Live-Cell Invasive Phenotyping Uncovers ALK2 as a Therapeutic Target in LKB1-Mutant Lung Cancer

Three-dimensional invasion-linked gene expression analysis reveals a therapeutic vulnerability to inhibition of ALK2/BMP6 signaling in LKB1-mutant lung cancer that can be rapidly translated to the clinic.

Abstract

The acquisition of invasive properties is a prerequisite for tumor progression and metastasis. Molecular subtypes of KRAS-driven lung cancer exhibit distinct modes of invasion that contribute to unique growth properties and therapeutic susceptibilities. Despite this, preclinical strategies designed to exploit growth within the context of invasion are lacking. To address this, we designed an experimental system to screen for targetable signaling pathways linked to active early 3D invasion phenotypes in different molecular subtypes of KRAS-driven lung adenocarcinoma. Combined live-cell imaging of human bronchial epithelial cells in a 3D invasion matrix and transcriptomic profiling identified mutant LKB1-specific upregulation of BMP6. LKB1 loss increased BMP6 signaling, which induced the canonical iron regulatory hormone hepcidin. Intact LKB1 was necessary to maintain BMP6 signaling homeostasis and restrict ALK2/BMP6-fueled growth. Preclinical studies in a Kras/Lkb1-mutant syngeneic mouse model and in a xenograft model showed potent growth suppression by inhibiting the ALK2/BMP6 signaling axis with single-agent inhibitors that are currently in clinical trials. Lastly, BMP6 expression was elevated in tumors of patients with LKB1-mutant early-stage lung cancer. These results are consistent with those of a model in which LKB1 acts as a “brake” to iron-regulated growth and suggest that ALK2 inhibition can be used for patients with LKB1-mutant tumors.

Significance: Three-dimensional invasion-linked gene expression analysis reveals a therapeutic vulnerability to inhibition of ALK2/BMP6 signaling in LKB1-mutant lung cancer that can be rapidly translated to the clinic.

Graphical Abstract

Introduction

Tremendous progress has been made toward treating KRAS-mutant cancers, including the discovery of both direct and indirect targeting strategies (1). Despite these advances, objective response rates for currently approved targeted therapies remain low. KRAS-driven lung cancers are particularly challenging due to the presence of heterogeneous tumor-suppressor landscapes that confer unique biology and compensatory pro-growth and -survival pathways (2). Variable responses and intrinsic resistance to direct KRASG12C inhibition (3, 4) indicate the necessity of incorporating co-occurring tumor suppressor mutations into clinically relevant screening approaches.

Comparative gene expression analysis has been used successfully to discover novel biological aspects of heterogeneous KRAS-mutant lung cancers, including those carrying either TP53 or LKB1 mutations (5, 6). Integrative approaches using mouse and human tumor cell lines have led to discoveries for novel targeted therapeutic vulnerabilities, particularly for patients with LKB1 mutations who are largely resistant to standard-of-care treatments and immunotherapies (7). Efforts to target pro-growth kinase cascades have been met with mixed results (8). One possible explanation for this is the lack of 3D cell culture models in preclinical approaches. 3D models more accurately mimic the 3D physiology of lung tissue (9), hence promoting the prioritization of signaling molecules driving both cell growth and proliferation and invasion.

In order to identify pro-growth targets that function within a social cell biological context, we took a novel approach that relied upon 3D live-cell phenotyping coupled with comparative gene expression analysis of engineered genetic subsets of invasive human bronchial epithelial cells (HBEC). Using this approach, we identified BMP6 signaling as uniquely upregulated in partially transformed and invasive KRAS/LKB1-mutant spheroids. To show clinical relevance, we probed gene expression data from patients with early-stage lung cancer to show that BMP6 expression is highly elevated in patients with LKB1-mutant lung cancer independent of driver mutation. We coupled this approach with preclinical studies in a novel Kras/Lkb1-mutant syngeneic mouse model and a xenograft model to show that single-agent growth suppression was achieved by inhibiting the ALK2/BMP6 signaling axis with two agents that are currently in clinical trials. Thus, image-guided 3D phenotyping is a powerful new tool to advance clinically relevant targeted treatment strategies in lung cancer.

Materials and Methods

Reagents and Antibodies

LKB1 (#3050 for human, #3047 for mouse), p53 (rodent-specific, #32532), RASG12D (#14429), E-cadherin (#3195), vimentin (#3932), p-mad1/5/9 (#13820), ID-3 (#9837), and AKR1C2 (#13035) Abs were purchased from Cell Signaling Technology. BMP6 (ab155963), Smad1/9 (ab108965), ID-1 (ab192303), transferrin receptor (ab84036), SCD1 (ab236868), ferritin (ab75973), epithelial cell adhesion molecule (EpCAM; ab71916), and TTF1 (ab76013) Abs were from Abcam. KRAS (sc-30), p53 (sc-126 for human), and α-tubulin (sc-8035) Abs were from Santa Cruz Biotechnology. Hepcidin antimicrobial peptide Ab (NBP1-59337) was from Novus Biologicals. Actin (A2066) was from Sigma-Aldrich. Horseradish peroxidase–conjugated secondary Abs (Jackson ImmunoResearch) were used for Western blotting.

Neutralizing BMP6 Abs (MAB507 for human, MAB6325 for mouse) were purchased from R&D Systems. LDN214117 (S7627) for in vitro and in vivo assays was from Selleckchem, and Cell Counting Kit-8 (CK04-11) was from Dojindo Laboratories.

Cell culture conditions and viral transductions

The HBEC3-KT (RRID: CVCL_X491) cell line was kindly provided by Dr. John Minna (The University of Texas Southwestern Medical Center, Dallas, TX). HBEC3-KT cells were cultured with keratinocyte serum-free medium (K-SFM; Life Technologies) containing 50 µg/mL of bovine pituitary extract (BPE; Life Technologies) and 5 ng/mL EGF (Life Technologies) as described previously. Custom-designed lentiviral shRNA-LKB1 (target seq; 5′-GAC​AAC​ATC​TAC​AAG​TTG​T-3′) and shRNA-p53 (target seq; 5′-ACA​TTC​TCC​ACT​TCT​TGT​T-3′) vectors expressing GFP plasmids were purchased from ATCGbio Life Technology Inc. pLenti6-KRAS^G12D lentiviral vector was a gift from Dr. John Minna. Large-scale production of high-titer lentivirus was generated using Lenti-X 293T cells (TaKaRa, #632180; RRID: CVCL_0063) according to an established protocol. Transduced cells were selected with hygromycin B (20 μg/mL) and blasticidin (2 μg/mL). Stable LKB1 or p53 knockdown and moderate KRAS^G12D expression by lentivirus was introduced stepwise, as described in Fig. 1A, and confirmed by Western blotting. BMP6 shRNA (TRCN0000432077) was purchased from Sigma-Aldrich. Stable knockdown BMP6 JK43-P and JK43-M cells were selected with 5 µg/mL puromycin. H157 (RRID: CVCL_0463), A549 (RRID: CVCL_0023), and H1299 (RRID: CVCL_0060) human non–small cell lung cancer (NSCLC) cells were obtained from ATCC. H157 cells stably expressing LKB1 [wild-type LKB1 (LKB1-WT)] and the kinase-dead LKB1 domain (LKB1-K78I) were generated as described previously (10). Stable knockdown LKB1 H1299 cells were generated by pLKO.1-shRNA LKB1. Transfected cells were selected with 2 μg/mL puromycin. LKB1 deletion was verified by Western blotting. JK43-P and JK43-M KL-GEMM cells were cultured in RPMI-1640 media or DMEM with 10% FBS and 100 U/mL of penicillin and streptomycin and maintained at 37°C and 5% CO₂. H157, A549, H1299, JK43-P, and JK43-M cells were used for experiments within 25 passages. HBEC-3KT and isogenic derivatives were used for experiments within 15 passages. All cell lines were tested for Mycoplasma using a MycoAlert kit (Lonza).

Figure 1.

Generation and characterization of in vitro transformation phenotypes and 3D invasive phenotyping. A, Schematic depicting knockdown and overexpression strategy to generate isogenic subsets of HBECs for invasive phenotyping. Scale bar, 200 μm. B, Western blot to verify tumor suppressor, oncogene status, and epithelial–mesenchymal transition markers in isogenic HBECs. E-cad, E-cadherin. C, Graph depicting quantitative analysis of cell-cycle distribution from each isogenic HBEC genotype. D, Graph depicting the mean cell growth rate from isogenic HBECs over the course of 72 hours. Error bar represents the ± SEM of data obtained from quadruplicate samples. E, Schematic representation of 3D invasive phenotyping assay. F, Brightfield and corresponding fluorescent images of live-cell 3D phenotype of isogenic HBECs 4 days post spheroid formation. Scale bar, 200 μm. G, Representative still images of focal point time-lapse microscopy (36 hours post matrix embedding) show protruding cell clusters at the leading branching edge, followed by budding formation or ductal elongation (red line). Scale bar, 16 μm. ***, P < 0.001; ****, P < 0.0001.

Western blot analysis

Three-dimensional cell lysates were extracted from the 3D spheroids after 3D invasion assay in ultra-low attachment 96-well plates. Culture media on the top of the matrix was removed and washed in cold PBS, and the matrix was depolymerized with 200 μL of Cultrex organoid harvesting solution (R&D Systems) at 4°C for 2 hours. Then the isolated cell spheroids were lysed with 1% NP-40 cell lysis buffer containing protease and phosphatase inhibitor cocktail (Thermo Fisher Scientific, #78442) and by microtip sonication. For detecting BMP6 and hepcidin from mouse tumor tissues, isolated subcutaneous tumors were minced and lysed in RIPA buffer containing protease and phosphatase inhibitor cocktail by mechanical homogenization. Samples were centrifuged at 12,000 rpm for 10 minutes at 4°C, and then the supernatant (lysate) was transferred to a new tube. The Bradford protein assay (Bio-Rad Laboratories Inc.) was used to measure the concentration of total protein in samples.

Cell-cycle analysis

Cells were harvested and fixed in 70% cold ethanol at −20°C overnight. Cells were washed in PBS and stained with DAPI (4 μg/mL DAPI, 0.25% Triton-X 100 in 1× PBS). DNA content was measured by flow cytometry with FACSCalibur (BD Biosciences), and data were analyzed to determine the distribution of cells with sub-G1, G1/G0, S, and G2/M peaks using FlowJo (RRID: SCR_008520) software.

Cell growth and viability assays

For examining the cell proliferation rate of the transformed HBEC3-KT cells, the cells were seeded at a density of 4,000 cells/well in 96-well plates in four replicates, and the cell growth rate was determined using Cell Counting Kit-8 (Dojindo Laboratories) at the indicated time point according to the manufacturer’s instructions. For checking cell viability after treating small molecules, with LDN214117, 3,000 cells were seeded in 96-well culture plates in four replicates and treated the next day with the given concentration of LDN214117. Viable cell numbers were determined using a sulforhodamine B assay.

2D cell migration and invasion assays

For 2D cell migration assay, 1 × 10⁴ cells were seeded in growth factor–free KSFM medium in the top transwell chamber (#353097, Corning Inc.) and allowed to migrate along a concentration gradient through a polycarbonate membrane with 8-µm pores to the bottom chamber containing KSFM medium with 100 μg/mL of BPE, 10 ng/mL of EGF, and 10% of FBS. After 24 hours, cells were fixed in 4% formalin and stained with crystal violet. The stained cells were counted from the four randomly chosen fields on the bottom surface of the membrane under 20× magnification of a IX-51 microscope. For 2D cell invasion assay, the transwell membranes were coated with Growth Factor Reduced Matrigel (#356231, Corning Inc.) and incubated for 30 minutes at 37°C, and 1.5 × 10⁵ cells were seeded in growth factor–free KSFM medium in the top chamber. Cells were allowed to migrate for 48 hours across Matrigel-coated membranes to the bottom chamber in KSFM medium containing 100 μg/mL of BPE, 10 ng/mL of EGF, and 10% of FBS. Cells were fixed and stained with crystal violet. Each sample was triplicated.

3D spheroid invasion assays and live-cell imaging

For the transformed HBEC3-KT cell spheroid invasion assay, 6,000 cells in 100 μL of K-SFM medium were seeded in a 96-well clear round bottom ultra-low attachment microplate (#7007, Corning Inc.) and centrifuged at 200 × g for 3 minutes in a swinging bucket rotor. When cells assembled loose-aggregated spheroid at 37°C incubator in 4 to 5 days after seeding, 50 µL of the medium was removed carefully. Matrigel (#356237, Corning Inc.) and Rat Tail Type I Collagen (#354249, Corning Inc.) mixed invasion matrix was embedded directly into the well on ice and then centrifuged at 300 × g for 5 minutes in a swinging bucket rotor. A measure of 100 µL of K-SFM medium was added on the top of the matrix after gel polymerization at 37°C in an incubator for 1 hour and then allowed to invade. Images were taken using an Olympus IX51 microscope 10× (0.30 NA air) with an Infinity2 CCD camera. Live-cell imaging was performed using the PerkinElmer UltraView system. For human NSCLC cells and KL-GEMM cells, spheroids were formed in ultra-low attachment 96-well round bottom plates and embedded in the invasion matrix in a 35-mm glass bottom dish (Cellvis) and allowed to invade for up to 72 hour while being incubated at 37°C. Images were taken using an Olympus IX51 microscope 10×. For neutralizing Ab treatment, Abs were pretreated in the medium 48 hours before embedding the invasion matrix and added in the media on the top of the matrix. For small-molecule treatment, LDN214117 was added directly to the invasion matrix during the embedding process, as well as to the growth media added on the top of the matrix. Invasive area was quantified by measuring the total spheroid area around the outer perimeter in ImageJ/Fiji.

3D spheroid immunofluorescence

Preparation of spheroids and immunofluorescence staining were performed as described previously (11). The spheroids were stained with rabbit anti-pYFAK397 (ab39967, Abcam), rabbit anti-vimentin (3932S, Cell Signaling Technology), and DAPI (D1306, Invitrogen). Secondary Abs were purchased from Thermo Fisher Scientific (anti-rabbit Alexa Fluor 647 and anti-rabbit Alexa Fluor 555 conjugate). After primary and secondary Ab staining, spheroids were imaged with the Leica TCS SP8 inverted confocal microscope (10×) using 1.0-mm z-stack intervals, line scanning (405 nm, 488 nm, 561 nm, 633 nm lasers), and photomultiplier tube (PMT) detectors.

RNA-seq from 3D spheroids and gene expression analysis

RNA sequencing (RNA-seq) was performed in triplicate on HBECK3-KT C, L, P, K, KL, and KP. Total RNA was extracted from the spheroids at day 7 of the 3D spheroid invasion assay using RNeasy Kit (Qiagen). The quantitation, integrity, and purity of the extracted total RNA samples were assessed by the Emory Integrated Genomics Core using the 2100 Bioanalyzer (Agilent), and RNA-seq was performed by Novogene, Co., Ltd. Data processing, quality control (QC), read alignment, and statistical analyses were performed by the Emory Biostatistics and Bioinformatics Shared Resource, as previously described (11). Raw read data (fastq) QC was performed using FastQC (RRID: SCR_014583) v0.11.7. Post-filtered reads were mapped against Ensemble Human GRCh38.p12 release 95 reference genome using the STARaligner v2.7.0e. Expression quantification was obtained using featureCounts (RRID: SCR_012919). Normalization and pairwise differential analyses were determined using the median of ratios method in DESeq2 (RRID: SCR_000154) and log₂-transformed. Differentially expressed genes (DEG) were identified using a moderated t test as implemented in Limma (RRID: SCR_010943). Heatmaps were created by unsupervised clustering of log₂-transformed normalized expression data for the significant DEGs, and the resulting data were analyzed using Gene Ontology enrichment analysis and Kyoto Encyclopedia of Genes and Genomes (RRID: SCR_012773).

RNA extraction and RT-PCR

Total cellular RNA was extracted from cell monolayers with RNeasy Kit following the manufacturer’s instructions, and then cDNA was synthesized by iScript Select cDNA Synthesis Kit (Bio-Rad). RT-PCR was performed with iTaq Universal SYBR Green Supermix (Bio-Rad) on 7500 Fast Real-Time PCR System (Applied Biosystems) following the manufacturer’s instruction. The primer sequences are as follows:

• BMP6 (human)

  • forward 5′-CGT​GAA​GGC​AAT​GCT​CAC​CT-3′,

  • reverse 5′- CCT​GTG​GCG​TGG​TAT​GCT​GT-3′,

• BMP6 (murine)

  • forward 5′- AAG​ACC​CGG​TGG​TGG​CTC​TA-3′,

  • reverse 5′-CTG​TGT​GAG​CTG​CCC​TTG​CT-3′,

• Hepcidin (human)

  • forward 5′-GAC​GGG​ACA​ACT​TGC​AGA​GC-3′,

  • reverse 5′-GCC​TCT​GGA​ACA​TGG​GCA-3′,

• Hepcidin (murine)

  • forward 5′-AAC​AGA​TAC​CAC​ACT​GGG​AA-3′,

  • reverse 5′-CCT​ATC​TCC​ATC​AAC​AGA​TG-3′,

• β-Actin (human)

  • forward 5′-ACG​GTG​AAG​GTG​ACA​GCA​GTC​G-3′,

  • reverse 5′-AAT​GTG​CAA​TCA​AAG​TCC​TCG​GC-3′,

• β-Actin (murine)

  • forward 5′-CCG​TGA​AAA​GAT​GAC​CCA​GA-3′,

  • reverse 5′-AGG​CAT​ACA​GGG​ACA​GCA​CA-3′

Generation of primary and metastatic lung tumor cell lines from KL-GEMM

A KRAS^G12D and LKB1^fl/fl genetically engineered mouse model (KL_luc-GEMM) was generated as previously described (12). Lung tumors were induced by intratracheal intubation of LV-CMV-Cre-GFP (SL100277, SignaGen Laboratories). From 5 weeks after lentiviral-Cre infection, tumor development was monitored and imaged on an IVIS Spectrum imager (PerkinElmer) once a week until the onset of clinical symptoms (hunched posture, tachypnea, and weight loss more than 20%), and the mice were euthanized according to Institutional Animal Care and Use Committee guidelines.

Lung tumor nodules and metastatic mediastinal lymph nodes were isolated from the late stage of male KL-mouse (JK-43) after perfusing via the right atrium with 10 mL of ice-cold PBS using a 10-mL syringe fitted with a 21G needle. The isolated tumor tissues were minced into small pieces and further incubated in collagenase/dispase (#07449, STEMCELL Technologies) and elastase (#07453, STEMCELL Technologies) solution at 37°C, 5% CO₂ incubation for 16 hours. One mg/mL of DNase I (#11284932001, Sigma-Aldrich) was added to cell suspensions for 15 minutes at room temperature (15–25°C), and then the cells were was filtered through sterile cell strainers with pore sizes of 70 and 40 µm. The final filtrate was centrifuged at 300 × g for 3 minutes at 4°C, and the resulting pellet was suspended in 10 mL of complete culture medium (DMEM with 10% FBS and 100 U/mL of penicillin and streptomycin). The fibroblasts were removed using FibrOut Fibroblast Growth Inhibitors (#4-21501, CHI Scientific) at 37°C, 5% CO₂ incubation for 3 days. The EpCAM-positive cell selection was performed by magnetic-activated cell sorting using mouse CD326 (EpCAM) MicroBeads (#130-105-958, Miltenyi Biotec). The KL-primary lung tumor EpCAM⁺ cells (JK-43P) and KL-metastatic lung tumor EpCAM⁺ cells (JK-43M) were cultured in complete culture medium and maintained at 37°C and 5% CO₂. EpCAM⁺, TTF1, KRAS^G12D, and LKB1 expression levels were confirmed by Western blotting. The lung and metastatic draining lymph nodes from JK-43 mouse were stained with hematoxylin and eosin and scored by a board-certified lung pathologist using criteria previously published in ref. 13.

Allograft animal experiment

KL-GEMM–derived allograft experiments were approved by the Institutional Animal Care and Use Committee of Emory University. Seven-week-old female FVB/NJ mice (#001800; RRID: IMSR_JAX:001800) were ordered from Jackson Laboratory. JK-43M cells (2.5 × 10⁶ cells in 30% Growth Factor Reduced Matrigel in 100 μL of PBS) were injected subcutaneously in the right flank of each mouse. Mice (n 5/group) received the following treatment: vehicle control and LDN214117 formulated in 2% DMSO + 30% PEG400 + 2% Tween 80 + ddH2O (25 mg/kg, i.p. once daily) for 22 days. Tumor size and body weight were measured twice a week.

IHC

Formalin-fixed paraffin-embedded mouse tumor sections were stained with rabbit anti-BMP6, rabbit anti-hepcidin, and rabbit anti-transferrin receptor Abs. Secondary Ab (MP-7401-50) and horseradish peroxidase–based 3,3'-diaminobenzidine (DAB) substrate (SK-4105) kits were purchased from Vector Laboratories and were used followed the manufacturer’s instructions.

Statistical analysis

A two-tailed unpaired Student t test was used to analyze statistical significance between two conditions in an experiment. For experiments with three or more comparisons, an ordinary one-way ANOVA with a Tukey multiple comparisons test was used. Significance was assigned to P values < 0.05; *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001. Error bars represent the mean ± SEM.

Data availability

The data generated in this study have been deposited in the NCBI’s Gene Expression Omnibus and are accessible through Gene Expression Omnibus series accession number GSE271368. All other raw data are available upon request from the corresponding author. The data analyzed in this study [Lung Adenocarcinoma (LUAD), The Cancer Genome Atlas, PanCancer Atlas] were obtained from cBioPortal at cbioportal.org.

Results

Generation and characterization of a panel of subtype-specific human bronchial cells for invasive phenotyping

To generate a series of isogenic cell lines that model clinically relevant early invasion events, immortalized HBEC3-KT cells were stably depleted for LKB1 (L) and TP53 (P) both alone and in a mutant KRAS^G12D background (Fig. 1A). These isogenic lines include control (C), KRAS^G12D (K), KRAS^G12D/shRNA-LKB1 (KL), and KRAS^G12D/shRNA-TP53 (KP) lines, which were validated by Western blotting (Fig. 1B). To begin to evaluate in vitro transformation phenotypes related to invasion, we assessed the expression of epithelial–mesenchymal transition markers, E-cadherin, and vimentin. E-cadherin was greatly reduced in KL and KP cells, and vimentin was increased. Analysis of cell-cycle phasing showed an increase in the percentage of cells in the G2/M phase, with the highest percentage in KL and KP cells (Fig. 1C). K, KL, and KP lines showed an increase in cell growth rates (Fig. 1D). To characterize 2D migration and invasion phenotypes, transwell assays with and without the invasion matrix were performed. Similar to the cell growth and cell-cycle assays above, KL and KP lines had significantly increased 2D migration and invasion compared with control cells (C, L, P, and K; Supplementry Fig. S1A–S1D). These data suggest that KL and KP cells are partially transformed in vitro and can be used to screen for targetable growth pathways linked to early invasion.

KL and KP isogenic HBEC3-KT spheroids exhibit distinct autonomous invasion phenotypes

To establish a 3D platform for invasive phenotyping, we created a mixture of Matrigel and collagen I that supported the 3D morphogenesis and invasion of partially transformed HBEC3-KT cells without the need for feeder fibroblasts (Fig. 1E). Initially, the K, KL, and KP cells formed partially aggregated loose spheroids; however, over the course of several days, they developed budding and branching morphologies consistent with lung morphogenesis (Fig. 1F; Supplementry Fig. S1E). We continued to monitor the behavior of spheroids over several days by live-cell imaging. KP spheroids contained unique lobular structures that continued to grow, spin, and display single invading cells into the surrounding matrix, whereas KL spheroids exhibited strong collectively invading cellular strands (Fig. 1G; Supplementry Fig. S2; Supplementry Movie File S1). K spheroids continued to exhibit branching morphogenesis but no invasion, and C, L, and P spheroids remained in loose (C) or tightly associated forms (L and P). Taken together, these results confirm that our subtype-specific spheroids exhibit unique and cell-autonomous growth and invasion phenotypes in our 3D platform.

Transcriptomics of subtype-specific invasion reveals upregulation of BMP6 in LKB1-mutant LUAD

To investigate differential gene expression in spheroids during early invasion, we first confirmed the subtype-specific genotypes in 3D by Western blotting (Fig. 2A). Next, we isolated RNA from invasive spheroids 7 days after embedding in the 3D matrix, the point at which K, KL, and KP spheroids exhibited either morphogenesis or distinct invasion phenotypes. RNA was subject to RT-PCR genotyping of the KRAS allele and QC analysis (Supplementry Fig. S3A–S3D).

Figure 2.

BMP6 expression is uniquely upregulated in response to loss of LKB1 in invasive HBECs and patients with LUAD. A, Western blot analysis of indicated proteins in a 3D invasive spheroid panel. B, Volcano plot depicting significant DEGs (upregulated, N = 583; downregulated, N = 401) between invasive KRAS/LKB1 (KL) vs. KRAS/TP53 (KP) 3D spheroids with respect to noninvasive control (C) HBEC spheroids. C, Heatmap depicting the mean log₂-transformed expression levels of selected differentially upregulated genes (KL > KP and KL > K > C) between isogenic 3D spheroids. D, Graph depicting fold change in BMP6 gene expression from qRT-PCR validation in isogenic HBECs. E, Confocal images of immunofluorescence for BMP6 protein expression (red) in 3D HBECs of the indicated genotypes (DAPI labels nuclei, and all cells express cytoplasmic GFP). Scale bar, 100 μm. F, Graph generated using cBioPortal depicting the mean BMP6 mRNA expression from genetic subtypes of patients with LUAD. Each circle represents an individual patient sample. Error bars, SD. G, Western blot analysis of indicated proteins in isogenic 3D spheroids. H, Graph of pathway enrichment analysis sorted by significance. ****, P < 0.0001. RESM, RNA-seq by expectation maximization; TCGA, The Cancer Genome Atlas.

RNA-seq was performed, and DEGs were identified between KL and KP with respect to the C line. There was a total of 984 significant DEGs (up = 583; down = 401) from the KL versus KP comparison (Fig. 2B). We further narrowed our candidates to 32 targetable hits that were significantly and uniquely upregulated in the KL versus KP (in comparison with K; Fig. 2C; Supplementry Table S1). We validated BMP6 upregulation by qRT-PCR, immunofluorescence, and Western blotting analysis in our KL HBECs (Fig. 2D and E; Supplementry Fig. S3G). To show that BMP6 is clinically relevant in LKB1-mutant LUAD we assayed expression from The Cancer Genome Atlas patient data in genetic subtypes of human LUAD. Our analysis revealed that expression of BMP6 is upregulated in KL versus KP spheroids and in LKB1 mutant LUAD when compared with other molecular subtypes (Fig. 2F; Supplementry Fig. S3E). In addition, BMP6 expression is upregulated in KRAS/LKB1-mutant LUAD across the three major KRAS alleles (G12C, G12D, and G12V; Supplementry Fig. S3F). BMP6 is a ligand of the (TGFβ superfamily and plays an important role in hepcidin expression and iron metabolism, an important mediator of aggressive and treatment-resistant lung cancer (14, 15). To test autonomous activation of BMP6 signaling, we tested hepcidin protein expression, which was greatly increased in KL 3D lysates and increased to a lesser extent in 2D lysates, although we were able to detect hepcidin transcript upregulation in 2D lysates (Fig. 2G; Supplementry Fig. S3H). Pathway enrichment analysis of DEGs identified altered BMP receptors, Jak-STAT signaling, and the known LKB1-regulated focal adhesion pathway (Fig. 2H). Taken together, our data show that 3D invasive phenotyping coupled with transcriptomics can be used to identify clinically relevant targets in LKB1-mutant cancers and that expression of BMP6 and its downstream target hepcidin (gene HAMP) is upregulated in KL mutated cells.

LKB1 restricts BMP6 expression and BMP type I receptor signaling using a kinase-dependent mechanism

BMP6 ligand activates activin A receptor type I (ALK2) and is a critical regulator of hepcidin expression and iron-regulated metabolism in normal and cancerous tissues (16, 17). Despite this knowledge, the link between LKB1, iron homeostasis, and BMP6 signaling in cancer cells is unexplored. We used shLKB1 H1299 cells (18) to verify upregulation of BMP6 ligand, Smad signaling, and p38 activation in response to LKB1 downregulation in lung cancer cells (Fig. 3A). Next, we restored LKB1-WT in LKB1-deficient H157 NSCLC cells. BMP6-Smad1/5/8 pathway-mediated gene targets, including ID-1, ID-3, and BMP6 itself, were decreased by exogenously restored LKB1-WT (Fig. 3B). We expressed kinase-dead LKB1 (LKB1-K78I) in H157 cells, which was unable to inhibit BMP6-Smad1/5/8 signaling. These data suggest that LKB1 negatively regulates the BMP6-Smad1/5/8-regulated pathway using a kinase-dependent mechanism.

Figure 3.

LKB1 restricts BMP6 signaling using a kinase-dependent mechanism, and suppression of BMP6 signaling and ALK2 inhibition suppresses 3D proliferation and invasion in multiple KL cell lines. A, Western blot analysis of BMP6 pathway activation in stable control pLKO.1 and shLKB1 H1299 lung cancer cells. B, Western blot of BMP6-regulated Smad signaling components in LKB1-null H157 cells that express vector control, LKB1-WT, or kinase-dead LKB1 (LKB1-K78I). C, Western blot analysis of indicated proteins in control IgG (−)- or anti-BMP6 (+)–treated KL, JK43-P, and JK43-M cells. D, Representative brightfield images (top) and quantitative graphs (bottom) of control IgG-treated or anti-BMP6–treated invasive KL, JK43-P, or JK43-M spheroids. E, Graph depicting cell viability of indicated cell lines with increasing concentrations of LDN214117 (top). Table of LDN214117 IC₅₀ in indicated cell lines. F, Representative images (left) and quantitative graphs (right) of A549 3D spheroids assayed for Ki67. Scale bar, 70 μm. G, Representative images (left) and quantitative graphs (right) of JK43-M 3D spheroids assayed for Ki67. Scale bar, 70 μm. H, Brightfield images of 3D spheroids of the indicated cell lines either treated with vehicle control (−) or treated with the indicated concentration of LDN214117 and embedded in the invasion matrix for 72 hours. Scale bar, 100 μm. I, Quantitation of the invasive area for LDN214117-treated KL spheroids (the graph depicts the mean of three biological replicates). J, Quantitation of the invasive area for LDN214117-treated A549 spheroids (the graph depicts the mean of three biological replicates). K, Western blot of BMP6/ALK2-regulated Smad signaling in KL HBECs and A549 (LKB1-null) cells treated with increasing concentrations of LDN214117 for 24 hours. L, Western blot analysis of BMP6/ALK2-regulated Smad signaling in JK43-P and JK43-M mouse tumor cell lines (KrasG12D/Lkb1-null) treated with increasing concentrations of LDN214117 for 24 hours. **, P < 0.01; ***, P < 0.001; ****, P < 0.0001.

Inhibiting BMP6/ALK2 receptor signaling decreases 3D invasion and proliferation in multiple LKB1-mutant cell models

BMP6 has a paradoxical role in cancer and has been described as both a tumor suppressor and a promoter of invasion and metastasis (19). To explore the functional role of BMP6 in LKB1-mutant lung cancer cells, we created two mouse tumor cell lines from the Kras^G12D/Lkb1^fl/fl genetically engineered mouse model (KL GEMM). JK43-P and JK43-M were generated from a primary lung tumor and a tumor-positive mediastinal draining lymph node, respectively (Supplementry Fig. S4A). The status of oncogene and tumor suppressor expression and epithelial identity was confirmed by Western blotting (Supplementry S4B). We treated KL HBECs and JK43-P and JK43-M cells with a neutralizing anti-BMP6 Ab, which resulted in decreased Smad1/5/8 activation, decreased target gene expression, and suppression of 3D invasion compared with IgG control Ab across all cell lines (Fig. 3C and D).

ALK2 is a BMP type I receptor that is a known oncogenic driver in 25% of diffuse pediatric intrinsic pontine glioma tumors, but its role in lung cancer is largely unexplored. Several small molecules target BMP type I receptors, including LDN214117, which is a small-molecule kinase inhibitor that exhibits a high degree of selectivity for BMP6-activated ALK2 (20).

Treating our panel of KL cell lines with LDN214117 and a second ALK2 inhibitor, LDN193189, led to a dose-dependent inhibition of cell viability in 2D and cell proliferation in 3D (Fig. 3E–G; Supplementry Fig. S4C and S4D). Sublethal doses of LDN214117 in 3D culture conditions were confirmed by staining with PI and acridine orange (Supplementry Fig. S4E and S4F). Treatment of A549 and JK-M cells with LDN214117 suppressed 3D invasion and activation of Smad1/5/8 signaling across our panel of KL cell lines (Fig. 3H–L). Lastly, we confirmed dose-dependent inhibition of 3D invasion using the ALK2 inhibitor LDN193189 (Supplementry Fig. S4G and S4H).

ALK2 inhibition inhibits KL tumor growth in vivo that is accompanied by alterations in iron homeostasis

To determine the in vivo efficacy of targeting ALK2 activity in Kras^G12D/Lkb^fl/fl LUAD, we created an immune competent syngeneic mouse model using the JK43-M tumor cell line. JK43-M cells were subcutaneously injected into FVB/NJ mice and treated with 25 mg/kg of LDN214117 (Supplementry Fig. S5A). Tumor volume and weight were significantly inhibited in LDN214117-treated mice compared with the vehicle control (Fig. 4A–C; Supplementry Fig. S5B and S5C). To test whether an intact immune system is required for tumor response, we repeated our experiment using an immunodeficient A549 xenograft model (Supplementry Fig. S5D). In addition, we used the additional ALK2 inhibitor LDN193189. Treatment of A549 xenograft tumors with both ALK2 inhibitors reduced tumor volume and weight (Fig. 4D–F). LDN214117-treated tumors exhibited downregulated BMP6 and hepcidin expression, decreased intracellular iron and cell proliferation, and increased cell death (Fig. 4G and H).

Figure 4.

Efficacy of targeting ALK2 in LKB1-mutant lung cancer in vivo. LKB1 restricts iron homeostasis pathways using a kinase-dependent mechanism. A, Mean tumor volume from syngeneic mice (JK-M cells) treated with vehicle (6 mice/group) and LDN214117 (7 mice/group). B, Graph depicting the mean tumor weight from vehicle- and LDN214117-treated mice. C, Brightfield images of tumors isolated from vehicle- and LDN214117-treated mice. D, The mean tumor volume from NSG mice with A549 lung tumor xenografts treated with either vehicle (7 mice/group), LDN214117 (7 mice/group), or LDN193189 (7 mice/group). E, Graph depicting the mean tumor weight from vehicle-, LDN214117-, and LDN193189-treated mice. F, Brightfield images of A549 tumors isolated from vehicle-, LDN214117-, and LDN193189-treated mice. G, Representative brightfield images of tumor sections treated with vehicle or LDN214117 and stained by the indicated Ab IHC or special stain. Scale bar, 200 μm or 2 mm (TUNEL). H, Western blot to assess BMP6 and hepcidin levels in tumors from vehicle- and LDN214117-treated mice. I, Western blot of indicated proteins from JK43-M cells treated with 1 uM of LDN214117 for 24, 48, and 72 hours. J, Western blot analysis of indicated proteins in vector, LKB1-WT, or LKB1-K78I add-back H157 lung cancer cells. K, Model depicting the mechanism of altered iron homeostasis signaling in LKB1-mutant tumor cells. **, P < 0.01; ****, P < 0.0001.

In cancerous tissues, the expression of excess hepcidin results in accumulation of intracellular iron (Fe2⁺) pools to promote cancer cell growth and proliferation (17). Hepcidin accomplishes this by promoting the degradation of the ferroportin receptor (FPN), which acts to export excess intracellular Fe2⁺. This is accompanied by an increase in the expression of the transferrin receptor (TfR1), which acts to import ferric iron (TF-Fe3⁺) into the cell. We treated JK-M cells in vitro with LDN214117 to show that ALK2i modulates cell-autonomous BMP6 and iron homeostasis pathways in LKB1-mutant tumor cells (Fig. 4I). To test whether LKB1 is sufficient to impact hepcidin-regulated iron homeostasis in lung cancer cells, we expressed LKB1-WT in H157 cells, which resulted in the downregulation of hepcidin, TfR1, and the proteins stearoyl-CoA desaturase 1 (SCD1) and aldo-keto reductase-1C (AKR1C2; Fig. 4J). SCD1 and AKR1C2 are enzymes involved in lipid and steroid metabolism, respectively, that also have a role in protecting cells from ferroptosis. We treated JK-M 3D invasive spheroids with the ferroptosis-inducer MMRi62, which phenocopied the decreased invasion detected with AKL2 inhibitors (Supplementry Fig. S5E and S5F). Lastly, we show that LKB1 modulates iron homeostasis pathways via its kinase activity (Fig. 4J). Taken together, these data support our conclusions that LKB1-mutant lung cancer cells deregulate iron homeostasis to drive growth and invasion (Fig. 4K) and support the future testing of ALK2 inhibitors as a targeted treatment strategy for LKB1-mutant lung cancer.

Discussion

In this study, we used live-cell phenotyping and gene expression profiling in a panel of partially transformed HBECs to identify a clinically relevant targeting strategy for treatment-resistant LKB1-mutant LUAD. We show that LKB1 loss upregulates BMP6 levels in invasive HBECs and patient tumors and that LKB1 restricts the BMP6/hepcidin iron homeostasis pathway through its kinase activity. Furthermore, we created a novel Kras/Lkb1-mutant syngeneic mouse model using a highly metastatic KL tumor with intact p53. Using this model, we show that tumor invasion and growth can be potently inhibited using a BMP6 monoclonal Ab and two TGFβ-family ALK2 kinase inhibitors. Our results suggest partially transformed epithelial cells grown in 3D can be used successfully to screen and prioritize therapeutic strategies that have the potential to be rapidly translated into the clinic.

We show here that LKB1 is required and sufficient to restrict BMP6-regulated Smad signaling in a kinase-dependent manner. This is consistent with a previous report showing that LKB1 can negatively regulate BMP type 1 receptors, including ALK2, in multiple cell lines and in a Drosophila model, including downstream SMAD signaling (21). This report shows a potential protein–protein interaction between LKB1 and ALK2, which could suggest that the LKB1–BMP6 axis characterized here is due to the LKB1 interaction with ALK2 directly and would further explain why KL cells are sensitive to ALK2 inhibitors in vivo and in vitro. Therefore, when synthesizing the findings here with others, one proposed model is one in which LKB1 loss potentiates a ALK2–BMP6 receptor–ligand interaction that drives downstream BMP6 signaling.

ALK2 is a clinically actionable target in several diseases, and small-molecule antagonists are orally bioavailable and well-tolerated in preclinical studies with patient-derived models (22). In diffuse intrinsic pontine glioma, treatment with ALK2 inhibitors LDN-193189 or LDN-214117 extended survival compared with vehicle control (23). Trials have completed or are ongoing using ALK2 small-molecule or biologic inhibitors in different clinical settings (24). Other ALK2 inhibitors, such as saracatinib, has improved selectivity for ALK2 over ALKs 3, 4, 5, supporting the concept of highly selective ALK2 targeting in the clinical setting (25). Based upon our findings, we would propose that the KL genetic subtype of lung cancer would be sensitive to this drug class. In addition, the ability of ALK2 inhibitors to penetrate the blood–brain barrier suggests they may be efficacious in treating primary lung cancers that have metastasized to the brain. Additional preclinical studies to further test this are warranted.

Our data are consistent with a model in which LKB1 can apply the “breaks” to finely tune a cancer cell’s need for iron-regulated metabolic growth with protection from cell death. Although we were able to detect upregulation in TUNEL positivity in ALKi-treated mouse tumors, we saw no evidence of lipid peroxidase activity. In addition, we were able to stop the growth of LKB1-mutant tumors in an immunodeficient mouse model. This suggests that further studies are warranted to test whether ALKi can synergize with ferroptosis inducers or immunotherapies to shrink or eliminate tumors. Our data also show that BMP6 protein levels are sensitive to LKB1 genetic perturbation as well as pharmacologic inhibition with the ALK2 inhibitor. This suggests that changes in BMP6 protein levels may be exploited in the future in patients with KL mutations as a positive predictive biomarker of response to ALK2 inhibition.

Supplementary Material

Supplementary Video 1

Invasive phenotyping of K, KL, and KP HBECs

Supplementary Figure Legends

Legends for Supplementary Figures

Figure S1

Figure S1

Figure S2

Figure S2

Figure S3

Figure S3

Figure S4

Dose response in 2D and 3D

Figure S5

Figure S5

Table S1

List of upregulated transcripts in KL vs. KP invasive spheroids with respect to K>C.

Authors’ Disclosures

S.S. Ramalingam reports grants from Bristol Myers Squibb, AstraZeneca, Merck, Pfizer, and Amgen outside the submitted work. No disclosures were reported by the other authors.

Authors’ Contributions

J. Koo: Formal analysis, investigation, visualization, methodology, writing–original draft. C.-S. Seong: Formal analysis, investigation. R.E. Parker: Data curation, formal analysis, visualization. A. Herrera: Investigation. B. Dwivedi: Data curation, formal analysis. R.A. Arthur: Data curation, formal analysis, visualization. A.R. Dinasarapu: Data curation, formal analysis. H.R. Johnston: Data curation, formal analysis, supervision. H. Claussen: Data curation, formal analysis, visualization. C. Tucker-Burden: Resources. S.S. Ramalingam: Resources, funding acquisition, writing–review and editing. H. Fu: Resources, funding acquisition, writing–review and editing. W. Zhou: Resources, funding acquisition, writing–review and editing. A.I. Marcus: Conceptualization, resources, supervision, funding acquisition, writing–original draft, project administration. M. Gilbert-Ross: Conceptualization, resources, formal analysis, supervision, funding acquisition, validation, visualization, writing–original draft, project administration, writing–review and editing.