Transcriptomic landscape profiling of metformin‐treated healthy mice: Implication for potential hypertension risk when prophylactically used

Abstract Recently, the first‐line anti‐diabetic drug metformin shows versatile protective effects against several diseases and is potentially prescribed to healthy individual for prophylactic use against ageing or other pathophysiological processes. However, for healthy individuals, it remains unclear what effects metformin treatment will induce on their bodies. A systematic profiling of the molecular landscape of metformin treatment is expected to provide crucial implications for this issue. Here, we delineated the first transcriptomic landscape induced by metformin in 10 tissues (aorta, brown adipose, brain, eye, heart, liver, kidney, skeletal muscle, stomach and testis) of healthy mice by using RNA‐sequencing technique. A comprehensive computational analysis was performed. The overrepresentation of cardiovascular disease‐related gene sets, positive correlation with hypertension‐related transcriptomic signatures and the associations of drugs with hypertensive side effect together indicate that although metformin does exert various beneficial effects, it would also increase the risk of hypertension in healthy mice. This prediction was experimentally validated by an independent animal experiments. Together, this study provided important resource necessary for investigating metformin's beneficial/deleterious effects on various healthy tissues, when it is potentially prescribed to healthy individual for prophylactic use.

including but not limited to AMP-dependent fructose-1-6-bisphosphatase inhibition 4 and redox-dependent mitochondrial glycerol-3-phosphate dehydrogenase inhibition. 5 Recently, metformin has been demonstrated to have versatile protection against many complex diseases beyond diabetes. For example, the hypoglycaemic and hypolipidemic effects of metformin also contribute to the risk reduction of cardiovascular diseases as evidenced by animal models. 6 Clinical trial supports the protective effects of metformin against cardiovascular diseases like myocardial infarction in diabetic patients. 7,8 Moreover, metformin also inhibits several oncogenic pathways, and therefore, it has been suggested for the treatment of various cancers. [9][10][11][12] Indeed, novel usage of metformin against diseases like polycystic ovary syndrome, 13 neurodegenerative diseases, 14 lung fibrosis, 15 multiple sclerosis 16 and fragile X syndrome 17 have been continuously proposed, with novel mechanism-of-action proposed like exerting its anti-cancer effect through mTOR regulation 18 and anti-inflammatory effect through IKKβ suppression, 19 highlighting metformin as one of the top versatile drugs in the field.
Recently, metformin has been reported to have anti-ageing effects in animal studies. [20][21][22] A clinical trial which aimed at validating the anti-ageing effects of metformin had also been newly launched. 23 However, this novel proposal of metformin usage also signifies that the extensive investigations are needed to evaluate the impact of long-term usage of metformin on healthy individual.
Metformin is widely distributed to a variety of tissues after intake 24 and therefore may elicit unexpected deleterious effects in these tissues. For example, the maternal exposure of metformin interferes the development of offspring's testis in normal mice. 25 Indeed, concerns about the metformin's influence on offspring during polycystic ovary syndrome treatment has been aroused in clinical studies. 26 Moreover, metformin's action is context-dependent. 27 For example, it has been reported that metformin exerts beneficial effects on breast cancer only in patients with metabolic syndrome but not in those without. 28 In addition, the dual-roles of AMPK signalling on cancer cells' metabolism and growth have also been observed.
Beyond inhibiting tumour cell growth, AMPK activation would also promote tumour cell growth by maintaining NADPH level under energy stress. 29 Metformin has considerable influence on the composition of gut microbiota, 30 and its therapeutic effect would be partly attributed to its effect on gastrointestinal tract and gut microbiota. 31 However, the underlying mechanisms are complicated and involve gut-oriented factors like FXR, 32 and its impact on gut microbiota is clearly context-dependent, varying among different treatment conditions or settings. 30,32,33 Collectively, these findings suggested that although metformin has beneficial effects in many pathophysiological processes such as diabetes and cancers, it may also cause unexpected deleterious effects in physiological condition. Due to its versatile protective effects on many diseases and anti-ageing potentials, metformin might be prescribed to healthy individual for prophylactic use or for lifespan prolonging purpose. 23,34 Long-term clinical investigation with respect to metformin's diabetes prevention effect on high-risk population like Diabetes Prevention Program Outcomes Study has been performed. 35 One latest small-scale clinical trial has stepped further to evaluate the potentials of metformin, together with other treatment, for anti-ageing purpose on old population and observed positive effects on reversing aging-related markers. 36 However, another latest investigation has again raised the concern about the context-dependent effect of metformin for anti-ageing settings, where the metformin's effect is significantly modified due to the aerobic exercise training in older adults. 37 Indeed, it remains largely unknown about the effects, particularly the unexpected deleterious effects of long-term metformin treatment on healthy body and tissues. Which molecules will be influenced by metformin? How will these molecules change? What outcomes would be resulted from these changes? The answers to all these important questions are still unknown and thus should be emergently addressed.
To answer these questions, molecular landscape profiling, especially transcriptomic profiling, could be an efficient way. For example, calorie restriction is the most validated anti-ageing factor. 21 By comparing the mouse liver and skeletal muscle transcriptomes after metformin treatment, with those after calorie restriction, the similar anti-ageing effect of metformin has been suggested and further experimentally validated. 21 However, current available transcriptome data about metformin are performed under pathophysiological conditions such as obesity and diabetes. Furthermore, these transcriptome data often cover very few tissues. To date, no systematic transcriptome profiling of the normal tissues after metformin treatment is currently available, which limits the evaluation of the potential beneficial and deleterious effects of metformin treatment on healthy tissues. In the current study, normal healthy mice had been orally treated with moderate dose of metformin for one month, and then the transcription profiles in 10 metformin-treated tissues had been determined using high throughout RNA sequencing. Computational models had been further developed to predict both the beneficial and deleterious effects of metformin in 10 tissues based on the transcription profiles, highlighting the risk for inducing unexpected diseases in normal mice. The predictions for metformin-induced hypertension and cardiac hypertrophy and in healthy mice were further validated by independent animal experiments. Overall, the transcriptome data delineated a landscape of metformin-induced molecular profiles in healthy condition and the findings provided a useful resource for interrogating the potential effects of long-term metformin usage in healthy human beings.

| Experimental animal details
Eight-week-old male/female C57BL/6 and old male mice (60-62 weeks) were fed on normal diet under the condition of constant temperature of 25°C. The mice were randomly divided into the experimental and control groups. Metformin hydrochloride tablets (Glucophage, 0.5 g/tablet, manufactured by Sino-American Shanghai Squibb Pharmaceuticals Ltd.) were dissolved in doubledistilled water to the concentration of 30 mg/mL. Generally, when metformin was used to treat animals, the daily dosage varied from 100 to 600 mg/kg bodyweight. [38][39][40] In the current study, the experimental group of mice were orally administrated by metformin at the daily dosage of 300 mg/kg bodyweight for one or three months, whereas the control group of mice were orally administrated with the same volume of double distilled water. According to the previous researches, it has been suggested that the effect of the standard dose of metformin used in humans for the treatment of type-2 diabetes (~20 mg/kg) is equivalent to that of the ~250 mg/kg dose used in mice, despite the absolute dose in mice is considerably higher than the dose in humans. 41  Raw reads were trimmed for adaptor sequence and masked for low-complexity or low-quality sequence, including those containing more than 10% of unknown nucleotides or more than 50% of low-quality (ie Q-value no larger than 20%) bases and then were filtered against rRNA database to remove potential rRNA contamination. Reads that passed the above filtration and quality control were

| Transcriptome profile comparisons
The transcriptome profiles (FPKM normalized expression values) were clustered by hierarchical clustering method and principle component analysis (PCA) by using the hclust and pca functions of R (https://www.r-proje ct.org, version 3.4.0) and were illustrated by using the pheatmap and ggplot2 packages of R, respectively. For comparative analysis with public transcriptome data, we defined the signature of transcriptome alteration by one treatment as the array of log 2 (fold change) comparing experimental group vs control group.
The transcriptomic signatures were comprehensively deduced from the curated transcriptome data in GEO database (queried from September, 2016 to December, 2017). The samples in one transcriptome data set were manually assigned to experimental or control groups based on the annotations on GEO data sets and samples, and the related publication if applicable. For two-colour arrays with paired case-vs-control design, the fold changes were directly deduced from the signal ratio. Two-colour arrays with universal reference were treated as the same as one-colour array. The fold changes from one colour arrays and RNA-seq data were deduced from the normalized expression values. All expression values, if not log 2transformed, were transformed before calculating fold changes.
Finally, the fold change signatures by metformin in this study and the curated signatures were compared by Spearman's correlation coefficient (SCC), in which only genes shared by two studies were taken in consideration. We removed the redundant signatures (intersignature SCC > 0.5 and of the same topic) and corrected the P-value by the Bonferroni method for multiple hypothesis correction.

| Identification and analysis of DEGs
The differentially expressed genes (DEGs) were identified by edgeR package of R with default parameters. Because the transcriptome profiles were used for hypothesis generation rather than validation, a relaxed threshold of P < .05 was applied. For functional enrichment analysis, mouse DEGs were firstly mapped to human orthologous genes based on the mapping function provided by g:Profiler (http:// biit.cs.ut.ee/gprof iler). The functional enrichment analysis for GAD disease gene sets and KEGG pathways was performed by using the DAVID tool (https://david.ncifc rf.gov, version 6.8) with human official gene symbols as the input. We also compared the DEGs with curated gene sets in response to chemical and genetic perturbations from the MSigDB online platform (http://softw are.broad insti tute. org/gsea/msigdb). Finally, the overrepresented transcription factors were analysed by using the enrichment analysis tool in ChIP-Atlas platform (https://chip-atlas.org). The result is visualized by using the gplots and riverplot packages in R.
The human orthologous genes of mouse DEGs were further mapped onto the human signalling network. The latest version of the human signalling network was obtained from the website of Wang lab (http://www.cance r-syste msbio logy.org/, update v7). The network topology metrics like PageRank centrality and betweenness centrality was calculated by the igraph package in R. Furthermore, the drug target genes from the DrugBank database (https://www. drugb ank.ca) and the DSigDB database (http://tanlab.ucden ver.edu/ DSigD B/DSigD Bv1.0) were also mapped onto the network. Then, the average distance between one DEG set D i = 1, …, m and one drug target set T j = 1, …, n was assigned by using the formula below: To assess the significance of the network distance, equal number of false drug targets were randomly selected from the network for

| Measurement of blood pressure
Blood pressure and heart rate of mice were measured by HD-X10 implantable transmitter after 3-month metformin treatment. Mice after 3-month metformin treatment were anesthetized with 1% sodium pentobarbital (i.p. injection). The left carotid artery was isolated from the surrounding tissue. The flank of the mouse was bluntly isolated with a vascular clamp to form a subcutaneous pocket. Rinse the pocket with sterile saline and place the HD-X10 transmitter in it.
The skin incision was sutured. Blood pressure levels were measured after 5 days. When blood pressure levels were measured by tail-cuff method, mice after 3-month metformin treatment were fixed in the canister in 37°C. Blood pressure and heart rate of mice were measured when mice were stabilized. Blood pressure and heart rate of every mouse were averaged from 10 measurements.

| Transcriptomic landscapes of 10 healthy tissues induced by one-month moderate metformin treatment
To investigate the potential impact of metformin on various healthy tissues, normal mice with chow diet were treated with moderate dosage of metformin (300 mg/kg) for one month. Ten tissues (aorta, brown adipose, brain, eye, heart, liver, kidney, skeletal muscle, stomach and testis) were harvested, and the transcriptomes were profiled by high-throughput RNA-seq ( Figure 1A). Each treatment (metformin or saline control) plus tissue combination were fortified by three biological replicates, which results in 60 transcriptome profiles depicting the landscape of transcriptome alteration by metformin treatment. Because the transcriptome profiles were used for hypothesis generation rather than validation, a relaxed threshold of P < .05 was applied. In all, there are 3079 genes showing differential expression in at least one of the 10 tissues (that is to say, there are 3079 genes in total which show differential expression in at least one of the 10 tissues). The overall similarity between transcriptome profiles of these genes is summarized in the heat map presented in

| Functional and disease associations of differentially expressed genes
In  Figure 2C. Liver is one of the primary target organs of metformin and harbours a considerable fraction of the DEGs in our study.
Accordingly, the liver selective genes and genes down-regulated in early liver development are overrepresented in the shared DEGs.
Besides, genes down-regulated in the diethylnitrosamine treatmentinduced or E2F1 overexpression-induced hepatocellular carcinoma models and HNF1A knockout model of type 1 diabetes are also overrepresented, supporting the liver protective roles of metformin. [42][43][44] Moreover, genes that are de-regulated in nasopharyngeal carcinoma are overrepresented, in line with the recently discovered anti-nasopharyngeal carcinoma action of metformin. 45 Finally, gene sets related to spermatogenesis are also presented in the top list. The metformin's impact on male reproduction has gained increasing notice recently, 46 and our data suggest that the risk of male fertility reduction after metformin treatment deserves serious assessments, both clinically and on animal models.
We further clustered the tissues based on their correlation of DEGs ( Figure 2D). The aorta, brain and brown adipose, in which  Figure 2E and Figure S2, respectively. We also analysed the overrepresented transcription factor behind each DEG module, and the results are shown in Figure S3A-C. A detailed discussion of the results is available in Appendix S1, where the significant associations with cardiovascular diseases like hypertension are highlighted and will be experimentally validated in the last section.

| Comparative analysis of the transcriptomic signatures predicts potential protective and deleterious effects of metformin on 10 healthy tissues
To investigate whether the metformin-induced transcriptome changes could imply beneficial or deleterious effects, for each tissue,  Table S1. In the next section, we will focus on the correlated signatures in liver and heart. The predicted potential protective and deleterious effects of metformin on the other eight tissues are described in Appendix S1, Figures S4 and S5.

| Potential protective and deleterious effects of metformin on healthy liver and heart
Since the liver is one primary known target tissue of metformin, we first performed the comparison in liver ( Figure 3A). correlations of the metformin intervention signature. 47 Therefore, the impact of metformin on mice liver seems context-dependent, and the influence of long-term usage of metformin on healthy livers would be assessed to rule out its potential liver toxicity. Besides, β-naphthoflavone is a known immune inhibitor that exerts anti-inflammatory effect by suppressing TNF-α pathway, 48 (Table S1). Indeed, although metformin has been proposed as an anti-ageing drug, 23 its effect on healthy lifespan is complicated by the onset and dosage of treatment. 50 It is possible that the context of metformin treatment in this study is not appropriate for anti-ageing assay, for example one study using the same dosage of metformin failed to expand healthy lifespan of mice. 51 We also analysed the overrepresented transcription factor behind DEGs  Figure S3E).

| Network analysis of differentially expressed genes indicates potential effects of metformin on blood pressure
We  Figure S6D). To be more precise, we further compare the previously mentioned metformin's transcriptomic signatures (Table S1) (Figure 4). A clinical study indicates that clozapine has noteworthy effect for elevating blood pressure. 56 Interestingly, according to the DeepDDI's prediction, metformin and clozapine, when used together, may increase the risk of adverse effects (Table S2). Given the high risk for a variety of fatal diseases of hypertension and the increasing medication rate of metformin, it is emergently needed to further experimentally investigate the long-term usage of metformin of healthy mice's blood pressure.

| Experimental validation of the prediction that long-term metformin treatment would induce hypertension in healthy mice
The above bioinformatics analyses of transcriptome alteration in mouse healthy tissues treated with metformin for 1 month revealed that long-term administration of metformin tended to cause a number of diseases including hypertension. To validate the prediction that long-time use of metformin causes hypertension, mice were orally treated with metformin for 3 months, and then the relaxation and constriction, and blood pressure levels were measured. In young male normal C57BL/6 mice, 3-month administration of metformin decreased bodyweight and fasting blood glucose levels ( Figure S7A-B). However, metformin treatment had no significant effect on overall glucose tolerance ( Figure S7C-D). In young male mice, 3-month metformin treatment significantly increased systolic, diastolic and mean blood pressure as measured by HD-X10 implantable transmitter ( Figure 5A-C).
Metformin treatment had little effect on the heart rate of young male mice ( Figure 5D). To further validate these observations, the effects of metformin treatment on the relaxation and constriction of mouse arteries were measured. Consistently, the results indicated that after 3-month metformin treatment, the constriction of mouse thoracic arteries in response to phenylephrine was significantly increased when compared with control mouse arteries ( Figure 5E). In contrast, the relaxation of mouse thoracic arteries in response to sodium nitroprusside dehydrate (SNP) and acetylcholine (Ach) was significantly impaired (Figure 5F-G).
These findings supported the blood pressure level increases after metformin treatment on young male mice. In support, the tailcuff method also revealed that 3-month metformin treatment significantly increased blood pressure levels in young male mice ( Figure 5H). Consistently, 3-month metformin treatment similarly F I G U R E 3 The significant correlations between metformin's transcriptomic signature and curated reference transcriptomic signatures in liver and heart. A, Correlations of transcriptomic signature (gene expression fold changes) between metformin-treated normal mice and other treatments or disease models in liver, the top 20 significant correlations are shown. B, The significant correlations in heart increased blood pressure levels in young female mice ( Figure 5I).
These findings suggested that long-term metformin administration did not impair renal functions in young mice.
To further verify the metformin-induced hypertension in normal mice, we investigated the effects of metformin treatment on old male mice as well. As a result, 3-month administration of metformin had little effects on bodyweight and white adipose weight/ bodyweight ratio ( Figure S8A-B). Moreover, metformin treatment also had limited effect on fasting blood glucose level and overall glucose tolerance ( Figure S8C-D). In old male mice, 3-month metformin treatment significantly increased systolic, diastolic and mean blood pressure as measured by HD-X10 implantable transmitter ( Figure 6A-C). Metformin treatment had little effect the heart rate of old young mice ( Figure 6D). Given that healthy people may take a less dosage of metformin treatment, it is also important to investigate whether less dose of metformin treatment also can induce hypertension. Indeed, we showed that 100 mg/kg dose of metformin treatment also induces hypertension in healthy young and old mice ( Figure S9). Overall, these findings further support that long-term use of metformin can cause hypertension in normal mice.

| D ISCUSS I ON
In the past decades, metformin had been reported to exert beneficial effects on various diseases beyond diabetes. More recently, metformin has also been shown to have anti-ageing effects and prolong lifespan. [20][21][22] Metformin is potentially prescribed to healthy individual for delaying ageing and prolonging lifespan based on recent studies, 57,58 and some healthy individuals start to take metformin as prophylactical use or anti-aging. However, the effects of metformin treatment on healthy tissues remain largely unknown. Therefore, it is of great significance and necessarily to evaluate the impact of long-term administration of metformin on various tissues under physiological condition. In the current study, we depicted the impact of long-term treatment with metformin on the transcriptomic F I G U R E 4 Potential links between drugs and DEGs in context of human signalling network. If the distance from one drug's known physical target genes to the DEGs from one specific tissue is significantly closer than random expectation, a potential functional link between the drug and the DEGs are assumed. The top five functionally linked drugs from the analysis in each tissue, together with those shared by multiple tissues, are shown in the bubble plot at the second panel to the right. The Spearman's correlation between the metformin's transcriptomic signature (differential expression induced by metformin) and those of other drug perturbations in the same tissue  63 We have also tried to deconvolute our transcriptome profile by using xCell method. 64 However, we could only find few cell types exhibiting significantly altered deconvolution scores and there is no agreement between different tissues (Table S3). Indeed, few interesting hypothesis could be noted, for example we noted the increased muscle cell and decreased adipocyte cell in skeletal muscle tissue after metformin treatment, which is in line with known effect of metformin. 50 Nevertheless, the cell type-specific transcriptome alteration should be directly addressed by single-cell RNA-seq analyses in the future.
In summary, the current study provided the first transcriptomic landscape of gene expression in 10 tissues of normal mice with or without metformin treatment. These results revealed that metformin deeply impacted gene expression profile to exert both beneficial and deleterious effects in various healthy tissues, which helps a lot to evaluate the potential impact of metformin on their functions.
Our findings revealed that long-term treatment of metformin will cause hypertension under normal condition. Clearly, cautions should be taken when metformin is potentially prescribed to healthy individual for the purposes of reducing bodyweight control, anti-ageing and prolonging lifespan.

ACK N OWLED G EM ENTS
We thank Prof. Zijian Li at the Department of Cardiology of Peking University Third Hospital for his constructive suggestions, and Dr Sisi Guo for her assistance in preparing the cartoons in Figure 1. Metformin treatment had little effect on heart rate of old male mice. 60-to 62-wk-old male mice were daily treated with metformin (300 mg/kg bodyweight) or water for 3 mo. Blood pressure levels were measured by HD-X10 implantable transmitter. Con, control mice treated with water; Met, mice treated with metformin. N = 9. The results are presented as the mean ± SEM. Statistical significance of differences between groups was analysed by t test. *P < .05 vs control mice treated with water