Untargeted Mass Spectrometry-Based Metabolomics Tracks Molecular Changes in Raw and Processed Foods and Beverages

A major aspect of our daily lives is the need to acquire, store and prepare our food. Storage and preparation can have drastic effects on the compositional chemistry of our foods, but we have a limited understanding of the temporal nature of processes such as storage, spoilage, fermentation and brewing on the chemistry of the foods we eat. Here, we performed a temporal analysis of the chemical changes in foods during common household preparations using untargeted mass spectrometry and novel data analysis approaches. Common treatments of foods such as home fermentation of yogurt, brewing of tea, spoilage of meats and ripening of tomatoes altered the chemical makeup through time, through both chemical and biological processes. For example, brewing tea altered its composition by increasing the diversity of molecules, but this change was halted after 4 min of brewing. The results indicate that this is largely due to differential extraction of the material from the tea and not modification of the molecules during the brewing process. This is in contrast to the preparation of yogurt from milk, spoilage of meat and the ripening of tomatoes where biological transformations directly altered the foods molecular composition. Comprehensive assessment of chemical changes using multivariate statistics showed the varied impacts of the different food treatments, while analysis of individual chemical changes show specific alterations of chemical families in the different food types. The methods developed here represent novel approaches to studying the changes in food chemistry that can reveal global alterations in chemical profiles and specific transformations at the chemical level. Highlights We created a reference data set for tomato, milk to yogurt, tea, coffee, turkey and beef. We show that normal preparation and handling affects the molecular make-up. Tea preparation is largely driven by differential extraction. Formation of yogurt involves chemical transformations. The majority of meat molecules are not altered in 5 days at room temperature.

A major aspect of our daily lives is the need to acquire, store and prepare our food. Storage and 63 preparation can have drastic effects on the compositional chemistry of our foods, but we have a 64 limited understanding of the temporal nature of processes such as storage, spoilage, 65 fermentation and brewing on the chemistry of the foods we eat. Here, we performed a temporal 66 analysis of the chemical changes in foods during common household preparations using 67 untargeted mass spectrometry and novel data analysis approaches. Common treatments of 68 foods such as home fermentation of yogurt, brewing of tea, spoilage of meats and ripening of 69 tomatoes altered the chemical makeup through time, through both chemical and biological 70 processes. For example, brewing tea altered its composition by increasing the diversity of 71 molecules, but this change was halted after 4 min of brewing. The results indicate that this is 72 largely due to differential extraction of the material from the tea and not modification of the 73 molecules during the brewing process. This is in contrast to the preparation of yogurt from milk, 74 spoilage of meat and the ripening of tomatoes where biological transformations directly altered 75 the foods molecular composition. Comprehensive assessment of chemical changes using 76 multivariate statistics showed the varied impacts of the different food treatments, while analysis 77 of individual chemical changes show specific alterations of chemical families in the different food 78 types. The methods developed here represent novel approaches to studying the changes in 79 food chemistry that can reveal global alterations in chemical profiles and specific 80 transformations at the chemical level. 81 Introduction: 82 We consume a variety of foods and beverages during any given day, such as fruits, vegetables, 83 dairy products and meats. Food is stored and processed in many different ways before 84 consumption, yet we know very little about the molecular impacts of such "normal" food 85 treatments before we consume them. There is a significant interest and awareness in the 86 population about the molecular contents of food. Consistent with this interest, there are >37,000 87 articles in Pubmed using the terms "food" and "mass spectrometry" but only ~250 when using 88 the search terms "food", "untargeted", and "mass spectrometry" or "metabolomics". Although as 89 many as 25,000 food molecules are known, the majority of food mass spectrometry studies 90 focus on the detection of insecticides, pesticides and toxins or particular compound classes 91 such as polyphenols to which healthy properties are attributed (Casida & Durkin, 2017, Giorio et 92 al., 2017, Scalbert et al., 2014) and are used to compare different food supplements such as the 93 coffee leaves (Souard, et al., 2018). As a consequence, much of the work is done by targeted 94 methods and/or GC-MS for untargeted methods. Nevertheless, the importance of mass 95 spectrometry as the most sensitive and selective tool currently available to decipher our food is 96 is only expected to grow (Yoshimura, Goto-Inoue, Moriyama, & Zaima, 2016) in areas such as 97 food monitoring during processing (Marshall et al., 2017), especially as the cost per data volume 98 of mass spectrometry has decreased by two orders of magnitude in the past 15 years and is 99 expected to continue to go down (Aksenov et al., 2017). An untargeted approach using LC-100 MS/MS has not been as widely used to analyze food types and effects of storage and 101 processing and never in conjunction with emerging untargeted mass spectrometry analysis 102 approaches such as mass spectral molecular networking to assess changes based on 103 processing. 104 Mass spectral molecular networking enables a broad overview of the molecular 105 information, that can be inferred from the MS/MS data (Watrous et al., 2012). For example, 106 molecular networking has been used in food analysis to study Siberian ginseng (Ge,Zhu,107 Yoshimatsu  visualize chemical relationships of detected compounds and  121  provide a powerful tool for in-depth interpretation of chemical transformations.  122  With these and other widely used mass spectrometry approaches, such as multivariate  123  statistics, we set out to investigate how the molecular make-up of foods is impacted by normal  124 handling before consumption, during processing and preparation. We hypothesize that many of 125 our methods of sourcing, handling and/or processing of food impact the molecular make-up of 126 the foods and that untargeted mass spectrometry combined with advanced analysis tools can 127 give us insight into these impacts on a molecular level. Building upon this hypothesis we aim to 128 address some of the following specific questions: 1) How does the molecular composition of a 129 food change based on how it was ripened off the vine or its particular sourcing ( Three packages of organic products for the beef and 3  149 packages with the labeling "without antibiotics and growth hormones" for the turkey as well as 150 three packages of products without any "organic" labeling for both meat types were selected. 151 Each package of product was sampled into two petri dishes: one of them was spiked with 152 tetracycline ( sampled based on processing types and/or longitudinal changes to determine molecular 343 variation associated with each processing method. Figure 1 highlights some representative 344 examples of images associated with the specific foods that were sampled. The numbers on the 345 tubes indicate the barcode number associated with each file, which was used to track the 346 information and metadata for the entire project. Figure 1a shows an example of ground beef at 347 3 days left at room temperature (a part of a 5 day time course to investigate meat spoilage). The 348 discoloration of the meat is non-uniform. The next sample type is tea, where twelve teas were 349 subjected to brewing for 0.5 min, 1 min, 4 min and 240 min. A representative sample point of 350 one tea at 1 min is shown in Figure 1b. The third sample type is tomatoes (Figure 1c). Both the 351 tomato origin and impact of time of storage at room temperature on the molecular make-up 352 were investigated. The fourth sample type studied was the home fermentation of yogurt, over 6 353 days, including controls of the milk and initial yogurts containing live active cultures. One 354 example for yogurt fermentation is shown in Figure 1e. Finally, we assessed different roasts of 355 coffee (the packaging for the medium/dark roast is shown in Figure 1d). Each of the samples 356 were subjected to extraction as outlined in the methods and the resulting extracts were 357 subjected to LC-MS/MS-based mass spectrometry. To obtain an overview of the data, we 358 created PCoA plots, heatmaps and molecular networks. image is provided as supporting information. 382 In parentheses the number of samples for 383 each group are shown. 384 385 The PCoA analysis of the yogurt and milk 386 samples shows distinct grouping between 387 yogurt and milk samples (Figure 3a-3d). differentiates the samples is the type of meat (Figure 3f). Samples with and without tetracycline 404 addition change similarly over time, indicating that based on this multivariate analysis, the 405 addition of tetracycline does not greatly impact the aging process (Figure 3g). Surprisingly, 406 even after leaving the meats out at room temperature for 5 days and the development of a 407 significant emanating odor from the samples, no trend could be spotted in the PCoA. A greater 408 change may be detected using other methods such as GC-MS that would detect volatile 409 compounds. As Figure 1a reveals, the aging process is non-uniform and thus the experimental 410 variation of the samples within the same time points appears to be larger than the overall 411 molecular variation associated with the 5 day aging process. 412 PCoA analysis of the tomato samples revealed that both source (Figure 3h) and storage 413 time (Figure 3i) affect the molecular composition of tomatoes. As expected, canned and 414 sundried data occupy very different PCoA space than fresh tomatoes. It is also notable that 415 differences exist for fresh tomatoes, with those from Farmer's market most closely resembling 416 home garden tomatoes and all store-bought tomatoes resembling one another, whether organic 417 or not. When organic tomatoes are left at room temperature they occupy the bottom left corner 418 in Figure 3i and gradually change to the lower right over the course of 6 days suggesting that 419 there are major molecular changes over this time period and upon inspecting Figure 2 it 420 appears that these changes are larger than the changes in meat over the same period. Notably, 421 despite the magnitude of these molecular changes the tomatoes did not change at all in either 422 their appearance or smell. 423 PCoA analysis of the coffee revealed a clear trend among the sample type liquid 424 "brewed coffee" or solid "ground coffee" (Figure 3j-l). If the coffee sample was a liquid from 425 brewing and the brew was extracted, then the sample appeared on the left side of the PCoA, 426 while extracts of the ground beans (picked up with clean spoon) or the cut beans (with sterile 427 knife) themselves directly appeared on the right. Besides sample type, the data suggest there is 428 clustering based on the roasting type, as there are clusters associated with clustering of dark 429 roast, light roast and the medium roast. This is particularly noticeable when the coffee is 430 extracted from the ground beans and/or directly from the beans (Figure 3k).

431
PCoA analysis of tea samples, Figure 3m-3o, revealed unambiguous differentiation of 432 solid from liquid samples, prepared using room temperature ethanol solution and 95°C water, 433 respectively. Note that water samples were most differentiated from the solid extract samples 434 along PC1. Twelve different teas were sampled over time (0.5 min, 1 min, 4 min, and 240 min) 435 to emulate the brewing process (Figure 3m). A water-only control at the same time points did 436 not change (Figure 3m) and direct extractions of the solid teas are also shown. Interestingly, 437 the samples appeared most similar to water blanks in the earliest time points and became more 438 similar to the solid samples over time (along the PC1 axis which explained 25.9% of total 439 variance), independent of tea type, indicating continued release of compounds from the leaves. 440 The kinetics of tea extraction were similar for all teas -interestingly, the observed chemical 441 differences between 240 min and 4 mins were minor for all teas which supports a steeping time 442 rationale which appears to be sufficiently effective for tea extraction of phytochemicals. A slight 443 deviation in overall kinetic trend was observed for oolong. The first two time points (0.5 and 1 444 min) appeared to be more similar to the blanks than other teas at the same time point. 445 Differences based on tea type were also observed. White, green, matcha, and black tea liquid 446 samples were more similar to each other than to oolong and pu'er, which were differentiated 447 along PC3 (6.81% of total variance); Figure 3n and 3o, illustrate the clear differences between 448 Chinese teas (oolong and pu'er), Figure 3n) and the American and British teas (Figure 3o). visualize key drivers in molecular patterns. The three store bought yogurts containing live active 458 cultures, the milk and the home ferments using the different yogurts as starter culture show 459 distinct groupings. The spheres are colored based on fermentation time from 0 to 58 hrs (a-d). 460 The meat samples separate by animal type (e), duration left at room temperature (f), but do not 461 show a clear trend based on tetracycline addition (g); tomato samples display differences based 462 on source (h) and storage time (i); coffee samples group based on collection device, which 463 tracks with brewed coffee vs. ground beans (j); the impact of the roast type is also depicted in 464 (k) and (l). Tea samples differentiated based on whether they were extracted with ethanol (tea 465 leaves) or first extracted with water, over a range of brewing times (m); different Chinese tea 466 varieties group separately (n) from the British and American teas (o). 467

Heatmaps: 468
We created heatmaps to visualize molecular changes driving differences between 469 samples for the time course experiments of tea brewing, yogurt fermentation, tomato ripening 470 and improper meat storage and gain more insight into groups of features that behave similarly 471 over time or in different sample types. In addition to a PCoA, heatmaps provide a visual 472 overview of the data to give more detailed insights behind molecular changes driving the 473 differences between sample types and within sample types. Because the tea and the milk-to-474 yogurt had the largest changes in abundances of groups of molecules they are shown in Figure  475 4. Other heatmaps are shown in the Supplementary Information (Supplementary Figure 1-4). 476 Consistent with the PCoA analysis, we observe different metabolite profiles between solid and 477 liquid samples in tea (Figure 4a). Furthermore, we observe that relative intensity of molecular 478 features increases with extraction time independent of the tea type. We assessed the 479 correlation of relative intensity per feature and tea type with extraction time. In tea this resulted 480 in a total of 2,045 significantly correlated features (spearman correlation, p-value < 0.05). 481 Figure 6c highlights selected molecular features for which we obtained a putative structure 482 annotation through GNPS library matching. For example, we observe that the relative intensity 483 of procyanidin B and theaflavin increase over time (Kruskal-Wallis, N=6, p-value ranging from 484 0.01 to 0.02, between brewing times 0.5 and 240). We also assessed the correlation of relative 485 intensity per feature and home ferment with different yogurt inoculums over time. For the Kroger 486 yogurt, this resulted in a total of 1,587 significantly correlated features (spearman correlation, p-487 value < 0.05). Figure 5b highlights selected molecular features for which we obtained a putative 488 structure annotation through GNPS library matching. For example, we observe that the relative 489 intensity of 4-O-beta-galactopyranosyl-D-mannopyranose decreases over time for each yogurt 490 type individually as well as overall (Kruskal-Wallis, N=9, p-value=0.0023, between 0 and 58 491 hrs). 492 493 Molecular changes during meat (beef and turkey) storage over five days were also 494 visualized (Supplementary Figure 2a). When comparing antibiotic vs non antibiotic treated 495 meat (both beef and turkey), the overall molecular differences as seen in PCoA space do not 496 vary much. However, there are some specific low intensity molecules that change, although with 497 minimal differences due to the addition of tetracycline, consistent with the observations from the 498 PCoA. We do observe differences between organic and non-organic beef. For example, in the 499 non-organic beef, oleoyl-taurine increases during the 5 days and does not appear by day 5 in 500 the organic samples, while the levels of acetyl-carnitine decrease in the non-organic beef but 501 are consistent across all time points for the organic beef. In the turkey the rate of appearance of 502 oleoyl-taurine and rate of disappearance of acetyl-carnitine are only slightly different 503 (Supplementary Figure 2b). The spectral match with parent mass difference 0.000 Da and 504 very strong cosine match of 0.84, to the fungal molecule, termitomycamide E (Choi et al., 2010), 505 increases over time, the presence of three analogues with mass differences pointing to different 506 acyl chain lengths, and minor suppression by tetracycline appears to have been detected and 507 would be consistent with increased microbial loads (Supplementary Figure 4). 508 Molecular differences between tomato samples were most striking when comparing sun 509 dried, canned and fresh tomatoes. In the heatmap visualizing molecular changes during the 510 ripening of fresh tomatoes (Supplementary Figure 1) no clear-cut large scale patterns were 511 observed. During the ripening process some individual molecular features were found to 512 decrease in their relative abundance. For example, 5'-methylthioadenosine, a molecule, which 513 can be used to produce ethylene, a key ripening hormone for plants (North,Miller,Wildenthal,514 Young, & Tabita, 2017) was found to decrease significantly in its relative abundance over the 5 515 day time course. Also plant flavonoids (including level 3 annotation of naringenin) and 516 tomatidine, a tomato-specific alkaloid, were found to decrease significantly in their relative 517 intensity over time. This is informative as many of the healthy properties assigned to 518 polyphenol-containing food are attributed to molecules like naringenin and it indicates that the 519 nutritional value of tomatoes may change over the time period we typically store tomato fruits at 520 home. process from milk to yogurt across different yogurt brands used as inocula, as well as the milk 529 as control. d) Metabolites increasing or decreasing significantly during the fermentation process 530 across different home ferments. Metabolite annotation was performed through mass spectral 531 molecular networking and spectral matching to reference spectra as indicated below. 532

Molecular networking and annotations: 533
To further explore specific molecules and molecular changes within each food type, we 534 subjected all LC-MS/MS data to mass spectral molecular networking. Mass spectrometry of the 535 tomato samples (120) resulted in 71,430 MS/MS spectra, 62,263 passed the filtering for a 536 minimum of 4 ions and a minimum of two identical MS/MS spectra in the data set and this 537 condensed to 2,611 unique spectra that are presented as nodes (Supplementary Figure 5). 212 538 of the nodes had an annotation. This is an 8.1% annotation rate and with an FDR for spectral 539 matches estimated using Passatuto to be 4.8% (Scheubert et al., 2017). All annotations are 540 level 2 or 3 according to the 2007 metabolomics standards initiative (Sumner et al., 2007). For 541 the milk to yogurt analysis, the 126 samples resulted in 78,203 MS/MS spectra, 63,241 passed 542 the minimal requirement of four ions and minimum of two identical spectra (Supplementary 543 Figure 6). Post clustering identical spectra, 4,142 nodes remain. 147 of the nodes had spectral 544 matches against the libraries (3.5% annotation rate, FDR 0.5%). The coffee analysis included a 545 total of 146 samples that resulted in a total of 50,929 MS/MS spectra. After filtering, 42,752 546 MS/MS spectra remained that condensed to 1,460 unique spectra in Supplementary Figure 7.

547
Of the 1,460 unique spectra, 72 had spectral matches to the reference libraries within a cosine 548 of 0.7. This is a 4.9% annotation rate and with an FDR estimated to be 0.09%. The meat 549 analysis included 119 samples, resulting in 72,083 MS/MS spectra, 54,663 of which passed the 550 filtering step (Supplementary Figure 8-9). Merging all identical spectra resulted in 5,035 unique 551 spectra of which 313 were annotated (6.2% annotation rate, FDR 1.5%). Finally, the tea 552 analysis had 185 samples resulting in 50,547 MS/MS spectra, 44,505 of which passed the 553 filtering (Supplementary Figure 10-11). After merging identical spectra, 1,834 unique MS/MS 554 spectra comprised the molecular network with 207 annotations (11.2% annotation rate, FDR 555 0.2%). 556 MS/MS belonging to the internal standard sulfadimethoxine was observed in all analyses 557 and correctly annotated through GNPS library matching. Molecules annotated as the amino 558 acids tryptophan and phenylalanine as well as phospholipids were widely distributed across all 559 samples and different food types. Other putatively annotated molecules were found to be food-560 specific. 561 In the tomato samples (Figure 5a) Figure 1), while other molecules increase. Only in the sun dried tomatoes did we observe a 574 spectral match to glucose, perhaps added as a sweetener. In both sun dried and fresh tomatoes 575 we detected azoxystrobin, a fungicide used as protectant against fungal diseases in agriculture. 576 In milk and yogurt, matches to six carbon sugars, disaccharides and oligosaccharides, 577 vitamins and acylated carnitines were observed (Figure 5b). In addition, large lipid molecular 578 families, such as sphingolipids, and glycerol conjugated with fatty acids such as monoolein and 579 linoleoylglycerol were annotated. Delvocid, also known as the clinically used antimicotic 580 natamycin, which is a known additive used to preserve dairy products from fungal growth, was 581 detected (Branen, Davidson, Salminen, & Thorngate, 2001), and did not change in relative 582 abundance over time. These annotations are all consistent with the animal, milk and yogurt 583 origin of the samples. However, we also obtained unexpected annotations. The bile acids 584 glycocholic acid and cholic acid formed an annotated molecular family. These were not 585 expected to be observed as they are primarily associated with the gut. Although level 3 586 annotations, manual inspection of the ions and retention time analysis reveal the data are 587 indeed consistent with these bile acids. 588 In coffee (Figure 5c) we observed caffeine as well as methyl-caffeine and a related 589 compound with a delta mass of m/z 14.01 (CH 2 ), corresponding to theobromine. Furthermore, 590 we detected several flavonoids and a large number of hydroxycinnamic acids and chlorogenic 591 acids, which are commonly observed in plants (Islam MT, et  corresponding to mass shifts associated with six carbon sugars, oxygen, and water, 598 respectively. 599 In the meat samples (Figure 6a), we observed MS/MS matches to tetracycline displayed 600 as a single node (no related spectra were detected), which were more abundant in the turkey 601 samples. Although tetracycline is commonly used as a growth promoter, here it was added to 602 see the effect of this antibiotic on a 5 day food spoilage test (Granados-Chinchilla & Rodríguez, 603 2017). We also have spectral matches to carnosine as well as a large cluster of acyl carnitines 604 with five spectral matches to different acylations. The acyl carnitines are predominantly 605 observed in beef when comparing to turkey. We also found a family of N-acyltaurines (NATs), a 606 recently discovered class of lipids (Turman, Kingsley what we would expect to observe in these sample types and we observed changes in molecular 620 compositions emerging after 2 days of storage but only for a small number of molecules. 621 A large range of phytochemicals were annotated in the tea samples (Figure 6c and  622 Supplementary Figure 10)  Note that the majority of nodes for this family were annotated with GNPS community contributed 632 library hits, indicating that for some compound classes library coverage is increasing due to the 633 growing publicly available spectra. As with coffee, caffeine was annotated in the tea samples as 634 well. Theaflavin, a polyphenol formed during fungal oxidation and its analogues often associated 635 with black tea (Zhang et al., 2018), were detected in white, green, black and oolong tea 636 samples, and as seen in Figure 4 and in molecular networks can be found as supporting information (Supplementary Figures 8-10) 662 and the GNPS links to the analysis jobs are provided in the data availability section. All 663 annotations shown are level 2 or 3 according to the 2007 metabolomics standards guidelines 664 (Sumner et al., 2007). 665 666 Discussion: 667 The untargeted mass spectrometry approach coupled with molecular networking allowed 668 us to assess large scale differences between sample type, find molecule-molecule links within 669 and between sample types, and identify different compound classes found within a sample type 670 -all useful for biochemical interpretations. We determined that different foods undergo different 671 molecular changes over time, exemplified by the tea and yogurt time courses. Furthermore, 672 mass spectral molecular networking could identify key metabolites, which differed based on 673 processing type, such as fermentation time in the yogurt samples and brewing time for tea. The 674 heatmaps as well as the molecular networks, while very different visualization techniques, 675 confirm and support each other. For example, theaflavin increases significantly in relative 676 abundance over time which can be visualized in the heatmap (Figure 4a,b) as well as the 677 molecular network displaying brewing time (Supplementary Figure 11). In Figure 6c one can 678 see that theaflavin is connected to two unannotated compounds, which allows us to understand 679 more about the compound family without understanding the exact identity. Furthermore, foods 680 within a group were found to undergo differential molecular changes over time, exemplified by 681 the extraction kinetics of oolong tea deviating from the rest of the tea samples. Two potential 682 explanations for the observed changes in extraction kinetics for tea are hypothesized. The 683 oolong tea bags might affect extraction kinetics; however, the observed differences were 684 observed in both manufacturers' brands. The second hypothesis is that the extraction kinetics of 685 oolong tea is different from those of other teas, which might result from the extensive drying, 686 physical changes of the leaves (e.g. twisting/curling), and oxidation. The molecular composition 687 of the teas changed over time, with observed patterns mainly consistent with continued 688 extraction of molecules as opposed to chemical modifications. While a range of compounds 689 increased in many of the tea types, there were signatures specific to tea type, such as the 690 increase in the relative abundance of coniferyl aldehyde only in oolong tea. 691 The changes observed are in contrast to the yogurt samples, where chemical alterations 692 over time vary significantly, likely due to the microbial activity. We can detect significant 693 changes in PCoA, molecular networks and heatmaps. In the PCoA, the home ferment 694 inoculated with Kroger yogurt resembled the original starting culture at a molecular level, and it 695 differed from the other home ferments, possibly because it contained a different set of yogurt 696 cultures. Interestingly, significant changes over time are not observed in the heatmap, when we 697 focus our analysis on annotated compounds only (Figure 4 and Supplementary Figure 3), 698 indicating that many of the molecular transformations during fermentation are not yet 699 characterized or that the reference spectra are not present in the available MS library 700 databases. Consistent with the lack of reference spectra in the public databases, the yogurt and 701 milk samples also had the lowest annotation rate at 3.5%. Among the compounds that were 702 annotated we found a broad range of compounds (Supplementary Figure 6), including food 703 additives and sugars, which are also found in other milk types within publically available 704 datasets on GNPS, such as breast milk. 705 One of the questions of interest was whether the different origins of tomatoes could be 706 distinguished on a chemical level. Expectedly, the processed tomatoes (canned and sun-dried) 707 were significantly different from fresh ones. Many molecules including added oils, sugars and 708 preservatives explain these differences. However, differences between fresh fruit are also 709 noticeable in the PCoA data. The private garden-grown tomatoes were used as an ideal case 710 scenario -these fruits were naturally grown, ripened on the vine and have not been treated with 711 any pesticides/herbicides. In PCoA space, farmer's market tomatoes most closely resemble 712 home-grown ones, while the store-bought tomatoes were all similar to each other. Also, different 713 brands could be distinguished. It is likely that the close similarity of garden and farmer's market 714 tomatoes results from similar treatment where the fruits are ripened on the vine and collected 715 and sold without any processing (this is known for the garden tomatoes and presumed for the 716 farmer's market ones). Conversely, the store-bought tomatoes are collected at an early stage, 717 often not fully ripened for ease of transportation, transported over long distances and treated 718 with exogenous ethylene (depending on the supplier). This appears to have a more significant 719 effect on the chemical composition of tomatoes than the "organic" designation. 720 Another question of interest was whether organic designation and the addition of an 721 antibiotic would impact meat spoiling over time. While the largest difference was that of beef 722 versus turkey, there are some minor trends that can be observed over time in the PCoA plots. 723 Because there was a large within intra-day sample variation, specific major trends could not be 724 detected with respect to organic or antibiotic addition. The data as well as visual inspection of 725 the meat indicate that there were non-uniform chemical transformations, possibly related to the 726 surface area and exposure to air. When the data from each time point is merged, as done with 727 molecular networking, and assessed for presence and absence of spectra, there are hints that 728 there are few low intensity molecular clusters, including oleoyl-taurine and acetyl-carnitine, that 729 change in both the molecular networking data and in a time-dependent manner 730 (Supplementary Figure 2). These were different for the different meat classification of organic 731 vs non-organic. Similarly the effect of tetracycline is observed in both the turkey and beef, but 732 only affects few molecules within the 5 day experiment. We expect that these observations are 733 just the tip of the iceberg that warrant further investigation. Future studies can further utilize the 734 mass spectral molecular networking data, with the ability to propagate annotations across a 735 network, to better understand the effect of time on spoilage. 736 Finally, we assessed the effect of roasting type of coffee across solid "ground" coffee as 737 well as liquid "brewed" coffee. The largest molecular changes were observed between the liquid 738 and solid samples, whereas, comparatively the roasting type only displayed minor molecular 739 changes. This finding suggests that extraction method has a larger effect on the molecular 740 composition of coffee than processing type such as roasting. Alternatively, molecular changes 741 induced by roasting might be predominantly observed in volatile components, not assessed in 742 this study. Indeed, changes in smell between the different roasting types could be readily 743 perceived. Further analyses which address aromatics, such as GC-MS, would be needed to 744 confirm this hypothesis. 745 In summary we have created five unique data sets that enable the molecular 746 assessment of five common foods and beverages, which are connected by frequently used 747 handling and processing practices. We show that the combination of molecular networking, and 748 multivariate statistical methods such as PCoA and heatmaps and univariate statistics 749 (correlation, significance testing) can be used to explore the molecular composition and the 750 effect of different processing methods, different products and storage conditions relative to all 751 other samples in the study or group. The data sets provided here serve as a reference data set 752 that can continue to be mined. One exemplary feature of the GNPS molecular networking 753 workflow is the search parameter 'Find Related Datasets'. As exemplified in this study, even the 754 most traditional food types contain a large number of unannotated molecules, therefore, we 755 expect that the increasing deposition of mass spectrometry datasets in the public domain would 756 allow the comparison to other complex mixtures and to narrow down the origin of molecular 757 features. In this spirit the projects are publicly available in GNPS (Wang et al., 2016). Anyone 758 who wishes to continue to learn about these data sets can subscribe to the projects as they will 759 be subjected to living data analysis. Living data is a strategy introduced in Wang et al., 2016  760 where the data is continuously reanalyzed and updates are provided automatically to all the 761 subscribers of these data sets. As 88-97% of all the signals are currently unannotated, we will, 762 as a community, continue to increase our knowledge about the molecular composition and 763 changes of our food.  samples. Top is non-organic turkey (Kroger brand). Bottom is turkey grown without antibiotics 999 and growth hormones (Empire brand). 1000 1001