Identification of Aging-Associated Food Quality Changes in Citrus

Dec 19, 2017 - discriminatory multivariate statistical techniques, projection partial least-squares discrimant analysis (PLS-DA) and machine learning ...
1 downloads 10 Views 918KB Size
Subscriber access provided by READING UNIV

Article

Identification of Aging-Associated Food Quality Changes in Citrus Products Using Untargeted Chemical Profiling Ian Ronningen, and Devin G. Peterson J. Agric. Food Chem., Just Accepted Manuscript • DOI: 10.1021/acs.jafc.7b04450 • Publication Date (Web): 19 Dec 2017 Downloaded from http://pubs.acs.org on December 20, 2017

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

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

Page 1 of 29

Journal of Agricultural and Food Chemistry

1

Identification of Aging Associated Food Quality

2

Changes in Citrus Products Using Untargeted

3

Chemical Profiling

4 5 Ian G. Ronningen1 and Devin G. Peterson1,*

6 7 8

1

Department of Food Science, University of Minnesota, MN 55108

9

*

Corresponding author

10

*

Current address: 317 Parker Building, Food Science & Technology, The Ohio State

11

University, 2015 Fyffe Rd, Columbus, OH 43210

12

*

Corresponding Author Email: [email protected]

13

ACS Paragon Plus Environment

Journal of Agricultural and Food Chemistry

14

Abstract

15

Chemometric techniques have seen wide application in biological and medical sciences,

16

but are still developing in the food sciences. This study illustrated the use of untargeted

17

LC/MS chemometric methods to identify features (retention time_m/z) associated with

18

food quality changes as products age (freshness). Extracts of three citrus fruit varietals

19

aged over four time points that corresponded to noted changes in sensory attributes, were

20

chemical profiled and modeled by two discriminatory multivariate statistical techniques,

21

projection partial least squares data analysis (PLS-DA) and machine learning random

22

forest (RF). Age-associated compounds across the citrus platform were identified.

23

Varietal was treated as a nuisance variable to emphasize aging chemistry, and further

24

variable selection using age-related piecewise model generation and meta filtering to

25

emphasize features associated with general aging chemistry common to all the citrus

26

extracts. The identified features were further replicated in a validation study to illustrate

27

the validity and persistence of these markers for applications in citrus food platforms.

28 29 30 31

Key Words: flavoromics, untargeted chemical profiling, MVA, flavor, freshness

ACS Paragon Plus Environment

Page 2 of 29

Page 3 of 29

Journal of Agricultural and Food Chemistry

32

Introduction

33

A core aim of flavor chemistry research is to understand the molecular basis for the

34

sensory properties of foodstuffs. Traditionally methods of flavor identification have

35

relied on ‘targeted’ analytical approaches that screen individual compounds and select

36

those with flavor activity1–3. While these focused, reductionist analytical approaches have

37

advanced our understanding of the compounds that contribute to food flavor, they are

38

also inherently limited in scope as compounds are evaluated individually (in isolation)

39

and out of context. Flavor is a complex sensation derived from a multitude of stimuli

40

from different sensory systems such as aroma (olfactory), taste (gustatory), and

41

tactile/irrigation/temperature (chemestetic). Consequently, analytical methods based on

42

singular compound evaluation can overlook drivers of flavor perception by ignoring

43

potential interactions among stimuli. Untargeted chemical profiling methods would

44

therefore provide a further basis to characterize flavor compounds/interactions. These

45

approaches typically utilize orders of magnitude more data compared to targeted methods

46

and are much more likely to identify interacting data features4,5.

47 48

Recent advancements in many scientific fields have benefited from inclusive data driven

49

‘omic’ research methods6–9. Untargeted chemical profiling methods are positioned as a

50

complementary method to the more traditional targeted approaches utilized in the

51

biological sciences10. For example, the use of metabolomics in plant physiology has

52

progressed the understanding of phenotypes by increasing understanding of contextual

53

relationships4. Paralleling advances seen in other fields with complex interrelated

54

outcomes, Reineccius et al11 discussed the term “Flavoromics” as a new frontier in flavor

ACS Paragon Plus Environment

Journal of Agricultural and Food Chemistry

55

research, initially suggesting the use of proven workflows seen in metabolomics and

56

other “omics” fields. Although this term has been present for a number of years, the field

57

has yet to see substantial growth as well as applications still lack a sensory verification

58

aspect or have a limited scope12,13. Emphasis has been placed on differentiating samples

59

without establishing causative relationships for the identified statistical features14. Even

60

as computational resources and analytical technology have progressed, untargeted

61

approaches often conclude investigations using only principle component analysis to

62

make conclusions whereas traditional targeted methods provide significantly stronger

63

conclusions15.

64 65

Combining untargeted chemical fingerprinting methods with multivariate analysis allows

66

for the inclusion of as many data points as possible, providing a hypothesis seeking

67

approach which relates inputs (e.g. compounds) to an output (i.e. flavor). Within flavor

68

research, data driven approaches has been applied to instrumental data for sensory panel

69

prediction12,16,17. The ability to gain novel information from flavor analysis while also

70

generating models for future prediction of flavor relevant outcomes is a major motivator

71

to apply untargeted chemometric methodology to food systems. Other applications in

72

food aim to understand the geographical region or type of food being analyzed14,18–20.

73

While there has been an increase in the application of untargeted chemometric based

74

research, there is still limited validation of the identified statistical features and

75

translatability.

76

ACS Paragon Plus Environment

Page 4 of 29

Page 5 of 29

Journal of Agricultural and Food Chemistry

77

The current study is the first phase of a two-part investigation to apply

78

discriminatory untargeted chemical profiling methods to identify compounds

79

related to flavor quality changes as products age (freshness). The first phase

80

focuses on chemical characterization, statistical methods of analyses and modeling

81

techniques for compound selection, whereas the second phase focused on sensory

82

validation and compound identification 21. In this current paper, discriminatory

83

untargeted liquid chromatography-mass spectrometry-time of flight (LC-MS-tof)

84

chemical fingerprinting methods coupled with multivariate analysis (MVA) were utilized

85

to characterize chemical changes in citrus extracts during storage to identify compounds

86

associated with flavor quality changes as the products aged (freshness). LC/MS

87

techniques were selected to provide further characterization (novel insights) of the

88

chemical changes of oranges related to the flavor attributes of oranges. Prior research on

89

identification of orange citrus flavor has primarily emphasized volatile aroma

90

characterization22,23. Two modeling methods, projection partial least squares (PLS) and

91

machine learning random forest (RF) were utilized, with further variable selection using

92

age-related piecewise model generation and meta filtering techniques. Finally, the

93

statistically significant chemical data was validated in a repeated experiment.

94 95

Materials and methods

96

Sample Preparation. Three common citrus hybrid fruit products were sourced form the

97

domestic market. Navel Oranges, Mineola Tangerines, and Valencia Oranges were

98

washed, rinsed, and then cut to be 97%) by multi-dimensional

365

LC fractionation and subsequently identified by TOF MS/MS and NMR, as ionone

366

glycoside and nomilin glucoside, respectively. Furthermore, when each compound was

367

added to fresh orange samples at quantities identified in aged sample, a trained sensory

368

panel reported the flavor attributes changed and mimicked that of the aged product.

369

Interestingly both compounds themselves in water did not report any sensory activity and

370

would not be expected to be identified by traditional targeted methods1–3. To the best of

371

the authors knowledge, this is a first report to identify and validate persistent LC/MS

372

chemical features (unknowns) using statistical directed methods linked to sensory

373

significance.

374

ACS Paragon Plus Environment

Journal of Agricultural and Food Chemistry

375

In summary, this work illustrates the utility of flavoromics (untargeted analysis) for

376

identifying data trends related to a biological (flavor) outcome and an approach to

377

establish those relationships as persistent through repeated modeling. Additionally, this

378

work illustrates how modeling can emphasize aging through treating varietal differences

379

a nuisance variable to identify chemistry related to a ‘global’ citrus platform and to

380

identify flavor-active compounds.

381 382

ACKNOWLEDGMENT

383

The authors would like to acknowledge the financial support provided by the Flavor

384

Research and Education Center at The Ohio State University and its supporting members.

385

ACS Paragon Plus Environment

Page 18 of 29

Page 19 of 29

Journal of Agricultural and Food Chemistry

386

Cited Literature

387

(1)

Ottinger, H.; Bareth, A.; Hofmann, T. Characterization of Natural “Cooling”

388

Compounds Formed from Glucose and l -Proline in Dark Malt by Application of

389

Taste Dilution Analysis. J. Agric. Food Chem. 2001, 49 (3), 1336–1344.

390

(2)

O. Frank; and H. Ottinger; Hofmann*, T. Characterization of an Intense Bitter-

391

Tasting 1H,4H-Quinolizinium-7-olate by Application of the Taste Dilution

392

Analysis, a Novel Bioassay for the Screening and Identification of Taste-Active

393

Compounds in Foods. 2000.

394

(3)

395 396

of gas chromatographic effluents. Food Chem. 1984, 14 (4), 273–286. (4)

397 398

(5)

Myers, C. L.; Troyanskaya, O. G. Context-sensitive data integration and prediction of biological networks. Bioinformatics 2007, 23 (17), 2322–2330.

(6)

401 402

Fiehn, O. Metabolomics – the link between genotypes and phenotypes. Plant Mol. Biol. 2002, 48 (1), 155–171.

399 400

Acree, T. E.; Barnard, J.; Cunningham, D. G. A procedure for the sensory analysis

Davies, H. A role for “omics” technologies in food safety assessment. Food Control 2010, 21 (12), 1601–1610.

(7)

Buczynski, M. W.; Dumlao, D. S.; Dennis, E. A. Thematic Review Series:

403

Proteomics. An integrated omics analysis of eicosanoid biology. J. Lipid Res.

404

2009, 50 (6), 1015–1038.

405

(8)

Roux, A.; Lison, D.; Junot, C.; Heilier, J.-F. Applications of liquid

406

chromatography coupled to mass spectrometry-based metabolomics in clinical

407

chemistry and toxicology: A review. Clin. Biochem. 2011, 44 (1), 119–135.

408

(9)

Trygg, J.; Holmes, E.; Lundstedt, T. Chemometrics in metabonomics. J. Proteome

ACS Paragon Plus Environment

Journal of Agricultural and Food Chemistry

409 410

Res. 2007, 6 (2), 469–479. (10)

411 412

2014, 1 (1), 2053951714528481. (11)

413 414

Kitchin, R. Big Data, new epistemologies and paradigm shifts. Big Data Soc.

Reineccius, G. Flavoromics- the next frontier? Abstr. Pap. 235th ACS Natl. Meet. 2008.

(12)

Charve, J.; Chen, C.; Hegeman, A. D.; Reineccius, G. a. Evaluation of

415

instrumental methods for the untargeted analysis of chemical stimuli of orange

416

juice flavour. Flavour Fragr. J. 2011, 26 (6), 429–440.

417

(13)

Majcher, M.; Siger, A.; Kaczmarek, A.; Gracka, A.; Jele, H. H. Flavoromics

418

approach in monitoring changes in volatile compounds of virgin rapeseed oil

419

caused by seed roasting. J. Chromatogr. A 2016, 1428, 292–304.

420

(14)

421 422

Aishima, T.; Nakai, S. Pattern Recognition of GC Profiles for Classification of Cheese Variety. 1987, 52 (4), 1985–1988.

(15)

Lindinger, C.; Pollien, P.; Labbe, D.; Rytz, A.; Juillerat, M. A.; Blank, I.

423

Prediction of the overall sensory profile of espresso coffee by online headspace

424

measurement using proton transfer reaction mass spectrometry (PTR-MS). Dev.

425

Food Sci. 2006, 43 (Flavour Science: Recent Advantages and Trends), 497–500.

426

(16)

Togari, N.; Kobayashi, A.; Aishima, T. Relating sensory properties of tea aroma to

427

gas chromatographic data by chemometric calibration methods. Food Res. Int.

428

1995, 28 (5), 485–493.

429

(17)

Tikunov, Y.; Bovy, a G.; Hall, R. D. Flavour Metabolomics : Holistic Approaches

430

in Flavour Research Versus Targeted Approaches in Flavor Research. Expr.

431

Multidiscip. Flavour Sci. Sci. 2008, 573–580.

ACS Paragon Plus Environment

Page 20 of 29

Page 21 of 29

432

Journal of Agricultural and Food Chemistry

(18)

Mehl, F.; Marti, G.; Boccard, J.; Debrus, B.; Merle, P.; Delort, E.; Baroux, L.;

433

Raymo, V.; Velazco, M. I.; Sommer, H.; et al. Differentiation of lemon essential

434

oil based on volatile and non-volatile fractions with various analytical techniques:

435

a metabolomic approach. Food Chem. 2014, 143, 325–335.

436

(19)

Lee, S. M.; Kwon, G. Y.; Kim, K.-O.; Kim, Y.-S. Metabolomic approach for

437

determination of key volatile compounds related to beef flavor in glutathione-

438

Maillard reaction products. Anal. Chim. Acta 2011, 703 (2), 204–211.

439

(20)

Ochi, H.; Naito, H.; Iwatsuki, K.; Bamba, T.; Fukusaki, E. Metabolomics-based

440

component profiling of hard and semi-hard natural cheeses with gas

441

chromatography/time-of-flight-mass spectrometry, and its application to sensory

442

predictive modeling. J. Biosci. Bioeng. 2012, 113 (6), 751–758.

443

(21)

Ronningen, I.; Miller, M.; Xia, Y.; Peterson, D. G. Identification and Validation of

444

Sensory-Active Compounds from Data-Driven Research: A Flavoromics

445

Approach. J. Agric. Food Chem. 2017, acs.jafc.7b00093.

446

(22)

447 448

Fresh and Processed Orange Juices. J. Agric. Food Chem 1990, 38 (4), 1048–1052. (23)

449 450

Perez-Cacho, P. R.; Rouseff, R. L. Fresh squeezed orange juice odor: a review. Crit. Rev. Food Sci. Nutr. 2008, 48 (7), 681–695.

(24)

451 452

Nisperos-carriedo, M.; Shaw, P. E. Comparison of Volatile Flavor Components in

Wold, S.; Sjostrom, M.; Eriksson, L. PLS-regression: a basic tool of chemometrics. Chemom. Intell. Lab. Syst. 2001, 58, 109–130.

(25)

Tautenhahn, R.; Patti, G. J.; Kalisiak, E.; Miyamoto, T.; Schmidt, M.; Lo, F. Y.;

453

McBee, J.; Baliga, N. S.; Siuzdak, G. metaXCMS: second-order analysis of

454

untargeted metabolomics data. Anal. Chem. 2011, 83 (3), 696–700.

ACS Paragon Plus Environment

Journal of Agricultural and Food Chemistry

455

(26)

Liland, K. H.; Snipen, L.; Mehmood, T.; Liland, K. H.; Snipen, L.; Sæbø, S. A

456

review of variable selection methods in Partial Least Squares Regression A review

457

of variable selection methods in Partial Least Squares Regression. Chemom. Intell.

458

Lab. Syst. 2012, No. 118, 62–69.

459

(27)

Wold, S.; Trygg, J. The PLS method -- partial least squares projections to latent

460

structures -- and its applications in industrial RDP ( research , development , and

461

production ). PLS Ind. RPD Prague 2004, 1 (June), 1–44.

462

(28)

Menze, B. H.; Kelm, B. M.; Masuch, R.; Himmelreich, U.; Bachert, P.; Petrich,

463

W.; Hamprecht, F. A. A comparison of random forest and its Gini importance with

464

standard chemometric methods for the feature selection and classification of

465

spectral data. 2009, 16, 1–16.

466

(29)

Breiman, L. Random forests. Mach. Learn. 2001, 45 (1), 5–32.

467

ACS Paragon Plus Environment

Page 22 of 29

Page 23 of 29

Journal of Agricultural and Food Chemistry

468 Table 1. Variable of Importance metrics for the identified statistical features for both PLS and Random Forrest (RF) modeling approaches in the initial and validation experiment Feature

Initial PLS

Initial RF

Validation PLS

Validation RF

(RT_M/ Z)

VIP

VIP

VIP

VIP

2.541_383.114

2.4

2.8

1.65

1.98

2.193_413.121

2.11

2.2

1.27

1.25

3.145_693.283

1.99

2.6

1.23

1.76

2.454_413.121

1.81

2.4

1.35

1.82

6.93_563.2393

1.62

0

1.02

1

4.909_191.091

1.52

2.4

NA

0

3.07_684.3070

1.5

0

1.07

1

2.38_ 661.2653

1.4

1

1.4

1

1.50_541.1749

1.33

0

0.99

0

5.48_457.2563

1.09

1.2

1.05

0

2.208_295.063

0.3

2.2

2.15

2.06

469

ACS Paragon Plus Environment

Journal of Agricultural and Food Chemistry

470

Figure Captions

471

Figure 1. Unsupervised PCA of chemical fingerprinting of citrus extracts (n=20 for each

472

varietal).

473 474

Figure 2. PLS-DA models of sample age for data subsets of each of the varietals; A.

475

Mineola (R2X= 0.571, R2Y=0.988, Q2=0.923), B. Navel (R2X=0.631, R2Y=0.988,

476

Q2=0.915), C. Valencia (R2X=0.632, R2Y=0.989, Q2=0.923)

477 478

Figure 3. PLS-DA of citrus aging with classification of sample age rather than varietal

479

and age; R2Y= 0.95 and a Q2 of 0.981.

480 481

Figure 4. Variable of importance for the top ten most features in the Random Forest

482

analysis, the model out of bag error was 2.38% and generated with piecewise

483

optimization of parameters.

484

ACS Paragon Plus Environment

Page 24 of 29

Page 25 of 29

Journal of Agricultural and Food Chemistry

Navel

Mineola

Valencia

485 486

Figure 1. Unsupervised PCA of chemical fingerprinting of citrus extracts (n=20 for each

487

varietal).

488

ACS Paragon Plus Environment

Journal of Agricultural and Food Chemistry

Page 26 of 29

A. Mineola Day 4

Day 6

Day 2

Day 0

B. Navel Day 2 Day 4 Day 6

Day 0

C. Valencia Day 2 Day 6 Day 4

Day 0

489 490

Figure 2. PLS-DA models of sample age for data subsets of each of the varietals; A.

491

Mineola (R2X= 0.571, R2Y=0.988, Q=0.923), B. Navel (R2X=0.631, R2Y=0.988,

492

Q2=0.915), C. Valencia (R2X=0.632, R2Y=0.989, Q2=0.923).

ACS Paragon Plus Environment

Page 27 of 29

Journal of Agricultural and Food Chemistry

Day 2 Day 6 Day 0 Day 4 493 494

Figure 3. PLS-DA of citrus aging with classification of sample age rather than varietal

495

and age; R2Y= 0.95 and a Q2 of 0.981.

496 497 498 499 500

ACS Paragon Plus Environment

Journal of Agricultural and Food Chemistry

Page 28 of 29

2.541_383.114 3.145_693.283 2.311_337.076 2.454_413.121 4.909_191.091 1.662_148.972 3.381_419.204 2.497_295.063 0.225_326.921 2.208_295.063 2.193_413.121 2

2.1

2.2

2.3 2.4 2.5 2.6 Mean Decrease In Accuracy

2.7

2.8

2.9

501 502

Figure 4. Variable of importance for the top ten most features in the Random Forest

503

analysis, the model out of bag error was 2.38% and generated with piecewise

504

optimization of parameters.

505 506 507 508 509 510 511 512 513 514

ACS Paragon Plus Environment

Page 29 of 29

515

Journal of Agricultural and Food Chemistry

TOC

516

ACS Paragon Plus Environment