Autofluorescence-based sorting removes senescent cells from mesenchymal stromal cell cultures

Mesenchymal stromal cells (MSC) are used in cell therapy, but results depend on the unknown quality of cell populations. Extended culture time of MSC increases their senescent levels, leading to a critical loss of cell fitness. Here, we tested the suitability of MSC-sorting based on their FACS autofluorescence profile, for a rapid and non-invasive method of senescent cell elimination. Cells were classified in low- (LA) and high- (HA) autofluorescence groups, and results compared to the original MSC population (control). Three days after sorting, cells were screened by replicative senescence markers (cell volume, SA-β-Gal assay and gene/protein expression) and MSC differentiation assays. The transcriptional profiles of sorted MSC were also analyzed by RNA‐Seq. Compared to control, LA cells had 10% lower cell volume and autofluorescence, and 50% less SA-β-Gal + cells. Instead, HA cells had 20% higher cell volume and autofluorescence, and 120% more SA-β-Gal + cells. No changes in replicative senescence and differentiation potentials were observed between all groups. However, 68 genes (16 related to senescence) were significantly differentially expressed (DEG) between LA and other groups. Biological network of DEG identified CXCL12 as topological bottleneck. In summary, MSC sorting may have practical clinical implications to enhance the results of MSC-based therapies.

Human mesenchymal stromal cells (MSC) are adult cells related to the musculoskeletal lineage, which in the recent years were associated numerous times to future therapeutical application, thanks to their fast replicative rate 1 , differentiation potential into mesodermal lineages 2 and secretion of trophic and anti-inflammatory factors 3 . The isolation of MSC-according to current criteria 4 -produces heterogeneous, non-clonal cultures of stromal cells containing stem cells with different multipotential properties, committed progenitors and differentiated cells 5 . Moreover, the use of MSC in regenerative therapies is hampered by the low starting number of MSC that can be isolated from adult tissues. The initial pool and functionality of MSC could be further reduced in quantity and quality (stemness) by diseases such as osteoporosis 6 , leukaemia 7 , or lung and breast cancers 8 . Among the factors underlying the loss of functionality of MSC in tissues, aging must also be seen as a major factor adversely affecting the quality of cells in tissues 9,10 . Nonetheless, decreased tissue regeneration is associated with an increased number of senescent cells, eventually having a negative impact on life expectancy 11 .
The mechanism leading to senescent phenotypes is triggered by multiple events, such as oxidative stress, DNA damage, telomere shortening, and oncogene activation 12,13 . It protects the dividing cells from undertaking a tumorigenic transformation 14 . MSC proliferative capacity is reduced and MSC in a senescent state secrete pro-inflammatory factors and facilitate the invasion of cancer cells into tissues 15 . Similarly to in vivo ageing, a significant proportion of MSC in vitro will also grow into a senescent phenotype when cultured for prolonged period of time 16 . As a result, the effectiveness and reliability of MSC in tissue regeneration may decrease due to age-related changes, both in vivo and in vitro, leading to a potential reduction in the MSC useful for cell-based therapy. This is relevant, in particular, for the medical use of autologous stem cell transplantation in elderly patients.
For efficacious therapies, a large number of cells are required, resulting from an extensive ex vivo cell expansion, and possibly leading to the accumulation of several senescent MSC phenotypes in culture. Identifying a tool for detecting and removing senescent cells in vitro would therefore be necessary for the development of new and more efficient cell-based therapies. Cell autofluorescence is optimal biomarker for screening and targeting

Results
Characterization of autofluorescence-based sorting of mesenchymal stromal cells (MSC) groups. Bone marrow isolated MSC (n = 3) were initially expanded to obtain a large number of cells to perform the sorting via FACS. During the sorting, cells were separated in three groups: low-(LA), mid-(MA) and high (HA) autofluorescence (Fig. 1). As comparison, the original MSC population was propagated in culture with (group unsorted) or without (group control) passing through the FACS. In order to characterize the five MSC groups, we investigated markers of phenotypic (Fig. 2) and genetic senescence (Fig. 3), and differentiation potential (Fig. 4), after three days in culture following cell sorting.
Senescent MSC show clear morphological features, such as enlarged aspect 24 and cell volume (Fig. 2a), higher cellular autofluorescence 17 (Fig. 2b), enhanced senescence-associated beta-galactosidase (SA-β-Gal) activity (Fig. 2c) and longer cell division time (Fig. 2d). In comparison to the control group, average cell volume and autofluorescence were reduced in LA cells (10%), and increased in HA cells (30% and 20% respectively). The average proportion of SA-β-Gal positive cells, as a percent of the total population, was remarkably decreased in LA group (~ 50%) and increased in HA group (~ 10%). In the unsorted and MA groups, the average variations in cell volume, autofluorescence and SA-β-Gal positive cells were always in the ± 10% range compared to control. Only cell division time was longer for all groups compared to control, which may be due to the cell sorting process. In LA group the time increase was ~ 20%, in the unsorted and MA groups ~ 70% and in the HA group was ~ 130%. However, in the following culture passages, we observed a gradual convergence of all groups to control, eventually all groups showing the same senescent profile (data not shown). www.nature.com/scientificreports/ To further define MSC features, we evaluated the gene expression of markers classically associated to replicative senescence ( Fig. 3a-f). Gene expression levels correlated positively (p16 INK4A , p21 CIP1 and ANKRD1) and negatively (p18 INK4C , CDCA7 and E2F1) with MSC replicative senescence. In all groups, we observed small-tomoderate average increases in p16 INK4A (range, 10-40%) and ANKRD1 (0-40%) as compared to control. On the other hand, the average expression of p21 CIP1 was nearly doubled in all groups, ranging from ~ 40% increase in the unsorted group to over 120% increase in the LA group. Also minor changes in the expression of p18 INK4C , CDCA7 and E2F1 were found. Compared to control, we observed an increased average expression of p18 INK4C (0-20%), CDCA7 (0-40%) and E2F1 (0-20%) in all groups. Protein extracts from all MSC groups were used to assess changes in expression of p21 CIP1 , p53, retinoblastoma (Rb) and p16 INK4A by western blot (Fig. 3g). Quantification of bands-normalized to β-actin-showed that in comparison to control, MSC in HA group had higher expression of p16 INK4A (4.5-fold), and lower expression of Rb (-1.0-fold). No relevant changes (± onefold) were observed in the other groups, except the lower expression of p16 INK4A in the MA group (-1.3-fold).
We also investigated the ability of MSC to differentiate towards adipogenic (Fig. 4a), chondrogenic (Fig. 4b) and osteogenic (Fig. 4c) lineages after 21 days in culture. In comparison to control, we observed negligible changes in MSC adipogenic differentiation in all groups, with average variations of ± 5%, assessed and quantified by Oil red O staining. Analysis of chondrogenic potential, evaluated by accumulation of proteoglycan and quantified by alcian blue staining, showed no changes in the unsorted and MA groups compared to control, but an average decline of ~ 50% in the HA group and an increase of ~ 250% in the LA group (despite high variation within the data). Osteogenic potential of MSC was evaluated by measuring calcium deposition within the extracellular matrix and, in comparison to control, all groups were in average 25% less potent, except HA group which equalled the control. We identified differentially expressed genes (DEG) between all groups using Cuffdiff program (p value < 0.0005; |log 2 |fold change > 1), and results are represented schematically by dendrogram of hierarchical clusters (Fig. 5a), table (Fig. 5b), and volcano plots (Fig. 5c). RNA-seq data showed that HA and unsorted cells were similar and closer to control, while LA cells were distinct from other groups. We observed only 9 and 17 genes differentially expressed in HA and unsorted groups respectively, compared to control. These results indicated the absence of discernible phenotypes between control and unsorted groups, demonstrating the lack of negative influences of FACS processing on MSC. In contrast, 171 genes in the control group were significantly changing in expression, compared to LA group. Always in comparison to LA cells, 158 and 155 genes were differentially expressed between unsorted and HA groups respectively. Among these genes, we observed 40 upregulated (Fig. 6a) and 28 downregulated (Fig. 6b) genes that were shared among all three groups (Fig. 6c). Of the 68 DEG, 16 genes (ITGB8, COL13A1, DUSP4, MYCT1, ESM1, FMO2, FMO3, NDNF, C1R, ESM1, CXCL12, VCAM1, NTN4, PLAT, KRT34, SERPINB2) have been already associate to senescence in MSC in vitro [25][26][27][28] and in vivo 29,30 . Identification of DEG among LA and HA cells using RNA-Seq. Based on system biology analysis, we identified the interaction network formed by the 155 DEG from the comparison between LA and HA cells (Fig. 7a) and the genes CXCL12, VCAM1 and LOX2 were recognized as a bottlenecks (Fig. 7b). The 40 genes with the greatest fold changes and significant p value are shown in Table 1.
The 155 differently expressed genes were analysed by ToppGene and were found to be involved in a number of biological processes, such as blood vessel development and cellular response to cytokine stimulus ( Table 2). Gene ontology (GO) term analysis of the DEG revealed that the most significant associations were with extracellular matrix and cell receptor regulatory activity. Overall, the results showed that the differences between groups reside in the way cells communicate and react with the environment, which is the peculiarity of MSC paracrine action. To validate RNA-Seq results, we evaluated by qRT-PCR the gene expression of the 18 genes with greatest fold changes between LA and HA cells, and represented in the clustergram (Fig. 8a). The validation analysis confirmed the results of the RNA-seq, and the hierarchical clustering analysis of the data also established LA group as distinct to the control, unsorted and HA groups, with MA group standing in the middle. Genes FGFR2, FMO2, TDPD52L1, NDNF, CXCL12, C1R, C7, VCAM1 were upregulated in the LA group, whereas POLG, SERIPNB2, PLAT, MYCT1, KRT34, BEX1, TMEM171, EPGN, ESM1, CAV1 were upregulated in the HA group.
Next, we investigated in all three MSC donors the expression of the genes CXCL12 26 (Fig. 8b), FGFR2 28 (Fig. 8c), FMO2 31 (Fig. 8d) and NDNF 32 (Fig. 8e) which had the largest fold increase in LA and HA cells and which have been shown to be important MSC senescence-related genes. In comparison to control, the average expression of CXCL12 (4.4 fold), FGFR2 (5.9 fold), FMO2 (13.6 fold) and NDNF (5.4 fold) was the highest in the LA group. In the MA group the average upregulation of CXCL12, FGFR2, NDNF (all ~ 2.0 fold) and FMO2 (3.9 fold) was less consistent, while in the HA and unsorted groups gene expression changes were minimal (± 0.3 fold).

Discussion
In this study, we showed that endogenous autofluorescence of human mesenchymal stromal cells (MSC) could be used as a non-invasive and easily applicable method to remove senescent cells from in vitro cultures. As we have previously demonstrated 17 , when cells become senescent in vitro they significantly change their phenotype and they accumulate undegredable molecules and key endogenous fluorophores-such as free nicotinamide adenine dinucleotide (NADH), flavin adenine dinucleotide (FAD) and lipofuscin 33 -resulting in a progressive increase of cellular autofluorescence.
Native cell fluorescence consists of a mixture of molecules that naturally display fluorescence, resulting in detailed fluorescent finger prints. For instance, by bioimaging of the autofluorescence of murine retinal www.nature.com/scientificreports/ cells-without any external biomarkers-one can identify biochemical constituents, their location and abundance within cytoplasm 34 . Likewise, based on this distinct autofluorescent signature, in our study we provide evidences that MSC with low average autofluorescence (LA) were less senescent than cell with higher autofluorescence (HA). In comparison to HA cells, LA cells were morphologically smaller, halved the proportion of senescenceassociated-β galactosidase (SA-β-gal) positive cells and overexpressed genes negatively associated with cellular senescence, such as CXCL12, FGFR2, FMO2 and NDNF 26,28,31,32 . However, HA cells did not exhibit the entire phenotype usually associated with replicative senescence. The gene expression of the characteristic markers of replicative senescence-positive genes p16 INK4A , p21 CIP1 and ANKRD1 and negative genes p18 INK4C , CDCA7 and E2F1-were unaffected in comparison to control. Only in the HA group compared to other groups the protein expression of p16 INK4A and Rb was higher and lower respectively. This means, that the autofluorescence method selects senescent cells based on a different phenotype, which do not involve substantial differences in cell cycle transcriptional regulators. Indeed, RNA-Seq transcriptome of HA cells found overexpression of genes considered senescent markers, but with no known function related to replicative senescence, such as Plasminogen Activator, tissue type (PLAT) 28 , or indirect function, such as serine protease inhibitor-B2 (SERPINB2) that binds to p21 CIP1 preventing its proteasome-mediated degradation 25 . Such more granulated level of evidence suggests that this senescence was either induced by different stimuli or detected before the classical cell cycle arrest/progression occurs. Indeed, comparing LA and HA cells, we see significant changes in cellular response to cytokine stimulation and receptor regulatory activity, as well as remodeling of extracellular collagen (Table 2), indicating a more likely presence of cell paracrine senescence caused by senescence-associated secretory phenotype (SASP). Previously, we demonstrated the significant increased secretion of IL-6 and MCP-1-involved in the SASP 35 -in those MSC with the highest cellular autofluorescence 17 . As a matter of fact, we have observed that after several weeks in culture, MSC in all groups aligned and converged to similar senescent levels to these in the control, despite significant differences in senescent frequencies few days after sorting. We suppose that autofluorescence-based sorting of cells interfered with the equilibrium and ratio between sub-populations of MSC, and as a compensatory www.nature.com/scientificreports/ mechanism, the secretion of SASP factors influenced the behavior of neighboring cells in a paracrine fashion, especially in the HA group. Among SASP factors, we found that CXCL12 (aka SDF-1; stromal cell-derived factor-1) was the bottleneck (the central node) in the network of differentially expressed genes between LA and HA cells (Fig. 7). The role of CXCL12 is to promote the processes of chemotaxis and migration 36 . For instance, in a murine model of arthritis, MSC expressing CXCL12 in an inflamed area of the bone suppressed the proliferation of osteoclasts and inflammatory stimuli 37 . Also senescent tumor cells produce CXCL12 to promote cancer cell migration and metastasis, and ultimately contributing to tumor development 38 . However, even though LA cells expressed higher levels of CXCL12, we did not observe any significant increase in the expression of matrix metalloproteinases (MMP) and other enzymes capable of degrading extracellular matrix, which are typical cellular signature of senescent tumor cells 39 . So, presumably, here MSC were not acquiring a tumorigenic phenotype, but rather reinforced their paracrine function, devoted to stimulate and support the fitness of the whole population.
It has been shown that the transplantation of senescent ear chondrocytes into the knee joint area of wildtype mice can exacerbate the effects of osteoarthritis, in comparison to transplantation of non-senescent chondrocytes 40 . On contrary, selective elimination of senescent cells potentially mitigates the effects of several age-dependent disorders 41 . For instance, the use of a potent senolytic drug-ABT263, inhibitor of the antiapoptotic proteins BCL-2 and BCL-xL-successfully alleviated the ageing effects of total-body irradiation in mice bone marrow hematopoietic stem cells and muscle stem cells 42 . Similar to senolytic drugs, the idea of sorting MSC before transplantation in vivo would benefit the therapeutical outcomes, by reducing locally the inflammatory response in adjacent cells. We propose that targeting senescent cells is a promising approach for preventing or treating age-related diseases. Additionally, we did not observe deleterious effects on MSC by the sorting process via FACS-except a temporary slight increase in cell division time-which makes the protocol safe and reliable in terms of cellular health and future clinical applications.
As limitations, we have found that autofluorescence-based sorting had no significant improvements in the tri-lineage in vitro differentiation potentials of MSC (Fig. 4). Despite the reduction of SA-β-gal positive cells  www.nature.com/scientificreports/  www.nature.com/scientificreports/ in the LA group, in comparison to control, the differentiation yield of MSC was almost identical, raising the hypothesis that cellular senescence and differentiation potential have mechanisms, which are not necessarily interconnected at all levels. Furthermore, in this study the results were affected by the inter-donor variability of the MSC populations. Starting cell population with different senescent levels resulted in high variance between data, which were normalized to the control. A population with a lower proportion of senescent cells will benefit less of the sorting because the removal of these cells will have a smaller impact on the overall cellular fitness of the population, in comparison to a population enriched with senescent cells.
In conclusion, we demonstrated the functionality of autofluorescence as a marker of MSC senescence. Sorting cell strategies to remove senescent cells or to delay senescence could be beneficial for both health span and life span of MSC. The method proposed here for extracting senescent cells from cultures is effective, fast and label-free. In perspective, MSC sorting based on autofluorescence is an economic assay which can be used on its own, or in combination with other tests, to control and improve cellular fitness and ultimately the outcomes of cell-based therapies.

Materials and methods
All methods were performed in accordance with the relevant guidelines and regulations of this journal.
MSC senescent phenotype characterization. After sorting, MSC were cultured in growing medium for three days and re-plated at the cell density of 1.25 × 10 3 cells/cm 2 . Cell number and cell volume were measured with the Scepter cell counter (Millipore, Thermo Fisher Scientific, Zug, Switzerland).
Autofluorescence data were acquired with CytoFLEX flow cytometer (Beckman Coulter Life Sciences, Nyon, Switzerland), using the excitation laser at 488 nm and detection optic at 525/50 nm. A 638 nm laser and 670/30 detector was used for live/dead stain TO-PRO-3 (Thermo Fisher Scientific). Resulting data files were analysed using FlowJo software v.10.0 software (Treestar, Ashland, OR, USA). To standardize cytometer settings between runs, 15 µm polypropylene calibration beads (PHCCBEADS, Millipore) were used and dead cells were excluded from analysis.
RNA isolation and RNA-seq. Total RNA was isolated from MSC (control, unsorted, low-and high-fluorescence) using QIAzol Lysis Reagent (Qiagen, Hombrechtikon, Switzerland) and Direct-zol RNA MiniPrep (Zymo, LucernaChem), according to the manufacturer's instructions and kept at − 80 °C. Potential genomic DNA contaminations were removed by treating samples with DNase during RNA isolation protocol.
RNA-Seq assays were performed by Fasteris SA (Plan-les-Ouates, Switzerland). Poly-adenylated transcripts selection purified from 1 µg of total RNA for each sample was followed by cDNA libraries construction using the TruSeq Stranded mRNA Library Prep kit (Illumina, San Diego, CA, USA). Libraries underwent high-throughput sequencing in an Illumina NextSeq 500 sequencer and for each 100 million paired-end 50 bp reads were generated during the sequencing runs. Sequence data was subjected to quality control using an indexed PhiX reference sequence to estimate sequencing error. RNA-seq reads were mapped with TopHat (https ://ccb.jhu.edu/softw