GOTO lifestyle intervention trial
The Growing Old TOgether (GOTO) trial, was a 13-week combined lifestyle intervention with 164 participants, of whom 163 completed the trial. One participant withrew from the trial due to knee surgery. The primary outcome of the trial was a significant reduction in fasting insulin, these effects have been reported in ref. 1. The GOTO trial is completed.
Baseline characteristics of the GOTO participants were similar among subsets
At baseline, the participants of the GOTO trial were non-diabetic overall healthy older adults, with a mean age of 63 for males and 62 for females (Table 1). Some of the participants were overweight and/or had elevated blood pressure. Transcriptome data are available at baseline and post intervention for different sets of participants of the complete GOTO trial group, i.e.: (1) 57 participants (33 males, 24 females) with transcriptome data of postprandial blood, SAT, and muscle samples, (2) 88 blood postprandial samples, containing 45 males and 43 females, (3) 78 SAT postprandial samples, which contains 38 males and 40 females, and 4) 82 muscle postprandial samples, 48 males and 34 females (Supplementary Fig. 1).
Table 1 Baseline measurements of metabolic health marker measurements per subset selection
Between the participants in the different subsets, there was only a slight variation in baseline measurements of the metabolic health markers (Table 1). There were differences between the baseline characteristics of males and females, especially in whole-body fat%, trunk fat%, HDL cholesterol level, which measurements were all higher in females than in males: 38.16%, 37.4% & 1.64 mmol/L versus 26.5%, 27.17% & 1.36 mmol/L, respectively. Fasting insulin, the primary outcome of the GOTO trial, was higher at baseline in males than in females: 9.42 mU/L versus 8.35 mU/L. These differences in baseline metabolic health between the two sexes, along with earlier published results on different activity patterns and differences in the intervention effect on metabolic health between the sexes1,41, prompted us to perform the analysis in a sex-stratified manner.
Health markers improve but improvement varies among subsets
Participants compliant to the intervention overall improved their metabolic health, with for some parameters slight sex differences in magnitude of effect (Table 2). The direction of effect on the health parameters in the subsets of participants for which data in different tissues is available was the same, though there were some differences in significance level. For example, in males, the intervention only had a significant effect on HDL cholesterol in the group with muscle transcriptome measurements. While in females, HDL cholesterol size was significantly different in only the participants with SAT transcriptome measurements, and the Framingham Risk Score (FRS) was significantly different in the complete set of measurements. These differences in effect could be partially explained by slight baseline classical metabolic health marker level differences between the groups (Table 1). To circumvent the slight differences in intervention effect on the classical metabolic health markers between the different groups, the associations between the transcriptome and the classical metabolic health markers will only be calculated on the participants with transcriptome data available in all three tissues (Supplementary Fig. 1).
Table 2 Effect of the 13-week lifestyle intervention on metabolic health marker measurements differs per subset selection
Strong sex-dependent transcriptional response in subcutaneous adipose tissue and muscle
We investigated the effect of the GOTO intervention by comparing the transcriptome levels post intervention to the baseline transcriptome levels matched for each participant. We studied the transcriptional differences in blood, SAT, and muscle (Fig. 1, differential gene expression analysis, see “Methods”). In males, the number of significantly differentially expressed genes (DEGs) across the intervention varied considerably between blood (1), SAT (89), and muscle (251) (Fig. 2, Table 3, and Supplementary Datasets 1–3). The ratio between up and downregulated DEGs also differed between the affected tissues: a slightly larger number of downregulated genes in SAT whereas over 80% of the DEGs were upregulated in muscle. Noteworthy, the muscle DEGs contained 14 collagen genes, all of which were upregulated (Supplementary Dataset 3).
Fig. 1: Outline of the transcriptomic analysis of the Growing Old TOgether trial.
A The participants of the Growing Old TOgether (GOTO) trial changed their lifestyle for a duration of 13 weeks. Both before and after the intervention measurements were performed, including anthropometrics, Dual-energy X-ray absorptiometry and blood metabolite levels. In addition, following a standardized nutritional challenge, postprandial blood, subcutaneous adipose tissue (SAT) and muscle tissue were sampled. Blood, SAT and muscle RNA-seq levels were studied and a differential analysis was performed to study the effect of the intervention on these three tissues. Subsequently, a pathway overrepresentation analysis was performed. B To study which tissue best represented the metabolic health changes, the association between the RNA-seq levels of the overrepresented pathways of the three tissues and metabolic health marker levels was studied. C Joint and Individual Variation Explained (JIVE) was applied to investigate whether there was a joint variation between the expression levels of the three tissues. JIVE decomposed the expression levels of each tissue (X) into the joint (J), individual (A), and the residual (R) effect, with the joint effect representing the shared variation across the expression levels of the three tissues. Subsequently, the drivers of the joint effect in blood were identified. Lastly, the association between the joint effect drivers in blood and the metabolic health markers was studied and compared to the postprandial blood individual effect expression levels (A).
Fig. 2: Intervention effect on postprandial blood, SAT and muscle gene expression levels in men and women.
The effect of the intervention was studied using two-sided linear mixed model analysis. The significance levels were adjusted for multiple testing using the false discovery rate (FDR). The log2 fold change is plotted on the x axis, the -log10 FDR-adjusted P value is plotted on the y axis. The dashed horizontal line represents the significance threshold of FDR 0.01, the two vertical lines represent the log2 fold change thresholds of −0.25 and 0.25. Genes with an absolute log2 fold change ≥0.25 and an FDR-adjusted P value < 0.01 were considered significantly differentially expressed. Each dot represents a gene. Yellow dots represent the significantly differentially expressed genes (DEGs). Gray dots represent genes with an absolute log2 fold change < 0.25 and/or FDR ≥ 0.01. Significantly differentially expressed genes with the strongest log2 fold changes were labeled with their gene symbol. SAT subcutaneous adipose tissue.
Table 3 The 13-week lifestyle intervention had the strongest effect on postprandial SAT and muscle transcriptomics
For females, the number of DEGs across the different tissues were somewhat similar as in males, but the effect sizes were smaller (Fig. 2 and Table 3): zero DEGs in blood, 168 in SAT, and 168 in muscle (Supplementary Datasets 4 and 5). Like the male samples, more than half of the DEGs in SAT were downregulated, while in muscle over two-thirds of significant DEGs were upregulated. The female muscle DEGs contained 7 collagen genes, all of them upregulated (Supplementary Dataset 5).
Overall, in both males and females, the intervention had hardly any effect on the blood transcriptome, while in SAT both up- and downregulated genes were observed, whereas in muscle the majority of the DEGs were upregulated.
Lipid and collagen-related pathways enriched in SAT and muscle response, respectively
We then investigated which pathways were associated with the found DEGs (pathway overrepresentation analysis, see Methods section). In male SAT, five pathways were significantly overrepresented (Fig. 3 and Supplementary Dataset 6). The three upregulated pathways were involved with HDL remodeling, nuclear receptor signaling and cholesterol transport, while the downregulated pathway was involved in fertilization, the remaining pathway (equally down -and upregulated) was involved in ligand clearance. In male muscle, all 18 enriched pathways were upregulated. The majority of these were involved in collagen formation or extracellular matrix formation/degradation (Fig. 3 and Supplementary Dataset 6). The remaining pathways were involved in myogenesis (NCAM1 pathways), mesenchymal stem cell differentiation (PDGF pathway), scavenger receptors (binding and uptake, class A receptors), and protein O-glycosylation.
Fig. 3: Significantly overrepresented pathways in postprandial SAT and muscle transcriptome.
A one-sided functional enrichment test was performed to identify the overrepresented pathways. The overrepresented pathways are plotted on the y axis. Overrepresented pathways in the male samples are plotted on the left, overrepresented samples in female samples, on the right. The fill color represents the normalized direction of effect: dark blue ( − 1) all significantly differentially expressed genes in the pathway were underexpressed, dark red (1) all significantly differentially expressed genes in the pathway were overexpressed. P values were adjusted for multiple testing using the Bonferroni correction method. Significance is indicated by the asterisks (*P adjust <0.05, **P adjust <0.01, ***P adjust <0.001). SAT subcutaneous adipose tissue.
In female SAT, seven pathways were significantly overrepresented (Fig. 3 and Supplementary Dataset 6). The three upregulated pathways were involved in signaling pathways and HDL remodeling. The four downregulated pathways were related to MAPK signaling and CO2/O2 exchange. However, the three CO2/O2 pathways were only overrepresented due to the same three genes (Supplementary Dataset 6). In female muscle, all eight overrepresented pathways were upregulated (Fig. 3 and Supplementary Dataset 6), six of which were involved in collagen formation and extracellular matrix organization/degradation. The remaining pathways were involved in in netrin-1 signaling and osteoblast differentiation.
Taken together, both in males and females, the overrepresented pathways in muscle were largely involved in collagen and extracellular matrix remodeling. In SAT, there were fewer enriched pathways which were less consistent in terms of direction (both -up and downregulated). The overrepresented pathways in SAT were involved in signaling, O2/CO2 exchange, HDL remodeling, and signaling.
Response of tissue-specific overrepresented pathways reproduced across SAT and muscle, not in blood
Next, we questioned whether there was a transcription response common to the three tissues. Hereto, we examined the response of the mean expression of the DEGs within each overrepresented pathway of SAT and muscle (Fig. 3). For example, the mean expression of the male-upregulated SAT DEGs involved in the HDL remodeling pathway was also significantly positively upregulated in muscle, whereas it was non-significantly upregulated in blood (second row in top panel of Fig. 4). Similarly, in females, we noted correspondence between SAT and muscle, albeit weaker than in males. The HDL remodeling pathway was weakly upregulated in both SAT and muscle but had no response in blood (second row, third panel from the top, Fig. 4). Since there was only one significant DEG in postprandial blood (Fig. 2 and Table 3), no overrepresented blood down- or upregulated pathways were present.
Fig. 4: Significantly overrepresented pathway expression levels in SAT and muscle are significantly associated with the intervention time point and metabolic health markers.
Associations between the overrepresented pathways and metabolic health markers were calculated using a two-sided linear mixed model. Overrepresented pathways are plotted on the y axis. Intervention time point and metabolic health markers are plotted on the x axis. Colors indicate the estimated effect between the mean expression in the pathways and the intervention effect/metabolic health markers: blue, a negative effect, white, no effect, red, a positive effect. P values were corrected for multiple testing using the false discovery rate. The significance level is indicated by the asterisks: *FDR <0.05, **FDR <0.01, ***<0.001.
In general, we noted a similar behavior of the overrepresented postprandial SAT pathways across the three tissues, both in males and females (Fig. 4). In addition, two overrepresented SAT pathways (HDL remodeling, NR1H3 & NR1H2 gene regulation), were significantly positively associated with the intervention time point in postprandial muscle expression of males. In contrast to the overrepresented SAT pathways, the intervention had an opposite effect on the overrepresented muscle pathways in the SAT transcriptome, when compared to the muscle expression levels. In addition, the ECM proteoglycans pathway, which was significantly upregulated in muscle, had a significant negative effect in SAT (in males only). The intervention had a weak effect on the remaining overrepresented muscle pathways, in either sex. Lastly, the intervention had hardly an effect on the postprandial blood expression levels of any of the overrepresented pathways, showing that the genes that were differentially expressed in SAT and/or muscle were not changed by the intervention in blood.
In summary, when investigating the intervention effect of the genes strongest affected by the intervention, no overlap of transcriptional response was observed between blood and SAT or muscle. Some relationships did exist within the transcriptional response between SAT and muscle, although not always consistent in the direction of effect.
At baseline, the transcriptomes of postprandial blood, SAT, and muscle are weakly associated with the metabolic health status
To investigate the interplay between the transcriptome and metabolic health at baseline, we calculated the association between the baseline mean expression of the overrepresented pathways to the baseline measurements of ten clinical metabolic health markers: I) two DEXA scan body composition measurements: whole-body fat%, trunk fat%; II) seven traditional metabolic health markers: BMI, Waist Circumference (WC) fasting insulin, fasting HDL cholesterol, fasting HDL cholesterol size, SBP, DBP; and III) the Framingham Risk Score (FRS), which is a cardiovascular risk score (Supplementary Fig. 2).
To account for the differences in the baseline metabolic health marker levels between the different subsets (Table 1), we performed this analysis only on the participants with transcriptome measurements in all three tissues (57; 33 males, 24 females) (Supplementary Fig. 1).
At baseline, none of the associations between the expression levels of the different pathways in the three tissues and health marker levels were significant. However, some trends were present. In males, the SAT transcriptome was strongest associated to the metabolic health marker levels. In SAT, three of the overrepresented pathways (nuclear receptor signaling, HDL remodeling, cholesterol transport) had a negative association with whole-body and trunk fat%, BMI, WC, fasting insulin, the FRS and a positive association with the HDL cholesterol level and size. Interestingly, the pathway scavenger receptor ligand binding and all the pathways overrepresented in muscle, showed an opposite trend with metabolic health, compared to the three aforementioned SAT pathways. The blood transcriptome was weakly associated to the metabolic health status. However, the association between the blood expression levels and metabolic health, was opposite to the SAT transcriptome associations reflecting that changes in blood cells poorly relate to those in the tissues. Lastly, the expression levels of muscle showed a weak association to the metabolic health levels, except for a positive association between signaling by the nuclear receptor pathway, the HDL measurements and DBP.
In females, the overrepresented SAT pathways were strongest positively associated with metabolic health markers when expressed in postprandial blood. The effect sizes were largest between blood expression of the signaling pathways, O2/CO2 exchange pathways, and whole-body fat%, trunk fat%, DBP and SBP. In postprandial SAT, only the pathway MAPK family signaling cascades showed a positive association with whole-body fat%, trunk fat%, BMI, and WC. In postprandial muscle, the overrepresented muscle pathways showed a weak negative association with the BMI, WC, fasting insulin and the body composition measurements.
Taken together, the baseline expression levels in none of the three tissues were significantly associated to the metabolic health status. However, when comparing the three tissues, the postprandial SAT transcriptome best captured the baseline metabolic health status in males, while in females the postprandial blood transcriptome was strongest associated with the baseline metabolic health status.
Especially SAT- and muscle-specific overrepresented pathways associate with health markers across the intervention
To better understand the differential expressed transcriptome, we investigated the associations between the mean expression of the overrepresented pathways and the metabolic health markers across the two time points of the intervention (“Methods”, Fig. 4). Since the intervention effect on some of the health markers differed between the subgroups (Table 2), which can cause differences in effects due to differences in subsets, the relation between the expression data and metabolic health markers was performed on the overlapping blood, SAT and muscle samples (57; 33 males, 24 females) (Supplementary Fig. 1).
In general, the postprandial blood expression levels were weakly associated to the metabolic health markers (Fig. 4). However, in males, the postprandial blood expression levels of two pathways (HDL remodeling, cholesterol transport) were significantly negatively associated to trunk fat%, while two other pathways (ligand binding & collagen fibril crosslinking) were significantly positively associated to trunk fat% and the fasting HDL cholesterol size. In addition, in females, the pathways involved in O2/CO2 exchange were significantly positively associated to the changes in DBP, while DBP and netrin-1 signaling were significantly negatively associated.
The postprandial SAT expression levels of the majority of both SAT and muscle-overrepresented pathways were significantly associated to the change in DEXA body composition parameters (whole-body fat%, trunk fat%), three of the seven traditional metabolic health markers (BMI, WC and fasting HDL cholesterol level) and the FRS, in both sexes, and with the fasting HDL cholesterol size, DBP and SBP only in men. The SAT pathways involved in nuclear receptor signaling, HDL remodeling cholesterol transport and signaling transduction were significantly negatively associated to whole-body fat%, trunk fat, BMI and WC, and positively associated to the HDL cholesterol measurements. Conversely, the remaining SAT pathways (fertilization, ligand binding and uptake, MAPK cascades) showed an opposite trend with similar strength of effect. The expression levels of the muscle-overrepresented pathways were positively associated to all metabolic health markers, excepting the fasting HDL level and size. Interestingly, only the overrepresented pathways of the muscle tissue were significantly positively associated to the two blood pressure measurements.
The transcriptome of postprandial muscle was weakly associated to the changes in the metabolic health markers. In males, pathways involved in HDL remodeling, NR1H3 & NR1H2 gene regulation, and TSR O-glycosylation and PpS, were significantly negatively associated to BMI and WC, respectively. In females, there was one significant negative association between the pathway nuclear receptor signaling and fasting HDL cholesterol size.
Interestingly, in males, the expression levels of HDL remodeling and NR1H3/NR1H2 gene regulation were significantly associated to weight-related metabolic health markers in all three tissues. Moreover, the direction of effect between the expression of these pathways and trunk fat%, whole-body fat%, WC, and BMI was negative in all three tissues.
Compared to the baseline associations between the pathway expression and metabolic health marker levels (Supplementary Fig. 2), the effects were stronger across the intervention for both the SAT and muscle transcriptomes. Contrarily, in postprandial blood, the expression of the overrepresented muscle pathways was strongly associated at baseline than across the intervention. The direction of effect remained the same for the overrepresented SAT pathways between the two analyses, while the associations between the muscle-overrepresented pathways and part of the health markers (whole-body fat%, trunk fat%, BMI, WC) were opposite at baseline and across the intervention, when expressed in blood and muscle.
Taken together, especially in the SAT transcriptome, the SAT and muscle-overrepresented pathways reflected changes in DEXA body composition markers (whole-body and trunk fat%), classical metabolic health markers (BMI, WC, fasting HDL cholesterol level and size, DBP, SBP) and the FRS. Many effects were similar in males and females, although more pronounced in males.
Integrated analysis reveals male postprandial blood, SAT, and muscle transcriptome share a joint variation
To explore the tissue dependency in a more intricate way, we captured the joint transcriptional reaction to the intervention across the three different tissues using Joint and Individual Variation Explained (JIVE) analysis (“Methods”, Supplementary Fig. 3). The joint representation is a linear subspace spanned by the genes, analogous to a principal component analysis (PCA), but that now also explains the joint variation in expression in all three tissues across the intervention (“Methods”). Hence, although informed by all three tissues, the blood expression of a participant (before and after the intervention) can be mapped to this joint subspace without needing information about the muscle or SAT transcriptome. In other words, the blood transcriptome of a participant is decomposed in three (independent) transcriptome terms: (1) a term that reflects the joint transcriptomic variation in blood, SAT, and muscle, (2) a term reflecting the remaining structured variation in blood only (also referred to as the individual variation), and (3) a term that represents the residual noise in the blood transcriptome, i.e., the signal that is not explained by the joint and individual terms. Subsequently, we can measure the differential expression across the intervention in either the joint subspace (of blood, SAT, and muscle) or the individual subspaces (for each tissue type). The number of joint and individual components are determined through permutation testing (“Methods”).
In the male samples, JIVE identified 2 joint components for the shared transcriptomic variation in blood, SAT and muscle, and 11, 15, and 15 individual components for the blood, SAT, and muscle, respectively (Supplementary Dataset 7). In postprandial blood and SAT, the joint components together explained 8% and 7.2% of the total variation in those tissues, respectively. In the postprandial muscle transcriptome, both joint components explained 31.3% of the variation. In female samples, JIVE did not identify a joint term that was independent of the individual terms (Supplementary Fig. 4).
In male samples, the drivers of the joint components were involved in phospholipid metabolism and the immune system
To investigate what genes expressed in postprandial blood were the drivers of the joint effect, the 250 genes that had the largest loading factors for each of the two joint components were selected, respectively (“Methods”). Both sets of genes were subsequently used in a pathway enrichment analysis. In postprandial blood, the genes driving the first joint component (JC1) that explains most of the joint variance across the tissues were involved in phospholipid metabolism, while the genes driving the second joint component (JC2) were involved in interferon signaling, inflammation, and other aspects of the immune system (Fig. 5).
Fig. 5: Significantly overrepresented pathways in the top 250 drivers of the postprandial blood joint effect.
A one-sided statistical test was performed to identify the overrepresented pathways. P values were adjusted for multiple testing using the Bonferroni correction method. The overrepresented pathways are plotted on the y axis. Y axis is ordered by significance, the strongest significantly overrepresenting pathways first. The x axis represents -log10 of the Bonferroni adjusted P value.
Phospholipid metabolism genes that drive the joint effect also capture the intervention effect individually
Next, we wondered whether the genes driving the significant effects of JC1 and JC2 also showed an intervention effect by themselves. Within both sets of 250 genes that had the highest loading factors for each of the joint components, we selected those genes that overlapped with the found enriched pathways. Interestingly, we found a remarkable difference between the intervention effect of the genes involved in phospholipid metabolism and those involved in immunity (Fig. 6): the joint effect of the genes involved in phospholipid metabolism was more strongly influenced by the intervention than the individual effect, however, neither effect crossed the significant threshold after adjustment for multiple testing (Fig. 6). Compared to the phospholipid metabolism genes, the effect sizes in both the joint nor individual effect of genes involved in the immune system were considerably weaker. These effects were also non-significantly associated with the intervention effect.
Fig. 6: In postprandial blood, the joint component 1 drivers strongly captured the intervention effect while joint component 2 drivers were significantly associated to whole-body fat% and trunk fat% levels.
Associations between the joint component drivers and the intervention effect & metabolic health markers were calculated using a two-sided linear mixed model. The genes were plotted on the y axis, the intervention effect and metabolic health markers were plotted on the x axis. Colors indicate the direction of the effect: blue, a negative effect, white; no effect, red; a positive effect, the darker the color, the stronger the effect. The P values were adjusted using the false discovery rate. Significance is indicated by the asterisks (*FDR <0.05).
Immune genes that relate to the joint effect capture metabolic health effects of the intervention
Next, we investigated whether changes in the genes involved in phospholipid metabolism and immunity associated with changes in the metabolic health markers. Here we found that neither the joint nor individual effect of the phospholipid genes associated to these health marker effects (Fig. 6). In contrast, the joint effect of the genes involved in the immune system did strongly associate with markers of body composition (whole-body and trunk fat%) while, except for RAB44 and RAPGEF3 (which was associated to the fasting HDL cholesterol level), the individual effect did not. Hence, it is possible to detect the effects of the intervention in the postprandial blood transcriptome, but only when the individual blood effect has been removed.
Variance in the muscle and SAT transcriptome explain the selection of contributing genes to the joint components of blood differently
To investigate which of the two tissues, SAT or muscle, correlated most with the joint components in blood, we calculated the correlations of the top 250 genes with the highest loading factors for each of the joint components (JC1 and JC2) in blood with those in muscle and SAT (Supplementary Fig. 5). The JC1 blood genes showed increasingly higher absolute correlations with the JC1 and JC2 muscle genes, whereas the JC2 blood genes showed an opposite behavior and were merely not correlated with the JC1 or JC2 muscle genes. On the other hand, the JC1 and JC2 SAT genes showed a stronger absolute correlation with the JC2 blood genes as opposed to the JC1 blood genes. These results indicate that the JC1 blood genes (dominated by genes involved in phospholipid metabolism) are largely correlated with the variance in the muscle transcriptome, the JC2 blood genes (dominated by genes involved in the immune system) are more correlated with the variances in the SAT transcriptome.
