Robust senescence evaluation by transcriptome-based hUSI to facilitate characterizing cellular senescence under various conditions

Robust senescence evaluation by transcriptome-based hUSI to facilitate characterizing cellular senescence under various conditions | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Robust senescence evaluation by transcriptome-based hUSI to facilitate characterizing cellular senescence under various conditions Ting Ni, Jing Wang, Weixu Wang, Jun Yao, Xiaolan Zhou, Gang Wei This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-3920908/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 29 May, 2025 Read the published version in Nature Aging → Version 1 posted You are reading this latest preprint version Abstract Despite the manifestation and contribution of cellular senescence to tissue aging and aging-related disease, the identification of in vivo senescent cells and the recognition of senescence-specific communication still remain challenging. Current senescence evaluation methods rely greatly on expression level of well-known senescence markers, enrichment of aging-related gene sets or weighted sum of curated genes. However, focusing on limited senescence aspects, these methods could not adequately capture the comprehensive senescence features. To evaluate senescence in a more general and unbiased way from the most common and easily accessible transcriptome data, we developed human universal senescence index (hUSI) to quantify human cellular senescence based on a series of weighted genes learned from representative senescence RNA-seq profiles using a machine learning algorithm. hUSI demonstrated its superior performance in distinguishing senescent samples under various conditions and robustness in handling batch effects and sparse profiles. hUSI could uncover the accumulation of senescent cells of various cell types in complex pathological conditions, and reflected the increasing senescence burden of patients and provided potential senotherapeutic targets. Furthermore, combined with gaussian mixture model, hUSI successfully inferred senescent tumor cells in melanoma and identified key target signaling pathways that are beneficial for patient prognosis. Overall, hUSI provides a valuable choice to improve our ability in characterizing cellular senescence under various conditions, illustrating promising implications in aging studies and clinical situations. Biological sciences/Cell biology/Senescence Biological sciences/Computational biology and bioinformatics/Computational models Cellular senescence Quantification Machine learning hUSI Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 1 Introduction Cellular senescence (CS) characterized by irreversible cell cycle arrest is considered a critical factor for aging and aging-related diseases 1 . For instance, by presenting senescence-associated secretory phenotypes (SASP) including increased secretion of pro-inflammatory proteins and other paracrine factors (such as TGF-β family ligands, VEGF, CCL2 and CCL20) 2 , senescent cells can stimulate immune response and cell-cell communication leading pleiotropic effects in various tissues 3 . Targeted clearance of accumulated senescent cells using senolytic drugs 4 or inducing tumor cells into senescence 5 have shown benefits for disease prognosis and healthy lifespan. However, despite several morphological (such as flattening and enlarging 6 ) and molecular markers (such as p16 7 , p21 8 , p15 9 and p27 9 ) are used to characterize senescent cells, identification of in vivo senescent cells still pose a great challenge due to its heterogeneity 10 . Depending on the real situations, CS in diverse cell types can be induced by various intrinsic and extrinsic stressors, such as replicative stress, oncogene activation, chemotherapeutic drugs and ionizing radiation 11 . Therefore, to properly quantify CS degree in multiple scenarios, there is an urgent need for a universal method which enables sensitive capture of comprehensive senescence features, especially in the era where single-cell transcriptome technology has been widely applied to construct cell atlas of human multiple tissues 12 , 13 . Quantifying senescence degree by senescence score is regarded as a convenient and efficient way to monitor senescence status and disease progression 14 . Considering there is no one-size-fits-all marker gene to exclusively indicate senescence, most attempts to evaluate senescent samples mainly depend on the expression level of aging or senescence associated genes derived from differential analysis and literature studies, giving rise to several aging or senescence-associated gene sets (such as CellAge 15 and SenMayo 16 ) and senescence scoring methods (such as DAS + MSS 14 , CS score 17 and lassoCS 18 ). However, due to the variation in gene composition and the limited study dataset (only focus on particular sample type or senescence type), these methods cannot reliably evaluate transcriptional signatures of senescent samples in various contexts and are susceptible to the absence of some pre-defined senescence-related genes. For example, only involving replicative senescence associated genes as the basis for scoring senescence status might produces bias in evaluating senescence in real aging tissues or induced senescence samples. Thus, we sought to utilize publicly available high-quality transcriptome profiles of senescent samples to learn the comprehensive senescence features and develop a reliable and universal senescence score for senescence evaluation. In the present study, to evaluate cellular senescence in a more general and unbiased way, we introduce hUSI that can accurately assess the burden of senescence at both bulk and single-cell levels. It started from collecting representative senescence transcriptome profiles encompassing multiple contexts, including those derived from different platforms, cell types, conditions, and senescence-induction factors. With the criteria of confirmed senescence status and involving diverse cell and senescence types, we finally collected bulk RNA-seq profiles from five cell types and six senescence types induced by both intrinsic and extrinsic stressors. Then, hUSI was developed based on features extracted from these representative senescence transcriptomes by a machine learning model. hUSI demonstrated high accuracy in distinguishing senescent samples from non-senescent samples in different context. Furthermore, hUSI outperformed other current methods in evaluating senescence at single-cell level and remained robust and reliable in multiple senescence samples. Intriguingly, hUSI can uncover senescent cell subpopulation, as illustrated in melanoma, that correlated with improved patient survival, indicating its promising potential in clinical situations. Notably, hUSI distinguished typical signaling pathways (such as TGF-β and BMP pathways) that could promote senescence-associated cell-cell communication in tumor microenvironment. Overall, hUSI provides a universal and robust way to measure senescence burden, enabling more comprehensive investigations into senescence in various experimental and clinical context. 2 Results Development and validation of hUSI To systematically learn and evaluate the comprehensive signature of CS, we developed a workflow including data collection, data re-processing, feature extraction and quantification of senescence degree (Fig. 1 a). We first collected RNA-seq data sets derived from representative human senescence types serving as senescence training samples, which encompassed five cell types (fibroblasts, endothelial cells, astrocytes, melanocytes, and keratinocytes) and six senescence types (ionizing radiation-induced senescence (IRIS), replicative senescence (RS), oxidative stress-induced senescence (OSIS), oncogene-induced senescence (OIS), natural senescence (NS) as well as compound-induced senescence (CIS)), along with the corresponding non-senescent samples serving as young controls 19 – 27 (Extended Data Fig. 1 a,b and Supplementary table S1 ). Next, considering these data sets were derived from different experimental methods and sequencing protocols, we re-preprocessed all the raw data with the same pipeline to generate standard and normalized profiles for feature extraction and validation (Extended Data Fig. 1 b and Methods). After processing, we found that, as expected, CDKN1A and CDKN2B (encode the well-known senescence marker p21 and p15, respectively) showed significant higher expression level, demonstrating reliability of samples and the overall analysis pipeline (Fig. 1 b). Then, we went on to acquire features of the senescence profiles by machine learning algorithms. Multiple machine learning algorithms have been employed for mining genes associated with individual aging or CS, such as regression, elastic net, and random forests 28 – 30 , facilitating quantification of senescence degree in tumor and normal cells. However, senescence, as a complicated and continuous state, its heterogeneity has not been fully considered in these models 31 . In this study, we selected one-class logistic regression (OCLR) algorithm to learn the features of senescence transcriptome profiles, as it has been demonstrated superior performance on capturing cell heterogeneity 32 . After training, OCLR learned the features of senescence samples in training set, in other words, all genes were respectively assigned with different weights representing their contributions to senescence (Methods). In our learned-senescence features, except for genes upregulated in CS and associated with SASP (such as APOD , EHF and SAA2 ) 33 – 35 were assigned with top weights, some genes with high positive or negative weights while their functions in CS were poorly reported (such as OLAH , CADM3 and HMSD ) (Extended Data Fig. 1 c). These results suggest that OCLR not only captures the known senescence features but also identifies potential novel senescence-associated genes. To our surprise, classical senescence marker, including CDKN1A , CDKN2B and SERPINE1 , are only assigned with slightly positive weights, probably because of relatively low expression levels in samples (Supplementary table S2 ). To further examine the biological interpretability of learned-senescence features, gene set enrichment analysis (GSEA) 36 was performed based on the weight of each gene (Methods). We found that multiple senescence associated gene sets were positively enriched, including interferon-gamma response 37 , KRAS signaling 38 , inflammatory response 39 , hypoxia and p53 pathway 40 , 41 (Fig. 1 c, left panel). On the contrary, the proliferation associated pathways (such as G2M checkpoint 42 , E2F targets 43 , mitotic spindles 44 and MYC targets 45 ) were in the negative enrichment terms (Fig. 1 c, right panel). These results supported the reliability of the features learned by OCLR in reflecting senescence. In terms of quantifying senescence degree, the Spearman correlation coefficient between gene weights and expression values was selected as the metric to quantify senescence degree, defined as human universal senescence index (hUSI) (Methods). To test the stability and reliability of hUSI, we used leave-one-out cross-validation (LOOCV) strategy to calculate average correctly rank probability (CRP) for each iteration, and the resulted CRP reached 0.9 (Extended Data Fig. 1 b, d and Methods). We next validated whether hUSI can be influenced by batch effects arising from variations in experimental conditions, sequencing platforms, or analysis pipelines. We compiled bulk RNA-seq datasets from seven independent studies, each comprising oncogene-induced senescent IMR90 cells induced by 4-hydroxytamoxifen (4-OHT), along with corresponding control cells 24 , 46 – 49 (Supplementary table S1 ). We calculated hUSI for each sample and found that all senescent groups have much higher hUSI values (hUSIs) than the non-senescent groups (Fig. 1 d). Besides, with a plethora of genes included in the senescence features for calculation, hUSI is technically more robust to profiles with limited or sparse gene signals such as microarray and single-cell RNA-seq (scRNA-seq) data. Therefore, we applied hUSI on six senescence-related microarray datasets and the most of them were derived from cell types which were not included in training set 50 – 55 . The results showed that hUSIs were consistently higher in all senescent groups compared to non-senescent ones (Extended Data Fig. 2 a). To test the robustness of hUSI, we generated simulated sparse profiles from these microarray transcriptome profiles by randomly zeroing-out expression signals. We found that even zeroing-out 50% of genes expression signals, hUSIs still represented higher levels in all senescent groups (Extended Data Fig. 2 b). All these results suggested that hUSI, based on comprehensive senescence features and effective nonparametric rank-based correlation 56 , is pretty stable and robust. hUSI shows reliable performance in quantifying senescence degree To assess the generalizability of hUSI, we gathered three bulk RNA-seq datasets (including immortal MDAH04 cells and senescent MDAH04 cells induced by different chemical compounds 57 , WI-38 cells treated with 4-OHT for different days 58 and proliferative WI-38 cells and senescent WI-38 cells induced by replication 58 ), notably the conditions of these samples are not exactly same as samples in the training set (Supplementary table S1 ). We found that most senescent groups exhibited significant higher hUSIs compared to non-senescent ones even the sample size is limited (Extended Data Fig. 2 a). Moreover, hUSI also demonstrated its ability to discern aggravated senescence in samples induced by extended 4-OHT exposure time (Fig. 1 e, middle panel). Next, we calculated hUSIs for a large normal samples dataset obtained from the Genotype-Tissue Expression Project (GTEx). We observed that hUSIs progressively and significantly elevated with increasing age, consisting with the continuous accumulation of senescent cells in aging process 59 (Fig. 2 a). To validate the reliability of hUSI in assessing senescence degree, we calculated Spearman correlation coefficient between hUSIs and CS scores, which was a tool based on conducting gene set variation analysis (GSVA) on a curated set of 1,259 genes derived from studies on replicative cell senescence 17 . The results showed overall positive correlations of these two methods (R = 0.7), and across 29/30 tissues (R from 0.28 to 0.85) (Fig. 2 b left panel and Fig. 2 c upper panel). The same strategy was applied on a large tumor samples dataset from The Cancer Genome Atlas (TCGA) datasets. Despite the heterogeneity in tumor samples, hUSIs still showed overall positive correlations with CS scores (R = 0.52) and across all cancer types (R from 0.12 to 0.94) (Fig. 2 b right panel and Fig. 2 c lower panel). Of note, we discovered that hUSIs demonstrated higher variations in different cancers compared to CS scores, which might indicate that hUSI enables to reveal more intrinsic heterogeneity of senescence across different tumor types (Fig. 2 c lower panel). hUSI has better performance in distinguishing senescence cells Given the reliable and robust performance of hUSI on scoring bulk samples under various conditions, we next applied hUSI on four scRNA-seq datasets derived from primary senescent cells induced by various stressors (including oncogene 49 , 60 , radiation 61 , and replication 61 , as well as secondary senescent cells triggered by paracrine signals 62 ) to assess the robustness of hUSI at single-cell level across diverse conditions. The senescence status of these cells had been confirmed in respective studies by examining senescence marker genes and senescence-associated β-galactosidase (SA-β-Gal) staining 49 , 60 – 62 (Supplementary table S1 ). Non-senescent cells from each dataset were also included for comparative analysis. After quantifying the senescence degree of each cell using hUSI, we observed significantly higher hUSI levels in senescent groups than non-senescent groups across all four datasets, supporting the applicability of hUSI on scRNA-seq data (Fig. 3 a). Next, we compared the performance of hUSI with other three groups of senescence qualification strategies (including those based on gene expression level, computed score and enrichment score of single sample GSEA (ssGSEA)) (Methods). First, we first obtained 12 well-known CS or proliferation associated markers ( GLB1 , TP53 , CDKN1A , CDKN2A , CDKN2B , CDK1 , CDK4 , CDK6 , MKI67 , LMNB1 , IL1A , and RB1 ) and separately used their normalized expression values to directly classify cells, as their upregulation or downregulation is widely employed to identify CS status 63 – 71 . We found that only CDKN1A exhibited a higher trend in senescent samples than control samples across all datasets (Extended Data Fig. 3 a, left panel). To better compare the performance of hUSI and the markers in classifying senescent cells in limited scRNA-seq datasets, we randomly split each dataset into 10 folds and replicated the process three times, and then calculated the ranks of average Area Under Curve (AUC) of all units in each dataset (Supplementary table S3 and Methods). We observed that hUSI exhibited excellent performance compared to all the tested classical markers (Fig. 3 b, left panel and Supplementary table S3). Second, we compared hUSI with five existing senescence score computing methods, including DAS, mSS and their combination (DAS + mSS) 14 , lassoCS 18 and CSS 28 . To our surprise, these methods only gave senescent group a higher score level than control group in certain datasets (Extended Data Fig. 3 a, middle panel). We then applied the same strategy above to calculate average AUC ranks. hUSI also achieved the highest average AUC rank compared to all computed senescence scores (Fig. 3 b, middle panel). Additionally, we observed that DAS + mSS, as expected, outperformed both DAS and mSS individually (Fig. 3 b, middle panel). Of note, except for hUSI, all these methods exhibited substantial variations across four datasets (Fig. 3 b, middle panel), supporting the more stable performance of hUSI. Finally, considering aging and senescence-associated gene sets have been commonly used to quantify CS by enrichment score using ssGSEA, in the present study, we collected eight publicly available senescence-associated gene sets (including CellAge 15 , GenAge 72 , ASIG 73 , SASP (downloaded from MSigDB under acessesion ID R-HSA-2559582), AgingAtlas 74 , SenUp 75 ,SenMayo 16 , and SigRS 76 ) to calculated their ssGSEA scores in four scRNA-seq datasets (Supplementary table S4). The result showed that only SenUp gave higher scores for senescent groups than control groups across four scRNA-seq datasets (Extended Data Fig. 3 a, right panel). After calculating average AUC ranks, hUSI still exhibited superior performance over all gene sets, with minimal variation observed across the four single-cell datasets (Fig. 3 b, right panel). Furthermore, we found that genes from all these gene sets can be found in our features, and genes had been assigned with different weights which enable hUSI to capture a broader spectrum of gene expression signals in the senescence evaluation process (Supplementary table S2 ,4). These results above combined to suggest that hUSI has relative superiority and stability across different scRNA-seq datasets comparing to other current methods. hUSI enables to evaluate senescence burden in complex conditions After validating the outperformance of hUSI in distinguishing senescent cells, we next sought to apply hUSI on single-cell data from real pathological tissues. The accumulation of senescent cells has been reported to increase the susceptibility to COVID-19 patients by contributing to SARS-CoV-2-mediated hyperinflammation and cytokine storm 77 , 78 . Consequently, the targeted elimination of these senescent cells has been proposed as a potential treatment strategy for COVID-19 77, 78 . However, the deconvolution of senescent status across various cell types in infected lung tissues and the study of detrimental effects of different senescent cells on patient survival are still lack. Thus, to evaluate the senescence burden of COVID-19 patients, we calculated the hUSIs for a single-nuclei RNA-seq (snRNA-seq) dataset (containing a total of 116,313 nuclei) derived from infected and normal lungs with donor age ranging from 58 ~ 84 years old 79 (Fig. 3 c and Methods). We found that most cell types (including epithelial cells, endothelial cells, fibroblasts, myeloid, and neuronal cells) from COVID-19 patients exhibited significantly higher hUSI values compared to those from normal donors (Fig. 3 d). Intriguingly, a reverse trend was observed in B cells and T cells, suggesting the activation of immune cells following COVID-19 infection 80 (Fig. 3 d). To better discern various senescence status, we applied a gaussian mixture model (GMM) to fit the distribution of hUSIs within all tested cells and successfully classified them into four distinct classes (C1 ~ C4) (Fig. 3 e, left panel and Methods). Their senescence degrees were further validated by the higher expression level of classical senescence-associated genes ( CDKN1A , IL1A , IL6 , IL8 , CCL2 , CXCL10 , MMP9 , SERPINE1 , THBS1 and TIMP1 ) and lower expression level of proliferation markers ( LMNB1 , MKI67 and DHFR ), consisting with the reported elevated cell senescence responses to SARS-Cov-2 infection 81 (Fig. 3 e, right panel). We also observed that COVID-19 lung tissue has a higher proportion of the most senescent cell class (denoted as C4) cells compared to normal tissue (Fig. 3 f), consistent with the reports suggesting a high accumulation of senescent cells in COVID-19 patients 77 , 82 . Besides, patients with faster disease progression showed more accumulation of senescent cells (Fig. 3 g). These results all suggested that hUSI successfully revealed survival-detrimental senescent cells accumulated in COVID-19 lung tissue across various cell types。 We then examined the difference in the fraction of four cell groups for each cell type between COVID-19 and the normal samples. The results showed there are higher fractions of senescent cells existed in the cell types with a higher risk of exposure to SARS-CoV-2 or hyperinflammatory microenvironment, such as monocyte-derived macrophages, inflamed endothelial cells, pathological fibroblasts and alveolar type 1 progenitor cells (AT1) (Fig. 3 h and Extended Data Fig. 4 a, b) 78 , 83 – 85 . While alveolar type 2 progenitor cells (AT2), which are targeted by SARS-CoV-2 through the angiotensin-converting enzyme 2 (ACE2), was reported to exhibit apparent senescence and a proinflammatory phenotypes 86 , AT1 accumulated in a higher proportion in COVID-19 lung tissue than AT2, possibly because AT2 can differentiate into AT1-like cells for alveolar regeneration in COVID-19 patients 87 . To investigate the alterations in senescent cells, we performed differential gene expression analysis between C4 and C2 (which is the second young class (Fig. 3 f)). We did not take C1 class as the control due to its very small cell numbers, which usually lead to some bias in differential analysis. Differential genes (DEGs) of AT1 and AT2 were respectively enriched on KEGG and GO databases. The results showed that senescent AT1 and AT2 cells have enriched on pathways associated with antigen process, extracellular matrix and immune cytotoxicity, especially AT1 has enriched on p53 signaling pathway, indicating a higher relevance of these senescent on infection response, cellular communication and cellular senescence (Fig. 3 i and Extended Data Fig. 4 c). In addition, DEGs were also enriched in senescent monocyte-derived macrophages as it showed largest fraction difference in C4, reaching 0.29, and was reported to drive the inflammatory response to SARS-CoV-2 and contribute to cytokine storms in severe COVID-19 88, 89 . The results showed that pathways, including positive regulation of T cell-mediated immunity and leukocyte-mediated cytotoxicity, was enriched in these cells, indicating their crucial roles of senescent cells in macrophage-mediated clearance of infected cells, which may also cause damages to infected tissues by hyperinflammatory 90 (Extended Data Fig. 4 d). All these results above demonstrated that hUSI enables to recognize senescent cells that abnormally accumulated in pathological tissue and reveal associated mechanisms. hUSI identifies immune associated senescent tumor cells in melanoma We then sought to apply it on tumor samples, as CS plays an important role in tumor development and can activate immune responses 91 . Significant progress has been made in immunotherapy of melanoma, especially with the application of immune checkpoint inhibitors, such as PD-1 antibodies and CTLA-4 antibodies, which result in significant durable responses and therapeutic efficacy 92 , 93 . However, the mechanisms underlying immunotherapy remain incompletely understood. Several studies have demonstrated the relationship between senescent tumor cells and immune recognition 94 – 97 , thus we sought to identify senescent tumor cells and investigate whether it could serve as potential targets for immunotherapy in melanoma. To explore the characteristics of senescent tumor cells in melanoma, we evaluated the senescence degree of tumor cells by applying hUSI on a melanoma scRNA-seq data set 98 . We then used GMM to infer three cell subpopulations which were denoted as cycling , transactional and senescent , based on the significantly increasing hUSI level (Extended Data Fig. 5 a, b). The senescence degree of these subpopulations was further validated by a microarray-based transcriptome dataset of melanocytes 52 . By overlapping DEGs of melanocytes bulk samples with specific highly expressing genes in our defined cell subpopulations (Methods), we found that genes up-regulated in senescent melanocytes were significantly enriched in senescent and transitional subpopulations (Fig. 4 a and Supplementary table S5). On the contrary, genes up-regulated in growing melanocytes were significantly enriched only in cycling subpopulation (Fig. 4 a and Supplementary table S5). We also validated the different senescence degree of these three subpopulations by inferring a senescence trajectory of tumor cells. We imputed 38 co-expression modules based on hUSI-related genes and diffusion map was used for dimensionality reduction and visualization. The senescence trajectory was characterized by the transition of tumor cells from cycling to senescent status (Fig. 4 b and Extended Data Fig..5c, d and Methods). Two well-known senescence hallmark genes, CDKN1A and SERPINE1 , showed higher expression level in senescent subpopulation than in the other two subpopulations ( cycling and transitional ) (Fig. 4 c). Moreover, GSEA results of specific highly expressing genes in each subpopulation indicated more frequent immune activities occurred in senescent tumor cells (Fig. 4 d and Methods). These results demonstrated that heterogeneity in senescence existed among melanoma tumor cells, and hUSI can reliably distinguish senescent tumor cells. We next analyzed the impact of senescent tumor cells on melanoma patient survival. We took three tumor cell subpopulations as a reference expression profile and deconvoluted RNA-seq profiles of melanoma cohort from TCGA-SKCM using EpiDISH 99 , obtaining the proportion of each subpopulation in each melanoma patient (Methods). Considering the potential relationship between senescent tumor cells and immune response, we also calculated abundances of 22 immune component using CIBERSORT 100 . We found that the proportion of senescent subpopulation have a higher positive correlation with the abundance of M1 macrophage cells, CD8 T cells, and activated immune cells (including activated CD5 memory T cells and activated dendritic cells) (Fig. 4 e). Survival analysis was then performed based on the inferred proportions of these subpopulations. The result showed that the higher proportion of senescent or transactional subpopulations in a patient, the more favorable it was for the patient’s survival, and the significance of senescent is higher than transactional (Fig. 4 f). In contrast, patients with higher proportion of cycling subpopulation have worse prognosis (Fig. 4 f and Extended Data Fig. 5 e). These results suggested that hUSI-aided senescence state evaluation of tumor cells can serve as a promising prognostic biomarker for melanoma patients. hUSI recognizes special signaling pathways in senescent melanoma cells In the above analysis, hUSI helps identify senescent tumor cells in melanoma. However, the role of senescent cells in tumor microenvironment is very complex and highly dependent on the physiological environment 101 – 103 . Senescent cells can communicate with neighbor cells and influence their behavior through paracrine signaling. Specifically, SASP presented by senescent tumor cells plays important roles in communication with immune system by attracting immune cells (such as T cells and NK cells) and then leading to the clearance of senescent tumor cells 95 – 97 . Besides, CS associated communication had been speculated to regulate immune surveillance and influence tumorigenesis 104 . Therefore, understanding how senescent cells interact with the microenvironment may provide additional clues for the relationship between senescence and tumorigenesis. To explore the cross-talk between senescent tumor cells and the microenvironment in melanoma, we investigated the cell-cell communication between these three tumor cell subpopulations ( cycling , transactional and senescent ) and their neighboring cells using CellChat 105 (Fig. 5 a). The results showed that the communication strength of hUSI-identified senescent subpopulation was higher than the other two relatively less senescent tumor cell subpopulations ( cycling and transactional ), indicating stronger cell-cell communication between senescent tumor cells and neighboring cells (Fig. 5 b and Extended Data Fig. 5 f). Furthermore, analysis of the global output communication patterns uncovered two different signaling patterns, with pattern 1 corresponding to the senescent tumor subpopulation and pattern 6 corresponding to the cycling and transitional subpopulations (Extended Data Fig. 5 g). To analyze which pathways were responsible for senescent tumor cells to receive communication signals from tumor microenvironment, we compared communication strength of each involved signaling pathway (Methods). Six pathways (including transforming growth factor β (TGFβ) pathway, leptin (LEP) pathway, chondroitin sulfate proteoglycan 4 (CSPG4) pathway, chemokine signaling pathways (CCL), CD6 pathway and bone morphogenetic protein (BMP) pathway) were found to have input signal strength to senescent subpopulation and not detected in cycling subpopulation, indicating that these signaling pathways are more likely to specifically function in senescent tumor cells (Fig. 5 c). In the two major pathways, TGF-β can induce senescent phenotype of tumor cells, which is secreted by macrophages originating in tumor stroma 106 , 107 , and BMP is a family of TGF-β superfamily, which has similarly been found to induce senescence of tumor cells 108 , 109 . Through signaling pathways pathway network, we found that senescent tumor cells receive TGF-β from macrophages (Fig. 5 d), which is consistent with previous report in lymphoma 107 . Interestingly, senescent tumor cells received more TGF-β from cancer-associated fibroblasts (CAFs) (Fig. 5 d). This may indicate that as a solid tumor, melanoma differs from lymphoma in microenvironment by the presence of a high number of fibroblasts. Moreover, senescent tumor cells received BMP from a variety of cell types in the microenvironment, of which the signal from T cells was the strongest (Fig. 5 d). Further investigation of ligand-receptor interactions in signaling pathways revealed that the expression level of genes encoding receptors for TGF-β and BMP were higher in senescent subpopulation than in cycling and transitional subpopulations, with TGFBR2 and BMPR1B showing more significant differences among these three subpopulations (Fig. 5 e). In addition, survival analysis performed on TCGA-SKCM also showed that patients with higher expression level of TGFBR1 , TGFBR2 and BMPR2 have better prognosis (Fig. 5 f), consistent with the idea that stronger interactions by these pathways between senescent tumor cells and microenvironment could benefit patient survival. We also noticed the other four signaling pathways which are also specific for senescent tumor cells. While transitional and senescent subpopulation interact with T cell by LEP signaling pathway and with CAFs by CSPG4 signaling pathway, the receptors involved in these two pathways did not show significant difference (Extended Data Fig. 6a). Notably, senescent subpopulation interacts with T cell by CD6 signaling pathway and with macrophages by CCL signaling pathway. CD6 receptor encoding gene ALMCAM and CCL receptor encoding gene CCR10 are specifically highly expressed in senescent subpopulation and benefit patients’ survival (Extended Data Fig. 6b, c). Although, these signaling pathways are reported associated with tumor progression 110 – 113 , their functions on tumor cell senescence need further study. Overall, these results highlight the clinical value of hUSI in identifying senescent tumor cells and the potentially involved signaling pathways. 3 Methods Data collection. Bulk RNA-seq datasets used for extraction of senescence features were collected from previously published papers 19 – 27 . We downloaded raw files from SRA database ( https://www.ncbi.nlm.nih.gov/sra ) (GSE53356, GSE56293, GSE58910, GSE61130, GSE63577, GSE64553, GSE113957, GSE130727, and GSE60883) and EMBL-EBI ( https://www.ebi.ac.uk/ebisearch/about ) (E-MTAB-5403). Only samples with confirmed senescence status were selected to build the representative senescence profiles. We also included corresponding non-senescent samples in each dataset for LOOCV. Seven microarray profiles used for batch effects test were download from Gene Expression Omnibus (GEO) database ( https://www.ncbi.nlm.nih.gov/geo/ ) (GSE101750, GSE101758, GSE61130, GSE122079, GSE113060, GSE72407, and GSE72404). The RNA-seq and microarray profiles of independent validation datasets were also downloaded from SRA database through GEO accession numbers: GSE60340, GSE130306, GSE19864, GSE16058, GSE83922, GSE11954, GSE100014, and GSE77239. Four scRNA-seq profiles used for benchmarking were downloaded from GEO (GSE119807, GSE115301, GSE94980, and GSE81547). More details of datasets mentioned above can be found in Supplementary Table S1 . The TPM normalized gene expression matrix, of which, TCGA was collected from UCSC Xena ( http://xena.ucsc.edu/public/ ), and GTEx was collected from GTEx Portal (version 8) ( https://www.gtexportal.org/home/downloads/adult-gtex ). The raw single-nuclei counts matrix of normal and COVID-19 patients’ lung tissues and the processed melanoma profiles were also respectively downloaded from GEO under accession numbers GSE171524 and GSE72056. Bulk RNA-seq data of training set processing. We used Prefetch v.3.0.2 to download SRA files and split them into FASTQ files by parallel-fastq-dump v.0.6.5. TrimGalore v.0.6.6 was used to filter out low quality reads and bases in 3’ end for the consequential alignment performed by STAR v.2.2.1. We chose GRCh38 as reference genome sequence and only unique mapping reads were included. StringTie was used to qualify the gene expression level and normalize values by TPM (transcripts per million). Only protein coding genes (annotated by gencode v.31) with TPM > 3 in 99% samples were included for next analysis. Considering bias introduced by batch effect, we also used log space transform to further reduce the disturbance for feature extraction. Quantification of cellular senescence based on OCLR. To quantify CS based on gene expression level, a predictive model was built using OCLR 32 in R package “gelnet” with the parameters y = NULL, l1 = 0, and l2 = 1. The input expression matrix of senescent cells was normalized by subtracting mean expression value across all samples. The Spearman correlation coefficient was defined as hUSI as its stable performance for minimizing possible batch effects across datasets 114 . Normalized gene expression matrix was used to calculate hUSI. Gene set enrichment analysis for learned-senescence features. Hallmark gene sets from The Molecular Signatures Database (MSigDB) database were included to perform GSEA for genes in OCLR learned-features which were sorted by their weights. Normalized enrichment score (NES) calculated by R package “fgsea” 115 was chosen to compare the enrichment degree in different gene sets. Performance evaluation of hUSI. The reliability of the acquired feature and quantification strategy were validated using LOOCV. For each training round, we excluded one senescent sample in our training set and trained the model to extract senescence features. Then hUSI was calculated based on features to score the leave-out sample as well as all the non-senescent samples. Finally, the probability that the score of senescent samples is higher than that of non-senescent samples 116 was used to measure the performance, denoted as correctly ranking probability (CRP). For the four scRNA-seq datasets GSE119807, GSE115301, GSE94980, and GSE81547 ) used for comparison of three types of scoring methods, the expression matrix was read and normalized using R package “Seurat”. Then, log-normalized gene expression value was used for directly classifying senescent cells or calculating senescence score. ssGSEA score for each gene set was produced by “gsea()” function in the R package “GSVA”. To better compare the performance of all scoring methods in four single-cell datasets, we randomly divided each dataset into 10 subsets and repeated for 3 times by “createMultiFolds()” function in R package “caret”, finally generating 30 data units in total. For each of the 30 data units, we computed the AUC based on the scores generated by the tested methods described below by “auc()” function in R package “pROC”. For the expression level of marker genes, the parameter “direction” was not assigned as these genes were reported up or down regulated in senescent cells. For the computed senescence score and ssGSEA score, we set the parameter “direction=<” to make sure it will have a higher AUC only if these scores are lower in non-senescent samples. We calculated the average AUC of all 30 units and ascendingly ranked the average AUCs derived from each method group for each dataset. The mean average AUC rank across four datasets was used to reflect their overall performance (Supplementary Table S3). Inference of cellular senescence states. For a large number of single cells, we adopted a GMM framework 117 to infer the number of potential senescence states according to the distribution of hUSI. Log transformation was firstly performed on the hUSI of each cell, that is, Logit(hUSI) = log2[(1 + hUSI)/(1-hUSI)]. The Logit(hUSI) values of all cells were then fitted under the framework of GMM (implemented in R package “mclust”), and the Bayesian information criterion (BIC) was used to estimate the optimal number of senescence states and the probability that each cell belonged to a specific state 118 . Analysis of snRNA-seq data of COVID-19 infected samples . The raw expression matrix was read using Python module "scanpy" to filter out low quality cells (min_genes = 200 and min_cells = 3). After removing mitochondrial and ribosomal genes, the expression matrix was normalized, and the highly variable gene matrix was considered for downstream scaling, batch effect removing and visualization. The "rpy2" module was used to call the "mclust" R package in python for cellular senescence state classification. GSEA for DEGs of senescence class was performed by python module “gseapy” 119 using log 2 -transformed Fold-Change as ranking metric. Analysis of scRNA-seq data of melanoma samples. The processed melanoma single-cell matrices, retaining only defined tumor cells (malignant = = 2) and non-tumor cells (malignant = = 1) (including T-cells, B-cells, macrophages, endothelial cells, cancer-associated fibroblasts and natural killer cells), were read, analyzed and visualized using the R package "Seurat". To validate the reliability of inferred tumor subpopulations, specifically highly expressing genes (logfc.threshold > 0.1) of each subpopulation were overlapped with DEGs derived from melanoma microarray data (senescent vs young), which were calculated by linear models “lm(gene expession ~ pheno)”. R function “phyper()” was used to test the overlapping significance. To observe the positional relationships of the different subpopulations of CS states on projected space, for each gene, we calculated the Pearson correlation coefficient between hUSIs and gene expression values of all cells, and the top 1500 genes ranked by absolute correlation coefficient values were selected as hUSI related genes. Then, using the tool ICAnet 120 , the tumor cells were integrated based on the 38 co-expression modules of the above hUSI related genes. Diffusion map based on five principal components of 38 co-expression modules was used to further reduce the dimension and infer senescence trajectory using R package “destiny” 121 . Specifically highly expressed genes (logfc.threshold > 0.1) of tumor subpopulation were enriched using R package “clusterProfiler ” 122 on the database Gene-Ontology (GO), Kyoto Encyclopedia of Genes and Genomes (KEGG), using log 2 -transformed Fold-Change as the ranking metric. Cell communication analysis was carried out using R package “CellChat”, and the communication intensity between tumor subpopulations and different non-tumor cell types in a signal network was quantified 105 . When filtered pathways specific for senescent tumor cells, we set three tumor subpopulations as “target” and T cell, NK cell, macro cell and CAF cell as “source”. Data analysis of melanoma patient cohort in TCGA. The normalized Level 3 RNA-seq data of a melanoma patient cohort (SKCM) with associated clinical data were downloaded from the TCGA (the Cancer Genome Atlas) database using R package TCGAbiolinks 123 . To analyze the proportion of cells with different senescence degree, three tumor subpopulations were used as reference to deconvolute the RNA-seq data of the SKCM patient cohort using EpiDISH 99 . The abundance of the 22 immune components were calculated by CIBERSORT 100 . Survival analysis was performed using R packages “survival” and “survminer” with “OS = vital_status” and “OS.time = days_to_last_followup”. 4 Discussion Though the characterization, identification, and pharmacological clearance of senescent cell are the basics of many senotherapies, the relative scarce of specific and efficient senescence marker keeps limiting the study of distinguishing and targeting senescent cells both in vitro and in vivo 1 , 10 , 124 , 125 . Additionally, in single-cell studies, the differential expression level of a single senescence marker is insufficient to identify senescent cells due to the heterogeneity of both cell types and senescence status 64 . Besides, apart from the absence or low expression of certain senescence markers, the possibility of improperly calculating senescence score in some methods also exists due to the absent detection or abnormal expression of key senescence genes. For example, in lassoCS 18 method, three out of ten genes ( SEMA3G , PCSK6 , and SLC44A4 ) are assigned with weight values for senescence score calculation, while they are absent when we applied it to another single-cell dataset 48 . Taking gene sets into consideration, while the enrichment score generated by GSEA or GSVA is widely utilized, different gene sets usually emphasize on different aspects of senescence. For instance, the SASP gene set focuses on the activated secretory phenotype, whereas signature of replicative senescence 76 specifically considers replicative-related changes. Notably, performing GSEA in large-scale single-cell atlases can be exceedingly time-consuming, due to the substantial number of permutations required to accurately estimate the nominal p-value 126 . To overcome these challenges above, the present study adopted OCLR machine learning algorithm 32 to acquire gene expression features of CS in relatively comprehensive senescence transcriptome profiles. Based on OCLR machine learned-features, we developed hUSI, a scoring method using Spearman correlation coefficient that can distinguish senescent samples or cells induced by different factors while less affected by batch effect. We validated the generalizability of hUSI by applying it to datasets of variety origins encompassing platforms, cell types, or induction factors that were not included in the training set. Moreover, the stability and the potential application of hUSI on single-cell data were further validated by simulated sparse profiles and real scRNA-seq profiles. Comparing with other currently available methods (including those based on senescence markers expression level, computed senescence score and ssGSEA score), hUSI manifested its reliability and superior performance in senescence evaluation. Unlike methods that rely on limited genes, hUSI takes14,638 protein-coding genes assigned with different weights learned by OCLR into senescence evaluation, which reduces bias when evaluating CS of distinct senescence types. Importantly, hUSI demonstrated reliable performance when applied to samples derived from both health and disease populations. Combination of hUSI with GMM provided a framework to reveal cell classes with different senescence level. hUSI identified sell types showing higher accumulation of senescent cells in COVID-19-infected lung tissue and provided a potential therapeutic avenue for selectively eliminating these senescent cells to mitigate senescence-induced hyperinflammation in COVID-19 patients 82 . For example, navitoclax, a senolytic drug, has been reported to target senescent alveolar epithelial cells and macrophages, and in turn reduces the secretion of pro-inflammatory SASP factors after SARS-CoV-2 infection 81 . Thus, our ongoing research will center on integrating senescence quantification into the cellular response to drugs to unveil the contribution of senescent cells to both drug resistance and drug sensitivity across various diseases. By investigating the communication network of melanoma cells of different senescence degree, combing with previous studies 106 , 107 , we hypothesize that senescent tumor cells can interact with tumor microenvironment by TGF-β and BMP signaling pathways in melanoma microenvironment and contribute to anti-cancer effects. TGF-β is well known being associated with the upregulated expression of p15, p21, and p27, which are known senescence markers and can inhibit cell proliferation 127 . BMP, a member of TGF-β family, has also been reported having a crucial role in paracrine induction in senescent cells 108 , 109 . We observed higher expression level of genes encoding TGF and BMP receptors in senescent cells, which support their specifical functions in senescence of tumor cells. Although hUSI has demonstrated superior performance in multiple aspects, we must admit that there are still some limitations in current status. First, though the stability of hUSI has been demonstrated by LOOCV and independent datasets, with more data coming in the big data era, the quality and quantity of training sets still has room for further improvement to optimize the performance of hUSI in the future. Second, cell-cell interactions networks are dynamic and intricate, no matter among senescent cells in different cell types or between senescent cells and non-senescent cells. However, in this study, we simplified the interactions networks by only focusing on senescent tumor cells due to the limited cell numbers of other cell types. Besides, the causal relationship between signaling pathways and the mechanism of how senescent tumor cells response to different signals also require further study. Third, hUSI was a human transcriptome-based scoring method. Although, using homologous genes when applied to other species is a feasible way, the reliability and the accuracy need extensive validation. We also believed senescence scoring tool can be developed based on transcriptomes of interested species following our stagey. Finally, with implementation of quantifying CS degree, a more detailed and standardized CS atlas supported by adequate experimental and multi-omics evidence is demanding to benefit the prevention of aging-related diseases and the application of senotherapeutics. 5 Conclusion In summary, we developed a senescence-evaluating tool that outperforms currently existing analogous methods, capable of robustly quantifying sample senescence degree based on bulk or single-cell transcriptome profiles. We also proposed a framework for classifying the senescence status of various cell types and recognizing senescence-specific intercellular communications. Based on the outperformance and applicability of hUSI, we believe that it will greatly help to evaluate senescence and benefit studies and even therapeutic strategies in senescence and age-related diseases. Declarations Funding This work was financially supported by the National Natural Science Foundation of China (92249302, 32370592), the National Key Research and Development Program of China (2023YFC3603300, 2021YFA0909300). Conflict of interest The authors declare no competing interests. Code availability hUSI can be implemented both in R and Python, the data and codes used to reproduce our analysis results are provided in GitHub (https://github.com/WJPina/HUSI), along with a detailed usage guideline. Author contribution The manuscript was written by J.W., W.W., and J.Y. and polished by G.W., and T.N.. The method was conceived by J.Y. and W.W. and the algorithm is implemented by J.W., and J.Y. Computational analyses and algorithm evaluations were conducted by J.W., W.W., J.Y., and X.Z. This work was supervised by T.N.. References van Deursen, J.M. The role of senescent cells in ageing. Nature 509 , 439-46 (2014). Acosta, J.C. et al. A complex secretory program orchestrated by the inflammasome controls paracrine senescence. Nat Cell Biol 15 , 978-90 (2013). Birch, J. & Gil, J. Senescence and the SASP: many therapeutic avenues. Genes Dev 34 , 1565-1576 (2020). Chaib, S., Tchkonia, T. & Kirkland, J.L. Cellular senescence and senolytics: the path to the clinic. Nat Med 28 , 1556-1568 (2022). Wang, L., Lankhorst, L. & Bernards, R. Exploiting senescence for the treatment of cancer. Nat Rev Cancer 22 , 340-355 (2022). Cho, K.A. et al. Morphological adjustment of senescent cells by modulating caveolin-1 status. J Biol Chem 279 , 42270-8 (2004). Petrova, N.V., Velichko, A.K., Razin, S.V. & Kantidze, O.L. Small molecule compounds that induce cellular senescence. Aging Cell 15 , 999-1017 (2016). Rossiello, F., Herbig, U., Longhese, M.P., Fumagalli, M. & D'Adda, D.F.F. Irreparable telomeric DNA damage and persistent DDR signalling as a shared causative mechanism of cellular senescence and ageing. Curr Opin Genet Dev 26 , 89-95 (2014). Munoz-Espin, D. et al. Programmed cell senescence during mammalian embryonic development. Cell 155 , 1104-18 (2013). Hernandez-Segura, A., Nehme, J. & Demaria, M. Hallmarks of Cellular Senescence. Trends Cell Biol 28 , 436-453 (2018). Huang, W., Hickson, L.J., Eirin, A., Kirkland, J.L. & Lerman, L.O. Cellular senescence: the good, the bad and the unknown. Nature reviews. Nephrology 18 , 611-627 (2022). Han, X. et al. Construction of a human cell landscape at single-cell level. Nature 581 , 303-309 (2020). He, S. et al. Single-cell transcriptome profiling of an adult human cell atlas of 15 major organs. Genome Biol 21 , 294 (2020). Lafferty-Whyte, K. et al. Scoring of senescence signalling in multiple human tumour gene expression datasets, identification of a correlation between senescence score and drug toxicity in the NCI60 panel and a pro-inflammatory signature correlating with survival advantage in peritoneal mesothelioma. BMC Genomics 11 , 532 (2010). de Magalhaes, J.P., Curado, J. & Church, G.M. Meta-analysis of age-related gene expression profiles identifies common signatures of aging. Bioinformatics 25 , 875-81 (2009). Saul, D. et al. A new gene set identifies senescent cells and predicts senescence-associated pathways across tissues. Nat Commun 13 , 4827 (2022). Wang, X. et al. Comprehensive assessment of cellular senescence in the tumor microenvironment. Brief Bioinform 23 (2022). Gong, Q., Jiang, Y., Xiong, J., Liu, F. & Guan, J. Integrating scRNA and bulk-RNA sequencing develops a cell senescence signature for analyzing tumor heterogeneity in clear cell renal cell carcinoma. Front Immunol 14 , 1199002 (2023). Hernandez-Segura, A. et al. Unmasking Transcriptional Heterogeneity in Senescent Cells. Curr Biol 27 , 2652-2660.e4 (2017). Casella, G. et al. Transcriptome signature of cellular senescence. Nucleic Acids Res 47 , 7294-7305 (2019). Rai, T.S. et al. HIRA orchestrates a dynamic chromatin landscape in senescence and is required for suppression of neoplasia. Genes Dev 28 , 2712-25 (2014). Alspach, E. et al. P38MAPK plays a crucial role in stromal-mediated tumorigenesis. Cancer discovery 4 , 716 Crowe, E.P. et al. Changes in the Transcriptome of Human Astrocytes Accompanying Oxidative Stress-Induced Senescence. Front Aging Neurosci 8 , 208 (2016). Herranz, N. et al. mTOR regulates MAPKAPK2 translation to control the senescence-associated secretory phenotype. Nat Cell Biol 17 , 1205-17 (2015). Marthandan, S. et al. Similarities in Gene Expression Profiles during In Vitro Aging of Primary Human Embryonic Lung and Foreskin Fibroblasts. Biomed Res Int 2015 , 731938 (2015). Marthandan, S. et al. Hormetic effect of rotenone in primary human fibroblasts. Immun Ageing 12 , 11 (2015). Fleischer, J.G. et al. Predicting age from the transcriptome of human dermal fibroblasts. Genome Biol 19 , 221 (2018). Lin, W. et al. Identification and validation of cellular senescence patterns to predict clinical outcomes and immunotherapeutic responses in lung adenocarcinoma. Cancer Cell Int 21 , 652 (2021). Park, H.S. & Kim, S.Y. Endothelial cell senescence: A machine learning-based meta-analysis of transcriptomic studies. Ageing Res Rev 65 , 101213 (2021). Jochems, F. et al. The Cancer SENESCopedia: A delineation of cancer cell senescence. Cell Rep 36 , 109441 (2021). Kumari, R. & Jat, P. Mechanisms of Cellular Senescence: Cell Cycle Arrest and Senescence Associated Secretory Phenotype. Front Cell Dev Biol 9 , 645593 (2021). Sokolov, A., Paull, E.O. & Stuart, J.M. ONE-CLASS DETECTION OF CELL STATES IN TUMOR SUBTYPES. Pac Symp Biocomput 21 , 405-16 (2016). Lim, S., Lim, J., Lee, A., Kim, K.I. & Lim, J.S. Anticancer Effect of E26 Transformation-Specific Homologous Factor through the Induction of Senescence and the Inhibition of Epithelial-Mesenchymal Transition in Triple-Negative Breast Cancer Cells. Cancers (Basel) 15 (2023). Takaya, K., Asou, T. & Kishi, K. Identification of Apolipoprotein D as a Dermal Fibroblast Marker of Human Aging for Development of Skin Rejuvenation Therapy. Rejuvenation Res 26 , 42-50 (2023). Hari, P. et al. The innate immune sensor Toll-like receptor 2 controls the senescence-associated secretory phenotype. Sci Adv 5 , eaaw0254 (2019). Liberzon, A. et al. The Molecular Signatures Database (MSigDB) hallmark gene set collection. Cell Syst 1 , 417-425 (2015). Kim, K.S., Kang, K.W., Seu, Y.B., Baek, S.H. & Kim, J.R. Interferon-gamma induces cellular senescence through p53-dependent DNA damage signaling in human endothelial cells. Mech Ageing Dev 130 , 179-88 (2009). Cisowski, J., Sayin, V.I., Liu, M., Karlsson, C. & Bergo, M.O. Oncogene-induced senescence underlies the mutual exclusive nature of oncogenic KRAS and BRAF. Oncogene 35 , 1328-33 (2016). Lasry, A. & Ben-Neriah, Y. Senescence-associated inflammatory responses: aging and cancer perspectives. Trends Immunol 36 , 217-28 (2015). Artandi, S.E. & Attardi, L.D. Pathways connecting telomeres and p53 in senescence, apoptosis, and cancer. Biochem Biophys Res Commun 331 , 881-90 (2005). Serrano, M., Lin, A.W., McCurrach, M.E., Beach, D. & Lowe, S.W. Oncogenic ras provokes premature cell senescence associated with accumulation of p53 and p16INK4a. Cell 88 , 593-602 (1997). Oshi, M. et al. G2M Cell Cycle Pathway Score as a Prognostic Biomarker of Metastasis in Estrogen Receptor (ER)-Positive Breast Cancer. Int J Mol Sci 21 (2020). Narita, M. et al. Rb-mediated heterochromatin formation and silencing of E2F target genes during cellular senescence. Cell 113 , 703-16 (2003). Dikovskaya, D. et al. Mitotic Stress Is an Integral Part of the Oncogene-Induced Senescence Program that Promotes Multinucleation and Cell Cycle Arrest. Cell Rep 12 , 1483-96 (2015). Wu, C.H. et al. Cellular senescence is an important mechanism of tumor regression upon c-Myc inactivation. Proc Natl Acad Sci U S A 104 , 13028-33 (2007). Guerrero, A. et al. Cardiac glycosides are broad-spectrum senolytics. Nat Metab 1 , 1074-1088 (2019). Hoare, M. et al. NOTCH1 mediates a switch between two distinct secretomes during senescence. Nat Cell Biol 18 , 979-92 (2016). Parry, A.J. et al. NOTCH-mediated non-cell autonomous regulation of chromatin structure during senescence. Nat Commun 9 , 1840 (2018). Georgilis, A. et al. PTBP1-Mediated Alternative Splicing Regulates the Inflammatory Secretome and the Pro-tumorigenic Effects of Senescent Cells. Cancer Cell 34 , 85-102.e9 (2018). Costarelli, L. et al. Different transcriptional profiling between senescent and non-senescent human coronary artery endothelial cells (HCAECs) by Omeprazole and Lansoprazole treatment. Biogerontology 18 , 217-236 (2017). Chicas, A. et al. Dissecting the unique role of the retinoblastoma tumor suppressor during cellular senescence. Cancer Cell 17 , 376-87 (2010). Orfanidis, K., Waster, P., Lundmark, K., Rosdahl, I. & Ollinger, K. Evaluation of tubulin beta-3 as a novel senescence-associated gene in melanocytic malignant transformation. Pigment Cell Melanoma Res 30 , 243-254 (2017). Garbe, J.C. et al. Molecular distinctions between stasis and telomere attrition senescence barriers shown by long-term culture of normal human mammary epithelial cells. Cancer Res 69 , 7557-68 (2009). Krizhanovsky, V. et al. Senescence of activated stellate cells limits liver fibrosis. Cell 134 , 657-67 (2008). Yuan, L. et al. Switching off IMMP2L signaling drives senescence via simultaneous metabolic alteration and blockage of cell death. Cell Res 28 , 625-643 (2018). Somekh, J., Shen-Orr, S.S. & Kohane, I.S. Batch correction evaluation framework using a-priori gene-gene associations: applied to the GTEx dataset. BMC Bioinformatics 20 , 268 (2019). Purcell, M., Kruger, A. & Tainsky, M.A. Gene expression profiling of replicative and induced senescence. Cell Cycle 13 , 3927-37 (2014). Sati, S. et al. 4D Genome Rewiring during Oncogene-Induced and Replicative Senescence. Mol Cell 78 , 522-538.e9 (2020). Borghesan, M., Hoogaars, W., Varela-Eirin, M., Talma, N. & Demaria, M. A Senescence-Centric View of Aging: Implications for Longevity and Disease. Trends Cell Biol 30 , 777-791 (2020). Aarts, M. et al. Coupling shRNA screens with single-cell RNA-seq identifies a dual role for mTOR in reprogramming-induced senescence. Genes Dev 31 , 2085-2098 (2017). Tang, H. et al. Single senescent cell sequencing reveals heterogeneity in senescent cells induced by telomere erosion. Protein Cell 10 , 370-375 (2019). Teo, Y.V. et al. Notch Signaling Mediates Secondary Senescence. Cell Rep 27 , 997-1007.e5 (2019). Minamino, T. et al. A crucial role for adipose tissue p53 in the regulation of insulin resistance. Nat Med 15 , 1082-7 (2009). Wiley, C.D. et al. Analysis of individual cells identifies cell-to-cell variability following induction of cellular senescence. Aging Cell 16 , 1043-1050 (2017). Orjalo, A.V., Bhaumik, D., Gengler, B.K., Scott, G.K. & Campisi, J. Cell surface-bound IL-1alpha is an upstream regulator of the senescence-associated IL-6/IL-8 cytokine network. Proc Natl Acad Sci U S A 106 , 17031-6 (2009). Diril, M.K. et al. Cyclin-dependent kinase 1 (Cdk1) is essential for cell division and suppression of DNA re-replication but not for liver regeneration. Proc Natl Acad Sci U S A 109 , 3826-31 (2012). Alessio, N. et al. Different Stages of Quiescence, Senescence, and Cell Stress Identified by Molecular Algorithm Based on the Expression of Ki67, RPS6, and Beta-Galactosidase Activity. Int J Mol Sci 22 (2021). Kansara, M. et al. Immune response to RB1-regulated senescence limits radiation-induced osteosarcoma formation. J Clin Invest 123 , 5351-60 (2013). McConnell, B.B., Starborg, M., Brookes, S. & Peters, G. Inhibitors of cyclin-dependent kinases induce features of replicative senescence in early passage human diploid fibroblasts. Curr Biol 8 , 351-4 (1998). Liu, S. et al. Senescence of human skin-derived precursors regulated by Akt-FOXO3-p27(KIP(1))/p15(INK(4)b) signaling. Cell Mol Life Sci 72 , 2949-60 (2015). Lee, B.Y. et al. Senescence-associated beta-galactosidase is lysosomal beta-galactosidase. Aging Cell 5 , 187-95 (2006). Tacutu, R. et al. Human Ageing Genomic Resources: new and updated databases. Nucleic Acids Res 46 , D1083-D1090 (2018). Saul, D. & Kosinsky, R.L. Single-Cell Transcriptomics Reveals the Expression of Aging- and Senescence-Associated Genes in Distinct Cancer Cell Populations. Cells 10 (2021). Aging Atlas: a multi-omics database for aging biology. Nucleic Acids Res 49 , D825-D830 (2021). Chatsirisupachai, K., Palmer, D., Ferreira, S. & de Magalhaes, J.P. A human tissue-specific transcriptomic analysis reveals a complex relationship between aging, cancer, and cellular senescence. Aging Cell 18 , e13041 (2019). Reyfman, P.A. et al. Single-Cell Transcriptomic Analysis of Human Lung Provides Insights into the Pathobiology of Pulmonary Fibrosis. Am J Respir Crit Care Med 199 , 1517-1536 (2019). Nehme, J., Borghesan, M., Mackedenski, S., Bird, T.G. & Demaria, M. Cellular senescence as a potential mediator of COVID-19 severity in the elderly. Aging Cell 19 , e13237 (2020). Lipskaia, L. et al. Evidence That SARS-CoV-2 Induces Lung Cell Senescence: Potential Impact on COVID-19 Lung Disease. Am J Respir Cell Mol Biol 66 , 107-111 (2022). Melms, J.C. et al. A molecular single-cell lung atlas of lethal COVID-19. Nature 595 , 114-119 (2021). Bartleson, J.M. et al. SARS-CoV-2, COVID-19 and the aging immune system. Nature Aging 1 , 769-782 (2021). Lee, S. et al. Virus-induced senescence is a driver and therapeutic target in COVID-19. Nature 599 , 283-289 (2021). Camell, C.D. et al. Senolytics reduce coronavirus-related mortality in old mice. Science 373 (2021). Li, S. et al. Cellular metabolic basis of altered immunity in the lungs of patients with COVID-19. Med Microbiol Immunol 211 , 49-69 (2022). D'Agnillo, F. et al. Lung epithelial and endothelial damage, loss of tissue repair, inhibition of fibrinolysis, and cellular senescence in fatal COVID-19. Sci Transl Med 13 , eabj7790 (2021). Parimon, T. et al. Potential mechanisms for lung fibrosis associated with COVID-19 infection. QJM 116 , 487-492 (2023). Evangelou, K. et al. Pulmonary infection by SARS-CoV-2 induces senescence accompanied by an inflammatory phenotype in severe COVID-19: possible implications for viral mutagenesis. Eur Respir J 60 (2022). Chen, J., Wu, H., Yu, Y. & Tang, N. Pulmonary alveolar regeneration in adult COVID-19 patients. Cell Res 30 , 708-710 (2020). Liao, M. et al. Single-cell landscape of bronchoalveolar immune cells in patients with COVID-19. Nat Med 26 , 842-844 (2020). Merad, M. & Martin, J.C. Pathological inflammation in patients with COVID-19: a key role for monocytes and macrophages. Nat Rev Immunol 20 , 355-362 (2020). Garcia-Nicolas, O., Godel, A., Zimmer, G. & Summerfield, A. Macrophage phagocytosis of SARS-CoV-2-infected cells mediates potent plasmacytoid dendritic cell activation. Cell Mol Immunol 20 , 835-849 (2023). Burton, D. & Stolzing, A. Cellular senescence: Immunosurveillance and future immunotherapy. Ageing Res Rev 43 , 17-25 (2018). Lo, J.A. & Fisher, D.E. The melanoma revolution: from UV carcinogenesis to a new era in therapeutics. Science 346 , 945-9 (2014). Robert, C. et al. Nivolumab in previously untreated melanoma without BRAF mutation. N Engl J Med 372 , 320-30 (2015). Hoenicke, L. & Zender, L. Immune surveillance of senescent cells--biological significance in cancer- and non-cancer pathologies. Carcinogenesis 33 , 1123-6 (2012). Kang, T.W. et al. Senescence surveillance of pre-malignant hepatocytes limits liver cancer development. Nature 479 , 547-51 (2011). Xue, W. et al. Senescence and tumour clearance is triggered by p53 restoration in murine liver carcinomas. Nature 445 , 656-60 (2007). Ruscetti, M. et al. NK cell-mediated cytotoxicity contributes to tumor control by a cytostatic drug combination. Science 362 , 1416-1422 (2018). Tirosh, I. et al. Dissecting the multicellular ecosystem of metastatic melanoma by single-cell RNA-seq. Science 352 , 189-96 (2016). Teschendorff, A.E., Breeze, C.E., Zheng, S.C. & Beck, S. A comparison of reference-based algorithms for correcting cell-type heterogeneity in Epigenome-Wide Association Studies. BMC Bioinformatics 18 , 105 (2017). Newman, A.M. et al. Robust enumeration of cell subsets from tissue expression profiles. Nat Methods 12 , 453-7 (2015). Coppe, J.P. et al. Senescence-associated secretory phenotypes reveal cell-nonautonomous functions of oncogenic RAS and the p53 tumor suppressor. PLoS Biol 6 , 2853-68 (2008). Acosta, J.C. et al. Chemokine signaling via the CXCR2 receptor reinforces senescence. Cell 133 , 1006-18 (2008). Kuilman, T. et al. Oncogene-induced senescence relayed by an interleukin-dependent inflammatory network. Cell 133 , 1019-31 (2008). Biran, A. et al. Senescent cells communicate via intercellular protein transfer. Genes Dev 29 , 791-802 (2015). Jin, S. et al. Inference and analysis of cell-cell communication using CellChat. Nat Commun 12 , 1088 (2021). Senturk, S. et al. Transforming growth factor-beta induces senescence in hepatocellular carcinoma cells and inhibits tumor growth. Hepatology 52 , 966-74 (2010). Reimann, M. et al. Tumor stroma-derived TGF-beta limits myc-driven lymphomagenesis via Suv39h1-dependent senescence. Cancer Cell 17 , 262-72 (2010). Buckley, S. et al. BMP4 signaling induces senescence and modulates the oncogenic phenotype of A549 lung adenocarcinoma cells. Am J Physiol Lung Cell Mol Physiol 286 , L81-6 (2004). Zhu, D., Wu, J., Spee, C., Ryan, S.J. & Hinton, D.R. BMP4 mediates oxidative stress-induced retinal pigment epithelial cell senescence and is overexpressed in age-related macular degeneration. J Biol Chem 284 , 9529-39 (2009). Korbecki, J. et al. in International Journal of Molecular Sciences (2020). Price, M.A. et al. CSPG4, a potential therapeutic target, facilitates malignant progression of melanoma. Pigment Cell Melanoma Res 24 , 1148-57 (2011). Gurrea-Rubio, M. & Fox, D.A. The dual role of CD6 as a therapeutic target in cancer and autoimmune disease. Front Med (Lausanne) 9 , 1026521 (2022). Zhang, C. et al. STAT3 Activation-Induced Fatty Acid Oxidation in CD8(+) T Effector Cells Is Critical for Obesity-Promoted Breast Tumor Growth. Cell Metab 31 , 148-161.e5 (2020). Malta, T.M. et al. Machine Learning Identifies Stemness Features Associated with Oncogenic Dedifferentiation. Cell 173 , 338-354.e15 (2018). Gennady, K., Vladimir, S. & Alexey, S. Fast gene set enrichment analysis. bioRxiv , 060012 (2019). Agarwal, S., Graepel, T., Herbrich, R., Har-Peled, S. & Roth, D. Generalization Bounds for the Area Under the ROC Curve. Journal of Machine Learning Research 6 , 393--425 (2005). Teschendorff, A.E. & Enver, T. Single-cell entropy for accurate estimation of differentiation potency from a cell's transcriptome. Nat Commun 8 , 15599 (2017). Yeung, K.Y., Fraley, C., Murua, A., Raftery, A.E. & Ruzzo, W.L. Model-based clustering and data transformations for gene expression data. Bioinformatics 17 , 977-87 (2001). Fang, Z., Liu, X. & Peltz, G. GSEApy: a comprehensive package for performing gene set enrichment analysis in Python. Bioinformatics 39 (2023). Wang, W. et al. Independent component analysis based gene co-expression network inference (ICAnet) to decipher functional modules for better single-cell clustering and batch integration. Nucleic Acids Res 49 , e54 (2021). Haghverdi, L., Buettner, F. & Theis, F.J. Diffusion maps for high-dimensional single-cell analysis of differentiation data. Bioinformatics 31 , 2989-98 (2015). Subramanian, A. et al. Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc Natl Acad Sci U S A 102 , 15545-50 (2005). Colaprico, A. et al. TCGAbiolinks: an R/Bioconductor package for integrative analysis of TCGA data. Nucleic Acids Res 44 , e71 (2016). Calcinotto, A. et al. Cellular Senescence: Aging, Cancer, and Injury. Physiol Rev 99 , 1047-1078 (2019). Gorgoulis, V. et al. Cellular Senescence: Defining a Path Forward. Cell 179 , 813-827 (2019). Nighat, N., Zhenqing, Y., Yidong, C., Xiaojing, W. & Siyuan, Z. Benchmarking supervised signature-scoring methods for single-cell RNA sequencing data in cancer. bioRxiv , 2021.06.29.450404 (2021). Zhang, Y., Alexander, P.B. & Wang, X.F. TGF-beta Family Signaling in the Control of Cell Proliferation and Survival. Cold Spring Harb Perspect Biol 9 (2017). Additional Declarations There is NO Competing Interest. Supplementary Files Extendeddatafigures.pdf Extended Data Fig.1 Information of training and validation data. Extended Data Fig.2 hUSI enables to distinguish senescent samples in complete and zeroing-out microarray profiles. Extended Data Fig.3 Comparison of three types of senescence qualification methods in four scRNA-seq datasets. Extended Data Fig.4 hUSI enables to uncover senescent cells accumulated in COVID-19 lung tissues. Extended Data Fig.5 hUSI enables to distinguish senescent tumor cells in melanoma. Extended Data Fig.6 Signal pathways are specific for senescent tumor cells. SupplementaryTables.xlsx Supplementary Table 1 Details of datasets used for training model, validation and evaluation. Supplementary Table 2 Details of learned-senescence features. Supplementary Table 3 AUC results of methods included in comparison. Supplementary Table 4 Eight gene sets included in comparison. Supplementary Table 5 DEGs in melanoma microarray data and specific highly expressed genes of three tumor subpopulations in melanoma scRNA-seq data. Cite Share Download PDF Status: Published Journal Publication published 29 May, 2025 Read the published version in Nature Aging → Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-3920908","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":272518496,"identity":"93717852-952f-489d-b9bf-ee1ecede53cd","order_by":0,"name":"Ting Ni","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA0UlEQVRIiWNgGAWjYDACZhBhYMPPwJAA5+IHPBAtaZINxGuBUIdJ0GLPznv4dUHBeQmD48nPHjBUWCc2sJ89QMBhfGnWMwxuSxiceWZuwHAmPbGBJy+BgBYeM2Meg9t1BjcSzCQY2w4nNkjwGBCj5ZyEwY30bxKM/4jTYvyYx+AAUEsO0JYGYrQc5jFj5jFIlpA886ZMIuFYunEbTw5+Lez9Z4w/8/yxk+A7nr5N4kONtWw/+xn8WoCATQLOTABxCakHAuYPRCgaBaNgFIyCkQwAxBQ7y8XAzI0AAAAASUVORK5CYII=","orcid":"https://orcid.org/0000-0001-7007-1072","institution":"Collaborative Innovation Center of Genetics and Development, Human Phenome Institute, School of Life Sciences, Fudan University","correspondingAuthor":true,"prefix":"","firstName":"Ting","middleName":"","lastName":"Ni","suffix":""},{"id":272518497,"identity":"de9e6a66-99a8-447e-b003-ec4b07312d00","order_by":1,"name":"Jing Wang","email":"","orcid":"","institution":"State Key Laboratory of Genetic Engineering, Collaborative Innovation Center of Genetics and Development, Human Phenome Institute, School of Life Sciences and Huashan Hospital, Fudan University","correspondingAuthor":false,"prefix":"","firstName":"Jing","middleName":"","lastName":"Wang","suffix":""},{"id":272518498,"identity":"886b6433-d5a8-4a34-96ac-2d9ca5ec546e","order_by":2,"name":"Weixu Wang","email":"","orcid":"","institution":"Institute of Computational Biology, Helmholtz Center Munich, Munich, Germany","correspondingAuthor":false,"prefix":"","firstName":"Weixu","middleName":"","lastName":"Wang","suffix":""},{"id":272518499,"identity":"cce90559-480b-4a5a-b718-44c67bac8929","order_by":3,"name":"Jun Yao","email":"","orcid":"","institution":"Department of Data System, 3D Medicines Inc, Shanghai, China","correspondingAuthor":false,"prefix":"","firstName":"Jun","middleName":"","lastName":"Yao","suffix":""},{"id":272518500,"identity":"76b1ca0c-d0ec-46e0-ab2d-193e05576ce5","order_by":4,"name":"Xiaolan Zhou","email":"","orcid":"","institution":"State Key Laboratory of Genetic Engineering, Collaborative Innovation Center of Genetics and Development, Human Phenome Institute, School of Life Sciences and Huashan Hospital, Fudan University","correspondingAuthor":false,"prefix":"","firstName":"Xiaolan","middleName":"","lastName":"Zhou","suffix":""},{"id":272518501,"identity":"bb2f8557-2c9a-4283-adfa-b3d9fea3be9e","order_by":5,"name":"Gang Wei","email":"","orcid":"","institution":"Fudan University","correspondingAuthor":false,"prefix":"","firstName":"Gang","middleName":"","lastName":"Wei","suffix":""}],"badges":[],"createdAt":"2024-02-02 12:46:42","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-3920908/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-3920908/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1038/s43587-025-00886-2","type":"published","date":"2025-05-29T04:00:00+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":51078396,"identity":"6e52dfe3-fab3-469c-b7df-504f136462ab","added_by":"auto","created_at":"2024-02-13 18:53:12","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":8191939,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eTraining and validation of hUSI.\u003c/strong\u003e \u003cstrong\u003ea.\u003c/strong\u003e Development workflow of hUSI. First, RNA-seq data sets derived from five cell types and six senescence inducing factors were included in the training datasets. Second, senescence profiles were re-processed by a unified RNA-seq data processing pipeline. Third, OCLR was chosen to learn the senescence features, generating a weighted gene set. Finally, Spearman correlation coefficient between weights and expression values of genes is defined as hUSI. \u003cstrong\u003eb. \u003c/strong\u003eSenescent samples in training set have significant higher expression level of \u003cem\u003eCDKN1A\u003c/em\u003e and \u003cem\u003eCDKN2B\u003c/em\u003ethan corresponding non-senescent samples (**** means p\u0026lt;1e-4 and *** means p\u0026lt;1e-3, \u003cem\u003et\u003c/em\u003e-test with Bonferroni correction). \u003cstrong\u003ec.\u003c/strong\u003e Senescence features learned by OCLR were significantly positively enriched in senescence associated gene sets (left panel) and negatively enriched in proliferation gene sets (right panel). Each line represents a different hallmark gene set with a variety of color shade of line depending on normalized enrichment sore (NSE). \u003cstrong\u003ed. \u003c/strong\u003eSenescent samples showed significant higher hUSIs than non-senescent (normal or proliferative) samples across seven studies, which demonstrated hUSI is robust against batch effects (*** means p\u0026lt;1e-3, one-tail \u003cem\u003et\u003c/em\u003e-test). Dotted lines represent paired senescent and control samples. \u003cstrong\u003ee. \u003c/strong\u003ehUSIs of senescent samples caused by CIS, OIS and RS in MDAH04 or WI-38 cell lines are all significantly higher than corresponding control samples (* means p\u0026lt;0.05, ** means p\u0026lt;0.01 and *** means p\u0026lt;1e-3, one-tail \u003cem\u003et\u003c/em\u003e-test).\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-3920908/v1/d89feaebdcf932cf33b41e5d.png"},{"id":51078397,"identity":"d52e84db-2aeb-4678-b330-ea248c2e89be","added_by":"auto","created_at":"2024-02-13 18:53:12","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":5733723,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003ehUSI gives reliable senescence evaluation for samples from GTEx and TCGA. a. \u003c/strong\u003ehUSIs progressively elevate with increased donor age for GTEx samples (* means p\u0026lt;0.05, ** means p\u0026lt;0.01 and *** means p\u0026lt;1e-3, one-tail \u003cem\u003et\u003c/em\u003e-test).\u003cstrong\u003e b. \u003c/strong\u003ehUSIs are significantly positively correlated with CS scores in both GTEx (left panel) and TCGA (right panel) samples. Spearman correlation coefficients (denoted as R values) were used for the evaluation.\u003cstrong\u003e c. \u003c/strong\u003ehUSIs and CS scores present high Spearman coefficients in most tissues (GTEx, upper panel) and cancer types (TCGA, lower panel).\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-3920908/v1/123ef1f26657d466dd30c429.png"},{"id":51078398,"identity":"70a12a4e-540b-4d87-add2-7f351c093ec2","added_by":"auto","created_at":"2024-02-13 18:53:12","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":10215130,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003ehUSI enable to distinguish senescent cells in various conditions. a. \u003c/strong\u003ehUSIs are significantly higher in senescent cells than growing cells across four single-cell datasets (**** means p\u0026lt;1e-4, \u003cem\u003et\u003c/em\u003e-test). \u003cstrong\u003eb. \u003c/strong\u003eAUC rank shows hUSI outperformed 12 senescence or proliferation marker genes, five methods and eight senescence-associated gene sets in evaluating senescence status. Marker gene expression value, computed senescence score and ssGSEA score are respectively used to calculate AUC on four scRNA-seq datasets. Error bars were based on the AUC ranks across four single-cell datasets. \u003cstrong\u003ec.\u003c/strong\u003ehUSI distribution of nine annotated cell types in the snRNA-seq dataset collected from COVID-19-infected lung tissues. \u003cstrong\u003ed\u003c/strong\u003e. Epithelial cells, endothelial cells, fibroblasts, myeloid, and neuronal cells from COVID-19 patients exhibited significantly higher hUSIs compared to normal donors (**** denote p\u0026lt;1e-4, Mann-Whitney tests followed by Bonferroni corrections). \u003cstrong\u003ee. \u003c/strong\u003eCells can be divided into four classes with significantly different senescence degree (C1~C4) (**** denote p\u0026lt;1e-4, Mann-Whitney tests followed by Bonferroni corrections) and the most senescent cells (C4) have apparent higher expression levels of core senescence- and SASP-related genes and lower expression level of proliferation-related genes. \u003cstrong\u003ef. \u003c/strong\u003eLung tissue from COVID-19\u003cstrong\u003e \u003c/strong\u003epatients has a higher fraction of C4 than normal sample. \u003cstrong\u003eg. \u003c/strong\u003eHigher\u003cstrong\u003e \u003c/strong\u003efraction of senescent cells (C4) (y-axis) existed\u003cstrong\u003e \u003c/strong\u003ein COVID-19\u003cstrong\u003e \u003c/strong\u003epatients with less days from symptom onset to death (x-axis). The spearman correlation coefficient (denoted as R) between days to death and C4 fraction is -0.26. \u003cstrong\u003eh. \u003c/strong\u003eTop 15 cell types which display higher fraction of senescent cells (C4)\u003cstrong\u003e \u003c/strong\u003ein COVID-19 lungs comparing to normal lungs. Top five cell types were highlighted whose fraction difference is larger than 0.2. \u003cstrong\u003ei. \u003c/strong\u003eUsing GSEA, DEGs of senescent AT1 cells (C4 vs C2) were enriched on GO terms (Biological process) and KEEG terms. Top ten terms were plotted (sorted by NES), all terms have false discovery rate (FDR) less than 0.005.\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-3920908/v1/ea7a1a89b60d02131de7b9c0.png"},{"id":51078403,"identity":"e4477f0d-10f0-4898-878e-63222cff2d03","added_by":"auto","created_at":"2024-02-13 18:53:12","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":10679930,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eIdentifying senescent tumor cells to benefit patient survival. a. \u003c/strong\u003eValidating the senescence degrees of \u003cem\u003ecycling\u003c/em\u003e, \u003cem\u003etransitional\u003c/em\u003e and \u003cem\u003esenescent\u003c/em\u003e subpopulations by overlap comparison with DEGs (up or down regulated) in bulk samples (senescence vs growing). Hypergeometric distribution test was used to calculate the p values. \u003cstrong\u003eb. \u003c/strong\u003eDiffusion map showing a trajectory from \u003cem\u003ecycling\u003c/em\u003e cells to \u003cem\u003esenescent\u003c/em\u003e cells in melanoma. \u003cstrong\u003ec. \u003c/strong\u003eTwo classical aging-related marker genes (\u003cem\u003eCDKN1A\u003c/em\u003e and \u003cem\u003eSERPINE\u003c/em\u003e) showed increased expression level along trajectory and in senescent subpopulation. \u003cstrong\u003ed. \u003c/strong\u003eDifferent GO terms (biological process) characterize \u003cem\u003ecycling\u003c/em\u003e, \u003cem\u003etransitional \u003c/em\u003eand \u003cem\u003esenescent\u003c/em\u003e subpopulations, with \u003cem\u003ereplication-associated\u003c/em\u003e terms enriched in \u003cem\u003ecycling\u003c/em\u003esubpopulation while \u003cem\u003eimmunity activation\u003c/em\u003e-related terms enriched in \u003cem\u003esenescent\u003c/em\u003esubpopulation. \u003cstrong\u003ee. \u003c/strong\u003eHeatmap of Spearman correlation coefficient between the three subpopulations and the abundance of 22 immune cell types indicates that senescent tumor cells were associated with immunity activation. \u003cstrong\u003ef. \u003c/strong\u003eSurvival curves of melanoma patients with different proportion of \u003cem\u003ecycling\u003c/em\u003e and \u003cem\u003esenescent\u003c/em\u003ecell subpopulations.\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-3920908/v1/df714c144ebe62aa9cd18258.png"},{"id":51078400,"identity":"131dd999-c16a-439d-88bb-8e97d5b3d3d7","added_by":"auto","created_at":"2024-02-13 18:53:12","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":259897,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eTNF-β and BMP signal pathways are specific for senescent tumor cells. a. \u003c/strong\u003eThree subpopulations of tumor cells and other six cell types presented in tumor microenvironment (T cell, B cell, natural killer (NK) cell, macrophage (Macro) cell, endothelial (Endo) cell and CAFs) are taken to infer cell-cell communications in tumor microenvironment. \u003cstrong\u003eb. \u003c/strong\u003eThe \u003cem\u003eSenescent\u003c/em\u003e subpopulation exhibits higher interaction strength than other two subpopulations (\u003cem\u003ecycling\u003c/em\u003e and \u003cem\u003etransitional\u003c/em\u003e). X-axis and y-axis represent log\u003csub\u003e10\u003c/sub\u003e-transformed outgoing and incoming interaction strength, respectively. \u003cstrong\u003ec. \u003c/strong\u003eTen ligand-receptor pairs showing specifically high communication probability in \u003cem\u003esenescent\u003c/em\u003e subpopulation, with two pairs belonging to TNF-β signaling pathway and four pairs belonging to BMP signal pathway. \u003cstrong\u003ed. \u003c/strong\u003e\u003cem\u003eSenescent\u003c/em\u003e subpopulation receives TGF-β and BMP signals mainly from CAFs and T cell in cell-cell communication networks. \u003cstrong\u003ee. \u003c/strong\u003e\u003cem\u003eSenescent\u003c/em\u003e subpopulation shows higher expression level of gene encoding receptors involved in TGF-β (\u003cem\u003eTGFBR1\u003c/em\u003e and \u003cem\u003eTGFBR1\u003c/em\u003e) and BMP (\u003cem\u003eBMPR1B \u003c/em\u003eand\u003cem\u003e BMPR2\u003c/em\u003e) signaling pathways. \u003cstrong\u003ef. \u003c/strong\u003eMelanoma patients with\u003cstrong\u003e \u003c/strong\u003ehigh expression level of \u003cem\u003eBMPR2\u003c/em\u003e, \u003cem\u003eTGFBR1\u003c/em\u003e or \u003cem\u003eTGFBR1\u003c/em\u003e have a significant better survival prognosis.\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-3920908/v1/9601a85fc0029cf0d180017a.png"},{"id":51078402,"identity":"e526fe78-6d7e-4a2c-8b02-abb5f0bbd661","added_by":"auto","created_at":"2024-02-13 18:53:12","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":7188447,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eExtended Data Fig.1\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eInformation of training and validation data.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eExtended Data Fig.2\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003ehUSI enables to distinguish senescent samples in complete and zeroing-out microarray profiles.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eExtended Data Fig.3\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eComparison of three types of senescence qualification methods in four scRNA-seq datasets.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eExtended Data Fig.4\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003ehUSI enables to uncover senescent cells accumulated in COVID-19 lung tissues.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eExtended Data Fig.5\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003ehUSI enables to distinguish senescent tumor cells in melanoma.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eExtended Data Fig.6\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eSignal pathways are specific for senescent tumor cells.\u003c/p\u003e","description":"","filename":"Extendeddatafigures.pdf","url":"https://assets-eu.researchsquare.com/files/rs-3920908/v1/8bf516afeac60688d5852c38.pdf"},{"id":51078399,"identity":"a099cee0-9471-4231-927f-e5c30b18edb1","added_by":"auto","created_at":"2024-02-13 18:53:12","extension":"xlsx","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":1244797,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSupplementary Table 1\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eDetails of datasets used for training model, validation and evaluation.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSupplementary Table 2\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eDetails of learned-senescence features.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSupplementary Table 3\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAUC results of methods included in comparison.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSupplementary Table 4\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eEight gene sets included in comparison.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSupplementary Table 5\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eDEGs in melanoma microarray data and specific highly expressed genes of three tumor subpopulations in melanoma scRNA-seq data.\u003c/p\u003e","description":"","filename":"SupplementaryTables.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-3920908/v1/5a30f6cc8211f79a4c7b44b5.xlsx"}],"financialInterests":"There is \u003cb\u003eNO\u003c/b\u003e Competing Interest.","formattedTitle":"Robust senescence evaluation by transcriptome-based hUSI to facilitate characterizing cellular senescence under various conditions","fulltext":[{"header":"1 Introduction","content":"\u003cp\u003eCellular senescence (CS) characterized by irreversible cell cycle arrest is considered a critical factor for aging and aging-related diseases\u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u003c/sup\u003e. For instance, by presenting senescence-associated secretory phenotypes (SASP) including increased secretion of pro-inflammatory proteins and other paracrine factors (such as TGF-β family ligands, VEGF, CCL2 and CCL20)\u003csup\u003e\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e, senescent cells can stimulate immune response and cell-cell communication leading pleiotropic effects in various tissues\u003csup\u003e\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u003c/sup\u003e. Targeted clearance of accumulated senescent cells using senolytic drugs\u003csup\u003e\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u003c/sup\u003e or inducing tumor cells into senescence\u003csup\u003e\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u003c/sup\u003e have shown benefits for disease prognosis and healthy lifespan. However, despite several morphological (such as flattening and enlarging\u003csup\u003e\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u003c/sup\u003e) and molecular markers (such as p16\u003csup\u003e7\u003c/sup\u003e, p21\u003csup\u003e8\u003c/sup\u003e, p15\u003csup\u003e9\u003c/sup\u003e and p27\u003csup\u003e9\u003c/sup\u003e) are used to characterize senescent cells, identification of \u003cem\u003ein vivo\u003c/em\u003e senescent cells still pose a great challenge due to its heterogeneity\u003csup\u003e\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e. Depending on the real situations, CS in diverse cell types can be induced by various intrinsic and extrinsic stressors, such as replicative stress, oncogene activation, chemotherapeutic drugs and ionizing radiation\u003csup\u003e\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u003c/sup\u003e. Therefore, to properly quantify CS degree in multiple scenarios, there is an urgent need for a universal method which enables sensitive capture of comprehensive senescence features, especially in the era where single-cell transcriptome technology has been widely applied to construct cell atlas of human multiple tissues\u003csup\u003e\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eQuantifying senescence degree by senescence score is regarded as a convenient and efficient way to monitor senescence status and disease progression\u003csup\u003e\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u003c/sup\u003e. Considering there is no one-size-fits-all marker gene to exclusively indicate senescence, most attempts to evaluate senescent samples mainly depend on the expression level of aging or senescence associated genes derived from differential analysis and literature studies, giving rise to several aging or senescence-associated gene sets (such as CellAge\u003csup\u003e\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u003c/sup\u003e and SenMayo\u003csup\u003e\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e) and senescence scoring methods (such as DAS\u0026thinsp;+\u0026thinsp;MSS\u003csup\u003e14\u003c/sup\u003e, CS score\u003csup\u003e\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e\u003c/sup\u003e and lassoCS\u003csup\u003e\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u003c/sup\u003e). However, due to the variation in gene composition and the limited study dataset (only focus on particular sample type or senescence type), these methods cannot reliably evaluate transcriptional signatures of senescent samples in various contexts and are susceptible to the absence of some pre-defined senescence-related genes. For example, only involving replicative senescence associated genes as the basis for scoring senescence status might produces bias in evaluating senescence in real aging tissues or induced senescence samples. Thus, we sought to utilize publicly available high-quality transcriptome profiles of senescent samples to learn the comprehensive senescence features and develop a reliable and universal senescence score for senescence evaluation.\u003c/p\u003e \u003cp\u003eIn the present study, to evaluate cellular senescence in a more general and unbiased way, we introduce hUSI that can accurately assess the burden of senescence at both bulk and single-cell levels. It started from collecting representative senescence transcriptome profiles encompassing multiple contexts, including those derived from different platforms, cell types, conditions, and senescence-induction factors. With the criteria of confirmed senescence status and involving diverse cell and senescence types, we finally collected bulk RNA-seq profiles from five cell types and six senescence types induced by both intrinsic and extrinsic stressors. Then, hUSI was developed based on features extracted from these representative senescence transcriptomes by a machine learning model. hUSI demonstrated high accuracy in distinguishing senescent samples from non-senescent samples in different context. Furthermore, hUSI outperformed other current methods in evaluating senescence at single-cell level and remained robust and reliable in multiple senescence samples. Intriguingly, hUSI can uncover senescent cell subpopulation, as illustrated in melanoma, that correlated with improved patient survival, indicating its promising potential in clinical situations. Notably, hUSI distinguished typical signaling pathways (such as TGF-β and BMP pathways) that could promote senescence-associated cell-cell communication in tumor microenvironment. Overall, hUSI provides a universal and robust way to measure senescence burden, enabling more comprehensive investigations into senescence in various experimental and clinical context.\u003c/p\u003e"},{"header":"2 Results","content":"\u003cp\u003e \u003cb\u003eDevelopment and validation of hUSI\u003c/b\u003e \u003c/p\u003e \u003cp\u003eTo systematically learn and evaluate the comprehensive signature of CS, we developed a workflow including data collection, data re-processing, feature extraction and quantification of senescence degree (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003ea). We first collected RNA-seq data sets derived from representative human senescence types serving as senescence training samples, which encompassed five cell types (fibroblasts, endothelial cells, astrocytes, melanocytes, and keratinocytes) and six senescence types (ionizing radiation-induced senescence (IRIS), replicative senescence (RS), oxidative stress-induced senescence (OSIS), oncogene-induced senescence (OIS), natural senescence (NS) as well as compound-induced senescence (CIS)), along with the corresponding non-senescent samples serving as young controls\u003csup\u003e\u003cspan additionalcitationids=\"CR20 CR21 CR22 CR23 CR24 CR25 CR26\" citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e\u003c/sup\u003e (Extended Data Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003ea,b and Supplementary table \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e). Next, considering these data sets were derived from different experimental methods and sequencing protocols, we re-preprocessed all the raw data with the same pipeline to generate standard and normalized profiles for feature extraction and validation (Extended Data Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eb and Methods). After processing, we found that, as expected, \u003cem\u003eCDKN1A\u003c/em\u003e and \u003cem\u003eCDKN2B\u003c/em\u003e (encode the well-known senescence marker p21 and p15, respectively) showed significant higher expression level, demonstrating reliability of samples and the overall analysis pipeline (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eb).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThen, we went on to acquire features of the senescence profiles by machine learning algorithms. Multiple machine learning algorithms have been employed for mining genes associated with individual aging or CS, such as regression, elastic net, and random forests\u003csup\u003e\u003cspan additionalcitationids=\"CR29\" citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e\u003c/sup\u003e, facilitating quantification of senescence degree in tumor and normal cells. However, senescence, as a complicated and continuous state, its heterogeneity has not been fully considered in these models\u003csup\u003e\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e\u003c/sup\u003e. In this study, we selected one-class logistic regression (OCLR) algorithm to learn the features of senescence transcriptome profiles, as it has been demonstrated superior performance on capturing cell heterogeneity\u003csup\u003e\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e\u003c/sup\u003e. After training, OCLR learned the features of senescence samples in training set, in other words, all genes were respectively assigned with different weights representing their contributions to senescence (Methods). In our learned-senescence features, except for genes upregulated in CS and associated with SASP (such as \u003cem\u003eAPOD\u003c/em\u003e, \u003cem\u003eEHF\u003c/em\u003e and \u003cem\u003eSAA2\u003c/em\u003e)\u003csup\u003e\u003cspan additionalcitationids=\"CR34\" citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e\u003c/sup\u003e were assigned with top weights, some genes with high positive or negative weights while their functions in CS were poorly reported (such as \u003cem\u003eOLAH\u003c/em\u003e, \u003cem\u003eCADM3\u003c/em\u003e and \u003cem\u003eHMSD\u003c/em\u003e) (Extended Data Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003ec). These results suggest that OCLR not only captures the known senescence features but also identifies potential novel senescence-associated genes. To our surprise, classical senescence marker, including \u003cem\u003eCDKN1A\u003c/em\u003e, \u003cem\u003eCDKN2B\u003c/em\u003e and \u003cem\u003eSERPINE1\u003c/em\u003e, are only assigned with slightly positive weights, probably because of relatively low expression levels in samples (Supplementary table \u003cspan refid=\"MOESM2\" class=\"InternalRef\"\u003eS2\u003c/span\u003e). To further examine the biological interpretability of learned-senescence features, gene set enrichment analysis (GSEA)\u003csup\u003e\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e\u003c/sup\u003e was performed based on the weight of each gene (Methods). We found that multiple senescence associated gene sets were positively enriched, including interferon-gamma response\u003csup\u003e\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e\u003c/sup\u003e, KRAS signaling\u003csup\u003e\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e\u003c/sup\u003e, inflammatory response\u003csup\u003e\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e\u003c/sup\u003e, hypoxia and p53 pathway\u003csup\u003e\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e, \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e\u003c/sup\u003e (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003ec, left panel). On the contrary, the proliferation associated pathways (such as G2M checkpoint\u003csup\u003e\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e\u003c/sup\u003e, E2F targets\u003csup\u003e\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e\u003c/sup\u003e, mitotic spindles\u003csup\u003e\u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e\u003c/sup\u003e and MYC targets\u003csup\u003e\u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e45\u003c/span\u003e\u003c/sup\u003e) were in the negative enrichment terms (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003ec, right panel). These results supported the reliability of the features learned by OCLR in reflecting senescence.\u003c/p\u003e \u003cp\u003eIn terms of quantifying senescence degree, the Spearman correlation coefficient between gene weights and expression values was selected as the metric to quantify senescence degree, defined as human universal senescence index (hUSI) (Methods). To test the stability and reliability of hUSI, we used leave-one-out cross-validation (LOOCV) strategy to calculate average correctly rank probability (CRP) for each iteration, and the resulted CRP reached 0.9 (Extended Data Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eb, d and Methods). We next validated whether hUSI can be influenced by batch effects arising from variations in experimental conditions, sequencing platforms, or analysis pipelines. We compiled bulk RNA-seq datasets from seven independent studies, each comprising oncogene-induced senescent IMR90 cells induced by 4-hydroxytamoxifen (4-OHT), along with corresponding control cells\u003csup\u003e\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e, \u003cspan additionalcitationids=\"CR47 CR48\" citationid=\"CR46\" class=\"CitationRef\"\u003e46\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e49\u003c/span\u003e\u003c/sup\u003e (Supplementary table \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e). We calculated hUSI for each sample and found that all senescent groups have much higher hUSI values (hUSIs) than the non-senescent groups (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003ed). Besides, with a plethora of genes included in the senescence features for calculation, hUSI is technically more robust to profiles with limited or sparse gene signals such as microarray and single-cell RNA-seq (scRNA-seq) data. Therefore, we applied hUSI on six senescence-related microarray datasets and the most of them were derived from cell types which were not included in training set\u003csup\u003e\u003cspan additionalcitationids=\"CR51 CR52 CR53 CR54\" citationid=\"CR50\" class=\"CitationRef\"\u003e50\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR55\" class=\"CitationRef\"\u003e55\u003c/span\u003e\u003c/sup\u003e. The results showed that hUSIs were consistently higher in all senescent groups compared to non-senescent ones (Extended Data Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ea). To test the robustness of hUSI, we generated simulated sparse profiles from these microarray transcriptome profiles by randomly zeroing-out expression signals. We found that even zeroing-out 50% of genes expression signals, hUSIs still represented higher levels in all senescent groups (Extended Data Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eb). All these results suggested that hUSI, based on comprehensive senescence features and effective nonparametric rank-based correlation\u003csup\u003e\u003cspan citationid=\"CR56\" class=\"CitationRef\"\u003e56\u003c/span\u003e\u003c/sup\u003e, is pretty stable and robust.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cb\u003ehUSI shows reliable performance in quantifying senescence degree\u003c/b\u003e \u003c/p\u003e \u003cp\u003eTo assess the generalizability of hUSI, we gathered three bulk RNA-seq datasets (including immortal MDAH04 cells and senescent MDAH04 cells induced by different chemical compounds\u003csup\u003e\u003cspan citationid=\"CR57\" class=\"CitationRef\"\u003e57\u003c/span\u003e\u003c/sup\u003e, WI-38 cells treated with 4-OHT for different days\u003csup\u003e\u003cspan citationid=\"CR58\" class=\"CitationRef\"\u003e58\u003c/span\u003e\u003c/sup\u003e and proliferative WI-38 cells and senescent WI-38 cells induced by replication\u003csup\u003e\u003cspan citationid=\"CR58\" class=\"CitationRef\"\u003e58\u003c/span\u003e\u003c/sup\u003e), notably the conditions of these samples are not exactly same as samples in the training set (Supplementary table \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e). We found that most senescent groups exhibited significant higher hUSIs compared to non-senescent ones even the sample size is limited (Extended Data Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ea). Moreover, hUSI also demonstrated its ability to discern aggravated senescence in samples induced by extended 4-OHT exposure time (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003ee, middle panel).\u003c/p\u003e \u003cp\u003eNext, we calculated hUSIs for a large normal samples dataset obtained from the Genotype-Tissue Expression Project (GTEx). We observed that hUSIs progressively and significantly elevated with increasing age, consisting with the continuous accumulation of senescent cells in aging process\u003csup\u003e\u003cspan citationid=\"CR59\" class=\"CitationRef\"\u003e59\u003c/span\u003e\u003c/sup\u003e (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ea). To validate the reliability of hUSI in assessing senescence degree, we calculated Spearman correlation coefficient between hUSIs and CS scores, which was a tool based on conducting gene set variation analysis (GSVA) on a curated set of 1,259 genes derived from studies on replicative cell senescence\u003csup\u003e\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e\u003c/sup\u003e. The results showed overall positive correlations of these two methods (R\u0026thinsp;=\u0026thinsp;0.7), and across 29/30 tissues (R from 0.28 to 0.85) (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eb left panel and Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ec upper panel). The same strategy was applied on a large tumor samples dataset from The Cancer Genome Atlas (TCGA) datasets. Despite the heterogeneity in tumor samples, hUSIs still showed overall positive correlations with CS scores (R\u0026thinsp;=\u0026thinsp;0.52) and across all cancer types (R from 0.12 to 0.94) (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eb right panel and Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ec lower panel). Of note, we discovered that hUSIs demonstrated higher variations in different cancers compared to CS scores, which might indicate that hUSI enables to reveal more intrinsic heterogeneity of senescence across different tumor types (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ec lower panel).\u003c/p\u003e \u003cp\u003e \u003cb\u003ehUSI has better performance in distinguishing senescence cells\u003c/b\u003e \u003c/p\u003e \u003cp\u003eGiven the reliable and robust performance of hUSI on scoring bulk samples under various conditions, we next applied hUSI on four scRNA-seq datasets derived from primary senescent cells induced by various stressors (including oncogene\u003csup\u003e\u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e49\u003c/span\u003e, \u003cspan citationid=\"CR60\" class=\"CitationRef\"\u003e60\u003c/span\u003e\u003c/sup\u003e, radiation\u003csup\u003e\u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e61\u003c/span\u003e\u003c/sup\u003e, and replication\u003csup\u003e\u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e61\u003c/span\u003e\u003c/sup\u003e, as well as secondary senescent cells triggered by paracrine signals\u003csup\u003e\u003cspan citationid=\"CR62\" class=\"CitationRef\"\u003e62\u003c/span\u003e\u003c/sup\u003e) to assess the robustness of hUSI at single-cell level across diverse conditions. The senescence status of these cells had been confirmed in respective studies by examining senescence marker genes and senescence-associated β-galactosidase (SA-β-Gal) staining\u003csup\u003e\u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e49\u003c/span\u003e, \u003cspan additionalcitationids=\"CR61\" citationid=\"CR60\" class=\"CitationRef\"\u003e60\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR62\" class=\"CitationRef\"\u003e62\u003c/span\u003e\u003c/sup\u003e (Supplementary table \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e). Non-senescent cells from each dataset were also included for comparative analysis. After quantifying the senescence degree of each cell using hUSI, we observed significantly higher hUSI levels in senescent groups than non-senescent groups across all four datasets, supporting the applicability of hUSI on scRNA-seq data (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003ea). Next, we compared the performance of hUSI with other three groups of senescence qualification strategies (including those based on gene expression level, computed score and enrichment score of single sample GSEA (ssGSEA)) (Methods).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eFirst, we first obtained 12 well-known CS or proliferation associated markers (\u003cem\u003eGLB1\u003c/em\u003e, \u003cem\u003eTP53\u003c/em\u003e, \u003cem\u003eCDKN1A\u003c/em\u003e, \u003cem\u003eCDKN2A\u003c/em\u003e, \u003cem\u003eCDKN2B\u003c/em\u003e, \u003cem\u003eCDK1\u003c/em\u003e, \u003cem\u003eCDK4\u003c/em\u003e, \u003cem\u003eCDK6\u003c/em\u003e, \u003cem\u003eMKI67\u003c/em\u003e, \u003cem\u003eLMNB1\u003c/em\u003e, \u003cem\u003eIL1A\u003c/em\u003e, and \u003cem\u003eRB1\u003c/em\u003e) and separately used their normalized expression values to directly classify cells, as their upregulation or downregulation is widely employed to identify CS status\u003csup\u003e\u003cspan additionalcitationids=\"CR64 CR65 CR66 CR67 CR68 CR69 CR70\" citationid=\"CR63\" class=\"CitationRef\"\u003e63\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR71\" class=\"CitationRef\"\u003e71\u003c/span\u003e\u003c/sup\u003e. We found that only \u003cem\u003eCDKN1A\u003c/em\u003e exhibited a higher trend in senescent samples than control samples across all datasets (Extended Data Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003ea, left panel). To better compare the performance of hUSI and the markers in classifying senescent cells in limited scRNA-seq datasets, we randomly split each dataset into 10 folds and replicated the process three times, and then calculated the ranks of average Area Under Curve (AUC) of all units in each dataset (Supplementary table S3 and Methods). We observed that hUSI exhibited excellent performance compared to all the tested classical markers (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eb, left panel and Supplementary table S3).\u003c/p\u003e \u003cp\u003eSecond, we compared hUSI with five existing senescence score computing methods, including DAS, mSS and their combination (DAS\u0026thinsp;+\u0026thinsp;mSS)\u003csup\u003e14\u003c/sup\u003e, lassoCS\u003csup\u003e\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u003c/sup\u003e and CSS\u003csup\u003e\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e\u003c/sup\u003e. To our surprise, these methods only gave senescent group a higher score level than control group in certain datasets (Extended Data Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003ea, middle panel). We then applied the same strategy above to calculate average AUC ranks. hUSI also achieved the highest average AUC rank compared to all computed senescence scores (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eb, middle panel). Additionally, we observed that DAS\u0026thinsp;+\u0026thinsp;mSS, as expected, outperformed both DAS and mSS individually (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eb, middle panel). Of note, except for hUSI, all these methods exhibited substantial variations across four datasets (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eb, middle panel), supporting the more stable performance of hUSI.\u003c/p\u003e \u003cp\u003eFinally, considering aging and senescence-associated gene sets have been commonly used to quantify CS by enrichment score using ssGSEA, in the present study, we collected eight publicly available senescence-associated gene sets (including CellAge\u003csup\u003e\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u003c/sup\u003e, GenAge\u003csup\u003e\u003cspan citationid=\"CR72\" class=\"CitationRef\"\u003e72\u003c/span\u003e\u003c/sup\u003e, ASIG\u003csup\u003e\u003cspan citationid=\"CR73\" class=\"CitationRef\"\u003e73\u003c/span\u003e\u003c/sup\u003e, SASP (downloaded from MSigDB under acessesion ID R-HSA-2559582), AgingAtlas\u003csup\u003e\u003cspan citationid=\"CR74\" class=\"CitationRef\"\u003e74\u003c/span\u003e\u003c/sup\u003e, SenUp\u003csup\u003e\u003cspan citationid=\"CR75\" class=\"CitationRef\"\u003e75\u003c/span\u003e\u003c/sup\u003e,SenMayo\u003csup\u003e\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e, and SigRS\u003csup\u003e\u003cspan citationid=\"CR76\" class=\"CitationRef\"\u003e76\u003c/span\u003e\u003c/sup\u003e) to calculated their ssGSEA scores in four scRNA-seq datasets (Supplementary table S4). The result showed that only SenUp gave higher scores for senescent groups than control groups across four scRNA-seq datasets (Extended Data Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003ea, right panel). After calculating average AUC ranks, hUSI still exhibited superior performance over all gene sets, with minimal variation observed across the four single-cell datasets (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eb, right panel). Furthermore, we found that genes from all these gene sets can be found in our features, and genes had been assigned with different weights which enable hUSI to capture a broader spectrum of gene expression signals in the senescence evaluation process (Supplementary table \u003cspan refid=\"MOESM2\" class=\"InternalRef\"\u003eS2\u003c/span\u003e,4). These results above combined to suggest that hUSI has relative superiority and stability across different scRNA-seq datasets comparing to other current methods.\u003c/p\u003e \u003cp\u003e \u003cb\u003ehUSI enables to evaluate senescence burden in complex conditions\u003c/b\u003e \u003c/p\u003e \u003cp\u003eAfter validating the outperformance of hUSI in distinguishing senescent cells, we next sought to apply hUSI on single-cell data from real pathological tissues. The accumulation of senescent cells has been reported to increase the susceptibility to COVID-19 patients by contributing to SARS-CoV-2-mediated hyperinflammation and cytokine storm\u003csup\u003e\u003cspan citationid=\"CR77\" class=\"CitationRef\"\u003e77\u003c/span\u003e, \u003cspan citationid=\"CR78\" class=\"CitationRef\"\u003e78\u003c/span\u003e\u003c/sup\u003e. Consequently, the targeted elimination of these senescent cells has been proposed as a potential treatment strategy for COVID-19\u003csup\u003e77, 78\u003c/sup\u003e. However, the deconvolution of senescent status across various cell types in infected lung tissues and the study of detrimental effects of different senescent cells on patient survival are still lack. Thus, to evaluate the senescence burden of COVID-19 patients, we calculated the hUSIs for a single-nuclei RNA-seq (snRNA-seq) dataset (containing a total of 116,313 nuclei) derived from infected and normal lungs with donor age ranging from 58\u0026thinsp;~\u0026thinsp;84 years old\u003csup\u003e\u003cspan citationid=\"CR79\" class=\"CitationRef\"\u003e79\u003c/span\u003e\u003c/sup\u003e (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003ec and Methods). We found that most cell types (including epithelial cells, endothelial cells, fibroblasts, myeloid, and neuronal cells) from COVID-19 patients exhibited significantly higher hUSI values compared to those from normal donors (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003ed). Intriguingly, a reverse trend was observed in B cells and T cells, suggesting the activation of immune cells following COVID-19 infection\u003csup\u003e\u003cspan citationid=\"CR80\" class=\"CitationRef\"\u003e80\u003c/span\u003e\u003c/sup\u003e (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003ed).\u003c/p\u003e \u003cp\u003eTo better discern various senescence status, we applied a gaussian mixture model (GMM) to fit the distribution of hUSIs within all tested cells and successfully classified them into four distinct classes (C1\u0026thinsp;~\u0026thinsp;C4) (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003ee, left panel and Methods). Their senescence degrees were further validated by the higher expression level of classical senescence-associated genes (\u003cem\u003eCDKN1A\u003c/em\u003e, \u003cem\u003eIL1A\u003c/em\u003e, \u003cem\u003eIL6\u003c/em\u003e, \u003cem\u003eIL8\u003c/em\u003e, \u003cem\u003eCCL2\u003c/em\u003e, \u003cem\u003eCXCL10\u003c/em\u003e, \u003cem\u003eMMP9\u003c/em\u003e, \u003cem\u003eSERPINE1\u003c/em\u003e, \u003cem\u003eTHBS1\u003c/em\u003e and \u003cem\u003eTIMP1\u003c/em\u003e) and lower expression level of proliferation markers (\u003cem\u003eLMNB1\u003c/em\u003e, \u003cem\u003eMKI67\u003c/em\u003e and \u003cem\u003eDHFR\u003c/em\u003e), consisting with the reported elevated cell senescence responses to SARS-Cov-2 infection\u003csup\u003e\u003cspan citationid=\"CR81\" class=\"CitationRef\"\u003e81\u003c/span\u003e\u003c/sup\u003e (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003ee, right panel). We also observed that COVID-19 lung tissue has a higher proportion of the most senescent cell class (denoted as C4) cells compared to normal tissue (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003ef), consistent with the reports suggesting a high accumulation of senescent cells in COVID-19 patients\u003csup\u003e\u003cspan citationid=\"CR77\" class=\"CitationRef\"\u003e77\u003c/span\u003e, \u003cspan citationid=\"CR82\" class=\"CitationRef\"\u003e82\u003c/span\u003e\u003c/sup\u003e. Besides, patients with faster disease progression showed more accumulation of senescent cells (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eg). These results all suggested that hUSI successfully revealed survival-detrimental senescent cells accumulated in COVID-19 lung tissue across various cell types。\u003c/p\u003e \u003cp\u003eWe then examined the difference in the fraction of four cell groups for each cell type between COVID-19 and the normal samples. The results showed there are higher fractions of senescent cells existed in the cell types with a higher risk of exposure to SARS-CoV-2 or hyperinflammatory microenvironment, such as monocyte-derived macrophages, inflamed endothelial cells, pathological fibroblasts and alveolar type 1 progenitor cells (AT1) (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eh and Extended Data Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ea, b)\u003csup\u003e\u003cspan citationid=\"CR78\" class=\"CitationRef\"\u003e78\u003c/span\u003e, \u003cspan additionalcitationids=\"CR84\" citationid=\"CR83\" class=\"CitationRef\"\u003e83\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR85\" class=\"CitationRef\"\u003e85\u003c/span\u003e\u003c/sup\u003e. While alveolar type 2 progenitor cells (AT2), which are targeted by SARS-CoV-2 through the angiotensin-converting enzyme 2 (ACE2), was reported to exhibit apparent senescence and a proinflammatory phenotypes\u003csup\u003e\u003cspan citationid=\"CR86\" class=\"CitationRef\"\u003e86\u003c/span\u003e\u003c/sup\u003e, AT1 accumulated in a higher proportion in COVID-19 lung tissue than AT2, possibly because AT2 can differentiate into AT1-like cells for alveolar regeneration in COVID-19 patients\u003csup\u003e\u003cspan citationid=\"CR87\" class=\"CitationRef\"\u003e87\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eTo investigate the alterations in senescent cells, we performed differential gene expression analysis between C4 and C2 (which is the second young class (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003ef)). We did not take C1 class as the control due to its very small cell numbers, which usually lead to some bias in differential analysis. Differential genes (DEGs) of AT1 and AT2 were respectively enriched on KEGG and GO databases. The results showed that senescent AT1 and AT2 cells have enriched on pathways associated with antigen process, extracellular matrix and immune cytotoxicity, especially AT1 has enriched on p53 signaling pathway, indicating a higher relevance of these senescent on infection response, cellular communication and cellular senescence (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003ei and Extended Data Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ec). In addition, DEGs were also enriched in senescent monocyte-derived macrophages as it showed largest fraction difference in C4, reaching 0.29, and was reported to drive the inflammatory response to SARS-CoV-2 and contribute to cytokine storms in severe COVID-19\u003csup\u003e88, 89\u003c/sup\u003e. The results showed that pathways, including positive regulation of T cell-mediated immunity and leukocyte-mediated cytotoxicity, was enriched in these cells, indicating their crucial roles of senescent cells in macrophage-mediated clearance of infected cells, which may also cause damages to infected tissues by hyperinflammatory\u003csup\u003e\u003cspan citationid=\"CR90\" class=\"CitationRef\"\u003e90\u003c/span\u003e\u003c/sup\u003e (Extended Data Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ed). All these results above demonstrated that hUSI enables to recognize senescent cells that abnormally accumulated in pathological tissue and reveal associated mechanisms.\u003c/p\u003e \u003cp\u003e \u003cb\u003ehUSI identifies immune associated senescent tumor cells in melanoma\u003c/b\u003e \u003c/p\u003e \u003cp\u003eWe then sought to apply it on tumor samples, as CS plays an important role in tumor development and can activate immune responses\u003csup\u003e\u003cspan citationid=\"CR91\" class=\"CitationRef\"\u003e91\u003c/span\u003e\u003c/sup\u003e. Significant progress has been made in immunotherapy of melanoma, especially with the application of immune checkpoint inhibitors, such as PD-1 antibodies and CTLA-4 antibodies, which result in significant durable responses and therapeutic efficacy\u003csup\u003e\u003cspan citationid=\"CR92\" class=\"CitationRef\"\u003e92\u003c/span\u003e, \u003cspan citationid=\"CR93\" class=\"CitationRef\"\u003e93\u003c/span\u003e\u003c/sup\u003e. However, the mechanisms underlying immunotherapy remain incompletely understood. Several studies have demonstrated the relationship between senescent tumor cells and immune recognition\u003csup\u003e\u003cspan additionalcitationids=\"CR95 CR96\" citationid=\"CR94\" class=\"CitationRef\"\u003e94\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR97\" class=\"CitationRef\"\u003e97\u003c/span\u003e\u003c/sup\u003e, thus we sought to identify senescent tumor cells and investigate whether it could serve as potential targets for immunotherapy in melanoma.\u003c/p\u003e \u003cp\u003eTo explore the characteristics of senescent tumor cells in melanoma, we evaluated the senescence degree of tumor cells by applying hUSI on a melanoma scRNA-seq data set\u003csup\u003e\u003cspan citationid=\"CR98\" class=\"CitationRef\"\u003e98\u003c/span\u003e\u003c/sup\u003e. We then used GMM to infer three cell subpopulations which were denoted as \u003cem\u003ecycling\u003c/em\u003e, \u003cem\u003etransactional\u003c/em\u003e and \u003cem\u003esenescent\u003c/em\u003e, based on the significantly increasing hUSI level (Extended Data Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003ea, b). The senescence degree of these subpopulations was further validated by a microarray-based transcriptome dataset of melanocytes\u003csup\u003e\u003cspan citationid=\"CR52\" class=\"CitationRef\"\u003e52\u003c/span\u003e\u003c/sup\u003e. By overlapping DEGs of melanocytes bulk samples with specific highly expressing genes in our defined cell subpopulations (Methods), we found that genes up-regulated in senescent melanocytes were significantly enriched in \u003cem\u003esenescent\u003c/em\u003e and \u003cem\u003etransitional\u003c/em\u003e subpopulations (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ea and Supplementary table S5). On the contrary, genes up-regulated in growing melanocytes were significantly enriched only in \u003cem\u003ecycling\u003c/em\u003e subpopulation (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ea and Supplementary table S5). We also validated the different senescence degree of these three subpopulations by inferring a senescence trajectory of tumor cells. We imputed 38 co-expression modules based on hUSI-related genes and diffusion map was used for dimensionality reduction and visualization. The senescence trajectory was characterized by the transition of tumor cells from \u003cem\u003ecycling\u003c/em\u003e to \u003cem\u003esenescent\u003c/em\u003e status (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eb and Extended Data Fig..5c, d and Methods). Two well-known senescence hallmark genes, \u003cem\u003eCDKN1A\u003c/em\u003e and \u003cem\u003eSERPINE1\u003c/em\u003e, showed higher expression level in \u003cem\u003esenescent\u003c/em\u003e subpopulation than in the other two subpopulations (\u003cem\u003ecycling\u003c/em\u003e and \u003cem\u003etransitional\u003c/em\u003e) (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ec). Moreover, GSEA results of specific highly expressing genes in each subpopulation indicated more frequent immune activities occurred in senescent tumor cells (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ed and Methods). These results demonstrated that heterogeneity in senescence existed among melanoma tumor cells, and hUSI can reliably distinguish senescent tumor cells.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eWe next analyzed the impact of senescent tumor cells on melanoma patient survival. We took three tumor cell subpopulations as a reference expression profile and deconvoluted RNA-seq profiles of melanoma cohort from TCGA-SKCM using EpiDISH\u003csup\u003e\u003cspan citationid=\"CR99\" class=\"CitationRef\"\u003e99\u003c/span\u003e\u003c/sup\u003e, obtaining the proportion of each subpopulation in each melanoma patient (Methods). Considering the potential relationship between senescent tumor cells and immune response, we also calculated abundances of 22 immune component using CIBERSORT\u003csup\u003e\u003cspan citationid=\"CR100\" class=\"CitationRef\"\u003e100\u003c/span\u003e\u003c/sup\u003e. We found that the proportion of \u003cem\u003esenescent\u003c/em\u003e subpopulation have a higher positive correlation with the abundance of M1 macrophage cells, CD8 T cells, and activated immune cells (including activated CD5 memory T cells and activated dendritic cells) (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ee). Survival analysis was then performed based on the inferred proportions of these subpopulations. The result showed that the higher proportion of \u003cem\u003esenescent\u003c/em\u003e or \u003cem\u003etransactional\u003c/em\u003e subpopulations in a patient, the more favorable it was for the patient\u0026rsquo;s survival, and the significance of \u003cem\u003esenescent\u003c/em\u003e is higher than \u003cem\u003etransactional\u003c/em\u003e (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ef). In contrast, patients with higher proportion of \u003cem\u003ecycling\u003c/em\u003e subpopulation have worse prognosis (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ef and Extended Data Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003ee). These results suggested that hUSI-aided senescence state evaluation of tumor cells can serve as a promising prognostic biomarker for melanoma patients.\u003c/p\u003e \u003cp\u003e \u003cb\u003ehUSI recognizes special signaling pathways in senescent melanoma cells\u003c/b\u003e \u003c/p\u003e \u003cp\u003eIn the above analysis, hUSI helps identify senescent tumor cells in melanoma. However, the role of senescent cells in tumor microenvironment is very complex and highly dependent on the physiological environment\u003csup\u003e\u003cspan additionalcitationids=\"CR102\" citationid=\"CR101\" class=\"CitationRef\"\u003e101\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR103\" class=\"CitationRef\"\u003e103\u003c/span\u003e\u003c/sup\u003e. Senescent cells can communicate with neighbor cells and influence their behavior through paracrine signaling. Specifically, SASP presented by senescent tumor cells plays important roles in communication with immune system by attracting immune cells (such as T cells and NK cells) and then leading to the clearance of senescent tumor cells\u003csup\u003e\u003cspan additionalcitationids=\"CR96\" citationid=\"CR95\" class=\"CitationRef\"\u003e95\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR97\" class=\"CitationRef\"\u003e97\u003c/span\u003e\u003c/sup\u003e. Besides, CS associated communication had been speculated to regulate immune surveillance and influence tumorigenesis\u003csup\u003e\u003cspan citationid=\"CR104\" class=\"CitationRef\"\u003e104\u003c/span\u003e\u003c/sup\u003e. Therefore, understanding how senescent cells interact with the microenvironment may provide additional clues for the relationship between senescence and tumorigenesis.\u003c/p\u003e \u003cp\u003eTo explore the cross-talk between senescent tumor cells and the microenvironment in melanoma, we investigated the cell-cell communication between these three tumor cell subpopulations (\u003cem\u003ecycling\u003c/em\u003e, \u003cem\u003etransactional and senescent\u003c/em\u003e) and their neighboring cells using CellChat\u003csup\u003e\u003cspan citationid=\"CR105\" class=\"CitationRef\"\u003e105\u003c/span\u003e\u003c/sup\u003e (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003ea). The results showed that the communication strength of hUSI-identified \u003cem\u003esenescent\u003c/em\u003e subpopulation was higher than the other two relatively less senescent tumor cell subpopulations (\u003cem\u003ecycling\u003c/em\u003e and \u003cem\u003etransactional\u003c/em\u003e), indicating stronger cell-cell communication between senescent tumor cells and neighboring cells (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eb and Extended Data Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003ef). Furthermore, analysis of the global output communication patterns uncovered two different signaling patterns, with pattern 1 corresponding to the \u003cem\u003esenescent\u003c/em\u003e tumor subpopulation and pattern 6 corresponding to the \u003cem\u003ecycling\u003c/em\u003e and \u003cem\u003etransitional\u003c/em\u003e subpopulations (Extended Data Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eg). To analyze which pathways were responsible for senescent tumor cells to receive communication signals from tumor microenvironment, we compared communication strength of each involved signaling pathway (Methods). Six pathways (including transforming growth factor β (TGFβ) pathway, leptin (LEP) pathway, chondroitin sulfate proteoglycan 4 (CSPG4) pathway, chemokine signaling pathways (CCL), CD6 pathway and bone morphogenetic protein (BMP) pathway) were found to have input signal strength to \u003cem\u003esenescent\u003c/em\u003e subpopulation and not detected in \u003cem\u003ecycling\u003c/em\u003e subpopulation, indicating that these signaling pathways are more likely to specifically function in senescent tumor cells (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003ec).\u003c/p\u003e \u003cp\u003eIn the two major pathways, TGF-β can induce senescent phenotype of tumor cells, which is secreted by macrophages originating in tumor stroma\u003csup\u003e\u003cspan citationid=\"CR106\" class=\"CitationRef\"\u003e106\u003c/span\u003e, \u003cspan citationid=\"CR107\" class=\"CitationRef\"\u003e107\u003c/span\u003e\u003c/sup\u003e, and BMP is a family of TGF-β superfamily, which has similarly been found to induce senescence of tumor cells\u003csup\u003e\u003cspan citationid=\"CR108\" class=\"CitationRef\"\u003e108\u003c/span\u003e, \u003cspan citationid=\"CR109\" class=\"CitationRef\"\u003e109\u003c/span\u003e\u003c/sup\u003e. Through signaling pathways pathway network, we found that senescent tumor cells receive TGF-β from macrophages (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003ed), which is consistent with previous report in lymphoma\u003csup\u003e\u003cspan citationid=\"CR107\" class=\"CitationRef\"\u003e107\u003c/span\u003e\u003c/sup\u003e. Interestingly, senescent tumor cells received more TGF-β from cancer-associated fibroblasts (CAFs) (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003ed). This may indicate that as a solid tumor, melanoma differs from lymphoma in microenvironment by the presence of a high number of fibroblasts. Moreover, senescent tumor cells received BMP from a variety of cell types in the microenvironment, of which the signal from T cells was the strongest (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003ed). Further investigation of ligand-receptor interactions in signaling pathways revealed that the expression level of genes encoding receptors for TGF-β and BMP were higher in \u003cem\u003esenescent\u003c/em\u003e subpopulation than in \u003cem\u003ecycling\u003c/em\u003e and \u003cem\u003etransitional\u003c/em\u003e subpopulations, with \u003cem\u003eTGFBR2\u003c/em\u003e and \u003cem\u003eBMPR1B\u003c/em\u003e showing more significant differences among these three subpopulations (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003ee). In addition, survival analysis performed on TCGA-SKCM also showed that patients with higher expression level of \u003cem\u003eTGFBR1\u003c/em\u003e, \u003cem\u003eTGFBR2\u003c/em\u003e and \u003cem\u003eBMPR2\u003c/em\u003e have better prognosis (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003ef), consistent with the idea that stronger interactions by these pathways between senescent tumor cells and microenvironment could benefit patient survival.\u003c/p\u003e \u003cp\u003eWe also noticed the other four signaling pathways which are also specific for senescent tumor cells. While \u003cem\u003etransitional\u003c/em\u003e and \u003cem\u003esenescent\u003c/em\u003e subpopulation interact with T cell by LEP signaling pathway and with CAFs by CSPG4 signaling pathway, the receptors involved in these two pathways did not show significant difference (Extended Data Fig.\u0026nbsp;6a). Notably, \u003cem\u003esenescent\u003c/em\u003e subpopulation interacts with T cell by CD6 signaling pathway and with macrophages by CCL signaling pathway. CD6 receptor encoding gene \u003cem\u003eALMCAM\u003c/em\u003e and CCL receptor encoding gene \u003cem\u003eCCR10\u003c/em\u003e are specifically highly expressed in \u003cem\u003esenescent\u003c/em\u003e subpopulation and benefit patients\u0026rsquo; survival (Extended Data Fig.\u0026nbsp;6b, c). Although, these signaling pathways are reported associated with tumor progression\u003csup\u003e\u003cspan additionalcitationids=\"CR111 CR112\" citationid=\"CR110\" class=\"CitationRef\"\u003e110\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR113\" class=\"CitationRef\"\u003e113\u003c/span\u003e\u003c/sup\u003e, their functions on tumor cell senescence need further study. Overall, these results highlight the clinical value of hUSI in identifying senescent tumor cells and the potentially involved signaling pathways.\u003c/p\u003e"},{"header":"3 Methods","content":"\u003cp\u003e \u003cb\u003eData collection.\u003c/b\u003e Bulk RNA-seq datasets used for extraction of senescence features were collected from previously published papers\u003csup\u003e\u003cspan additionalcitationids=\"CR20 CR21 CR22 CR23 CR24 CR25 CR26\" citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e\u003c/sup\u003e. We downloaded raw files from SRA database (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.ncbi.nlm.nih.gov/sra\u003c/span\u003e\u003cspan address=\"https://www.ncbi.nlm.nih.gov/sra\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) (GSE53356, GSE56293, GSE58910, GSE61130, GSE63577, GSE64553, GSE113957, GSE130727, and GSE60883) and EMBL-EBI (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.ebi.ac.uk/ebisearch/about\u003c/span\u003e\u003cspan address=\"https://www.ebi.ac.uk/ebisearch/about\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) (E-MTAB-5403). Only samples with confirmed senescence status were selected to build the representative senescence profiles. We also included corresponding non-senescent samples in each dataset for LOOCV. Seven microarray profiles used for batch effects test were download from Gene Expression Omnibus (GEO) database (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.ncbi.nlm.nih.gov/geo/\u003c/span\u003e\u003cspan address=\"https://www.ncbi.nlm.nih.gov/geo/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) (GSE101750, GSE101758, GSE61130, GSE122079, GSE113060, GSE72407, and GSE72404). The RNA-seq and microarray profiles of independent validation datasets were also downloaded from SRA database through GEO accession numbers: GSE60340, GSE130306, GSE19864, GSE16058, GSE83922, GSE11954, GSE100014, and GSE77239. Four scRNA-seq profiles used for benchmarking were downloaded from GEO (GSE119807, GSE115301, GSE94980, and GSE81547). More details of datasets mentioned above can be found in Supplementary Table \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e.\u003c/p\u003e \u003cp\u003eThe TPM normalized gene expression matrix, of which, TCGA was collected from UCSC Xena (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://xena.ucsc.edu/public/\u003c/span\u003e\u003cspan address=\"http://xena.ucsc.edu/public/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e), and GTEx was collected from GTEx Portal (version 8) (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.gtexportal.org/home/downloads/adult-gtex\u003c/span\u003e\u003cspan address=\"https://www.gtexportal.org/home/downloads/adult-gtex\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e). The raw single-nuclei counts matrix of normal and COVID-19 patients\u0026rsquo; lung tissues and the processed melanoma profiles were also respectively downloaded from GEO under accession numbers GSE171524 and GSE72056.\u003c/p\u003e \u003cp\u003e \u003cb\u003eBulk RNA-seq data of training set processing.\u003c/b\u003e We used Prefetch v.3.0.2 to download SRA files and split them into FASTQ files by parallel-fastq-dump v.0.6.5. TrimGalore v.0.6.6 was used to filter out low quality reads and bases in 3\u0026rsquo; end for the consequential alignment performed by STAR v.2.2.1. We chose GRCh38 as reference genome sequence and only unique mapping reads were included. StringTie was used to qualify the gene expression level and normalize values by TPM (transcripts per million). Only protein coding genes (annotated by gencode v.31) with TPM\u0026thinsp;\u0026gt;\u0026thinsp;3 in 99% samples were included for next analysis. Considering bias introduced by batch effect, we also used log space transform to further reduce the disturbance for feature extraction.\u003c/p\u003e \u003cp\u003e \u003cb\u003eQuantification of cellular senescence based on OCLR.\u003c/b\u003e To quantify CS based on gene expression level, a predictive model was built using OCLR\u003csup\u003e\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e\u003c/sup\u003e in R package \u0026ldquo;gelnet\u0026rdquo; with the parameters y\u0026thinsp;=\u0026thinsp;NULL, l1\u0026thinsp;=\u0026thinsp;0, and l2\u0026thinsp;=\u0026thinsp;1. The input expression matrix of senescent cells was normalized by subtracting mean expression value across all samples. The Spearman correlation coefficient was defined as hUSI as its stable performance for minimizing possible batch effects across datasets\u003csup\u003e\u003cspan citationid=\"CR114\" class=\"CitationRef\"\u003e114\u003c/span\u003e\u003c/sup\u003e. Normalized gene expression matrix was used to calculate hUSI.\u003c/p\u003e \u003cp\u003e \u003cb\u003eGene set enrichment analysis for learned-senescence features.\u003c/b\u003e Hallmark gene sets from The Molecular Signatures Database (MSigDB) database were included to perform GSEA for genes in OCLR learned-features which were sorted by their weights. Normalized enrichment score (NES) calculated by R package \u0026ldquo;fgsea\u0026rdquo;\u003csup\u003e\u003cspan citationid=\"CR115\" class=\"CitationRef\"\u003e115\u003c/span\u003e\u003c/sup\u003e was chosen to compare the enrichment degree in different gene sets.\u003c/p\u003e \u003cp\u003e \u003cb\u003ePerformance evaluation of hUSI.\u003c/b\u003e The reliability of the acquired feature and quantification strategy were validated using LOOCV. For each training round, we excluded one senescent sample in our training set and trained the model to extract senescence features. Then hUSI was calculated based on features to score the leave-out sample as well as all the non-senescent samples. Finally, the probability that the score of senescent samples is higher than that of non-senescent samples\u003csup\u003e\u003cspan citationid=\"CR116\" class=\"CitationRef\"\u003e116\u003c/span\u003e\u003c/sup\u003e was used to measure the performance, denoted as correctly ranking probability (CRP). For the four scRNA-seq datasets GSE119807, GSE115301, GSE94980, and GSE81547 ) used for comparison of three types of scoring methods, the expression matrix was read and normalized using R package \u0026ldquo;Seurat\u0026rdquo;. Then, log-normalized gene expression value was used for directly classifying senescent cells or calculating senescence score. ssGSEA score for each gene set was produced by \u0026ldquo;gsea()\u0026rdquo; function in the R package \u0026ldquo;GSVA\u0026rdquo;. To better compare the performance of all scoring methods in four single-cell datasets, we randomly divided each dataset into 10 subsets and repeated for 3 times by \u0026ldquo;createMultiFolds()\u0026rdquo; function in R package \u0026ldquo;caret\u0026rdquo;, finally generating 30 data units in total. For each of the 30 data units, we computed the AUC based on the scores generated by the tested methods described below by \u0026ldquo;auc()\u0026rdquo; function in R package \u0026ldquo;pROC\u0026rdquo;. For the expression level of marker genes, the parameter \u0026ldquo;direction\u0026rdquo; was not assigned as these genes were reported up or down regulated in senescent cells. For the computed senescence score and ssGSEA score, we set the parameter \u0026ldquo;direction=\u0026lt;\u0026rdquo; to make sure it will have a higher AUC only if these scores are lower in non-senescent samples. We calculated the average AUC of all 30 units and ascendingly ranked the average AUCs derived from each method group for each dataset. The mean average AUC rank across four datasets was used to reflect their overall performance (Supplementary Table S3).\u003c/p\u003e \u003cp\u003e \u003cb\u003eInference of cellular senescence states.\u003c/b\u003e For a large number of single cells, we adopted a GMM framework\u003csup\u003e\u003cspan citationid=\"CR117\" class=\"CitationRef\"\u003e117\u003c/span\u003e\u003c/sup\u003e to infer the number of potential senescence states according to the distribution of hUSI. Log transformation was firstly performed on the hUSI of each cell, that is, Logit(hUSI)\u0026thinsp;=\u0026thinsp;log2[(1\u0026thinsp;+\u0026thinsp;hUSI)/(1-hUSI)]. The Logit(hUSI) values of all cells were then fitted under the framework of GMM (implemented in R package \u0026ldquo;mclust\u0026rdquo;), and the Bayesian information criterion (BIC) was used to estimate the optimal number of senescence states and the probability that each cell belonged to a specific state\u003csup\u003e\u003cspan citationid=\"CR118\" class=\"CitationRef\"\u003e118\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003e \u003cb\u003eAnalysis of snRNA-seq data of COVID-19 infected samples\u003c/b\u003e. The raw expression matrix was read using Python module \"scanpy\" to filter out low quality cells (min_genes\u0026thinsp;=\u0026thinsp;200 and min_cells\u0026thinsp;=\u0026thinsp;3). After removing mitochondrial and ribosomal genes, the expression matrix was normalized, and the highly variable gene matrix was considered for downstream scaling, batch effect removing and visualization. The \"rpy2\" module was used to call the \"mclust\" R package in python for cellular senescence state classification. GSEA for DEGs of senescence class was performed by python module \u0026ldquo;gseapy\u0026rdquo;\u003csup\u003e\u003cspan citationid=\"CR119\" class=\"CitationRef\"\u003e119\u003c/span\u003e\u003c/sup\u003e using log\u003csub\u003e2\u003c/sub\u003e-transformed Fold-Change as ranking metric.\u003c/p\u003e \u003cp\u003e \u003cb\u003eAnalysis of scRNA-seq data of melanoma samples.\u003c/b\u003e The processed melanoma single-cell matrices, retaining only defined tumor cells (malignant\u0026thinsp;=\u0026thinsp;=\u0026thinsp;2) and non-tumor cells (malignant\u0026thinsp;=\u0026thinsp;=\u0026thinsp;1) (including T-cells, B-cells, macrophages, endothelial cells, cancer-associated fibroblasts and natural killer cells), were read, analyzed and visualized using the R package \"Seurat\". To validate the reliability of inferred tumor subpopulations, specifically highly expressing genes (logfc.threshold\u0026thinsp;\u0026gt;\u0026thinsp;0.1) of each subpopulation were overlapped with DEGs derived from melanoma microarray data (senescent vs young), which were calculated by linear models \u0026ldquo;lm(gene expession\u0026thinsp;~\u0026thinsp;pheno)\u0026rdquo;. R function \u0026ldquo;phyper()\u0026rdquo; was used to test the overlapping significance. To observe the positional relationships of the different subpopulations of CS states on projected space, for each gene, we calculated the Pearson correlation coefficient between hUSIs and gene expression values of all cells, and the top 1500 genes ranked by absolute correlation coefficient values were selected as hUSI related genes. Then, using the tool ICAnet\u003csup\u003e\u003cspan citationid=\"CR120\" class=\"CitationRef\"\u003e120\u003c/span\u003e\u003c/sup\u003e, the tumor cells were integrated based on the 38 co-expression modules of the above hUSI related genes. Diffusion map based on five principal components of 38 co-expression modules was used to further reduce the dimension and infer senescence trajectory using R package \u0026ldquo;destiny\u0026rdquo;\u003csup\u003e\u003cspan citationid=\"CR121\" class=\"CitationRef\"\u003e121\u003c/span\u003e\u003c/sup\u003e. Specifically highly expressed genes (logfc.threshold\u0026thinsp;\u0026gt;\u0026thinsp;0.1) of tumor subpopulation were enriched using R package \u0026ldquo;clusterProfiler \u0026rdquo;\u003csup\u003e\u003cspan citationid=\"CR122\" class=\"CitationRef\"\u003e122\u003c/span\u003e\u003c/sup\u003e on the database Gene-Ontology (GO), Kyoto Encyclopedia of Genes and Genomes (KEGG), using log\u003csub\u003e2\u003c/sub\u003e-transformed Fold-Change as the ranking metric. Cell communication analysis was carried out using R package \u0026ldquo;CellChat\u0026rdquo;, and the communication intensity between tumor subpopulations and different non-tumor cell types in a signal network was quantified\u003csup\u003e\u003cspan citationid=\"CR105\" class=\"CitationRef\"\u003e105\u003c/span\u003e\u003c/sup\u003e. When filtered pathways specific for senescent tumor cells, we set three tumor subpopulations as \u0026ldquo;target\u0026rdquo; and T cell, NK cell, macro cell and CAF cell as \u0026ldquo;source\u0026rdquo;.\u003c/p\u003e \u003cp\u003e \u003cb\u003eData analysis of melanoma patient cohort in TCGA.\u003c/b\u003e The normalized Level 3 RNA-seq data of a melanoma patient cohort (SKCM) with associated clinical data were downloaded from the TCGA (the Cancer Genome Atlas) database using R package TCGAbiolinks\u003csup\u003e\u003cspan citationid=\"CR123\" class=\"CitationRef\"\u003e123\u003c/span\u003e\u003c/sup\u003e. To analyze the proportion of cells with different senescence degree, three tumor subpopulations were used as reference to deconvolute the RNA-seq data of the SKCM patient cohort using EpiDISH\u003csup\u003e\u003cspan citationid=\"CR99\" class=\"CitationRef\"\u003e99\u003c/span\u003e\u003c/sup\u003e. The abundance of the 22 immune components were calculated by CIBERSORT\u003csup\u003e\u003cspan citationid=\"CR100\" class=\"CitationRef\"\u003e100\u003c/span\u003e\u003c/sup\u003e. Survival analysis was performed using R packages \u0026ldquo;survival\u0026rdquo; and \u0026ldquo;survminer\u0026rdquo; with \u0026ldquo;OS\u0026thinsp;=\u0026thinsp;vital_status\u0026rdquo; and \u0026ldquo;OS.time\u0026thinsp;=\u0026thinsp;days_to_last_followup\u0026rdquo;.\u003c/p\u003e"},{"header":"4 Discussion","content":"\u003cp\u003eThough the characterization, identification, and pharmacological clearance of senescent cell are the basics of many senotherapies, the relative scarce of specific and efficient senescence marker keeps limiting the study of distinguishing and targeting senescent cells both \u003cem\u003ein vitro\u003c/em\u003e and \u003cem\u003ein vivo\u003c/em\u003e\u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e, \u003cspan citationid=\"CR124\" class=\"CitationRef\"\u003e124\u003c/span\u003e, \u003cspan citationid=\"CR125\" class=\"CitationRef\"\u003e125\u003c/span\u003e\u003c/sup\u003e. Additionally, in single-cell studies, the differential expression level of a single senescence marker is insufficient to identify senescent cells due to the heterogeneity of both cell types and senescence status\u003csup\u003e\u003cspan citationid=\"CR64\" class=\"CitationRef\"\u003e64\u003c/span\u003e\u003c/sup\u003e. Besides, apart from the absence or low expression of certain senescence markers, the possibility of improperly calculating senescence score in some methods also exists due to the absent detection or abnormal expression of key senescence genes. For example, in lassoCS\u003csup\u003e\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u003c/sup\u003e method, three out of ten genes (\u003cem\u003eSEMA3G\u003c/em\u003e, \u003cem\u003ePCSK6\u003c/em\u003e, and \u003cem\u003eSLC44A4\u003c/em\u003e) are assigned with weight values for senescence score calculation, while they are absent when we applied it to another single-cell dataset\u003csup\u003e\u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e48\u003c/span\u003e\u003c/sup\u003e. Taking gene sets into consideration, while the enrichment score generated by GSEA or GSVA is widely utilized, different gene sets usually emphasize on different aspects of senescence. For instance, the SASP gene set focuses on the activated secretory phenotype, whereas signature of replicative senescence\u003csup\u003e\u003cspan citationid=\"CR76\" class=\"CitationRef\"\u003e76\u003c/span\u003e\u003c/sup\u003e specifically considers replicative-related changes. Notably, performing GSEA in large-scale single-cell atlases can be exceedingly time-consuming, due to the substantial number of permutations required to accurately estimate the nominal p-value\u003csup\u003e\u003cspan citationid=\"CR126\" class=\"CitationRef\"\u003e126\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eTo overcome these challenges above, the present study adopted OCLR machine learning algorithm\u003csup\u003e\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e\u003c/sup\u003e to acquire gene expression features of CS in relatively comprehensive senescence transcriptome profiles. Based on OCLR machine learned-features, we developed hUSI, a scoring method using Spearman correlation coefficient that can distinguish senescent samples or cells induced by different factors while less affected by batch effect. We validated the generalizability of hUSI by applying it to datasets of variety origins encompassing platforms, cell types, or induction factors that were not included in the training set. Moreover, the stability and the potential application of hUSI on single-cell data were further validated by simulated sparse profiles and real scRNA-seq profiles.\u003c/p\u003e \u003cp\u003eComparing with other currently available methods (including those based on senescence markers expression level, computed senescence score and ssGSEA score), hUSI manifested its reliability and superior performance in senescence evaluation. Unlike methods that rely on limited genes, hUSI takes14,638 protein-coding genes assigned with different weights learned by OCLR into senescence evaluation, which reduces bias when evaluating CS of distinct senescence types. Importantly, hUSI demonstrated reliable performance when applied to samples derived from both health and disease populations.\u003c/p\u003e \u003cp\u003eCombination of hUSI with GMM provided a framework to reveal cell classes with different senescence level. hUSI identified sell types showing higher accumulation of senescent cells in COVID-19-infected lung tissue and provided a potential therapeutic avenue for selectively eliminating these senescent cells to mitigate senescence-induced hyperinflammation in COVID-19 patients\u003csup\u003e\u003cspan citationid=\"CR82\" class=\"CitationRef\"\u003e82\u003c/span\u003e\u003c/sup\u003e. For example, navitoclax, a senolytic drug, has been reported to target senescent alveolar epithelial cells and macrophages, and in turn reduces the secretion of pro-inflammatory SASP factors after SARS-CoV-2 infection\u003csup\u003e\u003cspan citationid=\"CR81\" class=\"CitationRef\"\u003e81\u003c/span\u003e\u003c/sup\u003e. Thus, our ongoing research will center on integrating senescence quantification into the cellular response to drugs to unveil the contribution of senescent cells to both drug resistance and drug sensitivity across various diseases.\u003c/p\u003e \u003cp\u003eBy investigating the communication network of melanoma cells of different senescence degree, combing with previous studies\u003csup\u003e\u003cspan citationid=\"CR106\" class=\"CitationRef\"\u003e106\u003c/span\u003e, \u003cspan citationid=\"CR107\" class=\"CitationRef\"\u003e107\u003c/span\u003e\u003c/sup\u003e, we hypothesize that senescent tumor cells can interact with tumor microenvironment by TGF-β and BMP signaling pathways in melanoma microenvironment and contribute to anti-cancer effects. TGF-β is well known being associated with the upregulated expression of p15, p21, and p27, which are known senescence markers and can inhibit cell proliferation\u003csup\u003e\u003cspan citationid=\"CR127\" class=\"CitationRef\"\u003e127\u003c/span\u003e\u003c/sup\u003e. BMP, a member of TGF-β family, has also been reported having a crucial role in paracrine induction in senescent cells\u003csup\u003e\u003cspan citationid=\"CR108\" class=\"CitationRef\"\u003e108\u003c/span\u003e, \u003cspan citationid=\"CR109\" class=\"CitationRef\"\u003e109\u003c/span\u003e\u003c/sup\u003e. We observed higher expression level of genes encoding TGF and BMP receptors in senescent cells, which support their specifical functions in senescence of tumor cells.\u003c/p\u003e \u003cp\u003eAlthough hUSI has demonstrated superior performance in multiple aspects, we must admit that there are still some limitations in current status. First, though the stability of hUSI has been demonstrated by LOOCV and independent datasets, with more data coming in the big data era, the quality and quantity of training sets still has room for further improvement to optimize the performance of hUSI in the future. Second, cell-cell interactions networks are dynamic and intricate, no matter among senescent cells in different cell types or between senescent cells and non-senescent cells. However, in this study, we simplified the interactions networks by only focusing on senescent tumor cells due to the limited cell numbers of other cell types. Besides, the causal relationship between signaling pathways and the mechanism of how senescent tumor cells response to different signals also require further study. Third, hUSI was a human transcriptome-based scoring method. Although, using homologous genes when applied to other species is a feasible way, the reliability and the accuracy need extensive validation. We also believed senescence scoring tool can be developed based on transcriptomes of interested species following our stagey. Finally, with implementation of quantifying CS degree, a more detailed and standardized CS atlas supported by adequate experimental and multi-omics evidence is demanding to benefit the prevention of aging-related diseases and the application of senotherapeutics.\u003c/p\u003e"},{"header":"5 Conclusion","content":"\u003cp\u003eIn summary, we developed a senescence-evaluating tool that outperforms currently existing analogous methods, capable of robustly quantifying sample senescence degree based on bulk or single-cell transcriptome profiles. We also proposed a framework for classifying the senescence status of various cell types and recognizing senescence-specific intercellular communications. Based on the outperformance and applicability of hUSI, we believe that it will greatly help to evaluate senescence and benefit studies and even therapeutic strategies in senescence and age-related diseases.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis work was financially supported by the National Natural Science Foundation of China (92249302, 32370592), the National Key Research and Development Program of China (2023YFC3603300, 2021YFA0909300).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConflict of interest\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCode availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003ehUSI can be implemented both in R and Python, the data and codes used to reproduce our analysis results are provided in GitHub (https://github.com/WJPina/HUSI), along with a detailed usage guideline.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor contribution\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe manuscript was written by J.W., W.W., and J.Y. and polished by G.W., and T.N.. The method was conceived by J.Y. and W.W. and the algorithm is implemented by J.W., and J.Y. Computational analyses and algorithm evaluations were conducted by J.W., W.W., J.Y., and X.Z. This work was supervised by T.N..\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n \u003cli\u003evan Deursen, J.M. The role of senescent cells in ageing. \u003cem\u003eNature\u003c/em\u003e\u003cstrong\u003e509\u003c/strong\u003e, 439-46 (2014).\u003c/li\u003e\n \u003cli\u003eAcosta, J.C. et al. A complex secretory program orchestrated by the inflammasome controls paracrine senescence. \u003cem\u003eNat Cell Biol\u003c/em\u003e\u003cstrong\u003e15\u003c/strong\u003e, 978-90 (2013).\u003c/li\u003e\n \u003cli\u003eBirch, J. \u0026amp; Gil, J. Senescence and the SASP: many therapeutic avenues. \u003cem\u003eGenes Dev\u003c/em\u003e\u003cstrong\u003e34\u003c/strong\u003e, 1565-1576 (2020).\u003c/li\u003e\n \u003cli\u003eChaib, S., Tchkonia, T. \u0026amp; Kirkland, J.L. Cellular senescence and senolytics: the path to the clinic. \u003cem\u003eNat Med\u003c/em\u003e\u003cstrong\u003e28\u003c/strong\u003e, 1556-1568 (2022).\u003c/li\u003e\n \u003cli\u003eWang, L., Lankhorst, L. \u0026amp; Bernards, R. Exploiting senescence for the treatment of cancer. \u003cem\u003eNat Rev Cancer\u003c/em\u003e\u003cstrong\u003e22\u003c/strong\u003e, 340-355 (2022).\u003c/li\u003e\n \u003cli\u003eCho, K.A. et al. Morphological adjustment of senescent cells by modulating caveolin-1 status. \u003cem\u003eJ Biol Chem\u003c/em\u003e\u003cstrong\u003e279\u003c/strong\u003e, 42270-8 (2004).\u003c/li\u003e\n \u003cli\u003ePetrova, N.V., Velichko, A.K., Razin, S.V. \u0026amp; Kantidze, O.L. Small molecule compounds that induce cellular senescence. \u003cem\u003eAging Cell\u003c/em\u003e\u003cstrong\u003e15\u003c/strong\u003e, 999-1017 (2016).\u003c/li\u003e\n \u003cli\u003eRossiello, F., Herbig, U., Longhese, M.P., Fumagalli, M. \u0026amp; D\u0026apos;Adda, D.F.F. Irreparable telomeric DNA damage and persistent DDR signalling as a shared causative mechanism of cellular senescence and ageing. \u003cem\u003eCurr Opin Genet Dev\u003c/em\u003e\u003cstrong\u003e26\u003c/strong\u003e, 89-95 (2014).\u003c/li\u003e\n \u003cli\u003eMunoz-Espin, D. et al. Programmed cell senescence during mammalian embryonic development. \u003cem\u003eCell\u003c/em\u003e\u003cstrong\u003e155\u003c/strong\u003e, 1104-18 (2013).\u003c/li\u003e\n \u003cli\u003eHernandez-Segura, A., Nehme, J. \u0026amp; Demaria, M. Hallmarks of Cellular Senescence. \u003cem\u003eTrends Cell Biol\u003c/em\u003e\u003cstrong\u003e28\u003c/strong\u003e, 436-453 (2018).\u003c/li\u003e\n \u003cli\u003eHuang, W., Hickson, L.J., Eirin, A., Kirkland, J.L. \u0026amp; Lerman, L.O. Cellular senescence: the good, the bad and the unknown. \u003cem\u003eNature reviews. Nephrology\u003c/em\u003e\u003cstrong\u003e18\u003c/strong\u003e, 611-627 (2022).\u003c/li\u003e\n \u003cli\u003eHan, X. et al. Construction of a human cell landscape at single-cell level. \u003cem\u003eNature\u003c/em\u003e\u003cstrong\u003e581\u003c/strong\u003e, 303-309 (2020).\u003c/li\u003e\n \u003cli\u003eHe, S. et al. Single-cell transcriptome profiling of an adult human cell atlas of 15 major organs. \u003cem\u003eGenome Biol\u003c/em\u003e\u003cstrong\u003e21\u003c/strong\u003e, 294 (2020).\u003c/li\u003e\n \u003cli\u003eLafferty-Whyte, K. et al. Scoring of senescence signalling in multiple human tumour gene expression datasets, identification of a correlation between senescence score and drug toxicity in the NCI60 panel and a pro-inflammatory signature correlating with survival advantage in peritoneal mesothelioma. \u003cem\u003eBMC Genomics\u003c/em\u003e\u003cstrong\u003e11\u003c/strong\u003e, 532 (2010).\u003c/li\u003e\n \u003cli\u003ede Magalhaes, J.P., Curado, J. \u0026amp; Church, G.M. Meta-analysis of age-related gene expression profiles identifies common signatures of aging. \u003cem\u003eBioinformatics\u003c/em\u003e\u003cstrong\u003e25\u003c/strong\u003e, 875-81 (2009).\u003c/li\u003e\n \u003cli\u003eSaul, D. et al. A new gene set identifies senescent cells and predicts senescence-associated pathways across tissues. \u003cem\u003eNat Commun\u003c/em\u003e\u003cstrong\u003e13\u003c/strong\u003e, 4827 (2022).\u003c/li\u003e\n \u003cli\u003eWang, X. et al. Comprehensive assessment of cellular senescence in the tumor microenvironment. \u003cem\u003eBrief Bioinform\u003c/em\u003e\u003cstrong\u003e23\u003c/strong\u003e (2022).\u003c/li\u003e\n \u003cli\u003eGong, Q., Jiang, Y., Xiong, J., Liu, F. \u0026amp; Guan, J. Integrating scRNA and bulk-RNA sequencing develops a cell senescence signature for analyzing tumor heterogeneity in clear cell renal cell carcinoma. \u003cem\u003eFront Immunol\u003c/em\u003e\u003cstrong\u003e14\u003c/strong\u003e, 1199002 (2023).\u003c/li\u003e\n \u003cli\u003eHernandez-Segura, A. et al. Unmasking Transcriptional Heterogeneity in Senescent Cells. \u003cem\u003eCurr Biol\u003c/em\u003e\u003cstrong\u003e27\u003c/strong\u003e, 2652-2660.e4 (2017).\u003c/li\u003e\n \u003cli\u003eCasella, G. et al. Transcriptome signature of cellular senescence. \u003cem\u003eNucleic Acids Res\u003c/em\u003e\u003cstrong\u003e47\u003c/strong\u003e, 7294-7305 (2019).\u003c/li\u003e\n \u003cli\u003eRai, T.S. et al. HIRA orchestrates a dynamic chromatin landscape in senescence and is required for suppression of neoplasia. \u003cem\u003eGenes Dev\u003c/em\u003e\u003cstrong\u003e28\u003c/strong\u003e, 2712-25 (2014).\u003c/li\u003e\n \u003cli\u003eAlspach, E. et al. P38MAPK plays a crucial role in stromal-mediated tumorigenesis. \u003cem\u003eCancer discovery\u003c/em\u003e\u003cstrong\u003e4\u003c/strong\u003e, 716\u003c/li\u003e\n \u003cli\u003eCrowe, E.P. et al. Changes in the Transcriptome of Human Astrocytes Accompanying Oxidative Stress-Induced Senescence. \u003cem\u003eFront Aging Neurosci\u003c/em\u003e\u003cstrong\u003e8\u003c/strong\u003e, 208 (2016).\u003c/li\u003e\n \u003cli\u003eHerranz, N. et al. mTOR regulates MAPKAPK2 translation to control the senescence-associated secretory phenotype. \u003cem\u003eNat Cell Biol\u003c/em\u003e\u003cstrong\u003e17\u003c/strong\u003e, 1205-17 (2015).\u003c/li\u003e\n \u003cli\u003eMarthandan, S. et al. Similarities in Gene Expression Profiles during In Vitro Aging of Primary Human Embryonic Lung and Foreskin Fibroblasts. \u003cem\u003eBiomed Res Int\u003c/em\u003e\u003cstrong\u003e2015\u003c/strong\u003e, 731938 (2015).\u003c/li\u003e\n \u003cli\u003eMarthandan, S. et al. Hormetic effect of rotenone in primary human fibroblasts. \u003cem\u003eImmun Ageing\u003c/em\u003e\u003cstrong\u003e12\u003c/strong\u003e, 11 (2015).\u003c/li\u003e\n \u003cli\u003eFleischer, J.G. et al. Predicting age from the transcriptome of human dermal fibroblasts. \u003cem\u003eGenome Biol\u003c/em\u003e\u003cstrong\u003e19\u003c/strong\u003e, 221 (2018).\u003c/li\u003e\n \u003cli\u003eLin, W. et al. Identification and validation of cellular senescence patterns to predict clinical outcomes and immunotherapeutic responses in lung adenocarcinoma. \u003cem\u003eCancer Cell Int\u003c/em\u003e\u003cstrong\u003e21\u003c/strong\u003e, 652 (2021).\u003c/li\u003e\n \u003cli\u003ePark, H.S. \u0026amp; Kim, S.Y. Endothelial cell senescence: A machine learning-based meta-analysis of transcriptomic studies. \u003cem\u003eAgeing Res Rev\u003c/em\u003e\u003cstrong\u003e65\u003c/strong\u003e, 101213 (2021).\u003c/li\u003e\n \u003cli\u003eJochems, F. et al. The Cancer SENESCopedia: A delineation of cancer cell senescence. \u003cem\u003eCell Rep\u003c/em\u003e\u003cstrong\u003e36\u003c/strong\u003e, 109441 (2021).\u003c/li\u003e\n \u003cli\u003eKumari, R. \u0026amp; Jat, P. Mechanisms of Cellular Senescence: Cell Cycle Arrest and Senescence Associated Secretory Phenotype. \u003cem\u003eFront Cell Dev Biol\u003c/em\u003e\u003cstrong\u003e9\u003c/strong\u003e, 645593 (2021).\u003c/li\u003e\n \u003cli\u003eSokolov, A., Paull, E.O. \u0026amp; Stuart, J.M. ONE-CLASS DETECTION OF CELL STATES IN TUMOR SUBTYPES. \u003cem\u003ePac Symp Biocomput\u003c/em\u003e\u003cstrong\u003e21\u003c/strong\u003e, 405-16 (2016).\u003c/li\u003e\n \u003cli\u003eLim, S., Lim, J., Lee, A., Kim, K.I. \u0026amp; Lim, J.S. Anticancer Effect of E26 Transformation-Specific Homologous Factor through the Induction of Senescence and the Inhibition of Epithelial-Mesenchymal Transition in Triple-Negative Breast Cancer Cells. \u003cem\u003eCancers (Basel)\u003c/em\u003e\u003cstrong\u003e15\u003c/strong\u003e (2023).\u003c/li\u003e\n \u003cli\u003eTakaya, K., Asou, T. \u0026amp; Kishi, K. Identification of Apolipoprotein D as a Dermal Fibroblast Marker of Human Aging for Development of Skin Rejuvenation Therapy. \u003cem\u003eRejuvenation Res\u003c/em\u003e\u003cstrong\u003e26\u003c/strong\u003e, 42-50 (2023).\u003c/li\u003e\n \u003cli\u003eHari, P. et al. The innate immune sensor Toll-like receptor 2 controls the senescence-associated secretory phenotype. \u003cem\u003eSci Adv\u003c/em\u003e\u003cstrong\u003e5\u003c/strong\u003e, eaaw0254 (2019).\u003c/li\u003e\n \u003cli\u003eLiberzon, A. et al. The Molecular Signatures Database (MSigDB) hallmark gene set collection. \u003cem\u003eCell Syst\u003c/em\u003e\u003cstrong\u003e1\u003c/strong\u003e, 417-425 (2015).\u003c/li\u003e\n \u003cli\u003eKim, K.S., Kang, K.W., Seu, Y.B., Baek, S.H. \u0026amp; Kim, J.R. Interferon-gamma induces cellular senescence through p53-dependent DNA damage signaling in human endothelial cells. \u003cem\u003eMech Ageing Dev\u003c/em\u003e\u003cstrong\u003e130\u003c/strong\u003e, 179-88 (2009).\u003c/li\u003e\n \u003cli\u003eCisowski, J., Sayin, V.I., Liu, M., Karlsson, C. \u0026amp; Bergo, M.O. Oncogene-induced senescence underlies the mutual exclusive nature of oncogenic KRAS and BRAF. \u003cem\u003eOncogene\u003c/em\u003e\u003cstrong\u003e35\u003c/strong\u003e, 1328-33 (2016).\u003c/li\u003e\n \u003cli\u003eLasry, A. \u0026amp; Ben-Neriah, Y. Senescence-associated inflammatory responses: aging and cancer perspectives. \u003cem\u003eTrends Immunol\u003c/em\u003e\u003cstrong\u003e36\u003c/strong\u003e, 217-28 (2015).\u003c/li\u003e\n \u003cli\u003eArtandi, S.E. \u0026amp; Attardi, L.D. Pathways connecting telomeres and p53 in senescence, apoptosis, and cancer. \u003cem\u003eBiochem Biophys Res Commun\u003c/em\u003e\u003cstrong\u003e331\u003c/strong\u003e, 881-90 (2005).\u003c/li\u003e\n \u003cli\u003eSerrano, M., Lin, A.W., McCurrach, M.E., Beach, D. \u0026amp; Lowe, S.W. Oncogenic ras provokes premature cell senescence associated with accumulation of p53 and p16INK4a. \u003cem\u003eCell\u003c/em\u003e\u003cstrong\u003e88\u003c/strong\u003e, 593-602 (1997).\u003c/li\u003e\n \u003cli\u003eOshi, M. et al. G2M Cell Cycle Pathway Score as a Prognostic Biomarker of Metastasis in Estrogen Receptor (ER)-Positive Breast Cancer. \u003cem\u003eInt J Mol Sci\u003c/em\u003e\u003cstrong\u003e21\u003c/strong\u003e (2020).\u003c/li\u003e\n \u003cli\u003eNarita, M. et al. Rb-mediated heterochromatin formation and silencing of E2F target genes during cellular senescence. \u003cem\u003eCell\u003c/em\u003e\u003cstrong\u003e113\u003c/strong\u003e, 703-16 (2003).\u003c/li\u003e\n \u003cli\u003eDikovskaya, D. et al. Mitotic Stress Is an Integral Part of the Oncogene-Induced Senescence Program that Promotes Multinucleation and Cell Cycle Arrest. \u003cem\u003eCell Rep\u003c/em\u003e\u003cstrong\u003e12\u003c/strong\u003e, 1483-96 (2015).\u003c/li\u003e\n \u003cli\u003eWu, C.H. et al. Cellular senescence is an important mechanism of tumor regression upon c-Myc inactivation. \u003cem\u003eProc Natl Acad Sci U S A\u003c/em\u003e\u003cstrong\u003e104\u003c/strong\u003e, 13028-33 (2007).\u003c/li\u003e\n \u003cli\u003eGuerrero, A. et al. Cardiac glycosides are broad-spectrum senolytics. \u003cem\u003eNat Metab\u003c/em\u003e\u003cstrong\u003e1\u003c/strong\u003e, 1074-1088 (2019).\u003c/li\u003e\n \u003cli\u003eHoare, M. et al. NOTCH1 mediates a switch between two distinct secretomes during senescence. \u003cem\u003eNat Cell Biol\u003c/em\u003e\u003cstrong\u003e18\u003c/strong\u003e, 979-92 (2016).\u003c/li\u003e\n \u003cli\u003eParry, A.J. et al. NOTCH-mediated non-cell autonomous regulation of chromatin structure during senescence. \u003cem\u003eNat Commun\u003c/em\u003e\u003cstrong\u003e9\u003c/strong\u003e, 1840 (2018).\u003c/li\u003e\n \u003cli\u003eGeorgilis, A. et al. PTBP1-Mediated Alternative Splicing Regulates the Inflammatory Secretome and the Pro-tumorigenic Effects of Senescent Cells. \u003cem\u003eCancer Cell\u003c/em\u003e\u003cstrong\u003e34\u003c/strong\u003e, 85-102.e9 (2018).\u003c/li\u003e\n \u003cli\u003eCostarelli, L. et al. Different transcriptional profiling between senescent and non-senescent human coronary artery endothelial cells (HCAECs) by Omeprazole and Lansoprazole treatment. \u003cem\u003eBiogerontology\u003c/em\u003e\u003cstrong\u003e18\u003c/strong\u003e, 217-236 (2017).\u003c/li\u003e\n \u003cli\u003eChicas, A. et al. Dissecting the unique role of the retinoblastoma tumor suppressor during cellular senescence. \u003cem\u003eCancer Cell\u003c/em\u003e\u003cstrong\u003e17\u003c/strong\u003e, 376-87 (2010).\u003c/li\u003e\n \u003cli\u003eOrfanidis, K., Waster, P., Lundmark, K., Rosdahl, I. \u0026amp; Ollinger, K. Evaluation of tubulin beta-3 as a novel senescence-associated gene in melanocytic malignant transformation. \u003cem\u003ePigment Cell Melanoma Res\u003c/em\u003e\u003cstrong\u003e30\u003c/strong\u003e, 243-254 (2017).\u003c/li\u003e\n \u003cli\u003eGarbe, J.C. et al. Molecular distinctions between stasis and telomere attrition senescence barriers shown by long-term culture of normal human mammary epithelial cells. \u003cem\u003eCancer Res\u003c/em\u003e\u003cstrong\u003e69\u003c/strong\u003e, 7557-68 (2009).\u003c/li\u003e\n \u003cli\u003eKrizhanovsky, V. et al. Senescence of activated stellate cells limits liver fibrosis. \u003cem\u003eCell\u003c/em\u003e\u003cstrong\u003e134\u003c/strong\u003e, 657-67 (2008).\u003c/li\u003e\n \u003cli\u003eYuan, L. et al. Switching off IMMP2L signaling drives senescence via simultaneous metabolic alteration and blockage of cell death. \u003cem\u003eCell Res\u003c/em\u003e\u003cstrong\u003e28\u003c/strong\u003e, 625-643 (2018).\u003c/li\u003e\n \u003cli\u003eSomekh, J., Shen-Orr, S.S. \u0026amp; Kohane, I.S. Batch correction evaluation framework using a-priori gene-gene associations: applied to the GTEx dataset. \u003cem\u003eBMC Bioinformatics\u003c/em\u003e\u003cstrong\u003e20\u003c/strong\u003e, 268 (2019).\u003c/li\u003e\n \u003cli\u003ePurcell, M., Kruger, A. \u0026amp; Tainsky, M.A. Gene expression profiling of replicative and induced senescence. \u003cem\u003eCell Cycle\u003c/em\u003e\u003cstrong\u003e13\u003c/strong\u003e, 3927-37 (2014).\u003c/li\u003e\n \u003cli\u003eSati, S. et al. 4D Genome Rewiring during Oncogene-Induced and Replicative Senescence. \u003cem\u003eMol Cell\u003c/em\u003e\u003cstrong\u003e78\u003c/strong\u003e, 522-538.e9 (2020).\u003c/li\u003e\n \u003cli\u003eBorghesan, M., Hoogaars, W., Varela-Eirin, M., Talma, N. \u0026amp; Demaria, M. A Senescence-Centric View of Aging: Implications for Longevity and Disease. \u003cem\u003eTrends Cell Biol\u003c/em\u003e\u003cstrong\u003e30\u003c/strong\u003e, 777-791 (2020).\u003c/li\u003e\n \u003cli\u003eAarts, M. et al. Coupling shRNA screens with single-cell RNA-seq identifies a dual role for mTOR in reprogramming-induced senescence. \u003cem\u003eGenes Dev\u003c/em\u003e\u003cstrong\u003e31\u003c/strong\u003e, 2085-2098 (2017).\u003c/li\u003e\n \u003cli\u003eTang, H. et al. Single senescent cell sequencing reveals heterogeneity in senescent cells induced by telomere erosion. \u003cem\u003eProtein Cell\u003c/em\u003e\u003cstrong\u003e10\u003c/strong\u003e, 370-375 (2019).\u003c/li\u003e\n \u003cli\u003eTeo, Y.V. et al. Notch Signaling Mediates Secondary Senescence. \u003cem\u003eCell Rep\u003c/em\u003e\u003cstrong\u003e27\u003c/strong\u003e, 997-1007.e5 (2019).\u003c/li\u003e\n \u003cli\u003eMinamino, T. et al. A crucial role for adipose tissue p53 in the regulation of insulin resistance. \u003cem\u003eNat Med\u003c/em\u003e\u003cstrong\u003e15\u003c/strong\u003e, 1082-7 (2009).\u003c/li\u003e\n \u003cli\u003eWiley, C.D. et al. Analysis of individual cells identifies cell-to-cell variability following induction of cellular senescence. \u003cem\u003eAging Cell\u003c/em\u003e\u003cstrong\u003e16\u003c/strong\u003e, 1043-1050 (2017).\u003c/li\u003e\n \u003cli\u003eOrjalo, A.V., Bhaumik, D., Gengler, B.K., Scott, G.K. \u0026amp; Campisi, J. Cell surface-bound IL-1alpha is an upstream regulator of the senescence-associated IL-6/IL-8 cytokine network. \u003cem\u003eProc Natl Acad Sci U S A\u003c/em\u003e\u003cstrong\u003e106\u003c/strong\u003e, 17031-6 (2009).\u003c/li\u003e\n \u003cli\u003eDiril, M.K. et al. Cyclin-dependent kinase 1 (Cdk1) is essential for cell division and suppression of DNA re-replication but not for liver regeneration. \u003cem\u003eProc Natl Acad Sci U S A\u003c/em\u003e\u003cstrong\u003e109\u003c/strong\u003e, 3826-31 (2012).\u003c/li\u003e\n \u003cli\u003eAlessio, N. et al. Different Stages of Quiescence, Senescence, and Cell Stress Identified by Molecular Algorithm Based on the Expression of Ki67, RPS6, and Beta-Galactosidase Activity. \u003cem\u003eInt J Mol Sci\u003c/em\u003e\u003cstrong\u003e22\u003c/strong\u003e (2021).\u003c/li\u003e\n \u003cli\u003eKansara, M. et al. Immune response to RB1-regulated senescence limits radiation-induced osteosarcoma formation. \u003cem\u003eJ Clin Invest\u003c/em\u003e\u003cstrong\u003e123\u003c/strong\u003e, 5351-60 (2013).\u003c/li\u003e\n \u003cli\u003eMcConnell, B.B., Starborg, M., Brookes, S. \u0026amp; Peters, G. Inhibitors of cyclin-dependent kinases induce features of replicative senescence in early passage human diploid fibroblasts. \u003cem\u003eCurr Biol\u003c/em\u003e\u003cstrong\u003e8\u003c/strong\u003e, 351-4 (1998).\u003c/li\u003e\n \u003cli\u003eLiu, S. et al. Senescence of human skin-derived precursors regulated by Akt-FOXO3-p27(KIP(1))/p15(INK(4)b) signaling. \u003cem\u003eCell Mol Life Sci\u003c/em\u003e\u003cstrong\u003e72\u003c/strong\u003e, 2949-60 (2015).\u003c/li\u003e\n \u003cli\u003eLee, B.Y. et al. Senescence-associated beta-galactosidase is lysosomal beta-galactosidase. \u003cem\u003eAging Cell\u003c/em\u003e\u003cstrong\u003e5\u003c/strong\u003e, 187-95 (2006).\u003c/li\u003e\n \u003cli\u003eTacutu, R. et al. Human Ageing Genomic Resources: new and updated databases. \u003cem\u003eNucleic Acids Res\u003c/em\u003e\u003cstrong\u003e46\u003c/strong\u003e, D1083-D1090 (2018).\u003c/li\u003e\n \u003cli\u003eSaul, D. \u0026amp; Kosinsky, R.L. Single-Cell Transcriptomics Reveals the Expression of Aging- and Senescence-Associated Genes in Distinct Cancer Cell Populations. \u003cem\u003eCells\u003c/em\u003e\u003cstrong\u003e10\u003c/strong\u003e (2021).\u003c/li\u003e\n \u003cli\u003eAging Atlas: a multi-omics database for aging biology. \u003cem\u003eNucleic Acids Res\u003c/em\u003e\u003cstrong\u003e49\u003c/strong\u003e, D825-D830 (2021).\u003c/li\u003e\n \u003cli\u003eChatsirisupachai, K., Palmer, D., Ferreira, S. \u0026amp; de Magalhaes, J.P. A human tissue-specific transcriptomic analysis reveals a complex relationship between aging, cancer, and cellular senescence. \u003cem\u003eAging Cell\u003c/em\u003e\u003cstrong\u003e18\u003c/strong\u003e, e13041 (2019).\u003c/li\u003e\n \u003cli\u003eReyfman, P.A. et al. Single-Cell Transcriptomic Analysis of Human Lung Provides Insights into the Pathobiology of Pulmonary Fibrosis. \u003cem\u003eAm J Respir Crit Care Med\u003c/em\u003e\u003cstrong\u003e199\u003c/strong\u003e, 1517-1536 (2019).\u003c/li\u003e\n \u003cli\u003eNehme, J., Borghesan, M., Mackedenski, S., Bird, T.G. \u0026amp; Demaria, M. Cellular senescence as a potential mediator of COVID-19 severity in the elderly. \u003cem\u003eAging Cell\u003c/em\u003e\u003cstrong\u003e19\u003c/strong\u003e, e13237 (2020).\u003c/li\u003e\n \u003cli\u003eLipskaia, L. et al. Evidence That SARS-CoV-2 Induces Lung Cell Senescence: Potential Impact on COVID-19 Lung Disease. \u003cem\u003eAm J Respir Cell Mol Biol\u003c/em\u003e\u003cstrong\u003e66\u003c/strong\u003e, 107-111 (2022).\u003c/li\u003e\n \u003cli\u003eMelms, J.C. et al. A molecular single-cell lung atlas of lethal COVID-19. \u003cem\u003eNature\u003c/em\u003e\u003cstrong\u003e595\u003c/strong\u003e, 114-119 (2021).\u003c/li\u003e\n \u003cli\u003eBartleson, J.M. et al. SARS-CoV-2, COVID-19 and the aging immune system. \u003cem\u003eNature Aging\u003c/em\u003e\u003cstrong\u003e1\u003c/strong\u003e, 769-782 (2021).\u003c/li\u003e\n \u003cli\u003eLee, S. et al. Virus-induced senescence is a driver and therapeutic target in COVID-19. \u003cem\u003eNature\u003c/em\u003e\u003cstrong\u003e599\u003c/strong\u003e, 283-289 (2021).\u003c/li\u003e\n \u003cli\u003eCamell, C.D. et al. Senolytics reduce coronavirus-related mortality in old mice. \u003cem\u003eScience\u003c/em\u003e\u003cstrong\u003e373\u003c/strong\u003e (2021).\u003c/li\u003e\n \u003cli\u003eLi, S. et al. Cellular metabolic basis of altered immunity in the lungs of patients with COVID-19. \u003cem\u003eMed Microbiol Immunol\u003c/em\u003e\u003cstrong\u003e211\u003c/strong\u003e, 49-69 (2022).\u003c/li\u003e\n \u003cli\u003eD\u0026apos;Agnillo, F. et al. Lung epithelial and endothelial damage, loss of tissue repair, inhibition of fibrinolysis, and cellular senescence in fatal COVID-19. \u003cem\u003eSci Transl Med\u003c/em\u003e\u003cstrong\u003e13\u003c/strong\u003e, eabj7790 (2021).\u003c/li\u003e\n \u003cli\u003eParimon, T. et al. Potential mechanisms for lung fibrosis associated with COVID-19 infection. \u003cem\u003eQJM\u003c/em\u003e\u003cstrong\u003e116\u003c/strong\u003e, 487-492 (2023).\u003c/li\u003e\n \u003cli\u003eEvangelou, K. et al. Pulmonary infection by SARS-CoV-2 induces senescence accompanied by an inflammatory phenotype in severe COVID-19: possible implications for viral mutagenesis. \u003cem\u003eEur Respir J\u003c/em\u003e\u003cstrong\u003e60\u003c/strong\u003e (2022).\u003c/li\u003e\n \u003cli\u003eChen, J., Wu, H., Yu, Y. \u0026amp; Tang, N. Pulmonary alveolar regeneration in adult COVID-19 patients. \u003cem\u003eCell Res\u003c/em\u003e\u003cstrong\u003e30\u003c/strong\u003e, 708-710 (2020).\u003c/li\u003e\n \u003cli\u003eLiao, M. et al. Single-cell landscape of bronchoalveolar immune cells in patients with COVID-19. \u003cem\u003eNat Med\u003c/em\u003e\u003cstrong\u003e26\u003c/strong\u003e, 842-844 (2020).\u003c/li\u003e\n \u003cli\u003eMerad, M. \u0026amp; Martin, J.C. Pathological inflammation in patients with COVID-19: a key role for monocytes and macrophages. \u003cem\u003eNat Rev Immunol\u003c/em\u003e\u003cstrong\u003e20\u003c/strong\u003e, 355-362 (2020).\u003c/li\u003e\n \u003cli\u003eGarcia-Nicolas, O., Godel, A., Zimmer, G. \u0026amp; Summerfield, A. Macrophage phagocytosis of SARS-CoV-2-infected cells mediates potent plasmacytoid dendritic cell activation. \u003cem\u003eCell Mol Immunol\u003c/em\u003e\u003cstrong\u003e20\u003c/strong\u003e, 835-849 (2023).\u003c/li\u003e\n \u003cli\u003eBurton, D. \u0026amp; Stolzing, A. Cellular senescence: Immunosurveillance and future immunotherapy. \u003cem\u003eAgeing Res Rev\u003c/em\u003e\u003cstrong\u003e43\u003c/strong\u003e, 17-25 (2018).\u003c/li\u003e\n \u003cli\u003eLo, J.A. \u0026amp; Fisher, D.E. The melanoma revolution: from UV carcinogenesis to a new era in therapeutics. \u003cem\u003eScience\u003c/em\u003e\u003cstrong\u003e346\u003c/strong\u003e, 945-9 (2014).\u003c/li\u003e\n \u003cli\u003eRobert, C. et al. Nivolumab in previously untreated melanoma without BRAF mutation. \u003cem\u003eN Engl J Med\u003c/em\u003e\u003cstrong\u003e372\u003c/strong\u003e, 320-30 (2015).\u003c/li\u003e\n \u003cli\u003eHoenicke, L. \u0026amp; Zender, L. Immune surveillance of senescent cells--biological significance in cancer- and non-cancer pathologies. \u003cem\u003eCarcinogenesis\u003c/em\u003e\u003cstrong\u003e33\u003c/strong\u003e, 1123-6 (2012).\u003c/li\u003e\n \u003cli\u003eKang, T.W. et al. Senescence surveillance of pre-malignant hepatocytes limits liver cancer development. \u003cem\u003eNature\u003c/em\u003e\u003cstrong\u003e479\u003c/strong\u003e, 547-51 (2011).\u003c/li\u003e\n \u003cli\u003eXue, W. et al. Senescence and tumour clearance is triggered by p53 restoration in murine liver carcinomas. \u003cem\u003eNature\u003c/em\u003e\u003cstrong\u003e445\u003c/strong\u003e, 656-60 (2007).\u003c/li\u003e\n \u003cli\u003eRuscetti, M. et al. NK cell-mediated cytotoxicity contributes to tumor control by a cytostatic drug combination. \u003cem\u003eScience\u003c/em\u003e\u003cstrong\u003e362\u003c/strong\u003e, 1416-1422 (2018).\u003c/li\u003e\n \u003cli\u003eTirosh, I. et al. Dissecting the multicellular ecosystem of metastatic melanoma by single-cell RNA-seq. \u003cem\u003eScience\u003c/em\u003e\u003cstrong\u003e352\u003c/strong\u003e, 189-96 (2016).\u003c/li\u003e\n \u003cli\u003eTeschendorff, A.E., Breeze, C.E., Zheng, S.C. \u0026amp; Beck, S. A comparison of reference-based algorithms for correcting cell-type heterogeneity in Epigenome-Wide Association Studies. \u003cem\u003eBMC Bioinformatics\u003c/em\u003e\u003cstrong\u003e18\u003c/strong\u003e, 105 (2017).\u003c/li\u003e\n \u003cli\u003eNewman, A.M. et al. Robust enumeration of cell subsets from tissue expression profiles. \u003cem\u003eNat Methods\u003c/em\u003e\u003cstrong\u003e12\u003c/strong\u003e, 453-7 (2015).\u003c/li\u003e\n \u003cli\u003eCoppe, J.P. et al. Senescence-associated secretory phenotypes reveal cell-nonautonomous functions of oncogenic RAS and the p53 tumor suppressor. \u003cem\u003ePLoS Biol\u003c/em\u003e\u003cstrong\u003e6\u003c/strong\u003e, 2853-68 (2008).\u003c/li\u003e\n \u003cli\u003eAcosta, J.C. et al. Chemokine signaling via the CXCR2 receptor reinforces senescence. \u003cem\u003eCell\u003c/em\u003e\u003cstrong\u003e133\u003c/strong\u003e, 1006-18 (2008).\u003c/li\u003e\n \u003cli\u003eKuilman, T. et al. Oncogene-induced senescence relayed by an interleukin-dependent inflammatory network. \u003cem\u003eCell\u003c/em\u003e\u003cstrong\u003e133\u003c/strong\u003e, 1019-31 (2008).\u003c/li\u003e\n \u003cli\u003eBiran, A. et al. Senescent cells communicate via intercellular protein transfer. \u003cem\u003eGenes Dev\u003c/em\u003e\u003cstrong\u003e29\u003c/strong\u003e, 791-802 (2015).\u003c/li\u003e\n \u003cli\u003eJin, S. et al. Inference and analysis of cell-cell communication using CellChat. \u003cem\u003eNat Commun\u003c/em\u003e\u003cstrong\u003e12\u003c/strong\u003e, 1088 (2021).\u003c/li\u003e\n \u003cli\u003eSenturk, S. et al. Transforming growth factor-beta induces senescence in hepatocellular carcinoma cells and inhibits tumor growth. \u003cem\u003eHepatology\u003c/em\u003e\u003cstrong\u003e52\u003c/strong\u003e, 966-74 (2010).\u003c/li\u003e\n \u003cli\u003eReimann, M. et al. Tumor stroma-derived TGF-beta limits myc-driven lymphomagenesis via Suv39h1-dependent senescence. \u003cem\u003eCancer Cell\u003c/em\u003e\u003cstrong\u003e17\u003c/strong\u003e, 262-72 (2010).\u003c/li\u003e\n \u003cli\u003eBuckley, S. et al. BMP4 signaling induces senescence and modulates the oncogenic phenotype of A549 lung adenocarcinoma cells. \u003cem\u003eAm J Physiol Lung Cell Mol Physiol\u003c/em\u003e\u003cstrong\u003e286\u003c/strong\u003e, L81-6 (2004).\u003c/li\u003e\n \u003cli\u003eZhu, D., Wu, J., Spee, C., Ryan, S.J. \u0026amp; Hinton, D.R. BMP4 mediates oxidative stress-induced retinal pigment epithelial cell senescence and is overexpressed in age-related macular degeneration. \u003cem\u003eJ Biol Chem\u003c/em\u003e\u003cstrong\u003e284\u003c/strong\u003e, 9529-39 (2009).\u003c/li\u003e\n \u003cli\u003eKorbecki, J. et al. in International Journal of Molecular Sciences (2020).\u003c/li\u003e\n \u003cli\u003ePrice, M.A. et al. CSPG4, a potential therapeutic target, facilitates malignant progression of melanoma. \u003cem\u003ePigment Cell Melanoma Res\u003c/em\u003e\u003cstrong\u003e24\u003c/strong\u003e, 1148-57 (2011).\u003c/li\u003e\n \u003cli\u003eGurrea-Rubio, M. \u0026amp; Fox, D.A. The dual role of CD6 as a therapeutic target in cancer and autoimmune disease. \u003cem\u003eFront Med (Lausanne)\u003c/em\u003e\u003cstrong\u003e9\u003c/strong\u003e, 1026521 (2022).\u003c/li\u003e\n \u003cli\u003eZhang, C. et al. STAT3 Activation-Induced Fatty Acid Oxidation in CD8(+) T Effector Cells Is Critical for Obesity-Promoted Breast Tumor Growth. \u003cem\u003eCell Metab\u003c/em\u003e\u003cstrong\u003e31\u003c/strong\u003e, 148-161.e5 (2020).\u003c/li\u003e\n \u003cli\u003eMalta, T.M. et al. Machine Learning Identifies Stemness Features Associated with Oncogenic Dedifferentiation. \u003cem\u003eCell\u003c/em\u003e\u003cstrong\u003e173\u003c/strong\u003e, 338-354.e15 (2018).\u003c/li\u003e\n \u003cli\u003eGennady, K., Vladimir, S. \u0026amp; Alexey, S. Fast gene set enrichment analysis. \u003cem\u003ebioRxiv\u003c/em\u003e, 060012 (2019).\u003c/li\u003e\n \u003cli\u003eAgarwal, S., Graepel, T., Herbrich, R., Har-Peled, S. \u0026amp; Roth, D. Generalization Bounds for the Area Under the ROC Curve. \u003cem\u003eJournal of Machine Learning Research\u003c/em\u003e\u003cstrong\u003e6\u003c/strong\u003e, 393--425 (2005).\u003c/li\u003e\n \u003cli\u003eTeschendorff, A.E. \u0026amp; Enver, T. Single-cell entropy for accurate estimation of differentiation potency from a cell\u0026apos;s transcriptome. \u003cem\u003eNat Commun\u003c/em\u003e\u003cstrong\u003e8\u003c/strong\u003e, 15599 (2017).\u003c/li\u003e\n \u003cli\u003eYeung, K.Y., Fraley, C., Murua, A., Raftery, A.E. \u0026amp; Ruzzo, W.L. Model-based clustering and data transformations for gene expression data. \u003cem\u003eBioinformatics\u003c/em\u003e\u003cstrong\u003e17\u003c/strong\u003e, 977-87 (2001).\u003c/li\u003e\n \u003cli\u003eFang, Z., Liu, X. \u0026amp; Peltz, G. GSEApy: a comprehensive package for performing gene set enrichment analysis in Python. \u003cem\u003eBioinformatics\u003c/em\u003e\u003cstrong\u003e39\u003c/strong\u003e (2023).\u003c/li\u003e\n \u003cli\u003eWang, W. et al. Independent component analysis based gene co-expression network inference (ICAnet) to decipher functional modules for better single-cell clustering and batch integration. \u003cem\u003eNucleic Acids Res\u003c/em\u003e\u003cstrong\u003e49\u003c/strong\u003e, e54 (2021).\u003c/li\u003e\n \u003cli\u003eHaghverdi, L., Buettner, F. \u0026amp; Theis, F.J. Diffusion maps for high-dimensional single-cell analysis of differentiation data. \u003cem\u003eBioinformatics\u003c/em\u003e\u003cstrong\u003e31\u003c/strong\u003e, 2989-98 (2015).\u003c/li\u003e\n \u003cli\u003eSubramanian, A. et al. Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. \u003cem\u003eProc Natl Acad Sci U S A\u003c/em\u003e\u003cstrong\u003e102\u003c/strong\u003e, 15545-50 (2005).\u003c/li\u003e\n \u003cli\u003eColaprico, A. et al. TCGAbiolinks: an R/Bioconductor package for integrative analysis of TCGA data. \u003cem\u003eNucleic Acids Res\u003c/em\u003e\u003cstrong\u003e44\u003c/strong\u003e, e71 (2016).\u003c/li\u003e\n \u003cli\u003eCalcinotto, A. et al. Cellular Senescence: Aging, Cancer, and Injury. \u003cem\u003ePhysiol Rev\u003c/em\u003e\u003cstrong\u003e99\u003c/strong\u003e, 1047-1078 (2019).\u003c/li\u003e\n \u003cli\u003eGorgoulis, V. et al. Cellular Senescence: Defining a Path Forward. \u003cem\u003eCell\u003c/em\u003e\u003cstrong\u003e179\u003c/strong\u003e, 813-827 (2019).\u003c/li\u003e\n \u003cli\u003eNighat, N., Zhenqing, Y., Yidong, C., Xiaojing, W. \u0026amp; Siyuan, Z. Benchmarking supervised signature-scoring methods for single-cell RNA sequencing data in cancer. \u003cem\u003ebioRxiv\u003c/em\u003e, 2021.06.29.450404 (2021).\u003c/li\u003e\n \u003cli\u003eZhang, Y., Alexander, P.B. \u0026amp; Wang, X.F. TGF-beta Family Signaling in the Control of Cell Proliferation and Survival. \u003cem\u003eCold Spring Harb Perspect Biol\u003c/em\u003e\u003cstrong\u003e9\u003c/strong\u003e (2017).\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":false,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"nature-portfolio","isNatureJournal":true,"hasQc":false,"allowDirectSubmit":false,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"","title":"Nature Portfolio","twitterHandle":"","acdcEnabled":false,"dfaEnabled":false,"editorialSystem":"ejp","reportingPortfolio":"","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"Cellular senescence, Quantification, Machine learning, hUSI","lastPublishedDoi":"10.21203/rs.3.rs-3920908/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-3920908/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eDespite the manifestation and contribution of cellular senescence to tissue aging and aging-related disease, the identification of \u003cem\u003ein vivo \u003c/em\u003esenescent cells and the recognition of senescence-specific communication still remain challenging. Current senescence evaluation methods rely greatly on expression level of well-known senescence markers, enrichment of aging-related gene sets or weighted sum of curated genes. However, focusing on limited senescence aspects, these methods could not adequately capture the comprehensive senescence features. To evaluate senescence in a more general and unbiased way from the most common and easily accessible transcriptome data, we developed human universal senescence index (hUSI) to quantify human cellular senescence based on a series of weighted genes learned from representative senescence RNA-seq profiles using a machine learning algorithm. hUSI demonstrated its superior performance in distinguishing senescent samples under various conditions and robustness in handling batch effects and sparse profiles. hUSI could uncover the accumulation of senescent cells of various cell types in complex pathological conditions, and reflected the increasing senescence burden of patients and provided potential senotherapeutic targets. Furthermore, combined with gaussian mixture model, hUSI successfully inferred senescent tumor cells in melanoma and identified key target signaling pathways that are beneficial for patient prognosis. Overall, hUSI provides a valuable choice to improve our ability in characterizing cellular senescence under various conditions, illustrating promising implications in aging studies and clinical situations.\u003c/p\u003e","manuscriptTitle":"Robust senescence evaluation by transcriptome-based hUSI to facilitate characterizing cellular senescence under various conditions","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-02-13 18:53:07","doi":"10.21203/rs.3.rs-3920908/v1","editorialEvents":[],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"nature-aging","isNatureJournal":true,"hasQc":false,"allowDirectSubmit":false,"externalIdentity":"nataging","sideBox":"Learn more about [Nature Aging](https://www.nature.com/nataging/)","snPcode":"","submissionUrl":"","title":"Nature Aging","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"ejp","reportingPortfolio":"Nature Research","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"3425a1e4-57a1-4d9b-acb5-56754ca2cdd2","owner":[],"postedDate":"February 13th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[{"id":28724269,"name":"Biological sciences/Cell biology/Senescence"},{"id":28724270,"name":"Biological sciences/Computational biology and bioinformatics/Computational models"}],"tags":[],"updatedAt":"2025-05-30T07:07:01+00:00","versionOfRecord":{"articleIdentity":"rs-3920908","link":"https://doi.org/10.1038/s43587-025-00886-2","journal":{"identity":"nature-aging","isVorOnly":false,"title":"Nature Aging"},"publishedOn":"2025-05-29 04:00:00","publishedOnDateReadable":"May 29th, 2025"},"versionCreatedAt":"2024-02-13 18:53:07","video":"","vorDoi":"10.1038/s43587-025-00886-2","vorDoiUrl":"https://doi.org/10.1038/s43587-025-00886-2","workflowStages":[]},"version":"v1","identity":"rs-3920908","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-3920908","identity":"rs-3920908","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}