Tumor-initiating cells establish an IL-33-TGF-β niche signaling loop to promote cancer progression

Abstract

INTRODUCTION:

A small subset of tumor cells with long-term tumorigenic capacity, known as tumor-initiating cells (TICs), play a pivotal role in cancer development and therapy resistance. However, the development of effective TIC-targeted therapies is moving at a restricted pace due to the lack of identification of TIC vulnerabilities. Just as normal stem cells are regulated by external cues derived from specialized microenvironments, or stem cell niches, the stem-like state of TICs and the malignant phenotypes of their progeny are controlled by various factors emanating from the TIC-associated tumor microenvironment, the so-called TIC niche. Therefore, a mechanistic understanding of the cross-talk between TICs and the niche could accelerate the development of durable cancer therapeutics. Although the TIC niche is thought to evolve through reciprocal interactions with TICs, the mechanism by which the TIC–niche interaction emerges in the course of tumor development is poorly understood. Solid tumors are known to recruit immune cells in the stroma and create favorable conditions for their growth and survival. However, not much is known about how TICs regulate the localization and function of TIC-supportive immune cells in their spatial proximity.

RATIONALE:

Using a mouse model of squamous cell carcinoma (SCC), we previously showed that transforming growth factor β (TGF-β) induces a subset of drug-resistant TICs that give rise to invasive, poorly differentiated progeny. We observed that these TGF-β–responding tumor cells are spatially associated with localized TGF-β expression in the adjacent stroma. Therefore, the mechanisms that lead to “TGF-β–rich” tumor microenvironments may underlie the development of TIC–niche interactions and potentially be exploited as a new target for destabilizing TICs. Because normal stem cells coordinate their niches by sending short-distance signals, we hypothesized that TICs might send a specific signaling molecule to the adjacent stroma to induce a TIC-supporting niche.

RESULTS:

Focusing on the cytokine milieu and immune cells in the proximity of TGF-β–responding TICs, here, we address how TICs generate a spatially distinct niche microenvironment that is required for invasive progression and drug resistance of SCC. In a search for potential paracrine regulators of the neighboring tumor microenvironment, we identified interleukin-33 (IL-33) as the most highly up-regulated cytokine in TGF-β–responding TICs. Whereas IL-33 is stored in the nucleus under normal conditions, we found that it is released into the extracellular space in response to the NRF2-mediated antioxidant response, a hallmark of TGF-β–responding TICs. This TIC-derived IL-33 was required for invasive progression and drug resistance of SCC. Mechanistically, IL-33 induces the accumulation of a subset of tumor-associated macrophages expressing the IL-33 receptor ST2 and the high-affinity IgE receptor (FcεRIα) in close proximity to TICs (i.e., within a 50-μm radius). These previously unappreciated FcεRIα⁺ macrophages were differentiated and alternatively activated from bone marrow–derived cells and created a TGF-β–rich niche microenvironment through the IL-33–ST2–NF-κB pathway, inducing paracrine TGF-β signaling to TICs and further upregulating IL-33 expression. The abrogation of the pathway or the depletion of FcεRIα⁺ macrophages reduced the number of TGF-β–responding TICs, the rate of invasive tumor progression, and chemotherapy resistance.

CONCLUSION:

Therapy-resistant TICs are considered to be major culprits in cancer treatment failure. Studying a mouse model, we unveil the cellular and molecular basis for the formation of a TIC niche that promotes malignant progression and drug resistance of SCC. The discovery of the IL-33–TGF-β niche signaling loop between TICs and FcεRIα⁺ macrophages provides mechanistic insights into self-reinforcing TIC–niche interactions, which could be a potential target for destabilizing TICs to improve cancer treatment efficacy.

Targeting the cross-talk between tumor-initiating cells (TICs) and the niche microenvironment is an attractive avenue for cancer therapy. We show here, using a mouse model of squamous cell carcinoma, that TICs play a crucial role in creating a niche microenvironment that is required for tumor progression and drug resistance. Antioxidant activity in TICs, mediated by the transcription factor NRF2, facilitates the release of a nuclear cytokine, interleukin-33 (IL-33). This cytokine promotes differentiation of macrophages that express the high-affinity immunoglobulin E receptor FcεRIα and are in close proximity to TICs. In turn, these IL-33–responding FcεRIα⁺ macrophages send paracrine transforming growth factor β (TGF-β) signals to TICs, inducing invasive and drug-resistant properties and further upregulating IL-33 expression. This TIC-driven, IL-33–TGF-β feedforward loop could potentially be exploited for cancer treatment.

Graphical Abstract

Signaling loop in the TIC niche that promotes cancer progression in mice. TGF-β–responding TICs release IL-33 via the NRF2-mediated antioxidant response. This induces differentiation of immature myeloid cells into FcεRIα⁺ macrophages in close proximity to the TICs. In turn, FcεRIα⁺ macrophages send reciprocal paracrine TGF-β signals to TICs to promote invasive progression and drug resistance of SCC and further induce the release of IL-33, thereby establishing a self-reinforcing signaling loop.

A small subset of tumor cells with long-term tumorigenic capacity, known as tumor-initiating cells (TICs), play a pivotal role in cancer development and therapy resistance (1, 2). Just as normal stem cells are regulated by external cues derived from specialized microenvironments or stem cell niches (3), the stem-like state of TICs and the malignant phenotypes of their progeny are controlled by various factors emanating from the TIC-associated tumor microenvironment, the so-called TIC niche (4, 5). Therefore, a mechanistic understanding of the interaction between TICs and their niche would likely accelerate the development of effective cancer therapeutics. Although the TIC niche is thought to evolve through reciprocal cross-talk with TICs (6), the mechanism by which the TIC–niche relationship emerges in the course of tumor development is poorly understood. Solid tumors are known to recruit immune cells in the stroma and create favorable conditions for their growth and survival (7, 8). In fact, the spatial distribution of tumor-infiltrating immune cells can be predictive of poor patient survival and therapeutic responses (9, 10). However, not much is known about how TICs regulate the localization and function of TIC-supportive immune cells in their spatial proximity.

Squamous cell carcinoma (SCC), the second most prevalent cancer worldwide, harbors TICs at the tumor–stroma interface (11, 12), an ideal anatomical location for paracrine interactions with potential niche cells. One of the critical regulators of stem cells and SCC development is transforming growth factor β (TGF-β), which acts as a tumor suppressor or promoter at different stages of cancer development (13–16). To investigate the stage-specific role of TGF-β in tumorigenesis, we previously devised a mouse model of SCC that allowed us to label and lineage trace tumor cells responding to TGF-β (17). Using this approach, we showed that TGF-β induces slow-cycling tumor cells that give rise to invasive, poorly differentiated progeny. Moreover, TGF-β–responding tumor cells survived chemotherapy through a cytoprotective mechanism activated by NRF2, the master transcription factor for antioxidant responses, and their progeny repopulated the recurrent tumors (17), indicating that TGF-β–responding tumor cells function as drug-resistant TICs. We observed that these TGF-β–responding tumor cells are spatially associated with localized TGF-β expression in the adjacent stroma. In several types of human cancers, abundant TGF-β expression in the stroma is known to be associated with poor prognosis (18, 19). Therefore, the molecular and cellular mechanisms that lead to “TGF-β-rich” tumor microenvironments may underlie the development of TICβniche interactions and could potentially be exploited as a new target for destabilizing TICs.

Here, by focusing on the cytokine milieu and immune cells in the proximity of TGF-β–responding TICs, we unveiled the cellular and molecular basis of a TIC–niche interaction that promotes malignant progression and drug resistance in a mouse model of SCC. We identified a TIC-derived nuclear cytokine, interleukin-33 (IL-33), a member of the IL-1 family that has a pleiotropic role in tumorigenesis (20, 21), and IL-33–responding macrophages as key components of a signaling loop in the TIC niche that emerges during the early stages of SCC development and regulates tumor progression.

Results

IL-33 expression is up-regulated in TGF-β–responding tumor cells

To study TIC–niche interactions emerging in an early tumor microenvironment, we developed an oncogenic, HRAS-driven SCC mouse model using an in utero, epidermis-specific lentiviral (LV) transduction method (17, 22). Briefly, epidermal progenitor cells of TetO-HRAS^G12V (23); Rosa-LSL-EYFP (24) transgenic mouse embryos were transduced with LV vectors expressing rtTA and Cre constitutively and mScarlet fluorescent protein in response to TGF-β–SMAD2/3 signaling (fig. S1, A and B). Successful transduction was monitored by the expression of yellow fluorescent protein (YFP) (fig. S1C). Postnatally, the LV-rtTA induced HRAS^G12V expression in a doxycycline (Dox)–dependent manner in YFP⁺ cells, which initiated SCC formation (fig. S1D). We observed histopathologically distinct regions manifesting well-differentiated and invasive SCC (fig. S1E), and the LV fluorescent reporter (as well as SMAD2 phosphorylation; fig. S1F) illuminated TGF-β–responding tumor cells in invasive SCC (Fig. 1A, pink signal). Moreover, the frequency of TGF-β–responding tumor cells was positively correlated with the distribution of TGF-β ligand in the adjacent stroma (Fig. 1A, green signal, and 1B). This raised the question of whether TGF-β–responding TICs establish specific interactions with adjacent cells in the stroma, potentially niche cells, to maintain TGF-β–rich microenvironments.

Fig. 1. IL-33 is up-regulated in tumor cells adjacent to TGF-β–rich microenvironments in a mouse model of SCC.

(A) Fluorescent reporter illuminates higher TGF-β signaling activity in the invasive region of SCC. Note that increased TGF-β immunostaining (green) in the stroma correlates with the presence of TGF-β reporter⁺ cells (pink) at the invasive tumor front. (B) The proportion of TGF-β reporter⁺ cells and the intensity of TGF-β immunolabeling were measured by fluorescence microscopy of tumor sections with active and inactive TGF-β signaling (n = 7). (C) MA plot presentation of RNA-seq data comparison between TGF-β–responding and -nonresponding tumor basal cells. Biological replicates, n = 5. Red and blue dots indicate significantly up- and down-regulated genes, respectively (adjusted P < 0.01). Triangles indicate up-regulated NRF2-regulated genes involved in antioxidant responses (fig. S1H). Diamonds indicate genes encoding cytokines or chemokines (fig. S1I). (D) qPCR analysis of Il33 mRNA in FACS-isolated TGF-β reporter⁺ and reporter^neg tumor basal cells in vivo (the sorting strategy is shown in fig. S1G; n = 3). Data are expressed as relative mean expression with SEM and were analyzed with unpaired t test, ***P < 0.001. (E) Immunodetection of IL-33 protein in mosaic tissue with Tgfbr2 cKO (YFP⁺) and WT (YFP^neg) tumor cells. Mosaic tissues were induced by a partial activation of K14-CreER by limited dose of tamoxifen. (Right) Dotted lines denote the boundaries of WT and cKO tumor cell regions. (Graph) Quantification of IL-33 signal intensity (n = 3). Data are expressed as relative mean intensity with SD and were analyzed with unpaired t test, ***P < 0.001. Scale bars, 50 μm.

Normal stem cells coordinate their niches by sending short-distance cues (25, 26) and subsequently receive reciprocal signals, including TGF-β, to maintain stem cell properties (14, 16). Therefore, we hypothesized that TICs may send a specific signaling molecule to the adjacent stroma to directly induce a TGF-β–rich, TIC-supporting niche. To identify candidate molecules, we purified TGF-β reporter⁺ and reporter^neg tumor basal (YFP⁺, integrin α6^hi, β1^hi) cells by fluorescence-activated cell sorting (FACS) and compared their transcriptomes by RNA sequencing (RNA-seq) (fig. S1G). By combining current and previously published data (17), we obtained a highly reliable list of genes differentially expressed in TGF-β–responding tumor cells (Fig. 1C, significance cutoff, adjusted P < 0.01), including a battery of NRF2 target genes involved in antioxidant responses (Fig. 1C, triangles, and fig. S1H).

We then focused on genes encoding cytokines and chemokines (Fig. 1C, diamonds, and fig. S1I). There was no significant difference in Tgfb1, Tgfb2, and Tgfb3 expression between TGF-β reporter⁺ and reporter^neg tumor cells, supporting the notion that a paracrine signal from the adjacent stroma, but not an autocrine signal, may play a major role in generating TGF-β–responding TICs. We identified Il33 as the most significantly up-regulated and highly expressed gene in TGF-β–responding tumor cells. We verified elevated Il33 mRNA expression by quantitative PCR (qPCR) analysis of independent in vivo samples (Fig. 1D). To assess the association between TGF-β signaling and IL-33 expression in vivo, we locally deleted the Tgfbr2, the gene encoding the type II TGF-β receptor (27), in tumor epithelial cells by activating K14-CreER using a limiting dose of tamoxifen (28). The side-by-side comparison of YFP⁺ Tgfbr2 conditional knockout (cKO) and YFP^neg wild-type (WT) mosaic tissues revealed that the lack of the TGF-β receptor (Fig. 1E, green) was significantly correlated with reduced IL-33 immunolabeling (Fig. 1E, red). These results strongly suggest that TGF-β signaling is involved in the up-regulation of IL-33 in SCC.

NRF2-mediated antioxidant response triggers the extracellular release of IL-33

We next sought to identify what triggers IL-33 release from TGF-β–responding tumor cells. IL-33 is stored in the nucleus under normal conditions but is released into the extracellular space as an alarmin cytokine upon cell injury, especially by necrotic cell death (20). It remains unclear how IL-33 can be released from living tumor cells, particularly TICs. Immunostaining confirmed the nuclear localization of IL-33 in normal epidermal cells (fig. S2A). However, as tumors progressed to invasive SCC, cytoplasmic IL-33 staining became evident in keratin 5–positive (K5⁺) tumor basal cells, whereas IL-33 tended to remain in the nucleus of K5^dim suprabasal cells (Fig. 2A). Cytoplasmic IL-33 localization was more frequently observed in TGF-β–responding tumor cells compared with their nonresponding counterparts (Fig. 2B), suggesting that IL-33 may be preferentially released from TGF-β–responding TICs in the adjacent stroma.

Fig. 2. The NRF2-mediated antioxidant response induces extracellular release of IL-33.

(A) Immunolabeling of tumor tissues showing a range of IL-33 expression patterns. In well-differentiated SCC, IL-33 nuclear staining can be observed in both K5⁺ basal (arrows) and K5^dim suprabasal cells. In invasive SCC, cytoplasmic IL-33 staining can be observed in K5⁺ basal cells (arrowheads). (B) TGF-β reporter⁺ tumor cells at the invasive front exhibited cytoplasmic IL-33 localization (arrowheads), whereas suprabasal cells had nuclear IL-33 (arrows). (Graph) Quantification of IL-33 nuclear versus cytoplasmic localization in TGF-β reporter⁺ and reporter^neg tumor basal cells. (C) Immunostaining of invasive SCC tissue showing the correlation between nuclear NRF2 expression and cytoplasmic IL-33 localization (arrowheads). Note that nuclear IL-33–expressing K5^neg stromal cells are negative for NRF2. (D) Immunolabeling of human SCC section showing the correlation between cytoplasmic IL-33 localization and nuclear NRF2 expression (arrowheads). (E) Western blots showing that nonlethal oxidative stress induced by 500 μM H₂O₂ reduces IL-33 in total cell lysates over 3 hours. Arrowhead indicates an NRF2-specific band. Asterisk indicates a nonspecific band. (Graph) Quantification of IL-33 protein (n = 3). The relative intensity of IL-33 bands normalized based on the α-tubulin (α-tub) loading control is shown. Data are shown as mean with SD and were analyzed with unpaired t test, *P = 0.0195, **P < 0.01. (F) Western blot analysis of IL-33 protein in the concentrated conditioned medium (CM) of cells with and without H₂O₂ treatment. (G) Nrf2 (Nfe2l2) KD prevents H₂O₂-induced IL-33 protein reduction in total cell lysates. (Graph) Quantification of IL-33 protein (n = 4). Data are shown as mean with SD. (H) Western blots show that Keap1 KD cells increase NRF2 and decrease intracellular IL-33 protein levels regardless of oxidative stress. (I) Immunolabeling of IL-33 and NRF2 in scramble control, Nrf2 KD, or Keap1 KD cells with or without TGF-β treatment (50 pM for 36 hours). Note that Nrf2 KD results in an exclusive nuclear localization of IL-33, whereas TGF-β–treated and Keap1 KD cells show granular nuclear and cytoplasmic IL-33 localization. Dotted lines denote the tumor–stroma boundaries. Scale bars, 50 μm.

It has been reported that nonlethal mechanical stress can induce the secretion of nuclear IL-33 (29). We explored the possibility of nonlethal oxidative stress and the NRF2-mediated antioxidant response as a potential mechanism of IL-33 release, because the latter is a hallmark of TGF-β–responding TICs (17). The activity of NRF2 is primarily controlled at the level of protein stability (30). K5⁺ tumor basal cells in invasive SCC accumulated stabilized NRF2 protein in the nucleus, and these NRF2⁺ cells exhibited cytoplasmic localization of IL-33 (Fig. 2C, arrowheads). Such a correlation was also observed in invasive SCC of a chemical carcinogenesis mouse model (fig. S2B) and human SCC (Fig. 2D). It is noteworthy that stabilized NRF2 was detected explicitly in tumor epithelial cells, but not in stromal cells, and that IL-33 protein expressed in stromal cells was sequestered in the nucleus (Fig. 2C, arrows), further supporting the potential link between NRF2 activity and IL-33 release.

In vitro, H₂O₂-induced NRF2 stabilization in HRAS^G12V-expressing cells was correlated with the reduction of intracellular IL-33 protein levels (Fig. 2E) without changing Il33 mRNA levels (fig. S2C). At the same time, IL-33 protein became detectable in the conditioned medium of H₂O₂-treated cells (Fig. 2F), suggesting that IL-33 is released from the cells. To assess the functional relevance of NRF2 in the subcellular localization of IL-33, we depleted NRF2 or KEAP1 (an E3 ubiquitin ligase that leads to NRF2 degradation) (30). Short hairpin RNA (shRNA)–mediated Nfe2l2 (Nrf2) or Keap1 knockdown (KD) did not affect Il33 mRNA levels (fig. S2D). However, Nrf2 KD hampered H₂O₂-induced reduction of intracellular IL-33 (Fig. 2G). Conversely, Keap1 KD stabilized NRF2 and decreased intracellular IL-33 levels even without H₂O₂ (Fig. 2H).

As we previously reported, prolonged TGF-β treatment of HRAS^G12V-expressing cells induced NRF2 stabilization and the expression of NRF2 target genes such as Hmox1 and Me1 (Fig. 2I and fig. S2E) (17). TGF-β treatment also up-regulated Il33 mRNA levels (fig. S2E), and IL-33 protein was detected in both the nucleus and cytoplasmic space. This TGF-β–induced cytoplasmic IL-33 signal was suppressed by Nrf2 KD, resulting in sharp nuclear IL-33 localization. By contrast, Keap1 KD facilitated both NRF2 stabilization and cytoplasmic IL-33 localization (Fig. 2I). Together, the data from mouse and human tissues and results from in vitro experiments strongly suggest that the NRF2-mediated antioxidant response activated in TGF-β–responding TICs facilitates the extracellular release of IL-33.

Tumor-derived IL-33 promotes paracrine TGF-β signaling, invasive progression, and drug resistance

To determine the role of tumor-derived IL-33 in SCC development, we generated LV vectors carrying potent Il33 shRNA, which did not cause noticeable differences in cell morphology or proliferation rates in vitro (fig. S3, A to D). We then depleted IL-33 from tumor epithelial cells in vivo through the transduction of the LV vectors in utero (fig. S1, A and B). Immunostaining confirmed the efficient depletion of IL-33 in YFP⁺ tumor cells, whereas infrequent nuclear IL-33 expression remained in YFP^neg cells in the stroma (Fig. 3A). To validate Il33 KD efficacy in vivo, we purified tumor basal cells (YFP⁺, integrin α6^hi, β1^hi), CD45⁺ immune cells, CD140a⁺ fibroblasts, and CD31⁺ endothelial cells by FACS (fig. S3E). qPCR analysis of FACS-purified cells verified the tumor cell–specific Il33 KD and indicated that IL-33 is most highly expressed in tumor epithelial cells (fig. S3F), among which the TGF-β–responding TIC population is the dominant source (Fig. 1D).

Fig. 3. Depletion of tumor-derived IL-33 impairs tumor progression and paracrine TGF-β signaling.

(A) Immunolabeling of tumor sections showing efficient IL-33 protein depletion in Il33 shRNA-transduced (YFP⁺) cells. Note that some YFP^neg stromal cells maintain nuclear IL-33 (arrows). (B) HRAS^G12V-driven tumors in control and Il33 KD mice were sized at the time of euthanasia. Scramble control, n = 46; Il33 KD, n = 31. Approximation curves were drawn by applying the Michaelis-Menten kinetics curve. (C) Hematoxylin and eosin (H&E) staining of tumor sections showing smoother edges in Il33 KD tumors compared with control tumors, suggestive of reduced invasive capacity. Scale bars, 100 μm. (D) Immunolabeling of K10 showing a loss of differentiation property in control tumors, but not in Il33 KD tumors. (E) Detection of EdU incorporated 4 hours before euthanasia. (Graph) Quantification of EdU⁺ cells in integrin α6⁺ tumor basal cells of control and Il33 KD tumors (n = 3) (n = 806 to 1747 cells). Data are shown as mean with SEM and were analyzed with unpaired t test, **P = 0.004. (F) Flow cytometry analysis of TGF-β reporter (mScarlet) expression in YFP⁺ tumor epithelial cells. YFP+ cells in Il33 KD tumors show fewer and reduced TGF-β reporter activity compared with those in control tumors. (Graph) Proportion of TGF-β reporter⁺ tumor cells. Each dot indicates values from an individual tumor. Scramble control, n = 30; Il33 KD, n = 19. Data are shown as mean with individual values and were analyzed with unpaired t test, **P = 0.007. (G) Tumor-bearing mice were treated with cisplatin (10 mg/kg), and cells undergoing apoptosis (cleaved caspase-3⁺) in K5⁺ tumor basal cells were quantified 2 days after treatment (n = 3). Data are shown in box-and-whisker plots (midline, median; box, 25th and 75th percentiles; whiskers, min and max) and were analyzed with unpaired t test, **P = 0.0061, *P = 0.0334. (H) Volume of tumors after cisplatin administration. (Left) Spider plots showing the changes for each tumor. (Right) Normalized tumor volume of each condition. Scramble control, n = 3 (15 tumors), Il33 KD, n = 5 (24 tumors). Tu, tumor. St, stroma. Scale bars, 50 μm [except for (C)].

Transduced mice developed 8.2 ± 2.8 tumors (n = 46) and 6.1 ± 3.4 tumors (n = 31) in scramble control and Il33 KD conditions, respectively. The average size of the five largest tumors indicated that control tumors grew larger over time, whereas Il33 KD tumors were stunted in growth (Fig. 3B). Histopathological analysis showed that control tumors progressed to invasive, poorly differentiated states when assessed by a differentiation marker, keratin 10 (K10) (Fig. 3, C and D). By contrast, Il33 KD tumors showed smoother tumor edges and well-differentiated states. Despite smaller tumor size, a short 5-ethynyl-2′-deoxyuridine (EdU) pulse revealed that Il33 KD tumor cells are more proliferative than those in control tumors (Fig. 3E).

We previously demonstrated that TGF-β–responding tumor cells are a slow-cycling TIC population that give rise to invasive, poorly differentiated progeny (17), raising the intriguing possibility that stunted progression of Il33 KD tumors may be due to the reduction of paracrine TGF-β signaling toward tumor cells. To determine whether tumor-derived IL-33 is responsible for evoking TGF-β signaling, we quantified TGF-β–responding tumor cells by flow cytometry. In YFP⁺ tumor epithelial cells (Fig. 3F, green gates), the frequency of TGF-β reporter⁺ (mScarlet⁺) cells (Fig. 3F, gray versus orange gates) were significantly decreased in Il33 KD tumors compared with control tumors (Fig. 3F, graph). In vitro, TGF-β–induced SMAD2 phosphorylation and cytostatic responses were still intact in Il33 KD cells (fig. S3, D and G), strongly suggesting that the reduction of TGF-β signaling in Il33 KD tumors is likely due to a non-cell-autonomous effect, potentially the lack of TGF-β–rich tumor microenvironments.

Because TGF-β can exert drug resistance effects on TICs in vivo (Fig. 3G) (17), we next assessed the efficacy of cisplatin chemotherapy in Il33 KD tumors. Consistent with the reduction of TGF-β–responding tumor cells, K5⁺ basal cells of Il33 KD tumors underwent apoptosis more frequently (Fig. 3G). After cisplatin treatment, most tumors shrank regardless of Il33 status; however, control tumors typically recurred in 3 weeks. By contrast, almost all Il33 KD tumors failed to regrow (Fig. 3H). Our data suggest that tumor-derived IL-33 plays an important role in generating TGF-β–responding TICs that promote invasive progression and drug resistance in SCC.

FcεRIα⁺ macrophages accumulate in close proximity to TGF-β–responding tumor cells

We previously observed dense CD11b⁺ and Ly6C⁺ cell localizations at TGF-β–rich stroma (17), raising the possibility that CD11b⁺Ly6C⁺ immature myeloid cells (31) differentiate into TGF-β–producing immune cells in response to the cytokine milieu around TICs. Therefore, we investigated whether TGF-β–responding TICs are associated with a specific subset of immune cells and, if so, how TIC-derived IL-33 is involved in the development of spatially distinct microenvironments. To address this, we measured the frequency of TGF-β reporter⁺ tumor cells and the density of immune cells within a 50-μm radius of the tumor edges by immunofluorescence analysis of tumor sections (fig. S4, A and B). Our data showed that the peritumoral areas were dominated by CD11b⁺ myeloid cells rather than CD3⁺ T cells. Despite these lower cell numbers, the density of FoxP3⁺ regulatory T (Treg) cells was positively correlated with the frequency of TGF-β–responding tumor cells. This is not surprising because TGF-β promotes the differentiation and function of Treg cells (32), but it suggests the presence of a TGF-β gradient in the stroma. The strongest positive correlation was found in cells expressing the high-affinity IgE receptor FcεRIα (P = 0.0018) (fig. S4B). FcεRIα⁺ cells accumulated in the proximity (~50-μm radius) of TGF-β–responding tumor cells that showed nascent invasive phenotypes (Fig. 4A, i). Furthermore, their close spatial relationship was sustained during invasive progression at the tumor–stroma interface (Fig. 4A, ii and iii, and B).

Fig. 4. FcεRIα⁺ macrophages accumulate in the proximity of TGF-β–responding invasive tumor cells.

(A) Immunolabeling of tissue sections from different SCC stages showing clusters of FcεRIα⁺ cells accumulated within a 50-μm radius of TGF-β reporter⁺ invasive tumor cells (dotted circles and box). (B) Quantification of FcεRIα⁺ cells in the stroma adjacent to the tumor leading edge with TGF-β reporter high (≥30%) versus low (<30%) showing higher density of FcεRIα⁺ cells near TGF-β–responding cells. n = 4 mice, total 12 images that included both TGF-β reporter high and low areas and were analyzed with unpaired t test, **P = 0.0015. (C) Flow cytometry analysis of FcεRIα expression in different immune cell types in SCC. Gating strategy is shown in fig. S4D. (D) Immunolabeling of invasive SCC section showing that most FcεRIα staining overlaps with F4/80, an established murine macrophage marker, and present within a 50-μm radius of tumor leading edges. Note that F4/80⁺ cells farther than the 50-μm radius from tumor edges were negative for FcεRIα. Asterisks indicate rare FcεRIα⁺F4/80^neg cells. Dotted lines denote the tumor–stroma boundaries. (E) FACS plots of F4/80⁺FcεRIα⁺ (green dotted box) and F4/80⁺FcεRIα^neg (gray dotted box) macrophages in SCC. (F) MA plot presentation of RNA-seq data comparison between FcεRIα⁺ and FcεRIα^neg macrophages in SCC. Biological replicates, n = 3. Red and blue dots indicate significantly up- and down-regulated genes, respectively (adjusted P < 0.05). Dark red and dark blue dots highlight some of the “M2” and “M1” marker genes, respectively. (G) qPCR analysis of FACS-isolated macrophage populations in SCC (n = 3). Data are shown as mean with SD. Fcer1a, **P = 0.0019. Arg1, **P = 0.0032. Tgfb1, *P = 0.0266. (H) Immunolabeling of invasive SCC section showing that most FcεRIα⁺ cells overlap with TGF-β protein localization in the stroma. Magnified images are shown in fig. S4E. (I) Immunolabeling of human SCC section showing that FcεRIα⁺ cells cluster around invasive phospho-SMAD2 (pSMAD2)⁺ tumor cells (dotted circles). (J) Immunolabeling of human SCC section showing that most FcεRIα⁺ cells overlap with CD206, an established marker of alternatively activated macrophages, at the tumor-stroma interfaces, whereas CD206⁺FcεRIα^neg cells present in the stroma farther than a 50-μm radius from tumor edges. (Graph) Quantification of the density of FcεRIα⁺ cells and CD206⁺ cells 0 to 50 or 50 to 100 μm away from tumor edges. Data were analyzed with paired t test, **P = 0.0035. Tu, tumor. St, stroma. Scale bars, 50 μm.

Mast cells make up most of the FcεRIα⁺ cells in the normal dermis, but it is known that their density declines in invasive SCC and mast cells become dispensable for tumor progression (33, 34). We found most of the FcεRIα⁺ cells in invasive SCC to be negative for mast cell tryptase, an indication that these are not mast cells but another myeloid cell lineage (fig. S4C). To identify the major FcεRIα⁺ cells in SCC, we analyzed a panel of immune cell markers for FcεRIα expression by flow cytometry (fig. S4D). This comprehensive analysis revealed that a fraction of CD11b⁺Ly6C⁺Ly6G^neg immature myeloid cells/monocytes and F4/80⁺ macrophages express FcεRIα (Fig. 4C). In particular, a subset of macrophages with low major histocompatibility complex (MHC)-II surface expression showed the highest FcεRIα levels. Immunofluorescence analysis validated that FcεRIα⁺ cells are mostly F4/80⁺ macrophages, particulary those in the ~50-μm peritumoral areas (Fig. 4D, arrowheads).

Low MHC-II expression suggests that FcεRIα⁺ cells may resemble alternatively activated macrophages, also known as M2-like tumor-associated macrophages (35, 36). To characterize FcεRIα⁺ macrophages, we purified FcεRIα⁺ and FcεRIα^neg macrophages from SCC by FACS (Fig. 4E) and analyzed their transcriptomes by RNA-seq. Compared with FcεRIα^neg macrophages, FcεRIα⁺ macrophages up-regulated the expression of multiple genes, including Arg1, Mrc1 (Cd206), Cd163, Stab1, and Il10, which are known to be elevated in alternatively activated macrophages (37) (Fig. 4F). By contrast, FcεRIα⁺ macrophages significantly down-regulated the expression of classically activated (M1-like) macrophage markers such as Tnf, Cd40, and MHC-II. We further validated RNA-seq results by qPCR analysis of marker genes. Although an established M1 marker, Nos2, was similarly expressed regardless of FcεRIα expression, the M2 markers Arg1 and Tgfb1 were significantly up-regulated in the FcεRIα⁺ macrophages compared with their FcεRIα^neg counterparts (Fig. 4G). Indeed, immunostaining showed that most FcεRIα⁺ cells were overlapped with prominent TGF-β localization in the stroma (Fig. 4H and fig. S4E).

These previously unappreciated FcεRIα⁺ macrophages were also found in invasive SCC induced by DMBA–TPA chemical carcinogenesis (38) and in K14-HPV16–transgenic models (33), where they accumulated around phospho-SMAD2⁺ tumor cells (fig. S4, F and G). Moreover, we detected FcεRIα⁺CD206⁺ tumor-associated macrophages in human SCC patient samples, which were also clustered around phospho-SMAD2⁺ TGF-β–responding tumor cells (Fig. 4I). FcεRIα⁺CD206⁺ macrophages in human SCC were also significantly enriched within a 50-μm radius from the tumor edges (Fig. 4J, arrowheads), whereas FcεRIα^negCD206⁺ macrophages were distributed more evenly in the stroma (Fig. 4J, arrows and graph). Collectively, these results strongly suggest that FcεRIα⁺ cells are a subset of tumor-associated macrophages that create spatially distinct, TGF-β–rich tumor microenvironments in the proximity of TICs.

IL-33 induces FcεRIα⁺ macrophages that activate proinvasive TGF-β signaling

Macrophages are known pro-tumorigenic immune cells and a major component of the TIC niche (39, 40), and previous studies showed that IL-33 can induce macrophage recruitment and alter their phenotype to resemble alternatively activated macrophages (41, 42). We next sought to determine whether TIC-derived IL-33 is involved in FcεRIα⁺ macrophage differentiation. We found that FcεRIα⁺ macrophages in SCC expressed high levels of the IL-33 receptor ST2 (43) (Fig. 5A). By contrast, ST2 was barely detected in TGF-β reporter⁺ and reporter^neg tumor cells (fig. S5A). Therefore, IL-33 released from TICs may act as a short-distance paracrine signal to induce the differentiation of FcεRIα⁺ macrophages.

Fig. 5. IL-33-induced FcεRIα⁺ macrophages activate paracrine TGF-β signaling and epithelial cell invasion in vitro.

(A) Histogram presentation of ST2 expression in FcεRIα⁺ and FcεRIα^neg macrophages in SCC. (B) Flow cytometry analysis of FcεRIα and ST2 expression in CSF1- and IL-33–induced F4/80⁺ macrophages from bone marrow (BM)–derived cells (see fig. S5B). (C) qPCR analysis of CSF1-induced, IL-33–induced, and CSF1-induced→IL-4–treated macrophages in vitro (n = 3). Data are shown as relative mean expression with SD. (D) qPCR analysis of IL-33–induced macrophages treated with a MEK inhibitor (U0126, 0.5 –M), a JNK inhibitor (SP600125, 0.5 –M), a p38 inhibitor (SB202190, 0.5 –M), or an NF-κB inhibitor (BAY11-7082, 1 μM) during macrophage differentiation (n = 3). Data are shown as relative mean expression with SD and were analyzed with unpaired t test, **P = 0.0022. (E) qPCR analysis of IL-33–induced and IL-4–activated macrophages treated with the NF-κB inhibitor (n = 3). Data are shown as relative mean expression with SD and were analyzed with unpaired t test, ***P < 0.001. (F and G) qPCR analysis of IL-33–induced macrophages from ex vivo expanded hematopoietic progenitor cells that were LV transduced with scramble control, (F) Rela (NF-κB p65) shRNA, or (G) Fcer1a shRNA (see fig. S5E). (H) LV-transduced YFP⁺ MKs were cocultured with CSF1- or IL-33–induced YFP^neg macrophages (MΦs) for 24 hours. (Graph) Quantification of TGF-β fluorescent reporter intensity in YFP⁺ MKs. Data are shown in box-and-whisker plots (midline, median; box, 25th and 75th percentiles; whiskers, 5th and 95th percentiles with outliers); ***P < 0.001. ns, not significant. Scale bars, 50 μm. (I) Quantification of the circularity of YFP⁺ MKs cocultured with CSF1- or IL-33–induced MFs. Data are shown in box-and-whisker plots, ***P < 0.001. (J and K) Quantification of YFP⁺ MKs invaded through Matrigel-coated membranes. (J) MKs were cultured with CSF1- or IL-33–induced MΦs or with (K) IL-33–induced MΦs transduced scramble control or Tgfb1 shRNA (n = 3). Data are shown as mean with SD, *P < 0.05, **P < 0.01, ***P < 0.001. Scale bar, 50 μm.

To address this possibility, we used an in vitro culture system utilizing bone marrow–derived hematopoietic progenitor cells. Similar to macrophage colony-stimulating factor (CSF1), IL-33 treatment induced macrophage differentiation in 6 days (fig. S5B). Flow cytometry, qPCR, and immunofluorescence analyses showed that FcεRIα and ST2 expression were most robustly elevated in IL-33–induced macrophages (Fig. 5, B and C, and fig. S5C). Moreover, CSF1-induced macrophages could acquire FcεRIα expression by subsequent IL-33 stimulation (fig. S5, C and D). These results suggest that IL-33–ST2 signaling directly induces the differentiation of FcεRIα⁺ macrophages from immature myeloid cells and preexisting macrophages.

Similar to FcεRIα⁺ macrophages in vivo, IL-33–induced FcεRIα⁺ macrophages in vitro also expressed several markers of alternative activation, e.g., Arg1 and Tgfb1, at higher levels comparable to those of macrophages activated by IL-4 (Fig. 5C). However, IL-4–activated macrophages expressed significantly less FcεRIα and ST2, suggesting that IL-33 may induce FcεRIα⁺ M2 macrophages through a specific signaling pathway. To understand the mechanism of FcεRIα⁺ macrophage differentiation, we examined the impact of small-molecule inhibitors that target ST2 downstream signaling pathways, including extracellular signal-regulated kinase (ERK), c-Jun N-terminal kinase (JNK), p38 mitogen-activated protein kinase (MAPK), and nuclear factor κB (NF-κB) (44). When we used inhibitor concentrations that permit bone marrow–derived cells to differentiate into macrophages, an NF-κB inhibitor, BAY 11-7082, significantly down-regulated FcεRIα expression (Fig. 5D). The NF-κB inhibitor also repressed Arg1 expression up-regulated by IL-33, but not by IL-4 (Fig. 5E), suggesting that the ST2–NF-κB pathway is critical for FcεRIα⁺ macrophages. Moreover, we transduced hematopoietic progenitor cells expanded ex vivo (45) with LV-puro vectors containing NF-κB p65 (Rela) shRNA. After puromycin selection, we induced macrophage differentiation in vitro (fig. S5E). Again, p65 KD significantly repressed Fcer1a and Arg1 expression induced by IL-33 without affecting F4/80 expression (Fig. 5F). Furthermore, Fcer1a KD reduced the expression of Arg1 and Tgfb1 (Fig. 5G), suggesting that FcεRIα may be involved in alternative activation of macrophages induced by IL-33.

Next, we tested whether IL-33–induced macrophages could activate paracrine TGF-β signaling by coculture with YFP⁺ mouse keratinocytes (MKs) harboring the TGF-β reporter (fig. S5F). TGF-β reporter expression (Fig. 5H) and SMAD2 phosphorylation (fig. S5G) in keratinocytes were significantly induced by IL-33–induced macrophages, and this effect was mitigated by the addition of anti–TGF-β-neutralizing antibody in the culture medium (fig. S5H). TGF-β reporter⁺ keratinocytes cultured with IL-33–induced macrophages exhibited an elongated cell shape (Fig. 5I), suggesting an induction of invasive properties. We previously showed that TGF-β–responding tumor cells in vivo show partial epithelial-mesenchymal transition (EMT) phenotypes (17, 46). Such phenotypes, including the redistribution of E-cadherin from cell–cell junctions to cytoplasmic granules and the expression of an EMT transcription factor ZEB2 (fig. S5, I and J), were recapitulated in keratinocytes cultured with IL-33–induced macrophages. To functionally assess invasive properties, we performed a transwell assay by seeding YFP⁺ keratinocytes and YFP^neg macrophages on a layer of Matrigel. The number of keratinocytes that invaded to the other side of the transwell significantly increased in the presence of IL-33–induced macrophages compared with those with CSF1-induced macrophages (Fig. 5J), and this effect was mitigated by Tgfb1 KD in IL-33–induced macrophages (Fig. 5K and fig. S5K). On the basis of these data, we conclude that IL-33 can induce the differentiation of immature myeloid cells into FcεRIα⁺ macrophages that are capable of activating paracrine TGF-β signaling and invasive properties in epithelial cells.

TIC-derived IL-33 accumulates FcεRIα⁺ macrophages to create a TGF-β-rich, tumor-promoting niche

Our in vitro experiments provided the mechanistic basis of potential TIC–niche interactions that drive invasive tumor progression. We then sought to determine the impact of tumor-derived IL-33 on stromal cells and the IL-33–related mechanisms on SCC progression in vivo. Flow cytometry analysis showed that the overall proportions of YFP⁺ tumor cells, CD45⁺ immune cells, and the double-negative cells (including fibroblasts and endothelial cells) did not significantly differ between Il33 KD and control tumors (fig. S6A). Additionally, the frequency of F4/80⁺ macrophages in the CD45⁺ population was not affected by Il33 KD in vivo (Fig. 6A). However, the proportion of FcεRIα⁺ST2⁺ cells in the F4/80⁺ macrophages was significantly lower in Il33 KD tumors compared with control tumors (Fig. 6B). Moreover, immunofluorescence analysis of Il33 KD tumors revealed a significant reduction of FcεRIα⁺TGF-β⁺ cells in the stroma compared with control tumors (Fig. 6C). These results suggest that IL-33 released from the TIC population increases the density of FcεRIα⁺TGF-β⁺ macrophages in the adjacent stroma.

Fig. 6. FcεRIα⁺ macrophages mediate IL-33–induced paracrine TGF-β signaling and invasive tumor progression.

(A) No difference is observed in the frequency of F4/80⁺ macrophages in live, CD45⁺ cells in control and Il33 KD tumors. Scramble control, n = 30; Il33 KD, n = 20. Data are shown as mean with SEM. ns, not significant. (B) F4/80⁺ macrophages in Il33 KD tumors have a smaller FcεRIα⁺ST2⁺ double-positive population than control tumors. (Graph) Quantification of FcεRIα⁺ST2⁺ macrophages. Scramble control, n = 27; Il33 KD, n = 20. Data are shown as mean with SEM and were analyzed with unpaired t test, ***P < 0.001. (C) Immunolabeling of tumor sections showing fewer FcεRIα⁺TGF-β⁺ cells in the stroma of Il33 KD tumors compared with control. (Graph) Quantification of FcεRIα⁺ cells in the stromal area within a 50-μm radius of tumor edges analyzed with unpaired t test, *P = 0.0394. (D) Immunolabeling of tumor sections showing that sST2-overexpressing (OE) tumors have fewer FcεRIα⁺ cells in the adjacent stroma. (Graph) Quantification of FcεRIα⁺ cells in the stromal area within a 50-μm radius of tumor edges analyzed with unpaired t test, ***P < 0.001. (E) HRAS^G12V-driven tumors overexpressing sST2 or control were sized. Control, n = 6; sST2 OE, n = 6. Approximation curves were drawn by applying the Michaelis-Menten kinetics curve. (F) Immunolabeling of K10 showing a sustained differentiation property in sST2-overexpressing tumors. (G) Immunolabeling of IL-33 and FcεRIα in Nrf2 (Nfe2l2) or Keap1 KD tumors. Whereas Nrf2 KD tumors show mostly nuclear IL-33, Keap1 KD tumors show cytoplasmic IL-33 localization. (Graph) Quantification of FcεRIα⁺ cells in the stromal area within a 50-μm radius of the tumor leading edges (n = 3-4). Data are shown in box-and-whisker plots, *P = 0.0394 (control versus Il33 KD) or 0.0372 (Nrf2 KD versus Keap1 KD). (H) Immunolabeling of tumor sections showing tdTomato⁺ immune cells infiltrated from the circulation at 7 d after injection. Note that control tdTomato⁺ cells expressed FcεRIα (arrowheads) but ST2 KD cells were largely negative (arrows). Magnified images are shown in fig. S6I. (Graph) Quantification of tdTomato⁺ cells in the stroma and FcεRIα-expressing tdTomato⁺ cells (n = 4) analyzed with unpaired t test, **P = 0.0019. (I) FcεRIα⁺ cell depletion by anti-FcεRIα antibody injection in vivo. Quantification of cells expressing ST2 (a surrogate marker for FcεRIα) within the stromal area in a 50-μm radius of tumor basal cells (n = 3), ***P < 0.001. (J) Quantification of TGF-β fluorescent reporter intensity in K5 basal cells. Intensity values in each image were normalized by the median values (set as 1) (n = 3) and analyzed with unpaired t test, ***P < 0.001. (K) Tumors were sized before and 1 week after anti-FcεRIα antibody injection. Isotype control, n = 3, 10 tumors; anti-FcεRIα, n = 3, 10 tumors. **P = 0.0016. (L) Model of TIC-driven feedforward mechanism of invasive SCC progression. TGF-β–responding TICs release IL-33 through the NRF2-mediated antioxidant response, which induces differentiation of immature myeloid cells into FcεRIα⁺ macrophages in their close proximity. In turn, FcεRIα⁺ macrophages send reciprocal paracrine TGF-β signaling to TICs to promote invasive progression and drug resistance of SCC, and further induce the release of IL-33, establishing a self-reinforcing niche signaling loop between TICs and FcεRIα⁺ macrophages. Dotted lines denote the tumor–stroma boundaries. Scale bar, 50 μm.

To further substantiate the function of IL-33 released from TICs toward FcεRIα⁺ macrophage differentiation and tumor progression, we generated LV vectors expressing soluble ST2 (sST2), a decoy receptor for IL-33, to sequester extracellular IL-33 and to block ST2 signaling (43) (fig. S6B). In vitro, cells transduced with this LV vector secreted sST2 in the culture medium in a Dox-dependent manner (fig. S6C), and the conditioned medium of sST2-expressing cells suppressed macrophage differentiation induced by IL-33, but not by CSF1 (fig. S6D). Using this LV vector, we induced HRAS^G12V-driven tumors that ectopically expressed sST2 in vivo and detected a significant reduction of FcεRIα⁺ cells in the stroma (Fig. 6D). Moreover, the growth of sST2-overexpressing tumors was stunted (Fig. 6E), and they maintained a well-differentiated status similar to that of Il33 KD tumors (Fig. 6F), further supporting the function of IL-33 in FcεRIα⁺ macrophage differentiation and tumor progression in vivo. We also evaluated the impact of the NRF2-mediated IL-33 release on the accumulation of FcεRIα⁺ macrophages in the stroma by depleting Nrf2 or Keap1 in vivo. Consistent with our in vitro data (Fig. 2I), Nrf2 KD tumors largely maintained nuclear IL-33 (Fig. 6G). By contrast, Keap1 KD tumors showed cytoplasmic localization of IL-33 in tumor cells and increased numbers of FcεRIα⁺ macrophages in the stroma.

It has been reported that elevated IL-33 serum levels are associated with various inflammatory diseases (20, 47). Therefore, tumor-derived IL-33 might systemically induce myelopoiesis or differentiation of immature myeloid cells in hematopoietic organs such as the spleen and bone marrow (48). However, tumor-induced splenomegaly was observed at similar levels in control and Il33 KD tumor-bearing mice (fig. S6E). Likewise, myelopoietic states in the spleen and bone marrow were similarly exacerbated in the two groups (fig. S6F). Additionally, FcεRIα⁺ cells remained scarce (< 0.3%) in the spleen and bone marrow of tumor-bearing mice in both groups (fig. S6G). These results suggest that immature myeloid cells could be recruited into tumor tissues through IL-33–independent mechanisms, and that IL-33 acts as a short-distance cue in the TIC microenvironment to generate FcεRIα⁺ macrophages. To test these possibilities, we injected ex vivo expanded hematopoietic progenitor cells marked by cell membrane–localized tdTomato into the circulation of tumor-bearing mice. At 7 days after injection, transplanted tdTomato⁺ cells were detected at the tumor stroma and a fraction of them expressed FcεRIα (Fig. 6H). Moreover, ST2 KD in tdTomato⁺ cells (fig. S6H) did not prevent their infiltration into the stroma but rather suppressed FcεRIα expression (Fig. 6H and fig. S6I), supporting the idea that recruited immature myeloid cells are influenced by TIC-derived IL-33 to differentiate into FcεRIα⁺ macrophages.

Finally, we sought to determine whether FcεRIα⁺ macrophages are a critical mediator of IL-33–induced paracrine TGF-β signaling and tumor progression in vivo. We injected the anti-FcεRIα antibody into tumor-bearing mice to deplete FcεRIα⁺ macrophages through antibody-dependent cellular cytotoxicity. At 3 days after injection, efficient depletion was confirmed by immunofluorescence analysis of tumor sections using an anti-ST2 antibody (Fig. 6I). Both the frequency and intensity of TGF-β reporter expression were significantly decreased in FcεRIα⁺ cell-depleted tumors compared with control (Fig. 6J). By 7 days after injection, the growth of FcεRIα⁺ cell-depleted tumors was stunted compared with control tumors (Fig. 6K). To exclude the potential effect of FcεRIα⁺ mast cells, we also injected the anti-FcεRIα antibody into tumor-bearing, mast cell–deficient (Kit^w-sh) mice (49). We confirmed that FcεRIα⁺ macrophages exist around invasive, TGF-β–responding cells in a mast cell–deficient condition and were depleted by the anti-FcεRIα antibody (fig. S6, J to L). Moreover, FcεRIα⁺ cell depletion under this condition still exhibited growth-suppressing effects, similar to the effect seen in control tumors (fig. S6M).

Our studies provide evidence of a feedforward TIC–niche interaction model. The TIC-intrinsic, NRF2-mediated antioxidant response triggers the extracellular release of IL-33 from TGF-β–responding TICs, which induces TGF-β–rich tumor microenvironments by accumulating FcεRIα⁺ macrophages in their close proximity. This self-reinforcing, TIC–niche signaling loop may prove to be a critical driver of invasive cancer progression and drug resistance in human SCC (Fig. 6L).

Discussion

The development of effective TIC-targeted therapies is moving at a restricted pace because of the lack of identification of TIC vulnerabilities. Some of these vulnerabilities may stem from the cross-talk between TICs and their niches, which is critical in maintaining, or potentially inducing, stem-like properties in tumor cells. Here, focusing on TGF-β–responding tumor cells in a mouse model of SCC, a previously characterized TIC population (17), we uncover a mechanism by which TICs generate the niche microenvironment that is required for invasive progression and drug resistance in SCC.

We identified IL-33 as the most significantly up-regulated and highly expressed cytokine in TGF-β–responding tumor cells. IL-33 is known to play multiple and sometime opposing roles in tumorigenesis (21). It has been reported that overexpression of IL-33 inhibits tumor growth by enhancing cytotoxic T and NK cell activation (50, 51), whereas mice lacking ST2 also attenuated tumor progression by increasing T helper 1 (T_H1)/T_H17 cytokines and cytotoxic activity of NK cells (52). Here, we demonstrate that IL-33 has a dominant role in driving tumor progression in SCC and describe how this cytokine evokes pro-tumorigenic changes in the tumor microenvironment. In addition to IL-33, other cytokines might be regulated by TGF-β–responding tumor cells to modulate tumor-promoting and immunosuppressive microenvironments. For example, TSLP, a significantly down-regulated cytokine in TGF-β–responding cells (Fig. 1C), is known to reduce the incidence of chemically induced skin squamous carcinogenesis (53), and the loss of TSLP signaling accelerates malignancy by recruiting CD11b⁺Gr1⁺ myeloid cells (54). Therefore, it is likely that the function of IL-33 in tumorigenesis is tightly regulated by the cytokine milieu where IL-33–responding cells are located.

We have demonstrated that the NRF2-mediated antioxidant response activated in TGF-β–responding TICs is involved in the extracellular release of IL-33. It was previously reported that mechanical stress triggers the extracellular secretion of nuclear IL-33 (29), and that mechanical stress also induces reactive oxygen species (ROS) and activates the NRF2-mediated antioxidant responses (55, 56). Our previous work showed that TGF-β–responding tumor cells maintain lower ROS levels compared with their nonresponding counterparts, most likely because of enhanced NRF2 activity (17). Although the molecular mechanism of IL-33 release remains to be clarified, our study has revealed an interesting link between this TIC-intrinsic, stress-resistant activity and spatially restricted cytokine release as a mechanism for TIC niche formation.

The cellular components and molecular pathways that are essential for building the TIC niche remain unclear; however, tumor-associated macrophages are considered to be crucial players (40, 57, 58). Thus, the mechanisms underlying the emergence of a spatial and functional relationship between TICs and tumor-associated macrophages may provide a critical target for destabilizing TICs. The discovery of FcεRIα⁺ macrophages and the IL-33–TGF-β signaling loop that connects TICs and those macrophages is of great importance because it adds to the understanding of heterogeneity in tumor-associated macrophages and provides mechanistic insights into the development of the TIC niche. Therefore, targeting IL-33- or FcεRIα⁺ macrophage–relevant pathways may be an attractive avenue to develop durable cancer treatments, especially when conventional approaches to modulating tumor-associated macrophages as a whole are ineffective. Recently, it has been reported that a subset of macrophages differentiated from human peripheral blood monocytes express FcεRIα and markers of alternative activation (59) and that IgE plays a dual role in tumor suppression and promotion (60, 61). Therefore, it is becoming increasingly important to understand tumor stage–specific regulation of FcεRIα⁺ macrophages by IgE-related immune mechanisms.

Finally, normal stem cell niches not only protect tissue-resident stem cells from differentiation and apoptotic stimuli, but also act as a safeguard against the excessive stem cell production that can lead to cancer. The TIC niche may differ in that it can be expanded to accommodate an increasing number of TICs. The TIC-driven, self-reinforcing TIC–niche interactions may underlie the ability of cancers to expand and progress. The question of whether IL-33 or TGF-β can trigger the inception of the niche signaling loop remains open. Given that IL-33 is stored in the nucleus of normal and premalignant cells, the release of IL-33 through either NRF2-dependent or- independent mechanisms might precede the deposition of TGF-β in the stroma. Moreover, IL-33 and TGF-β are initially released in less potent and latent forms, respectively, and both are activated in the extracellular space by various mechanisms (20, 62). Therefore, it is likely that the establishment of a robust IL-33–TGF-β signaling loop is regulated by another layer of cellular and molecular interplay, which could also be a potential target to destabilize TIC–niche interactions.

Materials and Methods

LV construction, production, and concentration

For generation of an LV rtTA-2A-Cre construct, we replaced the puromycin cassette of pLKO.1 between BamHI and NsiI sites with an open reading flame containing synthesized cDNA of the reverse tetracycline transactivator (rtTA3), a P2A linker sequence, and a PCR-amplified cDNA of codon-improved Cre recombinase (iCre). To generate an LV vector containing TGF-β reporter, we upgraded the construct with a brighter version of nuclear red fluorescent protein, NLS-mScarlet-I (Addgene plasmid 85044) (46) and inserted the chicken hypersensitivity site 4 insulator (cHS4) (47) that improves sensitivity and transgene expression. To generate an LV-rtTA-2A-Cre–TGF-β reporter construct, we created a multicloning site sequence containing PmeI-NheI-BmtI-XbaI-SalI-MluI sites between NsiI and KpnI, and then inserted a DNA fragment containing 12× repeated SMAD-binding elements (SBE, 5′-AGCCAGACA-3′), a 113-bp minimal CMV promoter, mScarlet-I, and cHS4 insulator sequences between KpnI and NheI sites. Depending on the experiments, we also used the LV-rtTA–TGF-β reporter construct (no Cre), particularly when the Rosa-YFP expression was not ideal and when we used K14-CreER transgenic mice. For gene KD experiments, we used LV-puro-shRNA vectors that allowed us to select transduced cells by puromycin selection. For Dox-inducible sST2 expression, we cloned the sST2 coding sequence under the tetO promoter and inserted this fragment by replacing the TGF-β reporter cassette. DNA plasmid mini/midi/maxi-prep, gel extraction, and PCR clean-up kits from Zymo Research were used. Large-scale production of VSV-G pseudotyped LV was performed by calcium phosphate transduction of 293T cells (Clontech/Takara, 632180) plated on poly-l-lysine–coated cell culture dishes with pLKO.1 and helper plasmids: pMD2.G and psPAX2 (Addgene plasmid, 12259 and 12260). Viral supernatant was collected 30 to 36 hours after transfection, filtered through a 0.45-μm polyvinylidene difluoride (PVDF) filter (Millipore), concentrated first by passing through a Vivacell 70/100 centrifugal concentrator (100-kDa molecular weight cutoff, Sartorius, VS6041), and then by ultracentrifugation in an Ultra-Clear tube (Beckman Coulter, 344057) for 90 min in an MLS-50 rotor at 45,000 rpm or a SW-55 Ti rotor at 42,000 rpm (Beckman Coulter).

Mice, ultrasound-guided in utero microinjection, and tumor formation analysis

The following mice were used: Tg(TetO-HRAS)65Lc/Nci (TetO-Hras^G12V, NCI Mouse Repository, donated by L. Chin) (23), Gt(ROSA)26Sor^tm(EYFP)Cos/J (Rosa26-lox-stop-lox(LSL)-EYFP, The Jackson Laboratory, donated by F. Costantini) (24), B6;129-Tgfbr2^tm1Karl/J (Tgfbr2-floxed, shared by S. Karlsson), Tg(KRT14-cre/ERT)20Efu/J (K14-CreER, shared by E. Fuchs), Gt(ROSA)26Sor^{tm4(ACTB-tdTomato,-EGFP)Luo}/J (mT/mG, The Jackson Laboratory, donated by L. Luo), and B6.Cg-Kit^W-sh/HNihrJaeBsmJ (Kit^Wsh/Wsh, The Jackson Laboratory). Mice were backcrossed into the FVB/NJ for more than five generations or the C57BL/6J for 10 generations. Ultrasound-guided in utero microinjection was performed as described previously (21). Briefly, females at embryonic day 9.5 of gestation were anesthetized with isoflurane on a heated stage before injection. Then, 300 to 500 nl of LV suspension was injected into each embryo’s amniotic sac. Surgical procedures were limited to 30 min to maintain high survival rates. Postnatally, the LV-transduced mice were given mouse chow containing Dox (2 mg/kg, 5AKR, TestDiet), which activates rtTA to induce the expression of oncogenic HRAS^G12V from the TetO-Hras transgene. Cisplatin (Sigma-Aldrich, P4394) was dissolved in saline at 0.5 mg/ml and administrated by intraperitoneal injection (10 mg/kg). For cell proliferation assessment in vivo, EdU (25 μg/g, Invitrogen, A10044) was injected intra–peritoneally 4 hours before lethal administration of CO₂. For FcεRIα⁺ macrophage depletion, an in vivo grade anti-FcεRIα antibody (Tonbo Biosciences, MAR-1) was subcutaneously injected twice a week (100 μg/injection), and an isotype control antibody (Bio X Cell, Armenian hamster IgG) was used in the control cohorts. For intradermal orthotropic injection, 1 × 10⁶ cells (TetO-Hras^G12V) transduced with the LV-rtTA–TGF-β reporter with Matrigel (Cultrex, type 3, R&D systems) were injected. For inoculation of macrophage progenitor cells in the circulation, ex vivo expanded tdTomato⁺ hematopoietic progenitor cells (2 × 10⁵ cells) were suspended in phosphate-buffered saline (PBS, 50 μl) and injected into the retro-orbital sinus of tumor-bearing mice. The Oregon Health & Science University (OHSU) Animal Care and Use Committee approved the animal experimentation protocols used in this study.

Tissue harvest, sectioning, and immunofluorescence microscope imaging

Because of humane end points based on tumor size, number, and location on the body surface, mice were euthanized at different time points between 23 and 58 days after Dox-induced HRAS^G12V expression. Before tumor collection, dorsal hairs were removed with either an electric hair clipper or hair removal cream. Mice were then photographed, and tumor size was measured with a ruler. Tumors were dissected and fixed with 4% paraformaldehyde (16% solution, Electron Microscopy Sciences) in PBS for 15 min at room temperature. After washing with PBS overnight, tumor tissues were embedded and cryo-frozen in Tissue-Tek OCT compound (Sakura, 4583) in a cryomold (Sakura, 4557). Then, 5- to 10-μm cryosections were cut with a Cryostat (Leica, CM1850), mounted on SuperFrost Plus slides (Fisherbrand), and permeabilized for 30 min in 0.3% Triton X-100 in PBS. Tissues to be stained using the TSA Plus signal amplification kit (PerkinElmer, NEL744001KT) were treated with 1% H₂O₂ for 30 min at room temperature. When immunolabeling with mouse antibodies, sections were incubated with the M.O.M. blocking kit according to the manufacturer’s instructions (Vector Laboratories, BMK-2202). The following primary antibodies were used: chicken anti-GFP (to detect YFP) (Abcam, ab13970), chicken anti-keratin 5 (Covance, SIG-3475), rabbit anti–TGF-β (R&D Systems, AB-100-NA), rabbit anti–phospho-SMAD2 (Cell Signaling Technologies, 138D4), goat anti–IL-33 (R&D Systems, AF3626), rabbit anti-NRF2 (MBL International, PM069), rat anti-CD45 (BD Biosciences, 30-F11), rat anti-CD11b (BD Biosciences, M1/70), rat anti-CD3 (BD Biosciences, 17A2), rat anti-CD8a (eBioscience, 4SM15), rat anti-CD4 (eBioscience, 4SM95), rat Ly6G (BioLegend, HK1.4), rat anti-F4/80 (Bio-Rad, Cl:A3-1), Armenian hamster anti-FcεRIα (BioLegend, MAR-1), rat anti-integrin α6 (BD Biosciences, GoH3), mouse anti–α-tubulin (Sigma-Aldrich, DM1A), rabbit anti–keratin 14 (Abcam, EPR17350), rabbit anti–keratin 10 (Abcam, EP1607IHCY), goat anti-CD31 (R&D Systems, AF3628), goat antitryptase/MCP-6 (R&D Systems, AF3736), goat antihuman IL-33 (R&D Systems, AF3025), rabbit anti-CD206 (Abcam, ab64693), rabbit anti–E-cadherin (Cell Signaling Technologies, 24E10), rabbit anti SIP1/ZEB2 (Bethyl, IHC-00691), and rabbit anti–V5-tag (Abcam, ab9116). Sections were treated with primary antibody mixes and incubated at 4°C overnight. After washing with PBS with 0.1% Triton X-100 (PBST), sections were treated for 30 min at room temperature with secondary antibodies conjugated with Alexa Fluor 488, 546, or 647 (Jackson ImmunoResearch and ThermoFisher Scientific). Slides were washed, counterstained with 4′6′-diamidino-2-phenylindole (DAPI), and mounted in Prolong Gold (ThermoFisher Scientific). Imaging was performed on a Zeiss Axio Observer.Z1 microscope equipped with Apotome.2 using ×10/0.45, ×20/0.8, or ×40/0.95 air. Images were collected using Zeiss ZEN software and processed using an image-processing package, Fiji (https://fiji.sc/). For reporter intensity analysis, nuclei were selected and mean intensity per area was calculated after background subtraction. For immune cell quantification, images were framed to capture regions of the tumor–stroma interface containing high (>30%)- and low (<30%)-reporter–positive areas of the tumors. This was done to minimize the effects of variation in image quality or settings between regions being compared. Using ImageJ’s ruler and measurement tools, a continuous area extending 50 μm radially from the border of the tumor into the stroma was manually marked out for quantification for both the high- and low-reporter regions. For reporter-positive regions, the area began at the first reporter-positive cell and ended once there were no more reporter-positive cells within 50 μm. For reporter-negative regions, the area was set to capture as much area as possible without coming within 50 μm of the reporter-positive region. The Alexa Fluor 647/Cy5 channel was used for most of the immune cell markers to minimize autofluorescence. Immune marker–positive cells were quantified by counting stromal cells that stained with both the immune cell marker and had clear, single-nucleus DAPI staining. To analyze circularity, a mask was created for the YFP channel, individual cells were selected with the “analyze particles” function, and then circularity [4π(area/perimeter²)] was measured.

In vitro cell culture experiments

Keratinocytes isolated from neonatal TetO-Hras^G12V; Rosa26-LSL-YFP transgenic mouse epidermis were maintained in E medium with 10% fetal bovine serum (FBS) and 50 μM CaCl₂ at 37°C with 7.5% CO₂. For LV transduction in culture, cells were plated in six-well plates at 1.0 × 10⁵ cells per well, incubated with viruses in the presence of polybrene (10 μg/ml) for 30 min at 37°C and 7.5% CO₂, and then plates were spun at 1100g for 30 min at 37°C in an Eppendorf 5810R centrifuge. After centrifugation, virus-containing medium was removed and replaced with fresh growth medium. For H₂O₂-induced oxidative stress, 100 to 500 μM H₂O₂ was added to the culture medium. For bone marrow–derived cell culture, bone marrow cells in femurs from adult mice (7 to 10 weeks old) were collected by flushing with Dulbecco’s modified Eagle’s medium (DMEM) with 10% FBS, and filtered through a 40-μm cell strainer. The EasySep mouse CD117 (c-Kit) positive selection kit (STEMCELL Technologies, 18757) was used to enrich c-Kit⁺ hematopoietic stem/progenitor cells that were cultured in DMEM with 10% FBS at 37°C and 7.5% CO₂. For ex vivo expansion of c-Kit⁺ cells, we followed the culture protocol described in (45). For stimulation experiments, medium was supplemented with either recombinant murine TGF-β1 (50-100 pM, R&D Systems, 7666-MB), recombinant murine IL-33 (50 ng/ml, PeproTech, 210-33), recombinant murine M-CSF/CSF1 (10 ng/ml, PeproTech, 315-02), or recombinant murine IL-4 (10 ng/ml, PeproTech, 214-14). The anti–TGF-β–neutralizing antibodies (R&D Systems, AB-100-NA) were used for blocking experiments. For immuno-labeling and visualization of in vitro culture cells, cells were seeded onto coverslips and fixed with prewarmed 4% paraformaldehyde for 10 min at 37°C. Cells were permeabilized with 0.3% Triton X-100 in PBS and incubated with the primary antibodies listed above at 4°C overnight. After washing with PBST, cells were treated for 30 min at room temperature with secondary antibodies conjugated with Alexa Fluor 488, 546, or 647. Coverslips were washed, counterstained with DAPI, and mounted in Prolong Gold.

FACS and flow cytometry

Tumors were dissected from mice, minced by scalpels in a chilled dish on ice, and treated with 0.2 mg/ml collagenase (Sigma-Aldrich C2674 or Roche 10103586001) in Hank’s balanced salt solution with 10 mM Hepes for 30 min at 37°C or overnight at 4°C. Mouse back skin (the dermis side) was also digested in the same collagenase solution. The collagenase-digested tissue fraction was vigorously pipetted up and down by adding ice-cold PBS and pelleted at 400g for 15 min at 4°C. Precipitated tissue fractions were resuspended in 10× TrypLE (Invitrogen, A12177) to dissociate into single cells at 37°C for 15 min, and filtered through 100-, 70-, and 40-μm cell strainers. Cells were pelleted at 400g for 10 min, and precipitated cells were washed with staining buffer (0.1% bovine serum albumin in PBS) and pelleted again at 400g for 5 min. Total cell number was counted with a Countess Automated Cell Counter (Invitrogen), and cell density was adjusted to 2 million cells per 100 μl. Cells were blocked with 1 μg/ml TruStain FcX (anti-mouse CD16/32) antibody in staining buffer for 15 min at 4°C. To stain for ST2, biotinylated anti-ST2 antibody (1:1000, MD Biosciences, DJ8) was used. After removing unbound antibodies, cells were resuspended in staining buffer and incubated with antibodies for cell surface markers and Live/Dead Aqua (1:250, Invitrogen, L34957) for 30 min at 4°C. The following antibodies were used: CD49f/integrin α6-BV421 (1:250, BD Biosciences, clone GoH3), CD45-PerCP/Cy5.5 (1:100, BD Biosciences, clone 30-F11), F4/80-APC (1:200, BioLegend, clone BM8), FcεRIα-PE/Cy7 (1:2000, BioLegend, clone MAR-1), CD326/EpCAM-FITC (1:200, eBioscience, clone G8.8), MHCII-PerCP (1:200, BioLegend, clone M5/114.15.2), Ly6C-PerCP/Cy5.5 (1:400, eBioscience, clone HK1.4), Ly6G-APC/Cy7 (1:400, BD Biosciences, clone 1A8), CD11c-BV605 (1:200, BioLegend, clone N418), CD11b-BV650 (1:1000, BD Biosciences, clone M1/70), Siglec-F-BV711 (1:250, BD Biosciences, clone E50-2440), CD45-BV786 (1:2000, BD Biosciences, clone 30-F11), and CD117/c-Kit-PE (1:500, BioLegend, clone ACK2). Cell isolations were performed on a BD Influx cell sorter equipped with BD FACS software (BD Biosciences). Sorted cells were used for RNA preparation, RNA-seq, and real-time qPCR. Flow cytometry analysis was performed on a BD LSRFortessa Cell analyzer (BD Biosciences).

RNA purification, RNA-seq, and real-time qPCR

Total RNA from FACS-purified cells directly sorted into TRI Reagent (Zymo Research) was purified using the Direct-zol RNA MicroPrep Kit (Zymo Research) per the manufacturer’s instructions. The quality of the total RNA for sequencing was determined with a bioanalyzer, and all samples used had RNA Integrity Score numbers above the standard required value. Library preparation using the Illumina TruSeq mRNA sample preparation kit was performed at the OHSU Massively Parallel Sequencing Shared Resource (MPSSR), and RNAs were single-end sequenced on Illumina HiSeq 2500 machines. Alignment of reads was done using the STAR aligner with the mm10 build of the mouse genome. Transcript assembly and differential expression was determined using DESeq2 with Refseq mRNAs to guide assembly. For reverse transcription (RT)–qPCR, equivalent amounts of RNA were reverse-transcribed using the SuperScript IV VILO cDNA synthesis kit (Invitrogen, 11756050). cDNAs were mixed with the indicated primers and PowerUp SYBR Green Master Mix (Applied Biosystems, A25742), and RT-qPCR was performed on a ViiA 7 Real-time PCR System (Applied Biosystems). cDNAs were normalized to equal amounts using primers against Actb, Gapdh, or Ppib. The following primer sequences were used (5′-3′): Actb forward: GGCTGTATTCCCCTCCATCG, Actb reverse: CCAGTTGGTAACAATGCCATGT; Fcer1a forward: GAGTGCCACCGTTCAAGACA, Fcer1a reverse: GTAGATCACCTTGCGGACATTC; Nos2 forward: ACATCGACCCGTCCACAGTAT, Nos2 reverse: CAGAGGGGTAGGCTTGTCTC; Arg1 forward: CTCCAAGCCAAAGTCCTTAGAG, Arg1 reverse: AGGAGCTGTCATTAGGGACATC; Tgfb1 forward: GAGCCCGAAGCGGACTACTATG, Tgfb1 reverse: CAGCCACTGCCGTACAACTCC; Il33 forward: ACTGCATGAGACTCCGTTCTG, Il33 reverse: CCTAGAATCCCGTGGATAGGC; Adgre1 (F4/80) forward: CTTTGGCTATGGGCTTCCAGTC, Adgre1 (F4/80) reverse: GCAAGGAGGACAGAGTTTATCGTG; Il1rl1 (ST2) forward: TGACACCTTACAAAACCCGGA, Il1rl1 (ST2) reverse: AGGTCTCTCCCATAAATGCACA; Nfe2l2 (Nrf2) forward: TCTTGGAGTAAGTCGAGAAGTGT, Nfe2l2 (Nrf2) reverse: GTTGAAACTGAGCGAAAAAGGC; Keap1 forward: TGCCCCTGTGGTCAAAGTG, Keap1 reverse: GGTTCGGTTACCGTCCTGC; Rela (p65) forward: ACTGCCGGGATGGCTACTAT, Rela (p65) reverse: TCTGGATTCGCTGGCTAATGG; Hmox1 forward: AAGCCGAGAATGCTGAGTTCA, Hmox1 reverse: GCCGTGTAGATATGGTACAAGGA; Me2 forward: TCAACAAGGACTTGGCTTTTACT, Me2 reverse: TGCAGGTCCATTAACAGGAGAT; Krt14 forward: AGCGGCAAGAGTGAGATTTCT, Krt14 reverse: CCTCCAGGTTATTCTCCAGGG; Ptprc (CD45) forward: GTTTTCGCTACATGACTGCACA, Ptprc (CD45) reverse: AGGTTGTCCAACTGACATCTTTC; Pdgfra (CD140a) forward: TCCATGCTAGACTCAGAAGTCA, Pdgfra (CD140a) reverse: TCCCGGTGGACACAATTTTTC; and Pecam1 (CD31) forward: ACGCTGGTGCTCTATGCAAG, Pecam1 (CD31) reverse: TCAGTTGCTGCCCATTCATCA.

shRNA sequences for gene KD

For gene KD experiments, DNA oligos of following target sequences were synthesized (5′-3′): the scramble shRNA: CAACAAGATGAAGAGCACCAA; Il33 sh1: GCATCCAAGGAACTTCACTTT; Il33 sh2: GCCAATAGAATGGGATCTCAT; Nfe2l2 (Nrf2) sh1: CCAAAGCTAGTATAGCAATAA; Nfe2l2 (Nrf2) sh2: GCCTTACTCTCCCAGTGAATA; Keap1 sh1: GCCCAATTCATGGCTCACAAA; Keap1 sh2: GGATGATCACACCGATGAATA; Rela (p65) sh: AGAAGACATTGAGGTGTATTT; Fcer1a sh: CCCACCATGGATTAGAATATT; Tgfb1 sh: GCTGCGCTTGCAGAGATTAAA; and Il1rl1 (ST2) sh: GCTGCAATATCCCTGATTATT.

Protein gel electrophoresis and Western blotting

Total cell lysates were prepared using radio-immunoprecipitation assay buffer (20 mM Tris-HCl, pH 8.0, 150 mM NaCl, 1 mM EDTA, 1 mM EGTA, 1% Triton X-100, 0.5% deoxycholate, 0.1% SDS) supplemented with protease inhibitors (Halt Protease Inhibitor, Thermo Scientific, 78430), or used directly with 1X Bolt LDS Sample Buffer (Invitrogen, B0007). The protein concentrations of clarified supernatants were measured with the BCA Protein Assay Kit (Pierce, 23227). Proteins in lysates were separated by Bolt 4 to 12% Bis-Tris Plus gels (Invitrogen) and transferred to a PVDF membrane (Millipore, Immobilon-FL, IPFL00010) using a Pierce Power Blotter XL System (Invitrogen, PB0013). Transferred membranes were blocked for 1 hour in Odyssey blocking buffer (LI-COR, 927-50000) diluted 1:1 in Tris-buffered saline (TBS, containing 20 mM Tris-HCl, pH 7.4, 150 mM NaCl), and then incubated with primary antibodies in the blocking buffer or TBS overnight at 4°C. After washing with TBS, membranes were incubated with secondary antibodies in Odyssey blocking buffer diluted 1:1 in TBS for 1 hour at room temperature. Membranes were washed in TBS and imaged on an Azure c600 Infra-Red Imaging System (Azure Biosystems). Primary antibodies used were as follows: anti–IL-33 (1:500, goat polyclonal, R&D Systems, AF3626), anti–phospho-SMAD2 (1:2000, rabbit monoclonal, Cell Signaling Technology, 3108), anti-SMAD2/3 (1:2000, mouse monoclonal, BD Biosciences, 610842), anti-NRF2 (1:3000, rabbit polyclonal, MBL International, PM069), and anti–α-tubulin (1:3000, mouse monoclonal, Sigma-Aldrich, DM1A). IRDye680- or IRDye800-conjugated secondary antibodies were used (1:10,000, LI-COR).

Data and materials availability:

Raw gene expression data and processed counts are available on GEO, accession no. GSE151783.