Addition of chocolate milk to diet corresponds to protein concentration changes in human saliva

Abstract

Salivary proteins have the potential to alter oral sensory perception of foods. In rodents, dietary polyphenol exposure increases salivary concentrations of polyphenol-binding proteins and several cystatins, which correlate with less aversion to polyphenol-rich solutions. If similar salivary shifts occur in humans, then increasing dietary polyphenols may improve orosensory experience of polyphenol rich foods. We hypothesized that small dietary changes, focused on polyphenols, would increase expression of salivary binding proteins for polyphenols and thus suppress unpleasant polyphenol sensations. However, analogs of salivary polyphenol binding proteins are found in foods. Thus, we also hypothesized that food-sourced analogs of the salivary proteins would mitigate changes in saliva and sensation. Human subjects (N=55) alternated weeks of consuming a low polyphenol diet and then a regular diet plus a polyphenol-rich chocolate milk (almond, containing no polyphenol binding proteins, or bovine, containing polyphenol binding proteins). Statistical analyses revealed both chocolate milk interventions corresponded to changes in relative expression of 96 proteins and calculated concentration of 146 proteins (both after correction for false discovery rate), out of 1,176 proteins identified through proteomics. Of the proteins that changed, proline-rich proteins and cystatins were noticeable, which reflects prior work in animal studies. Subjects rated all chocolate milks as less flavorful after the bovine chocolate milk intervention week compared to low polyphenol weeks, but generally sensory changes were minimal. However, the results confirm that dietary changes coincide with salivary changes, and that some of those changes occur in proteins that have potential to influence oral sensations.

Introduction

Polyphenols can be beneficial for human health, as they protect against cancer, neurodegenerative disease, type 2 diabetes, and cardiovascular disease (Aron & Kennedy, 2008; Cory et al., 2018). These compounds are abundant in the food supply, particularly in green, leafy vegetables (kale), tea, fruits (blueberries and blackberries), wines, and dark chocolates. However, polyphenols cause astringency, (a dry, rough, puckery sensation on oral surfaces) and bitterness, both of which are generally negative sensory qualities (Lee & Lawless, 1991; Preys et al., 2006). Due to these unappealing sensations of polyphenols, the food industry historically treated high polyphenol content in foods as a defect, and some companies would remove polyphenols from food products or pursue plant varieties with lower polyphenol contents (Drewnowski & Gomez-Carneros, 2000). Conversely, recognition of polyphenols as health compounds has made increasing polyphenol content in foods more desirable. Yet whether effective doses for health benefits can be reached without making foods unpalatable is questionable. Thus, while increasing polyphenol content in the diet could be beneficial for human health, many people would reject polyphenol-rich formulations of foods in favor of more palatable, lower polyphenol alternatives (for examples with chocolate, see Harwood, Ziegler, & Hayes, 2012, 2013).

Nonetheless, some people enjoy foods rich in polyphenols. In fact, bimodal patterns of likers and dislikers are common for teas, dark chocolate, red wines, and many other polyphenol rich foods (Borchgrevink & Sherwin, 2017; Harwood et al., 2012, 2013). Since sensory appeal is a main driver of food choice (IFIC Foundation, 2018; Maarsman, 2016; Steptoe et al., 1995), increasing palatability of polyphenol-rich foods could increase their consumption, especially for individuals currently in the “disliker” category. Certainly, data already indicate that exposure to foods improves palatability (Fildes et al., 2014; Holley et al., 2015; Liem & de Graaf, 2004; Spill et al., 2019). However, animal data indicate that for bitter polyphenols and alkaloids, exposure may alter saliva in ways that suppress the negative sensory quality of these molecules (Martin et al., 2018, 2019; Torregrossa et al., 2014). If also true in humans, then increased consumption of polyphenols in the diet could provide a method to make bitter and astringent foods more palatable through changes in oral physiology, in addition to the centrally mediated (i.e., brain) effects from increased familiarity or post-ingestive feedback.

We designed this experiment as a counterbalanced, crossover design, where participants alternated weeks of lower and higher polyphenol consumption. At the end of each week, we collected saliva for proteomic analysis and sensory ratings for sweetness, sourness, bitterness, dryness/roughness, overall flavor, and overall liking of polyphenol-rich beverages. The beverages were chocolate almond and bovine milks. We chose these two, very different types of “milk” in order to test whether the delivery matrix for the chocolate could influence the type and degree of salivary proteomic changes observed, as food proteins such as bovine milk casein bind polyphenols though similar mechanisms to salivary polyphenol-binding proteins (Dumas et al., 1972; Huang et al., 2017; Jöbstl et al., 2006). Thus, the intervention stimuli we used included: chocolate almond milk (no binding proteins) and chocolate bovine milk (naturally occurring binding proteins). Originally, we included an almond milk with gelatin (which also binds polyphenols) as well, but preparation and appearance of this milk was inconsistent, and so those data were excluded as unreliable.

While chocolate milks are complex, we designed this study with the intent of establishing whether small, realistic dietary adjustments of polyphenol content would result in salivary protein changes and alterations in the sensory experience of polyphenols. We wished to ascertain whether, in the noisy environment of a typical human diet, such small changes targeted at polyphenol content would have any effects on the outcomes of interest. Certainly, this means that multiple confounding factors influence our outcomes, but this is the first study to attempt to change saliva through diet. Thus, we elected to use realistic beverages in order to maximize practical applicability. Moreover, we chose to add these beverages to people’s regular diets, rather than provide the entire diet, allowing us to observe whether any salivary changes are large enough to be observed through a noisy background. This study was designed not to assess causality of the chocolate milks on saliva. Rather, we wanted to observe if salivary protein concentrations would change at all in response to diet, and whether those changes would be in directions as expected per the literature on polyphenols influencing saliva in animal studies. Additionally, we wished to establish preliminary evidence of whether the matrix in which polyphenols are delivered would influence any salivary protein changes. Further work will be required to establish mechanistic and targeted cause-effects relationships, but this is the first work to investigate how small changes in human diets relate to concentrations of salivary proteins, particularly those of interest for oral sensations.

Compared to low polyphenol consumption weeks (“washout” weeks), we hypothesized that subjects would

1. Increase content of salivary binding proteins (particularly proline-rich proteins) and S-type cystatins after drinking chocolate almond milk for one week, with lesser changes after drinking chocolate bovine milk.

2. Give lower dry/rough (a simplified term for astringency) and bitterness intensity ratings of the chocolate beverages after drinking chocolate almond milk for one week, with lesser changes after drinking chocolate bovine milk.

We also expected that subjects would give lower bitterness/astringency ratings to the chocolate bovine milk in general compared to chocolate almond milk, regardless of the intervention. Additionally, we explored concentrations in saliva for other proteins of potential interest for orosensation, including other cystatins (Dsamou et al., 2012; Morzel et al., 2014), amylases, mucins (Davies et al., 2014; Engelen et al., 2007; Fabian et al., 2015), lipocalins (including lipocalin-1/von Ebner’s gland protein; Kock et al., 1994; Schmale et al., 1993), and carbonic anhydrases (including carbonic anhydrase 6/gustin; Barbarossa et al., 2015; Calò et al., 2011; Fabian et al., 2015; Feeney & Hayes, 2014).

Materials and Methods

All methods were approved by the Purdue University Institutional Review Board, and all subjects provided written informed consent. This study was listed in clinicaltrials.gov under NCT03501238.

Subjects

All subjects were recruited through a combination of online and print recruiting materials on the Purdue campus and from the Saliva, Perception, Ingestion, and Tongues (SPIT) lab participant pool. Exclusion criteria included individuals who were current smokers; had nut, dairy, or other food allergies or sensitivities; had known problems with taste or smell; had piercings in the cheek/lips/tongue; had issues with choking or swallowing; had too much or too little saliva production (intended to screen for dry mouth or excessive salivation as can be seen due to certain neurological conditions; however, note that no participants were excluded by this criteria); or were under 18 or over 65 years of age. In total 64 participants (25 male, 39 female, 0 other) were recruited and completed the study, with an average age of 24 years (range: 18–60). Average BMI was 24.6 kg/m² (range: 17.2 – 40.7 kg/m²). Full details on subjects are provided in supplemental files.

Study Procedures

Subjects were recruited to attend an initial visit before being invited to participate in the 6-week study. During the initial visit subjects were briefed on study details, given samples of the beverages of the study to taste along with exemplars for astringency (tannic acid, malic acid and alum), and practiced saliva collection. Full details on this visit can be found in the supplemental files.

Critically, we did not train subjects on the sensations of bitterness and astringency, as training would require repeated exposure to bitter and astringent stimuli. Per our hypotheses, this could fundamentally change oral physiology and salivary binding protein concentration. As no data are currently available regarding any potential changes in saliva due to orosensory exposures, let alone details on the timeline of those changes to appear or to return to baseline, it was not possible to train participants to correctly identify the sensory modalities they experienced. Thus, subjects were exposed to the tannic acid, malic acid, and alum samples during this initial screening/qualifying visit to familiarize them with the sensations of interest, but further training was not conducted. We also used the term “Dryness or roughness,” as these were the astringent sub-qualities of interest for our work and because our preliminary tests indicated this verbiage term may be easier to understand than the word “astringency” (Lawless & Corrigan, 1994). A pilot project tested bovine milk, two commercial almond milks, and one homemade almond milk, all of which were either chocolate flavored or just flavored with sugar and brown food coloring. In those data, untrained participants could discriminate astringency intensity among milk types (commercial almond, homemade almond, and bovine) as well as brown vs. chocolate flavors. Additionally, the term “dryness or roughness” did not change these ratings compared to “astringency (dry, rough, or puckering)”. Those results, and a power analysis for the current study using those data, are in the supplemental file. Per those data, we concluded we could adequately capture differences in astringency using untrained participants. Finally, by asking about multiple sensations, both pleasant and unpleasant, we increased our likelihood to document a sensory change, even if that change was ascribed by participants to the wrong modality (i.e., astringency is bad, and so untrained participants may assign astringency sensory intensity into other ratings typically associated with negative affect, such as bitterness or sourness, rather than “astringency”; more details on the sensory ratings in subsequent paragraphs).

After completing the initial screening/qualifying visit, qualified participants who were still interested enrolled in the 6-week intervention study. Over the course of the 6-week study, subjects alternated weeks of avoiding polyphenol rich foods (washout weeks) and weeks of consuming three 236 mL (8 fl oz) glasses of chocolate beverage per day (intervention weeks, one week per each beverage type; order of beverages was counterbalanced so that all beverages were tested first, second, or third an equal number of times, and all beverages were tested before/after all other beverages an equal number of times). A visual representation of the study design is shown in Fig 1. A complete list of ingredients and sources used in this study is available in Table 1. Note that while almonds do contain endogenous polyphenols, most of those polyphenols are found in the almond skins (Milbury et al., 2006). These polyphenols generally do not make it into almond milk in substantial quantities. Indeed, the amount of polyphenols in almond milk is actually low enough to be undetectable by common assays (Plank et al., 2012).

Figure 1:

6-Week Study Design

Table 1.

List of Products Used

Product		Company	Company Location
Ingredients in sensory samples	Unsweetened Original Almond Breeze® almond milk	Blue Diamond®	Sacramento, CA
	1% Bovine Milk	Various Brands
	Spring Water	Culligan®	Rosemont, IL
	White Sugar	Various Brands
	100% Unsweetened Dark Chocolate Cocoa	Scharffen Berger Chocolate Maker®	Robinson, IL
	Malic Acid	Milliard™	Lakewood, NJ
	Tannic Acid	Sigma Aldrich	St Louis, MO
	Alum Powder	Hoosier Hill Farm	Fort Wayne, IN
Reagents in protein analysis	Acetone	Fisher Scientific®	Pittsburgh, PA
	Urea	Fisher Scientific®	Pittsburgh, PA
	Formic Acid	Sigma Aldrich®	St. Louis, MO
	DL-Dithiothreitol	Sigma Aldrich®	St. Louis, MO
	Ammonium Bicarbonate	Fisher Scientific®	Pittsburgh, PA
	Iodoacetamide	Sigma Aldrich®	St. Louis, MO
	Trypsin/Lys-C Mix, Mass Spec Grade	Promega®	Madison, WI

Subjects were provided a list of common polyphenol-rich foods to avoid during washout weeks. The list included: red wine, tea, cocoa powder and dark chocolate products, dark colored berries (blueberries, blackberries, raspberries), plums, oranges, parsley, red cabbage, kale, radishes, and capers.

During intervention weeks subjects were instructed to drink three 236 mL servings of the provided chocolate beverage at routine times around breakfast, lunch, and dinner. A portioned, reusable plastic cup was provided to each subject, and they were instructed to use the cup to measure and drink all servings. Milks had the macronutrient distribution shown in Table 2, per their Nutrition Facts panels, and the interventions were prepared per the formulas that follow. The macronutrient contents of these milks differed, which is unavoidable due to inherent differences in the products. However, the 1% bovine milk was intentionally selected to be as close in fat content as possible to the almond milks. Milks were portioned into milk jugs, refrigerated, and distributed to subjects within 24 hours of preparation. Subjects picked up milks twice per week. Subjects were provided with the exact amount of milk needed for the study and instructed to keep the milk refrigerated. Subjects were not informed which beverage they were consuming each week, though many of them could likely tell the difference between the flavors of the bovine and almond milks. The total milk we asked subjects to consume (3×236 ml = 708 mL, 24 fl oz) is the recommended amount of servings of milk per day per the Dietary Guidelines for Americans (USDA & HHS, 2015), and the total daily volume we gave participants is similar to the size of “large” milk-based beverages at popular US coffee and beverage restaurants such as Starbucks®.

Table 2.

Milk composition and formulations

Composition (per nutrition facts panels; per one serving of 236mL [8 fl oz])
	Almond milk	Bovine milk (1% fat)
Calories	30	103
Total Fat  Saturated fat  Trans fat  Polyunsaturated fat  Monounsaturated fat	2.5g  0g  0g  0.5g  1.5g	2.4g  1.5g  0g  0.1g  0.7g
Total carbohydrate  Dietary fiber Sugars	1g  1g  0g	12g  0g  13g
Protein	1g	8g
Intervention formulations (percents are weight/weight)
Milk Type	Milk (%)	Cocoa (%)	Sugar (%)
1% Bovine	93.5	2.5	4
Almond Breeze	93.5	2.5	4

At the end of each washout and intervention week, subjects provided saliva and completed sensory evaluations of all milks. RedJade® (Redwood City, CA) sensory software was used for data collection and on-screen prompts. All testing was done under normal lighting with subjects seated at tables with cardboard dividers between stations. During sensory testing subjects were provided a tray with samples in counterbalanced order, and an iPad mini™ 2 (Apple, Cupertino, CA) that led them through study instructions and sample intensity ratings. Subjects were asked to abstain from eating or drinking for 1 hour prior to sensory testing. At the beginning of the tasting portion of the study, subjects were prompted to rinse their mouths with spring water (Culligan Spring water, purchased locally in 19 L (5 gal) jug as used in water coolers). Next, saliva was collected by swishing (10 s) and expectorating 15 mL of spring water. This expectorated water + saliva sample was placed on ice. We stimulated saliva in this way because we wanted to replicate the experience of drinking, and typically salivary protein content is high enough that it requires dilution for use in BCA assays (BCA standard assay works to 2mg/mL, typical stimulated saliva contains 4mg/mL protein). Thus, using the water to both stimulate oral surfaces and dilute the sample was a practical approach that we successfully pilot tested prior to beginning the study. In hindsight, some participants had very low protein content in these expectorated water samples, so we would recommend future researchers use less water and/or add a mild taste as an additional salivary stimulus.

A 90 s break, which included time to rinse with spring water, was enforced between the saliva collection and the tasting portion of the study. The tasting portion of the study included all three chocolate milk beverages, presented in counterbalanced order and contained in clear cups with clear lids. Samples were provided in two cups of 15 mL each (30 mL total). The first 15 mL cup of each sample was swished and expectorated while the second of each pair was swallowed and rated for the samples’ overall intensity, dryness or roughness, bitterness, sweetness, sourness, and overall liking on inset generalized visual analog scales. While we were most interested in bitterness, dryness or roughness (astringency), and overall liking, we included the other terms in order to avoid “dumping effects,” where a participant augments or suppresses a rating for the incorrect modality because the sensory questionnaire did not ask about a different modality that the participant experienced (Lawless & Heymann 2010). For example, a vanilla flavored milk may be rated as sweeter if a questionnaire only asks about sweetness intensity, but not vanilla flavor intensity. Intensity ratings were given directly on iPad mini™ 2 screens. The intensity scales were denoted from “None” to “Strongest Ever.” These labels on the scales were indented by 10 points, with a total scale length of 120pts. The overall liking scale was the same except the anchors were “Worst ever” and “Best ever” (inset by 10 pts from the end of the scale), with an additional anchor “Neutral” in the exact center. Examples of these intensity scales are included in the supplemental files. A 90 s break was enforced between milk types, during which subjects rinsed with spring water. These visits lasted between 15–20 min. Expectorated samples (kept on ice for the entire visit) were collected by research staff and three 200 μL aliquots of each sample were saved in microcentrifuge tubes. Aliquots were flash frozen using a CoolRack M90 (Corning©, Corning, NY) on dry ice. Samples were stored at −80 °C until preparation for proteomic analysis. Our preliminary work indicates that flash freezing samples before centrifuging yields greater protein content, and other pilot data indicate that proline-rich proteins in particular are found at greater concentrations when samples are frozen immediately, rather than centrifuging first (Davis & Running 2019). Subjects were scheduled for approximately the same time of day for each visit in order to minimize circadian effects on salivary flow rates. Aliquotted samples were weighed in order to be able to calculate absolute concentration of protein in the expectorated samples, but the total expectorate was not weighed.

During chocolate milk consumption weeks, participants came to the lab twice per week to pick up the milk. This ensured frequent contact with the researchers.

Proteomic Analysis

The expectorated water+saliva samples were used for proteomic analysis. By analyzing the protein concentration in this mixture, we were able to determine the salivary protein concentration in the oral cavity after the stimulus mixes with saliva. This mixture is what would interact with receptors for gustatory, textural, and chemesthetic sensations as well as what would influence the release of aroma compounds for retronasal olfaction. Thus, while we refer to “salivary protein concentrations” throughout this report, those concentrations are not of pure saliva. Rather, they are of the mixed saliva and water that would have been in each person’s mouth when sampling the water. This was intended to allow us to better approximate the salivary protein concentration that would be present when sampling the chocolate milks, which were presented with the same instructions as the water used to collect saliva (the salivary protein in the actual milk + saliva mixtures could not be tested as the concentration of milk proteins overwhelms the salivary proteins, making them pass under the detection limit for our proteomic approach; we verified this in a preliminary analysis).

Saliva samples from before (end of washout week) and after (end of milk consumption week) the chocolate bovine milk and almond milk testing weeks were subjected to proteomic analysis to determine salivary protein changes between washout and intervention weeks. Notably, despite keeping the samples cold during and after collection, several samples showed signs of gels/films forming within the saliva (anecdotally due to mucins gelling, a common problem in saliva samples). The gels/films interfered with pipetting of the samples, leading to wildly inconsistent protein content within the three technical replicates from the same aliquots (coefficients of variation as high as 70%). These samples, along with samples that had very low protein content after preparation (<0.6 μg, too low for our LC/MS/MS setup) and samples whose partner sample was gelled or low protein (i.e., the bovine sample gelled so the washout sample preceding that week was not analyzed) were excluded from the analysis (total 50 samples not analyzed out of 242 collected). Total, 192 samples were prepared for proteomic analysis.

Proteomic preparation and analysis was conducted as previously described by (Aryal et al., 2017). Briefly, proteins were precipitated in cold acetone, dried, resuspended/denatured in 8 M urea, and analyzed in triplicate for total protein content by bicinchoninic acid (BCA) assay (Thermo Scientific Pierce BCA Protein assay). Cysteines were reduced using dithiothreitol and alkylated with iodoacetamide before protein was digested with trypsin/lysC. Protein samples were cleaned using C18 spin columns and analyzed by reverse-phase HPLC-ESI-MS/MS as described by (Aryal et al., 2017). A Dionex UltiMate 3000 RSLC Nano System coupled to the Q Exactive™ HF Hybrid Quadrupole-Orbitrap Mass Spectrometer (Thermo Scientific, Waltham, MA) was used for analysis with an Acclaim™ C18 PepMap™ 100 analytical column (Thermo Scientific, Waltham, MA). Sample preparation (a 3–4 day process) was conducted in our lab, while LC/MS/MS work was done by the Purdue Proteomics Facility. Detailed instructions for the preparation procedure for the protein samples, with annotations of where issues are likely to arise, are included in the supplemental files.

Cocoa Polyphenol Analysis

Catechin and epicatechin content of the cocoa powder used to formulate our milks was analyzed by UHPLC/MS/MS. Two lots of cocoa powder were analyzed, in duplicate. Full methods and results for this procedure can be found in the supplemental files. The average epicatechin and catechin content of the cocoa powders were 5.14 mg/g and 2.96 mg/g, respectively. Per the formulation in Table 2, each serving of chocolate beverages would contain approximately 31.1 mg epicatechin and 17.9 mg catechin, for respective totals of 93.3 mg and 53.7 mg per day. Note these totals include just these monomeric polyphenols, which are more important for bitter taste, and does not account for the oligomeric polyphenols, which are more important for astringency (Robichaud & Noble, 1990). However, protocols for analyzing oligomeric polyphenols are much more complex (Neilson et al., 2016) and beyond the capabilities of our current analytical setup and scope of this work.

Data Analysis

MaxQuant 1.6.5 software (Cox et al., 2011; Cox & Mann, 2008) was used to analyze the digested protein spectra obtained from the mass spectrometer. The software identifies the peptides found in the digest and maps those to the groups of peptides that would be expected from digesting the proteins that are encoded by the human genome. Thus, while we will discuss these results as identified “proteins,” they are actually groups of peptides that are linked to likely proteins from the human genome that would yield those peptides. The parameters used in the MaxQuant setup are also included in supplemental files. The fasta file used to match the identified peptides to proteins in the human genome was downloaded from UniProt (https://www.uniprot.org/) on May 8^th, 2018 and is available upon request.

Protein groups reported from the MaxQuant software were analyzed for their relative intensity by calculating the ratio of that protein’s spectral intensity divided by the total spectral intensity of that sample (a typical proteomics approach). We also analyzed proteins by a calculated, quasi-absolute concentration. By taking the ratio of spectral intensity for a protein (a mass based-ratio of the total) and multiplying it by total protein content from the BCA assay (a mass-based concentration), we calculated what we are calling “Calculated Concentrations.” This method will not perfectly represent absolute concentrations in mg/mL of specific proteins in saliva, as different proteins are digested with varying effectiveness through our protocol and as different peptides may be detected with differing efficiency by the mass spectrometer. However, this value will hopefully offer new insight into the actual chemical concentration of different proteins in saliva. This is a very different concept than a particular protein’s concentration as a ratio of total protein, especially as different individuals can have very different total protein concentrations. As we are interested in relating salivary proteins to their chemical activities, potentially in modulating orosensory sensations, a concentration of protein is potentially more interesting than a ratio of the total protein.

All zero values for spectral intensity, ratios of spectral intensity, and our calculated concentrations were replaced with half-minimum values, as these zeros are not necessarily true zeros, but rather may have been below the limit of detection for the mass spectrometer.

Statistical analysis

Changes in relative (“Ratio of Spectral Intensity”) and quasi-absolute (“Calculated Concentration”) protein concentrations from washout to testing weeks were analyzed using Python 3 and SAS 9.4 (Cary, NC) in Jupyter Lab (https://jupyter.org/). In the SAS code, the proc mixed procedure was used to generate and test linear mixed models. For Ratios of Spectral Intensity, the distribution of residuals in our models was not improved by transformations, so the raw, original values were used. For Calculated Concentrations, residuals were more normally distributed when a logarithm (base 10) was applied.

Proteomic data were analyzed for each protein using the model statements:

$$\text{Ratio}\text{of}\text{Spectral}\text{Intensity}=\text{Intervention}\text{WeekOrderGp}\text{Age}\text{Gender}\text{Intervention}*\text{WeekorderGp}$$

$${\text{log}}_{10}(\text{Calculated}\text{Concentration})=\text{Intervention}\text{WeekOrderGp}\text{Age}\text{Gender}\text{Intervention}*\text{WeekorderGp}$$

“Intervention” is whether the saliva is from the end of a washout week, bovine milk consumption week, or almond milk consumption week. “WeekOrderGp” is the order in which that type of milk was tested. In other words, if a participant was assigned the bovine milk intervention first, almond milk second, and almond milk with gelatin third, the week order groups would be 1, 2, and 3 respectively with the washout weeks preceding each of those also assigned 1, 2, or 3. This week order group value was treated as nominal data, as we did not know whether any order effects would be linear in nature. Subjects were included in the model as a repeated factor. The Kenward Roger approximation was used for degrees of freedom, and the autoregressive covariance structure was used (with data sorted by protein id, subject, and week order). The model was run individually for each protein, thus comparative models were not run. Only proteins that were found in at least 50% of all samples were analyzed. The Benjamini-Hochberg procedure was used to correct for false discovery rate (set at 0.1), given the large number of proteins we analyzed (total 244 found in at least 50% of all samples) (Benjamini & Hochberg, 1995).

Sensory data were also analyzed using SAS 9.4 using linear mixed models with the mixed procedure to determine differences in sensory intensity ratings among beverage types and intervention weeks. Our sensory scales had a range of 120 pts, and the bottom of the scale was assigned a value of 1 and the top a value of 121 (avoiding zeros and negatives, thus allowing for logarithmic transformations). The logarithm (base 10) of ratings was used for sensory intensity ratings to improve the distribution of the residuals. Residual distribution was acceptable for hedonic ratings, so no transformation was required.

We used the model statement (except for liking, which was not logged):

$${\text{log}}_{10}(\text{Rating})=\text{Beverage}\text{Intervention}\text{WeekOrderGp}$$

Where Beverage is the milk type being rated (chocolate bovine, chocolate almond). Other terms are as described above. Subjects were included in the model as a repeated factor. The Kenward Roger approximation was used for degrees of freedom, and the compound symmetry covariance structure was used (compound symmetry structure yielded AIC and BIC ~100 lower than auto-regressive structure across all modalities). Alpha was set to 0.05, and Bonferroni adjustments were used in the case of multiple comparisons.

We also ran analysis to see if changes in proteins were correlated with changes in sensation after the chocolate milk exposures. The code for that analysis can be found in the supplemental files, but data are not reported as no changes passed the Benjamini-Hochberg correction for false discovery rate.

OriginPro 2019 (Northhampton, Massachusetts, USA) was used to create figures with boxplots and data points.

Annotated statistical code, details on the manipulations required to pull the MaxQuant output into the Python and SAS analysis, and to apply the Benjamini-Hochberg procedure are in the supplemental files. Full results tables, with F values and degrees of freedom, can also be found in the supplemental files.

Results

Protein Changes

In this study, we found 146 proteins that changed in quasi-absolute quantification of concentration (Calculated Concentration) and 96 proteins that significantly changed their relative concentration (Ratio of Spectral Intensity), after applying the Benjamini-Hochberg false discovery rate procedure. Additionally, for calculated concentrations, 2 showed interaction effects of intervention and week order group; for relative concentration, no other effects were found after applying the Benjamini-Hochberg procedure. Full lists of these proteins, as well as the full output from the statistical analysis, are included in the supplemental files. Here, we will focus only on proteins that were of known interest for sensory outcomes, namely proline-rich proteins, cystatins (and cathepsins, which are inhibited by cystatins), amylase, carbonic anhydrases, and lipocalins/lipid binding proteins. The full list we used to look for these proteins of interest is included in the supplemental files. Fig 2 – 5 show the proteins of interest that differed by intervention. None of these proteins of interest passed the Benjamini-Hochberg false discovery correction procedure for age, week order group, or the interaction term of intervention*week order group. A few proteins of interest differed by gender, but as this was not a primary outcome, those figures are in the supplemental files.

Figure 2:

Salivary proline-rich proteins that showed significant differences in concentration among interventions. Top panels are by using calculated concentration, and bottom panel using ratio of total protein. Panels with same letter are the same protein by these different measurements. A: Profilin-1, B: Basic salivary proline-rich protein 3, C: Salivary acidic proline-rich phosphoprotein/Peptide P-C, D: Proline-rich protein 4, E: Proline-rich protein 27, F1/F2: Statherin

Figure 5:

Other proteins of sensory interest showing significant differences in concentration among interventions. Top panels are by using calculated concentration, and bottom panel using ratio of total protein. Panels with same letter are the same protein using analyzed by these different measurements. A1/A2: Salivary α-amylase, B1/B2: Fatty acid binding protein 5

Proline-rich Proteins

Intervention

For calculated concentrations (which are intended reflect a more “absolute,” mg/mL type quantity of protein, rather than being relative to total protein), profilin-1 (PFN1), basic salivary proline-rich protein 3 (PRB3), salivary acidic proline-rich phosphoprotein (peptide P-C, PRH1), proline-rich protein 4 (PRR4), proline-rich protein 27 (PRR27), and statherin (STATH) showed differences across intervention weeks (Fig 2). Excluding STATH, these proline-rich proteins generally were found at higher concentrations after chocolate bovine milk intervention week compared to washout weeks and chocolate almond milk intervention week. STATH, on the other hand, had highest concentrations after chocolate almond milk intervention week compared to the other week types. Looking at ratios of total spectral intensity, only statherin (STATH) showed significantly different relative concentrations, with higher abundance after washout weeks and chocolate almond milk intervention week compared to chocolate bovine milk intervention week (Fig 2).

Cystatins

Intervention

Observing the data for the calculated concentrations, cystatins-SN, SA, C, S, D, and A (CST1, 2, 3, 4, 5, and A respectively) all showed greater concentrations after chocolate bovine milk week compared to washout weeks, with S, C, and D also showing greater concentrations after chocolate bovine milk week compared to chocolate almond milk week (Fig 3). Cystatin-A (CSTA) was slightly different from the others, showing lowest calculated concentrations after chocolate almond milk week compared to bovine milk and washout weeks. For the ratios of total spectral intensity measures, cystatins SN, S, SA, and D were different across our interventions (Fig 3). For cystatin-SN and -S, the higher relative abundance was observed after chocolate almond milk intervention week compared to washout weeks. For cystatin-SA and D, higher relative abundance was observed after chocolate bovine milk intervention week compared to washout weeks, and higher abundance was observed for cystatin-D was also seen for chocolate almond milk intervention week compared to washout weeks.

Figure 3:

Cystatins that showed significant differences in concentration among interventions. Top panels are by using calculated concentration, and bottom panel using ratio of total protein. Panels with same letter are the same protein using analyzed by these different measurements. A1/A2: Cystatin-SN, B1/B2: Cystatin-SA, C: Cystatin-C, D1/D2: Cystatin-S, E1/E2: Cystatin-D, F: Cystatin-A

Notably, cathepsins B and Z, which are proteins inhibited by cystatins but not to our knowledge by the specific cystatins that we observed to change in saliva for this study, also showed altered calculated concentrations and ratios of spectral intensity (Fig 4). We observed lowest concentrations for cathepsin Z (CTSZ) after chocolate almond milk intervention week compared to both other week types. Cathepsin B (CTSB) also showed lowest concentrations after chocolate almond milk intervention week compared to both other week types for the ratio of spectral intensity values, but only showed lower calculated concentration after chocolate almond milk intervention week compared to chocolate bovine milk intervention week.

Figure 4:

Cathepsins showing significant differences in concentration among interventions. Top panels are by using calculated concentration, and bottom panel using ratio of total protein. Panels with same letter are the same protein using analyzed by these different measurements. A1/A2: Cathepsin-B, B1/B2: Cathepsin-Z

Other Proteins of interest

Intervention

Salivary α-amylase (AMY1) was found at highest concentrations after chocolate bovine milk intervention week, regardless of which data type (calculated concentration or ratio of spectral intensity) was used (Fig 5). Fatty acid binding protein 5 (FABP5) was also found in greatest concentration after chocolate bovine milk week, but only when observing calculated concentrations (the more “absolute” concentration value, rather than the relative to total protein value). Using the ratio of spectral intensity, FABP5 was lowest after the almond milk week intervention compared to washout weeks as well as bovine milk week intervention.

Other primary proteins of interest from the sensory perspective, notably carbonic anhydrase 6 (CA6, gustin) and lipocalin-1 (LCN1, von Ebner’s gland protein) did not show any significant differences based on any of our model effects. All other proteins that showed potential differences by intervention, gender, week order, age, or the interaction of the intervention*week order are included in the supplemental files.

Sensory Ratings

Summary of effects for sensory ratings are shown by sample type (Fig 6) and intervention (Fig 7). Intensity ratings are displayed on logarithm base 10 transformed y-axis because those data were analyzed with that transformation, while disliking/liking ratings are scaled with the original scale because the residual of that statistical model were better distributed in their native form.

Figure 6:

Sensory ratings of the chocolate milks (main effect of milk type, across all intervention weeks), only ratings from subjects/visits with an analyzed saliva sample. Boxes indicate 25–75%, whiskers go from minimum to maximum values, horizontal line indicates median. Intensity ratings are shown on a logarithmic scale (base 10), as that is how they were analyzed.

Figure 7:

Sensory ratings of all milks by intervention type (i.e., these are all milks’ ratings, after the washout weeks, the almond intervention week, and the bovine milk intervention week), only ratings from subjects/visits with an analyzed saliva sample. Boxes indicate 25–75%, whiskers go from minimum to maximum value, horizontal line indicates median.

Sample

The different milk types (almond, bovine) significantly influenced the sensory ratings for the chocolate milks (Fig 6). Bitterness and dry/rough (astringent) ratings averaged under the midpoint of the (original, un-log’d) scale for both milks, with bitterness ratings higher for almond than bovine (p=9.1×10⁻⁴). Flavor was near the mid-point for both milks, with bovine rated higher (p=1.2×10⁻⁸). Sweetness was rated just under the mid-point for almond milk, and just over the mid-point for bovine milk (significantly different, p=3.2 ×10⁻¹⁷). Almond milk was rated near neutral for liking, while bovine milk was rated above neutral (significantly different, p=3.9×10⁻¹⁰).

Week order group

No effects of time (measured as week order group) passed the Bonferroni correction for multiple comparisons. Data and a figure are available in the supplemental files.

Intervention

Flavor was less intense after the bovine milk intervention compared to the “washout” (low polyphenol) weeks (Fig 7). Some trends were apparent for bitterness to be more intense after the bovine milk intervention (see supplemental tables), but these did not survive the Bonferroni adjustment for multiple comparisons.

Discussion

In this study, we tested whether one week of regular polyphenol consumption plus chocolate almond or bovine milk, compared to a week of low polyphenol consumption, would result in changes in concentrations of salivary proteins that might also relate to changes in sensory perception of bitter, astringent polyphenols. We did indeed observe many changes in salivary proteins, yet not all of these changes match our hypotheses. Additionally, we did not observe the expected sensory changes in the bitterness or astringency of chocolate milks comparing our interventions to washout weeks, though milks were rated as less flavorful after the chocolate bovine milk intervention week compared to the washout weeks. We suspect this may be partially because our milks were reasonably well liked (above neutral on the scale)—not very bitter, and not very astringent even at the beginning of the experiment (below the midpoint of the scale). Additionally, the need to use untrained participants (as necessitated because training would expose participants to polyphenols) adds error to the sensory ratings, as untrained participants often confound “bitter” with other aversive sensations. Though we do see some salivary changes similar to animal data on polyphenol exposure, this study does not establish what part of our interventions (the almond milk, the bovine milk, the cocoa, energy addition to the diet, general hydration status, the general reduction of polyphenols during the “washout” weeks, etc.) actually caused the various changes in salivary proteins we observed. It does, however, provide the first direct evidence in humans that diet influences saliva. Furthermore, the different outcomes observed from weeks of different chocolate milk types (almond vs. bovine), indicate that food matrices and the specific context or delivery method of a dietary change can alter the influence of diet on saliva.

Protein changes

In prior work, proline-rich proteins (PRPs) and α-amylase are probably the best studied salivary proteins that interact with food to influence flavor, with some studies also noting cystatins, carbonic anhydrase-6, and lipocalin-1 as relating to flavor as well. We will discuss ours and others’ findings related to PRPs most extensively, as there is more literature to consider in context of our current findings and study design.

Animal data indicate that cystatins and PRPs are found at higher concentrations in saliva after rats consume a tannic acid rich diet (Martin et al., 2018, 2019; Torregrossa et al., 2014). Our study did find some PRPs in humans at higher concentrations after higher polyphenol exposure, though we cannot say if these changes were due to the higher polyphenol content or other intervention effects. Our observed differences were primarily apparent when we analyzed the data as calculated concentrations (i.e., more equivalent to mg/mL units) rather than as ratios of total protein (i.e., more like percents of total protein). As ratios do not reflect actual concentrations of protein available for chemical reactions, we find this outcome intriguing. By analyzing the data using a mass-to-volume, rather than percent style, concentration, we saw outcomes more consistent with the animal literature on PRPs increasing after polyphenol exposure. This suggests that methods testing these phenomena in humans should carefully consider what type of metric is used for analyzing protein contents. Additionally, only one basic PRP (PRB3), which are the most commonly studied PRPs in relation to astringency (Condelli et al., 2006; de Freitas & Mateus, 2001; Dinnella et al., 2010; Soares et al., 2012), was observed to increase after either intervention. That particular PRP was found at higher concentration after chocolate bovine milk, not almond milk as we expected. Many non-alkaline PRPs have been observed to interact with polyphenols (Ramos-Pineda et al., 2019; Soares et al., 2019). These other families of PRPs have simply not been as fully studied in relationship to actual human perception of the polyphenols. However, why most PRPs in our current work (PFN1, PRB3, PRH1, PRR4, PRR27) increased the most after chocolate bovine milk exposure rather than almond milk is unclear. Statherin was the only PRP found in greatest concentration after almond milk interventions, which is more in line with our expectations. Notably, statherin does interact with condensed tannins, which are present in cocoa (Soares et al., 2011, 2013). As statherin is present at the mucosal layer (Nayak & Carpenter, 2008), it may have the potential to influence astringency. Yet, we did not observe any substantial or significant changes in the dry/rough astringent sub-qualities in our current work.

Notably, acute studies of tannic acid and cranberry juice also indicate different outcomes in terms of subsequent salivary protein concentrations for specific salivary PRPs and a cystatin, not broad categories of basic vs. acidic PRPs (Melis et al., 2017). That study found greater concentrations of these proteins and one other basic salivary PRP (PS 1, not identified as different in our study) to be greater after cranberry juice exposure than after tannic acid exposure. These patterns are consistent with other acute research indicating that salivary proteins are in general depleted after tannic acid stimulation, and that work would indicate that greater depletion results in greater astringency ratings (Fleming et al., 2016). Other data on saliva and astringency confirms that not all potentially astringent foods differ by saliva parameters (Dinnella et al., 2011). Overall, it appears that specific salivary proteins may interact with specific compounds in foods, rather than generalized salivary protein interacting with all astringent stimuli (Kallithraka et al., 2001). This is potentially congruent with our salivary protein data, which did not indicate a generalized increase in protein concentration but rather changes in specific proteins within saliva. Thus, food and polyphenol types are certainly important when considering the influence of astringent products on saliva. While our milk interventions did not have the expected effects overall, the food matrix did influence alterations in saliva. Future work should investigate how this occurs, whether as a direct result of the different composition of the milks in the oral cavity or though some other dietary or physiological compensatory mechanism.

The patterns we observed in salivary PRPs are not consistent with our expected results based on milk type. Potentially, the differences between baseline diets compared to washout week (low polyphenol) diets could have been strong enough to mute any milk-specific effects on salivary PRPs. However, then we would not have expected to observe different outcomes by the different milk types at all. Perhaps the caseins in bovine milk do not interfere with upregulation of PRPs after polyphenols exposure, contrary to our original hypothesis, or perhaps these PRPs play a different role in human management of polyphenols in the orogastrointestinal tract. Alternatively, perhaps the carbohydrates in the bovine milk influenced the outcome. However, while our bovine milk certainly contained more sugar than the almond milk, the main sugar of bovine milk lactose. We know of no evidence that lactose would interfere with protein-polyphenol binding. Indeed, if anything, we would expect the polysaccharides of the almond milks (which are added as stabilizers) to have greater interference with the polyphenol-protein binding than simple sugars (Brandão et al., 2020; Luck et al., 1994; Renard et al., 2017). Nonetheless, the sensory ratings would indicate that the almond milk was still experienced as more dry/rough. Thus, if the differing types and quantities of carbohydrate did influence polyphenol-protein binding, the patterns of influence do not match between the sensory and salivary protein data.

The relationship between salivary PRPs and astringency or bitterness is somewhat confusing if the entire literature is considered. In vitro binding of PRPs and polyphenols is extensively established—but most of that in vitro work seems to imply that salivary PRPs binding polyphenols should induce astringency. This is, however, backwards from actual sensory work demonstrating that individuals with more binding protein in their saliva actually experience less astringency (Horne et al., 2002). Comments in the article from Horne, Hayes, and Lawless reflect that this conclusion was a surprise, considering all the in vitro earlier work. That article tested the development of turbidity, or “haze,” that appears when polyphenols are interacting with PRPs or other similar proteins in solution. More haze in saliva indicates the presence of more salivary PRPs/other binding proteins in the sample, due the scattering of light by these large proteins/polyphenol complexes. Originally, the authors of the work on turbidity formation and astringency anticipated that “haze development might parallel astringency”; however, instead their data revealed the reverse. Saliva samples with more haze came from participants who experienced less astringency (Horne et al., 2002). Thus, they concluded that salivary PRPs protect against astringency. This is congruent with animal literature indicating that increased salivary PRPs reduce the aversiveness of tannic acid solutions (Torregrossa et al., 2014). Together with animal data on quinine exposure, this protection offered by PRPs for astringency could occur for bitterness as well (Martin et al., 2018, 2019). Binding of bitterants by PRPs is also supported by in vitro binding assays for smaller, bitter phenolic molecules (Baxter et al., 1997). We suspect that the confusion over the protective vs causative outcome for PRPs and unpleasant sensations is due to the fact that in vitro complexes of salivary PRPs and polyphenols will precipitate, especially when centrifuged. However, the mouth is not a centrifuge. Precipitation would have to happen rapidly for dispersed aggregates of PRPs/polyphenols to be causative for astringency. Thus, salivary PRPs are more likely protective rather than causative for astringency.

Notably, many individuals did not have detectable quantities of several PRPs in their saliva, as indicated by the large numbers of half minimum values observable in Figs 2–5. PRPs can be difficult to accurately quantify, as they undergo a wide variety of post-translational modifications and proteolytic events, and they also have many isoforms are encoded by the same genetic loci (Amado et al., 2010; Campese et al., 2009; Lyons et al., 1988; Manconi et al., 2016; Stubbs et al., 1998). Consequently, methods to quantify PRPs and their numerous posttranslational products are still advancing, and applying new methods in future studies could yield substantially different results. Pilot data from our lab would indicate that freezing prior to centrifugation (as done in the current study) yields greater amounts of PRPs in our proteomic methodology compared to centrifuging prior to freezing (Davis & Running, 2019), which is another common approach. However, given the numerous samples we encountered that were not analyzable due to the formation of gels in the tubes (which occurred after centrifugation, despite immediate flash freezing on a CoolRack and despite being kept cold during the entire protocol), more work is suggested on optimal preparation methods for stimulated saliva.

Our study also observed increased concentrations (in both ratio and calculated concentrations) of cystatins in saliva, particularly after the bovine milk interventions. We had actually hypothesized that these differences would be greatest after the almond milk intervention, which was not the case. This, as discussed above for PRPs, implies that binding of polyphenols by milk casein does not disrupt potential dietary related changes in salivary cystatin concentrations. Other cross-sectional work has noted higher cystatins in individuals who were more accepting and/or less sensitive to bitterness (Dsamou et al., 2012; Morzel et al., 2014). While we did not observe substantial or significant changes in bitterness in our current study, the increase in cystatins that we observed is consistent with the theories of these prior studies. Other work has considered various cystatins as an indicator of immune response and improved tumor prognosis (for review see Ochieng & Chaudhuri, 2010), with cystatin-C noted as a potential indicator of low inflammation. Thus, the cystatin family of proteins may be associating with the positive health outcomes that come from eating healthy foods typically rich in bitter molecules, rather than directly with the sensation of bitterness itself.

Two cathepsins (CTSB and Z) were decreased substantially after chocolate almond milk exposure but increased after chocolate bovine milk, a pattern not explained by their inhibitory cystatins. The increase in cathepsins B and Z after bovine milk might be explained as compensatory, as cystatins were generally also highest after bovine milk. However, that pattern is not apparent for the almond milk exposure weeks, and thus the relationship of these proteins to polyphenol exposure or milk exposure is unclear.

Polyphenols are known to inhibit the activity of α-amylase (Naz et al., 2011). Thus, the increase we observed in salivary α-amylase may have been compensatory, especially if the binding of chocolate polyphenols by caseins in milk was not effective in reducing the activity of the polyphenols. The increase in statherin after almond milk intervention weeks may have alleviated any polyphenol inhibition of α-amylase during those interventions, potentially explaining why we only observed a significant, potentially compensatory increase in α-amylase after bovine milk interventions.

The differences we observed in patterns of PRPs, cystatins, cathepsins, and other proteins of interest based on how we quantified their concentrations implies that our observed differences are not due to a simple change in salivary flow rate. If salivary flow rate was in general altered, we would have expected to see changes in total salivary protein concentration (greater flow results in lower salivary protein concentration, as the flow dilutes the protein) across interventions and washout weeks, which we did not observe (full statistical code and tables available in supplemental files). In hindsight, salivary flow rate data could have been measured to confirm this, but we were more interested in the biochemistry of the samples (i.e., protein concentration in the mixture of water+saliva, representing a similar value to what we would expect to find in the mixture of milk+saliva). Salivary flow rates vary by collection method (Dawes et al., 2000; Froehlich et al., 1987; Jones et al., 2000) and attitude of the participant toward the sample (Running & Hayes, 2016). Cannulating or Lashley cup (a device placed directly over parotid ducts to collect isolated parotid saliva) type methods for measuring flow of saliva in our study was not feasible, as our fundamental research questions are regarding how saliva mixes with food. Sensory science has known for decades that the environment of testing alters human behavior and human sensory ratings of foods (Lawless & Heymann, 2010), and our former work demonstrates that the least invasive method for measuring salivary flow (spitting and weighing the spit) is variable depending on subjects’ expectations and context for the stimulus (Running & Hayes, 2016). Thus, data from gravimetric or volumetric analysis of saliva for establishing a flow rate is highly variable and subject to problems such as partial swallowing of the sample. Moreover, we were more interested in salivary protein concentration of the mixture in the mouth, rather than these volume measures of salivary flow. Additionally, concentration of the mixture is the same, regardless of whether some of the mixture is swallowed, as long as saliva is thoroughly mixed with the oral stimulus. Consequently, focusing on concentration can avoid issues with incomplete expectoration. Further work should attempt to ascertain which changes in salivary proteins may be due to dilution/concentration effects of flow and which to more specific regulation of protein expression, but the differences we observed in both calculated and ratio concentrations indicate that flow by itself is unlikely to explain our data. Moreover, while flow of saliva into the mouth certainly influences salivary protein concentrations and thus the kinetics of those protein’s interactions, that flow is accounted for by mixing the saliva with our water sample to be expectorated. A higher flow rate would add more saliva to the water sample, but we would still be measuring the protein concentration of that full mixture, and thus the higher flow is still accounted for in our analysis, even though we cannot report it directly. Thus, while we only measured the protein concentrations in saliva collected in this study, we fully acknowledge that other factors we did not measure could be influencing those concentrations—including salivary flow rates. This is indeed part of why this field of research is complex, as a myriad of factors contribute to the composition and flux of saliva. We show in our current work that the composition of saliva mixed with an oral stimulus is influenced by ingested foods, yet the mechanisms for exactly why certain proteins increase, others decrease, and others remain unchanged need to be further investigated.

Sensory Changes

Though we did find a significant increase in some PRPs and cystatins after chocolate beverage consumption, we did not find evidence that this up-regulation also caused a decrease in astringency (termed “Dryness or Roughness” in our study) or bitterness. Since astringency is not a well-understood sensation, it is possible this was due to the minimal training for our subjects, or to the fact that our beverages were not very astringent (both under the midpoint of the scale). The untrained nature of our participants was unavoidable, as training them to these sensations would require exposure to astringent/bitter molecules, which we hypothesize could alter that sensation by altering saliva. Our preliminary data, however, indicated we had enough power to find differences in astringency, even using untrained participants. Even if our lack of astringency differences were due to increased variability from the untrained participants, our prior work (supplemental files) implies any false-negative effects are due to very small changes in astringency.

There was an indication of less overall flavor after the bovine milk intervention compared to washout week. We originally hypothesized there would be less salivary binding proteins for polyphenols after bovine milk interventions, which would have resulted in greater unpleasant flavors after that week. However, this was not the case. Moreover, these sensory changes do not match the observed pattern of changes observed in salivary protein concentrations for PRPs (except statherin) or cystatins. We suspect the sensory differences observed due to interventions in our current study may have more to do with context effects, as the chocolate bovine milk was overall the better liked, less bitter, more flavorful, and sweeter milk. Thus, after a week of drinking chocolate bovine milk (the best milk), the other milk may have been less enjoyable, reflected in the lower flavor intensity. It is also possible that our hypothesis was in part correct—that binding of polyphenols to bovine milk proteins prevented physiological changes that caused lower flavor intensity after consumption of the chocolate almond milk. However, given the very small and only marginally significant effects of our interventions, we must conclude the changes we observed in salivary proteins did not strongly influence sensory ratings in chocolate milks.

Limitations

Importantly, many other confounding factors may influence our results. The matrices of both almond and bovine milk are complex, with ingredients other than just polyphenols that could potentially influence saliva. However, we intended this study as a first step to observe whether an actual food (rather than model system) would influence saliva. Certainly, other work will be needed to identify which changes (if any) in salivary proteins are truly due to polyphenols versus other components of the milks we used. Potentially, all changes we observed in saliva could be due to the contrast between the low polyphenol diet and regular diet of the participants (which they returned to during chocolate milk weeks), though then we would not have expected to see any differences by milk type. Additionally, work will be needed to understand whether changes in saliva are due to oral sensory exposure or actual ingestion of various food components. However, at a practical level we have demonstrated that saliva does indeed change in response to fairly small and realistic dietary changes, focused on polyphenols. Thus, further investigations are warranted into how and why such changes occur.

With our current study results, we cannot exclude the idea that the almond vs. bovine milks themselves, hydration status, or caloric intake caused the differences in salivary proteins. Given the prior animal work and human cross-sectional data (Dsamou et al., 2012; Martin et al., 2018, 2019; Morzel et al., 2014; Torregrossa et al., 2014), we hypothesize that polyphenols are the likely stimulus for inducing cystatin and PRP changes. However, we cannot currently say whether these are induced by oral, gastrointestinal, or other systemic exposures to the polyphenols. Additionally, the changes in salivary α-amylase and fatty acid binding protein 5 indicate that other proteins of interest from a sensory perspective were also altered by our interventions, and how these might or might not influence sensation in other settings is unknown. Many other proteins, not discussed above as they were not of a priori interest from a sensory perspective, also changed across our intervention and control weeks. Future work will need to investigate whether the milks, hydration, or other compensatory dietary or behavior changes status could be contributing to our observed effects.

While a simpler system, such as simply polyphenols in water, would have allowed us to better isolate the effects of polyphenols from other dietary components, we used milks because they are a plausible intervention with a relatively pleasant beverage. Additionally, from a food science and nutritional perspective, a comparison of commercially available milk and milk alternatives is a dietarily and agriculturally relevant investigation, especially as bovine milk contains proteins likely to interfere with the known dietary component of interest (polyphenols) and almond milk does not. Our current investigation certainly cannot assign any causality to what components of these milks are responsible specifically for which salivary changes, or indeed if some other dietary compensation is the causative agent. However, the results indicate that future interventions with simpler stimuli are warranted in order assess the mechanistic underpinnings of our findings.

Data on the baseline consumption of polyphenols would also have been informative for our study. If diet does indeed influence the concentration of specific salivary proteins, then our participants’ baseline dietary habits may have influenced their baseline salivary protein profiles. We attempted to mitigate these effects by having all participants go through the “washout” week at the beginning of the experiment. Yet, data on the timeline of how diet may influence saliva are as of yet unavailable. Consequently, we cannot be certain our initial washout week was long enough to eliminate prior diet effects on saliva or, potentially, sensation. Moreover, our instructions for all participants to avoid polyphenols during the washout weeks (even if they would typically have consumed many polyphenols) could also have influenced the results. We gave this instruction attempting to minimize the influence of the baseline diet on our outcomes. For example, if we instructed all participants to avoid polyphenols for the entire study, then any high polyphenol consumers would have been on a reduced overall polyphenol diet for all 6 weeks. This could have potentially masked any effect of adding chocolate to the diet. On the other hand, if we had given no instructions about polyphenols in the diet, then the addition of the chocolate milks may not have been adequate to overcome baseline diet effects. Thus, we set up this study in an attempt to have the best chance to observe outcomes of consuming a diet relatively higher and relatively lower in polyphenols. However, this certainly means that we cannot ascribe all outcomes specifically to the milks, as these effects could be due to changes in low-polyphenol washout weeks to regular diet weeks (when participant also happened to consume our milks). Future work will need to address these limitations, and separate the confounding factors.

The large number of saliva samples that either gelled, leading to low and highly variable protein content, or simply had too little protein to analyze is a serious limitation in this work. Notably, we did not screen using any clinical measures to observe whether participants had low salivary flow (our screening on “too much or too little saliva” was all self-report). However, both self-reported and measured rates of hyposalivation among populations similar in age and general health status to our sample group are around 10–13% (Flink et al., 2008, 2020; Thomson et al., 2006). Thus, to lose over 20% of our samples due to hyposalivation seems unlikely.

Finally, there are a myriad of measurements that can be collected in studies of saliva. We chose to focus on protein composition, through a proteomics approach, in order to maximize the amount of information we could gain about the protein composition of saliva in response to a dietary intervention. However, salivary flow rates would certainly also influence the amount of actual protein present in the mouth, as would swallowing rates, oral volume, amount of saliva retained on oral surfaces, and many other factors. Improving the precision of salivary measures related to chemical kinetics of saliva-food mixtures is an important goal in improving understanding of these phenomena.

Conclusions

Our interventions of regular diet plus chocolate milks compared to low polyphenol diet demonstrate that the concentration of many salivary proteins can be influenced by diet. Similar findings in controlled animal studies (Martin et al., 2018, 2019; Torregrossa et al., 2014) and plausible mechanisms related to several of our proteins of interest would indicate that some of our observed protein changes in saliva may be due to the polyphenols in the diet, but further work is required to confirm this as well as to isolate polyphenol from other dietary effects. Future work should also explore how long these salivary protein changes are sustained, whether changes would be more dramatic after longer exposures, as well as how early in the exposure the changes can be detected. Our study was designed to answer whether changes in saliva or sensation could be observed from relatively small dietary modifications that is dietarily relevant as well as practically available. Now that we have observed salivary protein changes in conjunction with a diet higher in polyphenols and containing these real (but chemically complex) beverages, further work in less complicated matrices should be pursued in order to validate, expand, and point to mechanisms for these findings. If saliva does indeed respond to diet, and if it responds to wider array of dietary compounds beyond just polyphenols, this could be a useful tool for monitoring dietary habits as well as for potentially measuring diet-induced health outcomes such as inflammation.

Highlights.

• Prior rat data show polyphenols alter salivary proteins that influence sensation

• We tested if chocolate polyphenols altered human saliva and sensation

• Chocolate almond and bovine milk consumption altered salivary proteome

• Several salivary proline-rich proteins and cystatins increased in concentration

• Minimal sensory changes were observed