SMALL CELL LUNG CANCER PLASTICITY ENABLES NFIB-INDEPENDENT METASTASIS

Abstract

Metastasis is a major cause of morbidity and mortality in patients with cancer, highlighting the need to identify improved treatment and prevention strategies. Previous observations in pre-clinical models and tumors from patients with small cell lung cancer (SCLC), a fatal form of lung cancer with high metastatic potential, identified the transcription factor NFIB as a driver of tumor growth and metastasis. However, investigation into the requirement for NFIB activity for tumor growth and metastasis in relevant in vivo models is needed to establish NFIB as a therapeutic target. Here, using conditional gene knockout strategies in genetically engineered mouse models of SCLC, we found that upregulation of NFIB contributes to tumor progression, but NFIB is not required for metastasis. Molecular studies in NFIB wild-type and knockout tumors identified the pioneer transcription factors FOXA1/2 as candidate drivers of metastatic progression. Thus, while NFIB upregulation is a frequent event in SCLC during tumor progression, SCLC tumors can employ NFIB-independent mechanisms for metastasis, further highlighting the plasticity of these tumors.

Statement of significance

Small cell lung cancer cells overcome deficiency of the pro-metastatic oncogene NFIB to gain metastatic potential through various molecular mechanisms, which may represent targets to block progression of this fatal cancer type.

Introduction

Most cancer deaths result from the growth and spread of metastatic tumors, which impair organ function and disrupt physiological processes. Metastasis is a hallmark of cancer, but the mechanisms involved in this process are not fully understood. The difficulty in studying the metastatic process is in part due to its complexity: To metastasize, cancer cells employ a series of molecular programs that not only vary between different cancer types, but also between different contexts within a cancer type (1-3). Direct comparisons between metastases and their originating primary tumors are also often limited by the paucity of paired primary and metastatic tumor samples that can be obtained from patients (3,4). Developing new therapies to treat metastatic disease requires a better understanding of both universal and cancer type-specific mechanisms underlying metastatic progression.

Small cell lung cancer (SCLC) is a highly metastatic carcinoma associated with over 200,000 deaths per year worldwide (5). Most patients already have widespread metastases at diagnosis, most commonly in the liver, brain, and bone (6). Patients with metastatic disease have a median survival of only ~1 year, and their tumors are rarely surgically resected (7,8). Study of SCLC metastasis is hindered by the limited number of models of SCLC progression. Surgical resection in SCLC is usually limited to early-stage patients without distant metastases, and xenografts derived from resected tumors often do not readily metastasize. Cell lines and xenografts derived from circulating tumor cells (CTCs) or metastatic tumors (9,10) also cannot easily model earlier steps of tumor progression, such as local invasion and intravasation.

Genetically engineered mouse models can address some of these limitations. The tumor suppressors RB1 and TP53 are inactivated in most human SCLC tumors (11), so conditional knockout mouse models of SCLC delete Lox-flanked alleles of Rb1 (Rb1^f/f) and Trp53 (Trp53^f/f) in lung epithelial cells via intratracheal instillation of adenoviral vectors expressing the Cre recombinase (12,13). Other alleles can be added to the Rb1^f/f;Trp53^f/f (RP) mouse model to reflect the contributions of other recurrent alterations in human SCLC. Adding a conditional knockout of the Rbl2 tumor suppressor gene to generate Rb1^f/f;Trp53^f/f;Rbl2^f/f (RPR2) mutant mice accelerates tumor growth (14,15). Tumors in both RP and RPR2 mutant mice are of the SCLC-A subtype, a predominant subtype of SCLC characterized by expression of the transcription factor ASCL1 (Achaete-Scute Family BHLH Transcription Factor 1) (16). The Rb1^f/f;Trp53^f/f;H11^{CAG-LSL-MycT58A} (RPM) mouse model, with conditional expression of the Myc^T58A oncogene, generates tumors of the SCLC-N subtype, another common subtype of SCLC characterized by expression of the transcription factor NEUROD1 (17,18). In these SCLC mouse models, lung primary tumors eventually metastasize to distant organs, particularly the liver and lymph nodes, thereby providing models to investigate tumor progression in vivo.

In the RPR2 model, initiating tumors with Cre recombinase driven by the ubiquitous CMV promoter results in early primary tumors that express low levels of the NFIB transcription factor. As tumors progress, cells upregulate Nfib expression through genomic amplification, expand as NFIB^high clonal populations, and seed metastases (19). NFIB levels are also elevated in late RP mutant mouse tumors and metastases, RPM mutant mouse tumors, and in a majority of human metastatic SCLC, where high levels of NFIB correlate with lower survival (18-23). Overexpression of NFIB at the time of Rb1 and Trp53 deletion results in tumors that are more metastatic than RP mutant tumors or RP mutant tumors overexpressing the MYCL oncogene (22). These data suggest a model in which upregulation of NFIB is sufficient to endow SCLC cells with a greater ability to metastasize. However, metastatic tumors with little to no NFIB expression have been observed in RPR2 mutant mice given Ad-CGRP-Cre to specifically target pulmonary neuroendocrine cells (24), suggesting SCLC tumors are capable of metastasizing without NFIB in at least one context. While loss of NFIB methylation was recently shown to decrease metastases in the RPR2 model (21), only one experimental approach thus far has tested the potential direct requirement for NFIB in metastasis formation: In RPR2 mutant mouse cell lines, knock-down of NFIB decreases the ability of these cells to form liver metastases upon tail vein injection (19).

Here we investigated the consequences of Nfib deletion in RPR2 and RPM mutant mice. In these models, tumors with high NFIB expression show greater dependence on NFIB for growth, validating its role as an oncogenic factor in SCLC in vivo. Notably, while NFIB upregulation is a very frequent event in late tumors and metastases in the RPR2 model, NFIB is not required for the development of metastases. Molecular analyses of the transcriptome and the chromatin state of NFIB wild-type and knockout SCLC cells in the RPR2 model indicate that, when SCLC cells cannot use NFIB upregulation as a metastatic switch, they rely on other mechanisms such as the upregulation of transcriptional networks around ASCL1 and chromatin accessibility around FOXA1/2 binding to promote proliferation and metastasis.

Materials and Methods

Mouse models of SCLC

Mice were maintained according to practices prescribed by the National Institute of Health at Stanford’s Research Animal Facility. Experiments with mice for this work were approved by the Institutional Animal Care and Use Committee (IACUC) at Stanford (APLAC protocol #13565). Additional accreditation of Stanford Research Animal Facility was provided by the Association for Assessment and Accreditation of Laboratory Animal Care.

The Rb1^flox/flox;Trp53^flox/flox;Rbl2^flox/flox (RPR2, or TKO) mouse model and the Rosa26^mTmG (R26^mTmG) allele have been previously described (15,19). RPM mice were provided by Trudy Oliver (18). Nfib^flox/flox (Nfib^f/f thereafter) mice were provided by Richard Gronostajski (25). Mice were maintained in a mixed C57BL6;129SVJ background.

Nfib^f mice were crossed to RPR2;R26^mTmG mice to generate RPR2;R26^mTmG;Nfib^f/+ mice, which were then crossed to each other to generate RPR2;R26^mTmG;Nfib^f/f mice and RPR2;R26^mTmG;Nfib^+/+ littermate controls. Nfib^f mice were similarly crossed to RPM mice to generate RPM;Nfib^f/f mice and RPM;Nfib^+/+ littermate controls.

Tumors were initiated at 8-12 weeks of age with intratracheal instillation of Adeno-CMV-Cre (Baylor College of Medicine, Houston, TX) or Adeno-CGRP-Cre (University of Iowa). RPR2 mice were given 4x10⁷ plaque-forming units (pfu) of Adeno-CMV-Cre or 4x10⁸ pfu of Adeno-CGRP-Cre. RPM mice were given 1x10⁸ pfu of Adeno-CGRP-Cre. Animals were euthanized and tumors were collected at set time points post-initiation or when they showed signs of respiratory distress. Mice were housed at 22°C ambient temperature with 40% humidity and a light/dark cycle of 12 hours (7am – 7pm). Mice of both sexes were used in the experiments in approximately equal numbers.

Histology and immunostaining

Lung and liver lobes were fixed for 48 hours in 10% neutral buffered formalin before paraffin embedding. Paraffin sections were deparaffinized with Histo-Clear (National Diagnostics HS-200) and gradually rehydrated from ethanol to water. Antigen retrieval was carried out in citrate-based unmasking solution (Vector Laboratories H-3300) by microwaving at full power until boiling, then 30% power for 15 min, then left to cool at room temperature for 10 min before washing with water. Endogenous peroxidase was blocked by incubating slides in 3% hydrogen peroxide for 1 h. Sections were washed in PBST (PBS with 0.1% Tween-20) and blocked in 5% horse serum for 1 h. For immunostaining with anti-ASCL1 antibodies, endogenous mouse IgG were blocked by washing sections in PBST, incubating with Goat F(ab) Anti-Mouse IgG H&L (ab6668) diluted in PBS for 1 h, then washing in PBST for 30 min. All sections were incubated with primary antibody diluted in PBST overnight at 4°C.

For immunohistochemistry, sections (except those with anti-HES1 antibodies) were developed using ImmPRESS^®HRP Horse Anti-Rabbit IgG Polymer Detection Kit (Vector Laboratories MP-7401) following the manufacturer’s protocol. Sections with anti-HES1 were developed using ImmPRESS^® Excel Amplified Polymer Staining Anti-Rabbit IgG Peroxidase Kit (Vector Laboratories MP-7601) following the manufacturer’s protocol. All sections used DAB substrate kit (Vector Laboratories SK-4100) for color development. Sections were counterstained with hematoxylin (Sigma-Aldrich HHS32-1L), gradually dehydrated from water to ethanol to xylene, and mounted with Refrax mounting medium (Anatech Ltd711). Primary antibodies were used at the following dilutions: anti-ASCL1 (D-7, Santa Cruz sc-374104, 1:100), anti-CDH1 (E-Cadherin 24E10, CST 3195S, 1:500), anti-GFP (D5.1, CST 2956S, 1:100), anti-HES1 (D6P2U, CST 11988S, 1:200), anti-Ki-67 (SP6, ThermoFisher RM9106-S0, 1:200), anti-NFIB (Abcam ab186738, 1:2000), anti-UCHL1 (Sigma-Aldrich HPA005993, 1:2500). Research resource identifiers (RRIDs) for antibodies are listed in Table S1.

Tumor and immunostaining quantification

Sections were imaged using Keyence BZ-X700 all-in-one fluorescence microscope with BZ-X Viewer program version 1.3.1.1. Quantified sections were imaged at low magnification and stitched together with BZ-X Analyzer 1.4.0.1. Image analysis was conducted with Fiji 1.53t.

Stitched images were deconvolved using the color deconvolution plugin bundled with Fiji (26). Tumors were manually traced on each deconvolved hematoxylin image and saved as ROIs (regions of interest) in Fiji. At the 4.5-month timepoint, we used GFP expression from the R26^mTmG allele on serial sections to identify the ROIs that were tumors and filter out inflammation before quantifying tumor number, tumor burden, or immunostaining.

Tumor area and immunostaining signal were measured by thresholding deconvolved images and measuring only within the tumor regions. Tumor areas were quantified in deconvolved eosin images at a single threshold. H-scores for immunostaining signal were quantified as 3 x % tumor area with high staining intensity + 2 x % tumor area with medium intensity + 1 x % tumor area with weak intensity. The % tumor area with each staining intensity was measured in deconvolved DAB images by setting three different thresholds.

Cell culture

Cell lines were cultured in DMEM (Hyclone SH30243.01) supplemented with 10% bovine growth serum (Hyclone SH3054103HI) and penicillin-streptomycin-glutamine (Gibco 10378016) at 37°C in standard cell culture incubators. All cells were routinely tested (MycoAlert Detection Kit, Lonza) and confirmed to be free of mycoplasma contamination. N2N1G and 12N1G murine SCLC cell lines have been described previously (19).

AlamarBlue cell proliferation assay

Cells were seeded at 8,000 cells per well in a 96-well plate in triplicate with four time points, and AlamarBlue reagent was added to a new time point every two days (with the day 0 timepoint starting immediately after seeding). Plates were incubated at 37°C for two hours after reagent addition, then read using a plate reader.

Cell line generation

Individual tumors (up to ~5 mm in diameter, or a smaller piece from the edge of a large tumor to avoid necrotic cores) were dissected from lungs or livers of RPR2 or RPR2;Nfib^f/f mutant mice. Tumors or tumor pieces were finely chopped with a razor blade and digested for 15 min at 37°C in 3 mL of L-15 medium (Sigma L1518) containing 4.25 mg/mL Collagenase I (Sigma C0130), 1.4 mg/mL Collagenase II (Sigma 6885), 4.25 mg/mL Collagenase IV (Sigma 5138), 0.625 mg/mL Elastase (Worthington LS002292), and 0.625 mg/mL DNase I (Roche 10104159). The digested mixture was filtered through a 40 μm filter and centrifuged at 400 g for 5 min at room temp, and the resulting pellet was resuspended in 500 μL RBC lysis buffer (eBioscience 00-4333-57) for 30 sec, diluted in 10 mL PBS, and centrifuged at 400 g for 5 min at room temp. The pelleted cells were then resuspended in 1.5 mL cell culture media (DMEM with 10% BGS and 1X PSQ) and plated in a single well of a 12-well plate. Tumor cells were left alone until they showed visible signs of growth by forming tight spherical aggregates typical of most murine SCLC cell lines or proliferating as adherent cells. All cell lines grew as spherical aggregates or as a mix of spherical aggregates and adherent cells.

SCLC CellMiner Cross Database

The heatmap of mRNA expression (log2) of MYC and mRNA expression (log2) of NFIB from the SCLC NCI-DTP cell line set was obtained using the SCLC-CellMinerCDB online tool (v1.2) (27).

RNA-seq analysis

Individual tumors were dissected from the lungs and livers of RPR2 or RPR2;Nfib^f/f mutant mice at morbidity, snap-frozen, and sent to Azenta Life Sciences for RNA extraction and sequencing. Samples are listed in Table S1. Total RNA isolation, polyA selection, quality control, library preparation, and sequencing were performed by Azenta using Illumina HiSeq platform (2 x 150bp, ~350M paired-end reads). Transcript quantification was conducted with Salmon (v0.12.0) (28) with mouse genome version GRCm38. DESeq2 (v1.34.0) was used to calculate differential expression (29) (Supplementary Data 1).

Gene set enrichment analysis (30,31) was performed through the GSEAPreranked module (v7.3.4) on GenePattern (32). From the default parameters, permutation type was changed to gene_set, and the metric for ranking genes was changed to log2_Ratio_of_Classes. GSEA was performed on pre-ranked lists of genes that were differentially expressed between metastatic and primary tumors with an adjusted p value < 0.05. The gene set is a custom list of genes classically associated with axonogenesis and axon guidance, as described in (33).

Enriched terms and cell types were identified from the KEGG 2021 Human pathway and the Mouse Gene Atlas in differentially expressed genes using Enrichr (34-36). For the KEGG 2021 Human enrichment, differentially expressed genes were first filtered for those that were not listed as liver genes in the Mouse Gene Atlas.

Low-pass whole genome sequencing analysis

Genomic DNA was extracted from RPR2 and RPR2;Nfib^f/f cell lines using the AllPrep DNA/RNA Mini Kit (Qiagen 80204) following the manufacturer’s protocol. The library was prepared using the Illumina DNA Prep, (M) Tagmentation kit (Illumina 20018704, formerly Nextera DNA Flex) and IDT for Illumina DNA/RNA UD Indexes Set B (Illumina 20027214, formerly IDT for Illumina Nextera DNA Unique Dual Indexes Sets B) following the manufacturer’s protocol. Individual libraries were sent to Novogene for quality control, pooling, and sequencing on a partial NovaSeq 6000 lane (PE150) at 0.8X coverage. Reads were trimmed of adapters using Trimmomatic (v.0.39) (37), aligned to the mm39 genome using BWA-MEM (v.0.7.17) (38), and PCR duplicates removed using Picard’s MarkDuplicates (v.2.26.10). The resulting BAM files were used to call copy number variation using CNVkit (v.0.9.9) (39).

ATAC-Seq library preparation and sequencing

To perform ATAC-seq with mouse SCLC cell lines, ~100,000 suspension cells per cell line were collected, trypsinized, and processed using the OMNI-ATAC-Seq protocol (40). Briefly, cells were washed with 1 ml of ice-cold ATAC-RSB buffer and centrifuged at 500rcf for 5 minutes at 4°C. Next, the pellets were resuspended in 100μl of chilled ATAC-RSB-LYSIS and incubated on ice for 3 minutes. Lysis was stopped by adding 1mL of ATAC-RSB-WASH to each sample, which were again centrifuged for 5 minutes at 500rcf. The supernatant was removed, and the pellet was resuspended in 50μl of OMNI-ATAC Mix and incubated in a mixing (500rpm) thermoblock at 37°C for one hour. Tagmented DNA was purified using a Qiagen MinElute Kit (#28204), with 21μl of elution buffer warmed to 55°C. Library amplification PCR was performed with the NEBNext Ultra II Q5 2X Master Mix (NEB #M0544S) using Nextera primers for 12 cycles. DNA concentration was measured using a Qubit DNA HS Assay (ThermoFisher #Q33230). Individual samples were pooled sequenced on an Illumina NextSeq500 using the 75bp kit in a paired end configuration.

ATAC-Seq sequence processing

Raw sequencing reads were demultiplexed and trimmed using CutAdapt (v2.10) in paired-end mode with the forward and reverse Nextera sequencing adapters (Fwd – CTGTCTCTTATACACATCT, Rev- AGATGTGTATAAGAGACAG) and a minimum read length of at least 25. Next, Bowtie2 (41) was used to align paired-end reads to the ENSEMBL mm10 genome (GRCm38), with the following settings: “--local --very-sensitive-local --no-unal --no-mixed --no-discordant”. PicardTools (https://github.com/broadinstitute/picard) was used to mark duplicates, which were then filtered from the BAM files using samtools (42). MACS2 was used to call peaks genome-wide, with a q value of 0.05 and the arguments “-f BAMPE -g $GENOME_SIZE --nomodel --shift 37 --extsize 73” (43). Downstream analysis, including generation of consensus peakset and determining differential peaks between conditions, was performed using DiffBind (44). Motif enrichment analysis was performed using HOMER.

CUT&RUN library preparation, sequencing, and analysis

CUT&RUN was performed using on mouse SCLC cell lines using the CUTANA ChIC/CUT&RUN Kit (EpiCypher 14-1048) and the CUTANA CUT&RUN Library Prep Kit (EpiCypher 14-1001) according to the manufacturer’s protocols. FOXA1 binding was detected using anti-FOXA1 (Abcam, ab170933). Libraries were sequenced by an Illumina NovaSeq PE150 (100G) platform, with ~20M paired reads per sample.

The demultiplexed CUT&RUN sequencing reads underwent preprocessing steps as follows: In paired-end mode, CutAdapt (v2.10) was used to trim sequences. Trimming involved the removal of TruSeq sequencing adapters from both the forward (5’-AGATCGGAAGAGCACACGTCTGAACTCCAGTCA-3’) and reverse (5’-AGATCGGAAGAGCGTCGTGTAGGGAAAGAGTGT-3’) ends, and reads were retained only if they met a minimum length threshold of 25 bases. Subsequently, alignment to the ENSEMBL mm10 genome was performed using Bowtie2 (41) with specific options, including '–local –very-sensitive-local –no-unal –no-mixed –no-discordant,' and alignments exceeding 1000 base pairs were excluded. To identify duplicate reads, PicardTool (https://github.com/broadinstitute/picard) was utilized, followed by the removal of duplicates from the BAM files using samtools. Genome-wide peak calling was conducted using MACS2, employing a significance threshold (q value) of 0.05, along with the arguments '-f BAMPE -g $GENOME_SIZE -q 0.05 –call-summits.' Subsequent analyses, such as motif enrichment, generation of peak profiles, and quantification of FOXA1 binding, were performed using Homer, Deeptools (45), and the DiffBind (44) packages, respectively.

Quantitative immunoassay

Cells were lysed in RIPA buffer (Thermo Fisher Scientific 89900) supplemented with protease and phosphatase inhibitors (Roche cOmplete ULTRA tablets, Mini, EASYpack 05892970001 and PhosSTOP EASYpack 04906845001). Protein concentration was measured using the Pierce BCA Protein Assay Kit (ThermoFisher #23227). 1 μg of protein was analyzed using the capillary-based Simple Western assay on the Wes system (ProteinSimple) and the Compass software v4.0.0 (ProteinSimple), according to the manufacturer's protocol. Anti-NFIB (Abcam ab186738, 1:10000) and anti-FOXA1 (Abcam ab170933, 1:5000) were run simultaneously with anti-HSP90 (CST 4874S, 1:5000) as a loading control.

Statistics and reproducibility

Statistical significance was assayed with GraphPad Prism software. The specific tests used are indicated in the figure legends. Data are presented as mean ± standard deviation or mean ± standard error as indicated in the figure legends. Mice were randomized where applicable, and quantifications from in vivo experiments were performed in at least two biological replicates as indicated in the figure legends. Two liver tumors (1548-liver from RPR2 and 997-liver from RPR2;Nfib^f/f) were excluded from the RNA-sequencing data because they were determined to not be SCLC: They had low expression of all SCLC markers by RNA-sequencing, and liver sections from the same mice showed presence of metastases that did not resemble SCLC by histology. Investigators were not blinded to allocation during experiments and outcome assessment.

Data availability

RNA sequencing, ATAC sequencing, and CUT&RUN datasets generated in this study are available at Gene Expression Omnibus (GEO) under Super Series GSE228334 [https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE228334]. Whole genome sequencing datasets are available at the NCBI Sequence Read Archive (SRA) under accession PRJNA945077 [https://www.ncbi.nlm.nih.gov/sra/PRJNA945077]. The code for the whole genome sequencing, RNA sequencing, ATAC sequencing, and CUT&RUN analyses are available at Zenodo, DOI: 10.5281/zenodo.7878261. mRNA expression (log2) from the SCLC NCI-DTP cell line set is publicly available through the SCLC-CellMinerCDB online tool (v1.2) (27). All other raw data are available upon request from the corresponding author.

Results

NFIB is not required for cancer initiation in the RPR2 mouse model of SCLC-A

To investigate the role of NFIB in SCLC tumor development in vivo, we crossed conditional Nfib knockout mice (Nfib^f/f mice) to the RPR2 mouse model of SCLC with the Rosa26^mTmG/mTmG (R26^mTmG) fluorescent reporter allele (hereafter, RPR2 mice) to generate Rb1^f/f;Trp53^f/f;Rbl2^f/f;R26^mTmG;Nfib^f/f mice (hereafter, RPR2;Nfib^f/f mice). In these RPR2;Nfib^f/f mutant mice, Nfib is knocked out at tumor initiation by concurrent deletion of Nfib along with the Rb1, Trp53, and Rbl2 tumor suppressor genes by Ad-CMV-Cre. Cells expressing Cre also switch from tdTomato-positive to GFP-positive (19), and in this model, tumors can be identified by histology and by immunostaining for GFP (Fig. 1A) (see Methods).

Fig. 1. NFIB is not required for the development of SCLC tumors in the RPR2 mouse model.

A. Representative images of serial sections from RPR2 and RPR2;Nfib^f/f mutant tumors stained with H&E (hematoxylin and eosin) or immunostained for GFP (brown signal, with blue hematoxylin counterstain), 4.5 months post Ad-CMV-infection. Scale bar, 100 μm. B. Representative images of the lungs of RPR2 and RPR2;Nfib^f/f mutant mice stained with H&E, 4.5 months post Ad-CMV-Cre infection. Arrows point to representative tumors. Scale bar, 1mm. C. Quantification of (B) for tumor number per mm². Here and in (D), tumors were confirmed by GFP immunostaining of serial sections as in (A). Point shapes represent lungs from n=2 cohorts of mice given virus at separate times. Outlined points represent the averages from each cohort. D. Quantification of (B) for tumor burden (% tumor area relative to lung area). E. Representative images of serial sections from RPR2 and RPR2;Nfib^f/f mutant tumors stained with H&E or immunostained for NFIB. Scale bar, 100 μm. F. Quantification of NFIB expression from (E). Here and below, point shapes represent tumors from two lung sections, and outlined points represent the averages from each lung. G. Representative images of RPR2 and RPR2;Nfib^f/f mutant tumors immunostained for ASCL1. Scale bar, 100μm. H. Quantification of ASCL1 expression from (G). I. Representative images of RPR2 and RPR2;Nfib^f/f mutant tumors immunostained for Ki67. Scale bar, 100 μm. J. Quantification of Ki67 expression from (I). K. Representative images of RPR2 and RPR2;Nfib^f/f mutant tumors immunostained for the neuroendocrine marker UCHL1 or the non-neuroendocrine marker HES1. Scale bar, 100 μm. L. Quantification of HES1 expression from (K). All p values calculated by two-tailed Mann-Whitney test. Error bars in (C) and (D) indicate mean ± SEM. Error bars in (F), (H), (J), and (L) indicate mean ± SD.

NFIB is required for lung development in mice (25) and is expressed in early primary RPR2 mutant tumors (before Nfib gene amplification) (19). This raised the possibility that NFIB is required for tumor initiation and development in this mouse model. We compared primary lung tumors in RPR2 and RPR2;Nfib^f/f mutant mice 4.5 months after Ad-CMV-Cre infection, at a time when primary tumors are relatively small. We found that both the number of detectable lesions and total tumor burden in the lungs of RPR2;Nfib^f/f mutant mice were not significantly different compared to RPR2 mutant controls (Fig. 1B-D). Immunostaining for NFIB confirmed that all RPR2;Nfib^f/f mutant tumors had lost expression of NFIB (Fig. 1E,F), showing efficient recombination of the Nfib^f/f alleles and suggesting little to no selection pressure to retain NFIB in early stages of tumor development in this mouse model when NFIB levels are low.

Loss of NFIB in RPR2;Nfib^f/f mutant tumors may also lead to changes in cell fate, but we found that the tumors in RPR2;Nfib^f/f mutant mice histologically resembled tumors in RPR2 mutant mice. Immunostaining for ASCL1, a key marker of the SCLC-A subtype, showed no difference between RPR2 and RPR2;Nfib^f/f mutant lung tumors (Fig. 1G,H). In addition, there was no significant difference in the expression of the proliferation marker Ki67 between RPR2 and RPR2;Nfib^f/f mutant lung tumors (Fig. 1I,J). RPR2 mutant tumors initiated by Ad-CMV-Cre are comprised of neuroendocrine and less/non-neuroendocrine cancer cells (46,47), and immunostaining for the neuroendocrine marker UCHL1 and the non-neuroendocrine marker HES1 showed no difference in heterogeneity between RPR2 and RPR2;Nfib^f/f mutant lung tumors (Fig. 1K,L).

Together, these experiments indicate that NFIB is not required for the initiation and the early development of SCLC-A tumors in the RPR2 mouse model.

NFIB promotes tumor progression when upregulated in the RPR2 mouse model of SCLC-A but is not required for metastasis

To determine if loss of NFIB affected late tumor progression in the RPR2 model, we compared tumor development in mutant mice 6.5 months after initiation, when some RPR2 mutant tumors usually have gained metastatic potential. At this time point, we observed a significant reduction in tumor burden in the lungs of RPR2;Nfib^f/f mutant mice compared to RPR2 mutant mice (Fig. 2A,B). As expected for this time point (19), RPR2 mutant tumors showed a wide range of NFIB expression, indicating that some clonal populations in the tumors have upregulated NFIB expression. In contrast, nearly all RPR2;Nfib^f/f mutant tumors remained negative for NFIB expression with very rare exceptions (Fig. S1A,B). Immunostaining for Ki67 showed an association between higher NFIB expression and increased proliferation in RPR2 mutant tumors (Fig. S1C-E). This difference in proliferation is likely contributing to the difference in tumor burden between RPR2;Nfib^f//f and RPR2 mutant mice. In cell lines derived from RPR2 and RPR2;Nfib^f/f tumors, there was a trend for cell lines with high expression of NFIB to proliferate faster than cell lines with no NFIB expression (Fig. S1F,G). Furthermore, mice in the RPR2 model generally die from their lung tumor burden. Accordingly, when we aged RPR2 and RPR2;Nfib^f/f mutant mice until moribund, we found that RPR2;Nfib^f/f mutant mice lived significantly longer than RPR2 mutant mice (Fig. 2C). RPR2;Nfib^f/f mutant tumors expressed no detectable NFIB at that time by immunostaining (Fig. S1H).

Fig. 2. NFIB contributes to tumor development when upregulated but is not required for metastasis in the RPR2 mouse model.

A. Representative images of RPR2 and RPR2;Nfib^f/f mutant lungs stained with H&E, 6.5 months post Ad-CMV-Cre infection. Scale bar, 1 mm. B. Quantification of (A) for tumor burden. Here and below, point shapes represent lungs from n=3 cohorts of mice given virus at separate times. Outlined points represent the averages from each cohort. C. Survival curves and median survival time of RPR2 and RPR2;Nfib^f/f mutant mice. D. Representative images of RPR2 and RPR2;Nfib^f/f mutant lungs stained with H&E at morbidity. Scale bar, 1 mm. E. Quantification of (D) for tumor burden. F. Proportion of RPR2 and RPR2;Nfib^f/f mutant livers with SCLC tumors at morbidity. G. Quantification of metastatic tumor burden in RPR2 and RPR2;Nfib^f/f mutant livers at morbidity. H. Quantification of metastatic tumor number per mm² in RPR2 and RPR2;Nfib^f/f mutant livers at morbidity. I. Representative images of livers (left) and metastatic tumors (right) from RPR2 and RPR2;Nfib^f/f mutant mice at morbidity, immunostained for NFIB. Scale bar, 1 mm (left), 100 μm (right). J. Representative images of serial sections of metastatic tumors as in (I), stained with H&E or immunostained for ASCL1. Scale bar, 100 μm. P values for (B), (E), (G), and (H) calculated by two-tailed Mann-Whitney test. P value for (C) calculated by logrank test. P value for (F) calculated by two-tailed Fisher’s exact test. Error bars indicate mean ± SEM.

The difference in tumor growth at later stages of SCLC development in RPR2 and RPR2;Nfib^f/f mutant mice made it more challenging to comparatively assess the seeding of metastases between NFIB wild-type and knockout tumors, as fewer metastases would be expected to be seeded in mice with lower tumor burdens. However, tumor burden and tumor histology at death in RPR2 and RPR2;Nfib^f/f mutant mice were similar between the two genotypes (Fig. 2D-E and Fig. S1I), allowing us to compare metastases while controlling for primary tumor size, with the limitation that metastases had on average more time to develop in RPR2;Nfib^f/f mutant mice. At death, 76% of RPR2 mutant mice (n=21) and 50% of RPR2;Nfib^f/f mutant mice (n=14) had liver metastases (Fig. 2F). Metastatic tumor burden by area and by number were similar at this time point (Fig. 2G,H). None of the RPR2;Nfib^f/f mutant liver metastases expressed detectable levels of NFIB by immunostaining (Fig. 2I) but were still histologically similar to RPR2 mutant liver metastases, with similarly high ASCL1 immunostaining and little to no HES1 immunostaining (Fig. 2J and Fig. S1J).

Thus, tumor growth in the RPR2 model is more dependent on NFIB as NFIB levels increase during tumor progression, but tumors in RPR2;Nfib^f/f mutant mice are still metastatic and can still form liver metastases without NFIB.

NFIB is not required for tumor development or metastasis in RPR2 tumors initiated in pulmonary neuroendocrine cells

To test if NFIB is necessary for tumor progression and metastasis in SCLC tumors originating from pulmonary neuroendocrine cells (PNECs), we infected RPR2 and RPR2;Nfib^f/f mice with Ad-CGRP-Cre. At morbidity, we found that both Ad-CGRP-Cre RPR2 and RPR2;Nfib^f/f mutant lungs had similar tumor burden (Fig. S2A,B). Because of the lower number of primary tumors compared to Ad-CMV-Cre (24), Ad-CGRP-Cre RPR2 and RPR2;Nfib^f/f mutant mice were less likely to die from lung tumor burden and frequently died from other health issues, and their survival times were similar (Fig. S2C). Primary lung tumors in RPR2;Nfib^f/f mutant mice given Ad-CGRP-Cre were almost all NFIB^neg (Fig. S2D,E), and all RPR2;Nfib^f/f mutant mice harvested at morbidity had NFIB^neg liver metastases (Fig. S2F-H). We note that in this analysis, RPR2 mutant mice had a greater proportion of NFIB^high tumors than previously reported (24), possibly due to differences in genetic backgrounds. Two of the RPR2;Nfib^f/f mutant mice had lung tumors and liver metastases with low but detectable levels of NFIB (Fig. S2G,H), but these tumors were the exception and may come from the few primary tumors with no deletion of Nfib (Fig. S2E).

Overall, our findings show that in the RPR2 model, NFIB is not required for tumor initiation or metastasis in tumors initiated from PNECs.

NFIB is required for cancer development in a mouse model of SCLC-N with high levels of NFIB at initiation

Our current work and other previous studies have focused on the role of NFIB in mouse models of SCLC-A (19,21-23), and whether NFIB contributes to tumor growth and metastasis in other SCLC subtypes remains unknown. The Rb1^f/f;Trp53^f/f;H11^{CAG-LSL-MycT58A} (RPM) mouse model of SCLC is characterized by high levels of the NEUROD1 transcription factor and is thus a relevant model for the SCLC-N subtype (16,18). Notably, RPM mutant tumors express high levels of NFIB from the time they are initiated, which is likely a consequence of the direct transcriptional activation of Nfib by MYC in this model (18).

To investigate whether NFIB is necessary for tumor development in the RPM model, we crossed Nfib^f/f mutant mice to RPM mice to generate RPM;Nfib^f/f mutant mice. Around 30 days after tumor initiation with Ad-CGRP-Cre, when tumors in the RPM model are readily visible, we observed a trend for fewer tumors in lung sections from RPM;Nfib^f/f mutant mice compared to RPM controls (Fig. 3A,B). However, unlike in RPR2;Nfib^f/f mutant mice, in which nearly all tumors efficiently knocked out Nfib, most RPM;Nfib^f/f mutant tumors retained similarly mid/high levels of NFIB as in control RPM tumors (Fig. 3C,D). The large number of NFIB^high tumors in RPM;Nfib^f/f lungs suggests there is strong selective pressure at initiation for tumors to retain NFIB in this mouse model. RPM mice die at a median of 60 days after tumor initiation from their tumor burden, and at that time, RPM;Nfib^f/f mutant lungs had significantly lower tumor burden by area compared to RPM mutant lungs (Fig. 3E,F). As a result, RPM;Nfib^f/f mutant mice survived longer before succumbing to their lung tumor burden (Fig. 3G,H). At the time of death, some large tumors in RPM;Nfib^f/f mutant lungs were NFIB^neg (Fig. 3I) and had similarly near-ubiquitous Ki67 positive staining as NFIB^high tumors (Fig. 3J), possibly using an alternate molecular mechanism to increase proliferation in place of NFIB. Human cell lines with low MYC expression have almost invariably mid-to-high NFIB expression, while cell lines with high MYC expression have a much higher range (Fig. S3) (27), suggesting that the oncogenic expression of MYC allows SCLC cells to grow without a need for NFIB.

Fig. 3. NFIB loss is counterselected for at the time of tumor initiation in RPM mutant mice.

A. Representative images of RPM and RPM;Nfib^f/f mutant lungs stained with H&E, 1 month post Ad-CGRP-Cre infection. Arrows point to representative tumors. Scale bar, 1 mm. B. Quantification of (A) for tumor number per mm². C. Representative images of RPM and RPM;Nfib^f/f mutant tumors immunostained for NFIB, 1 month post Ad-CGRP-Cre infection. The arrow points to a lesion with low/no NFIB expression. Scale bar, 100 μm. D. Staining intensity for NFIB expression from (C) from a scale of 0 (no signal) to 3 (high signal). E. Representative images of RPM and RPM;Nfib^f/f mutant lungs stained with H&E, 2 months post Ad-CGRP-Cre infection. Scale bar, 1 mm. F. Quantification of (E) for tumor burden. G. Survival curves and median survival time of RPM and RPM;Nfibf/f mutant mice. H. Representative images of RPM and RPM;Nfib^f/f mutant lungs stained with H&E at morbidity. Scale bar, 1 mm. I. Representative images of RPM and RPM;Nfib^f/f mutant lungs immunostained for NFIB at morbidity. The arrow points to a large tumor with low/no NFIB expression. Scale bar, 1 mm. J. Representative images of RPM and RPM;Nfib^f/f mutant tumors immunostained for NFIB and Ki67. The near-ubiquitous staining for Ki67 made it challenging to accurately quantify the number of positive cells. Scale bar, 100 μm. P values in (B) and (F) calculated by two tailed Mann-Whitney test. P value in (G) calculated by logrank test. Error bars indicate mean ± SEM.

These data in RPM mutant mice indicate that, similar to the RPR2 model, the presence of NFIB affects SCLC tumor growth at the time when NFIB levels are elevated, with a more critical role for NFIB at the time of initiation in the RPM model. Previous studies in human SCLC cell lines expressing high levels of MYC also showed that NFIB is required for the growth of these cell lines (20), and its overexpression is sufficient to promote further growth (19). These observations support the role of NFIB as an oncogenic driver of SCLC in different subtypes of this cancer.

The rapid death of RPM mutant mice and the frequent retention of NFIB in RPM;Nfib^f/f mutant mice, however, made it difficult to conclusively determine the role of NFIB in metastatic progression in this model. While mice in the RPM model develop metastases with high NFIB expression (18), we rarely found metastases at morbidity in our RPM or RPM;Nfib^f/f mutant mice. Furthermore, because of the presence of NFIB^high RPM;Nfib^f/f primary tumors, we would not have been able to determine if any NFIB^high metastases in RPM;Nfib^f/f mice arose because NFIB was contributing to metastatic ability or by chance from primary tumors that retained NFIB expression. Thus, we decided to pursue our investigation of NFIB in SCLC metastatic progression solely in RPR2 mutant mice.

NFIB is not required for EMT in the RPR2 mouse model of SCLC-A

E-Cadherin/CDH1 is a marker of epithelial cells and is involved in cell adhesion (48,49). In SCLC and other cancers, CDH1 expression is inversely correlated with an epithelial-to-mesenchymal transition (EMT) signature (48,50), and low CDH1 levels have been associated with poorer survival in patients with SCLC (51). In mouse models of SCLC-A, primary tumors are mostly NFIB^low CDH1^high, and upregulation of NFIB during tumor progression has been found to correlate with low levels of CDH1 (22,52).

Consistent with those previous results, primary tumors in the lungs of RPR2 and RPR2;Nfib^f/f mutant mice were CDH1^high at an early time point (Fig. S4A). At morbidity, lung tumors in RPR2 mutant mice showed the expected heterogeneity in CDH1 expression, with a shift towards NFIB^high CDH1^low/neg tumors (Fig. S4B), and the majority of liver metastases were CDH1^neg (Fig. S2C,D). Notably, in RPR2;Nfib^f/f mutant lungs, tumors showed a similar heterogeneity in CDH1 expression despite being entirely NFIB^neg (Fig. S4B), and NFIB^neg metastases in RPR2;Nfib^f/f mutant livers expressed a range of CDH1 expression levels by immunostaining (Fig. S4C,D). Thus, while NFIB^neg metastases in RPR2;Nfib^f/f mutant mice had higher levels of CDH1 on average, the large proportion of CDH1^low/neg tumors suggests that SCLC tumors can gain more mesenchymal features independently of NFIB levels. Furthermore, in RPR2 and RPR2;Nfib^f/f mutant mice, the presence of metastases with high levels of CDH1 indicates that sustained loss of CDH1 is not absolutely required for metastatic progression.

Together, these data indicate that neither upregulation of NFIB nor loss of CDH1 are necessary for metastatic progression in this mouse model of SCLC, which led us to further investigate molecular phenotypes in NFIB wild-type and knockout SCLC metastases.

NFIB^neg tumors are molecularly distinct from NFIB^high tumors in the RPR2 model

In mouse models of SCLC-A, late-stage tumors and metastases frequently have genomic rearrangements on chromosome 4, including independent focal amplifications of the Nfib and Mycl loci, which correlates with high expression levels of NFIB and MYCL (20,22,53). Previous work in the RP model has shown that targeted deletion of the Pten gene on chromosome 19 is sufficient to abrogate the loss of chromosome 19, which is otherwise frequently observed during tumor progression (53). Similarly, targeted loss of Nfib would be expected to prevent amplification of its chromosomal locus, unless other genes near Nfib are SCLC drivers that benefit from amplification during tumor progression.

We performed low-pass whole genome sequencing (WGS) on 5 cell lines derived from individual RPR2;Nfib^f/f mutant large primary tumors or liver metastases as well as one RPR2 mutant cell line (N2N1G) derived from a lymph node metastasis (19) (Fig. 4A). As expected, the RPR2 mutant cell line N2N1G showed amplification at the Nfib locus (Fig. 4B), which correlated with high levels of the NFIB protein (Fig. 4C). Of the RPR2;Nfib^f/f mutant cell lines analyzed, 4/5 did not amplify the chromosomal region including the Nfib locus (Fig. 4B and Fig. S5A), and 0/5 cell lines had elevated protein levels of NFIB (Fig. 4C), indicating that loss of Nfib decreases the selective pressure for amplification of this locus. One of the RPR2;Nfib^f/f mutant cell lines showed amplification of the Nfia locus, also on chromosome 4 (Fig. 4B). This observation suggests that this family member of Nfib may sometimes compensate for the inability of these cells to upregulate Nfib, as suggested before in other contexts (54,55), but we did not pursue the analysis of NFIA function in this cell line. None of the cell lines analyzed showed focal amplification of the other NFI family members (Nfic located on chromosome 10, Nfix located on chromosome 8) (Fig. 4D).Of the RPR2;Nfib^f/f cell lines, 2/5 showed loss of chromosome 19 where Pten is located, and 3/5 showed focal amplification of Mycl (Fig. 4D), as previously described in RP mutant tumors (53). While the number of samples we analyzed were limited, the observed heterogeneity suggested that the ability to metastasize without NFIB is not limited to tumors driven by a specific known chromosomal alteration.

Fig. 4. NFIB^neg and NFIB^high SCLC metastases display distinct chromosomal and transcriptional features.

A. Cell lines used for low-pass whole genome sequencing, their genotypes, and the type of tumor from which they were derived. B. Plots showing chromosome 4 gains and losses from three cell lines, as noted. Yellow lines indicate loci for Nfib, Mycl, and Nfia for 1399M2. N2N1G cells are derived from lymph node metastases from the RPR2 model and show Nfib genomic amplification as well as Mycl amplification. Cell lines derived from liver metastases from the RPR2;Nfib^f/f model show no genomic amplification of the Nfib locus with the exception of 1399M2. C. Immunoassay for NFIB levels in cell lines in (A). Iso1 and iso3 indicate protein isoforms 1 and 3 of NFIB, respectively. HSP90 serves as a loading control. D. Plots showing genome-wide gains and losses from low-pass whole genome sequencing of cell lines in (A). Note the loss of chr19 in two of the RPR2;Nfib^f/f cell lines (1398M2 and 1399M2). E. Principal component analysis of RNA sequencing from primary lung tumors and liver metastases macrodissected from the RPR2 and the RPR2;Nfib^f/f models. F, G. Top enriched terms from KEGG 2021 Human pathways in genes upregulated in liver metastases over primary lung tumors in the RPR2 (F) and RPR2;Nfib^f/f (G) models. Differentially expressed genes (log2 fold change > 1.5, adjusted p value < 0.05) were filtered for those that did not overlap with the liver gene set in the Mouse Gene Atlas to account for contaminating normal liver tissue. P values were calculated by Enrichr (Fisher’s exact test).

In RPR2 mutant tumors, amplification of Nfib and upregulation of NFIB have been associated with changes in genes programs related to cell motility, cell migration, and neuronal differentiation, which may all partly contribute to the enhanced metastatic ability of NFIB^high cells (19,33). To determine if similar gene programs were being upregulated during metastasis in RPR2;Nfib^f/f mutant tumors in the absence of NFIB upregulation, we compared gene expression programs in macrodissected RPR2 and RPR2;Nfib^f/f mutant lung tumors and liver metastases by RNA sequencing (RNA-seq). RPR2 mutant metastases clustered separately from RPR2;Nfib^f/f mutant metastases, indicating that the tumors were transcriptionally distinct (Fig. 4E and S5B). We compared the genes that were differentially up- or downregulated between primary tumors and metastases in RPR2 and in RPR2;Nfib^f/f mutant mice and found little overlap (Fig. S5C and Table S2). The genes that did overlap were primarily liver or lung-specific (Fig. S5D and Table S3), likely due to the presence of contaminating normal lung or liver tissue in the tumor samples collected by macrodissection. Genes that were upregulated in metastases compared to primary tumors in RPR2 mutant mice were enriched for neuronal terms as expected (19,33) (Fig. 4F and Table S3), whereas genes upregulated in RPR2;Nfib^f/f mutant metastases compared to primary tumors were enriched for cell cycle terms (Fig. 4G and Table S3). To specifically interrogate neuronal enrichment between primary and metastatic tumors, we performed Gene Set Enrichment Analysis (GSEA) on the differentially expressed genes using an axon-related gene signature expressed in SCLC (33). We found that unlike in RPR2 mutant mice, the axon-related signature is not enriched in RPR2;Nfib^f/f mutant metastases compared to RPR2;Nfib^f/f mutant primary tumors (Fig. S5E and Table S4).

These DNA and RNA analyses suggest that the NFIB^neg SCLC cells that metastasize have gained some proliferative potential compared to cells in RPR2;Nfib^f/f mutant primary tumors, but they did not readily identify mechanisms underlying this potential or the ability of NFIB^neg SCLC cells to metastasize.

NFIB^neg metastases display a chromatin signature associated with ASCL1 and FOXA1/2 activity

In RPR2 mutant tumors, upregulation of NFIB is associated with an increase in global chromatin accessibility at NFIB binding sites; in our previous ATAC-seq (Assay for Transposase-Accessible Chromatin followed by sequencing) analysis, over 20,000 sites were differentially open between NFIB^high metastases and NFIB^low primary tumors (19). We wondered if the analysis of chromatin accessibility in RPR2;Nfib^f/f mutant tumors and metastases may reveal a similar increase in chromatin accessibility during metastatic progression and/or point to factors implicated in the metastasis of NFIB^neg SCLC cells. We performed an ATAC-seq analysis of cell lines from individual lung tumors and liver metastases from RPR2;Nfib^f/f mutant mice, using cell lines from liver and lymph node metastases from RPR2 mutant mice as controls. This analysis showed few differences between metastases and primary tumors in the RPR2;Nfib^f/f model (Fig. 5A), as well as large differences between metastases when comparing the RPR2;Nfib^f/f model against the RPR2 model (Fig. 5B). As expected, RPR2;Nfib^f/f mutant metastases showed less accessibility at peaks previously found to be associated with NFIB (19) compared to RPR2 mutant metastases (Fig. S6A). RPR2 mutant metastases clustered separately from RPR2;Nfib^f/f metastases, suggesting that their chromatin state is not a shared phenotype of metastasis (Fig. 5C and Fig. S6B). Overall, metastatic progression in the absence of NFIB does not involve the increase in global accessibility characteristic of NFIB upregulation.

Fig. 5. NFIB^neg SCLC metastases show distinct chromatin states enriched for FOXA1/2 and ASCL1-related motifs.

A, B. Differential accessibility (log2 fold change in reads per peak) plotted against the mean reads per peak by ATAC sequencing. Comparisons are between cell lines derived from RPR2;Nfib^f/f mutant metastases (mets) and cell lines derived from (A) RPR2;Nfib^f/f mutant primary tumors and (B) RPR2 mutant metastases. Number of peaks that are differentially accessible are indicated. C. Correlation of ATAC sequencing with cell lines derived from RPR2;Nfib^f/f mutant liver metastases, RPR2;Nfib^f/f mutant primary lung tumors, RPR2 mutant liver metastases, and RPR2 mutant lymph node (LN) metastases. D, E. Motif enrichment in regions differentially open in RPR2;Nfib^f/f metastases from (A) and (B), respectively. Orange bars indicate motifs for transcription factors that are expressed in SCLC tumors as determined by RNA sequencing (see Figure S6C). Asterisks indicate basic helix-loop-helix (bHLH) transcription factors. Duplicate motifs identified for TCF12 (C), TCF3 (D), and FOXA1 (C,D) are not shown but can be found in Table S5.

To identify candidate transcriptional regulators that might be contributing to metastasis in RPR2;Nfib^f/f mutant mice, we searched for binding motifs enriched in regions that were more accessible in RPR2;Nfib^f/f mutant metastases compared to RPR2;Nfib^f/f mutant primary tumors (Fig. 5D and Table S5) or RPR2 mutant metastases (Fig. 5E and Table S5). The comparison between RPR2;Nfib^f/f mutant metastases and RPR2;Nfib^f/f mutant primary tumors (Fig. 5D and Table S5) identified an enrichment for binding sites for basic helix-loop-helix (bHLH) transcription factors (group A, (56)). Our RNA-seq analysis shows that of the bHLH factors with enriched motifs, only ASCL1, TCF3, TCF4, and TCF12 are expressed in our SCLC-A tumors (Fig. S6B and Table S5). ASCL1 is the only one expressed at high levels, and TCF3, TCF4, and TCF12 are known heterodimer partners for ASCL1 (57,58). ASCL1 is a known driver of SCLC development (59,60), so our enrichment analysis suggests that RPR2;Nfib^f/f mutant tumors rely on ASCL1 activity to progress and metastasize. The comparison between RPR2;Nfib^f/f mutant metastases and RPR2 mutant metastases (Fig. 5E and Table S5) identified an enrichment for motifs for the FOX and SOX families of transcription factors. Of the top enriched motifs, the transcription factors expressed in SCLC-A tumors in our RNA-seq analysis are FOXA1, FOXA2, FOXM1, and SOX2 (Fig. S6C and Table S5). FOXM1 and SOX2 are both ASCL1 targets (59,61) and have oncogenic roles in SCLC (62,63). Similarly, FOXA1 and FOXA2 have both been shown to be part of the ASCL1 regulatory network in SCLC-A tumors (57,60), which further supports the importance of the ASCL1 network in RPR2;Nfib^f/f mutant tumors.

To further investigate possible changes in the activity of FOXA1/2 factors in SCLC metastases with no NFIB, we performed a Cleavage Under Targets and Release Using Nuclease (CUT&RUN) (64) analysis for FOXA1 in two of the cell lines on which we performed the ATAC-seq analysis: N2N1G (cell line from an RPR2 mutant metastasis) and 1399M2 (cell line from an RPR2;Nfib^f/f mutant metastasis). First, we identified a consensus peak set from the open chromatin regions in the ATAC-seq analyses (Fig. 5C, S7A). FOXA1 binding to this consensus peak set was marginally higher overall in 1399M2 compared to N2N1G (Fig. 6A,B). When we focused on chromatin regions that were differentially accessible between N2N1G and 1399M2, we found that at sites differentially open in N2N1G, FOXA1 binding was enriched in N2N1G compared to 1399M2 (Fig. 6C and Fig. S7B), and at sites differentially open in 1399M2, FOXA1 binding was enriched in 1399M2 compared to N2N1G (Fig. 6D and Fig. S7C). In tandem, these suggest that FOXA1 binding is independently involved in chromatin accessibility in both N2N1G and 1399M2. Furthermore, when we analyzed consensus binding motifs for transcription factors at FOXA1 binding sites, we found FOXA binding motifs as expected in both N2N1G and 1399M2 (Fig. 6E,F and Table S6). In contrast, NFI binding motifs were enriched at FOXA1 binding site in N2N1G (Fig. 6E and Table S6), but not in 1399M2 (Fig. 6F and Table S6).

Fig. 6. FOXA1 binding sites are enriched in open chromatin regions in NFIB^neg SCLC cells.

A. Tornado plot showing IgG and FOXA1 binding signal by CUT&RUN analysis at all accessible ATAC peaks in N2N1G (NFIB^high cell line derived from RPR2 metastasis) and 1399M2 (NFIB^neg cell line derived from RPR2;Nfib^f/f metastasis). B. Quantitation of genome-wide FOXA1 binding in (A) shows no significant difference between N2N1G and 1399M2. C, D. Heatmap and profile plots showing FOXA1 binding at ATAC peaks that lose (C) and gain (D) accessibility in 1399M2 compared to N2N1G. E, F. Motif enrichment in FOXA1 binding sites in N2N1G (E) and 1399M2 (F). Green bars in (E) indicate motifs for NFI family members. Duplicate motifs for FOXA1 and motifs for transcription factors without mouse orthologs are not shown but can be found in Table S6. G. Model for metastasis of SCLC-A tumors. When SCLC cells can upregulate NFIB (top), high NFIB activity acts as a switch that promotes tumor growth and metastasis. NFIB upregulation opens the chromatin at numerous loci in the genome, with FOXA1/2 as possible co-factors. When SCLC cells are unable to upregulate NFIB (bottom, for example when the gene is knocked out), tumor progression is slower, and cells that metastasize may rely on transcription factors such as ASCL1 and FOXA1/2, which may work in independent programs.

Thus, SCLC-A cells can metastasize by turning on an “NFIB switch”, potentially in conjunction with FOXA1/2. When this switch is not possible, these cells rely on oncogenic programs centered around the ASCL1 and the FOXA1/2 transcription factors (Fig. 6G).

Discussion

Metastases are a major cause of morbidity and mortality in patients with SCLC. Clinical progress for this recalcitrant cancer is thus likely to come from a better understanding of the mechanisms that allow SCLC cells to metastasize. The identification of NFIB as a pro-metastatic oncogene in a majority of SCLC cases (19,22) has provided for the first time a molecular handle to comprehend metastatic progression in this fatal disease. Here we sought to determine if NFIB is required for SCLC cells to gain metastatic potential. To this end, we used relevant mouse models in which upregulation of NFIB is associated with tumor growth and metastasis. Our results conclusively show that there is no absolute requirement for NFIB in these models for metastasis. These observations underscore the highly plastic nature of SCLC and the ability of SCLC cells to take various molecular paths during tumor progression.

NFIB is not required for the growth of NFIB^high SCLC populations in culture or in subcutaneous tumors, but SCLC cells in circulation rely on NFIB to successfully metastasize to the liver (19). One mechanism underlying this requirement for liver metastasis may originate from some of the neuronal features of NFIB^high cells, which have been linked to new migratory capabilities (33). NFIB is also required for optimal growth of SCLC cells in the brain microenvironment, which again can be linked to neuronal features of NFIB^high SCLC cells (65). Our new data with the same RPR2 model indicate that NFIB plays a role in the growth of primary tumors in the lungs, especially after NFIB levels are upregulated during tumor progression, and the decreased growth of NFIB^neg tumors in this context is correlated with a decreased proliferative potential. It is thus possible that some of the pro-metastatic effects of NFIB simply come from its pro-proliferative role, which becomes more visible in comparison to NFIB^neg tumors. A similar role for NFIB in proliferation has been observed before in mouse and human SCLC cells in culture (20,23), including MYC-driven human SCLC cell lines with varying levels of NFIB (19,20). In the MYC-driven RPM mouse model, the high Ki67 expression we found in late NFIB^neg tumors suggests that MYC-driven SCLC may be less reliant on NFIB to increase proliferation, but NFIB may still play a critical role in tumor progression, as we also observed a strong pressure to retain NFIB expression at the time of initiation.

Recent work has further shown that methylation of NFIB by the CARM1 methyltransferase promotes the oncogenic potential of NFIB in the RPR2 model; a germline knock-in mutation in this methylation site inhibits tumor progression and metastasis (21). Notably, the inhibition of metastasis in this model was observed at an earlier time point when there was a still a difference in primary tumor burden. In contrast, all of our metastatic tumors were collected at morbidity While we did not measure metastatic burden at a set time point, it is likely that we would have seen a similar difference in metastatic tumor burden between RPR2 and RPR2;Nfib^f/f mutant mice.

Altogether, these data indicate that NFIB has oncogenic and pro-metastatic roles in multiple contexts, including via its role as a pro-proliferative factor and as a pro-neuronal factor. These data also support the notion that NFIB may remain a viable therapeutic target in SCLC. The development of strategies such as small molecule degraders (66) offers new possibilities to target this transcription factor in developing tumors and in metastases.

Our ATAC-seq data have identified ASCL1 and FOXA1/2 as candidate regulators of the metastatic process in SCLC cells in which NFIB levels cannot be upregulated. ASCL1 expression is not only the defining characteristic of SCLC-A tumors, but also necessary for the development of neuroendocrine SCLC (59,60). ASCL1 can directly activate cell cycle factors such as E2F and FOXM1 (61,67) and known SCLC oncogenes such as SOX2 and MYCL (57,59). It is likely that enhanced ASCL1 activity can promote tumor progression through multiple mechanisms in its transcriptional network, including FOXA1 and FOXA2. In SCLC, ASCL1, FOXA1, and FOXA2 expression are positively correlated (59,60), and FOXA motifs are enriched at ASCL1 binding sites (59), suggesting these three factors may co-localize to regulate oncogenic transcriptional networks.

FOXA1 has roles in multiple cancers, notably other neuroendocrine cancers (68,69), and works as a pioneer factor in conjunction with many other transcriptional regulators (70). In prostate cancer, FOXA1 and ASCL1 are known to co-localize on regulatory regions to promote the transition from prostate adenocarcinoma to neuroendocrine prostate cancer, and these FOXA1-bound regulatory regions are similarly H3K27-acetylated in SCLC (71). FOXA2 has long been known to be expressed in SCLC (72), and FOXA2 is itself an ASCL1 target (59,60). FOXA2 levels have been found to be high in cell lines derived from SCLC circulating tumor cells (73), and human SCLC cell lines are dependent on FOXA2 for their growth in culture (57). Notably, our previous ATAC-seq analysis had found FOXA1/2 binding motifs enriched at NFIB binding sites in metastases from the RPR2 model (19), which initially suggested a model in which it may be acting as a pioneer factor to promote NFIB binding. This is further supported by our CUT&RUN analysis, which shows NFI binding motifs enriched in FOXA1 binding sites in a cell line derived from an RPR2 metastasis. FOXA1/2 can also potentially promote tumor progression in the absence of NFIB, as our CUT&RUN data suggest that FOXA1 is bound at independent chromatin regions that are differentially accessible in both RPR2 and RPR2;Nfib^f/f tumors. Thus, FOXA1/2 may be critical in SCLC metastasis in both NFIB-dependent and NFIB-independent contexts. It will be important in future studies to investigate the consequences of manipulating FOXA1 and FOXA2 during tumor progression and metastasis in relevant models, including models for not only SCLC-A but also other SCLC subtypes with low or no ASCL1 expression, similar to what has been done previously in lung adenocarcinoma (74).

The vast majority of metastatic tumors in the Ad-CMV-Cre RPR2 model upregulate NFIB (19). In these tumors, FOXA1/2 may facilitate NFIB binding at new sites throughout the genome to increase chromatin accessibility, proliferation, and pro-metastatic and neuronal gene programs (Fig. 5G, top). NFIB upregulation accelerates tumor progression and metastasis, and the resulting decrease in survival limits the window of opportunity for tumors that do not upregulate NFIB to metastasize. If given enough time, cells that do not or cannot upregulate NFIB may progress through an NFIB-independent model of metastasis by relying on their high ASCL1, FOXA1, or FOXA2 expression. The ASCL1 transcriptional network includes multiple oncogenic mechanisms, including cell cycle regulation, and FOXA1/2 binding may also play an important and possibly independent role in regulating chromatin accessibility(Fig. 5G, bottom).

Metastatic tumors with NFIB deletion may have been expected to follow an evolutionary path that is dominated by one alternative mechanism. For example, upregulation of a family member of NFIB such as NFIA could have been a recurrent mechanism, especially since Nfia is on the same chromosomal region as Nfib. However, our low-pass whole genome sequencing, RNA sequencing, and chromatin analyses do not support the existence of a single major alternative mechanism (for example, just one of five tumors amplified Nfia, and we did not investigate other NFI family members). Furthermore, amplification of Nfia does not appear to be directly compensatory for amplification of Nfib, as the cell line with the Nfia amplification (1399M2) did not show enrichment of the NFI family consensus motifs at FOXA1 binding sites (Fig. 6F). Instead, our data suggest that NFIB mutant tumors can take various paths during tumor progression and metastasis. These paths may be different from metastases in the same RPR2 model where tumors are initiated by Ad-CGRP-Cre, targeting specifically pulmonary neuroendocrine cells. In this model, NFIB is less frequently upregulated during tumor progression, but metastases also upregulate neuronal programs compared to primary tumors (unlike NFIB^neg tumors, in which the only signature observed is a cell cycle signature) (24). This ability to reach a diversity of genetic and transcriptional states during tumor progression and metastasis is reminiscent of the mechanisms of resistance to treatment (e.g., chemotherapy), where resistant cancer cells reach transcriptional states that are more diverse than naïve tumors (75). These observations underscore the need for better analysis of SCLC heterogeneity both during tumor progression and in response to treatment.

Acknowledgments

We thank Alyssa Ray for administrative support; Dr. Trudy Oliver for sharing the RPM mouse model; Dr. Richard Gronostajski for sharing the Nfib^flox mouse model; Pauline Chu and the Animal Histology Service Center at Stanford University for help with histology; the Stanford Shared FACS Facility for flow cytometry services (NIH S10 Shared Instrument Grant S10RR027431-01); the Stanford Veterinary Service Center for expert animal care; the Stanford Genomics Service Center as well as the Protein and Nucleic Acid Facility for help with Bioanalyzer runs; members of the laboratory of J.S. for their help and support throughout this study. This work was supported by the NIH (J.S., CA231997 and J.H.K., F31CA257169-01, and P30-CA124435 to the Stanford Cancer institute). M.C.L. was supported by a Tom and Susan Ford Stanford Graduate Fellowship in Science and Engineering and a Tobacco-Related Disease Research Program (TRDRP) Predoctoral Fellowship (T32DT4747). D.B. is a fellow of the Damon Runyon Cancer Research Foundation (DRG-2466-22). J.S. is the Elaine and John Chambers Professor in Pediatric Cancer.