecules associated with AD.

2. Methods

2.1. Patient sample collection

This study used 148 AD patients and 152 controls from theDementia Case Register at King’s College London and theEU-funded AddNeuroMed study [10]. All individuals withAD patients met criteria for either probable (NINCDS-ADRDA, DSM-IV) or definite (CERAD) AD. All nonpopu-lation individuals who were controls were screened fordementia using the MMSE or ADAS-cog or were determinedto be free from dementia at neuropathologic examination orhad a Braak score �2.5. Diagnosis was confirmed by patho-logic examination for a proportion of cases and cognitivelynormal elderly controls. All AD cases had an age of onset�60 years, and controls were�60 years at examination. A to-tal of 102 AD cases and 104 controls had HDL-c, LDL-c, TC,and TG serum levels (mmol/L) available. Nonoverlapping in-dividuals from these cohorts have been previously reported[9]. Each individual was required to fast for 2 hours beforesample collection, and 10 mL of blood was collected in tubescoated with sodium ethylenediaminetetraacetic acid to pre-vent clotting. Whole blood was centrifuged at 2000 g for10 minutes at 4�C to separate plasma, which was removedand stored at 280�C. All samples were centrifuged withinapproximately 2 hours of collection.

2.2. Lipidomics

Sample treatment has been described elsewhere [4,9,11]and is explained in detail in Supplementary Methods 1.Briefly, 20 mL of plasma was added to a glass HPLC vialcontaining a 400-ml glass insert (Chromacol, UK). Ten mi-croliters of high purity water and 40 mL of MS grade meth-anol were added to each sample, followed by a 2-minute

vortex mix to precipitate proteins; 200 mL of Methyl tert-Butyl Ether (MTBE) containing 10 mg/mL of internalstandard Tripentadecanoin (TG45:0) was added, and thesamples were mixed via vortex at room temperature for1 hour. After addition of 50 mL of high purity water, a finalsample mixing was performed before centrifugation at3000 g for 10 minutes. The upper, lipid-containing,MTBE phase was then injected onto the LC-MS systemdirectly from the vial by adjustment of the instrument nee-dle height (17.5 mm from bottom).

Lipidomics was performed by a Waters ACQUITYUPLC and XEVOQTOF system. The method has previouslybeen published [4,12] and has been shown to quantitate.4500 metabolite species (Supplementary Methods 1).Samples were analyzed in a randomized order, in fourbatches, with pooled plasma sampled (QC) at regular inter-vals throughout the run (n 5 30 for both positive and nega-tive ionization). Features were extracted from netCDFfiles using the R package “XCMS” [13] which performedfiltration, peak identification, matching of peaks across sam-ples, and retention time correction. Positive and negativeionization mode data were extracted separately and quantilenormalized.

2.3. Structural magnetic resonance imaging

Volumes of whole brain and the hippocampi and entorhi-nal cortices were obtained using FreeSurfer 5.1.0 from 123subjects (53 AD patients and 70 Controls) who had under-gone sMRI. Regions were normalized by intracranial vol-ume [14]. The volumetric data were not used to aid in theclinical diagnosis of AD. Detailed information regardingdata acquisition, pre-processing, and quality control assess-ment has been described elsewhere [15,16]. Before analyses,sMRI measures were standardized to have a mean of 0 and astandard deviation (SD) of 1.

2.4. Calculation of rate of cognitive decline

The ROD was available for 118 AD patients with analytedata and has been described elsewhere [17]. The ROD wasbased on longitudinal mini mental state examination(MMSE) assessments [18], and only samples with at leastthree MMSE measures were included in the calculation us-ing linear mixed effect models. After covariate adjustment[17], the slope coefficient for each sample was used as theROD defined as the change in MMSE per day.

2.5. Statistical analysis

2.5.1. Quality controlData QC has been previously described [9] and included

filtering of features and individuals, data transformation,batch effect correction, outlier detection, and imputation(Supplementary Methods 2.1 and SupplementaryFigure 1). All analyses took place in R.3.01.


P. Proitsi et al. / Alzheimer’s & Dementia - (2016) 1-122

135

136

137

138

139

140

141

142

143

144

145

146

147

148

149

150

151

152

153

154

155

156

157

158

159

160

161

162

163

164

165

166

167

168

169

170

171

172

173

174

175

176

177

178

179

180

181

182

183

184

185

186

187

188

189

190

191

192

193

194

195

196

197

198

199

200

201

202

203

204

205

206

207

208

209

210

211

212

213

214

215

216

217

218

219

220

221

222

223

224

225

226

227

228

229

230

231

232

233

234

235

236

237

238

239

240

241

242

243

244

245

246

247

248

249

250

251

252

253

254

255

256

257

258

259

260

261

262

263

264

265

266

267

268

2.5.2. Single-analyte statistical analysisLogistic regression investigated the association of each

metabolite with clinical AD diagnosis and linear regressionthe association with cognitive decline and sMRI measures.Logistic regression and linear regression models for thesMRI measures were adjusted for age at sampling, gender,presence of the apolipoprotein E (APOEQ4 ) ε4 allele, batch,and study site. For the ROD models, covariate adjustmentwas only applied for batch as the rest of the covariateswere included in the ROD calculation [17]. sMRI measure-ments were not adjusted for diagnosis to allow identificationof features associated with brain atrophy caused by AD.False discovery rate (FDR) correction (0.05) was appliedto correct for multiple testing (“fdrtool”). Secondary modelsinvestigated whether any associations were modified by theAPOE 4 allele or by gender.

Logistic regression results (summary statistics) for thepositive ionization metabolites were combined with the re-sults (summary statistics) from the Proitsi et al data set [9]using inverse variance weighted fixed effect meta-analysis(“metafor”). The published data set [9] was restricted to576 features extracted using Mass-Lynx, and therefore, theanalysis presented here includes a large number of previ-ously unreported molecules extracted using XCMS. All as-sociations are reported as the change per one metabolitestandard deviation (SD).

2.5.3. Multivariate statistical analysisA random forest (RF) classifier approach (using “rf” and

“rfe” in “CARET”) was used to develop a clinical diagnosisclassifier as previously described [9] (SupplementaryMethods 2.2). Briefly, AD cases and controls were divided

into a training data set (2/3 of the sample) matched for age,gender, and site and an independent data set (rest 1/3 of thesample). An RF model was built on the training data set(100 bootstraps), and in each iteration, each variable was as-signed a variable importance (VI) score. The summedVI ranksprovided an indication of the predictive power for each vari-able, and the top 10% molecules were selected for RF withrecursive feature elimination (rfe; 100 bootstraps) from 250down to two features. For each subset of predictors, themean bootstrap testing performance was calculated, and theoptimal number of variables was identified using “sizeToler-ance” that picks a subset of variables that is small withoutsacrificing too much performance. Subsets of variables within2.5% and 5% of the optimum performancewere examined andused to build final models in the complete training data, whichwere tested on the test set. The final model was also tested inthe Proitsi et al data set [9] which was used as a validation dataset, after excluding metabolites in the negative ionizationmode. The area under curve (AUC) was used to test the perfor-mance of each classifier. Receiver operator curves (ROCs)were plotted using “ROCR”. Models including APOE ε4and the six features in Proitsi et al [9] were also tested.

Random forest regression (RFR) models were built forcognitive decline and sMRImeasures following the same strat-egy as for clinical diagnosis. The data set was split randomlyinto a training (2/3 of the data) and test set (1/3 of the data)for each endophenotype such that the training and test datasets were stratified for each endophenotype and containedequal representation of each site. Age, gender, and APOE ε4presence were included in the model development for thesMRI models, and the root mean squared error (RMSE) wasused to evaluate the performance of the models.

Table 1

Sample demographics

AD (N 5 142) Controls (N 5 135)

Difference between AD patients and

controls*

Age, mean (SD) 77 (6.5) 74 (5.9) t 5 24.8 (270), P value 5 2.62 ! 1026

Gender (males/females) 48/87 47/95 c2 5 0.09 (1), P value 5 .761

APOE ε4 allele (absence/presence) 54/81 99/43 c2 5 23.53 (1), P value 5 1.23 ! 1026

MMSE, mean (SD; Range) 20.1 (4.6; 10–27) 29.2 (0.9; 27–30) t 5 22.58 (143), P value , 2.0 ! 10216

ROD (per year), mean (SD)y 21.46 (1.26) NA NA

Entorhinal cortex right, mean (SD)zx 0.00092 (0.0003) 0.0013 (0.0003) t 5 6.02 (99), P value 5 2.87 ! 1028

Entorhinal cortex left, mean (SD)zx 0.00094 (0.0002) 0.0013 (0.0004) t 5 5.26 (86), P value 5 1.58 ! 1026

Hippocampus right, mean (SD)zx 0.0019 (0.0004) 0.0025 (0.0003) t 5 9.06 (100), P value 5 1.3 ! 10214

Hippocampus left, mean (SD)zx 0.0018 (0.0004) 0.0025 (0.003) t 5 10.57 (104), P value , 2.0 ! 10216

Mean HDL-c (SD), mmol/Lk 1.58 (0.37) 1.55 (0.38) b 5 0.109 (SE 5 0.33), P value 5 .068

Mean LDL-c (SD), mmol/Lk 3.42 (1.01) 3.07 (0.82) b 5 0.092 (SE 5 0.15), P value 5 .529

Mean TC (SD), mmol/Lk 5.69 (1.17) 5.29 (1.01) b 5 0.209 (SE 5 0.173), P value 5 .229

Mean TG (SD), mmol/Lk 1.64 (1.04) 1.52 (0.67) b 5 0.021 (SE 5 0.146), P value 5 .885

Statins (yes/no) 38/97 34/108 c2 5 0.436 (1), P value 5 .509

Abbreviations: AD, Alzheimer’s disease; MMSE, mini-mental state examination score; ROD, rate of cognitive decline; SD, standard deviation.

*Differences in the means/frequencies of clinical/demographic variables were tested using t test t(df), x2(df) test, or linear regression analyses after adjusting

for age, gender, the APOE ε4 allele, and study site.yRate of decline data was available for a subset of AD patients (N 5 118).zsMRI data were available for a subset of study participants (N 5 123 [AD 5 53, controls 5 70]).xNormalized to intracranial volume.kSerum HDL cholesterol, LDL cholesterol, total cholesterol, and triglyceride levels were available for a subset of study participants (N 5 208 [AD 5 102,

Controls 5 106]).


P. Proitsi et al. / Alzheimer’s & Dementia - (2016) 1-12 3

269

270

271

272

273

274

275

276

277

278

279

280

281

282

283

284

285

286

287

288

289

290

291

292

293

294

295

296

297

298

299

300

301

302

303

304

305

306

307

308

309

310

311

312

313

314

315

316

317

318

319

320

321

322

323

324

325

326

327

328

329

330

331

332

333

334

335

336

337

338

339

340

341

342

343

344

345

346

347

348

349

350

351

352

353

354

355

356

357

358

359

360

361

362

363

364

365

366

367

368

369

370

371

372

373

374

375

376

377

378

379

380

381

382

383

384

385

386

387

388

389

390

391

392

393

394

395

396

397

398

399

400

401

402

web4C=FPO

web4C=FPO

Fig. 1. Associations of previously reported molecules (Proitsi et al 2015) with clinical AD diagnosis in the current data set and associations of putatively annotated

molecules, selected through random forest analyses, with the respective phenotype. (A) Association of Mass 856 with clinical AD diagnosis; (B) Association of

Mass 866 with clinical AD diagnosis; (C) Association of Mass 868 with clinical AD diagnosis; (D) Association of Mass 882 with clinical AD diagnosis; (E) As-

sociation ofMass 894with clinical AD diagnosis; (F) Association ofMass 970with clinical AD diagnosis; (G) Association ofMass 882 (2) (PC 40:4) with clinical

AD diagnosis; (H) Association of Mass 948 (1) TG (57:1) with clinical AD diagnosis; (I) Association of Mass 919 (1) TG 50:2 with Hippocampus (Right); (J)

Association ofMass 943 (1) (ChoE/TG) withHippocampus Left; (K) Association ofMass 367 (sterol) with Entorhinal Cortex (Right); (L) Association ofMass 816

(1) with Entorhinal Cortex (Left); (M) Association ofMass 771 (1) PC 36:3 with the rate of cognitive decline (ROD). The P values displayed are for the univariate

regressions after adjusting for covariates. All molecules are scaled to have a mean of 0 and a standard deviation of 1.



403

404

405

406

407

408

409

410

411

412

413

414

415

416

417

418

419

420

421

422

423

424

425

426

427

428

429

430

431

432

433

434

435

436

437

438

439

440

441

442

443

444

445

446

447

448

449

450

451

452

453

454

455

456

457

458

459

460

461

462

463

464

465

466

467

468

469

470

471

472

473

474

475

476

477

478

479

480

481

482

483

484

485

486

487

488

489

490

491

492

493

494

495

496

497

498

499

500

501

502

503

504

505

506

507

508

509

510

511

512

513

514

515

516

517

518

519

520

521

522

523

524

525

526

527

528

529

530

531

532

533

534

535

536

3. Results

A total of 2539 positive ionization and 358 negative ioni-zation features were initially extracted from 300 individuals.

After QC, 2216 positive and 289 negative ionization featuresfrom 277 individuals (142 AD cases and 135 controls) wereused in subsequent analyses. Of these, 53 AD patients and 70controls had sMRI data available, and 118 AD patients had

web4C=FPO

web4C=FPO

Fig. 1. (Continued)



537

538

539

540

541

542

543

544

545

546

547

548

549

550

551

552

553

554

555

556

557

558

559

560

561

562

563

564

565

566

567

568

569

570

571

572

573

574

575

576

577

578

579

580

581

582

583

584

585

586

587

588

589

590

591

592

593

594

595

596

597

598

599

600

601

602

603

604

605

606

607

608

609

610

611

612

613

614

615

616

617

618

619

620

621

622

623

624

625

626

627

628

629

630

631

632

633

634

635

636

637

638

639

640

641

642

643

644

645

646

647

648

649

650

651

652

653

654

655

656

657

658

659

660

661

662

663

664

665

666

667

668

669

670

ROD data available. Sample demographics are displayed inTable 1.

3.1. Univariate analyses results

Logistic regression analyses were initially used to inves-tigate the association of each lipid with AD. We then per-formed fixed-effects meta-analyses between the results ofthis data set and our previously published data set [9], usingfixed-effects meta-analyses. Briefly, 425 features wereassociated with AD at P value ,.05 in this data set; ofthese, 87 features passed correction for multiple testing atQ value ,0.05. After meta-analysis, 377 features wereassociated with AD at P value ,.05 and 125 at Q val-ue,0.05. All six features from Proitsi et al [9] were associ-ated with AD at Q value ,0.05 in meta-analysis (Fig. 1(A–F) and Table 2).

Linear regression investigated the association of eachlipid with brain atrophy and the rate of cognitive decline.A total of 266 features were associated with the ROD at Pvalue ,.05, but none passed multiple testing correction. Atotal of 181 features were associated with right hippocampusvolume and 224 were associated with left hippocampus vol-ume; only six features were associated with left hippocam-pus at Q value ,0.05. Finally, 156 and 124 features wereassociated with EC volume (left and right, respectively) atP value,.05, but no associations passed correction for mul-tiple testing. Results for all logistic and linear regression an-alyses are provided in Supplementary Table 1.

Overall, most lipids were reduced in AD compared tocontrols (54 of the 87 features associated at Q-value,0.05 were reduced in AD). Additionally, we observed sub-stantial overlap between features associated with clinicalAD diagnosis and brain atrophy (Supplementary Figure 2).

We further investigated whether the APOE ε4 and gendermodified the associations between lipids and clinical ADdiagnosis. APOE ε4 modified the association of 231 featureswith AD, and gender modified the association of 191 fea-tures with AD (P value ,.05); none of these associationswas significant at Q-value , 0.05. There were only 3 indi-viduals with the ε2/ε4 genotype, and there were thereforeno differences in lipids levels between ε4 and non-ε4 carriersafter excluding ε2/ε4 individuals.

3.2. Multivariate analysis results

A RF approach was used to identify a panel of moleculesassociated with clinical AD diagnosis. After an initial RFpre-selection step on the training data set, the top 10%lipids (250 features), in terms of their variable importance,were selected (after 100 Bootstraps). Furthermore, randomforest with recursive feature elimination (RF-rfe) on thetraining data set showed that the best training performancewas for a model with 240 features. To choose a model withhigh accuracy while reducing the number of features as lowas possible, a 5% tolerance RF-rfe model (25 features) was

fitted on the whole training data set (AUC, 0.87) and clas-sified the test data set with 73% accuracy (SupplementaryFigure 3 and Table 3). The model was then fitted on thetraining data set, excluding one negative ionization modeanalyte and classified the test training data set with 74% ac-curacy and the Proitsi et al [9] validation data set with 71%accuracy. There was no increase in accuracy when covari-ates and the features from Proitsi et al [9] were added tothe models (Table 3).

Random forest regressions using the same pipeline wereapplied to the ROD and brain atrophy measures. AfterRFR-rfe on the training data set, the lowest mean RMSEfor RODwas for a model with 40 features. The 5% tolerancemodel of the lowest RMSE model (10 features) was fitted tothe whole training data set (R25 0.49) and predicted the testdata set with R2 5 0.10 (Table 4).

For right hippocampus, the lowest RMSE was with amodel with 70 features that included age. A 5% tolerancemodel (12 features) was fitted to the training data set(R2 5 0.55) and predicted the test data set with R2 5 0.15(Table 3). For left hippocampus, the lowest training RMSEwas with 100 features that also included age. The 5% toler-ance model (12 features) was fitted to the training data set(R2 5 0.59) and predicted the test data set with R2 5 0.15(Table 3). The performance of the models was almost iden-tical when age was excluded.

For the right EC, the lowest mean training RMSE waswith 70 features; a 5% tolerance model (12 features) wasfitted to the training data set (R2 5 0.54) and predicted thetest data set with R2 5 0.14. Finally, for left EC, a modelwith 90 features had the lowest RMSE, and a 5% toler-ance model (12 features) was fitted to the training dataset (R2 5 0.42) and predicted the test data set withR2 5 0.01 (Table 3). Results of all 2.5% models are pre-sented in Supplementary Tables 2 and 3, and the list ofmolecules included in each classifier is found inSupplementary Table 1. The strength of association be-tween selected features and each model is shown inFig. 2, and the scaled VI of each lipid after RF-RFE/RFR-RFE for each phenotype is shown inSupplementary Figure 4.

3.3. Lipid annotation and putative identification

We opted to annotate the top features, in terms of VI fromeach model and features selected in more than one model,using our in-house lipid database and MS/MS fragmentationpatterns [4,11,12]. These features were annotated as mainlyLCTs and ChoEs, some were PCs and a sterol. Fig. 1(G–M) and Table 2 present the univariate associations of thesemolecules with the respective phenotypes. The association ofthe annotated molecules with all phenotypes is shown inSupplementary Figure 5. The raw intensity counts for eachAD associated lipid across AD and controls, along with thecoefficients of variation (relative standard deviation [RSD])



671

672

673

674

675

676

677

678

679

680

681

682

683

684

685

686

687

688

689

690

691

692

693

694

695

696

697

698

699

700

701

702

703

704

705

706

707

708

709

710

711

712

713

714

715

716

717

718

719

720

721

722

723

724

725

726

727

728

729

730

731

732

733

734

735

736

737

738

739

740

741

742

743

744

745

746

747

748

749

750

751

752

753

754

755

756

757

758

759

760

761

762

763

764

765

766

767

768

769

770

771

772

773

774

775

776

777

778

779

780

781

782

783

784

785

786

787

788

789

790

791

792

793

794

795

796

797

798

799

800

801

802

803

804

of the pooled samples (QCs) are shown in SupplementaryTable 4.

4. Discussion

This is to our knowledge, the largest nontargeted bloodlipidomics study in AD to date. Here, we expanded our

recent work [9], and using univariate and multivariate ap-proaches, we replicated the associations between six previ-ously reported blood lipids and AD [9] and reported theirassociation with brain atrophy. We further identified combi-nations of lipids that classified AD patients with relativelygood accuracy when tested in both a test and a validationdata set (.70%), and combinations of molecules that

Table 2

List of putatively identified metabolite molecules selected by the six random forest models

Logistic

regression

analysis

m/z

(ionization

mode)

Putative

metabolite

molecule

Present study data set Proitsi et al 2015 data set Meta-analysis

OR 95% CI P value OR 95% CI P value OR 95% CI P value

Clinical AD

diagnosis

882 (2) PC 40:4 1.996 1.35–3.01 6.79E204* NA NA NA NA NA NA

948 (1) TG 57:1 0.514 0.36–0.71 8.09E205* 0.522 0.28–0.92 3.01E202 0.516 0.39–0.69 6.83E206*

856 (1) ChoE/TGy 0.711 0.51–0.99 4.34E202 0.141 0.04–0.43 1.75E203 0.632 0.46–0.87 4.94E203*

866 (1) ChoE/TGy 0.663 0.46–0.94 2.42E202 0.251 0.10–0.52 7.51E204* 0.569 0.41–0.79 7.37E204*

868 (1) ChoE/TGy 0.65 0.47–0.89 7.74E203* 0.218 0.07–0.55 3.15E203 0.591 0.44–0.80 7.16E204*

882 (1) ChoE/TGy 0.57 0.41–0.79 7.99E204* 0.231 0.08–0.53 1.56E203 0.517 0.38–0.71 3.05E205*

894 (1) ChoE/TGy 0.732 0.51–1.04 8.05E202 0.151 0.05–0.38 3.14E204* 0.615 0.44–0.86 4.65E203*

970 (1) ChoE/TGy 0.643 0.47–0.87 4.96E203* 0.362 0.18–0.67 2.56E203 0.58 0.44–0.77 1.27E204*

Linear regression

analysis

Beta 95% CI P value NA NA NA NA NA NA

Hippocampus (right) 919 (1) TG 50:2 0.396 0.20–0.59 7.91E205* NA NA NA NA NA NA

Hippocampus (Left) 943 (1) ChoE/TG 0.320 0.13–0.51 9.97E204* NA NA NA NA NA NA

Entorhinal Cortex

(Right)

367 (1) Sterol 20.201 20.39 to 20.01 3.88E202 NA NA NA NA NA NA

Entorhinal Cortex

(Left)

816 (1) NA 0.218 0.03–0.41 2.47E202 NA NA NA NA NA NA

ROD 771 (1) PC 36:3z 20.412 20.65 to 20.17 9.92E204 NA NA NA NA NA NA

Abbreviations: AD, Alzheimer’s disease; ChoE, cholesteryl ester; CI, confidence interval; m/z, mass-to-charge ratio; OR, odds ratio; PC, phosphatidylcho-

line; ROD, rate of cognitive decline; TG, Triglyceride.

NOTE. The six random forest models were for the clinical AD diagnosis, ROD, hippocampus (R/L), and entorhinal cortex (R/L) phenotypes. The association

of eachmolecule is presented with the respective phenotype (i.e., primary phenotype of association). The association of the six molecules previously reported by

Proitsi et al 2015 with AD is also presented.

*Q value ,0.05.yFeatures identified by Proitsi et al 2015 and for the Proitsi et al., data set semiquantified values are presented.zPC 36:3 has m/z 770, and m/z 771 is its C13 isotope. ChoE/TG indicates co-elution of ChoE and TG molecules.

Table 3

Random forest classifier model results (clinical AD diagnosis) for the training data set and predictions on the test data set and the Proitsi et al data set

Model (5% tolerance)

Training data set (N 5 179) Test data set (N 5 98) Validation data set (N 5 75)

Sens. Spec. AUC Acc. Sens. Spec. AUC PPV NPV Acc. Sens. Spec. AUC PPV NPV

Covariates only* 0.72 0.71 0.77 0.56 0.54 0.58 0.56 0.5 0.62 0.6 0.57 0.63 0.6 0.57 0.63

25 featuresy 0.82 0.82 0.87 0.73 0.77 0.69 0.73 0.66 0.79 NA NA NA NA NA NA

24 featuresz 0.81 0.82 0.86 0.74 0.74 0.73 0.74 0.68 0.78 0.71 0.69 0.73 0.71 0.69 0.73

25 featuresy 1 covariates* 0.83 0.83 0.88 0.75 0.79 0.71 0.75 0.68 0.81 NA NA NA NA NA NA

24 featuresz 1 covariates* 0.84 0.82 0.88 0.75 0.79 0.71 0.75 0.68 0.81 0.71 0.69 0.73 0.71 0.69 0.73

25 featuresy 1 6 ChoE/TGx 0.82 0.82 0.87 0.71 0.74 0.67 0.71 0.64 0.77 NA NA NA NA NA NA

24 featuresz 1 6 ChoE/TGx 0.82 0.81 0.87 0.71 0.79 0.66 0.72 0.65 0.80 0.72 0.71 0.73 0.72 0.69 0.74

25 featuresz 1 covariates*

1 6 ChoE/TGx0.83 0.83 0.88 0.74 0.77 0.71 0.74 0.68 0.80 NA NA NA NA NA NA

24 featuresz 1 covariates*

1 6 ChoE/TGx0.83 0.82 0.88 0.74 0.79 0.69 0.74 0.67 0.81 0.71 0.69 0.73 0.71 0.69 0.73

Abbreviations: Acc, accuracy; AUC, area under the curve; ChoE, cholesteryl ester; NPV, negative predictive value; PPV, positive predictive value; Sens,

sensitivity; Spec, specificity; TG, triglyceride.

*Age, sex, ε4.y5% tolerance model including negative ionization molecule.z5% tolerance model excluding negative ionization molecule.xSix features identified by Proitsi et al 2015.



805

806

807

808

809

810

811

812

813

814

815

816

817

818

819

820

821

822

823

824

825

826

827

828

829

830

831

832

833

834

835

836

837

838

839

840

841

842

843

844

845

846

847

848

849

850

851

852

853

854

855

856

857

858

859

860

861

862

863

864

865

866

867

868

869

870

871

872

873

874

875

876

877

878

879

880

881

882

883

884

885

886

887

888

889

890

891

892

893

894

895

896

897

898

899

900

901

902

903

904

905

906

907

908

909

910

911

912

913

914

915

916

917

918

919

920

921

922

923

924

925

926

927

928

929

930

931

932

933

934

935

936

937

938

predicted changes in disease progression (R2 5 0.10 for testdata set) and brain atrophy (R2 � 0.14 for all test data setsexcept for left EC). Although these signatures cannot beused for diagnostic purposes yet, they suggest important bio-logical mechanisms associated with AD.

4.1. Identification and role of lipids in AD

We putatively identified two PC molecules; additionally,ChoEs and triglycerides (TGs) were tentatively annotateddue to chromatographic coelution, and finally, we putativelyannotated a molecule as a sterol. The higher MS-MS sensi-tivity achieved here enabled the detection of a number ofadditional lipids that co-eluted with ChoE; these were anno-tated as TGs (Table 2).

The association of PCs with AD and cognition has beenextensively described [4,8]. Here, one of the moleculesmost strongly associated with AD is a putative PC (PC40:4), and the top lipid in the ROD model is also a putativePC (PC 36:3). In contrast to the same species of molecules,we have previously identified, both PCs are increased inAD, and PC 36:3 is associated with faster ROD. Althoughmost studies to date have reported a reduction of PC levelsin AD, an increase in CSF PCs has been observed in ADcompared to control brains [19] and recently in “AD-like” pa-tients based on their CSFAmyloid-beta42, Tau, and Phospho-Tau-181 levels [20]. A recent study also reported a parallelincrease of PCs containing saturated and short-chain fattyacids in serum from AD patients [21]. These suggest deregu-lation in the biosynthesis, turnover, and acyl chain remodelingof phospholipids, in accordance with increased phospholipidbreakdown due to PLA2 [21] overactivation.

We have also reported associations with low-chain andvery-low-chain triglycerides (LCTs/VLCTs; fatty acid chainlength .16 carbons). One of the most interesting findingswas that due to the higher MS-MS sensitivity achieved inthis study, we were able to observe putative VLCTs thatwere coeluting with ChoEs (Table 2). We have previously re-ported on the synthesis of ChoEs [9]; briefly, it takes place bytransfer of fatty acids from PC to cholesterol, a reaction cata-lyzed by lecithin cholesterol acyl transferase in plasma and byacyl-coenzyme A: cholesterol acyl transferase 1 and 2(ACAT1 and ACAT2) in other tissues, including the brain.The association of LCTs/VLCTs with AD is noteworthy.Although overall TGs are seen as risk factors for many disor-ders including cardiovascular disease (CVD) and type 2 dia-betes (T2D), numerous investigations point to the diverse roleof TGs with different chain lengths. It is known for examplethat medium-chain triglycerides (MCTs) and LCTs havedifferent metabolic pathways in digestion and absorption[22]. Moreover, although LCTs of lower carbon numberand double bond content have been associated with increasedCVD [23] and T2D risk [24], LCTs with higher carbon num-ber and double bond content, like the ones here, have beenassociated with decreased risk of T2D [24], whereas no asso-ciations between T2D and total triglyceride levels wereobserved in the same individuals [24]. Furthermore,decreased concentration of LCTs and an increased concentra-tion of VLCTs have been associated with longevity [25].These findings are particularly interesting as most vegetableoils are comprised of long-chain fatty acids; however, onlyMTCs have to our knowledge been implicated in AD,although findings are controversial [26]. When we previouslyinvestigated the association of total cholesterol and TGs withAD in overlapping individuals using Mendelian randomiza-tion, we found no evidence for an association with AD [27].Additionally, we observed no difference in serum triglycer-ides, total cholesterol, LDL cholesterol, and HDL cholesterolbetween AD patients and controls for a random subset ofstudy participants in this study (AD 5 102, controls 5 106)that had serum lipid measures available for the same visit,as well as no difference in the frequency of AD patients andcontrols who were taking statins. On the other hand, a recentstudy reported an overlap between genes involved in elevatedplasma lipid levels and inflammation and the risk for AD [28].All these highlight the relevance of investigating smaller lipidfractions as they highlight specific steps in their biosynthesisand metabolism that may be associated with AD.

Finally, we observed an association between most pheno-types and a feature of m/z 367. We previously described amolecule with the same mass and similar retention time tobe reduced in AD [9]. The molecule discovered here wasassociated with an increased risk for AD in both data setsand with reduced brain volume and was included in theERC and clinical diagnosis models. We believe this featureis a fragment and a sterol, specifically an isomer of desmos-terol. Desmosterol is a precursor of cholesterol and seladin(DHCR24), which governs the metabolism of desmosterol

Table 4

Random forest regression model results for the training data set and

predictions on the test data set for each AD endophenotype

Phenotype

Model

(5% tolerance)

Train data

set (n 5 93)

Test data

set (n 5 28)

RMSE R2 RMSE R2

Hippocampus

(right)

Covariates only* 0.92 0.28 1.09 0.02

12 featuresy 0.58 0.55 0.9 0.15

Hippocampus

(Left)

Covariates only* 0.89 0.34 1.21 0.01

12 featuresy 0.64 0.59 0.99 0.15

Entorhinal

cortex (right)

Covariates only* 0.95 0.22 1.14 ,0.01

12 features 0.66 0.54 0.92 0.14

Entorhinal

cortex (left)

Covariates only* 1.00 0.22 1.17 ,0.01

12 features 0.77 0.42 1.07 0.01

ROD 10 featuresz 0.93 0.49 1.09 0.10

Abbreviations: RMSE, root mean squared error; ROD, rate of cognitive

decline.

*Age, sex, ε4.yAge was included in the final model. There was no difference in either

train or test data set performance when age was excluded.zCovariates were already included in the calculation for the ROD.



939

940

941

942

943

944

945

946

947

948

949

950

951

952

953

954

955

956

957

958

959

960

961

962

963

964

965

966

967

968

969

970

971

972

973

974

975

976

977

978

979

980

981

982

983

984

985

986

987

988

989

990

991

992

993

994

995

996

997

998

999

1000

1001

1002

1003

1004

1005

1006

1007

1008

1009

1010

1011

1012

1013

1014

1015

1016

1017

1018

1019

1020

1021

1022

1023

1024

1025

1026

1027

1028

1029

1030

1031

1032

1033

1034

1035

1036

1037

1038

1039

1040

1041

1042

1043

1044

1045

1046

1047

1048

1049

1050

1051

1052

1053

1054

1055

1056

1057

1058

1059

1060

1061

1062

1063

1064

1065

1066

1067

1068

1069

1070

1071

1072

to cholesterol in specific brain areas. Desmosterol has beenshown to inhibit b-secretase cleavage of APP, and the forma-tion of amyloid-b and lower desmosterol levels has beenfound in the plasma and brains of AD patients comparedto controls [29–32].

Although the association of aberrant lipid metabolism inAD pathogenesis is undisputed [33–35]; at this stage, themechanisms by which these changes in lipids might occur inAD are unclear. One possibility involves AD selectivealterations to circulating lipid metabolism. However, another

**************************************

*

*

***

*

**

**

*********************************************************************

**************************

*****

*

*********

**

*********************************************************************************************

****************

**

*************

********

**********

**

*

***

****

*

********

***

*

*

***

*********

*********************

********

*

*

****

****

*

***************************************************

*************

***

***

****

*

*

*

*

**

*****

*

**

*

**

*

**

*

********

**

***

*******

****

*

**

972.8970.8947.8948.8949.8942.8943.8966.8967.8682.5919.8934.8897.7923.8840.7866.7856.7868.7894.8882.8752.5779.6570.4542.3543.3702.6818.5828.6576.5577.5603.5895.8918.9937.8810.8524.5552.5991.9XN778.56XN856.62XN882.64342.3382.3267.3341.3954.8996.3550.6578.7807.6687.6941.7943.7718.7719.7785.6786.6680.6700.5771.3771.6XN787.58803.7816.7891.7720.6766.2367.3289371.4629.6643.6XN659.52XN818.66774.6876.7892.7765.5934.9861.7888.7 Classifier Phenotype

aaaaaaaaaaaa

AD-CTL

AD-CTL (Proitsi et al)

AD-CTL & ERC_R

AD-CTL & HIP_R

ERC_L

ERC_L & ERC_R

ERC_R

HIP_L

HIP_L & HIP_R

HIP_R

HIP_R & ERC_R

ROD

-1

0

1

beta

AD-C

TLAD

-CTL

(Pro

itsi e

t al)

AD-C

TL M

eta-

analy

sis

ROD

HIP_

L

HIP_

R

ERC_

L

ERC_

R

web4C=FPO

web4C=FPO

Fig. 2. Heatmap of the univariate associations between features selected during random forest analyses for each phenotype. The color of each box represents the

univariate logistic regression beta coefficient (log[OR]) for clinical diagnosis or the univariate scaled linear regression beta coefficients for the rate of cognitive

decline and brain atrophy, after adjusting for covariates. The stars on each box represent the strength of the association: *P value,.05; **P value,.01; ***P

value,.001; ****P value,.0001; *****P value,.00001. The order of the metabolite molecules on the y-axis is based on a hierarchical clustering using the

metabolites pairwise correlations. XN denotes negative ionization mode feature. Abbreviations: AD-CTL, Clinical AD diagnosis; ROD, rate of cognitive

decline; HIP_L, left hippocampus; HIP_R, right hippocampus; ERC_L, left entorhinal cortex; ERC_R, right entorhinal cortex.



1073

1074

1075

1076

1077

1078

1079

1080

1081

1082

1083

1084

1085

1086

1087

1088

1089

1090

1091

1092

1093

1094

1095

1096

1097

1098

1099

1100

1101

1102

1103

1104

1105

1106

1107

1108

1109

1110

1111

1112

1113

1114

1115

1116

1117

1118

1119

1120

1121

1122

1123

1124

1125

1126

1127

1128

1129

1130

1131

1132

1133

1134

1135

1136

1137

1138

1139

1140

1141

1142

1143

1144

1145

1146

1147

1148

1149

1150

1151

1152

1153

1154

1155

1156

1157

1158

1159

1160

1161

1162

1163

1164

1165

1166

1167

1168

1169

1170

1171

1172

1173

1174

1175

1176

1177

1178

1179

1180

1181

1182

1183

1184

1185

1186

1187

1188

1189

1190

1191

1192

1193

1194

1195

1196

1197

1198

1199

1200

1201

1202

1203

1204

1205

1206

possibility relates to cellular lipid production. A number ofphospholipids are synthesized within a specialized region ofthe endoplasmic reticulum (ER) that is closely associatedwith mitochondria, the mitochondria-associated ER mem-branes (MAM). The close association of MAM to mitochon-dria facilitates Ca21 and phospholipid exchange between thetwo organelles [36–38]. Recent studies have shown thatMAM contacts are damaged in AD [39–42]. Because ER-mitochondria contacts are required for the synthesis of certainlipids [36–38], such changes may affect lipid metabolism andlead to some of the changes described here. Indeed, differentAPOE alleles have been shown to influence MAM [43].

4.2. Strengths and limitations

Here, we have used a large well-characterized AD cohortand a careful and systematic analysis pipeline. Throughbootstrapping, we have reduced over-fitting, and subse-quently, we validated our results in an unseen data set foreach phenotype and an additional validation data set for clin-ical AD diagnosis. Our AD diagnostic classifier achieved88% accuracy on the training data set (summary of 100 boot-straps) and predicted the test and validation data sets with.70% accuracy. Our training data set comprised of individ-uals matched on age and gender. On the other hand, the testdata set consisted primarily by females, and AD patientswere significantly older than controls. Additionally, the vali-dation data set [9] included AD patients and controls of UKorigin only, older than the individuals of the current data set.These findings highlight robustness in the model. For the ADendophenotypes, the Random Forest Regression models hada very good performance on all training data sets. Althoughthe performance dropped significantly in the test data sets,we observed R2 . 0.10 for all phenotypes except for LeftEC (R250.01). The drop in performance can be attributedto over-fitting of the training data sets and the smaller num-ber of individuals with ROD/brain atrophy measures. Thepoor performance of the Left EC is in agreement with ourunivariate analyses that highlighted weaker associationswith the EC for the whole sample; however, it is in contrastto the overall right-to-left asymmetry in AD [17].

A limitation of this study is that we were not able to deci-pher the exact fatty acid chain structure of some features.Owing to the higher MS-MS sensitivity, we observed a num-ber of putative ChoEs and TGs co-eluting, which iscommonly observed in lipidomics studies due to hundredsof lipids detected in one analysis; to minimize co-elutionproblems, our chromatographic run is 2 hours long using ul-tra pressure chromatography [11,44].

Additionally, although this is the largest AD lipidomicsstudy to date, we acknowledge that the sample size is stillmodest and further replication is required, especially forthe ROD and brain atrophy phenotypes. Moreover, althoughwe had information on the ROD, this calculation was basedon the MMSE, which is a crude measure of measurement ofcognition. Furthermore, the present study did not contain an

MCI cohort or information on conversion to MCI/AD, andtherefore, we do not know whether these features are associ-ated with initiation of AD. This study additionally suffersfrom limitations inherent to AD case-control studies, suchas the large number of comorbidities in old age, the possibil-ity that some of the elderly controls may already carry pa-thology, and that some of the clinically diagnosed AD maybe pathologically non-AD dementias. Finally, this studylacks information on BMI and body fat distribution thatcould potentially explain some of the differences betweenAD patients and controls.

However, through the longitudinal nature of these co-horts, we know that all the AD patients used for our analysismaintained the diagnosis of AD as did all controls for at least3 years from their baseline visit. Additionally, our informa-tion on disease progression and brain atrophy provide uswith more precise phenotypes that capture different stagesof disease pathology including the early preclinical stages.Given the good performance of these models, we believethat enrichment with additional individuals and pathologyinformation would increase their performance. Finally,although we did not have BMI information for our cohort,we observed no difference in statin use or serum lipids be-tween AD cases and controls.

5. Conclusion

In conclusion, the findings of this study deepen our knowl-edge of AD disease mechanisms and emphasize the impor-tance of investigating in detail different lipid fractions indementia research. As it is not known whether the observedchanges in lipid levels are causally related to or are just amarker of changes in lipoprotein dynamics and composition,studies that address causality are essential, as the success oftargeting specificmolecules and identifying potentially causalpathways amenable to intervention is predicated on thesemolecules being on the causal pathway. Finally, integratingadditional types of biological modalities such as protein,gene expression, and genotype information may increasethe fit of these models and help us to understand more aboutthe biological context in which these molecules operate.

Acknowledgments

The authors thank the individuals and families who took partin this research. The authors thank Professor Chris Morrisfor his input on Mitochondrial ER Membranes (MAM).We would like to acknowledge the use of the computationalLinux cluster and the Biomedical Research Centre NucleusInformatics Team supported by National Institute for HealthResearch (NIHR) Mental Health Biomedical ResearchCentre and Dementia Unit at South London and MaudsleyNHS Foundation Trust and (Institute of Psychiatry) King’sCollege London.This work was supported by the National Institute for HealthResearch (NIHR)Mental Health Biomedical ResearchCentre



1207

1208

1209

1210

1211

1212

1213

1214

1215

1216

1217

1218

1219

1220

1221

1222

1223

1224

1225

1226

1227

1228

1229

1230

1231

1232

1233

1234

1235

1236

1237

1238

1239

1240

1241

1242

1243

1244

1245

1246

1247

1248

1249

1250

1251

1252

1253

1254

1255

1256

1257

1258

1259

1260

1261

1262

1263

1264

1265

1266

1267

1268

1269

1270

1271

1272

1273

1274

1275

1276

1277

1278

1279

1280

1281

1282

1283

1284

1285

1286

1287

1288

1289

1290

1291

1292

1293

1294

1295

1296

1297

1298

1299

1300

1301

1302

1303

1304

1305

1306

1307

1308

1309

1310

1311

1312

1313

1314

1315

1316

1317

1318

1319

1320

1321

1322

1323

1324

1325

1326

1327

1328

1329

1330

1331

1332

1333

1334

1335

1336

1337

1338

1339

1340

and Dementia Unit at South London and Maudsley NHSFoundation Trust and [Institute of Psychiatry]Q5 King’s CollegeLondon and the 7th Framework Programme of the EuropeanUnion (ADAMS project, HEALTH-F4-2009-242257). Add-NeuroMed was funded through the EU FP6 programme. Pet-roula Proitsi is an Alzheimer’s Society Post-Doctoral Fellow.Richard JB Dobson and Cristina Legido-Quigley are partiallysupported from the Innovative Medicines Initiative Joint Un-dertaking under EMIF grant agreement No. 115372, re-sources of which are composed of financial contributionfrom the European Union’s Seventh Framework Programme(FP7/2007-2013) and EFPIA companies’ in-kind contribu-tion. The funding bodies had no role in the design, collection,analysis, interpretation of the data, writing of the manuscriptand the decision to submit the manuscript.

Supplementary data

Supplementary data related to this article can be found athttp://dx.doi.org/10.1016/j.jalz.2016.08.003.

RESEARCH IN CONTEXT

1. Systematic review: The authors reviewed the litera-ture using PubMed and reported key publications.There is a pressing need to identify noninvasive Alz-heimer’s disease (AD) biomarkers, and blood metab-olites are promising biomarkers that could aid earlydiagnosis and ultimately lead to the development ofmore effective interventions. Recent blood metabo-lomic studies have highlighted the role of lipid com-pounds in AD. However, most studies are small andrelatively heterogeneous.

2. Interpretation: This study replicated previous associ-ations between blood lipids and AD and reportednovel associations between blood lipids and clinicalAD diagnosis, the rate of cognitive and brain atrophy.These findings deepen our knowledge of AD diseasemechanisms and suggest novel targets for futurework.

3. Future directions: Results of this study could be com-plemented with protein and genetic data. Futurestudies should address whether these changes arecausally related to AD or are just a marker of changesin lipoprotein dynamics and composition.

References

[1] Mapstone M, Cheema AK, Fiandaca MS, Zhong X, Mhyre TR,

MacArthur LH, et al. Plasma phospholipids identify antecedent mem-

ory impairment in older adults. Nat Med 2014;20:415–8.

[2] Oresic M, Hyotylainen T, Herukka SK, Sysi-Aho M, Mattila I, Sep-

panan-Laakso T, et al. Metabolome in progression to Alzheimer’s dis-

ease. Transl Psychiatry 2011;1:e57.

[3] Trushina E, Dutta T, Persson XM, Mielke MM, Petersen RC. Identifi-

cation of altered metabolic pathways in plasma and CSF in mild cogni-

tive impairment and Alzheimer’s disease using metabolomics. PLoS

One 2013;8:e63644.

[4] Whiley L, Sen A, Heaton J, Proitsi P, Garcia-Gomez D, Leung R, et al.

Evidence of altered phosphatidylcholine metabolism in Alzheimer’s

disease. Neurobiol Aging 2014;35:271–8.

[5] Zipser BD, Johanson CE, Gonzalez L, Berzin TM, Tavares R,

Hulette CM, et al. Microvascular injury and blood-brain barrier

leakage in Alzheimer’s disease. Neurobiol Aging 2007;28:977–86.

[6] Lindon JC, Holmes E, Nicholson JK. Metabonomics in pharmaceu-

tical R&D. FEBS J 2007;274:1140–51.

[7] Suhre K, Shin SY, Petersen AK, Mohney RP, Meredith D, Wagele B,

et al. Human metabolic individuality in biomedical and pharmaceu-

tical research. Nature 2011;477:54–60.

[8] Simpson BN, Kim M, Chuang YF, Beason-Held L, Kitner-Triolo M,

Kraut M, et al. Blood metabolite markers of cognitive performance

and brain function in aging. J Cereb Blood Flow Metab 2016;

36:1212–23.

[9] Proitsi P, Kim M, Whiley L, Pritchard M, Leung R, Soininen H, et al.

Plasma lipidomics analysis finds long chain cholesteryl esters to be

associated with Alzheimer’s disease. Transl Psychiatry 2015;5:e494.

[10] Lovestone S, Francis P, Kloszewska I, Mecocci P, Simmons A,

Soininen H, et al. AddNeuroMed–the European collaboration for the

discovery of novel biomarkers for Alzheimer’s disease. Ann N Y

Acad Sci 2009;1180:36–46.

[11] Whiley L, Godzien J, Ruperez FJ, Legido-Quigley C, Barbas C. In-vial

dual extraction for direct LC-MS analysis of plasma for comprehen-

sive and highly reproducible metabolic fingerprinting. Anal Chem

2012;84:5992–9.

[12] Whiley L, Legido-Quigley C. Current strategies in the discovery of

small-molecule biomarkers for Alzheimer’s disease. Bioanalysis

2011;3:1121–42.

[13] Smith CA, Want EJ, O’Maille G, Abagyan R, Siuzdak G. XCMS: pro-

cessing mass spectrometry data for metabolite profiling using

nonlinear peak alignment, matching, and identification. Anal Chem

2006;78:779–87.

[14] Westman E, Aguilar C, Muehlboeck JS, Simmons A. Regional mag-

netic resonance imaging measures for multivariate analysis in Alz-

heimer’s disease and mild cognitive impairment. Brain Topogr 2013;

26:9–23.

[15] Simmons A, Westman E, Muehlboeck S, Mecocci P, Vellas B,

Tsolaki M, et al. MRI measures of Alzheimer’s disease and the Add-

NeuroMed study. Ann N YAcad Sci 2009;1180:47–55.

[16] Simmons A, Westman E, Muehlboeck S, Mecocci P, Vellas B,

Tsolaki M, et al. The AddNeuroMed framework for multi-centre

MRI assessment of Alzheimer’s disease : experience from the first

24 months. Int J Geriatr Psychiatry 2011;26:75–82.

[17] Sattlecker M, Kiddle SJ, Newhouse S, Proitsi P, Nelson S, Williams S,

et al. Alzheimer’s disease biomarker discovery using SOMAscan mul-

tiplexed protein technology. Alzheimers Dement 2014;10:724–34.

[18] Wilson RS, Arnold SE, Schneider JA, Kelly JF, Tang Y, Bennett DA.

Chronic psychological distress and risk of Alzheimer’s disease in old

age. Neuroepidemiology 2006;27:143–53.

[19] Walter A, Korth U, HilgertM, Hartmann J,Weichel O, HilgertM, et al.

Glycerophosphocholine is elevated in cerebrospinal fluid of Alzheimer

patients. Neurobiol Aging 2004;25:1299–303.

[20] Koal T, Klavins K, Seppi D, Kemmler G, Humpel C. Sphingomyelin

SM(d18:1/18:0) is significantly enhanced in cerebrospinal fluid sam-

ples dichotomized by pathological amyloid-beta42, tau, and

phospho-tau-181 levels. J Alzheimers Dis 2015;44:1193–201.

[21] Gonzalez-Dominguez R, Garcia-Barrera T, Gomez-Ariza JL. Combi-

nation of metabolomic and phospholipid-profiling approaches for the

study of Alzheimer’s disease. J Proteomics 2014;104:37–47.



1341

1342

1343

1344

1345

1346

1347

1348

1349

1350

1351

1352

1353

1354

1355

1356

1357

1358

1359

1360

1361

1362

1363

1364

1365

1366

1367

1368

1369

1370

1371

1372

1373

1374

1375

1376

1377

1378

1379

1380

1381

1382

1383

1384

1385

1386

1387

1388

1389

1390

1391

1392

1393

1394

1395

1396

1397

1398

1399

1400

1401

1402

1403

1404

1405

1406

1407

1408

1409

1410

1411

1412

1413

1414

1415

1416

1417

1418

1419

1420

1421

1422

1423

1424

1425

1426

1427

1428

1429

1430

1431

1432

1433

1434

1435

1436

1437

1438

1439

1440

1441

1442

1443

1444

1445

1446

1447

1448

1449

1450

1451

1452

1453

1454

1455

1456

1457

1458

1459

1460

1461

1462

1463

1464

1465

1466

1467

1468

1469

1470

1471

1472

1473

1474


[22] St-Onge MP, Mayrsohn B, O’Keeffe M, Kissileff HR, Choudhury AR,

Laferrere B. Impact of medium and long chain triglycerides consump-

tion on appetite and food intake in overweight men. Eur J Clin Nutr

2014;68:1134–40.

[23] Stegemann C, Pechlaner R, Willeit P, Langley SR, Mangino M,

Mayr U, et al. Lipidomics profiling and risk of cardiovascular disease

in the prospective population-based Bruneck study. Circulation 2014;

129:1821–31.

[24] Rhee EP, Cheng S, Larson MG, Walford GA, Lewis GD, McCabe E,

et al. Lipid profiling identifies a triacylglycerol signature of insulin

resistance and improves diabetes prediction in humans. J Clin Invest

2011;121:1402–11.

[25] Montoliu I, SchererM, Beguelin F, DaSilva L,Mari D, Salvioli S, et al.

Serum profiling of healthy aging identifies phospho- and sphingolipid

species as markers of human longevity. Aging (Albany NY) 2014;

6:9–25.

[26] Henderson ST, Vogel JL, Barr LJ, Garvin F, Jones JJ, Costantini LC.

Study of the ketogenic agent AC-1202 in mild to moderate Alz-

heimer’s disease: a randomized, double-blind, placebo-controlled,

multicenter trial. Nutr Metab (Lond) 2009;6:31.

[27] Proitsi P, Lupton MK, Velayudhan L, Newhouse S, Fogh I, Tsolaki M,

et al. Genetic predisposition to increased blood cholesterol and triglyc-

eride lipid levels and risk of Alzheimer disease: a mendelian random-

ization analysis. PLoS Med 2014;11:e1001713.

[28] Desikan RS, Schork AJ, Wang Y, Thompson WK, Dehghan A,

Ridker PM, et al. Polygenic Overlap Between C-Reactive Protein,

Plasma Lipids, and Alzheimer Disease. Circulation 2015;131:2061–9.

[29] Crameri A, Biondi E, Kuehnle K, Lutjohann D, Thelen KM, Perga S,

et al. The role of seladin-1/DHCR24 in cholesterol biosynthesis, APP

processing and Abeta generation in vivo. EMBO J 2006;25:432–43.

[30] Greeve I, Kretzschmar D, Tschape JA, Beyn A, Brellinger C,

Schweizer M, et al. Age-dependent neurodegeneration and Alzheimer-

amyloid plaque formation in transgenic Drosophila. J Neurosci 2004;

24:3899–906.

[31] Sato Y, Suzuki I, Nakamura T, Bernier F, Aoshima K, Oda Y. Identifi-

cation of a new plasma biomarker of Alzheimer’s disease using metab-

olomics technology. J Lipid Res 2012;53:567–76.

[32] Wisniewski T, Newman K, Javitt NB. Alzheimer’s disease: brain des-

mosterol levels. J Alzheimers Dis 2013;33:881–8.

[33] Reitz C, Mayeux R. Alzheimer disease: epidemiology, diagnostic

criteria, risk factors and biomarkers. Biochem Pharmacol 2014;

88:640–51.

[34] Reitz C. Dyslipidemia and the risk of Alzheimer’s disease. Curr Athe-

roscler Rep 2013;15:307.

[35] Reitz C. Dyslipidemia and dementia: current epidemiology, genetic

evidence, and mechanisms behind the associations. J Alzheimers Dis

2012;30:S127–45.

[36] Rowland AA, Voeltz GK. Endoplasmic reticulum-mitochondria con-

tacts: function of the junction. Nat RevMol Cell Biol 2012;13:607–25.

[37] Helle SC, Kanfer G, Kolar K, Lang A, Michel AH, Kornmann B. Or-

ganization and function of membrane contact sites. Biochim Biophys

Acta 1833;2013:2526–41.

[38] van Vliet A, Verfaillie T, Agostinis P. New functions of mitochondria

associated membranes in cellular signalling. Biochim Biophys Acta

2014;1843:2253–62.

[39] Area-Gomez E, Del Carmen Lara Castillo M, Tambini MD, Guardia-

Laguarta C, de Groof AJ, Madra M, et al. Upregulated function of

mitochondria-associated ER membranes in Alzheimer disease.

EMBO J 2012;31:4106–23.

[40] Zampese E, Fasolato C, Kipanyula MJ, Bortolozzi M, Pozzan T,

Pizzo P. Presenilin 2 modulates endoplasmic reticulum (ER)-mito-

chondria interactions and Ca21 cross-talk. Proc Natl Acad Sci U S

A 2011;108:2777–82.

[41] Sepulveda-Falla D, Barrera-Ocampo A, Hagel C, Korwitz A, Vinueza-

Veloz MF, Zhou K, et al. Familial Alzheimer’s disease-associated pre-

senilin-1 alters cerebellar activity and calcium homeostasis. J Clin

Invest 2014;124:1552–67.

[42] Hedskog L, Pinho CM, Filadi R, Ronnback A, Hertwig L,Wiehager B,

et al. Modulation of the endoplasmic reticulum-mitochondria interface

in Alzheimer’s disease and related models. Proc Natl Acad Sci U S A

2013;110:7916–21.

[43] Tambini MD, Pera M, Kanter E, Yang H, Guardia-Laguarta C,

Holtzman D, et al. ApoE4 upregulates the activity of mitochondria-

associated ER membranes. EMBO Rep 2016;17:27–36.

[44] Sen A, Wang Y, Chiu K, Whiley L, Cowan D, Chang RC, et al. Meta-

bolic phenotype of the healthy rodent model using in-vial extraction of

dried serum, urine, and cerebrospinal fluid spots. Anal Chem 2013;

85:7257–63.



1475

1476

1477

1478

1479

1480

1481

1482

1483

1484

1485

1486

1487

1488

1489

1490

1491

1492

1493

1494

1495

1496

1497

1498

1499

1500

1501

1502

1503

1504

1505

1506

1507

1508

1509

1510

1511

1512

1513

1514

1515

1516

1517

1518

1519

1520

1521

1522

1523

1524

1525

1526

1527

1528

1529

1530

1531

1532

1533

1534

1535

1536

1537

1538

1539

1540

1541

1542

1543

1544

1545

1546

1547

1548

1549

1550

1551

1552

1553

1554

1555

1556

1557

1558

1559

1560

1561

1562

1563

1564

1565

1566

1567

1568

1569

1570

1571

1572

1573

1574

1575

1576

1577

1578

1579

1580

1581

1582

1583

1584

1585

1586

1587

1588

1589

1590

1591

1592

1593

1594

1595

1596

1597

1598

1599

1600

1601

1602

1603

1604

1605

1606

1607

1608

King s Research Portal - CORE fileKing s Research Portal DOI: 10.1016/j.jalz.2016.08.003 Document...

Documents

Transcript of King s Research Portal - CORE fileKing s Research Portal DOI: 10.1016/j.jalz.2016.08.003 Document...