Stratifying cellular metabolism during weight loss: an interplay of metabolism, metabolic flexibility and inflammation

Stratification of the metabolic profiles

Using the affinity network fusion (ANF) algorithm for clustering we generated a two-cluster system with samples across individuals and time points clustered with each other. The primary purpose of this clustering was to cluster the transcriptomics and proteomics expression patterns of the 291 metabolic flexibility related genes and proteins. The clustering thus generated two expression profiles, with individuals changing cluster membership over the duration of the study (visualised in Fig. 1A). Figure 1B shows the respective individuals in each cluster membership pattern found. In this subfigure, we can see that at the beginning of the caloric restriction (TP1) the majority of individuals are in Cluster B compared to Cluster A. Immediately after weight loss, at time point 2 (TP2), the cluster membership for Cluster A surges to become larger than Cluster B, with both clusters settling close to even after weight maintenance (TP3). This indicates that the metabolic profile of a lar ge number of individuals changed during weight loss with some of them maintaining that change whereas others reverted to their original profile.

Figure 1

(A) Sankey diagram showing the proportion of individuals changing or not changing clusters across the time points of the dietary intervention. Eight individuals had missing samples at different time points and thus, were not clustered at those time points. These samples are shown as black blocks at the respective time points. (B) The various cluster membership patterns of the thirty individuals observed across the three time points, as well as the respective number of individuals following said pattern.

We next compared the phenotypic/clinical measurements between the clusters with 101 samples instead of 106 due to missing/incomplete data for two individuals. We identified significant differences in sex, height, BMI, fat mass, fat free mass, waist size and hip circumference. From the forearm venous plasma measurements, total cholesterol, insulin, free fatty acids, triglycerides levels as HOMA-IR index were found to be significantly different between the clusters. However, none of the adipose tissue arteriovenous flux measurements were found to be significant post multiple testing correction. Lastly, the s-value, defined as a value for weight regain and/or maintenance, was also not found to be significantly different between the two clusters. These results are collectively provided as Supplementary file 3.

Differential expression analysis of the 18,113 gene transcriptomics data between the samples of the two clusters showed that only 1,343 genes were differentially expressed between the two clusters (|Fold change| ≥ 1.2 and p-value < 0.05). 150 of these differentially expressed genes had an absolute fold change greater than 1.5, and 28 of those had it greater than 2. The number of differentially expressed genes have been tabulated in Table 1. A gene ontology enrichment analysis34 on the 1,343 genes revealed that apart from the expected metabolic processes (fatty acid and other lipid metabolism, oxidoreductase activity, chemical homoeostasis etc.) there were processes related to tissue morphology and inflammatory response also being enriched. Figure 2 shows a pie chart of the gene ontology terms that were enriched.

Table 1 The number of differentially expressed genes across different groups for absolute fold changes of 1.2, 1.5 and 2. All counted genes are significant at p-value < 0.05. Cluster A_st or Cluster B_st: individuals staying in the respective cluster A or B throughout the dietary intervention. Comparisons 1–5 are illustrated in Fig. 3. Figure 2

Pie chart showing the proportions of genes enriched for each of the gene ontology term for the differentially expressed genes between Cluster A and B. The differentially expressed genes were significant at p-value < 0.05 and an absolute fold change of at least 1.2.

A deeper look at the differentially expressed genes in the context of biological pathways (using PathVisio35 for pathway overrepresentation analysis) also revealed several metabolic pathways being differentially affected between the clusters. It appears that the overall cellular metabolism in Cluster A is decreased across the board; in the pathway overrepresentation results, the majority of the genes in the electron transport chain, fatty acid biosynthesis, fatty acid beta oxidation, glycolysis, and citric acid pathways are expressed lower (at least with a fold change of 1.2) in Cluster A compared to Cluster B. However, these are to be expected due to the way the ANF algorithm constructed the clusters based on the 291 metabolic flexibility-, and thus metabolism-related genes. On the other hand, we observed that pathways associated with the immune system and tissue restructuring were expressed higher in Cluster A compared to Cluster B. In particular, Complement C6 and C7, inflammat ory interleukins IL1B and IL8, and matrix metalloproteinases MMP2 and MMP9 were found to be expressed higher in Cluster A. The differential expression analysis results, gene ontology enrichment results and pathway overrepresentation results are provided as Supplementary files 4–6 respectively.

Cluster membership patterns Cluster membership changers

Figure 1B shows the number of individuals in each cluster membership pattern found. We observed that a total of 17 individuals change their cluster membership from Cluster B to Cluster A when moving from time point 1 (TP1) to time point 2 (TP2); 9 of these individuals then stay in Cluster A until time point 3 (TP3), while the remaining 8 revert to their original cluster. We further analysed this pattern in the following comparisons (illustrated in Fig. 3; numbers of differentially expressed genes provided in Table 1).

Figure 3

Illustration of the 17 subjects which switched from Cluster B to Cluster A during weight loss, after which 8 of the subjects reverted to Cluster B. The five comparisons used to break down and analyse the changes in the gene expression are also labelled.

Comparison 1

We first performed a paired differential expression analysis of all 17 individuals of this pattern between TP2 and TP1 to see the first changes that occurred when the individuals changed their cluster membership. A total of 1,286 differentially expressed genes were found in this comparison at absolute fold change equal to or greater than 1.2 and p-value less than 0.5. Gene ontology enrichment of these showed a similar pattern to the gene ontology enrichment when comparing the two clusters as a whole – the processes were broadly categorised into metabolic, immune/inflammatory and tissue morphology. This is understandable as these changes would be the reason for these 17 individuals being clustered in Cluster A at TP2 as opposed to Cluster B at TP1. Similarly, the pathway enrichment results also follow the same pattern as that of the whole cluster comparison. Genes in metabolic processes such as fatty acid beta oxidation, electron transport chain complexes, glycolysis, lipogenesis , gluconeogenesis and amino acid metabolism were down regulated at TP2 Cluster A compared to when these individuals were in Cluster B at TP1. Genes involved in inflammation, on the other hand, were found to be upregulated in the same pattern as the whole cluster comparison.

Comparison 2

We performed a paired differential expression analysis between TP3 and TP2 for the 9 individuals that retained their new cluster to see what changes occurred even when the cluster membership did not change. In this comparison, we found 282 differentially expressed genes. Considering that this comparison is between the 9 individuals which stay in Cluster A at both TP2 and TP3, the low number is expected. Consequently, the gene ontology enrichment and pathway overrepresentation analysis also do not show any major differences between these samples. Of the minor differences found, peroxisome proliferator activated receptor gamma (PPARγ), fatty acid synthase (FASN) and stearoyl-CoA desaturase (SCD) were found to be upregulated in these individuals at TP3 compared to TP2. This is interesting, especially in the light of our previous studies covering the role of these proteins in metabolic flexibility. However, the lack of strong results from the two analyses overall make it difficult to infer the effects of their upregulation concretely.

Comparison 3

We performed a paired differential expression analysis between TP3 and TP2 for the 8 individuals that changed back to the original cluster to see what changes occurred when these individuals returned to their original cluster. In this comparison, we found 897 differentially expressed genes. The gene ontology enrichment analysis for these genes showed that these were also tied to cellular metabolic processes, with two immune system related terms also enriched. Compared to the previous two parts, the individuals in this analysis showed a reversed trend where the immune system related pathways were being downregulated at TP3 compared to TP2. The human complement system pathway and toll-like receptor (TLR) associated pathways showed downregulation of respective components. The matrix metalloproteinase MMP9 was also strongly downregulated. On the other hand, upregulated genes showed an upregulation of metabolism related processes across the board with lipid metabolism and biosynthesis showing the highest z-scores in pathway overrepresentation. Sterol regulatory element-binding protein (SREBP) signalling also showed strong upregulation compared to both TP2 in these individuals, as well as the previous two parts. Also in contrast to the previous two parts, as well as the whole cluster differential expression, was the upregulation of leptin (LEP) and downregulation of leptin receptor (LEPR). In addition, glycolysis and gluconeogenesis enzymes were mostly found to be slightly upregulated.

Comparison 4

We performed a differential expression analysis between the 17 individuals at TP3 to see which changes were retained by these individuals by the final time point. In this comparison, a total of 669 genes were found to be differentially expressed genes between 17 individuals, 8 at TP3 in Cluster B and 9 at TP3 in Cluster A. The gene ontology enrichment showed a limited number of terms, all associated with cellular metabolism. Pathway overrepresentation showed genes from several immune system related processes, such as interferon signalling and interleukin signalling to be downregulated. Metabolism associated processes, on the other hand, were found to be upregulated. Overall, the results of this analysis mirror those of the previous part.

Comparison 5

We performed a paired differential expression analysis between the 9 individuals at TP3 and TP1 to see the changes in metabolic processes, if any, between their old cluster at TP1 and their new cluster at TP3. In this comparison, we observed a total of 660 differentially expressed genes, with the gene ontology enrichment and pathway overrepresentation results following the same pattern as that in Comparison 1. The reduction in the number of differentially expressed genes only affected the intensity of the gene expression and the z-score in the overrepresentation analysis. Due to this, the pathways results are difficult to interpret as considerable parts of many pathways have varied expression patterns between the individuals and high p-values.

Cluster membership maintainers

We also observed that a number of individuals do not change their cluster membership at all and stay within Cluster A (4 individuals) or Cluster B (8 individuals) throughout the study. The differential gene expression analysis between individuals staying in Cluster A and individuals staying in Cluster B yields the largest number of differentially expressed genes, at 2,838, compared to the previous analyses. Gene ontology enrichment analysis showed a combination of metabolic, inflammatory and tissue morphology terms that were enriched. The pathway overrepresentation showed that individuals staying in Cluster A persistently had a lower overall expression of cellular metabolism with various enzymes across lipid metabolism, SREBP signalling, mitochondria electron transport chain, TCA cycle, and glycolysis being downregulated compared to individuals in Cluster B. Immune system processes, such as components of the human complement system and interleukin signalling were found to be upreg ulated in Cluster A. In addition, matrix metalloproteinases MMP2, MMP7 and MMP9 were also found to be upregulated in Cluster A.

The differential expression analyses output, gene ontology enrichments output, as well as the pathway overrepresentation analysis output for the aforementioned results are provided as Supplementary files 4–6 respectively.