Proteomics reveals the effects of sustained weight loss on the human plasma proteome

Abstract

Sustained weight loss is a preferred intervention in a wide range of metabolic conditions, but the effects on an individual's health state remain ill-defined. Here, we investigate the plasma proteomes of a cohort of 43 obese individuals that had undergone 8 weeks of 12% body weight loss followed by a year of weight maintenance. Using mass spectrometry-based plasma proteome profiling, we measured 1,294 plasma proteomes. Longitudinal monitoring of the cohort revealed individual-specific protein levels with wide-ranging effects of losing weight on the plasma proteome reflected in 93 significantly affected proteins. The adipocyte-secreted SERPINF1 and apolipoprotein APOF1 were most significantly regulated with fold changes of -16% and +37%, respectively (P < 10^-13), and the entire apolipoprotein family showed characteristic differential regulation. Clinical laboratory parameters are reflected in the plasma proteome, and eight plasma proteins correlated better with insulin resistance than the known marker adiponectin. Nearly all study participants benefited from weight loss regarding a ten-protein inflammation panel defined from the proteomics data. We conclude that plasma proteome profiling broadly evaluates and monitors intervention in metabolic diseases.

Keywords: diabetes; mass spectrometry; metabolic syndrome; obesity; plasma proteome profiling.

Figure 1. Study design and the plasma proteome profile pipeline

1. The study cohort consisted of 52 obese individuals, who lost on average 12% of their body mass during 8 weeks of calorie restriction (800 kcal/day). The acute weight loss was followed by a 52‐week weight maintenance period by 43 of the study participants with longitudinal blood sampling at the indicated time points.

2. Quadruplicates of the samples and the establishment of a matching library resulted in 1,294 plasma proteomes, which were separately prepared by an automated liquid handling platform. The LC‐MS/MS data, which we analyzed by MaxQuant and Perseus, resulted in 319 individual plasma proteome profiles for 52 participants.

Figure EV1. MS‐based measurements and sample quality

1. Total MS measuring time leading to usable raw data or lost to HPLC issues, MS cleaning, or other problems. The 6% cleaning time for the MS instrument resulted from a single cleaning instance, which, however, subsequently led to a 5‐day downtime.

2. Analysis of the reproducibility between samples by color‐coded pairwise comparison (Pearson correlation coefficients) of 1,276 samples (without the matching library). The high reproducibility within quadruplicates and the individuals can be seen by the red diagonal, indicating high correlation.

3. Levels of four highly abundant erythrocyte‐specific proteins were used to assess the sample quality with regard to erythrocyte lysis. The red arrow indicates the only sample with strong erythrocyte lysis.

4. Displaying the levels of fibrinogens allows the detection of (partial) blood coagulation. Five cases with decreased fibrinogen levels and therefore coagulation events are highlighted by red arrows.

Figure 2. Individual‐specific plasma protein levels

1. Coefficients of variation (CVs) for all proteins were calculated in all participants for the five longitudinal samples. Combination of these longitudinal CVs and fold differences allows estimation of how many proteins are individual‐specific.

2. CVs plotted against the log₂‐fold difference of one participant compared to the average label‐free quantitation (LFQ) intensity of the study cohort. Considering a fivefold difference and a CV below 30% yields the proteins in the blue boxes as specific for this participant.

3. LFQ intensities of seven proteins plotted during weight maintenance (weeks 4–52) for all 43 individuals. These proteins are stable over time within individuals, but strongly vary between individuals. A2M: alpha‐2‐macroglobulin; PZP: pregnancy zone protein; IGJ: immunoglobulin J chain; LPA: apolipoprotein(a); FETUB: fetuin‐B; CRP: C‐reactive protein; GDI1/GDI2: Rab GDP dissociation inhibitor alpha/beta.

Figure 3. The effect of weight loss

1. Volcano plot of proteomes before (week −8) and directly after weight loss (week 0), with the x‐axis depicting the fold change in protein levels and the y‐axis the –log₁₀ t‐test P‐value of the quadruplicates.

2. Changes in protein abundance are shown for each individual. LFQ intensities from before (yellow) and after weight loss (green) are connected by a line to highlight individual‐specific protein levels and their changes. The median increase or decrease for each protein over the population is indicated in red.

Figure 4. Long‐term effects of weight loss on the plasma proteome profile

Hierarchical clustering of Z‐scored median LFQ intensities for highly significant proteins (P < 0.0005) resulted in seven longitudinal weight regulated protein clusters. Scale bar for Z‐scored weight is displayed below the protein clusters. Insets show Z‐scores for proteins in different clusters as a function of time, color‐coded for the distance from the center. The highlighted protein names with the same color indicate functionally connected proteins (red: inflammatory markers; brown: serine protease inhibitors; orange: complement system; blue: apolipoproteins; purple: anti‐inflammatory‐acting proteins; green: steroid transport proteins).

Figure EV2. Proteins with significant long‐term effects due to weight loss

1. Proteins with a significant increase in at least five out of six time points, comparing before and after weight loss.

2. Proteins with a significant decrease over a long time period due to weight loss.

3. Long‐term behavior of specific proteins. The means are plotted with SEM as error bars over time. Upregulated proteins are indicated in orange and downregulated in blue.

Figure 5. Plasma proteins and the longitudinal inflammation profile

1. A

   Correlation of body mass index (BMI) with all quantified proteins in our dataset.

2. B–F

   Correlation analysis of the indicated clinical parameter with plasma protein levels over time.

3. G

   For each individual, Z‐scores were calculated for each of the proteins of the ten‐protein panel over the seven time points. The proteins were arranged in the indicated order and hierarchical clustering was performed on the level of the different individuals, resulting in a longitudinal inflammation profile.

4. H

   Dot plots for the ten‐protein panel in the same order as in (G) for the central cluster. The black line indicates the median Z‐score of the inflammation panel for each time point.

Data information: Significant proteins are displayed by red dots, and non‐significant ones with gray dots. Red letters indicate inflammation factors that correlate with the BMI and that were used to generate panel (G). A Benjamini–Hochberg FDR of 0.05 was used for significance in all correlation analyses (A–D).

Figure 6. Insulin resistance and systemic inflammation

1. The five proteins with the highest positive and the four proteins with the highest negative correlation with IR were used to define a pro‐ and an anti‐IR panel. These panels separated the study cohort in a high and a low IR group and the 24 individuals that were present in at least one of the IR panels are highlighted (numbered in red at the y‐axes). Likewise, the ten‐protein inflammation panel separates the cohort in individuals with high and low systemic inflammation levels and individuals with high levels were highlighted (numbered in red). Of 16 study participants with high inflammation levels, 14 were also present in the high IR group as illustrated by a Venn diagram, indicating a high metabolic burden group as defined by plasma proteome profiling.

2. HOMA‐IR levels and the BMI are compared between the high and the low metabolic burden group with indicated changes in percent at the study endpoint. These changes are linked to longitudinal changes in the anti‐IR, the pro‐IR, and the inflammation profile. The means are plotted with SEM as error bars over time.

Figure 7. Effect of weight loss on the apolipoprotein family

1. The initial LFQ intensity before weight loss (week −8) was set to 100% to normalize protein abundance within each participant to account for individual‐specific protein levels. Colors derive from clusters in Fig 4. The means are plotted with SEM as error bars over time. For the acute phase protein SAA1, the peak at week 13 was caused by very high levels in one individual (gray curve and right y‐axis in the 6^th panel). Excluding this individual results in the red curve. Numbers below the protein name refer to their presence in the lipoprotein particle in panel (C).

2. Annotation for gene ontology cellular component (GOCC) was Z‐scored and filtered for main lipoprotein particles.

3. Gene ontology biological process (GOBP) annotations were filtered for the keywords lipid, lipoprotein, fat, and cholesterol, leading to the displayed group of GOBPs, which decreased due to weight loss.

Figure EV3. MS‐measured protein intensities compared to standard clinical measurements

1. Correlation of APOA1 with HDL (according to the literature each HDL particle should contain approximately one APOA1 molecule). The imperfect correlation can partially be explained by the lack of resolution of the reported HDL values, which can be seen by the non‐continuous values for the individuals, resulting in vertical lines.

2. MS‐measured APOB intensities were correlated with LDL values (each LDL particle should contain one APOB molecule).

3. Correlation of APOB intensities with cholesterol measurements.

Figure EV4. Hierarchical clustering of lipid metabolism‐related GOBP terms

GOBP terms annotations were Z‐scored over time and filtered for the keywords lipid, lipoprotein, fat, and cholesterol. Hierarchical clustering indicates lipid metabolism‐related biological processes in response to weight loss and maintenance.