Senescence-directed nanotherapy ameliorates fibrosis and overcomes immune exclusion in cancer

Abstract

Fibrotic remodeling of tissues and tumors establishes immune-suppressive microenvironments that drive organ dysfunction and, in cancer, limit responses to immunotherapy. Cells exhibiting features of cellular senescence are conserved drivers of fibrotic remodeling and thus represent therapeutic targets, yet senescent states are heterogeneous and can exert both beneficial and pathogenic effects, complicating therapeutic intervention. Here, we show that P-selectin is selectively expressed by subsets of senescent-like cells in fibrotic tissues and fibrotic tumor microenvironments. Leveraging fucoidan-based nanoparticles that bind P-selectin, we develop senescence-modulating nanoparticles (SMNPs) to selectively target these disease-associated cell states. SMNPs exhibit potent antifibrotic and immunomodulatory activity while markedly improving therapeutic index. Mechanistically, we identify a pathogenic, immune-suppressive macrophage population as a principal functional target of SMNPs in vivo. In fibrotic tumors, niche remodeling restores immune infiltration and sensitizes tumors to immune checkpoint–based therapies. More broadly, SMNPs establish a generalizable nanotherapeutic strategy for selectively targeting pathogenic senescent cell subsets across fibrotic disease and cancer.

Introduction

Chronic inflammation of the liver and lung frequently culminates in fibrosis, progressive organ dysfunction, and premature mortality, yet disease-modifying therapies remain limited (1–4). Fibrosis is marked by the accumulation of activated stromal and immune cells, excessive extracellular matrix deposition, vascular remodeling, and distortion of tissue architecture, collectively reshaping local cellular ecosystems (5–7). Additionally, fibrotic environments are potently immune suppressive, form barriers to immune surveillance that constrain host defense and limit the efficacy of immunotherapies in cancer (8–11). These observations imply that chronic inflammation, fibrosis, immune dysfunction, and cancer progression may reflect convergent biological programs rather than distinct pathological entities. Accordingly, identifying and targeting the cellular states that sustain fibrotic, immune-suppressive niches represents a shared challenge across chronic inflammatory disease and oncology.

One cellular program implicated in the fibrotic progression is cellular senescence, a stress response engaged by diverse forms of cellular and tissue injury (12–15). Senescence is associated with stable proliferative arrest and broad rewiring of cell physiology, including activation of tissue remodeling programs that alter how cells signal and sense their microenvironment (16, 17). While the presence of cells exhibiting senescence features can support tissue repair following acute injury, their persistence in chronically damaged tissues sustains inflammation, matrix remodeling, and architectural distortion, thereby accelerating fibrotic progression and organ decline (16, 18–20). In cancer, senescent-like cells accumulate within the tumor microenvironment (TME), where they are linked to stromal remodeling, altered myeloid cell composition, and immune suppression (21–28). Together, these observations position senescent cells as conserved driver of fibrotic remodeling, immune dysfunction, and tissue decline across organs and disease contexts.

Pharmacologic strategies to eliminate or modulate senescent cells show promise in preclinical models (29–34), yet translation has been limited by two fundamental challenges. First, while senescence was defined in vitro, senescence-like states in tissues are heterogeneous, varying across cell types, tissue, and association with distinct functional consequences (14, 35). Second, existing senotherapeutics lack cellular precision, leading to predictable on-target toxicity in normal or reparative cell populations (29, 33, 36–38). These limitations underscore the need for strategies that selectively target pathogenic senescent states while sparing beneficial counterparts, a goal that requires improved resolution of senescent-state heterogeneity in vivo and its relationship to tissue pathology.

Motivated by our finding that P-selectin (SELP) is selectively upregulated on cells with senescent features that accumulate in fibrotic tissues and drive pathological remodeling, we reasoned that this molecule could provide a tractable entry point for selective therapeutic targeting. P-selectin is induced on activated endothelium in inflammatory and fibrotic settings, where it mediates leukocyte recruitment and transcytosis via PSGL-1 and related glycosylated ligands (39–44). Nanoparticles bearing potent, high-affinity P-selectin ligands can exploit this pathway to concentrate therapeutic cargo within inflamed tissues and tumors (45–47). We therefore hypothesized that P-selectin could function both as a conduit for tissue-selective drug delivery and a molecular handle to preferentially access senescent-like, pathology-driving cell states within fibrotic niches. In this study, we define the landscape of P-selectin expression across senescent-like states in fibrotic tissues and evaluate the therapeutic potential of P-selectin–directed senescence-modulatory nanoparticles (SMNPs) in preclinical models of fibrosis and cancer.

Results

P-selectin marks senescent-like cells in fibrotic tissues

Given emerging links among P-selectin biology, senescence, and fibrosis (41–44), together with the potential to exploit P-selectin-dependent pathways for targeted drug delivery, we sought to systematically define the relationship between P-selectin expression and senescent cell states in fibrotic disease. We analyzed 58 human liver and lung fibrosis specimens, alongside healthy controls, using conventional immunohistochemistry and multiplexed immunofluorescence (mIF) panels incorporating canonical senescence markers. Across both organs, P-selectin expression was significantly elevated in fibrotic tissues compared to non-fibrotic counterparts (Fig. 1A–C and Fig. S1 A–B). Within these tissues, a subset of P-selectin-positive cells exhibited multiple hallmark features of cellular senescence, including expression of the cell-cycle inhibitor p21, absence of the proliferation marker Ki67, and reduced expression of the nuclear lamina protein LaminB1 (Fig. 1A–C and Fig. S1C). We refer to such states as “senescent-like” to acknowledge the heterogeneity of senescence programs and their incompletely defined nature in vivo (48, 49).

Figure 1: Senescent-like cells in fibrotic tissues that express P-selectin and are selectively targeted by fucoidan nanoparticles.

(A-C) Multiplex immunofluorescence (mIF) analysis of tissue microarrays of human liver (heathy, n=21; fibrotic/cirrhotic, n=30 cases) and lung (heathy, n=21; fibrotic, n=28 cases). The figure shows representative images of fibrotic organs (A) and quantification of SMA⁺ activated fibroblasts and P-selectin⁺ senescent cells (defined as p21⁺, Ki67⁻, LaminB1^low) (B-C). Scale bars, 500 μm. (D-G) Collagen deposition (Sirius Red), senescence-associated-B-galactosidase activity (X-Gal), SMA, and P-selectin expression in senescent cells (defined as C12RG⁺) in murine models of healthy liver (oil treated), liver fibrosis (6 weeks after start of CCl₄ treatment) (D) and healthy lung (saline) and lung fibrosis (28 days after bleomycin treatment) (E). The figure shows representative images (D,E) and quantification in control and fibrotic liver (F) and lung (G). Data represent 13–15 mice per group from three independent experiments. Scale bars, 100 μm. (H) Pearson correlation analysis between the expression of P-selectin⁺ senescent cells and activated fibroblasts in human and murine fibrotic and healthy liver and lung tissues. (I) Near-infrared fluorescence imaging of indocyanine green-positive (ICG⁺) nanoparticle enrichment (FiNav and DexNav) across tissues in murine models of liver fibrosis (top) and lung fibrosis (middle) at 12h post nanoparticle injection. (Bottom) Quantification of ICG⁺ nanoparticle enrichment in each tissue (n= 3 mice/group; one experiment). Scale bars, 500 μm. (J) Representative fluorescent images of FiNav enrichment (ICG⁺) in P-selectin⁺ senescent cells (C12RG⁺) of murine fibrotic liver (top) and lung (bottom). Scale bars, 50 μm. (K) Signal intensity quantification of ICG in P-selectin⁺ senescent cells vs all other cells (n=3 mice per group, left) and receiver operating characteristic (ROC) curve of ICG specificity for P-selectin⁺ senescent cells (n=3 mice per group, right). Each dot represents one patient or one mouse (B-C; F-G, respectively) or one cell (K). Data are presented as mean ± s.e.m. Mann Whitney test used for (B-C, F-G, K), Welch’s t test for (F, G), Simple linear regression used for (H), one-way ANOVA with Sidak’s multiple comparisons test (I) and ROC analysis in (K).

We next asked whether the presence of P-selectin-positive senescent-like cells was conserved in murine fibrosis models, where senescent cells have been implicated as drivers of tissue pathology (49, 50). In CCl₄- and bleomycin-induced models of liver and lung fibrosis (51, 52), we observed increased collagen deposition and an increase in the fraction of cells expressing senescence-associated beta-galactosidase (SA-β-gal), a canonical senescence marker (53, 54) (Fig. 1D–G). Consistent with human tissues, P-selectin expression was markedly increased in fibrotic murine livers and lungs across both species and organs. P-selectin expression was enriched in regions densely staining for α-smooth muscle actin (α-SMA), a marker of activated fibroblasts involved in extracellular matrix (ECM) deposition (55, 56) (Fig. 1D–G). Notably, a subset of P-selectin-positive cells also exhibited SA-β-gal activity, detected using the fluorescent substrate C12FDG (Fig. 1D–F). Collectively, these data establish that P-selectin marks a subset of senescent-like cells that accumulate in fibrotic environments, revealing a conserved association between P-selectin expression, senescence-associated features, and fibrosis across organs and species (Fig. 1H).

SMNPs selectively target senescent-like cells in fibrotic organs

Having established that P-selectin is upregulated on senescent-like cells in fibrotic tissues, we next asked whether this feature could be exploited for targeted drug delivery. Leveraging our previously described fucoidan (Fi)-based, P-selectin–targeting nanoparticles (47), we engineered nanoparticles harboring agents known to impact senescence-associated phenotypes; we term these senescence-modulating nanoparticles (SMNPs). One SMNP type encapsulated the senolytic agent Navitoclax, which targets BCL-2 family anti-apoptotic proteins induced in senescent states (57, 58), generating FiNav (59). The second SMNP type encapsulated the senomorphic agent dBET6, a BRD4-targeting proteolysis targeting chimera (PROTAC) that disrupts enhancer function and suppresses expression of SASP genes, generating FidBET6 (46). Nanoparticles were synthesized by nanoprecipitation together with the near-infrared dye indocyanine green (ICG) for particle stabilization and in vivo tracing (Fig. S1D). Control nanoparticles were produced using dextran (Dex), which is biochemically similar to fucoidan but does not bind P-selectin (DexNav and DexdBET6) (47). All SMNP and Dex nanoparticle formulations exhibited uniform size distributions, similar hydrodynamic diameters, and similar surface charge properties (Fig. S1E–G).

To assess biodistribution in vivo, SMNPs or control nanoparticles were administered intravenously to mice with established liver or lung fibrosis, and ICG signal was quantified in 6 vital organs at 12 or 24 hours post-treatment. In the CCl₄–induced liver fibrosis model, FiNav and FidBET6 nanoparticles preferentially accumulated in the liver, whereas in the bleomycin-induced lung fibrosis model, SMNPs selectively enriched in the lung (Fig. 1I and Fig. S1H, I). By contrast, control DexNav and DexdBET6, which do not bind P-selectin, showed no organ-specific accumulation (Fig. 1I and Fig. S1H, I). At the cellular level, FiNav and FidBET6 nanoparticles were enriched within P-selectin+/SA-β-gal+ cells (Fig. 1J, K and Fig. S1J). Together, these data establish P-selectin can be leveraged to selectively deliver senescence-modulating payloads to senescent cells in fibrotic organs, providing a foundation for evaluating therapeutic efficacy and safety.

SMNPs reverse established fibrosis while limiting systemic toxicities

Prior work has shown that Navitoclax and BET inhibitors can attenuate fibrosis in preclinical models (15, 60, 61), but their clinical development in fibrosis and cancer has been impeded by dose-limiting hematologic and systemic toxicities (37, 38, 62–66). We therefore asked whether P-selectin-directed SMNPs could preserve the antifibrotic efficacy of these agents while mitigating their toxicity. To this end, we performed treatment studies in CCl₄- and bleomycin-induced models of liver and lung fibrosis, respectively (Fig. 2A–B). After fibrosis was established, mice were treated with vehicle, SMNPs (FiNav or FidBET6), control nanoparticles (DexNav or DexdBET6), matched doses of free Navitoclax or dBET6, or free fucoidan. In both organs, treatment efficacy was evaluated by changes in collagen deposition (via Sirius red staining) and abundance of α-SMA+ activated fibroblasts. In mice with liver fibrosis, resolution of liver damage was further assessed by measuring portal vein flow (67) and changes in liver enzymes (ALT and AST). Systemic toxicities were assessed by hematologic profiling, body-weight monitoring, and treatment-associated mortality.

Figure 2: FiNav nanoparticles selectively deplete senescent cells and attenuate fibrosis with minimal toxicity.

(A,B) Schematic of nanoparticle (NP) treatment schedules in murine fibrotic livers (A) and lungs (B); cNP: control nanoparticles. (C-J) Representative images of P-selectin⁺ senescent (SA-β-gal⁺ per C12RG assay) cells, collagen deposition (Sirius Red), and myofibroblast activation (SMA), in fibrotic livers (C) and lungs (D) treated as indicated, with corresponding quantifications (livers: E, G, H and lungs: F, I, J) (n= 4–6 mice/group). Scale bars, 150 μm. (K-L) Neutrophil (K) and platelet (L) counts from blood of mice with fibrotic livers (left) or lungs (right) treated with free navitoclax or FiNav nanoparticles (n= 5–6 mice/group). (M) Toxicity of free navitoclax and FiNav assessed by Kaplan-Meier survival curves for mice with fibrotic livers (left) or fibrotic lungs (right) treated with free navitoclax or FiNav nanoparticles. Each dot represents one mouse. Data are presented as mean ± s.e.m. One-way ANOVA and Sidak’s multiple comparisons test for (E-L) and log-rank Mantel Cox test (M).

Across both, liver and lung fibrosis models, FiNav treatment efficiently reduced the abundance of P-selectin+/SA-β-gal+ double-positive cells, whereas FidBET6 modestly reduced or did not alter this compartment (Fig. 2C–F and Fig. S2A–D). Despite these differences, both FiNav and FidBET6 significantly reduced fibrotic burden, as evidenced by decreased collagen deposition and reduced numbers of α-SMA+ activated fibroblasts (Fig. 2C–D, G–J and Fig. S2A–B, E–H). This effect was at least as great as that achieved with free drug, despite SMNPs targeting only a fraction of cells in vivo (Fig. 2G–H and Fig. S2E–H). Notably, at doses matched to the nanoparticle formulations, free fucoidan had no effect. Similarly, dextran-based control nanoparticle encapsulation abolished the activity of both agents and failed to ameliorate fibrosis in either organ. Together, these findings demonstrate that nanoencapsulation preserves therapeutic efficacy and highlights the requirement for P-selectin–dependent targeting to achieve therapeutic benefit.

We next asked whether histopathologic improvements translated into restoration of organ function. In mice with liver fibrosis, free fucoidan and dextran-based control nanoparticles had no measurable impact on portal vein flow or serum ALT and AST levels, which remained comparable to untreated control mice with liver fibrosis (Fig. S2I–K). By contrast, animals treated with P-selectin–directed SMNPs showed improvement of portal vein flow and reduced levels of ALT and AST, with FiNav producing a significant advantage when compared to free Navitoclax (Fig. S2I–K). These findings indicate that nanoencapsulation preserves antifibrotic efficacy and may confer additional functional benefits in restoring organ physiology.

Finally, we evaluated whether SMNP-mediated delivery mitigates the dose-limiting toxicities associated with Navitoclax and BRD4-targeting agents, including neutropenia and thrombocytopenia. In both liver and lung fibrosis models, FiNav and FidBET6 treatment eliminated these toxicities (Fig. 2K, L and Fig. S3A, B). Consistent with preserved platelet function, SMNP-treated mice exhibited significantly shorter bleeding times compared to free-drug–treated animals (Fig. S3C, D). Extended toxicity profiling further revealed that SMNPs produced with either Navitoclax or dBET6 protected from free drug-associated leukopenia, lymphopenia, body-weight loss, and treatment-associated mortality (Fig. 2M and Fig. S3 E–M). Therefore, SMNPs harboring either a small molecule senolytic or senomorphic drug can reverse established fibrosis in both liver and lung while markedly improving the therapeutic index.

SMNPs target a subset of senescent-like macrophages

To better define the cell types and states targeted by SMNPs in vivo, we performed single-cell RNA sequencing (scRNA-seq) on cells isolated from murine fibrotic liver and lung tissues. Owing to variability in cellular autofluorescence across cell types and the low fraction of SMNP target cells, fluorescence activated cell sorting (FACS) of cells based on their signal of the far-red (ICG) dye was not consistently achievable. Based on our observation that SMNPs preferentially accumulate in P-selectin/SA-β-gal double-positive cells (Fig. 1J–K and fig. S1J), we instead leveraged endogenous P-selectin surface expression in combination with a fluorescent β-galactosidase probe (SPiDER, functionally analogous to C12FDG (54, 68) to enrich for potential target populations. Following organ digestion, four distinct cells populations were isolated by FACS: P-selectin+/SPiDER+ (P⁺sen⁺), P-selectin+/SPiDER⁻ (P⁺sen⁻), P-selectin⁻/SPiDER+ (P⁻sen⁺), and P-selectin⁻/SPiDER⁻ (P⁻sen⁻). scRNA-seq analysis revealed that the P⁺sen⁺ compartment from both liver and lung was composed predominantly of macrophages (Adgre1⁺, Cd68⁺, Csf1r⁺) and endothelial cells (Pecam1⁺, Cdh5⁺, Icam1⁺), with smaller contributions from fibroblasts and epithelial cells in the fibrotic lung (Fig. 3A–D; Fig. S4B–G). The identities of P⁺sen⁺ macrophages and endothelial cells inferred by scRNA-seq were validated in situ using C12RG-based SA-β-gal staining combined with immunofluorescence for macrophage (CD68, F4/80) and endothelial (CD31) markers (Fig. 3E–F).

Figure 3: SMNPs target a P-selectin/SA-b-gal double-positive macrophage subpopulation in fibrotic liver and lungs.

(A-B) Visualization of unsupervised clustering in a UMAP of cells isolated from murine fibrotic (A) livers and (B) lungs. (C-D) Identification of P+sen+ cells (green) in A and B. (E-F) IF analysis of SA-β-gal (C12RG⁺) and P-selectin expression in CD31⁺ endothelial cells (EC), F4/80⁺ macrophages (MΦ) (liver), and CD68⁺ MΦ (lung) of murine fibrotic (E) livers and (F) lungs. Scale bars, 100 μm (G) Representative gating on LiveDead- CD45⁺ F4/80⁺ cells based on P-selectin expression and SA-b-gal activity (SPiDER). (H) BD FACSDiscover S8 Cell Sorter images of surface P-selectin expression (red) and SPiDER activity (green) in MΦ populations sorted in (G). (I-J) Ex-vivo killing assays of sorted P-selectin⁺ senescent (P+sen+) or P-selectin⁺ (P+sen−) ECs and P+sen+ and P-selectin-senescent− (P−sen−) MΦs from murine fibrotic (I) livers and (J) lungs treated with DexNav or FiNav nanoparticles. Representative data from three independent experiments. (K-L) IF images (left panels) and quantification (left panels) of nanoparticle enrichment (ICG), SPiDER activity, and Brd4 expression in sorted P+sen+ EC and MΦ from murine fibrotic livers (K) or lungs (I) following treatment with DexdBET6 or FidBET6 nanoparticles (left). Quantification of Brd4 expression upon DexdBET6 or FidBET6 treatment in sorted P+sen+ or P+sen− EC and P+sen+ and P−sen− MΦ from murine fibrotic (K) livers or (L) lungs (right) (three biological replicates). Scale bars, 5 μm. Individual dots each represent one cell. Data are presented as mean ± s.e.m. Mann Whitney test or Welch’s t were used at endpoint (18h) for (I,J). Mann Whitney test for (K,L).

To directly establish the SMNP-targeted populations, we assessed SMNP uptake and pharmacologic activity across senescent-like and non-senescent subsets of macrophage and endothelial lineages. Using a modified FACS strategy incorporating P-selectin surface expression, SPiDER-β-Gal, and lineage-specific markers, we isolated P⁺sen⁺, P⁺sen⁻, P⁻sen⁺, and P⁻sen⁻ subsets within macrophage and endothelial compartments from fibrotic liver and lung (Fig. 3G,H, Fig. S5A). Macrophages were defined as CD45+/F4/80+ cells in the liver and CD45+/CD64+ cells in the lung, whereas endothelial cells were defined as CD45⁻/CD31+ (Figure S5A). Purified populations (Fig. S5A,B) were subjected to short-term ex vivo treatment with Navitoclax-based or dBET6-based nanoparticles. Targeting was evaluated using Incucyte-based cell viability assays for cell preparations treated with Navitoclax-based nanoparticles or Brd4 immunofluorescence for those treated with dBET6-based nanoparticles.

These studies pinpointed a P-selectin+/SPiDER+ (P⁺sen⁺) macrophage (MΦ) population as the principal cellular target of SMNPs in both liver and lung fibrosis. Accordingly, FiNav selectively induced cell death in MΦP+sen+ while sparing all other macrophage subsets and sparing endothelial subsets irrespective of P-selectin or SPiDER status (Fig. 3I, J and fig. S5 C,D). Consistent with this selectivity, complementary analysis with FidBET6 revealed selective degradation of Brd4 protein within MΦP+sen+ cells (Fig. 3K, L and Fig. S5E, F). By contrast, neither P+sen+ nor P−sen− endothelial populations showed evidence of functional perturbation by either SMNP, a finding consistent with prior work showing that fucoidan-based nanoparticles entering endothelial cells undergo caveolin-dependent transcytosis (45–47). Extending these observations in vivo, FiNav treatment led to depletion of MΦP+sen+ populations in liver and lung within 36 hours, with minimal effects on the other populations analyzed (Fig. S5G–I). Notably, depletion of MΦP+sen+ populations preceded fibrosis resolution, supporting the idea that MΦP+sen+ depletion is a cause rather consequence of tissue remodeling (Fig. S5J).

SMNP-targeted macrophage populations are related to lipid-associated macrophages

The identification of a senescent-like macrophage population as a key driver of fibrosis is consistent with studies demonstrating that genetic ablation of p16-expressing macrophages reverses liver fibrosis in mice (35). Other studies implicate pathogenic roles for p16-positive macrophages in lung disease (28, 69), collectively supporting a conserved role for senescent-like macrophages in tissue pathology across organs. To better define the molecular identity and functional properties of this population, we performed differential gene expression analysis using MAST (70) comparing MΦP+sen+ vs MΦP⁻sen⁻ macrophages.

In the liver, MΦP+sen+ macrophages displayed a robust senescence-associated transcriptional program, including increased expression of Cdkn1a, Cdkn2a, and Jun, induction of SASP-associated factors (Cxcl1, Cxcl2, Il1a, Mmp27), and reduced expression of Lmnb1, and Ccnd2 (Fig. 4A). Further gene expression analysis revealed enrichment for pathways linked to p53 signaling, lysosomal function, and reactive oxygen species, with depletion of those linked to cell-cycle progression, E2F targets, and DNA repair (Fig. S6A) (71–73). In MΦP+sen+ macrophages isolated from fibrotic lungs we observed a largely concordant transcriptional profile, including upregulation of Cdkn1a, Il1a, Cxcl2 and Mmp12 and downregulation of Lmnb1, E2f2 and Ccnd3. (Fig. 4B). Pathway analysis revealed an enrichment of p53 signaling, lysosomal pathways, and suppression of apoptosis, cell-cycle regulation, and DNA repair (Fig. S6B).

Figure 4: P-selectin/SA–β-gal double-positive macrophages exhibit senescence and LAM-related programs and shape immunosuppressive niches.

(A-B) Volcano plot showing representative senescence-associated genes highly expressed in P+sen+ macrophages compared to P−sen− macrophages in (A) liver and (B) lung fibrosis. (C-H) Gene signature analysis of P+sen+ macrophages vs P−sen− macrophages. The UMAP plots show enrichment of signatures for P+sen+ macrophages (C-D), SenNet (E-F), and lipid-associated macrophages (G-H) in murine liver fibrosis (C, E, G) and lung fibrosis (D, F, H). (I) P+sen+ macrophage signature in pooled scRNA datasets of human liver and lung fibrosis. (J) Violin plot of P+sen+ macrophage signature score in human healthy and fibrotic liver and lungs. (K) Enrichment of two lipid-associated macrophages gene signatures in human liver and lung fibrosis. (L-M) (left) Representative images of multiplex analysis mapping different immune and stromal populations in murine fibrotic livers (K) and lungs (L) treated for four weeks with DexNav or FiNav nanoparticles. Scale bars, 250 μm. (right) Quantification of different immune and stomal populations. (N-O) Cellular neighborhood analysis measuring the distance between the the indicated immune and stromal populations in murine fibrotic livers (N) and lungs (O) treated for four weeks with DexNav or FiNav nanoparticles. Wilcoxon ranked sum test was used for (J).

Consistent with these features, MΦP⁺sen⁺ gene signatures strongly overlapped with the consensus senNET senescence signature (48) and with transcriptional signatures derived from p16+ pathogenic macrophages in fibrotic livers (35) (Fig. 4C–F, S6C,D). Irrespective of organ site, these cells were also enriched for gene signatures linked to lipid, fatty acid and cholesterol metabolism pathways, and depleted in antigen processing and presentation programs (Fig S6A,B). Concordantly, MΦP+sen cells genes involved in matrix remodeling (Cd36, Pdgfc, Il1a, Marco, Timp2, Cd5l, Cd163, Vcam1) and a plethora of immunomodulatory and immunosuppressive genes (Cd274, Mrc1, Gas6, Hmox1, Spp1, Cd38, and Il10), many of which are hallmark features of lipid-associated macrophages (LAMs) (74–76) (Fig. S6E,F). Consistently MΦP⁺sen⁺ transcriptional programs showed strong enrichment with published LAM signatures (Fig. 4G,H) (75, 77).

LAMs are considered key organizers of fibrotic and immunosuppressive niches, where they coordinate cytokine production, lipid handling, matrix remodeling, and immune cell exclusion across diverse inflammatory contexts (35, 74–76, 78, 79). Using mIF, we orthogonally validated LAM-associated proteins CD36, CD206 (encoded by Mrc1) and the immunoPD-L1 (encoded by Cd274) on MΦP+sen+ cells (defined by P-selectin+, p21+, macrophage markers CD68+ and/or F480+, and LaminB1− and Ki67−) (Fig. S6G,H).

To assess conservation in human disease, we analyzed publicly available scRNA-seq datasets from human liver (44 healthy and 5 cirrhotic patients across five studies) and lung (51 healthy and 66 fibrotic patients across four studies) (78). Scoring myeloid cells using a humanized MΦP+sen+ signature revealed a human myeloid subpopulation transcriptionally concordant to the murine state in fibrotic settings (Fig. 4I). This population accumulated in fibrotic human tissues and was enriched for LAM-associated gene expression programs (Fig. 4J,K). The convergence between gene expression programs in senescent-like and LAM implies shared pathogenic functions in fibrosis and immune suppression.

SMNPs remodel the immune landscape of fibrotic tissues

To define the spatial organization MΦP+sen+ populations within fibrotic tissues, we examined their local cellular neighborhoods in situ. mIF of fibrotic liver and lung, coupled with spatial neighborhood analysis using MΦP⁺sen⁺ cells as anchor points, revealed preferential localization adjacent to activated α-SMA⁺ fibroblasts and endothelial cells in both organs, defining a conserved pro-fibrotic stromal niche. In contrast, adaptive immune populations, including dendritic cells, conventional CD4⁺ and CD8⁺ T cells, and B cells, were largely excluded from these regions, consistent with localized immune suppression (Fig. S6I,J).

To determine how SMNP-mediated targeting impacts these niches, fibrotic livers and lungs were isolated from mice following treatment with SMNPs (FiNav or FidBET6) or control nanoparticles (DexNav or DexdBET6) and were analyzed by mIF for immune cell markers. SMNP treatment induced profound tissue remodeling associated with an overall reduction in macrophage content, shifting the residual macrophage composition from anti-inflammatory phenotypes (MHCII⁻CD206⁺ and MHCII⁻LYVE1⁺PD-L1⁺) toward proinflammatory/antigen-presenting cell (APC) states (MHCII⁺CD11c⁺ and MHCII⁺CD206⁺) (Fig. 4L–O and Fig. S7A–D). Concomitantly, SMNP treatment caused an influx of dendritic cells and B cells and expanded CD4⁺ and CD8⁺ T cells populations within regions once dominated by profibrotic stroma (Fig. 4L–O and Fig. S7A–D). Neither free drugs nor control nanoparticles produced these effects, indicating targeting a P-selective population was required (Fig. 4L–O; Fig. S7E–G). Consistent with predicted depletion of MΦP⁺sen⁺ cells, SMNP treatment reduced pro-fibrotic and immunosuppressive mediators (including amphiregulin, CCL2, CCL5, IL-1β, and GAS6) while inducing cytokines associated with adaptive immunity (including IL-2, IL-4, IL-7, and IFN-γ) (Fig. S7H–K). Together, these coordinated cellular and secretory changes indicate that SMNPs resolve fibrosis while reprogramming immunosuppressive niches toward immune-permissive microenvironments.

SMNPs unlock immunotherapy responses in solid tumors

Many solid tumors that arise within, or actively induce, a fibrotic stroma develop immune-excluded microenvironments that limit T cell infiltration and confer resistance to immune checkpoint blockade (ICB) (8–11). This is particularly evident in liver and lung cancers, which frequently develop in chronically inflamed, fibrotic organs, where stromal desmoplasia and myeloid-driven immune suppression are strongly linked to poor clinical outcomes (10, 11). Although immune-suppressive macrophage populations are central mediators of these barriers, strategies to selectively target these populations and reprogram the fibrotic TME remain limited (75, 80). Based on our findings that MΦP⁺sen⁺ macrophages sustain fibrotic, immune-excluded niches and are efficiently engaged by SMNPs, we hypothesized that SMNP-mediated remodeling of the fibrotic TME might restore sensitivity to ICB in fibrotic tumors.

To test this, we engineered murine models of liver and lung cancer that are responsive to clinically established, immunotherapy based first-line therapies in non-fibrotic settings and then embedded these tumors within experimentally induced fibrotic, immunosuppressive microenvironments (Fig. 5A–B). In the liver, fibrosis was established by chronic CCl₄ administration, followed by orthotopic transplantation of syngeneic Myc-GFP; Trp53^−/− (MP) hepatocellular carcinoma (HCC) tumor cells generated by hydrodynamic tail vein injection (81) (Fig. 5C). In the lung, fibrosis was induced by intratracheal bleomycin administration Kras^G12D/+; Trp53^−/− (KP) non-small cell lung cancer (NSCLC) tumor cells (82) (Fig. 5D). To exclude tumor-cell–intrinsic uptake of SMNPs as a potentially confounding factor, MP and KP cells were further engineered to express validated Selp-targeting shRNAs (MPS and KPS) (Fig. S8A,B). Tumor-bearing mice were randomized by baseline imaging and treated with SMNPs to remodel the fibrotic niche before and during administration of clinically relevant, first-line ICB-based regimens: anti-PD-L1 combined with anti-VEGF in liver tumors, or anti-PD-L1 combined with cisplatin in lung tumors. Tumor growth, survival, and immune responses were monitored longitudinally (Fig. 5C, D).

Figure 5: SMNP-mediated niche remodeling overcomes resistance to immune checkpoint blockade.

(A-B) Multiplex IF analysis of vimentin, Lyve1, Collagen type I (Col1A1), and smooth muscle actin (SMA) expression on fibrotic and non-fibrotic murine liver (A) and lung tumors (B). Scale bars, 250 μm. (C-D) Schematic of treatment schedule with nanoparticles and first-line treatment (Tx) in murine fibrotic liver (C) and lung (D) tumor models. MPS: Myc-GFP;p53^KO;shSelp murine liver cancer cells. KPS: Kras^G12D;p53^KO; shSelp murine lung cancer cells. (E) Fluorescence imaging of GFP⁺ tumors and ICG⁺ nanoparticles in murine fibrotic liver tumors treated with DexNav or FiNav nanoparticles in combination with liver first-line Tx. (F) Representative pictures of hematoxylin and eosin macroscopic sections of murine fibrotic lung tumors treated with DexNav or FiNav nanoparticles in combination with lung first-line Tx. (G) Quantification of tumor volume change between d0 and d14 in murine fibrotic liver tumors treated with DexNav or FiNav nanoparticles in combination with liver first-line Tx (n= 5–10 mice/group). (H) Kaplan-Meier analysis comparing overall survival of mice with fibrotic liver tumors treated with DexNav or FiNav nanoparticles in combination with first-line Tx. (I) Quantification of tumor area at d21 of murine fibrotic lung tumors treated with DexNav or FiNav nanoparticles in combination with lung first-line Tx (n= 5–10 mice/group). (J) Kaplan-Meier analysis comparing overall survival of mice with fibrotic lung tumors treated with DexNav or FiNav nanoparticles in combination with first-line Tx. (K-L) Multiplex IF analysis of stromal (Vimentin, Lyve1, Col1a1, SMA) and immune (B220, F4/80, Cd11c, CD8, CD4) markers in murine fibrotic liver (K) or lung (L) tumors treated with DexNav or FiNav nanoparticles in combination with first-line Tx. Scale bars, 100 μm. (M-N) Quantification of active/cytotoxic/exhausted CD8⁺ and CD4⁺ T cells in murine fibrotic liver (M) or lung (N) tumors treated with DexNav or FiNav nanoparticles in combination with first-line Tx (n= 5 mice/group). Individual bars (G) or dots (I, M-N) each represent one mouse. Data are presented as mean ± s.e.m.. Log-Rank Mantel Cox test used for (H, J), one-way ANOVA with Sidak’s multiple comparisons test used for (I) and Mann Whitney test or Welch’s t test for (M,N).

Mirroring resistance patterns observed in patients (10, 83), MPS and KPS tumors that responded robustly to ICB+ therapy in non-fibrotic liver and lung were refractory when embedded within fibrotic TMEs (Fig. 5E–J). SMNP monotherapy with FiNav induced modest tumor growth inhibition and survival benefits, whereas control nanoparticles (DexNav) were ineffective (Fig. 5E–J). Strikingly, combining FiNav with ICB-based therapies induced durable tumor regression in both fibrotic tumor models, accompanied by a significant improvement in overall survival (Fig. 5E–J and Fig. S8 C–E). Comparable therapeutic synergy was observed using FidBET6 in the liver cancer model (Fig. S8F), indicating that both senolytic and senomorphic SMNPs can sensitize fibrotic tumors to immunotherapy.

To define how SMNPs remodel the fibrotic TME, we performed mIF profiling of MPS and KPS tumors and adjacent liver and lung stroma, respectively (Fig. 5K, L). Untreated fibrotic tumors were characterized by expanded αSMA+ activated fibroblasts, dense collagen I deposition, increased LYVE1+ lymphatic endothelium, and CD68+ or F4/80+ macrophage accumulation at the tumor border, with concomitant exclusion of dendritic cells and CD4+ and CD8+ T cells (Fig. 5 K,L and fig. S9 A–H). This architecture persisted in DexNav treated mice (Fig. 5K, L and Fig. S9 A–D). In contrast, FiNav treatment alone dismantled these stromal barriers, reducing collagen type I deposition, αSMA+ myofibroblasts, and macrophage density, while enabling infiltration of dendritic cells and CD4+ and CD8+ T cells. Spatial neighborhood analysis at the tumor border revealed collapse of the pro-fibrotic, immune-suppressive niche and its replacement by antigen presenting cells, dendritic cells, B cells, and effector T cells (Fig. S9G–J).

Despite extensive remodeling of the fibrotic tumor environment, FiNav monotherapy was insufficient to induce durable tumor control, indicating that niche disruption alone is insufficient to fully restore anti-tumor immunity. Accordingly, immunophenotyping revealed that, while FiNav produced a modest increase in activated and proliferating CD4+ and CD8+ T cells, the majority of T cell populations displayed markers of exhaustion (Fig. S9K,L). By contrast, combining FiNav with ICB+ triggered a marked expansion of cytotoxic, proliferating T cells and a concomitant contraction in exhausted T-cell populations in both liver and lung tumors (Fig. 5M–N). Collectively, these results establish that SMNP-mediated niche reprogramming can create a permissive context in which immune checkpoint blockade can elicit robust and sustained antitumor immunity.

SMNPs target a conserved senescent-like macrophage state in human tumors

To assess whether macrophages analogous to the MΦP+sen+ populations identified in fibrotic murine tissues are present in human liver and lung tumors, we applied the MΦP⁺sen⁺ gene signature to published scRNA-seq datasets from treatment-naïve HCC (n=79) (84) and NSCLC (n=12) (85). In both tumor types, the MΦP⁺sen⁺ signature identified a distinct myeloid subcluster (Fig. 6A, C). When mapping our signature onto a previously described framework stratifying the HCC tumor immune microenvironment (TIME) into distinct stages ranging from immune activation to immune suppression, we observed a preferential alignment with the myeloid-driven immune-suppressive niche (Fig. S10A) (84). Comparison with established tumor-associated macrophage programs (77) revealed strongest overlap with Kupffer-like TAMs, interstitial resident tissue macrophages, and LAMs (Fig. 6B, D), all previously implicated in immune suppression (77, 86–88).

Figure 6: P-selectin⁺ senescent-like macrophages mark immunosuppressive, ICB-refractory human tumors and are vulnerable to SMNPs.

(A) Enrichment of the P+sen+ macrophage gene signature in myeloid cells from human liver cancer. (B) P+sen+ macrophage signature enrichment in previously described tumor-associated macrophage (TAM) subsets from (A). (C) Enrichment of the P+sen+ macrophage gene signature in myeloid cells from human lung cancer. (D) P+sen+ macrophage signature enrichment in previously described TAM subsets from (C). (E-F) Multiplex IF analysis of SMA, p21, p53, P-selectin, and immune markers (CD8, CD11c, CD163, CD206, CD68) on human tissue microarrays (TMAs) of hepatocellular carcinomas (HCC) (E) and lung adenocarcinomas (LUAD) (F). Fibrotic burden was defined based on SMA expression. Scale bars, 500 μm. (G-H) (Left) Representative core picture of multiplex analysis mapping P-selectin⁺ senescent-like macrophages, all macrophages and tumors between high and low fibrotic human HCC (G) and LUAD (H) TMAs. (Right) Cellular neighborhood analysis measuring the distance between the different immune and stromal populations in human HCC (G) and LUAD (H) TMAs. (I) Schematic of human non-small cell lung cancer (NSCLC) biopsy from MSK patients. (J) Representative core picture of multiplex analysis mapping P-selectin⁺ senescent-like macrophages, all macrophages and lung tumors from human NSCLC biopsy of responder and non-responder patients to ICB. Scale bars, 250 μm. (K) Quantification of P+sen+ macrophages in human lung cancer of responder and non-responder patients to ICB. (L) Likelihood ratio of the different immune population presence in human lung cancer of responder and non-responder patients to ICB. (M) Ex-vivo killing assays of sorted P-selectin⁺ senescent (P+sen+) or P-selectin⁺ (P+sen−) endothelial cells (EC) and macrophages (Mac) from human NSCLC biopsy treated with FiNav nanoparticles (n= 3 human NSCLC patients). Individual dots each represent one human patient. Data are presented as mean ± s.e.m. Pearson correlation for (G,H), Mann Whitney test for (K), likelihood ratio, cut-off defined using Youden’s J statistic based on Receiver operating characteristic (ROC) for (L) and one-way ANOVA and Sidak’s multiple comparisons test for (M).

We next validated these observations at tissue level by mIF on tumor samples from 16 HCC and 48 NSCLC patients, in which MΦP⁺sen⁺ cells comprised ~2% and ~0.4% of all cells, respectively (Fig. S10B). Across both tumor types, higher MΦP+sen+ abundance was associated with highly proliferative, p53-altered tumors (i.e. Ki67+ p53+ tumor cells) and fibrotic TMEs characterized by increased collagen deposition, αSMA+ fibroblasts, and higher macrophage infiltration (Fig. 6, E–H). MΦP+sen+ abundance was positively associated with CD163+ macrophages and inversely correlated with CD11c+ APC-like macrophages, dendritic cell abundance, and CD4+ and CD8+ T cell density and proliferation (Fig. 6. E–H). Spatial neighborhood analysis of highly fibrotic tumors mirrored murine findings, placing MΦP+sen+ macrophages within multicellular niches enriched for stromal activation and immune exclusion (Fig. S10 C). Furthermore, in a prospective cohort of 19 NSCLC patients receiving neoadjuvant treatment with anti–PD-1 plus carboplatin and pemetrexed (Fig. 6I), mIF revealed that MΦP⁺sen⁺ macrophages were significantly enriched in non-responding tumors (Fig. 6J, K). Whereas intratumoral CD4⁺ and CD8⁺ T-cell abundance showed the expected positive correlation with the efficacy of ICB-based therapy, intratumoral MΦP⁺sen⁺ abundance was a strong negative predictor of response (Fig. 6L).

Finally, to functionally probe SMNP activity in a human context, we analyzed freshly resected NSCLC specimens from three patients and isolated macrophage and endothelial populations based on SPiDER activity, P-selectin expression, and lineage markers. MΦP⁺sen⁺ macrophages were readily detected in all tumors, and their presence independently validated in tissue by mIF (Fig. S10 D). Ex vivo treatment with FiNav, but not DexNav, selectively induced cell death MΦP+sen+ macrophages but had minimal effects on other macrophage subsets or endothelial cells (Fig. 6M). Together, these data establish that MΦP+sen+ macrophages are conserved in both human HCC and NSCLC, occupy pro-fibrotic and immune-excluded niches, and predict sensitivity to immunotherapy while remaining selectively vulnerable to SMNP-based targeting. This defines a conserved senescent-like macrophage axis linking fibrosis, immune exclusion, and immunotherapy resistance, and provides a translational rationale for targeting senescent-like macrophage states to reprogram fibrotic tumor ecosystems.

Discussion

Here we develop senescence-modulating nanoparticles (SMNPs) as a platform for selectively reprogramming fibrotic and immune-excluded tissues present in chronic inflammation and cancer. By exploiting the induction of P-selectin on subsets of senescent-like cells and the established ability of fucoidan to bind P-selectin, fucoidan-encapsulated nanoparticles enable targeted delivery of senolytic or senomorphic agents to disease-relevant cellular niches. In models of liver and lung fibrosis, as well as models of HCC and NSCLC arising in fibrotic environments, SMNPs reduce fibrotic burden, restore immune infiltration, and sensitize tumors to immune checkpoint blockade. This biological efficacy is coupled to a marked improvement in therapeutic index: nanoparticle delivery largely dissociates the senescence-modulatory activity of Navitoclax and BET inhibitors from the severe hematologic and systemic toxicities that have constrained their clinical development (37, 38, 62–66), thereby permitting sustained in vivo exposure and durable therapeutic benefit.

Notably, SMNP delivery preserves, and in some contexts enhances, therapeutic efficacy despite restricting drug exposure to a small, disease-driving subset of cells. This contrasts sharply with chemically similar control nanoparticles lacking P-selectin binding, which abrogate antifibrotic and antitumor activity associated with the free drugs underscoring that efficacy is governed less by whole tissue exposure than by selective engagement of pathogenic cellular states within fibrotic niches. The concordant activity of senolytic (Navitoclax) and senomorphic (dBET6) SMNPs further indicates that acute suppression of the SASP is sufficient to confer key therapeutic benefits. While senescent cell elimination may be advantageous in settings where persistence sustains pathology, our data suggest that transient SASP inhibition, without tissue-disruptive cell loss, may represent a favorable therapeutic mode in indications where preserving tissue integrity is paramount.

Mechanistically, P-selectin delineates restricted subsets of senescent-like cells predominantly within macrophage and endothelial lineages. Integrated single-cell, spatial, and uptake analyses identify a senescent-like macrophage subpopulation as a dominant functional target of SMNPs. This state, which is conserved across murine and human fibrotic tissues and tumors, transcriptionally aligns with previously characterized, pathogenic p16⁺ macrophages and LAM programs that have been implicated in chronic inflammation, fibrosis and immune suppression (28, 35, 69, 74–76, 78). Notably, although endothelial cells can express P-selectin and a subset of these express senescence markers, they do not accumulate SMNPs nor are they functionally perturbed, consistent with prior work indicating that fucoidan nanoparticles undergo P-selectin-dependent transcytosis rather than intracellular retention in activated endothelial cells (45–47). These findings emphasize that surface marker expression alone is necessary but not sufficient to predict SMNP therapeutic engagement and point to additional cellular determinants, such as endocytic routing, that govern selective vulnerability.

Finally, our findings underscore that senescence is not a monolithic cellular program but rather a heterogeneous spectrum of cell states with distinct, and sometimes opposing, roles in tissue remodeling, inflammation, and regeneration (16, 49, 89). Within this landscape, SMNPs preferentially engage a macrophage subpopulation that couples senescent features with profibrotic and immunosuppressive activity and is conserved in fibrotic human tissues. Although disease-driving macrophage states contribute to diverse pathologies (28, 35, 69, 90–92), the ability to selectively target these populations while sparing beneficial and reparative myeloid cells has remained a long-standing and largely unmet therapeutic goal. In this sense, SMNPs represent a second-generation senotherapeutic strategy that combines state-selective targeting with modular niche-directed drug delivery, overcoming both biological and toxicity barriers that may limit other approaches. More broadly, our findings establish a general framework for reprogramming fibrotic and immune-excluded tissue ecosystems and suggest a strategy for enhancing immunotherapy responses in otherwise refractory cancers.

Material and Methods:

Patient samples

Human FFPE TMAs were purchased from TissueArray (www.tissuearray.com). We used a liver TMA containing normal and cirrhotic human liver tissues (#LV805b), adjacent normal and fibrotic pulmonary tissue (#BC04118a), as well as TMA of human HCC (#LV487). Detailed clinicopathologic information for each case is available from the vendor. In addition, we used two TMA generated at Memorial Sloan Kettering Cancer Center (MSK). One TMA comprising primary LUAD enriched in untreated patient and previously published (93) as well as a TMA comprising chemotherapy with or without immunotherapy treated LUAD patients. In addition, whole tissue sections and fresh human lung tumors specimens were obtained from patients who were enrolled at MSK. For all MSK specimens used in this study, all enrolled patients gave consent to an institutional biospecimen banking protocol, and all analyses were performed according to a biospecimen research protocol. All protocols were approved by the MSK institutional review board (IRB). Patients were asked for consent following the IRB-approved standard operating procedures for informed consent. Written informed consent was obtained from all patients before conducting any study-related procedures.

Synthesis of Navitoclax and dBet6-encapsulated nanoparticles

Fucoidan nanoparticles were synthesized via methods adapted from Tannan et. al. Nanoprecipitation was conducted by adding 450 μL of fucoidan (Thermo Scientific, 412541000, 10 mg/mL), 100 μL of 0.01 mM sodium bicarbonate (Fisher, S233), 100 μL of IR-125/indocyanine green (ICG; Thermo Scientific, 412541000, 2 mg/mL), and 50 μL of the drug dissolved in DMSO (20 mg/mL), which was added dropwise to the master mix while vortexing. The solution was centrifuged (20,000g for 10 min for dBet6; 30,000g for 10 min for Navitoclax), and the supernatants were separated from the pellet. The pellet was re-suspended in water or 0.9% saline solution (Sigma-Aldrich,S8776). The concentration of Navitoclax and dBET6 nanoparticles was quantified using either the Alliance 2695e System and H-Class UPLC (Waters) or the Agilent 1260 Infinity Bio-inert Quaternary LC (Agilent). Two-fold serial dilutions of drug were run on LC to generate standard curves, which were used to calculate the drug concentration of the nanoparticle from its LC analysis. The nanoparticles are then diluted with 0.9% sodium chloride (Sigma, Cat #S8776) to a working concentration of 3.75 mg/mL.

Microscopy for nanoparticle characterization

Atomic force microscopy (AFM) was used to both visualize and characterize the size and uniformity of the nanoparticles. A volume of 20 μl of the sample was deposited onto pretreated mica with 3-aminoproply-trietoxy silane (APTES) (Pelco Mica Disc, V1, Ted Pella) and left to air dry. AFM images were captured using the Nanowizard V (JPK-Bruker) in AC Mode Imaging at room temperature. AFM probe (FMV-A, Bruker) with nominal frequencies of approximately75 kHz and nominal spring constant of 3N/m was used for imaging. Images were collected at a 1 Hz Line Rate with an image size of 1 × 1 μm to 3 × 3 μm at 512 × 512-pixel resolution. The images were processed with JPK Data Processing software.

Bench characterization of Navitoclax and dBet6-encapsulated nanoparticles

To ensure nanoparticle batch consistency nanoparticles were characterized before initial use in experiments, intensity-based hydrodynamic size, polydispersity index of the nanoparticles, and zeta potential were measured in water using a Zetasizer Nano ZS (Malvern). Encapsulation efficiency was quantified by Ultra-Performace Liquid Chromatography (UPLC) using an Acquity C18 column (50 mm × 2.1 mm internal diameter; 1.7 μm; Waters Corporation) or by High-Performance Liquid Chromatography (HPLC) using C18 analytical columns (150 mm × 2.1 mm internal diameter, 3.5 μm; Agilent Technologies) and a mobile phase consisting of acetonitrile and deionized water, both containing 0.1% trifluoroacetic acid or formic acid. For UPLC, the program ran a gradient of 5% to 95% acetonitrile over 2 minutes, returning to 5% acetonitrile at 3.5 minutes with a flow rate of 0.5 mL/min. Retention times were 1.9 min for Navitoclax and 2.1 min for dBet6. For HPLC,the gradient elution program ran from 5% to 90% acetonitrile over 5 minutes, followed by 90% to 95% acetonitrile over 3 minutes, with a flow rate of 1 mL/min. Retention times were 5.3 minutes for Navitoclax and 5 minutes for dBET6. Absorbance was monitored at 275 nm for Navitoclax and 256 nm for dBET6 for both UPLC and HPLC.

Nanoparticle Stability Assay

At timepoints 0, 3, 6, 24, and 48 hours; both Navitoclax and dBet6 fucoidan nanoparticles suspended in 100 μL of 0.9% saline were diluted 1:100 in water. Size of the nanoparticles was measured by DLS using the diluted samples.

Generation and authentication of shSelp cell lines

shRNAs targeting mouse Selp were generated as previously described. In brief, the SplashRNA algorithm (94) was used to generate 10 top-scoring shRNA and each was cloned into the miR-30 backbone. Renilla luciferase shRNA, previously described, was used as a control (59, 95). shRNA efficacy was tested by qPCR and western blot. The shRNA with the highest knock down efficacy(TGCTGTTGACAGTGAGCGCTCTCTGTAGTTTAAAACAAAATAGTGAAGCCAC AGATGTATTTTGTTTTAAACTACAGAGAATGCCTACTGCCTCGGA) was used in experiments. Expression of shSelp was linked to a BFP reporter system, allowing for the sort and enrichment of shSelp+ populations based on high BFP expression before experiments.

Quantitative PCR with reverse transcription

Total RNA was isolated using the RNeasy Mini Kit (Qiagen) and cDNA was obtained using TaqMan reverse-transcription reagents (Applied Biosystems). Quantitative PCR (qPCR) was performed in triplicates using SYBR green PCR master mix (Applied Biosystems) on the ViiA 7 Real-Time PCR System (Invitrogen). 18S (18S forward (5’-GTAACCCGTTGAACCCCATT-3’) and 18S reverse (5’- CCATCCAATCGGTAGTAGCG-3’)served as endogenous normalization controls for mouse samples.

Animal Studies

Housing conditions

All animal experiments in this study were performed in accordance with protocols approved by the Memorial Sloan Kettering Cancer Center (MSKCC) Institutional Animal Care and Use Committee (protocol number 11–06-011, 12–04-006). MSKCC guidelines for the proper and humane use of animals in biomedical research were followed. Mice were housed in a temperature-controlled room (22 ± 1°C), with a 12-hr light/dark cycle and free access to food and water. All mouse experiments were done using 8- to 14-week-old male mice of C57BL/6N background. Sample sizes were determined on the basis of previous experiments and published studies to ensure adequate power to detect biologically relevant differences. Mice were randomly assigned to experimental groups. Investigators were blinded to group allocation during data collection and analysis whenever possible.

In vivo induction of mouse liver and lung fibrosis models

To induce liver fibrosis, mice were treated twice a week with 12 consecutive intraperitoneal injections of 1 ml/kg carbon tetrachloride (CCl₄) in peanut oil. Control mice received peanut oil (100 μL) alone on the same schedule. Mice were continuously treated with 1 ml/kg of CCl₄ bi-weekly while on treatment with nanoparticles. To induce lung fibrosis, bleomycin was intratracheally administered at 1.5 U/kg. Saline (50 μL) was administered as control. For nanoparticle treatment, 100 μL of Navitoclax-encapsulated (15 mg/kg) or dBET6-encapsulated (25 mg/kg) fucoidan, free fucoidan (1 mg/mL) or dextran control nanoparticles were injected intravenously via the tail vein once week for four weeks starting after liver fibrosis establishment (six weeks on CCl₄). Samples were collected 4 days after the last injection.

Fibrotic liver cancer model and treatment

Mice were first injected twice a week for 6 weeks with 1 ml/kg CCl₄ to induce liver fibrosis (see above). Upon liver fibrosis establishment, 2.5 × 10⁵ murine liver cancer cells (96) (Myc-GFP; sg-Trp53; sh-Selp) were orthotopically transplanted into fibrotic livers. Upon tumor growth (2 weeks post transplantation), mice were randomized into different groups based on tumor size, measured via ultrasound. Navitoclax-encapsulated (15 mg/kg) or dBET6-encapsulated (25 mg/kg) fucoidan (or dextran control) nanoparticles were injected intravenously at day 14, 17, 19 post transplantation then once a week until the end of the study. For first-line based ICB, anti-PD-L1 (BE0101, BioXCell, 5 mg/kg) and anti-VEGF-A (BE0399, BioXCell, 50 μg/mouse) depletion antibodies were administered intravenously at day 18 post transplantation twice or once a week, respectively. Mice remained on CCl₄ administration throughout the treatment period until collection and humane endpoint. Mice were monitored for liver tumor volume by ultrasound (Fig. 5A,C).

Fibrotic lung cancer model and treatment

Mice were first injected once intratracheally with 1.5 U/kg bleomycin to induce lung fibrosis. Upon lung fibrosis establishment (day 7 post bleomycin injection), 1.5 × 10⁵ murine lung cancer cells (97) (KRas^G12D; Trp53^KO; sh-Selp) were injected intravenously. Upon tumor growth (1 week post transplantation), mice were randomized into different groups. Navitoclax-encapsulated (15 mg/kg) or dBET6-encapsulated (25 mg/kg)fucoidan (or dextran control) nanoparticles were injected intravenously at day 7, 10, 12 post cancer cell injection, then once a week for 3 weeks. For first-line based ICB, anti-PD-L1 depletion antibodies (BE0101, BioXCell, 5 mg/kg) and cisplatin (S1166, Selleck Chemicals, 3 mg/kg) were administered intravenously at day 18 post transplantation twice or once a week, respectively. Mice were monitored for lung tumor volume by micro-computed tomography (Fig. 5B.D).

Ultrasound and micro-computed tomography (μCT) imaging

For ultrasound, mice were first placed in a chamber for anaesthesia by isoflurane. Once mice were adequately anaesthetised, all abdominal hairs were removed using hair removal cream. The abdomen was cleaned off all excess cream with alcohol wipes to avoid skin irritation. Anaesthetisedmice were then placed under constant nasal anaesthetic administration on a heated pad monitoring heart rate and breathing during the ultrasound scan. The shaved abdomen was covered in ultrasound gel before scanning to enhance the transducer detection. 2D scanning was then performed onthe liver area to record the long (D) and short (d) tumor diameter using a Vevo F2 ultrasound (Fujifilm - VisualSonics). After scanning, mice were placed into an empty recovery cage and then transferred back to their original cage. Liver tumor volume was calculated using the following formula: tumor volume = (D x d²)/2. For μCT, scans were performed using a Mediso Nano SPECTCT system covering only the lung fields of each mouse as previously described (98).

Animal tissue harvesting and blood/serum analysis

At indicated time points, mice were euthanized by carbon dioxide inhalation followed by cervical dislocation, and blood was collected by cardiac puncture. For lung collection, mice were systemically perfused with PBS to remove intravascular blood prior harvest. Livers were excised and rinsed in PBS. The liver right lobe was divided into pieces for downstream processing as follows: 1. Formalin-fixed paraffin-embedded tissues (10% formalin (Epredia, 5705) for 24 hours then stored in 70% EtOH until paraffin embedding); 2. 4% paraformaldehyde (PFA)-fixed followed by cryopreservation. (4% PFA for 3 hours followed by fixation in 30% sucrose overnight before embedding in Tissue Plus O.C.T Compounds (Fisher HealthCare, 4585) on dry ice); 3. flash freezing on dry ice for immediate use or storage at −80°C. Blood samples were processed to obtain EDTA-plasma and serum. Complete blood count were performed using the Element HT5^™ hematology analyzer (Heska). Plasma and serum were isolated by centrifugation at 900g for 10 min at 4°C and stored at −80°C. Serum samples were submitted to the Laboratory of Comparative Pathology (Memorial Sloan Kettering Cancer Center [MSKCC]. For alanine aminotransferase (ALT) and aspartate aminotransferase (AST) enzymatic rate methods using automated analyzers were used.

In vivo nanoparticle bioluminescence analysis

Upon tissue collection at indicated time point tissue was rinsed in PBS and kept on ice until imaging. Nanoparticle fluorescence in tissues was captured using tiBright FL1000 (invitrogen), 745em/765ex, and quantified using Fiji/ImageJ.

Liver portal vein flow analysis

Mice were anesthetized and prepared for ultrasound as described before. The liver portal vein flow was recorded using the Pulse Wave Doppler function of the Vevo F2 ultrasound (Fujifilm - VisualSonics)

Histological analysis: tissue preparation

Tissues for FFPE processing were fixed in 10% neutral buffered formalin (Epredia, 5705) for 24 h, transferred to 70% ethanol, and submitted to IDEXX or the Memorial Sloan Kettering Cancer Center (MSKCC) Molecular Cytology Core for paraffin embedding and sectioning. Hematoxylin and eosin (H&E) staining was performed using standard protocols. For immunohistochemistry (IHC) and immunofluorescence (IF) on FFPE tissues, blocks were sectioned at 5 μm, deparaffinized, and rehydrated through graded ethanol to water. Antigen retrieval was performed using a pressure cooker with either a dual–pH protocol (10 mM citrate buffer, pH 6.0 [Vector, H-3300], and EDTA–Tris buffer, pH 9.0 [Vector, H-3301]) or a single 10 mM citrate buffer, pH 6.0. For cryopreserved tissues, samples were fixed in 4% paraformaldehyde (PFA) in PBS (Thermo Fisher Scientific, J61899.AK) for 4–8 h, cryoprotected, and further fixed overnight in 30% sucrose in 4% PFA in PBS. Tissues were embedded in Tissue-Plus O.C.T. Compound (Fisher HealthCare, 4585) on dry ice and stored at −80°C. Cryosections (5–10 μm) were cut using a cryostat onto SuperFrost Plus microscope slides (Fisher Scientific, 12–550-15) and stored at −80°C until use. Prior to staining, slides were thawed at room temperature for ~15 min or until condensation evaporated and rehydrated in PBS.

Histological analysis: conventional immunohistochemistry and immunofluorescence

FFPE and cryosections, tissue sections were outlined using an ImmEdge Hydrophobic Barrier PAP Pen (Vector Laboratories, H-4000) and blocked in PBS containing 3–5% BSA (GeminiBio, 700–100P) and 0.01–0.3% Tween-20 (PBST). Primary antibodies were diluted in blocking solution at the indicated concentrations and incubated on sections overnight at 4°C. Slides were washed in PBST and incubated with appropriate secondaryantibodies diluted in blocking buffer for 1 h at room temperature or overnight at 4°C. Following washing, nuclei were counterstained with DAPI (1 μg/mL in PBS) for 5–10 min. Slides were then washed and coverslips mounted using ProLong^™ Gold Antifade Mountant (Invitrogen, P36930). Slides were scanned using a Pannoramic Scanner (3DHistech) with a 20×/0.8 NA objective at the MSKCC Molecular Cytology Core, or images were acquired manually using a fluorescence microscope 20x,40x or 63x using an SP8 (Leica) or a Zeiss microscope (Imager.Z2, Zeiss camera (AxioCamMRc). For general quantification, 5 independent fields of views per mouse were used.

Sirius Red staining

Sirius red staining was performed using 5-um FFPE tissue section and the Sirius Red / Fast Green Collagen Staining Kit (Chondrex, 9046) following manufacturer’s instructions. Briefly, liver or lung tissue sections were baked, deparaffinized, and rehydrated through graded ethanol to water. Slides were stained with Dye solution for 30 minutes, washed and incubated with the dye extraction buffer. After rinsing in water, slides were dehydrated, cleared in xylene, and mounted with a resinous medium. Brightfield images were acquired under consistent illumination/white balance with Zeiss microscope (Imager.Z2, Hamamatsu camera (ORCA-R2)) at 20x magnification. Collagen area fraction (red) was quantified using ImageJ in 5 independent fields of views per mouse. Data were analyzed and visualized using GraphPad Prism.

Detection of senescence-associated β-galactosidase activity

Chromogenic senescence-associated β-galactosidase activity (SAβ-gal) was performed on cryopreserved using X-gal (Invitrogen, 15520018) as previously reported (PMBID: 30573629, 36302222, 40299790). Briefly, tissues were washed twice with 1 mM MgCl2 PBS solution (pH 5.5 for mouse, pH 6.0 for human, adjusted using HCl). An X-gal working solution consisting of 1 mM MgCl2 PBS, 5 mM potassium ferricyanide, 5 mM potassium hexacyanoferrate (II), and X-gal (40X from 40 mg/mL stock diluted in N,N-dimethylformamide) was prepared, filtered (Steriflip Vacuum tube Top Filter Millipore, SE1M179M6) and applied to sections. Slides were incubated at 37ºC in a humidified chamber for 2–6 hours. Upon signal detection, slides were washed in PBS 2X and coverslips were mounted using Fluoromount-G Mounting Medium (SouthernBiotech, 0100–01). Fluorescent detection of SA-βgal activity was carried out utilizing the ImaGene Red C₁₂RG lacZ Gene Expression Kit (Invitrogen, I2906). Cryosections (5–10 μm) were incubated with 1% chloroquine at 37 °C for 30 min, washed twice with pre-cooled PBS, and incubated with C12RG (1:50 dilution) at 37 °C for 30 minto 2 h. Sections were rinsed immediately with PBS, where indicated, slides were subsequently processed for immunostaining following the protocol above. Fluorescence images were acquired as whole slide scans with a Pannoramic P250 Flash scanner (3DHistech, Hungary) using 20x/0.8NA objective lens by the Molecular Cytology Core. Fluorescent detection of SA-βgal activity following single-cell tissue dissociation as described below was performed using the SPiDER-βgal kit (Dojindo, SG04). Cells were resuspended in FACS buffer containing Baflomycin A1 1:1000 and incubatedfor 1 h at 37 °C followed by incubation with SPiDER-βgal substrate (1:1000) for 30 min at 37 °C. Cells were washed with FACS buffer and processed for flow cytometry and sorting as described below.

Mutiplex IF imaging and analysis

Multiplex-IF experiments were performed using the cyclic IF CellDive platform (Lecia) following the manufacturer’s protocol using either 5 μm FFPE or cryopreserved sections. Cryopreserved tissue sections were baked at 64C for 30 minutes and FFPE sections were prepared as above using a double pH6/pH9 antigen retrieval. Next all following steps were performed after securing the slides in the Leica Clickwell slide holder. Slides were washed with PBST and blocked with 3% BSA in PBST for 30 min to 1 h followed by counterstain DAPI for 5 to 10 min. After washes, slides were imaged unstained to acquire background autofluorescence. Samples were then sequentially stained with antibodies and imaged using DAPI, Cy3, Cy5, and FITC channels on the CellDive instrument with CellDive image acquisition and processing software. Each field of view (FOV) was imaged in each staining round, followed by autofluorescence removal, registration with baseline DAPI, and stitching. Unconjugated primary antibodies were used in the first round of staining, followed by secondary antibody staining. After imaging, dye inactivation was performed using 0.1 M Na₂CO₃ 3% H2O2 solution for 15 minutes at RT before starting the next round of autofluorescence imaging and staining. All rounds of imaging and slide storage were done in a solution of PBS with 50% glycerol. Staining quality and fluorescence removal were verified after each round. The fully stitched images were imported into HALO image analysis software (Indica Labs) for analysis. Cell segmentation was performed using the “traditional” nuclear segmentation option, with analysis settings optimized for each staining category. Cell segmentation was performed using the “traditional” nuclear segmentation option, with analysis settings optimized for each staining category. Phenotypic classifications were used to define cell types. When analyzing TMAs a limited number of cores or tissue areas were excluded from the analysis due to poor tissue quality, antibody aggregates, tissue folds or absence of tissue. Each patient’s data represents the average cell count across single or multiple replicate cores. After image processing, data were analyzed and visualized using GraphPad Prism. Spatial analysis were performed using the HALO image analysis software (Indica Labs). After cell type annotation, using predefined marker combinations, we determined the average distance between two cell populations, using the nearest neighbor algorithm. Data were analyzed and visualized using GraphPad Prism.

Cytokine profiling

After harvest, fresh liver or lung tissues were rinsed with PBS and placed in Tissue Lysis buffer (0.5% Igepal, 0.5% sodium deoxycholate, 0.1% sodium dodecyl sulfate, 50 mM Tris-HCl [pH 7.5], and 150 mM NaCl) with Protease Inhibitor Cocktail. Tissue was homogenized using mechanical homogenization and centrifuged at 1000g for 10 min at 4 °C. Supernatant was then quantified using BCA Protein kit following manufacturer instructions. The supernatant was used immediately or aliquoted and stored at −80 °C until use. Assays were using the Proteome Profiler Mouse XL Cytokine Array (R&D Systems, ARY022B) and the BioRad Imager (ChemiDoc MP Imaging System) with automated optimal exposure following the manufacturer’s protocols. To generate raw values signal intensity was quantified using ImageJ. After image processing, data were analyzed and visualized using GraphPad Prism.

Lung and liver dissociation

Single cell suspension of liver non-parenchymal cells was generated from normal or CCl_4-induced fibrotic livers, 48 h after the final CCl₄ or oil injection, using an in situ pronase-collagenase retrograde perfusion technique followed by Nycodenz density separation, as previously described (99, 100). Briefly, mice were anesthetized with isoflurane. The inferior vena cava (IVC) at the hepatic outflow was cannulated with a catheter IV (Terumo, SR-OX2225CA), the portal vein was cut to provide outlow, and the supra-diaphragmatic IVC was clamped to confine perfusion to the liver. Livers were perfused with pre-warmed PBS at 42 °C (5 ml/min), followed by pre-warmed pronase (0.4 mg/mL, Sigma, P5147) and finally by collagenase D (1.8 mg/mL, Roche, #11088874103) solutions. Perfused livers were excised and incubated ex vivo in pronase(0.5 mg/mL) and collagenase (0.4 mg/mL) solution for 30 min at 40 °C with gentle agitation, then mechanically dissociated and filtered through 70-μm strainers. Cell suspensions were twice washed in Gey’s balanced salt solution B, GBSS/B (Sigma-Aldrich, G9779) containing DNAase I (0.02 mg/mL, Roche, 10104159001). Cell suspensions were mixed 1:1 with 34% w/v Nycodenz to a final concentration of 17% w/v, overlaid with GBSS/B, and centrifuged at 1,380 × g for 17 min at 4°C with brake off; the interphase was collected. After wash, the final non-parenchymal cell pellets were resuspended in FACS buffer (PBS, 2 mM EDTA (Invitrogen 15575), 0.5% BSA (GeminiBio, 700–100P)) for staining with SPIDER and fluorescent antibodies.

Dissociation of normal or fibrotic lungs (14 days after bleomycin or saline administration), was performed using a dispase-collagenase buffer was used as previously described (101). Briefly, following systemic perfusion with PBS to clear lungs of blood, through the trachea lungs were inflated with 3 mL of lung digestion buffer (S-MEM (Thermo Fisher Scientific, 11380037)) with 1.7 U ml−1 dispase (Corning, 354235), 500 U ml−1 collagenase IV (Thermo Fisher Scientific, 17104019) and 10 μg ml−1 DNase I (StemCell Technologies, 07900)), finely minced and further dissociated in a 37°C oven with gentle agitation for 45 min. The dissociated cells were passed through a 100-μM filter and red blood cell lysis was performed using ACK lysis buffer (Thermo Fisher Scientific, A1049201), following the manufacturer’s protocol. Cells were washed with S-MEM containing 2% heat inactivated fetal bovine serum (HI-FBS; Hyclone, SH30910.03), filtered through a 40 μm strainer (Corning, 431750) and pelleted by centrifuging at 300g for 5 min at 4°C. Cell pellets were resuspended in FACS buffer for staining with SPIDER and fluorescent antibodies as described below.

Human lung tissue samples were obtained from 3 NSCLC patients undergoing pulmonary resections. Normal tissue was obtained outside of the tumor margin, determined by intraoperative pathology observation. Tissue samples were washed with pre-chilled PBS, minced, and centrifuged at 250g for 5 min at 4°C. The supernatant was drained and tissue samples were further minced and dissociated in 5 mL-10 mL of lung digestion buffer, described above. Samples were further dissociated in a 37°C oven with gentle agitation for 1–1.5hr. Following, dissociated tissue samples were processed as lung dissociation described above, and resuspended in FACS buffer for downstream processing as described below.

Flow cytometry

Following SPiDER staining when applicable, single-cell solutions of lung or liver cells were washed with pre-chilled PBS and stained with either LIVE/DEAD^™ Fixable Near-IR Dead Cell Stain Kit (Liver digestions, Thermo Fisher Scientific, L34994) or LIVE/DEAD Fixable Aqua (Lung digestion, Thermo Fisher Scientific, L34957) per manufacturer’s instructions for 30 min on ice. Cells were then washed with PBS, then with FACS buffer and incubated blocked with anti-mouse Fc (BD Biosciences, 553142) in FACS buffer for 10 min on ice, followed by antibody panels for 30 min on ice. Stained cells were centrifuged at 250g for 5 min at 4°C, washed with FACS buffer and filtered into 5 mL polystyrene tubes with strainer Caps (Falcon, 352235). Single-cell suspension samples were acquired using a 5-laser Cytek Aurora spectral analyzer (RRID:SCR_019826) or sorted with a 5-laser BD FACSymphony S6 SE or BD FACSDiscover S8 (RRID:SCR_026674) cell sorters. 6-way sorting was performed using 4-way purity mode for the S6 and purity mode for the S8, with a 100 μm nozzle at 20 psi (137.9 kPa) and a drop drive frequency of 30 kHz. Spectral unmixing was done using vendor-specific acquisition software, with autofluorescence of non-senescent cells incorporated as an additional parameter. The resulting unmixed data were subsequently analyzed using FlowJo (RRID:SCR_008520).

For single cell RNA-sequencing sorts, cells were stained with SPiDER, BUV563-anti-CD45 (BD Biosciences, 612924, 1:200) and BB700-anti-CD62P (BD Biosciences, 742128, 1:200). Live/Dead− singlets events were first split into CD45− and CD45+ gates; within each gate, P-selectin+ (CD62P+) and P-selectin− (CD62P−) as well as senescent (SPiDER+) and non-senescent cells (SPiDER−) subsets were defined and sorted simultaneously (see Fig. S4a, S5a for gating strategy). Because autofluorescence differs between CD45− and CD45+ immune compartments, SPiDER negative thresholds were set separately in each CD45 gate, per the established method (14). For library preparation, CD45+ and CD45− cells were pooled together into 4 main populations: SPiDER-/P-selectin−, SPiDER+/P-selectin−, SPiDER−/P-selectin+ and SPiDER+/P-selectin+ populations (Fig. S5a).

For sorts of lung and liver single cell suspensions used in ex vivo nanoparticle treatment experiments, panels included BUV563-anti-CD45 (BD Biosciences, 612924, 1:200), BB700-anti-CD62P (BD Biosciences, 742128, 1:200), APC-anti-CD31 (Biolegend, 102409, 1:200), and either PE-anti-CD64 (Biolegend, 139314, 1:200, for lung macrophages) or PE-anti-F4/80 (Invitrogen, 61–4801-82, 1:200, for liver macrophages). Endothelial cells were defined as Live/Dead-, CD45−, CD31+; macrophage as Live/Dead-, CD45+, CD64+ (lung) or Live/Dead-, CD45+, F4/80+ (liver) populations (Fig. S5a). Within each population SPiDER−/P-selectin−, SPiDER+/P-selectin−, SPiDER-/P-selectin+ and SPiDER+/P-selectin+ subsets were sorted and used for ex vivo nano-treatment (Fig. S5a).

Nanoparticle Ex Vivo Assays

Sorted cell populations, as described above, were centrifuged at 250g for 5 min at 4°C. Cell pellets were resuspended in DMEM (Gibco, 11995–065) supplemented with 10% FBS and 1% Penstrep (Gibco, 50–753-3040) containing either DexNav (1:200), FiNav (1:200), DexdBET6 (1:200), or FidBET6 (1:200) and treated for 30 min. Cells were then centrifuged at 250g for 5 min at 4°C and washed 3 times with DMEM. Cells were resuspended in DMEM supplemented with 10% FBS and 1% PenStrep and seeded at 0.01×106 into a seeded in a 96-well plate (Cellvis, P96–1.5P),that was previously coated with collagen (30 min, 37°C (Advanced Biomatrix PureCol).

Following, to measure cell toxicity in samples treated with Nav loaded particles, propidium Iodide (Thermo Fisher Scientific, P1304MP) was added to the milieu. After seeding the plate was loaded onto the IncuCyte S3 (Sartorius) and the Incucyte Software (v2020C) was used to quantify cell death and confluency over time.

In experiments using dBET6 particles, following seeding and treatment, the plate was placed in a humidified incubator at 37°C with 5% CO2 for 2hrs. After 2hrs, cells were fixed for 15 min on ice with 4% PFA in PBS. Subsequently, cells were blocked and permeabilized using 3 % BSA with 0.03 %Tween-20 for 30 min and then stained with anti-BRD4 (Abcam, ab128874, 1:200) in blocking buffer overnight at 4°C. After overnight incubation, cells were washed 3 times with blocking buffer and then incubated with Alexa Fluor 647-conjugated-anti-rabbit (Invitrogen, A32795, 1:1000) in blocking buffer for 1 h with gentle agitation. Cells were then washed with PBS and stained with DAPI (1 ug/mL) diluted in PBS for 5–10 minutes. After washes, cells were imaged using SP8Leica microscope. Quantification of BRD4 signal was done using ImageJ software as described above.

Single Cell RNA-sequencing library preparation

Cells were stained with Trypan blue and Countess II Automated Cell Counter (ThermoFisher) was used to assess both cell number and viability. Following QC, the single cell suspension was loaded onto Chromium GEM-X 3’ Chip (10x Genomics PN 1000690) and GEM generation, cDNA synthesis, cDNA amplification, and library preparation of 2–15,000 cells proceeded using the Chromium GEM-X Single Cell 3’ Kit v4 (10X Genomics PN 1000691) according to the manufacturer’s protocol. cDNA amplification included 11–12 cycles and 88–327 ng of the material was used to prepare sequencing libraries with 10 cycles of PCR. Indexed libraries were pooled equimolar and sequenced on a NovaSeq 6000 (16588) or X (both projects) in a PE28/88 paired end run using the NovaSeq 6000 S4 or X 10B or 25B Reagent Kit (Illumina). An average of 26 thousand reads was generated per cell.

Single Cell RNA-sequencing (scRNA-seq) data processing

Data preprocessing and quality control:

All mouse scRNA-seq datasets were barcode-corrected, aligned and UMI-corrected using CellRanger (v8.0.1) using mouse genome mm10 and default parameters for 3’ v4 GEM-X scRNA-seq kit.

Empty droplet removal and ambient RNA subtraction:

Empty droplets and ambient RNA subtraction were performed using the remove-background function of CellBender (v0.3.0) with expected_cells equal to estimated number of cells detected by CellRanger. Total-droplets-included was based on estimated UMI count plateau of ranked cell barcodes, (fpr = 0.01 (default), learning-rate = 0.0001 (default) and epochs = 150 (default). Cellbender background-corrected count matrices were used for downstream analysis, and counts were summed for genes sharing the same gene symbol.

Low-quality cell removal, doublet detection & normalization:

scRNA-seq data was processed using scanpy (v1.10.4) and rapids_singlecell (0.10.10) (hereinafter rsc.). To remove low quality cells, quality control metrics for library size, percentage of mitochondrial (mt), ribosomal (ribo) and the number of genes with at least 1 count in a cell (cngeneson) were first calculated using (rsc.pp.calculate_qc_metrics). An additional ‘size_factor’ was calculated for downstream batch integration by dividing the total counts per cell (total_counts) by the average of the total counts for all cells in the dataset (hereinafter size_factor). Genes detected in fewer than 3 cells were filtered for downstream analysis. For quality control boundary criteria,loglibrary size (log1p_total_counts) < 7.5 and mitochondrial count percentage (pct_counts_mt) > 15 were used and cells falling below these criteria were excluded from downstream analysis. Doublets were removed using scrublet (rsc.pp.scrublet) with the following parameters (log_transform=True, expected_doublet_rate=0.1, n_prin_comps=30 (default), threshold=0.25, sim_doublet_ratio=2 (default)). Count data was normalized (rsc.pp.normalize_total – target_sum = 1×10⁴) and log transformed with pseudocount 1 (rsc.pp.log1p). This normalization was used for downstream processing with the exception of MAST differential expression analysis where target_sum = 1×10⁶.

Gene exclusion for post-cleaning preprocessing:

The following classes of genes were excluded for downstream feature selection and embedding generation: (i) mitochondrial and (ii) ribosomal transcripts. In excluding these genes, we sought to minimize variation stemming from quality control metrics alone.

Embedding, batch integration and cell type annotation:

To generate refined and batch corrected embeddings and batch corrected annotations across samples and conditions single-cell ANotation using Variational Inference (scANVI) was used (104). First a kNN graph was constructed on non-batch corrected count matrices. This was done by first selecting highly-variable genes (HVG) (rsc.pp.highly_variable_genes) with the following parameters: flavor=Seurat_v3, n_top_genes=3000, followed by dimensionality reduction using PCA (rsc.tl.pca) ((n_comp=500). A kNN graph was then generated using function rsc.pp.neighbors with n_neighbors=20 from the top 17 PCs as specified from the inflection point in the cumulative explained variance curve (explaining 58% of variance). To generate first approximate cell type annotations CellTypist (102) cell annotation (‘CellTypist_annot_voting’) was performed using celltypist.annotate function with majority_voting = True and using Immune_All_Highmodel. Immune_All_High model was converted from human to mouse using CellTypist’s convert function. Batch corrected latent space was calculated on CellBender corrected counts using only HVGs with the following model (scvi.model.SCVI.setup_anndata) parameters: batch_key=‘SampleID’, size_factor=‘size_factor’, labels_key= ‘CellTypist_annot_voting’ and (scvi.model.SCVI) with parameters: n_layers=2, n_latent=30. scVI latent space was then computed using scvi.model.train() function. Following, scANVI latent space was computed using scvi.model.SCANVI.from_scvi_model() and scvi.model.SCANVI.train(). kNN graph was then generated on scANVI latent space using rsc.pp.neighbors() with n_neighbors = 30. UMAP embeddings were generated with rsc.tl.umap() using default parameters. For downstream re-embedding of fibrotic liver/lung scRNA-seq datasets cell annotations were first re-generated utilizing batch corrected kNN graph and used to regenerate a scANVI latent space for liver and lung fibrotic samples with the same parameters used above. Final annotations for cell types were generated by first performing phenograph clustering using the phenograph package’s (v1.5.7) on scANVI latent space with function phenograph.cluster() with the following parameters (k=30, clustering_algo=‘leiden’, resolution_parameter=0.8). Phenograph clusters were then annotated based on known markers for tissue cell type markers.

Differential expression:

To perform single-cell differential expression 1×10⁶ normalized logtransformed gene counts was performed using MAST (v3.22) by first fitting hurdle model “~1 + total_counts + SortID” with MAST zlm() function using parameters: method=‘bayesglm’ and regularize=TRUE. Then likelihood ratio tests (LRT) were performed using MAST summary() function with doLRT corresponding to differential expression contrast of interest. FDR values from hurdle component < 0.05 and logFC > 0.5 or < −0.5 were used to determine differentially expressed genes. Differential expressed gene (DEG) analysis was used to generate gene signatures for MΦP+sen+ populations (compared to MΦP−sen−) in both the liver and lung (Fig 4A,B and Fig S6E,F).

Analysis of single cell RNA sequencing data from fibrotic mouse models

scRNA-seq analysis in Mouse Fibrosis Models:

We analyzed scRNA-seq data from mouse liver and lung fibrotic tissues to characterize the myeloid landscape during organ fibrosis. Preprocessed data (in .h5ad format) were integrated into Seurat for downstream analysis. Dimensionality reduction and visualization were performed using UMAP, with cell types annotated through multiple methods, including CellTypist and PhenoGraphclustering. Module scores for MΦP+sen+Liver and MΦP+sen+Lung signatures were calculated using the Seurat AddModuleScore function based on custom mouse-specific gene signatures. Comparative analysis of signature enrichment was conducted between specific sorted populations (e.g., P+sen+ vs. P−sen−), with statistical significance assessed via Wilcoxon rank-sum tests and visualized through violin plots and feature plots to identify senescence-associated myeloid subsets across different fibrotic organs.

Analysis of single cell RNA sequencing data from public datasets

scRNA-seq analysis in HCC patients:

We analyzed scRNA-seq data of liver cancer patient samples from GSA-Human Study HRA001748 (National Genomics Data Center). Specifically, we focused on the myeloid compartment of HCC patients, including macrophages, monocytes, and dendritic cells (84). To manage the computational scale of the dataset, count matrices were processed using BPCells. Data processing included standard Seurat normalization, cell cycle scoring, and regression of cell cycle effects and mitochondrial gene percentages. Integration across samples was performed using Harmony through Seurat’s IntegrateLayers function. Module scores were calculated for myeloid-specific signatures using the Seurat AddModuleScore function. Cells were classified into high/low groups based on the 75th percentile of MΦP+sen+Liver signature. Furthermore, we utilized the TIMELASER clinical classification to summarize signature scores across different patient groups, evaluating the mean expression levels of senescence-related pathways within each clinical category. Statistical significance of signature score differences was assessed using Wilcoxon rank-sum tests, with results visualized through violin plots, box plots, and lollipop plots to display effect sizes and adjusted p-values.

scRNA-seq analysis in LUAD patients:

We analyzed a previously published scRNA-seq dataset comprising 9 LUAD patient samples accessed through Gene Expression Omnibus (GEO; GSE253013), specifically focusing on tumor derived myeloid cell (including macrophages and monocytes) (103). To ensure computational efficiency for large-scale data, count matrices were managed using BPCells. Data processing included standard Seurat normalization, cell cycle scoring, and regression of cell cycle effects and mitochondrial gene percentages. Integration across samples was performed using Harmony through Seurat’s IntegrateLayersfunction. Module scores were calculated for senescence-associated and myeloid-specific signatures using the Seurat AddModuleScore function. Cells were categorized into high/low groups based on the 75th percentile of MΦP+sen+Lung gene signature. Statistical significance of signature score differences was assessed using Wilcoxon rank-sum tests, with results visualized through violin plots, box plots, and lollipop plots to illustrate median differences and adjusted p-values.

scRNA-seq analysis in fibrotic and healthy tissues:

We analyzed a previously published and pre-integrated scRNA-seq dataset of human liver and lung myeloid cells from healthy and fibrotic tissues (78). and available through the Broad Institute Single Cell Portal (https://singlecell.broadinstitute.org/single_cell/study/SCP2156/identification-of-a-broadly-fibrogenic-macrophage-subset-induced-by-type-3-inflammation-human-liver-lung-myeloid-fibrosis-scrnaseq-atlas). To facilitate the analysis of this large-scale dataset, count matrices were processed using BPCells for efficient storage and computation. Data processing included standard Seurat normalization, cell cycle scoring, and regression of cell cycle effects. Standardized embeddings from the original study were used for UMAP visualization. Module scores for senescence and progenitor signatures were calculated using Seurat’s AddModuleScore function with MΦP+sen+ combined signatures. This signature was generated by taking the overlap of genes with logFC of > 1 and FDR values of < 0.05 from the MΦP+sen+Liver and MΦP+sen+Lung signatures. To evaluate the enrichment of these signatures in disease states, cells were grouped based on health status (Healthy/Normal vs. Fibrosis), and statistical significance was determined using Wilcoxon rank-sum tests. Results were visualized through violin plots, feature plots, and average module score heatmaps to characterize the myeloid landscape across tissues.

Quantification and statistical analysis

For IHC/IF including multiplex IF studies, staining was analyzed using HALO (Indica labs) or Fiji softwares. For all in vivo studies, data are shown as means ± s.e.m.. Before performing any statistical tests, we tested normal distribution by Shapiro-Wilk test. Statistical significances were determined by two-tailed unpaired student’s or Welch’s t test for two-group comparison, Multiple group comparisons were preformed using one-way ANOVA with Sidak’s multiple comparisons test. Overall survival (OS) was calculated using the Kaplan-Meier method and log-rank (Mantel-Cox) test was used for survival comparison. The predictive value of SMNP and MΦP⁺sen⁺ enrichment as a prognostic factor was evaluated by examining the area under the receiver-operator characteristic (ROC) curve using a confidence interval of 95%. Likelihood ratio was calculated using a cut-off defined via Youden’s J statistic based on Receiver operating characteristic (ROC) analysis. All statistical tests were considered statistically significant when P was less than 0.05. Statistical analysis was performed using GraphPad Prism (v10).

Data, code, and materials availability:

The mouse scRNAseq datasets generated in this study are in the process of being deposited in the NCBI Gene Expression Omnibus (GEO). This study reanalyzes publicly available single-cell RNA-sequencing datasets: human liver and lung fibrosis (PMID: 37027478), dataset available from the Broad Institute Single Cell Portal: https://singlecell.broadinstitute.org/single_cell/study/SCP2156/identification-of-a-broadly-fibrogenic-macrophage-subset-induced-by-type-3-inflammation-human-liver-lung-myeloid-fibrosis-scrnaseq-atlas; human liver cancer (PMID: 36352227), GSA-Human Study HRA001748, available via the National Genomics Data Center and human lung adenocarcinoma (PMID:38335304) available from the NCBI GEO (GSE253013). Due to their size, multiplex IF images cannot be easily deposited in a public repository and are available from the corresponding author upon reasonable request.