SOX9 drives KRAS-induced lung adenocarcinoma progression and suppresses anti-tumor immunity

SOX9 significantly contributes to LUAD progression

SOX9 is overexpressed in numerous cancers [33] and its upregulation is associated with worse survival in human NSCLC [34]. Utilizing the TCGA LUAD dataset, we confirmed that patients with SOX9-high tumors (top 20% of tumor samples) showed significantly shorter survival (p = 0.0039, Fig. 1a) and SOX9-low patients (lowest 15% of tumor samples) showed significantly longer survival.

We utilized the Kras^LSL-G12D mouse LUAD model to determine if SOX9 directly contributes to lung carcinogenesis. To test this, we first achieved Sox9 loss-of-function through a pSECC CRISPR-mediated genome editing system [35], in which CRISPR and Cre were combined to knockout Sox9 and activate Kras^G12D, respectively. We designed three guide RNAs to target Sox9, taking tdTomato guide RNA (sgTom) as the control [35], and one was selected for the in vivo studies (sgSox9.2-pSECC). At weeks 18, 24 and 30 after intratracheal delivery, both tumor number (Fig. 1b, d) and burden (Fig. 1c, d) were significantly decreased (p = 0.018 and 0.029, respectively) when Sox9 was knocked out.

To determine if SOX9 accelerates Kras-induced lung tumor progression, we compared the SOX9 expression patterns to tumor grade distributions [36]. By 18 weeks, both sgTom and sgSox9.2 lungs showed only grade 1 or 2 tumors and did not display SOX9 expression by immunohistochemistry (IHC) (data not shown). Strikingly, by 24 and 30 weeks, 12 grade 3 tumors were observed in the control lungs, while only one grade 3 tumor was identified in the Sox9 knockout lungs (Fig. 1d–f). Most of the grade 2 tumors at 24 and 30 weeks were SOX9-negative (SOX9⁻, Fig. 1e, f) but some showed a low percentage of SOX9-positive (SOX9⁺) cells (0.52–19.05%, Fig. 1d, f). In contrast, SOX9 was highly upregulated in all the grade 3 tumors (9.87–48.77%), including the one grade 3 tumor from the Sox9 knockout mouse (Fig. 1f). Because cell proliferation is a clinical marker for tumor progression, we next asked if SOX9-associated tumor progression correlates with increased cell proliferation. Indeed, we observed a significantly higher percentage of Ki67 positive (Ki67⁺) cells in SOX9⁺ tumors (p = 0.00092, Fig. 1e, g). In conclusion, Sox9 loss-of-function significantly decreases both tumor number and tumor burden, as well as suppresses LUAD tumor progression, and SOX9 is mainly expressed in larger, proliferative, and high-grade tumors. These data suggest Sox9 is a driver of LUAD tumor growth and progression.

To confirm that loss of Sox9 reduces Kras^G12D-induced tumorigenesis, we utilized Kras^LSL-G12D;Sox9^w/w (KS^w/w) and Kras^LSL-G12D; Sox9^flox/flox (KS^f/f) genetically engineered mouse models (GEMM), with lenti-Cre intratracheal delivery to achieve Kras^G12D-driven tumors with or without Sox9. As shown in Fig. 1h, KS^f/f mice had significantly longer survival than KS^w/w mice (p = 0.0012). At the endpoint (380 days), we sacrificed all surviving mice (9 KS^w/w and 18 KS^f/f) and compared their tumor burden and tumor grade distribution. Consistent with the CRISPR-generated Sox9 knockout model, lenti-Cre induced Sox9 knockout GEMMs showed significantly reduced lung tumor burden (p = 0.011, Fig. 1i, j). Moreover, in KS^w/w mice, we confirmed that SOX9 was highly upregulated in all grade 3 tumors, and these high-grade tumors showed dysplastic features such as larger nuclear-cytoplasmic ratio, pleomorphic nuclei, prominent nucleoli, nuclear molding and/or signet ring structures. Consistently, most of the grade 2 tumors were SOX9⁻, or had a very low percentage of SOX9⁺ cells (Fig. 1i). We noted that SOX9⁺ tumors were much larger than the SOX9⁻ tumors and they were responsible for the higher tumor burden (Fig. 1i). Once more, significantly increased Ki67⁺ cells were observed in SOX9⁺ tumors (Fig. 1g). In contrast, in KS^f/f mice, we observed significantly fewer grade 3 tumors compared to KS^w/w mice (p = 0.049; p = 0.0008, SOX9⁺ tumors excluded from KS^f/f mice). Noteworthy, 33% of these grade 3 tumors were SOX9⁺ (Fig. 1k). The few SOX9⁻ grade 3 tumors that were present displayed only a portion of grade 3 features such as pleomorphic nuclei, prominent nucleoli and nuclear molding, but they did not show a high nuclear-cytoplasmic ratio (Fig. 1i). Therefore, these data demonstrate that SOX9 contributes to Kras^G12D-driven lung tumor development and progression.

SOX9 increases Kras^G12D tumor cell proliferation

To demonstrate that Sox9 upregulation directly contributes to lung tumor cell growth, we utilized two Kras^G12D mouse lung tumor mTC11 and mTC14 cell lines with low endogenous SOX9 expression (Fig. 2a and Fig. S1a) and performed in vitro gain- and loss-of-function assays (Fig. 2; Fig. S1). In a two-dimensional (2D) culture system, we did not observe growth differences between control and Sox9 overexpression in either of the models (data not shown). We further performed growth comparisons through a three-dimensional (3D) tumor organoid culture assay as well as an allograft transplantation model in syngeneic C57BL/6 J mice. As shown in Fig. 2b, d, f, i, the average tumor organoid sizes and number of cells per organoid of mTC11-mSox9OE were significantly more than those of mTC11-EV control. We examined SOX9 and Ki67 expression by IHC (Fig. 2f) and found that Ki67 expression was significantly correlated with SOX9 expression in mTC11-EV (r = 0.75, Fig. S1b) and mSox9OE organoids (r = 0.58, Fig. S1b), and mTC11-mSox9OE organoids had significantly higher Ki67 expression than in mTC11-EV organoids (p = 0.0039, Fig. 2h). These data confirmed that higher SOX9 expression contributes to lung tumor cell growth.

Since most of the tumor organoids harbored SOX9⁺ cells (Fig. S1c), we knocked out Sox9 in these models by CRISPR-Cas9 genome editing (Fig. 2a; Fig. S1a) to test if loss of Sox9 decreases lung tumor organoid and allograft growth. Two guide RNAs were selected (sgSox9.2 and sgSox9.3). To avoid potential clonal selection bias and to match the in vivo studies, early passage knockout cells at the population level were used. As expected, both mTC11-sgSox9.2 and sgSox9.3 organoids were significantly smaller (Fig. 2c, e), with fewer cells (Fig. 2g, i), and decreased Ki67 expression (Fig. 2g, h) as compared to controls. To exclude the possibility of off-targeting effects of the Sox9 CRISPR guide RNAs, and to confirm SOX9 function in the observed tumor cell growth, we reintroduced human SOX9 (hSOX9), which we designed to avoid the cleavage by the guide RNAs, to Sox9 knockout cell lines (sgSox9.2 and its control sgTom), and we confirmed endogenous, transgene and knockout SOX9 expression (Fig. S1d). Consistently, hSOX9 significantly increased the size of organoids compared to empty vector (all p < 0.01, Fig. S1e, f). Through further SOX9 and Ki67 IHC examination (Fig. S1g), exogenous overexpression of hSOX9 significantly restored the Sox9 knockout phenotypes, including significantly higher proliferation compared to knockouts, to levels that were similar to controls (Fig. S1h–k).

To test if Sox9 directly contributes to tumor burden, we measured tumor growth using Sox9-overexpressing and Sox9-knockout lung tumor cells in vivo through subcutaneous allografts. We observed that mTC11-mSox9OE tumors grew significantly faster (p = 0.0016) whereas mTC11-sgSox9.2 and mTC11-sgSox9.3 tumors grew significantly slower (both p values < 0.05), as compared to their respective control tumors (Fig. 2j). These data were replicated with mTC14 tumors (Fig. S2a, b). As expected, Sox9 overexpressed subcutaneous tumors displayed significantly more Ki67⁺ cells (p = 0.045, Fig. 2k) and fewer apoptotic cells (p < 0.001, Fig. 2l). Additionally, in lung cancer SOX9 was reported to promote cancer stem cells (CSC) [25], driving tumor initiation and progression. To examine if our observed phenotypes are partially due to CSC expansion, we examined expression of CD44 (a CSC marker) and found its expression was not significantly altered by Sox9 expression in the mouse lung tumor cells (Fig. S3a) although there was a significant correlation between expression of SOX9 and CD44 in a comprehensive panel of NSCLC cell lines (Fig. S3b). These results in total confirm that Sox9 contributes to lung tumor progression and increased tumor burden at least partly through increased proliferation and decreased apoptosis.

RNA sequencing revealed immune suppression in Sox9-overexpressed Kras^G12D lung tumors

To explore the mechanisms by which SOX9 contributes to lung tumor growth and progression, we performed RNA sequencing (RNA-seq) on mTC11-EV and mTC11-mSox9OE subcutaneous syngeneic allografts. A total of 373 significantly up-regulated and 471 significantly down-regulated genes (Fig. 3a, Supplementary Table 1) from mSox9OE tumors were detected and utilized to perform gene set enrichment analysis (GSEA). All differentially regulated pathways with adjusted p values less than 0.05 are shown in Fig. 3b and Supplementary Table 2. As expected, mSox9OE significantly upregulated the cell proliferation pathway (Fig. 3b), and other known SOX9-regulated pathways, including response to extracellular stimulus, RAS signal, and duct morphogenesis. Notably, the pathway immune system process displayed the most significant downregulation (Fig. 3b). To better understand how SOX9 modulates immune suppression, we performed GSEA on immune-related gene sets (Supplementary Table 3). Using cutoff p value < 0.05 and FDR < 0.25, we found that 14 immune subsets were significantly downregulated in mSox9OE tumors (Fig. 3c, Supplementary Table 2). Finally, 15 immune gene subsets and 65 included genes (Fig. 3c) were found. We investigated these genes in Kras^G12V mouse lung tumors, which were collected by microdissection at 10 months after tumor initiation (GSE65258) [37]. Compared to normal mouse lung tissues, we selected 3 tumors as Sox9-high and 3 as Sox9-low (Fig. 3d). Through the comparison of Sox9-high vs. Sox9-low tumors, we found that 22 out of 65 (34%, labeled in red) were replicated in Kras^G12V mouse LUAD (Fig. 3c, Supplementary Table 4). These 22 genes included CD8⁺ T cell specific genes (Cd3g, Zap70, and Cd8a) and natural killer (NK) cell specific genes (Nkg7 and Klrd1), suggesting that Sox9 expression in lung tumors is correlated with suppression of CD8⁺ T cells and NK cells.

SOX9 is significantly associated with suppression of tumor immunity in LUAD

Based on the observed suppression of immune signatures, we performed subcutaneous transplantation using Sox9-overexpressing and Sox9-knockout mouse-derived lung tumor cells in NOD-SCID-gamma (NSG) mice, lacking mature T, NK and B cells [38]. Although mSox9OE increased and Sox9 knockout decreased tumor volumes in NSG mice, the tumor growth differences were much more substantial in immunocompetent C57BL/6J mice (around twofold change) than in the immunocompromised NSG mice (Fig. 4a, Fig. S2b, c). These data suggested that SOX9 might contribute to resistance to anti-tumor immunity.

To better understand SOX9-related immune suppression, we performed flow cytometric analysis of T cells, B cells, NK cells, neutrophils, monocytes, macrophages, and dendritic cells in mTC11-EV and mSox9OE syngeneic allografts (Supplementary Table 5). We also investigated immune cell subtypes (naïve, memory and effector) and their functions (exhaustion markers PD-L1, LAG3 and Tim3, and cell proliferation) within the CD8⁺, CD4⁺ T cell and T regulatory cell populations. As shown in Fig. 4b, c, overexpression of Sox9 caused a significant decrease in tumor infiltrating T cells (p = 0.030). Within the T-cell population, we noted that the percentage of CD8⁺ T cells was significantly decreased (p = 0.025) in mSox9OE tumors. These tumor infiltrated CD8⁺ T cells expressed more exhaustion marker LAG3 (p = 0.014) and less proliferative marker Ki67 (p = 0.050), which indicate they were in more exhausted states and had weaker proliferative capability in mSox9OE tumors. We did not observe significant differential expression of LAG3 or Ki67 in CD4⁺ T cells or T regulatory cells. In addition, we normalized the percentages of immune cell sub-populations to the whole tumor (Fig. 4d). In the mSox9OE tumors, CD45⁺ immune cells were decreased 2.2-fold, and T cell subpopulations showed the largest decrease, especially CD8⁺ T cells (6.4-fold decrease). We verified the decreased CD8⁺ cell infiltration in Sox9 overexpressed lung tumors by both CD8a RT-qPCR and IHC, which were both statistically significant (Fig. 4e, f). We further assessed representative genes for CD8⁺ T cells (Zap70 and Btla), NK cells (Nkg7 and Ncam1), and M1 macrophages (Tnf and Nos2) by RT-qPCR in Sox9-overexpressed and Sox9-knockout tumors. In addition, we examined functional markers of CD8⁺ T cells, NK cells and dendritic cells in the subcutaneous tumors and Kras^G12V mouse lung tumors (Fig. 4g, Supplementary Table 4). All these genes showed an inversely correlated association with Sox9 (Fig. 4e, g). To confirm our observations that SOX9 induces immunosuppression, we performed immunoprofiling on mTC11-Sox9 knockout subcutaneous tumors. Loss of Sox9 significantly increased tumor immune cell infiltration (Fig. S4a, b). CD8⁺ effector T cells and CD80⁺ dendritic cells were significantly increased in Sox9-knockout tumors (Fig. S4b, c, d). Thus, flow cytometry along with RNA-seq, RT-qPCR, and IHC confirmed that SOX9 suppresses anti-tumor immunity in Kras-driven mouse LUAD.

Recent studies have established that the chemokine receptor CXCR3 plays a dominant role facilitating CD8⁺ T cells across tumor vessels and infiltrating into the tumors [39, 40]. CXCR3 ligand CXCL9 has been demonstrated to be downregulated in tumors lacking a CD8⁺ T cell infiltrate [41]. CXCR6 signaling can also increase retention of memory CD8⁺ T cells in the tumor microenvironment [42]. We noticed that Cxcl9 and Cxcr6 levels were significantly lower in Sox9-overexpressed tumors. Therefore, we assessed the expression of Cxcl9, its receptor Cxcr3, Cxcr6 and its ligand Cxcl16, and confirmed SOX9-related suppression of these genes in allografts (Fig. 4e, g). Lower abundance of Cxcr3 and Cxcr6 is likely correlated with lower CD8⁺ T cell infiltrate in the Sox9 overexpressing tumors. Since dendritic cells also secret chemokines and a developmentally important gene of dendritic cells, Flt3 [43], was significantly downregulated (Fig. 3c), we evaluated the function of dendritic cells by Flt3 and Itgae (CD103), and found suppressed expression of both in the allografts (Fig. 4e, g). These findings suggest a potential mechanism whereby SOX9 upregulation inhibits tumor infiltrating dendritic cells and subsequently suppresses CD8⁺ T cells and NK cells.

To test for a cell intrinsic regulation of the immune suppression, we examined expression of the immune-related genes in mTC11 cells, in which Sox9 overexpressed or knocked out, that were cultured in vitro. Except for Ncam1, the immune-related gene expressions were not detectable or not significantly changed (Fig. S3c), indicating that the observed immune suppression is likely regulated by a SOX9-related tumor microenvironment rather than the tumor cells.

We next examined if altered tumor matrix stiffness and tumor associated angiogenesis contribute to SOX9-induced immune suppression. As previously reported, Stc1 is involved in cancer cell proliferation [44], tumor associated angiogenesis [45] and tumor immune resistance [46]. Gprc5b is a hallmark of Kras signaling, and it promotes cell proliferation and collagen production [47]. Col2a1 [48], Col4a5 [49] and Trf [50] are involved in extracellular matrix pathway and increase cell proliferation. Notably, these genes were significantly upregulated in mSox9OE tumors (Supplementary Table 1). We selected them for RT-qPCR validation in both mSox9OE and Sox9-knockout subcutaneous tumors, together with Igfbp5, Fosl1 and Bcl2 (from cell proliferation pathway, Fig. 3b, Supplementary Table 2) as references. We found that all these collagen-related genes were positively correlated with Sox9 expression (Fig. 4e, Fig. S3d), suggesting that SOX9 enhances lung tumor stiffness. These genes were likely intrinsic to the tumor cells because they were correlated with Sox9 expression in vitro as well (Fig. S3c, d). We performed collagen staining in KS^f/f and KS^w/w lung tumors, and SOX9⁺ tumors had abundant collagen fibers whereas SOX9⁻ tumors had no to very low collagen fibers (Fig. 4h). These data consistently demonstrate that SOX9 increases collagen production in LUAD, facilitating a tumor microenvironment with increased tumor stiffness that is known to inhibit anti-tumor immunity.

SOX9 is significantly associated with immune suppression in human LUAD

To test if SOX9 has immunosuppressive function in human lung cancer, we explored immune signatures in human LUAD with SOX9 high (top 20%) and low (lower 15%) using a CIBERSORT dataset [51]. We observed significantly fewer CD8⁺ T cells, NK cells, plasma cells, as well as more macrophages and memory CD4⁺ T cells in SOX9-high LUAD tumors (Fig. 5a, Supplementary Table 6). To test if the immunosuppressive effects of SOX9 are independent of mutant KRAS, we performed CIBERSORT analysis on the KRAS-wildtype LUAD subset, and the results were identical, except for CD4⁺ T cells (Supplementary Table 6). We validated these data by IHC analysis of CD8⁺ T cells in 23 unselected human LUAD specimens and found that SOX9⁺ LUAD had significantly less infiltration of CD8⁺ T cells (Fig. 5b, c). Thus, SOX9 suppresses CD8⁺ T cells in human LUAD. We also performed CIBERSORT analysis in 32 additional human cancers and observed a significant association between high SOX9 expression and CD8⁺ T cell suppression in 3 additional cancers, Glioblastoma Multiforme, Sarcoma and Thyroid Carcinoma (Supplementary Table 6), suggesting the immunosuppressive effects of SOX9 extend to other cancer types but are not universal. To examine if the functions of CD8⁺ T cells, NK cells and dendritic cells are suppressed in SOX9-high tumors in human LUAD, we investigated their functional markers using a single cell RNA-seq dataset (GSE131907) [52]. Based on the percentages of SOX9⁺ epithelial cells over total epithelial cells in each tumor, we defined the top 3 as SOX9-high and the lower 3 as SOX9-low tumors (Fig. S5). For CD8⁺ T cells, we found that functional markers GZMA, GNLY, IFNG, CD44 and CD69 were significantly downregulated in SOX9-high tumors (Fig. 5d). Similarly, dendritic cells and NK cells from SOX9-high tumors showed a significant downregulation of the functional markers CD80/CD86/LAMP3 (Fig. 5e) and GNLY/KLRD1 (Fig. 5f), respectively. Therefore, in human LUAD, upregulation of SOX9 results in decreased immune cell infiltration as well as suppression of CD8⁺ T cell, NK cell and dendritic cell functions in the tumor microenvironment.

Lentiviral production, cells and cell culture, and animal studies

Detailed information on lentiviral production, cell culture and the accompanying experimental conditions in animals can be found in the Supplementary information.

Mouse tumor organoid culture and image analysis

Mouse 3D organoid cultures were performed as described [71] with some modifications. Cells were suspended in cold complete medium at 250 cells/μl, and then mixed with Matrigel (Corning, 356237) at ratio 1:4 on ice. We dropped 10 μl of the cell/Matrigel into each well in 24-well plate (Greiner Bio-one, 662102) in triplicates and incubated at 37 °C. After 20 min, 800 μl of complete medium were added and changed every 2 days. The tumor organoids were imaged under ×4 magnification with bright-field microscope and organoid size was evaluated with function “morphological segmentation” of plugins “MorphoLibJ” in ImageJ1.52a.

Histology, immunohistochemistry and image analysis

Details related to preparation of tissues, methods for immunohistochemistry and analysis of images can be found in the Supplementary information.

Supplementary methods

Additional experimental procedures including RNA-seq and analysis, real-time quantitative PCR and flow cytometry are available in the Supplementary information.

Statistical tests

In this study, statistical tests were performed using R or GraphPad Prism. We evaluated the normality of the samples using a Normality and Lognormality test. For the normally distributed data, two-tailed t test was used or ordinary two-way ANOVA was performed to compare two datasets. For the data that were not normally distributed, we used the two-tailed Mann–Whitney test to compare two datasets. P values <= 0.05 were considered statistically significant. Data are expressed as mean ± standard error of the mean (SEM). All experiments were repeated at least twice to confirm the repeatability, and each group had at least 3 biological replicates (specified in figure panels). For the statistical analysis of the RNA-seq data, it was described in the supplementary methods.