Proteomic Analysis of Plasma Reveals Fat Mass Influences Cancer-Related Pathways in Healthy Humans Fed Controlled Diets Differing in Glycemic Load

The human diet is a complex exposure that plays a recognized role in the risk of several types of cancer (1). Glycemic index (GI) and glycemic load (GL) may be a relevant component of dietary recommendations for cancer prevention due to the known effects of GI/GL on blood glucose and insulin concentrations and the effects of these on oxidative stress and immune response (2). One study examining GI and colorectal cancer identified that a sedentary lifestyle in addition to a high-GI diet was associated with a higher risk of colorectal cancer than an active lifestyle plus high-GI diet or sedentary and low GI (2). Additional studies have also shown GI and GL as risk factors for colorectal, breast, and pancreas cancer (2, 3). Observational studies suggest that we need to understand the biological response to a total dietary pattern to provide better public health recommendations (4, 5).

Increased adiposity is a risk factor for at least 13 types of cancers (1, 6) and has been shown to increase inflammatory responses, with obesity characterized as a “low-grade chronic inflammatory state” (7, 8). In addition, growth factors and hormones such as insulin-like growth factor 1 (IGF1), insulin, and leptin (LEP) are increased with higher adiposity (9). Identifying dietary patterns that contribute to, as well as measures to prevent, the predisposition of obese and overweight individuals to cancer is a high priority worldwide (10).

We previously showed that response to a low-GL diet higher in whole grains and fresh fruits and vegetables, compared with a more refined grain high-GL diet, resulted in reduced fasting concentrations of the acute-phase protein C-reactive protein (CRP; ref. 11), reduced IGF1 (12), reduced postprandial glycemic response (13), and improved satiety (14). These results have suggested that these diets could have a broader effect on integrated cellular processes and cellular signaling linked to cancer risk. Omics approaches, such as proteomics, can interrogate the effects of dietary changes on a broader scale, allowing us to understand how signaling pathways are modulated in ways that might affect disease risk.

Utilizing proteomics to measure response to exposure that may influence cancer risk has been reported, including studies to discover plasma biomarkers of cancer risk (15–17). Other studies in prospective cohorts have identified biomarkers present in plasma years prior to cancer diagnosis, suggesting that differences in certain pathways may reflect increased cancer susceptibility (18, 19). Proteomics approaches have also been used to examine responses to interventions, including those testing diet (20) and drugs (21), as well as to examine the effect of body mass on overall pathway biology (7). These studies suggest that in the context of interventions, proteomics can provide insight into the complexity of multiple responses to a dietary intervention and the possible mechanisms driving intervention effects. The objective of this study was to use proteomic approaches to identify proteins and pathways that are altered in the plasma of participants fed low- and high-GL diet patterns and to evaluate whether response differed by participant adiposity and any potential associated cancer risk.

The plasma samples used in this study were obtained from the Carbohydrate and Related Biomarkers (CARB) study. The CARB study was a randomized crossover–controlled feeding trial designed to test the effects of low- versus high-GL diets on known biomarkers of cancer-risk pathways; complete details on recruitment and study design have been published previously (11). This trial was registered at ClinicalTrials.gov as NCT00622661. Briefly, healthy, nonsmoking men and women, aged 18–45 years, consumed two controlled diets for 28 days each, with at least a 28-day washout period between intervention periods (11). One diet was high in whole grains, fresh fruits, and vegetables, and low in GI carbohydrate sources (WG). The other diet substituted refined grains for whole grains and provided carbohydrates from mostly high-GI sources (RG). The two diets differed primarily in GL and total dietary fiber content (22). Mean daily energy intake (estimated for each participant based on height, weight, sex, and usual activity level) and percent energy from carbohydrate, protein, and fat was designed to be similar on the two diets. Complete details on diet menus and consumption have been published previously (11, 12). All food was prepared under consistent and carefully controlled conditions and provided to the participants. Fasting blood was collected at the beginning and end of each diet period, and each participant served as their own control in this cross-over design (11). A total of 82 participants were recruited for the CARB study and 80 completed both feeding periods (11). In our analysis, three plasma samples collected in EDTA from each of the 80 participants who completed the CARB study were used: day 1 of the study (before diet intervention), as well as after 28 days on each diet. At the baseline clinic visit, all participants completed measures of height, weight, and body circumference, and underwent dual X-ray absorptiometry (DXA), which yielded measures of total body fat mass (FM). High FM was defined using the accepted cut points of >25% and >32% for men and women, respectively (23), consistent with other evaluations of effects of FM in the CARB study (11).

The antibody arrays were populated with 3,504 distinct antibodies acquired primarily from commercial suppliers such as SDIX (now sold by Novus Biologicals), Aviva Biosciences, R&D Systems, Abnova, Sigma-Aldrich, and others. The 3,504 antibodies correspond to 2,072 different human proteins that participate in diverse signaling pathways. Details on array fabrication have been previously reported (24, 25).

Relative protein levels were detected as described previously (24–27). Briefly, albumin and IgG were depleted from plasma, and the depleted plasma was concentrated to its original volume, measured for total protein concentration, labeled with the amine-reactive dyes Cy3- and Cy5-maleimide, and unincorporated dye was removed. Individual Cy5-labeled participant samples were incubated with an equal amount of Cy3-labeled reference (a common pool of plasma comprised of samples collected from four women and three men was used as a reference). Labeled lysates were incubated on arrays for 90 minutes, the arrays were washed serially to remove excess dye, and then scanned in an Axon GenePix 4000B microarray scanner and data extracted using GenePix Pro 6.0 Software (Molecular Devices).

For each antibody, the fold change of signal (red channel) compared with reference (green channel), the M value, was calculated as log₂(Rc/Gc); where Rc is red corrected, and Gc is green corrected using the “normexp” background correction method (28). Technical sources of variation were normalized by loess procedures for microarrays using R limma package (RRID:SCR_010943; ref. 29), including within-array print-tip loess and between-arrays reference channel quartile normalization. Following normalization, triplicate features were summarized using their median. All statistical analyses were conducted on M values.

Linear mixed models (LMM; ref. 30) were fit for each antibody using the pre-intervention baseline and two post-diet timepoint measurements for each of the 80 participants. After initially adjusting for baseline values for each participant, the two post-diet timepoint measurements in this crossover study were accounted for by a random-effect term per participant in the LMM, performing repeated-measures analysis and allowing each participant to serve as their own control. The baseline protein concentrations prior to diet intervention and the diet sequence, together with sex, age, fat mass by DXA (defined as 0 if <25% for male or <32% for female, 1 if ≥25% for male or ≥32% for female) and the experimental hybridization days, were adjusted in the models. Antibodies for which the mean difference in M values between the diets differed from zero (at P < 0.05; ref. 31) were selected. Further stratification analysis was performed on subgroups of participants with low and high FM. In each group, LMMs were fit as described above, omitting DXA measures.

To obtain pathway membership information for all candidate proteins, we utilized Enrichr (RRID:SCR_001575; ref. 32), a comprehensive web-based resource for curated gene sets. Specifically, we input the list of proteins that were candidates for differential abundance (P < 0.05 unadjusted for multiple comparisons) between diets and Enrichr provided a list of pathways [as identified by the Reactome (RRID:SCR_003485) 2016 database (33, 34)] containing these proteins. Because our antibody array does not represent all proteins in every pathway, our testing procedure accounted for this reduced pathway representation. Specifically, we accounted for the total number of proteins analyzed on the array (N = 2072), the number of proteins in the pathway represented on our array (J), the total number of proteins of interest (K; 172 for all participants, 221 for high FM, and 266 for low FM), and the number of the proteins of interest in the pathway (X). Using a hypergeometric test [Microsoft Excel 2016 (HYPGEOM.DIST)], we computed the probability that when K proteins (among all N on the array) are deemed candidates (at the unadjusted level of P < 0.05), then a pathway of size J contains at least X of these antibodies. We defined pathways of interest for further investigation as those exhibiting an overrepresentation based on a hypergeometric test P <0.05, following correction for multiple comparisons by the method of Benjamini–Hochberg (35).

Plasma protein concentrations were determined using the BCA Protein Assay Kit (Thermo Fisher Scientific). For the leptin (LEP) blot, 50 μg of protein per sample was prepared in Laemmli sample buffer plus 3% β-mercaptoethanol with human LEP (Sigma-Aldrich) as a positive control. Lysates were separated on a NuPage Novex 4–12% Bis-Tris Gel (Invitrogen) and transferred to nitrocellulose. The blot was subsequently blocked in 3% nonfat dry milk/PBS and then incubated with a LEP primary antibody (Aviva Systems Biology; catalog no. ARP41698; RRIS:AB_2046546). For the EGFR blot, 20 μg of protein was prepared, separated on Novex Wedgewell 10% Tris-Glycine gel, transferred to nitrocellulose, and the blot was subsequently incubated with an EGFR primary antibody specific to the extracellular domain of the protein [Novus Biologicals; catalog no. 19060002 (formerly from SDI); RRID:AB_10004630]. Donkey anti-rabbit Alexa Fluor 790 (Invitrogen; catalog no. A11374; RRID:AB_2534145) was used as a secondary antibody for both blots. The immunoblots were scanned and densitometry measurements were conducted using a LI-COR Odyssey Infrared Imager (LI-COR Biosciences) and associated software. A two-tailed paired Student t test was used to compare band intensities between WG and RG diets on the LEP blot.

Previous reports involving the CARB study found that diet intervention effects differed between participants with high and low FM (11, 36); therefore, we analyzed all 80 participants as a group as well as stratified by FM into two separate analyses (Table 1). The mean age of participants was slightly (∼4 years) higher in the high-FM group (P = 0.006). The high-FM group had more females and the low-FM group had more males (P < 0.05; Table 1). Age and sex presented potential confounding factors, so we adjusted for both in our LMM analysis.

The difference in protein expression between RG and WG diets was investigated using 2,072 different targeted proteins. Analysis in all 80 participants identified 190 antibodies, corresponding to 172 unique proteins that showed evidence of different expression due to diet (unadjusted P < 0.05; Supplementary Table 1A). Analyses stratified by high- and low-FM revealed 251 and 306 antibodies, respectively, which showed evidence of different expression between the diets, corresponding to 221 and 266 unique proteins (unadjusted P < 0.05; Supplementary Fig. S1A and S1B; Supplementary Tables S1B and S1C). Proteins showing possible differential expression due to diet were defined by an unadjusted P <0.05 with the goal of identifying pathways that may be overrepresented for these proteins.

Of the proteins identified in the analysis of all 80 participants, 134 (78%) overlapped with proteins identified in the two FM categories (Fig. 1A). The two analyses by FM group contained 48 proteins in common, 16 of which were detected by more than one antibody in the low- and/or high-FM analysis (Fig. 1A). Among these 48, 25 (52%) appeared higher in the same diet in each analysis (i.e., higher in RG diet or higher in WG diet in both FM groups; Fig. 1B). Leptin (LEP), a hormone used as a marker in obesity that is involved in energy homeostasis and a key player in body weight control (37), was higher after the consumption of the WG diet as compared with RG in both stratified and unstratified analyses (P < 0.01 in FM-stratified analyses, P < 0.05 in all 80-participant analysis; Fig. 1C; Supplementary Tables S1A–S1C). As our previous work in these participants found no significant diet difference in plasma LEP concentration when employing a different assay (11), we wanted to confirm our findings by using an additional assay that employed the same antibody as was on our array. We validated the specificity of the antibody used on our array and recapitulated our observation using a Western immunoblot assay, confirming higher expression of LEP in plasma after consumption of the WG diet compared with RG, regardless of FM (P < 0.05; Fig. 1D). Of the 48 proteins in common between the high- and low-FM analyses, 23 (48%) appeared higher in opposite diets between FM analyses (Fig. 1B). One of these 23 proteins, EGFR, was higher after consumption of the RG diet among participants with high FM (P < 0.05), whereas among participants with low FM it was higher after the WG diet (P < 0.05; Supplementary Fig. S2A; Supplementary Tables S1A–S1C). We wanted to confirm that we were seeing the soluble form of the transmembrane protein EGFR (sEGFR, ∼100kDa) in the participants' plasma, as the overexpression of EGFR is commonly observed in cancers (38) and could indicate a change in cancer-related predisposition in our participants. We observed that the primary size of EGFR measured in participant plasma was approximately 100 kDa compared with full-length EGFR around 160 kDa (Supplementary Fig. S2B). While not tested for significance, as we only examined samples from two participants in each FM group, in addition to identifying the form of EGFR present in our samples, we also observed higher signal after consumption of the RG diet (compared with WG) in samples from high-FM individuals and higher signal after consumption of the WG diet in samples from low-FM individuals, consistent with our array results.

We first explored pathways of interest, according to our hypergeometric test, in all 80 participants. For identification purposes, all Reactome pathways are italicized. Sixty-nine pathways were identified as overrepresented at FDR <0.05 (Supplementary Table S2). Similar pathway analyses stratified by participant FM identified 142 pathways overrepresented in participants with high FM (FDR <0.05; Supplementary Table S3), and 72 pathways overrepresented in participants with low FM (FDR <0.05; Supplementary Table S4). Of the 69 pathways identified in all 80 participants, 61% were also present in one or both FM groups (34 in common with high FM, 16 with low FM; Fig. 2A). The pathways identified in each FM analysis were predominantly distinct from each other, with only 16 pathways in common, accounting for 11% and 22% of the high and low FM overrepresented pathways, respectively (Fig. 2A). The majority of the pathways the two FM overrepresentation analyses had in common (56%) were in pathways regarding the biology of protein modification, a feature of many cancer types.

Pathway overrepresentation analysis identified three primary pathway areas, DNA Repair, Cell Cycle, and DNA Replication, each with numerous individual subpathways that were exclusively overrepresented for proteins differing between WG and RG diets in participants with high FM (Fig. 2B; Supplementary Figs. S3A, S3B, and S3C). For cancer to arise, changes in DNA Repair (more active in RG in high FM), Cell Cycle, and DNA Replication (Supplementary Table S3) are required.

Although many pathways in Reactome could be viewed as cancer related, participants with high FM had pathways overrepresented for proteins differing between diet that were closely related to the initiation and/or predisposition of cancer. First, the pathway group Immune System (Supplementary Fig. S3D) had six pathways of the Adaptive Immune System overrepresented by proteins differing between diet in the high-FM group (Fig. 3A and B). Three of the Adaptive Immune System pathways are subpathways to the Class I MHC mediated antigen processing & presentation tertiary pathway. Second, some of the secondary Signal Transduction (Supplementary Fig. S3E) cancer-related pathway categories, including Signaling by WNT, contained predominantly high-FM overrepresented pathways (Fig. 4A and B).

While many of the cancer-related pathways identified in the overrepresentation pathway analysis of high-FM participants were not identified in the low-FM participants, post-translational protein modification had numerous pathways overrepresented for proteins differing between diets in both FM groups. Post-translational protein modification, a secondary pathway in Metabolism of proteins (Supplementary Fig. S3F), contained eight pathways overrepresented for proteins differing between diet in the high-FM participants and four in the low-FM participants (Fig. 5A and B). Two pathways, Post-translational protein modification and O-linked glycosylation, were identified in both high- and low-FM participants. In the high-FM group, 82% of the proteins differing between diets in the Post-translational protein modification pathway had higher expression after consumption of the RG diet (50% in low-FM group; Supplementary Tables S1B and S1C; Supplementary Table S3; Supplementary Table S4). Diseases of glycosylation, a secondary pathway in Disease (Supplementary Fig. 3G), contained six pathways overrepresented for proteins differing between diets in the high-FM participants and seven in the low-FM participants (Fig. 5C). Six pathways overlapped between the high and low-FM overrepresented pathways, including Diseases of glycosylation and Diseases associated with O-glycosylation of proteins. Diseases of glycosylation had 88% of proteins higher after the RG diet in the high-FM group (50% in low-FM group; Supplementary Tables S1B and S1C; Supplementary Table S3; Supplementary Table S4).

Here, we demonstrate measurable proteomic differences in plasma of healthy adults after 4 weeks of consumption of RG- versus WG-controlled diets. Proteomic differences due to diet were more robustly detected after stratifying participants based on participant DXA–measured FM. One striking observation was that proteomic differences observed in the two FM protein analyses were largely unique, with only approximately 10% of proteins identified per group overlapping and similarly higher with the same diet. One such protein, leptin, showed higher plasma levels after the WG diet in both FM groups. Leptin is an adipokine with a key role in the regulation of energy homeostasis and food intake (37). We previously reported that participants in the CARB study found the WG diet more satiating then the RG diet (14). We did not see a diet difference in leptin in a previous analysis in these participants, even when stratified by participant FM (11), possibly due to the use of different antibodies and detection platforms. Because most of the diet-differences in proteins identified in our analyses were unique to either high or low-FM participants, we hypothesized that there were differing biological responses to the diets based on FM. Pathway analysis was used to test this hypothesis and showed the following: (i) pathway overrepresentation analysis in the high-FM group identified many key processes in the initiation and predisposition of cancer that were altered by diet only in the high-FM participants and (ii) differing expression of proteins by diet in post-translational protein modification pathways (a process observed in the detection of many cancer types) occurred in both FM groups.

Pathway analysis among individuals with a high FM uniquely identified pathways involved in Cell Cycle, DNA Repair, and DNA Replication as overrepresented for proteins differently expressed between the WG and RG diets. DNA Repair, Cell Cycle, and DNA Replication pathways are critical for cell replication, and disruption in any of these processes can lead to cancer initiation (39). Sustained alterations in these cancer-related pathways could lead to the development of cancer (39). As individuals with high FM are at increased risk for some cancers (1, 6), finding that the difference between consuming a WG or RG diet could lead to the change in expression of proteins involved in these pathways highlights the need to further study diet in relation to the link between obesity and cancer. Overall, we saw that DNA Repair pathways were more active (had higher protein expression) after consumption of the RG diet compared with the WG diet in the high-FM participants. One protein that was higher after the consumption of the RG diet and is a member of DNA Repair pathways, proliferating cell nuclear antigen (PCNA), acts to recruit DNA damage repair proteins that allow completion of DNA replication after DNA damage (40). Note that because Cell Cycle and DNA Replication pathways are highly interconnected, the proteins that identify a change in these pathways are the same. Three proteins in DNA Replication and Cell Cycle pathways, PSMD7, PSMB5, and PSMB9 are components of the proteasome (41). In addition, UBC, a protein in DNA Replication and Cell Cycle pathways, is involved in protein degradation via the proteasome (42). Proteasome activity is essential for progression through the cell cycle and regulation of DNA replication (41, 42). Currently, there are several proteasome inhibitors being used in cancer therapy (43).

Inflammation, a product of the innate immune system, is a proposed mechanism for the association between obesity and cancer (7). Obesity is associated with low-grade inflammation (7, 8), and chronic low-grade inflammation also tends to contribute to a more active adaptive immune system (44). Of the seven adaptive immune system pathways overrepresented by proteins differing by diet in our participants, six are overrepresented in the high-FM group. The secondary pathway group Signaling by WNT had nine pathways overrepresented by proteins differing by diet in our participants and all nine were overrepresented in the high-FM group (two were also overrepresented in the low-FM group). Wnt signaling, a critical normal developmental pathway, when altered can contribute to cancer initiation and is an important regulator of cancer (45). Previous studies have shown a direct link between Wnt signaling and colorectal cancer, as well as highlighted evidence of Wnt signaling in other cancers (45).

Post-translational modification involves additional covalent modifications to proteins after biosynthesis. Post-translational modification alters the functional properties of proteins, increases diversity of the proteome, and can impact almost all parts of cell biology and pathogenesis (46). Routinely used cancer biomarkers (e.g., CA125, CA15-3, CA19-9, PSA, and CEA for ovarian, breast, pancreatic, prostate, and colon cancer, respectively) are glycoproteins (47, 48). Our data suggest that glycosylation pathways are more active (i.e., have higher protein expression) with the RG diet compared with the WG diet, particularly in individuals with high FM. This may be plausible given the higher GL of our RG diet and is supported by previous studies showing a reduction in glycosylation with a low-GI diet in patients with diabetes (49).

The parent study from which these samples were derived was designed in a manner that strengthens the impact of the results we present here. All foods were consistently prepared under controlled conditions; all participants received and consumed the same foods; participants were instructed to maintain similar physical activity levels across both diet intervention periods and to avoid weight change; both controlled diets were eucaloric and each participant received both diets—all characteristics of our crossover study that allowed us to measure plasma proteomic changes that were specific to diet differences within an individual. In addition, all participants were healthy with normal fasting blood glucose levels. By stratifying participants, we did lose statistical power with smaller FM group sample sizes; however, we gained insights into FM-dependent responses. In addition, the distribution of men and women and age differed between the FM groups, but we adjusted for sex and age in our LMM to limit these confounding factors. Overall, the participants in the study were younger (ages 18–45 years), thus our results may not be generalizable to older individuals with long-term obesity. Furthermore, the platform we utilized to measure the proteomic response, while a powerful way to measure thousands of proteins in the blood, does not cover all possible human proteins. We attempted to overcome this by using overrepresentation pathway analysis, allowing us to identify pathways overrepresented for the proteins identified on our arrays as differing between WG and RG diets in the high- and low-FM groups. One could surmise that pathway analysis in general might have a cancer bias; we specifically used Reactome pathways, as they are frequently cross-referenced with other pathway resources, providing one of the most comprehensive and all-encompassing pathway databases (34).

In conclusion, proteomic differences in the biological response to consumption of a WG as compared with a RG diet were measurable in the plasma of healthy participants and allowed for us to employ pathway analysis to determine biological processes that were affected by the consumption of one diet versus the other. We observed striking diet differences between participants with low and high FM in pathways previously implicated in cancer predisposition and initiation. For example, in both the high- and low-FM groups, we uncovered post-translational protein modification pathway alterations between the WG and RG diets. However, pathways involved in cell cycle, DNA repair/replication, the adaptive immune system, and WNT signaling were overrepresented solely in high-FM participants. Altogether, we identified potential mechanisms of dietary patterns that can affect cancer predisposition and initiation in a susceptible obese population. Our data suggest that an individual's level of adiposity affects the physiologic response to a WG or RG diet and that further studies of diet recommendations for cancer prevention may need to consider the impact of adiposity on response.

No potential conflicts of interest were disclosed.

The content of this report is solely the responsibility of the authors and does not necessarily represent the official views of the NIH.

Conception and design: C.B. Garrison, S.L. Navarro, M. Kratz, M.L. Neuhouser, D. Raftery, P.D. Lampe, J.W. Lampe

Development of methodology: C.B. Garrison, P.D. Lampe, J.W. Lampe

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): C.B. Garrison, M.L. Neuhouser

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): C.B. Garrison, Y. Zhang, S.L. Navarro, T.W. Randolph, M. Kratz, P.D. Lampe, J.W. Lampe

Writing, review, and/or revision of the manuscript: C.B. Garrison, S.L. Navarro, T.W. Randolph, M.A.J. Hullar, M. Kratz, M.L. Neuhouser, D. Raftery, P.D. Lampe, J.W. Lampe

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M.L. Neuhouser, J.W. Lampe

Study supervision: M.L. Neuhouser, P.D. Lampe, J.W. Lampe

This work was funded through the NIH [R01 CA192222, U54 CA116847, P30 CA015704].

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.