Targeted regeneration of post-radiation epithelium without promoting cancer recurrence

Abstract

Cancer radiotherapy inevitably damages normal tissues, yet therapeutic activation of pro-survival or regenerative pathways frequently increases the risk of cancer recurrence. Strategies that selectively restore normal tissue without fueling malignant regrowth remain largely undefined. Using the slow dividing yet radio-responsive salivary gland and adjacent oral squamous cell carcinoma (OSCC) as paired in vivo models, we established a high-throughput genetic screening platform to quantify post-irradiation clonal expansion in both normal and malignant epithelium. Comparative screening revealed depletion of the salivary gland dominant SWI/SNF core ATPase Smarca2 as the top-ranking selective driver of salivary epithelial regeneration, in contrast to most candidates that promoted expansion in both tissues. Smarca2 depletion specifically enhanced regeneration of acinar cells, the lineage most refractory to recovery after irradiation, by suppressing terminal differentiation programs and enabling transcriptional activation of key regenerative genes. In OSCC cells, however, this regenerative switch was uncoupled from Smarca2 depletion due to tumor-specific chromatin state remodeling. Notably, this mechanism is conserved in human salivary gland tissue and can be induced using SMARCA2-targeting PROTAC degrader, a therapeutic class under active development for cancer treatment. Together, our findings suggest that Smarca2 depletion is a potent lineage-specific mechanism to replenish radiation damaged salivary epithelium without promoting OSCC expansion, a paradigm that is underexplored.

Introduction

Efficiently suppressing cancer cell viability and expansion while preserving the often-limited regenerative capacity of co-exposed normal tissues remains a fundamental challenge in cancer therapy. Because malignant and normal cells frequently rely on overlapping growth, stress-response, and cell death pathways, therapeutic doses required to inhibit cancer growth inevitably injure adjacent healthy tissues. Moreover, indiscriminate protection or regeneration of damaged tissues carries the risk of facilitating tumor recurrence^1–6. This challenge is most evident in radiation therapy, where poorly regenerative or actively cycling while stress sensitive tissues, including exocrine glands, bone marrow, intestine, and others, are highly vulnerable to off-target injury^4,5. Among these, radiation induced salivary gland damage stands out due to the gland’s rapid response to radiation, extremely limited natural regenerative potential, and the resultant lifelong reduction of quality of life^7–9. Such injury is a major adverse effect of both localized radiotherapy for head and neck cancer^9,10 and systemic radioligand therapy used for other malignancies^11,12. Current FDA approved treatments for radiation induced salivary gland dysfunction fall into two categories: (1) saliva substitutes or secretion stimulants (e.g., pilocarpine), which are temporary and ineffective in patients with severe secretory cell loss^7–9,13, and (2) Amifostine, a broad spectrum radioprotector, which is not widely used due to its narrow therapeutic window, toxicity and potential to also protect cancer cells¹⁴. These limitations emphasize the need for in depth understanding of mechanisms that can prevent salivary gland damage and/or promote regeneration post radiation.

In preclinical murine models, several approaches have shown promise in restoring salivary gland structure and function^7–9,13. These include blocking DNA damage master regulators of the p53 gene family^15,16; inhibiting apoptosis^17,18 and senescence cascades^19,20; activating proliferation/cell survival signals like the IGF/KGF-PI3K/AKT axis^21–23 and regeneration pathways such as EDAR²⁴ and Yap^25–28. However, the central role of these signals in promoting cancer survival/growth^29–33, especially in the context of radiotherapy and recurrence^1–3 substantially limit their future applicability^7–9,13 and highlights the need for identifying mechanisms that can selectively protect/regenerate salivary gland without supporting cancer growth. Progress in this area has been further constrained by the limited number of perturbations that can be evaluated using traditional genetic models, making systematic in vivo interrogation of large candidate gene sets impractical.

The salivary gland placodes arise from the primitive oral epithelium at embryonic day 11 (E11)^34,35 and are directly exposed to amniotic fluid, creating a unique opportunity for ultrasound guided intra-amniotic lentiviral transduction^36–44 prior to placode formation, potentially enables multiplex delivery of gene-targeting constructs for genetic screening in murine models. Initially developed for epidermal targeting, we and others have demonstrated that intra-amniotic lentiviral injection at E9.5 efficiently transduces progenitors that give rise to both stratified epithelia (skin, oral mucosa) and placode-derived glandular epithelium (mammary gland)^36–44. The tissue stably transduced with shRNAs, sgRNAs and/or ORFs can then be used for rapid study of gene function during development and in adult animals^36–44.

Using lentiviral barcoding- and imaging-based lineage tracing, we discovered that each submandibular gland (SMG) originates from ~780 E9.5-transduced clones, while each parotid gland (PG) originates from ~520. These clones contribute uniformly to gland expansion during both development and post-irradiation replenishment. Consequently, a single mouse harboring paired major salivary glands can, in principle, accommodate ~2,600 uniquely traceable genetic perturbations, establishing a high-throughput in vivo screening platform. Leveraging this system, we stringently screened 255 candidate genes using an inducible shRNA pool in more than 150 transduced mice to identify regulators of post-irradiation clonal expansion. In parallel, we performed screens in OSCCs, the predominant form of head and neck cancer, to exclude perturbations that enhance tumor fitness following irradiation.

By contrasting results from the two screening paradigms, we identified depletion of Smarca2, a core SWI/SNF ATPase highly expressed in salivary glands, as the top-ranking candidate capable of selectively replenishing salivary glands without promoting OSCC growth. Smarca2 depletion enhanced the regenerative capacity of acinar cells, which normally exhibit minimal expansion potential in the absence of intervention. Mechanistically, Smarca2 directly binds promoter and enhancer regions of numerous terminal differentiation genes in acinar cells. Its depletion results in reduced H3K27Ac levels at differentiation-associated loci and suppression of their expression, while indirectly increased H3K27Ac levels at normally low-expressing regenerative gene loci and activated their transcription. In contrast, Smarca2 depletion in OSCCs fails to trigger this regenerative switch due to distinct chromatin state alterations. Using a PROTAC-based SMARCA2 degrader originally developed for cancer therapy^45–53, we efficiently activated a conserved regenerative program in human salivary gland xenografts. Together, these findings suggest that Smarca2 depletion represents a targeted regenerative mechanism for radiation induced salivary gland damage with minimal risk of promoting OSCC recurrence.

Results

Lentivirus based in vivo genetic targeting enables rapid gene function study in murine salivary glands.

To test whether the intra-amniotic lentiviral injection strategy^{36,37,39,41,44} can efficiently transduce salivary gland epithelium, we injected a Cre recombinase-expressing lentivirus (LV-Cre) into the amniotic cavity of mTmG Cre-reporter embryos at E9.5⁵⁴. We hypothesized that progeny of transduced early progenitor cells would be stably maintained within adult salivary glands (Fig. 1a and b). Light-sheet imaging of optically cleared salivary glands⁵⁵ (see Methods) revealed that, by titrating viral concentration, we could achieve a broad spectrum of transduction rate, ranging from sparse clonal labeling to near complete epithelial coverage, at postnatal day 42 (P42) (Fig. 1c). P42 represents a well-established developmental stage at which major salivary epithelial lineages (acinar, ductal, and myoepithelial) are fully specified^56–58 and is widely used as a benchmark for irradiation studies^57,59–62. 3D reconstruction of clonally labeled glands showed that transduced cells contributed to both ductal and acinar compartments, the basic functional units of the gland (Fig. 1c). Consistently, immunostaining for epithelial lineage markers confirmed that transduced cells gave rise to all three major epithelial lineages (acinar: Aqp5⁺; ductal: Krt19⁺ and/or Krt8^high; Myoepithelial: αSMA⁺ Krt14⁺) of the salivary epithelium (Fig. S1a).

Figure 1. Equipotent growth dynamics of transducible clones allows for large scale in vivo screens in murine salivary glands.

a and b, Representative image (a) and schematic (b) of ultrasound guided intra-amniotic lentiviral transduction of salivary gland progenitors. c, Light sheet imaging, 3D reconstruction of LV-Cre transduced SMG and SLG from R26^mTmG (mTmG) mice. Inset show 3D segregated ductal (magenta) and acinar structures (cyan) using Imaris. LV: lentivirus; MOI: multiplicity of infection; scale bars as indicated. d, Schematic and representative barcode composition of lineage tracer pool transduced salivary glands. e, Barcoded clone numbers are positively correlated with transduction rates across gender and gland types. f, Quantification of clone number per gland extrapolated from barcode composition and transduction rate. The results indicate that major salivary glands are originate from a defined number of transducible clones. P-values as indicated; Error bar: SD, center value: mean. g and h, 3D reconstruction of progenies derived from a single transduced clone generated by serial dilution of LV-Cre (g) and statistics across SMGs and PGs (h) validated the clonal number quantification from barcoded lineage tracing. P-values as indicated; Error bar: SD, center value: mean. i and j, Schematic (i) and representative FACS plot and post sorting cell suspension images (j) of “fix and sort” approach used isolate major epithelial lineages form salivary glands (also see Methods). k, Clone numbers extrapolated from barcode composition and transduction rate remained stable post radiation across salivary gland lineages. P-values as indicated; Error bar: SD, center value: mean.

To assess whether viral transduction was biased toward specific progenitor subsets, we injected an equal-titer mixture of four lentiviral constructs expressing distinct nuclear or membrane-localized fluorescent reporters (H2B-GFP/RFP and membrane-GFP/RFP) into the E9.5 amniotic fluid (Fig. S1b). We then plotted the fraction of cells receiving two or more viruses against the total infected population across a range of transduction rates (Fig. S1b and c). The resulting curve closely matched the prediction of a Poisson distribution (Fig. S1c; Kolmogorov-Smirnov (KS) test, P=0.9194), indicating that viral transduction of E9.5 salivary gland progenitors was random and unbiased. Together, these results demonstrate that intra-amniotic lentiviral injection enables efficient, unbiased, and tunable genetic targeting of salivary gland epithelium. This approach allows precise control over transduction clonality and provides a powerful platform for interrogating gene function across a wide dynamic range of epithelial coverage with minimal targeting bias.

Equipotent growth dynamics of transducible clones in homeostatic and irradiated salivary glands enable large-scale genetic screens.

To characterize the growth dynamics of transducible clones, we generated a barcoded lentiviral lineage-tracing library containing approximately 1.7 × 10⁶ unique barcodes (Fig. 1d). Low-multiplicity transduction (~10% transduced, MOI~0.11) resulted in single-barcode integration in the vast majority of targeted cells, enabling high-resolution quantification of individual clonal behaviors. In targeted mTmG glands, we isolated transduced (GFP⁺) cells and profiled their barcode content via our established barcode focused sequencing pipeline^36,37,39(Fig. 1d, see Methods). For each gland, we obtained comprehensive quantitative metrics, including transduction rate, total cell number, and barcode complexity. Glands with comparable GFP⁺ fractions were pooled to improve quantification accuracy (Fig. 1d and e). Barcode diversity was consistently high and stable in both SMGs and PGs, whereas sublingual glands (SLGs) exhibited substantial clonal drift. This variability likely reflects the later developmental timing of SLG placode formation⁶³, when viral titers had declined; or a mixed germ-layer origin^64,65 in which progenitors may not be uniformly transducible. We therefore focused subsequent analyses on SMGs and PGs.

Across both gland types, barcode number scaled linearly with transduction rate (R² range: 0.77–0.98), allowing extrapolation of the total number of transducible clones contributing to each gland (Fig. 1e and f). Quantitative analysis estimated that each SMG originates from approximately 780 transducible clones, while each PG arises from about 520 (Fig. 1f). These values were consistent between male and female mice, despite the significantly larger size of male SMGs⁶⁶ (Fig. 1f). Notably, average clone size was larger in male SMGs, suggesting that sex-specific clonal expansion contributes to the long-observed sexual dimorphism in murine SMG size^66–68 (Fig. S1d). Within the same gland type and sex, clone sizes were relatively uniform, indicating comparable expansion potential among transduced progenitors during development. (Fig. S1d). To validate these quantifications, we performed sparse labeling by serially diluting LV-Cre in mTmG reporter mice until minimal level or no transduction could be observed (Fig. 1g). Under these conditions, progenies of individual transduced clones were distributed throughout the gland, consistent with established models of salivary gland morphogenesis involving repeated epithelial expansion, folding, and clefting that progressively segregate clonal progeny⁶⁹(Fig. 1g). Importantly, clone numbers inferred by imaging closely matched estimates derived from barcoded lineage tracing, supporting the accuracy of our quantitative clonal analysis. (Fig. 1h).

To benchmark clonal dynamics following irradiation, we analyzed barcoded glands after exposure of P42 mice to 10 Gy head and neck focused X-ray beam followed by 90 days of recovery (Fig. S1e, see Methods). These parameters were selected because: (i) P42 represents a widely benchmarked stage at which all major epithelial lineages are fully established^56–62; (ii) 10 Gy is a commonly used single-dose regimen with effects comparable to higher or fractionated doses^57,60,70; and (iii) functional recovery of adult murine salivary glands plateaus between 60 and 90 days post-irradiation^{9,21,24,62,71}.

To fully capture clonal dynamics across epithelial lineages of the transduced gland, due to the limited availability of validated cell-surface markers, we adapted a plasma membrane selective permeabilization (“fix-and-sort”) approach^72–75 that preserves nuclear integrity for downstream gDNA and RNA analysis while enabling detection of intracellular lineage markers⁷⁶. Using this method, we isolated acinar (Aqp5⁺), ductal (Krt19⁺ and/or Krt8^high), and myoepithelial (αSMA⁺, Krt14⁺) cells from barcoded glands after irradiation and recovery (Fig. 1i and j, S1a). Barcode profiling of these purified populations revealed that the number of contributing transducible clones was preserved across all three epithelial lineages, indicating that post-irradiation regeneration is driven by equipotent clonal expansion without significant lineage bias (Fig. 1k). Collectively, the combined clonal contributions of bilateral SMGs (~780 clones each) and PGs (~520 clones each) provide the capacity to assay approximately 2,600 independent genetic perturbations encoded by pooled lentivirus in a single mouse. These results establish intra-amniotic lentiviral targeting as a powerful and scalable platform for multiplexed in vivo genetic screening in murine salivary glands.

Identification of mechanisms that selectively promote salivary gland but not OSCC expansion post radiation.

We hypothesized that a subset of regulatory mechanisms could promote salivary gland regeneration following irradiation without concomitantly enhancing cancer cell growth. To test this hypothesis, we performed parallel in vivo genetic screens in salivary glands and syngeneic OSCC models (Fig. 2a–d). To assemble a focused yet comprehensive candidate gene set, we first compared non-irradiated P42 controls with salivary gland 1- and 10-days post 10Gy irradiation at P42 (Fig. 2a). Differential expression analysis revealed extensive transcriptional change at both post-irradiation time points (Fig. 2a). Pathway enrichment analysis using g:Profiler⁷⁷ identified multiple well-characterized signaling programs with temporally distinct activation profiles (top 10 shown in Fig. 2b). In addition, we identified a substantial group of irradiation-responsive genes that did not significantly map to known gene sets (Fig. 2c). To streamline the screen while maintaining pathway coverage, we selected approximately eight core activators and suppressors for each enriched pathway, the top 60 functionally uncategorized irradiation-responsive genes. In addition, we incorporated 98 putative epithelial cell fate regulators identified by published bulk and single-cell RNA-seq datasets of salivary glands (GSE:150327/96747^78,79; Fig. 2c). This strategy yielded a final list of 255 candidate genes (Fig. 2c). We then constructed a doxycycline-inducible shRNA library targeting these candidates (Tet-On shRNA pool; Fig. 2d). Inducible knockdown allowed transient gene suppression, thereby avoiding interference with normal salivary gland development, minimizing artifacts associated with permanent loss of essential genes, and more closely modeling therapeutic intervention^80,81. Each gene was targeted by 5–7 independent shRNAs, all validated either by us or the TRC consortium^{36,37,39,42,44} (Fig. 2d).

Figure 2. Identification of salivary gland specific growth promoters post radiation.

a, Heatmap showing differentially expressed genes in salivary glands at 1 and 10 days following 10 Gy irradiation, quantified by bulk RNA-seq. b and c, Identification of 255 candidate regulators of epithelial growth after irradiation, derived from (b) highly enriched pathways identified in published bulk RNA-seq datasets and (c) putative salivary gland cell fate regulators curated from published bulk and single-cell RNA-seq datasets^78,79 (GSE150327 and GSE96747). d, Schematic of clonal expansion screens in salivary gland and OSCCs post radiation using a TetOn shRNA pool targeting 255 candidate regulators identified in c. e, Volcano plots showing the shRNA enrichment profiles from post-irradiation clonal expansion screens performed in both salivary glands and OSCC grafts. The top 10 promoters and inhibitors of clonal growth, ranked by P-value, are labeled by their corresponding target gene symbols. Red circles highlight all 6 shRNAs targeting Smarca2 across all screening groups. Significance threshold: P<1×10⁻⁴. f, Heatmap showing the enrichment fold changes of shRNAs targeting candidate genes that act as salivary gland specific growth promoters in our screens, ranked by the number of significant shRNAs per gene. The red box highlights Smarca2 as the top-ranked gene, exhibiting the highest number of salivary gland-promoting shRNAs and consistent growth promoting activity across all gland types. Significance threshold: P<1×10⁻⁴.

Using this pool, we screened for genes that promote clonal expansion post radiation in both the salivary epithelium and two radiosensitive OSCC lines, MOC1 and MOC22^82,83. To ensure that each cell received only one shRNA, we titrated the pool to achieve ~10% transduction rate (MOI~0.11) in both models^36,37,39(Fig. 1c). For salivary glands, the shRNA pool was delivered via intra-amniotic injection at E9.5 (Fig. 2d). For OSCCs, MOC1/22 cells were transduced in culture and transplanted syngeneically into the tongue of P28 C57BL/6 mice (Fig. 2d). shRNA expression was induced with doxycycline at P35, seven days prior to 10 Gy head and neck focused irradiation (Fig. S1e) at P42, allowing identification of genes that regulate radioprotection and/or regenerative outgrowth (Fig. 2d). Salivary glands were harvested at P132, and tumors were collected upon reaching ~250mm³. shRNA abundance in each sample was quantified as a proxy for clonal expansion (Fig. 2d). After normalization to baseline shRNA composition (T=0), shRNAs promoting clonal expansion were enriched, whereas those inhibiting expansion were depleted (Fig. 2e). We analyzed SMGs and PGs from 154 mice (79 males and 75 females), together with tumors from 16 animals (8 per OSCC line), and identified differentially enriched shRNAs using DESeq2⁸⁴ with a stringent significance threshold (P<1e-4, Fig. 2e). Comparison of screen hits between salivary glands and OSCCs revealed that the majority of clonal expansion promoting perturbations were shared across tissues (Fig. 2f, S2a and b). These included knockdown of established activators of p53, NF-κB, JNK, and Hippo signaling, as well as suppressors of AKT signaling, pathways previously implicated in both salivary gland regeneration and tumor growth following irradiation^{1–3,15,16,21–23,25–28,33}, thereby validating the robustness of our screening strategy (Fig. S2a). Notably, among the perturbations that selectively promoted clonal expansion in salivary glands, Smarca2 emerged as the top-ranked hit, supported by the highest number of salivary gland promoting shRNAs and consistent clonal expansion activity across all gland types (red box, Fig. 2f). In contrast, Smarca2 knockdown had no detectable growth promoting effect on OSCC clones post radiation (red box, Fig. 2f). This pronounced tissue specific divergence prompted us to focus on Smarca2 for subsequent functional and mechanistic analyses.

Smarca2 depletion selectively promotes acinar cell expansion after salivary gland irradiation.

To validate Smarca2 depletion as a driver of post-irradiation salivary epithelial expansion, we selected the most efficient shRNA from the screen (TRCN071402, Fig. 2f) for focused functional studies. To control for potential off-target effects, we generated a lentiviral rescue construct co-expressing a codon-edited, shRNA-resistant Smarca2 open reading frame (Smarca2 ORF^shRes, Fig. 3a and b). Immunostaining confirmed that doxycycline induction efficiently depleted Smarca2 protein in transduced cells and that this depletion was fully restored by co-expressing Smarca2 ORF^shRes (Figure 3A and B). Using the same temporal paradigm as in the genetic screens, we induced Smarca2 shRNA expression alone or together with Smarca2 ORF^shRes control at P35, exposed mice to 10 Gy head and neck focused irradiation at P42, and harvested salivary glands at P132. Quantification of whole gland cross-sections revealed that Smarca2 depletion robustly expanded transduced epithelial clones across salivary glands in both male and female mice compared to untreated controls (Fig. 3c and d). Importantly, co-expression of Smarca2 ORF^shRes completely suppressed this clonal expansion, confirming that the observed effect is dependent on Smarca2 depletion (Fig. 3c and d).

Figure 3: Depletion of Smarca2 specifically promotes acinar cell expansion in salivary glands post radiation.

a and b, Representative immunofluorescence image (a) and single-cell level quantification using Imaris in- and outside of lentivirus (LV) transduced clones (b) show that expression of a shRNA resistant Smarca2 ORF (Smarca2 ORF^shRes) fully rescues Smarca2 protein levels following doxycycline induced activation of TetOn shSmarca2 within the transduced clones. c and d, Representative images (c) and quantification (d) of TetOn-shSmarca2:Cre lentivirus-transduced cells (YFP⁺) showing that doxycycline induced Smarca2 shRNA expression significantly expanded the salivary epithelium post radiation. This effect was rescued by co-expression of an shRNA-resistant Smarca2 ORF (Smarca2 ORF^shRes). P-values as indicated; Error bar: SD, center value: mean. e and f, Representative images of lineage marker panel staining (e), quantification across entire gland sections (f), demonstrate that Smarca2 depletion specifically expanded acinar cells post radiation. g, tumor volume measurements in MOC1 and MOC22 allografts indicate that Smarca2 depletion does not affect MOC1 grafts but significantly suppresses MOC22 graft growth post radiation. P-values as indicated; Error bar: SD, center value: mean.

To determine which epithelial population drove this expansion, we performed lineage marker analysis on transduced clones (Fig. 3e and f). Quantification of epithelial subtypes revealed a marked and selective increase in acinar cells (Aqp5⁺), the population known to be particularly vulnerable and poorly regenerative following irradiation (Fig. 3f). In contrast, the relative abundance of ductal (Krt19⁺ and/or Krt8^high) and myoepithelial (αSMA⁺, Krt14⁺) cells was not significantly altered by Smarca2 depletion (Fig. 3f). Finally, we examined whether Smarca2 depletion affected post-irradiation growth of OSCC grafts. Consistent with the screen results, Smarca2 knockdown did not significantly alter the growth kinetics of MOC1 tumors and modestly suppressed expansion of MOC22 tumors four weeks after irradiation (Fig. 3g, S3a–c). Together, these data demonstrate that Smarca2 depletion selectively promotes post-irradiation regeneration of salivary glands by preferentially expanding the acinar cell population, while failing to support, and in some contexts suppressing, OSCC regrowth.

Smarca2 depletion does not protect salivary epithelium against radiation damage.

To define the cellular mechanisms underlying Smarca2 depletion driven acinar expansion after irradiation (Fig. 3), we examined whether this effect is mediated by enhanced epithelial protection and/or regenerative outgrowth, two well-established processes that promote salivary gland recovery following radiation injury^7–10,13,85 (Fig. 4 and 5).

Figure 4. Smarca2 depletion does not enhance protection against radiation induced damage in salivary epithelium.

a, Schematic of the experimental timeline: doxycycline was administered seven days prior to irradiation to induce shSmarca2 expression, followed by analysis at indicated time points post-irradiation to assess potential protective effects of Smarca2 depletion. b-g, Representative images (b, d, f) and corresponding quantifications (c, e, g) showed that Smarca2 depletion did not significantly impact DNA damage quantified by γH2AX, insets show zoomed in view of nucleus (b and c), apoptosis quantified by cleaved-caspase3 (d and e), and senescence quantified by senescence-associated β-galactosidase (SA-β-gal, f and g), across acinar, ductal and myoepithelial lineages of salivary epithelium. P-values as indicated; Error bar: SD, center value: mean.

Figure 5. Smarca2 depletion increases the proliferative capacity and division potential of acinar cells following irradiation.

a, Schematic of experimental design: shSmarca2 expression was induced post irradiation to control for untested protective mechanisms, followed by EdU pulse chase labeling assays to assess the contribution of regenerative mechanisms in Smarca2 depletion induced post radiation salivary epithelium expansion. b-d, Representative images (b), and quantification of EdU labeling rate immediately after 1day of EdU pulse (chase1, c), and fold change in EdU labeling rate 7 days later (chase2 versus chase1, d) show that Smarca2 depletion significantly increased both the fraction of actively dividing acinar cells (c) and their proliferative potential over time (d) following radiation, without significantly impacting ductal and myoepithelial population (c and d). P-values as indicated; Error bar: SD, center value: mean.

We first tested whether Smarca2 depletion protects salivary epithelial cells from radiation induced DNA damage, apoptosis, or senescence^{7–9,11,12,86} (Fig. 4). To establish appropriate time points for assessing these damage responses, we performed longitudinal analyses following 10 Gy head and neck focused irradiation at P42 (Fig. S4a). DNA damage (γH2AX⁺) and apoptosis (cleaved-caspase3⁺) cells peaked at P44, declined markedly by P49, and were reduced to near baseline levels by P98 (Fig. S4b and c). In contrast, senescence (SA-β-gal⁺) cells peaked at P49 and, although reduced, persisted through P98 (Fig. S4d). These temporal dynamics are consistent with previous reports^19,87–89 and defined the optimal windows for interrogating the impact of Smarca2 depletion (Fig. S4 a–d and 4a).

We therefore induced Smarca2 shRNA expression at P35, seven days prior to irradiation, and quantified DNA damage and apoptosis at P44 (2 days post-irradiation) and senescence at P49 (7 days post-irradiation) (Fig. 4a). In control glands, γH2AX staining revealed higher levels of DNA damage in ductal cells relative to acinar and myoepithelial cells, consistent with prior observations¹⁹. Smarca2 depletion did not significantly alter the proportion of γH2AX⁺ cells in any epithelial lineage (Fig. 4b and c). Similarly, analysis of apoptosis and senescence showed lineage specific patterns that mirrored those reported previously: apoptosis occurred preferentially in acinar and myoepithelial cells^9,10,90–92, whereas senescence was enriched ductal cells^{9,10,19,90,93}. Importantly, Smarca2 depletion did not significantly affect either the magnitude or lineage distribution of apoptotic or senescent cells (Fig. 4d–g). Collectively, these data indicate that Smarca2 depletion does not confer protection against radiation induced DNA damage, apoptosis, or senescence in salivary epithelium. Thus, enhanced epithelial protection is unlikely to be the primary mechanism underlying the Smarca2 depletion driven expansion of acinar cells following irradiation.

Smarca2 depletion increases the dividing acinar cell population and enhances their post-irradiation expansion potential.

We next examined whether Smarca2 depletion expand salivary gland by increasing post irradiation regeneration. To define the temporal window of observation, we first benchmarked proliferative activity using 24-hour EdU pulses administered every three days between P46 and P67 following 10 Gy irradiation at P42 (Fig. S5a). Quantification of EdU incorporation revealed that proliferative activity peaked between P49 and P52 in SMGs and at P49 in PGs (Fig. S5b and c), consistent with previous reports^{7–9,11,12,57,60,62,86,94}. Based on these results, we focused subsequent analyses on the peak proliferative window at P49. To assess the impact of Smarca2 depletion on this regenerative response, we induced Smarca2 shRNA expression after irradiation at P42, thereby also controlling for potential, untested radioprotective effects (Fig. 5a). A single 24-hour EdU pulse was then administered from P49 to P50 (Fig. 5a). Salivary glands were harvested either immediately after the pulse (chase 1) to quantify the fraction of cells entering the cell cycle, or after an additional 7-day chase (P57; chase 2) to assess the expansion potential of EdU-labeled cells during the initial pulse.

In control glands at chase 1, acinar cells exhibited significantly lower EdU labeling rates than ductal cells, consistent with the limited proliferative capacity of acinar cells reported in both murine and human salivary glands^{7–10,13,57,60,62,94} (Fig. 5b and c). In contrast, Smarca2 depletion nearly doubled the proportion of EdU⁺ acinar cells, indicating a marked increase in the number of acinar cells entering the cell cycle (Figure 5B and C). This effect was specific to acinar cells and was not observed in ductal or myoepithelial populations (Fig. 5b and c).

To evaluate the expansion capacity of dividing cells, we compared EdU labeling frequencies between chase 2 and chase 1 (Fig. 5b and d). In control glands, although a subset of acinar cells entered the cell cycle at P49, they did not expand substantially over the subsequent week, in contrast to ductal and myoepithelial cells (Fig. 5b–d). Strikingly, Smarca2 depletion significantly increased the expansion of EdU-labeled acinar cells, elevating their growth to levels comparable to those of ductal and myoepithelial lineages (Fig. 5b and d). Together, these results demonstrate that Smarca2 depletion enhances salivary gland regeneration after irradiation by both increasing the number of acinar cells that entered the cell cycle and boosting their subsequent expansion potential, thereby overcoming the intrinsic proliferative limitation of the acinar lineage^{7–10,13,90–92}.

Smarca2 depletion suppresses terminal differentiation programs and activates regenerative gene expression in acinar cells after irradiation.

Smarca2 is one of the two ATPase subunits of the SWI/SNF chromatin remodeling complex and is required for ATP-dependent nucleosome sliding or eviction at promoter and enhancer regions. Lineage-specific chromatin occupancy of Smarca2 has been shown to play a critical role in maintaining the activation of cell fate determinant transcriptional program in multiple contexts^51,95–97. Analysis of public RNA-seq datasets, together with our own data, revealed that Smarca2, rather than Smarca4, is the dominant SWI/SNF ATPase expressed in salivary gland tissue, exhibiting ~2–7-fold higher RNA levels in both human and mouse glands (Fig. S6a–d). We further validated this expression bias at protein level using quantitative western blotting with recombinant Smarca2 and Smarca4 standards (Fig. S6e and f). Given its dominant expression and known role in maintaining lineage-specific chromatin states, we hypothesized that Smarca2 depletion can induce substantial chromatin remodeling and transcriptional reprogramming that enables acinar cell regeneration following irradiation.

To directly assess chromatin state changes after irradiation and Smarca2 depletion, we performed H3K27Ac and Smarca2 ChIP-seq in acinar cells isolated from female PGs using the “fix-and-sort” approach⁷³ (Fig. 1i and j; Fig. 6a). H3K27Ac peaks identified by MACS2⁹⁸ were normalized and compared between conditions using deepTools⁹⁹; peaks with log2 fold change ≥1 and P < 0.05 were classified as differentially regulated (see Methods). Heatmap analysis separated H3K27Ac peaks into upregulated and downregulated groups following Smarca2 depletion (Figure 6A). Notably, H3K27Ac peaks that were downregulated upon Smarca2 depletion showed strong Smarca2 occupancy in control acinar cells (Fig. 6a). These downregulated peaks were highly enriched at regulatory regions of acinar and secretory lineage-defining genes, including the master transcription factors Bhlha15 (Mist1)^59,100–102 and Elf5^103–106, as well as the mucin family genes Prol1 and Mucl2, which rank among the top acinar transcripts in published single-cell RNA-seq datasets^78,107–109 (Fig. 6a and b). These findings are consistent with a model in which Smarca2 maintains active chromatin states at terminal differentiation genes, thereby constraining acinar cells in a highly differentiated, low regenerative state. In contrast, H3K27Ac peaks that were upregulated after Smarca2 depletion exhibited minimal Smarca2 occupancy in control cells and were enriched on key regulators of cell cycle progression and regeneration, including Ccnd1¹¹⁰, Cdk4¹¹⁰, Ska2^111–113 and Hmga1^114,115 (Fig. 6a and b), indicating activation of a previously restrained regenerative program.

Figure 6. Smarca2 depletion activates a regenerative transcriptional program in acinar cells.

a, ChIP-seq heatmaps of H3K27Ac and Smarca2 occupancy in acinar cells with or without Doxycycline induced Smarca2 depletion. Differentially regulated H3K27Ac peaks are ranked by signal intensity. The corresponding Smarca2 heatmap shows that regions with downregulated H3K27Ac are broadly occupied by Smarca2, whereas regions with upregulated H3K27Ac exhibit minimal Smarca2 binding in Dox- control cells. b, Representative ChIP-seq tracks illustrating that Smarca2 depletion leads to reduced H3K27Ac at key acinar differentiation gene loci, while increasing H3K27Ac at loci associated with cell cycle progression and regenerative programs. Differentially regulated ChIP peaks (P<0.05, fold change ≥1; see Methods) are marked by blue dash boxes. c-e, Experimental schematic of post-irradiation shSmarca2 induction and subsequent RNA-seq time point (c), volcano plot of differentially expressed genes (d), and Gene Ontology analysis (e), together demonstrating that Smarca2 depletion in acinar cells suppresses differentiation-associated gene expression while activating proliferation and regeneration-related programs. (f-i) Immunofluorescence staining and single cell level quantification via Imaris in and out of lentivirus transduced clones show that Smarca2 depletion significantly reduces expression of acinar differentiation genes Bhlha15(Mist1) and Prol1 (f, g), while increasing expression of proliferation/regeneration genes Ccnd1 and Hmga1 (h, i) within the shSmarca2 expressing clones post radiation.

We next asked whether fail of induction of chromatin state change driven regenerative switch explains the inability of Smarca2 depletion to promote OSCC regrowth after irradiation. In OSCC cell lines, acinar lineage genes (Bhlha15, Prol1, Mucl2, Elf5) displayed minimal H3K27Ac and Smarca2 occupancy and remained transcriptionally silent following Smarca2 depletion (Fig. S7a–c). In contrast, cell cycle and regeneration genes (Ccnd1, Cdk4, Ska2, Hmga1) were already associated with high H3K27Ac levels, indicating a constitutively active chromatin state (Figure S7A). Smarca2 occupancy at these loci was prominent in MOC22 but not MOC1 cells, and Smarca2 depletion led to reduced H3K27Ac levels and decreased expression of Ccnd1, Ska2, and Hmga1 specifically in MOC22 cells (Fig. S7b–g). This dependency likely underlies the observed suppression of MOC22 tumor growth after irradiation (Fig. 3g) and is consistent with emerging efforts to exploit SMARCA2 as a therapeutic target in cancers^45,51–53. Gene ontology analysis of OSCC transcriptomes further failed to identify activation of regenerative or cell cycle programs following Smarca2 depletion (Fig. S7h). Collectively, these results demonstrate that Smarca2 regulates distinct, lineage-specific chromatin and transcriptional programs in acinar epithelial cells versus OSCCs. Smarca2 depletion selectively suppresses terminal differentiation and activates regenerative programs in acinar cells, while failing to induce, and in some contexts disrupting, proliferative programs in OSCCs (Fig. 6 and S7).

Pharmacological degradation of SMARCA2 activates conserved regenerative program in human salivary gland xenografts.

Our genetic studies in murine salivary glands demonstrated that Smarca2 depletion robustly promotes acinar cell regeneration following irradiation. To determine whether this mechanism is conserved in human salivary tissue, we employed a short-term human salivary gland xenotransplantation model^116,117 that preserves native tissue architecture, cellular composition and transcriptional states while allowing controlled in vivo perturbation (Fig. 7a). Briefly, adult human salivary gland tissue was sectioned into ~2 mm³ fragments and transplanted beneath the kidney capsule of NSG mice^116,117 (Fig. 7a). Next, we tested the effects of pharmacological SMARCA2 depletion using ACBI2, a potent, orally bioavailable SMARCA2-targeting PROTAC originally developed for cancer therapy^45,51–53. Seven days after transplantation, the graft-bearing kidney was locally irradiated, followed by daily oral administration of ACBI2 for an additional seven days (Fig. 7a). Xenografts were then harvested, microdissected to enrich for acinar regions, and subjected to RNA-seq analysis (Fig. 7a).

Figure 7. PROTAC-mediated SMARCA2 depletion activates a conserved regenerative program in human salivary gland xenografts.

a, Schematic of the experimental workflow: human salivary gland tissue was transplanted under the kidney capsule, focused irradiated, treated with ACBI2 (SMARCA2 targeting PROTAC), and harvested for analysis to assess post-radiation regenerative responses. b, Immunofluorescence staining and single-cell quantification by Imaris confirm robust SMARCA2 depletion in ACBI2-treated human salivary gland xenografts. c-i, Volcano plot of differentially expressed genes (c), immunofluorescence staining and single-cell level quantification of acinar differentiation genes via Imaris (BHLHA15(MIST1), MUC7; d, e) and regeneration/proliferation genes (CCND1, HMGA1; f, g); Gene Set Enrichment Analysis (GSEA; h); and Gene Ontology (GO) enrichment analysis (i) demonstrate that SMARCA2 depletion induces a conserved regenerative transcriptional program post radiation. This is characterized by: downregulation of key acinar differentiation genes (highlighted in c, d, e); upregulation of regeneration and proliferation genes (highlighted in c, f, g); enrichment of murine shSmarca2-driven gene signatures (h) GO enrichment of cell cycle and stemness genes among upregulated transcripts, and epithelial differentiation genes among downregulated transcripts (i).

Immunofluorescence staining confirmed efficient SMARCA2 degradation in human xenografts treated with ACBI2 (80 mg/kg), with an average ~71% reduction in protein levels (fold change = 0.29), consistent with reported pharmacodynamic effects of ACBI2⁵² (Fig. 7b). Transcriptomic profiling revealed that ACBI2 treatment induced a gene expression program highly similar to that observed following genetic Smarca2 depletion in murine acinar cells. Specifically, acinar and secretory lineage defining transcription factors BHLHA15 and ELF5, along with human salivary gland enriched mucins MUC1 and MUC7, were markedly downregulated (Fig. 7c). Notably, murine acinar markers Prol1 and Mucl2 lack human homologs and were therefore not assessed. In parallel, we observed robust induction of regeneration and proliferation genes, including CCND1, CDK4, SKA2, and HMGA1 (Fig. 7c). These transcriptional changes were corroborated at the protein level. Immunostaining demonstrated reduced expression of BHLHA15 (MIST1) and MUC7, alongside increased CCND1 and HMGA1 levels in acinar cells of ACBI2 treated xenografts compared to controls (Fig. 7d–g). This molecular signature closely recapitulates the regenerative transcriptional switch observed in Smarca2 depleted murine salivary glands (Fig. 6c–i).

To assess whether this response reflected a conserved global transcriptional program, we performed Gene Set Enrichment Analysis (GSEA)¹¹⁸ using gene sets defined by Smarca2-induced and Smarca2-repressed transcripts in murine acinar cells (Fig. 6d and 7h). GSEA revealed strong enrichment of Smarca2-induced gene sets in ACBI2-treated human xenografts, whereas Smarca2 repressed gene sets were preferentially enriched in DMSO treated controls (Fig. 7h). Consistently, Gene Ontology analysis identified cell cycle progression and regenerative pathways as the top enriched categories following ACBI2 treatment, while epithelial differentiation programs were broadly suppressed (Fig. 7i). Together, these data demonstrate that pharmacological degradation of SMARCA2 using the orally available PROTAC ACBI2 activates a conserved regenerative transcriptional program in irradiated human salivary gland xenografts. These findings provide strong evidence that SMARCA2 targeting represents an efficient strategy to enhance human salivary gland regeneration following irradiation.

Discussion

A central focus of cancer biology over the past four decades has been the identification of tumor-specific growth and survival mechanisms as the foundation for molecularly targeted therapies¹¹⁹. These efforts have yielded substantial improvements in patient survival and quality of life across many cancer types¹¹⁹. In contrast, the molecular mechanisms that can selectively regenerate cancer therapy damaged normal tissues, and whether such programs can be activated without simultaneously promoting cancer growth, remain far less well understood. Recent work in non-cancer contexts has demonstrated that precise molecular targeting can elicit lineage-specific regenerative responses¹²⁰. A striking example showed that, following lung injury, Fzd5/6-selective agonists activate Wnt signaling in alveolar epithelial cells while avoiding fibrotic expansion of fibroblasts¹²¹. However, in the context of radiation induced salivary gland injury, among the most debilitating and clinically intractable side effects of radiotherapy, previous regenerative strategies have largely lacked such lineage and tissue specificity^7–9,13.

By systematically interrogating the function of 255 candidate genes across parallel in vivo screens in salivary glands and tumor grafts spanning more than 150 animals, we identified Smarca2 depletion as a selective and potent driver of salivary gland regeneration following irradiation (Fig. 2). Smarca2 depletion robustly enhanced the regenerative and proliferative capacity of acinar cells, the epithelial lineage that is most vulnerable to radiation damage and the least capable of spontaneous recovery (Fig. 3–5). Notably, this regenerative effect seems to be in contrast to the well-established role of SMARCA2 as a therapeutic vulnerability in cancer. For instance, in SMARCA4-deficient tumors, residual SWI/SNF activity becomes dependent on SMARCA2, resulting in synthetic lethality upon SMARCA2 loss^48–50. More recent studies have extended this paradigm to tumors with intact SMARCA4^45–47, such as AR⁺/FOXA1⁺ prostate cancer, where combined SMARCA2/4 inhibition disrupts oncogenic transcriptional programs sustained by AR, FOXA1, ERG, and MYC⁴⁵.

Our findings reconcile these seemingly opposite outcomes, growth suppression in cancer and regenerative expansion in normal epithelium, through a shared underlying mechanism. Disruption of Smarca2 dependent chromatin remodeling selectively suppresses cell state defining transcriptional programs, but the functional consequence of this suppression depends critically on cellular context. Consistent with the established role of SWI/SNF complexes in preserving lineage identity during development^51,95–97, we observed strong Smarca2 occupancy at regulatory regions of master transcription factors of secretory lineage (Bhlha15^59,100–102, Elf5^103–106) as well as acinar marker mucin genes (Prol1, Mucl2)^78,107–109. Depletion of Smarca2 suppressed these differentiation-associated programs in both murine salivary glands and human salivary gland xenografts (Fig. 6 and 7). This finding parallels observations in the pancreas, where silencing of Bhlha15 and subsequent attenuation of the acinar differentiation program are required for effective regeneration following pancreatitis¹²². Together, these results support the concept that terminal differentiation programs impose a barrier to regeneration, one that can be transiently relaxed through targeted chromatin remodeling. Importantly, Smarca2 depletion not only suppressed differentiation-associated genes but also activated cell cycle and regenerative regulators in acinar cells. In contrast, in OSCC and other cancer cells, many of these same genes were already embedded within active chromatin states and, some contexts, required Smarca2 to sustain their expression^45–47. Consequently, Smarca2 depletion failed to activate regenerative and cell cycle programs in OSCCs, underscoring the lineage specific dependency on Smarca2 mediated chromatin regulation.

The rapid development of SWI/SNF-targeting compounds for cancer therapy has yielded highly potent and selective inhibitors and degraders of SMARCA2¹²³. Leveraging this pharmacological toolkit, we demonstrate that the SMARCA2-targeting PROTAC ACBI2⁵² efficiently activates a conserved regenerative transcriptional program in irradiated human salivary gland xenografts (Fig. 7). In summary, our work establishes a robust genetic screening framework for identifying mechanisms that drive targeted regeneration of radiation-damaged salivary epithelium and positions Smarca2 depletion as a promising molecular strategy for salivary gland specific regeneration without promoting OSCC growth.

Methods

Animals.

We used approximately equal numbers of male and female animals throughout the study. R26^mTmG (JAX007576) and R26^lsl-YFP (JAX 007903) Cre-reporter mice were all crossed to the C57BL/6 background. NOD.Cg-Prkdc^scid Il2rg^tm1Wjl/SzJ (NSG, JAX 005557) mice were obtained from The Jackson Laboratory. Mice were housed and cared for in an AAALAC-accredited facility at the Icahn School of Medicine at Mount Sinai. All animal experiments were conducted in accordance with the ethical regulations of the Icahn School of Medicine at Mount Sinai and IACUC-approved protocols. Project license number: IPROTO202100000070.

Tissue clearing and light sheet imaging.

Tissue clearing was performed as previously described⁵⁵. Briefly, transduced salivary glands were perfused in PBS for 30 minutes at 4°C and fixed in 4% PFA at room temperature for 2 hours. Fixed glands were subsequently incubated in B1n (2% glycine, 0.1% Triton X-100, 1 μM NaOH, 0.08% NaN₃) overnight; SBiP (200 μM Na₂HPO₄, 0.08% SDS, 16% 2-methyl-2-butanol, 8% 2-propanol; pH 7.4) for 24 hours; PB (16 mM Na₂HPO₄, 4 mM NaH₂PO₄; pH 7.4) for 3 hours; PTS (20% 2,2′-thiodiethanol in PB) for 2 hours; and ACB (44.2% 2,2′-thiodiethanol, 48% Iohexol in PB) for 24 hours before imaging with a LifeCanvas SmartSPIM light sheet microscope. Image datasets were analyzed and 3D-reconstructed with Imaris (Oxford Instruments).

Lentiviral constructs.

Barcoded lineage tracer pools were subcloned as previously described³⁷. Briefly, random 10-mer barcodes were subcloned into the pLKO-Cre lentiviral backbone^37,39,41 between EcoRI and AgeI sites. The ligation products were electroporated into ElectroMAX (Invitrogen) competent cells using the Gene Pulser Xcell Electroporation System (Bio-Rad) and plated on six 500 cm² bioassay plates to achieve ~1.7×10⁶ random barcode inserts. To achieve inducible RNAi-mediated gene depletion and minimize off-target effects in the TetOn shRNA pool, we subcloned 5–7 top-scoring and/or validated shRNA hairpins from the TRCportal database of the Broad Institute¹²⁴ to target each candidate, as previously described^37,39. shRNA hairpins were subcloned into the Tet-pLKO-Cre virus, which we modified from Tet-pLKO-puro (Addgene: #21915) by replacing puroR with Cre recombinase. To ensure adequate representation of each shRNA within the TetOn shRNA pool, electroporated E. coli were plated on 20 500 cm² bioassay plates to ensure >100× coverage per construct. Co-expressing shSmarca2 and shRNA resistant Smarca2 ORF (Smarca2 ORF^shRes) was achieved by subcloning shSmarca2 hairpin into pTRE:shRNA-pGK:rtTA-Cre (Addgene: #68471) and replacing inducible GFP with Smarca2 ORF^shRes.

Lentivirus production and intra-amniotic injection.

Large-scale production, concentration, and titration of lentivirus were performed as described before^42,44. Briefly, 120 ml of viral supernatant produced by two T300 flasks of 293TN cells was concentrated using a Centricon Plus-70 centrifugal filter unit (100 kDa, Millipore). The concentrated supernatant was pelleted via ultracentrifugation (Beckman Optima MAX-TL ultracentrifuge, MLS-50 rotor; 45,000 rpm, 1.5 hr). The lentiviral pellet was resuspended in 70 μl of Viral Resuspension Buffer (VRB; 20 mM Tris pH 8.0, 250 mM NaCl, 10 mM MgCl₂, 5% sorbitol) for storage. Lentivirus stocks were serially diluted and titrated using genomic DNA from LV-Cre-infected, FACS-sorted salivary gland cells as qPCR standards. Barcoded virus was diluted in VRB according to titration results to ensure ~10% transduction efficiency intra-amniotic. Lentiviral transduction in vivo was performed via ultrasound-guided intra-amniotic microinjection as previously reported^41,42,44. Briefly, the developmental stage of mouse embryos was determined using a Vevo 3100 LT animal ultrasound imager with MS550D transducer (Fujifilm). One μl of diluted virus was injected into the amniotic cavity of E9.25–9.75 embryos using a glass capillary needle fitted on a Celltram Vario microinjector (Calibre Scientific).

Head and neck focused irradiation.

All irradiations were performed using an X-Rad 320 Precision Irradiator (Precision X-Ray) operating at 320 kV. Mice received a single dose of 10 Gy. Animals were anesthetized with ketamine and xylazine and positioned prone on the irradiation platform. The X-ray field, visualized using the built-in coaxial halogen light, was focused on the head and neck region to ensure complete coverage of the major salivary glands and oral cavity (Fig. S1e). Following irradiation, mice were placed on a 37.5 °C heating platform for recovery. Gray coat pigmentation of the irradiated region, indicative of melanocyte depletion and effective irradiation, was observed approximately 2 weeks post-irradiation (Fig. S1e).

In vivo lineage tracing and genetic screen.

Barcoded in vivo lineage tracing was performed as we previously reported in mammary glands³⁷. Briefly, pLKO-Cre-based barcoded lentiviral lineage tracing pool-infected R26^mTmG salivary glands were minced and digested in 0.25% collagenase (Worthington-biochemical) with 2 mg/ml dispase II (Gibco) at 37°C for 1 hour to release stromal cells. The tissue was washed with PBS and further incubated in 0.125% Trypsin-EDTA (Gibco) at 25°C for 50mins to release epithelial cells. The barcoded GFP⁺ cell population was isolated using a MACSQuant^® Tyto^® Cell Sorter (Miltenyi Biotech). Salivary glands with similar transduction rates were pooled to increase quantification accuracy (Fig. 1e and f). To further distinguish the three major epithelial populations from transduced cells, isolated salivary gland cells were fixed and stained with Aqp5 (AB15858, 1:100, Millipore); Krt19 (EP1580Y, 1:100; Abcam), Krt8 (TROMA-I, 1:50; DSHB); and Krt14 (BP5009, 1:100, Origene), α-SMA (1A4, 1:100, Cell Signaling) to identify acinar, ductal, and myoepithelial populations, respectively (Fig. 1i–k). For post radiation clonal expansion screens, transduced salivary glands were classified by gender and gland type, and pooled for tissue lysis using the Bead Mill 24 Homogenizer (Fisher) with ceramic beads in three 10-second pulses. Post radiation OSCC grafts were also pooled and lysed under the same conditions. gDNA from all samples was extracted using the QIAamp DNA Tissue Mini Kit (Qiagen), while fixed samples were pretreated with 5 M NaCl at 65°C to de-crosslink DNA. Barcode pre-amplification, sequencing, and normalization to T=0 counts were performed as previously described^37,39–41. Briefly, amplicons generated via PCR pre-amplification of the barcode region were made into sequencing libraries using the NEBNext Ultra II DNA Library Prep Kit (NEB). Sequencing was performed on an Illumina NovaSeq 6000 at the Icahn School of Medicine at Mount Sinai Genomics Core Facility. Short reads were trimmed to barcode regions with FASTX-Toolkit¹²⁵ and mapped to the predetermined barcode list using BWA¹²⁶. Relative barcode enrichment was calculated using DESeq2⁸⁴, and individual significant shRNA constructs were scored using a p-value cutoff of 1e-4.

Immunofluorescence staining and imaging analysis.

The following primary antibodies were used: Rabbit anti-Aqp5 (AB15858, 1:500; Millipore Sigma); Mouse anti-αSMA (1A4, 1:200; Cell Signaling); Rabbit anti-Mist1 (D7N4B, 1:200; Cell Signaling); Rat anti-E-cadherin (ECCD-2, 1:300; Thermo Fisher), Rabbit anti-Krt19 (EP1580Y, 1:400; Abcam), Rat anti-Krt8 (TROMA-I, 1:100; DSHB); guinea pig anti-Krt14 (BP5009, 1:100, Origene); Rabbit anti-Smarca2 (HL1115, 1:500; Thermo Fisher); Rabbit anti-γH2AX (3F2, 1:400; Thermo Fisher); Rabbit anti-active Caspase-3 (MAB835, 1:200; R&D Systems); Rat anti-Prol1 (EB10617, 1:200; Everest Biotech); Rabbit anti-Ccnd1 (E3P5S, 1:400; Cell Signaling); Rabbit anti-Hmga1 (5J0E6, 1:200; Thermo Fisher); Rabbit anti-MUC7 (PA5–52292, 1:200; Thermo Fisher). Tissues were processed for immunostaining as previously described^37,39,41,44 and mounted in ProLong Gold with or without DAPI (Life Technologies). Fluorescence-based senescence-associated β-galactosidase staining was performed using the Cell Meter^™ Cellular Senescence Activity Assay Kit (AAT Bioquest). Confocal images were captured on an Andor BC43 confocal system (Oxford Instruments) using 20×/NA 0.85 air objective (Nikon) and 60×/NA 1.4 oil objective (Nikon). Tiled image datasets were stitched using Fusion2 (Oxford Instruments) and analyzed with Imaris (Oxford Instruments) using cell segmentation and single-cell quantification functions.

ChIP-seq analysis.

For ChIP-seq analysis in acinar cells or OSCC cell lines, 1 × 10⁶ cells were used per replicate for H3K27Ac ChIP, and 5 × 10⁶ cells were used per replicate for Smarca2 ChIP. Cells were processed using the Pierce Magnetic ChIP Kit (Thermo Fisher Scientific) according to the manufacturer’s instructions, with rabbit anti-H3K27Ac antibody (D5E4, 1:200, Cell Signaling) or rabbit anti-Smarca2 antibody (HL1115, 1:200, Thermo Fisher). Purified ChIP DNA was used to generate sequencing libraries with the NEBNext Ultra II DNA Library Prep Kit (NEB) and sequenced on an Illumina NovaSeq 6000 platform. Sequencing reads were aligned to the reference genome using Bowtie2¹²⁷, and coverage tracks were generated and normalized using the bamCoverage function from deepTools⁹⁹. ChIP peaks were identified using MACS2⁹⁸ with an FDR q-value cutoff of 0.01. Peaks from biological replicates were merged using bedtools¹²⁸ and annotated with HOMER¹²⁹. Differentially regulated H3K27Ac peaks were identified using the bigWigAverageOverBed function from deepTools⁹⁹, with thresholds of log2 fold change ≥ 1 and P < 0.05. Sorted ChIP-seq signal heatmaps were generated using EaSeq¹³⁰.

Mouse kidney capsule transplantation of human salivary gland tissue.

Non-irradiated human salivary gland tissues were obtained from consented patients undergoing head and neck squamous cell carcinoma surgeries at the Department of Otolaryngology-Head and Neck Surgery at the Icahn School of Medicine at Mount Sinai. All specimens were procured, deidentified, and confirmed to be tumor free by the Icahn School of Medicine Biorepository and Pathology Core under an IRB-approved protocol: STUDY 12–00145. 8–10 weeks old NSG mice were housed in a specific pathogen free facility at the Icahn School of Medicine at Mount Sinai. During surgery, the left kidney was exposed, and ~2 mm³ of fresh human salivary gland tissue was implanted under the kidney capsule using a sharp PE50 tubing (BD), as previously described^116,117. Seven days post transplantation, the graft bearing kidney area was subjected to 10 Gy of focused X-ray irradiation using the X-Rad320 Precision Irradiator (Precision X-Ray). Transplanted mice were then treated daily with either ACBI2 (80 mg/kg) or vehicle control (2.5% DMSO) via oral gavage for 7 days before tissue collection for histological and RNA-seq analysis. A total of 6 transplants from 3 independent donors (2 grafts per donor) were used for each treatment condition (ACBI2 or DMSO).

RNA-seq and gene set analyses.

In R26^lsl-YFP salivary glands, we isolated Cre-expressing lentivirus-transduced acinar cells (YFP⁺ Aqp5⁺) using MACSQuant^® Tyto^® Cell Sorter. For fixed human salivary gland xenografts, acinar cell enriched regions were microdissected from ten 20 μm slides using the LMD6500 laser microdissection system (Leica). RNA from fixed cells/tissues was extracted using the miRNeasy FFPE Kit (Qiagen), as previously described for extracting RNA from fixed muscle stem cells^73,74. RNA quality was assessed with an Agilent 2100 Bioanalyzer; all samples had RNA integrity numbers (RIN) > 8. Libraries were prepared using the NEBNext Ultra II mRNA Sample Prep Kit (NEB) and sequenced on an Illumina NovaSeq 6000. Reads were mapped to reference mouse (mm10) and human (hg38) genomes using STAR¹³¹, summarized with featureCounts¹³², and normalized using DESeq2⁸⁴. Differential expression was calculated with DESeq2 using a p-value cutoff of 0.01 and a fold-change cutoff of 2. GSEA^118,133 analyses were performed using GSEA v4.3.2, and Gene Ontology enrichment was assessed with g:Profiler¹³⁴.

Statistics and reproducibility.

All quantitative data are expressed as mean ± s.d. Differences between two groups were assessed using a two-tailed Student’s t-test. Differences were considered significant when P < 0.05. Quantitative data were visualized using Prism (GraphPad Software). All quantitative data were collected from experiments performed with at least three samples or biological replicates. Sample size was not predetermined, and experiments were not randomized.

Data availability.

ChIP-seq data supporting the findings of this study have been deposited in the Gene Expression Omnibus (GEO) under accession code GSE304048.