King s Research Portal - CORE fileKing s Research Portal DOI: 10.1016/j.jalz.2016.08.003 Document...
Transcript of 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
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]
http://dx.doi.org/10.1016/j.jalz.2016.08.003
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
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.
FLA 5.4.0 DTD � JALZ2275_proof � 28 September 2016 � 2:14 pm � ce
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]).
FLA 5.4.0 DTD � JALZ2275_proof � 28 September 2016 � 2:14 pm � ce
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.
FLA 5.4.0 DTD � JALZ2275_proof � 28 September 2016 � 2:14 pm � ce
P. Proitsi et al. / Alzheimer’s & Dementia - (2016) 1-124
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)
FLA 5.4.0 DTD � JALZ2275_proof � 28 September 2016 � 2:14 pm � ce
P. Proitsi et al. / Alzheimer’s & Dementia - (2016) 1-12 5
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])
FLA 5.4.0 DTD � JALZ2275_proof � 28 September 2016 � 2:14 pm � ce
P. Proitsi et al. / Alzheimer’s & Dementia - (2016) 1-126
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.
FLA 5.4.0 DTD � JALZ2275_proof � 28 September 2016 � 2:14 pm � ce
P. Proitsi et al. / Alzheimer’s & Dementia - (2016) 1-12 7
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.
FLA 5.4.0 DTD � JALZ2275_proof � 28 September 2016 � 2:14 pm � ce
P. Proitsi et al. / Alzheimer’s & Dementia - (2016) 1-128
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.
FLA 5.4.0 DTD � JALZ2275_proof � 28 September 2016 � 2:14 pm � ce
P. Proitsi et al. / Alzheimer’s & Dementia - (2016) 1-12 9
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
FLA 5.4.0 DTD � JALZ2275_proof � 28 September 2016 � 2:14 pm � ce
P. Proitsi et al. / Alzheimer’s & Dementia - (2016) 1-1210
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.
FLA 5.4.0 DTD � JALZ2275_proof � 28 September 2016 � 2:14 pm � ce
P. Proitsi et al. / Alzheimer’s & Dementia - (2016) 1-12 11
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.
FLA 5.4.0 DTD � JALZ2275_proof � 28 September 2016 � 2:14 pm � ce
P. Proitsi et al. / Alzheimer’s & Dementia - (2016) 1-1212
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