Cellular senescence plays important roles in human diseases, including musculoskeletal disorders (1, 2). Recently we found that aged mice have markedly increased senescent cells (SCs) in fracture callus that exert the senescence-associated secretory phenotype (SASP). Clearance of SCs with senolytic drugs promotes fracture healing in aged mice (3) and young mice (4). However, fractures are also associated with inflammation. Inflammatory cells (Inf-Cs) produce pro-inflammatory factors, some of which overlap with SASP factors (5, 6).

Pre-clinic studies have revealed potential difference between SCs and Inf-Cs in fracture healing. For instance, the acute inflammatory response in the callus following fracture peaks at 24-48 hours and quickly disappears within a week, and deficiency or inhibition of inflammation at this stage causes delayed fracture healing by inhibiting endochondral ossification (7, 8). In contrast, SCs accumulate in the callus gradually and peaks around 10-14 days post-fracture and removing SCs by senolytic drugs, given at 3-7 days post-fracture before the peak of SC accumulation, improves fracture healing (3).

Distinguishing SCs from Inf-Cs in aged callus is very important, not only because both of them are increased and produce inflammatory factors, but also because they require different drug treatments, e.g. senolytic drugs for SCs, and non-steroidal anti-inflammation drugs (NSAIDs) for Inf-Cs. NSAIDs improve fracture healing in mice (9), but the drugs have been reported to significantly increase the risk of a second hip fracture, especially in old patients (10). The meta-analysis of the NSAID treatment in clinical trials reports a negative effect of NSAID, which is highly dose and time dependent because long term NSAID administration increase the rates of non-union fracture in the elder patients (11, 12). These studies indicate that anti-inflammatory drugs may not be a good choice for treating fractures in the elderly.

Thus, identification of any differences between the behavior of Inf-Cs and SCs for treatment of aging fracture is an important unmet need. Here, we analyzed our recently published single cell RNA-sequencing (scRNA-seq) data set of aged mouse callus stromal cells (13). We compared cells that express inflammatory genes and cells that express senescence-associated genes to study molecular signatures of Inf-Cs and SCs and their interactions using bioinformatic analyses. We validated our findings with cell culture experiments and callus cells from NF-κB-GFP reporter mice.

Callus stromal cells contain distinct cell sub-populations that express genes associated with inflammation, senescence, or combined inflammation and senescence. To investigate the differences between callus inflammatory cells (Inf-Cs) and senescent cells (SCs), we analyzed our recently published scRNA-seq dataset of CD45^-CD31^-Ter119^- callus stromal cells (13). We used data from aged callus cells because inflammation and cellular senescence are increased with aging (13, 21). We identified 4 major cell clusters by graph-based clustering ( Figure 1A ). Cluster1 and cluster2 expressed known osteogenic genes (Runx2, Acta2, Col5a1), cluster3 expressed osteogenic genes with a relative lower level than cluster1 and 3, and cluster4 expressed known adipogenic genes (Apoe, Lpl) ( Figure 1B ). Thus, we named cluster1, cluster2, cluster3 as osteogenic-1, osteogenic-2, osteogenic-3, and cluster4 as adipogenic cluster, respectively ( Figure 1A ). Heatmap of top differentially expressed genes (DEGs) showed that similar to our previous report that aged callus cells have a high inflammatory signature (13), cells in clusters1, 2 and 4 expressed inflammatory (S100a11, Retnlg, Mmp9, Cxcr2, and Ccr2) genes. Interestingly, cells in cluster3 expressed genes associated with chromosome modification and cellular senescence (Tuba4a and Dtymk) ( Figure 1C ).

To label inflammatory cells and SCs for further analysis, we defined cells that express the inflammation-related genes, Rela or Relb (22), as the inflammatory cell (Inf-C) subset and cells that express senescence-associated genes as the SC subset. We examined the expression of 10 genes that have been used to detect SCs of various cell types in the literature (23, 24) and decided to use Cdkn1a, Cdkn2c and Cdkn2a as senescence-associated genes because Cdkn1a and Cdkn2c are expressed mainly by cluster 3 ( Supplemental Figure 1 ) and Cdkn2a has been used in our previous study to label callus SCs (3). Thus, we defined cells that express Cdkn1a, Cdkn2a or Cdkn2c as SCs. Here Inf-Cs did not express Cdkn1a/Cdkn2a/Cdkn2c, while SCs did not express Rela/Relb. Inf-Cs were present in all clusters, while SCs were mainly localized in cluster3, the osteogenic-3, and cluster4, the adipogenic cluster. Some cells in the adipogenic cluster co-expressed both inflammation and senescence-associated genes and were named as the inflammatory SC (Inf-SC) subset ( Figure 1D ). Inf-Cs comprised 34% (2321), Inf-SCs 8% (579), and SCs 7% (452) of the total 6,834 cells analyzed ( Figure 1E ). The expression pattern of these inflammatory and senescence genes in the 3 subsets was illustrated in a violin plot ( Figure 1F ).

To investigate functional differences among Inf-C, Inf-SC, SC subsets, we compared their top 10 DEGs. A half of the top 10 DEGs in Inf-SCs and SCs overlapped ( Figure 2A ), while Inf-Cs expressed a different set of genes. The heatmap revealed that Inf-Cs highly expressed the matrix degrading gene, Mmp9, and other inflammation-associated genes (S100a6, S100a11). Inf-SCs and SCs highly expressed Cdkn2c, a gene associated with cellular senescence, and Tuba1b gene-associated with histone modification ( Figure 2B ). Although ~half of the genes expressed by Inf-SCs and SCs were similar, Tuba1b, Rrm1, Cdk1, Asf1b and Snrpd1 were distinct in Inf-SCs, while Kif11, Hist1hib, Cit, Fbxo5, and Nusap1 were expressed only by SCs ( Figures 2A, B ). Gene ontology (GO) analysis revealed that the top 6 GO terms corresponding to biological process in the Inf-C subset were mainly related to chemokine and immune cell-related signaling pathways, while those in the Inf-SC and SC subsets were mainly related to DNA damage and repair, oxidation-reduction and cellular senescence ( Figure 2C ). KEGG analysis of 6 upregulated pathways also showed a trend similar to GO analysis ( Figure 2D ).

SCs acquire senescence-associated secretory phenotype (SASP) (25). We reported higher expression of 12 SASP factors in callus tissue from aged mice than those from young mice, including Tgfβ1, Tnfα, Il1a, Il1b, Il4, Il6, Csf1, Cxcl1, Cxcl2, Ccl3, Ccl5, and Ccl8 (3). Among these SASP factors, only Tgfβ1 and Tnfα showed differential expression in Inf-C and Inf-SC, SC subsets. Tgfβ1 and Tnfα were expressed mainly by Inf-SC and SC subsets ( Figure 3A ). In addition, we examined the expression of 111 cytokine genes listed in a commercial mouse cytokine array (http://www.rndsystems.com/products/proteome-profiler-mouse-xl-cytokine-array_ary028) and found that Ccl6 and Vegfa had different expression profiles among the 3 cell subsets. Ccl6 was expressed mainly by the Inf-C and Inf-SC subsets, while Vegfa was expressed mainly by the Inf-SC and SC subsets ( Figure 3B ). As the heatmap shows in Figure 2B , Mmp9 was expressed by all 3 subsets, but the expression level was the highest in Inf-Cs ( Figure 3C ).

SCs produce SASPs, which include some pro-inflammatory factors produced by Inf-Cs. Thus, It is important to distinguish SCs from Inf-Cs because different drugs would be used to inhibit their adverse effects in some clinical conditions. The big discovery of this study is that Inf-Cs are transcriptionally different from SCs. In the current study, we analyzed scRNA-seq data from aged callus stromal cells and demonstrated that the cells include Inf-C, Inf-SC and SC subsets, based on their gene expression profiles. Cell-cell communication analyses predicted that SCs are the predominant cells that affect Inf-Cs. Further, our in vitro cell culture experiments indicated that SCs promote an inflammatory phenotype in callus mesenchymal progenitor cells, and these inflammatory callus stromal cells have reduced intrinsic osteoblast differentiation capacity. Thus, the big discovery of our study is that Inf-Cs are transcriptionally different from SCs. Our data suggest that prevention of inflammatory phenotypes in mesenchymal progenitor cells by blocking SCs may serve as an additional mechanism for senolytic therapy-enhanced fracture healing.

Why is it important to distinguish Inf-Cs from SCs? In the elderly, inflammation and senescence often co-exist due to age-associated chronic inflammation (28, 29) and age-associated cellular senescence (30, 31). However, inflammation and senescence have different causes, cellular processes, and therapies. Inflammatory processes are associated with immune responses and activation of NF-κB signaling in living organisms, including short-term acute inflammation and long-term chronic inflammation. Chronic inflammation is a low-grade unresolved immune response that occurs during aging, known as inflammaging, and is a risk factor for many aging-related diseases (32). Different from inflammation, cellular senescence is stimulated by multiple stress signals, including replicative senescence, DNA damage, programmed development senescence, and oncogene activation (33). Senescence-associated pathways are associated with activation of the cell cycle regulators, p53/p21^WAF1/CIP1 and p16^INK4A/pRB, and NF-κB and c/EBPβ are important downstream transcription factors (24, 34, 35). While both processes produce pro-inflammatory cytokines, Coleman et al. reported that senescent endothelial cells may have a protective role to limit local inflammatory responses (36). Indicative of the mixed nature of the cellular processes in inflammation and senescence, anti-inflammation treatment was reported to decrease senescence and increase osteogenic function in skeletal stem progenitor cells (9), and senolytic drugs had diminished anti-inflammatory effects on atherosclerotic lesions (37).

Senolytic drugs that kill SC cells have been used in 12 clinical trials for age-associated conditions, including skeletal diseases. For example, Khosla et al. are conducting a phase-2 clinical trial in healthy women over 70 years of age, in which subjects will take the senolytic drugs, dasatinib+quercetin or fisetin in a 3 day-on and 28 day-off cycle for 5 cycles. Serum bone turnover makers, C-terminal telopeptide of type I collagen (bone resorption) and A procollagen type I intact N terminal propeptide (bone formation), will be measured before and after the treatment as primary outcome measures (ClinicalTrials.gov Identifier: NCT04313634). This trial will be closed by April 2023. However, although there is no clinical trial examining the effects of senolytics in human fracture repair, recent studies reported that dasatinib+quercetin promote fracture healing in young (17) and aged mice (3).

Unlike consistent reports of beneficial effects of senolytic drugs in fracture healing (38, 39), conflicting findings have been reported for inflammatory drugs (NSAIDs) in fracture healing, despite their analgesic potency being well established (40). A systematic review of all available literature, including animal and clinical studies, reported a great diversity in the data presented in the studies (41), leaving clinicians puzzled over potential safety issues. Thus, well-randomized prospective clinical trials are warranted.

Chronic inflammation has been shown to induce senescence of mesenchymal stem/progenitor cells in fracture healing and rheumatoid arthritis (9, 42). Josephson et al. reported that mesenchymal stem/progenitor cells in middle-aged mice have an increased senescent phenotype and that NSAID treatment can improve fracture healing in middle-aged mice (9). However, emerging evidence also indicates that senescence is an upstream source of inflammatory factors (37, 43). Biavasco et al. reported that oncogene-activated senescence of human hematopoietic stem and progenitor cells promotes systemic inflammation by secreting TNFα (37, 43). In the present study, we observed that treatment of CaMPCs with SC CM only induced their inflammatory phenotype, but not cellular senescence ( Supplemental Figure 4 ). We suspect that either the SASP induced by H₂O₂ alone was not enough to trigger senescence of CaMPCs or the concentration of SASPs in the CM was too low to induce senescence of CaMPCs.

Can we distinguish Inf-Cs from SCs since both cell types express the same SASP factors? We initially attempted to use SASP expression to distinguish Inf-Cs from SCs in our data set, such as Tgfβ1, Tnfα, Il1a, Il1b, Il4, and Il6, but failed because there is no difference in the expression between Inf-Cs and SCs. We selected cells expressing NF-κB p65 as Inf-Cs because NF-κB is a well-known inducer of inflammation (22). Under our experimental conditions, most of the NF-κB⁺ cells did not stain positively for β-galactosidase, a marker of SCs ( Supplemental Figure 4 ). The finding that callus SCs also express genes associated with ROS and DNA damage can help to distinguish them from Inf-Cs.

Although it is generally accepted that inflammatory stimuli trigger cellular senescence (25, 44), the relationship between senescence and inflammation still be the “chicken or egg conundrum”. Our study demonstrated or partially demonstrated that SCs are different from Inf-Cs and SCs inflamed callus stromal cells, which caused the decreased osteogenic ability of these Inf-stromal cells. Our Cellchat analysis predicts that callus SCs and Inf-SCs affect Inf-Cs, based on the finding that SCs and Inf-SCs produce more ligands/factors than Inf-Cs ( Figure 4A ). Thus, it is possible that within a combined aging and injury micro-environment where the number of both Inf-Cs and SCs are increased, SCs likely function as effector cells to influence Inf-Cs. In addition, Nichenet analysis identified several target genes in Inf-Cs that may be regulated by ligands expressed by SCs and Inf-SCs, such as Itgβ1/Cebpβ, and Gip1/Mcl1 ( Figure 5C ). C/EBPβ is a transcription factor often involved in inflammation (45, 46). Mcl1 (Myeloid Cell Leukemia 1), a member of the Bcl-2 family for maintaining cell viability (47), can act as a molecular switch for double-strand break DNA repair (48) perhaps by having an anti-senescence effect.

We also found that NF-κB⁺ CaMPCs have reduced osteoblast differentiation capacity. This is not surprising because several mechanisms by which NF-κB activation inhibits mesenchymal progenitor cell differentiation to osteoblasts have been reported. For example, conditional deletion of TRAF3 (TNF Receptor Associated Factor 3), a negative regulatory protein of NF-κB in Prx1-expressing mesenchymal progenitor cells, inhibited osteoblast formation by promoting degradation of β-catenin (49). Activation of the Wnt-β-catenin signaling pathway is negatively regulated by formation of a β-catenin destruction complex composed of the proteins, APC (Adenomatous polyposis coli), Annexin and GSK3β (Glycogen synthase kinase 3 β) (50). Interestingly, we found high expression of Anxa1, Anxa2, Gsk3β, Apc in Inf-Cs ( Supplemental Figure 5 ), suggesting that the down-regulation of β-catenin signaling in inflammatory CaMPCs may be related to high expression of these negative regulators.

Limitations of the current study include the following. Since we used Rela/Relb-expressing cells to define Inf-Cs, it is possible that Inf-Cs are heterogenous with different biologic functions. More study is needed to investigate interactions between Inf-Cs and SCs and further sub-cluster analysis of Inf-Cs with a pro-senescent phenotype, such as pro-inflammatory factor-expressing Inf-Cs. We need to determine if these cell subsets exist in vivo. We currently do not have a set of surface markers for flow cytometry or immunostaining to identify these potential subsets, which also limits further determination of their function. We found Inf-SCs, cells that express genes related to both inflammation and cellular senescence. These Inf-SCs appear to have upregulated pathways similar to SCs. The origin of Inf-SCs is not clear and it is not known if they represent an intermediary state between Inf-Cs and SCs. Our trajectory analysis suggests that Inf-SCs do not originate from Inf-Cs ( Supplemental Figure 6 ). Functional differences among these subsets need to be investigated. Another limitation is that we are unable to confirm our findings in patients with bone fracture, which will require high quality of scRNA-seq data from human callus cells. In our scRNAseq dataset, we did not detect the expression of genes related to chondrogenesis (Sox9, Col2a1, Col10a1). We suspect that the digestion method used in this study does not allow us to isolate cells with chondrogenic property from callus cells. In the future, we should modify our isolation protocol to collect cells with chondrogenic property from callus cells and investigate cellular senescence in this cell population.

In summary, callus SCs not only express SASP factors, but also genes related to ROS and DNA damage, which distinguishes them from Inf-Cs. Bioinformatic analyses predict strong interactions between Inf-Cs and SCs and potential influence of SCs on Inf-Cs by producing active ligands, although further analysis of Inf-C sub-populations is needed to elucidate the contribution of Inf-Cs to the increased cellular senescence in aged bone fracture. Despite the limitations, this study has initially answered our original question: what is the difference between inflammatory cells and SCs in aged callus: they have different gene signatures, SCs produce more factors than Inf-Cs, and SCs play a dominant role in driving inflammation and decreased osteogenic capacity in callus stromal cells.

The original contributions presented in the study are included in the article/ Supplementary Material . Further inquiries can be directed to the corresponding author.

The animal study was reviewed and approved by the University of Rochester Committee for Animal Resources (protocol number: 2001-121R).

JL, XL, AM, HZ, BB, and LX designed the study. JL, YY, and HZ performed experiments. JL, XL, AM, and YY performed bioinformatic analyses. JL and LX wrote the original manuscript. JL, XL, LX wrote the original rebuttal for resubmission. All authors contributed to the article and approved the submitted version.