We incubated adipose-derived MSCs (replicative passage 3) with 3 mM, 6 and 9 mM of metformin and evaluated some biological parameters (cell cycle, apoptosis and senescence) at 3 weeks, 4 weeks and 6 weeks post-treatment. Control cultures showed a progressive decrease of dividing cells since the percentage of S-phase doubling dropped from 7.7% at 3 weeks to 2.1% at 6 weeks (Table 1). At two and 4 weeks of incubation with metformin, no significant differences between the control and treated samples was observed. However, at 6 weeks a higher percentage of cells in the S-phase was seen in treated samples. Specifically, MSCs incubated with 9 mM metformin showed 6.7% of cells in the S-phase (Table 1). The in vitro cultivation induces replicative senescence through telomer attrition. The percentage of senescent MSCs sharply increased from 11% at 2 weeks to more than 50% at 6 weeks. At this data-collection point, the metformin treatment had greatly reduced the percentage of senescent cells. In detail, incubation with 6 and 9 mM metformin almost halved the percentage of senescent cells (Figure 1). Prolonged cell cultivation leads to cell stress that, in addition to the onset of senescence phenomena, may trigger apoptosis. Indeed, after 6 weeks of cultivation, MSCs showed a significant growth of apoptotic cells (6.7% at 3 weeks vs. 14% at 6 weeks) (Figure 2). Of great interest, the metformin treatment also reduced the apoptosis level at all the analyzed time points (Figure 2).

Our study demonstrated that 6 weeks’ treatment with metformin produced the most striking changes in the analyzed biological functions; thus, we performed a LC-MS/MS analysis of whole cell proteome and secretome of MSCs cultivated in vitro for 6 weeks with or without metformin supplementation. For each experimental point we consider only proteins that were present both in biological and technical replicates. In control MSCs, the LC-MS/MS analyses of peptides identified 1,749 proteins in whole cell lysates and 448 proteins in the secretome. Similar numbers of proteins were identified in the metformin-treated samples (Supplementary Material S1). The Venn analysis of proteome demonstrated that in addition to a common core of proteins being present in all cell lysates or in all secretomes, each experimental condition exhibited specific proteins (Figure 3, Supplementary Material S2). This result indicates that metformin profoundly modified the cellular protein composition of MSCs as well as their secretomes. Next, we conducted several bioinformatics investigations to gain insights into the differences in protein composition between the control and metformin-treated MSCs.

We parsed the proteome profiles of our samples by Gene Ontology (GO) in order to determine the relative frequency of ontological terms associated with specific cellular functions. We accomplished this by performing analysis of the terms grouped in the GO biological process database. For each experimental condition, we identified one hundred ontologies and then used Venn-diagram analysis to combine the GO data and identify the biological processes that were affected by metformin (Supplementary Material S2). Interestingly, we found enrichments in metabolic processes—such as glycolysis, carbohydrate metabolic process, carboxylic acid catabolic process, and nucleotide biosynthetic process—in the whole-cell proteome of metformin-treated samples. These enriched ontologies were also identified in secretome samples of metformin-treated MSCs (Supplementary Material S2). This result is in keeping with the role of metformin in modifying cellular metabolism.

GO analysis found the enriched ontological terms in our samples but could not identify the most important proteins in several of the experimental conditions. In this context, we performed a canonical pathway analysis to identify the set of interactions for each protein in our datasets. This investigation allowed us to determine the most representative pathways that could have a functional impact on the observed biological phenomena.

For every experimental condition, the proteins present in cell lysate and the secretome could be attributed to hundreds of canonical pathways (Supplementary Material S3A, B). We fixed a high statistical significance (p < 0.001) cutoff in order to determine the pathways that were overrepresented in our samples. Also, with this limitation, we identified many pathways for each analyzed sample. The Venn diagram was then used to combine the data concerning all of the experimental conditions to find the specific canonical pathways of the control samples and of those treated with metformin (Supplementary Material S3A). We further limited our analysis by comparing the pathways exclusively present in the control or in metformin-treated samples, which were grouped together irrespective of drug concentration. This strategy allowed us to identify a few pathways that could be further investigated (Table 2). The control MSCs showed ten exclusive pathways, while the metformin-treated cells exhibited four specific canonical pathways. The majority (eight out of ten) of control pathways could be associated with the typical features of senescent cells. Senescent cells are resistant to death due to active anti-apoptotic signaling. They may lose their original functions, but they will remain highly metabolically active and synthesize dozens of factors that are released in the SASP (van Deursen, 2014; Kim and Kim, 2019; Myrianthopoulos et al., 2019). The control MSCs presented a nucleotide-excision repair pathway, a cysteine biosynthesis pathway and an iron homeostasis signaling pathway, all of which can be related to survival strategies used to cope with genotoxic stress events (Nakamura et al., 2019; Daher et al., 2020). These cells employed pathways associated with aerobic and anaerobic glucose catabolism (PFKFB4 signaling, Acetyl-CoA biosynthesis) and nucleotide metabolism (Ribonucleotide biosynthesis, pentose phosphate pathways, etc.). Additionally, they presented an active dopamine receptor signaling that could be responsible for impaired MSC functions. Indeed, dopamine signaling can block the wound healing activity of MSCs (Shome et al., 2012).

Treatment with metformin profoundly modified the intracellular signaling in MSCs. We identified four specific pathways, and three of them were related to the recovery of cellular functions and a reduced presence of senescence phenomena. Alpha-adrenergic signaling is important for the physiological secretory activity of MSCs (Tyurin-Kuzmin et al., 2016). The xenobiotic metabolism PXR signaling pathway is associated with the detoxification activity that MSCs can exert on the surrounding environment (Oladimeji and Chen 2018). Furthermore, the aspartate degradation pathway is related to optimal energy production, which is based on the transfer of cytosolic NADH into the mitochondrial matrix (Munoz et al., 2014).

Also, we found specific pathways in the secretome of control MSCs that can be associated with senescence and SASP activity. Indeed, the telomeres extension through the telomerase pathway is associated with replicative senescence as it is well known that telomeres’ erosion occurs after prolonged in vitro cultivation (Victorelli and Passos, 2017). The virus entry via endocytic pathway is indicative of active the endo/exocytosis events that are responsible for SASP production and paracrine signaling among senescent cells and the surrounding healthy cells (Ayaz-Guner et al., 2020). The intrinsic prothrombin activation pathway could be associated with the pro-inflammatory activity of SASP (Chu, 2010). In the secretome of metformin-treated MSCs, we found only one specific pathway: the role of PKR (protein kinase R) in Interferon Induction and Antiviral Response. There are findings showing that MSC immunosuppressive properties are based on the activation of indoleamine-2,3-dioxygenase-1 through interferon-beta and PKR (Opitz et al., 2009).

Following the identification of significant pathways in our experimental conditions, we performed a regulatory network analysis by determining key transcription factors that are likely to be responsible for the changes observed in our data (Supplementary Material S4). We conducted an IPA upstream regulatory investigation followed by Venn diagram analysis to determine the transcription factors exclusively present in the control samples or the metformin-treated cultures (Supplementary Material S4, Table 2). We found that six transcription factors were exclusively present in the whole cell proteome and five transcription factors were only identified in the secretome of control cultures, respectively. In the metformin-treated cells, we found four and one transcription factor(s), respectively (Table 2). Most of the identified transcription factors have pleiotropic activities since they regulate many biological functions including antagonistic tasks, such as cell survival and apoptosis, depending on environmental factors. In this context, it is not a straightforward matter of attributing them a role in our experimental conditions. We tried to circumvent this difficulty by examining the proteins they putatively regulated in our samples (Supplementary Material S5). For example, in metformin-treated cells the TP53 transcription factor putatively regulates 381 proteins; while in the control sample, the HOXD3 factor regulates eight proteins (Supplementary Material S5). We focused our attention on proteins that were regulated by at least two of the identified transcription factors. This allowed us to identify some key proteins and their associated transcription factors in our datasets (see Tables 3 and 4).

In control cultures of MSCs, FLI1, MEF2D and MYOD1 in cooperation with other factors regulate the expression of proteins involved in cytoskeletal and extracellular matrix formation (e.g., Actin, Collagen, Fibronectin) (Table 3). All these proteins play a key role in senescence and its associated extracellular remodeling. They are listed in the senescence gene database (https://senequest.net). Also in the control secretome, LEF1 and LMO2, together with other factors, regulate many proteins that are part of cytoskeletal and extracellular structures. In the samples besides the control secretome, these transcription factors regulate the expression of all proteases that are involved in senescence (Serpins, Metalloproteases) (Table 3). All the listed proteins play a role in senescence as indicated in Senequest database.

In metformin treated samples, CEBPA, TP53 and YY1 regulate the expression of the genes that may be negatively associated with senescence (COL1A1; HSPA5) (Bigot et al., 2012; Li et al., 2014). There were other senescence-associated proteins whose expression is regulated by the identified upstream factors, such as Serpins and Timp1 (Table 4). Nevertheless, we observed many factors (SOD1, SOD2, CAT, GLRX, GSTP1) that neutralize ROS and hence prevent the onset of senescence (Davalli et al., 2016). Additionally, there were proteins (DNAJB4, CRYAB) involved in the regulation of protein folding even after stressful stimuli, such as those inducing apoptosis or senescence, impact the cell (Miao et al., 2019; Zhang et al., 2019) (Table 4).

Adipose tissue-derived MSCs were obtained from the American Type Culture Collection (ATCC PCS-500-011) and were grown in DMEM containing 10% FBS, 4 mM L-glutamine, 100 U/mL penicillin-streptomycin, and 5 ng/ml bFGF. Cells were provided at replicative passage 3 (P3) and were cultivated for up to 6 weeks either in presence or absence of metformin. All the biological assays (apoptosis; senescence; cell cycle) were performed after 3, 4 and 6 weeks of treatment. For each biological assay, we performed three biological replicates.

In order to perform apoptosis tests, cells were detached by trypsinization and following the washing step dissolved in 100 μL 2% FBS containing PBS. 100 μL Annexin V reagent (Millipore, Burlington, MS, USA) was added on the cells and incubated for 20 min at room temperature. Analyses were performed by using BD FACSAria III flow cytometer (Millipore, Burlington, MS, USA). For each experimental point, we analyzed at least 5,000 cells to determine the percentage of apoptotic cells.

Detached cells were fixed with 70% ethanol at −20°C for at least 3 hours. After fixation step, cells were washed in order to remove the ethanol and stained with propidium iodide containing Cell Cycle Reagent (Millipore, Burlington, MS, USA). Analyses were performed by using Muse Cell Analyzer (Millipore, Burlington, MS, USA). For each experimental point, we analyzed at least 5,000 cells to determine the cell cycle profile.

Cells grown in six well plates were fixed using a 0.2% glutaraldehyde solution for 5 min at room temperature (RT). Then, the cells were washed with PBS and stained with 40 mg/ml X-gal staining solution, as reported (Debacq-Chainiaux et al., 2009). Blue-stained cells were counted from three/five different regions of each well, and the percentage of senescent cells was determined. In identifying senescent cells, we also considered other properties, such as cell size, multi-nuclei presence, and granularity. For each experimental point, we analyzed at least 1,000 cells to determine the percentage of senescent cells.

For every experimental point we performed two biological replicates and for each of them we did two technical replicates. Globally, we performed 32 MS analyses: 16 runs for whole cell proteome samples and 16 runs for the secretome ones. The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE (Perez-Riverol et al., 2019) partner repository with the dataset identifier PXD028349.

Samples were prepared with the InStage Tip digestion method described by Kulak and colleagues (Kulak et al., 2014). We collected 1 × 10⁶ cells each sample and, after PBS washing, cells dissolved in 100 μL Lysis Buffer (6 M Guanidinium chloride, 40 mM CAA, 10 mM TCEP, 25 mM Tris-HCl pH:8,5). Lysates were boiled for 5 min and then were sonicated in an ice-filled ultrasonic water bath for 5 min. Samples were then centrifuged at 20,000 g for 15 min, and proteins containing supernatants were collected.

We mixed 20 μL of each supernatant with 280 ng Lys-C (Promega, WI, USA) containing a 40 μL dilution buffer (25 mM Tris-HCl pH 8.5, % 10 ACN), which we put into InStage tips previously prepared by using 3 SDB-RPS extraction disks (3M Emporem, MN, USA). Mixtures were incubated overnight at 37°C. Subsequently, 1,000 ng Trypsin-Gold (Promega, WI, USA) was added to the Stage Tips, mixed well, and incubated for 4 h. Following the incubation step, a 140 μL loading buffer (1% TFA) was added to each tip and centrifuged at 2,000 g; then, the peptide-loaded disks were washed four times with a 100 μL washing buffer. Peptides were eluted from disks in three fractions, according to their hydrophobic properties, by using 60 μL of each of the three elution buffers: SDB-RPS1 (100 mM Ammonium formate, 35% ACN, 0.5% Formic Acid), SDB-RPS2 (100 mM Ammonium formate, 55% ACN, 0.5% Formic Acid), and Buffer X (80% ACN, 0.125% Ammonia). Samples were lyophilized with SpeedVac and stored at −20°C until the LC-MS/MS analysis.

MSC cultures were incubated in serum-free media for 24 h; then, 5 ml of culture medium (secretome) was collected from each culture dish without disturbing the attached cells. Culture debris was removed by centrifugation at 10,000 g for 10 min, and supernatants were used for the StartaClean beads protein pooling. Collected secretomes were incubated overnight with the beads; then, the beads were washed twice with TE Buffer (50 mM Tris 10 mM EDTA pH 7) and dried with a vacuum concentrator.

The dried beads were resuspended at 2% (w/v) in a RapiGest (Agilent, CA, USA) solution containing TEAB (Sigma, MO, USA). Then, TCEP (Sigma, MO, USA) was added to the solution at a final concentration of 20 mM. Samples were incubated at 60°C for 30 min and cooled on ice. IAA (Bio Rad, CA, USA) was added to sample solutions, and samples were incubated at RT for 15 min. Then, 200 ng Lys-C (Promega, WI, USA) was added to each sample and incubated for 4 h at 37°C. After Lys-C incubation, 800 ng Trypsin-Gold (Promega, WI, USA) was added to each sample and incubated overnight. Samples were centrifuged at 10,000 g for 1 min; peptides containing supernatants were collected and acidified with 1% TFA before being loaded into Stage Tips. These tips were prepared with C18 material: they were washed with buffer B (% 0.1 Acetic Acid, 80% ACN) and equilibrated with buffer A (% 0.1 Acetic Acid). Acidified samples were loaded onto Stage Tips, and peptide-bounded tips were washed twice with buffer A. Following the washing, buffer B was added to the tips, and samples were eluted into collecting tubes with a syringe. Samples were dried with a vacuum concentrator and stored at −20°C until LC/MS analysis.

LC-MS analysis was performed with AB Sciex Triple ToF 5600+ (AB SCIEX, CA, USA) integrated with LC-MS/MS Eksigent ekspert™ nanoLC 400 System (AB SCIEX, CA, USA). Peptides were separated using nanoACQUITY UPLC 1,8 μM HSS T3 C18 column (Thermo Fisher, MS, USA) in the trap-elute mode. In order to separate the peptides, 4–40% ACN gradient was used for 240 min. Data dependent acquisition (DDA) MS/MS analysis of separated peptides was performed after electrospray ionization. Raw data analysis—generated by instrument reporting—and multiple analytical data measurements in each sample were performed with Analyst® TF v.1.6 (AB SCIEX, CA, USA). The peptides and the ion-product of the MS and MS/MS data were evaluated with PeakView (AB SCIEX, CA, USA). Generated peak-lists were evaluated in consideration of the UniProtKB-based reference library of the Homo sapiens species on our server with ProteinPilot 4.5 Beta (AB SCIEX, CA, USA).

The protein content of whole cells and secretomes was analyzed with PANTHER (http://www.pantherdb.org) and with the Ingenuity Pathway Analysis (IPA) (http://www.ingenuity.com/products/ipa).

PANTHER allowed the GO analysis by classifying protein contents according to three ontological terms: biological processes, molecular functions, and molecular classes. For PANTHER analysis, we used the statistics overrepresentation, which compares classifications of multiple clusters of lists with a reference list to statistically identify the over- or under-representation of PANTHER ontologies. Significance was set to a p-value of 0.05.

Differentially expressed proteins in whole cell lysates and secretome were imported into IPA to identify canonical pathways and upstream regulators. Fischer’s exact test was used to calculate a p-value that would determine the probability that the association between genes in the dataset and canonical pathway could be explained by chance alone. Significance was set to a p-value of 0.001. The IPA Upstream Regulator analysis uses known molecular interactions in the datasets to identify upstream regulators. The z-score was used to identify the significant upstream regulators.

Statistical significance was determined with ANOVA analysis followed by Student’s t and Bonferroni’s tests. We used mixed-model variance analysis for data with continuous outcomes. All data were analyzed with a GraphPad Prism version 5.01 statistical software package (GraphPad, CA, USA).