Lung cancer is the leading cause of cancer-related deaths, killing more people in the U.S. than the next three most prevalent cancer types combined [1
]. The five-year survival rate for all lung cancer patients is about 18%, which has improved marginally over the past several decades even with the improvement of genomic profiling and rational implementation of targeted therapies. Thus, a better understanding of the complexities of lung cancer progression, the contributing microenvironmental factors and how to target them would benefit patient outcomes.
Research focusing on systemic and tumor-infiltrating immune cell populations and their impact on shaping cancer progression in solid tumor types has provided compelling evidence for immune escape as a crucial survival mechanism. These studies have revealed that tumors avoid immune detection through a variety of complex mechanisms. For example, tumors can recruit immunosuppressive populations of cells such as myeloid derived suppressor cells or CD4+ T regulatory cells, which secrete suppressive cytokines that interfere with the cytotoxic functions of CD8+ T cells [3
]. Additionally, tumors upregulate expression of PD-L1, which can occur de novo through oncogenic signaling or as a consequence of IFNγ-stimulation due to immune cell activation. PD-L1 binds to the PD-1 molecule on CD8+ T cells and blocks the full activation necessary for the cytotoxicity [6
], thus representing an avenue of therapeutic intervention to promote cytotoxic activity of T cells.
The implementation of immunotherapies to release immune system braking mechanisms like those described above has been paradigm-shifting for cancer therapeutics. Clinical studies in lung cancer have revealed that inhibiting the PD-L1/PD-1 axis results in a significantly improved clinical outcome in ~15–20% of patients with lung cancer when compared to standard of care chemotherapy [8
] and thus shows promise in improving patient prognosis. While some patients do show clinical benefit to checkpoint inhibitors when administered as single agents, the majority of patients either show no response or develop resistance to single agent checkpoint inhibition [9
]; thus, discovering mechanisms of resistance and tumor cell dependencies in the face of immune-related pressure is imperative in furthering the potential for immunotherapy in treating lung tumors.
Several factors have been identified as impacting response to immune checkpoint inhibitors. For example, tumor mutational burden significantly correlates with response to immunotherapy, likely due to the creation of neoantigens that activate the immune response. Additional work has focused on oncogenic drivers of lung cancer. Kirsten rat sarcoma (KRAS) mutations occur in about 30% of lung adenocarcinomas, and unlike other common oncogenic drivers (such as epidermal growth factor receptor (EGFR) and anaplastic lymphoma kinase (ALK)), effective targeted therapeutic strategies for KRAS mutant lung cancer have been limited [16
]. Interestingly, KRAS mutant lung tumors and their degree of immune system engagement and infiltration vary based upon the co-occurring mutations found within the tumor. Patients that present with a p53 mutation concurrently with oncogenic Kras (KP) exhibit higher expression of PD-L1 and other inflammatory markers when compared to other commonly co-occurring mutations such as STK11/LKB1 or CDKN2A [17
], and these patients respond better to PD-1/PD-L1 axis blockade [18
]. However, the mechanisms of tumor-regulated immunosuppression and the potential avenues of resistance in KP mutant lung cancer are vastly unknown, and the understanding of these factors is necessary for intelligent use of immunotherapies for maximum benefit to patients.
Previous work in our laboratory has focused on understanding the biology of KP mutant lung tumors through cancer cell intrinsic properties as well as extrinsic factors influencing cancer progression that are present within the tumor microenvironment. We have previously derived murine lung cancer cell lines from the primary or metastatic lesions of the KrasLA1/+/p53R172HΔg/+
genetically engineered mouse model of lung cancer [19
]. These cells demonstrate heterogeneity in their epigenetic state and propensity to metastasize when re-implanted syngeneically into wildtype mice. Specifically, the KP murine cell lines that have undergone an epithelial-to-mesenchymal transition (EMT) are not only more metastatic and aggressive, but they also have lower CD8+ T cell infiltration and an increase in an exhaustive signature when compared to cells in an epithelial state [21
]. This heterogeneity also translates to a response to PD-1 blockade, with mesenchymal cells responding initially to the anti-PD-1 antibody but ultimately acquiring resistance [22
]. Thus, our in vivo models closely mimic patient disease progression and immune checkpoint inhibitor response, providing the opportunity to discover novel mechanisms regulating tumor response to immune checkpoint blockade in KP mutant lung cancer.
To identify novel mechanisms of KP lung cancer cell resistance to PD-1 checkpoint inhibition, we performed a clinically relevant and powerful in vivo dropout screen. KP murine mouse cell lines stably expressing the FDAome, a library of barcoded shRNAs specific to genes that encode for clinically actionable targets, were implanted into wildtype mice and treated with an anti-PD-1 antibody. Tumors were sequenced and analyzed for depleted shRNA sequences when mice were treated with an anti-PD-1 antibody, thus revealing genes essential for tumor survival in the face of PD-1 blockade. From this screen, neurotrophic receptor tyrosine kinase 1 (Ntrk1) was identified as a top lead candidate as it dropped out significantly in anti-PD-1 treated tumors. Our data indicate that Ntrk1 regulates KP cell biology including cell growth and invasion in vitro while also impacting the tumor-infiltrating immune populations and their functionality with a consistent promotion of an exhausted microenvironment. Thus, we determined that Ntrk1 is a novel regulator of immune functionality in KP lung cancer, and combinatory treatment strategies could circumvent PD-1 blockade resistance.
The data generated from the in vivo FDAome dropout screen provide compelling evidence for the use of in vivo functional genomic screens to identify novel tumor cell genes and/or pathways that promote immunosuppression and thus may contribute to immunotherapy resistance. We and others have demonstrated the utility of these screens, in the context of immunotherapy or in the context of other tumor biology hallmark dependencies (i.e., cellular growth), to provide preliminary evidence for novel drug targets as well as stimulate new research questions. For example, human patient-derived xenografts and genetically engineered mouse models have been utilized to perform loss-of-function screens using a shRNA library targeting known epigenetic regulators in pancreatic cancer to identify novel tumor survival dependencies [26
]. This group provided strong evidence for WDR5
as being essential for pancreatic tumorigenesis and thus targeting it as a potential therapeutic strategy to further explore for pancreatic cancer patients.
Recently, functional genomic screens have moved towards the CRISPR-Cas9 system for gene editing as it has several advantages over shRNA-based screens, including complete gene knockout as well as greater genomic coverage with larger libraries. Two recent screens did so in the context of tumor cell-immune cell interactions. One was completed in vitro with human T cells in co-culture with melanoma cells to identify tumor genes essential for the effector functions of T cells [27
]. The other CRISPR-Cas9 screen was performed in vivo with anti-PD-1 treatment in combination with GM-CSF-secreting, irradiated tumor cell vaccine (GVAX) [28
]. This screen contained ~2500 genes and identified hits in the IFNγ response and antigen presentation pathways as expected, but also less understood hits such as PTPN2
. From these two studies alone, the breadth of knowledge and the resources available to develop hypothesis-driven research about tumor cell influence on immune system response was significantly expanded and will continue to provide additional knowledge about these complex processes.
To contribute to these efforts and better understand response and resistance to immune checkpoint inhibitors in a complex system, we performed a clinically relevant FDAome in vivo dropout screen using a Kras/p53 mutant syngeneic mouse model of lung cancer. This model and the GEM model from which the KP mutant cell lines were originally derived [19
] closely recapitulate the progression of human lung cancer disease, with development of metastatic lesions throughout the body, as well as heterogeneity of immune infiltrate and response to immunotherapy agents [19
]; thus, the similarity to KRAS
-driven human lung cancer validates the use of these models to address specific immune-related questions, as well as to perform mechanistic and therapeutic studies. We identified Ntrk1 as a top lead hit as being essential for tumor cell survival in vivo when challenged with an anti-PD-1 antibody and therefore a potential avenue of acquired resistance. Molecular studies revealed that Ntrk1 regulates KP lung cancer cell intrinsic biological processes such as cell signaling to AKT and MAPK to promote cellular growth as well as regulation of in vitro invasive capacity. In vivo analyses demonstrated that Ntrk1 also augments immune infiltrate and functionality. Specifically, Ntrk1 expression can promote CD8+ T cell exhaustion within the tumor microenvironment, suggesting that its expression may contribute to CD8+ T cell dysfunction and thus diminish response to PD-1 inhibition.
We now provide the first evidence that Ntrk1 can regulate the expression of the immunosuppressive molecule PD-L1, likely due to modulation of Jak/Stat signaling cascade. In melanoma, loss-of-function JAK1/JAK2
mutations were discovered in a minority of patients after relapse on the anti-PD-1 treatment pembrolizumab; therefore, it is a potential avenue of acquired resistance to immune checkpoint blockade [30
]. Our data may indicate a distinct mechanism by which this pathway becomes aberrantly hyperactivated in lung cancer cells to promote immunosuppression and resistance to immune checkpoint inhibition. Ntrk1 can promote activation of Jak1 and Stat3, and Ntrk1 depletion reduces this signaling and downstream PD-L1 expression. There is little evidence in the literature connecting Ntrk1 to Jak signaling, so this mechanism needs to be further explored. However, previous work in neuronal cells demonstrated that neurotrophin-dependent stimulation of downstream transcription and neuronal cell elongation can be blocked by depletion of Stat3 [31
], suggesting that Stat3 does function downstream of neurotrophin receptors. Additionally, knockout of gp130, a type I cytokine receptor, can diminish NGF-induced neurite extension, thus linking Ntrk1 signaling and cytokine signaling. However, both of these studies were limited to neuronal cells and lacked further mechanistic studies to determine whether Ntrk1 directly interacts with these cytokine response elements. Thus, the mechanism for Ntrk1-dependent activation of Jak/Stat signaling remains to be fully elucidated. Additionally, other immunosuppressive molecules that are upregulated as a result of Ntrk1 overexpression in addition to PD-L1 need to be explored to understand the full impact of Ntrk1 upregulation on various immune subpopulations and their functionality within tumors.
In vivo functional genomics screens to address specific immune-related questions such as undiscovered mechanisms of acquired resistance to the anti-PD-1 antibody will drive the field forward and bolster our understanding of the regulatory pathways driving tumor cell evasion of immune detection and death. The goal of the work described was to identify novel therapeutic combinations to improve patient response to immunotherapy. Our data indicate that Ntrk1 may be one such hit that could be carried forward clinically to improve patient response to a single agent PD-1 blocking antibody. In some solid tumor types such as lung cancer, Ntrk1 genetic rearrangements occur infrequently (~1–3% of lung adenocarcinomas), leading to fusion proteins with constitutive kinase activity. Targeted therapies have recently been FDA approved, with most patients showing durable responses [32
]. Importantly, our preliminary in vivo screen and supporting data suggest that a significantly broader patient population may benefit from these well-tolerated Trk inhibitors in the context of immune checkpoint blockade.
4. Materials and Methods
4.1. Cell Culture and Reagents
The human lung cancer cell lines used in these studies were H1299, H157, A549, H441, H358, and HCC827. Murine lung cancer cells were created from KrasLA1/+/p53R172HΔg/+
genetically engineered mice as previously described [19
]. All lung cancer cell lines were cultured in Roswell Park Memorial Institute (RPMI) + 10% Fetal Bovine Serum (FBS). A total of 293T cells were cultured in Dulbecco’s Modified Eagle’s Medium (DMEM) + 10% FBS and were used to generate lentiviral particles for creating stable cell lines. The miR-200ab inducible H1299 cells and Zeb1-inducible 393P cells were generated using the pTRIPz plasmid as previously described by our laboratory [21
]. Expression of mir-200ab or Zeb1 was induced using 2 μg/mL of doxycycline. Ntrk1 overexpression cells were generated by subcloning human Ntrk1 cDNA from pCMV5 TrkA (Addgene plasmid #15002; http://n2t.net/addgene:15002
; RRID:Addgene_15002 [37
]) into the pLenti-puro vector backbone using EcoRI and AgeI restriction cut sites. Human Ntrk3 cDNA was cloned into the pLD6E2F vector. The Ntrk1 shRNA sequences that were used in these studies were as follows: sh#1: 5′-TCAAGCGCCAGGACATCATT; sh#2: 5′-GTGGCTGCTGGTATGGTATATCT; sh#3: 5′-TCTATAGCACAGACTATTACC; sh#4: 5′-TTGGAGTCTGCGCTGACTAAT.
NGF and NT-3 ligands were obtained from Sigma (St. Louis, MO, USA) and used at a final concentration of 100 ng/mL. MK2206, LOXO-101, trametinib, and ruxolitinib inhibitors were obtained from SelleckChem (Houston, TX, USA). Anti-PD-L1 (clone 10F.9G2), PD-1 (clone RMP1-14), and isotype control antibodies (Rat IgG2b and IgG2a, respectively) were obtained from BioXCell (West Lebanon, NH, USA).
4.2. Animal Studies
Cancer cells were prepared at a concentration of 1 × 106 cells in 100μl of serum free media. The cells were subcutaneously implanted into the right flanks of male and female syngeneic 129/sv mice of at least three months of age. Tumors were allowed to grow for 3–4 weeks, depending on the study. Where indicated, mice were treated with an anti-PD-1 or isotype control antibody via i.p. injections biweekly (100 μL per dose for a total of 200 μg). After euthanasia, tumors were measured both by calipers and weight and subsequently collected for flow cytometry analyses or sequencing for the FDAome screen (see below). All animal experiments were reviewed and approved by the Institutional Animal Care and Use Committee at the University of Texas MD Anderson Cancer Center (protocol #00001271).
4.3. FDAome Dropout Screen
Murine lung cancer cell lines (393P and 344P) were infected at a multiplicity-of-infection (MOI) of 0.3 with a pooled shRNA lentiviral library targeting 192 genes associated with FDA-approved target therapies (10 independent shRNAs/gene). The FDAome expressing cells were implanted into 129/Sv mice at 1.0 × 106 cells/mouse in triplicate for each condition. Once tumor size reached ~150 mm3 as measured by calipers, mice were either treated with an isotype control or PD-1 blocking antibody as described above. Tumors were collected after 10 days (time point 1) or 17 days (time point 2) in the 344P model, or after 25 days in the 393P model.
The shRNA-coupled barcodes were detected deploying high-throughput sequencing technology (for detailed procedures and primer sequences see the following reference) [26
]. Raw counts for the screen endpoints and a reference population, isolated after transduction, were normalized using the variance stabilizing transformation in R (version 3.3.2) with the DESeq2 in R. A fold change in barcode abundance was estimated by dividing the normalized counts by the reference. Four independent shRNA targeting essential genes (Rpl30, Psma1) or the negative control luciferase (LUC) were cloned with five unique barcodes each and incorporated in the library as positive and negative controls (20 reagents/control, see Table S1
). One LUC hairpin showed an apparent off-target effect, whereas one hairpin for Psma1 did not show a robust drop out; however, this result was not reflective of poor screen performance as the trend was consistent across the five barcodes. After excluding those hairpins, the separation of positive and negative controls was evaluated by the robust strictly standardized mean (SSM, Table S1
). Fold change distribution was converted to percentiles, and biological replicates were collapsed for RSA analysis. The RSA logP-values and ranks are provided in Table S2
4.4. Cell Viability and Growth Assays
Cell growth in Ntrk1 knockdown or overexpression cells was measured by plating an equal cell number at day 0 and then counting viable cells (using Trypan blue exclusion) every day over a three-day period. Cell growth in Ntrk3 overexpression cells was measured in 3D cultures. These cells were plated on top of a thick Matrigel layer in 8 well chamber slides as single cells. NT-3 ligand was added at the time of cell seeding. Growth was monitored over six days and media refreshed every other day. Images from day 6 were used to measure 3D structure diameter using ImageJ software (version 1.5h). Two independent chambers were quantified for each condition.
Cell viability in the presence of either dimethyl sulfoxide (DMSO,) MK2206 (1 μM), or trametinib (1 μM) was measured using a 3-(4,5-Dimethylthiazol-2-yl)-2,5-Diphenyltetrazolium Bromide (MTT) reagent (Sigma, St. Louis, MO, USA). Cells were plated in a 96 well plate and the drug was added at the time of seeding at the indicated concentrations. After 72 h, an MTT reagent (1.5 mg/mL) was added to each well, incubated for one hour, then supernatant aspirated and precipitate solubilized in DMSO. Absorbance values were read using an Epoch plate reader at 570 nm and 630 nm (background), and background absorbance values were subtracted from 570 nm values.
4.5. Invasion and Migration Assays
Cells were plated in an equal number in 8 μm Transwell inserts (BD Biosciences) placed in 24 well plates as described previously [19
]. Cells were incubated overnight for 16 h and then stained with a crystal violet solution. Chambers were then imaged by brightfield microscopy on an Olympus IX73 (Olympus, Center Valley, PA, USA), and ImageJ software was used to count cells that had migrated/invaded through the insert pores.
3D cultures were completed as previously described [19
]. Briefly, either 100% Matrigel or a Matrigel/collagen type I mixture (50:50) were used to coat 8 well chamber slides. Collagen was used at a final concentration of 1.5–2.0 mg/mL. Single tumor cells were plated on top of the matrix (1500 cells/chamber) in media containing 2% Matrigel. Cells then grew over a period of 3–5 days and invasive structures were manually counted as a percentage of total structures in each well. 30–50 structures were counted per well, and each condition was plated in triplicate.
4.6. Co-Culture Assays
Spleens were extracted from 129/Sv mice bearing either 393P or 344P tumors. These were then mechanically processed and filtered to obtain single cells. Red blood cells were lysed using red blood cell (RBC) Lysis buffer (BioLegend, San Diego, CA, USA). Splenocytes were frozen in 90% FBS/10% DSMO. After thaw, viable splenocytes were counted using Trypan blue exclusion and incubated with a far red proliferation dye (Life Technologies, Carlsbad, CA, USA). After 30 min at 37 °C, the dye was washed out with complete media, and stained splenocytes were plated at various ratios with matched tumor cells (i.e., splenocytes from 393P tumors were plated with 393P tumor cells in co-culture) in media supplemented with 5 μg/mL of anti-CD-3 and anti-CD28 (Thermo). Where indicated, an anti-PD-L1 antibody was added at the time of seeding at a concentration of 20 μg/mL. The percentage of far red positive splenocytes was then measured by flow cytometry using a FACSCanto II machine (BD Biosciences, San Jose, CA, USA).
4.7. Flow Cytometry Analysis for Immune Subpopulations
Tumors were processed for flow cytometry into single cells using mechanical and enzymatic digestion (enzyme mixture—collagenase, DNAse, and hyaluronidase). Red blood cells were lysed as described above and viable tumor cells counted. The following antibodies were used to stain immune cell populations: Ghost violet 510 Live/Dead, Pacific Blue CD45, PE-594 CD3, PE/Cy7 CD8, APC/Cy7 CD4, BV605 PD-1, APC Tim3, FITC CD62L, BV711 CD44 (BioLegend).
Samples were run on an LSR Fortessa machine. Single color compensation controls were performed using compensation beads (Thermo Fisher Scientific, Waltham, MA, USA) to correct for overlap in signal among antibodies. Spleen samples were used to set a gating strategy for CD3+/CD4+ and CD3+/CD8+ T cells. FlowJo software (version 10) was used to perform all downstream analyses on subpopulations.
4.8. Western Blot Analysis
Cells were harvested and lysed in radioimmunoprecipitation assay buffer (RIPA) lysis buffer supplemented with phenylmethylsulfonyl fluoride (PMSF), a protease inhibitor (Cell Signaling), and phosphatase inhibitors (Sigma). Lysates were separated by SDS-PAGE (BioRad, Hercules, CA, USA), transferred to nitrocellulose or polyvinylidene fluoride (PVDF) (BioRad) membranes, and probed with the following primary antibodies: phospho-TrkA(Y674/675), phospho-TrkA(Y785), TrkA, phospho-MEK, MEK, phospho-ERK, ERK, phospho-AKT, AKT, phospho-GSK3b, phospho-Jak1, Jak1, phospho-Stat3, Stat3 (Cell Signaling, Danvers, MA, USA), phospho-TrkC, total TrkC (Thermo Fisher), PD-L1 (Abcam, Cambridge, MA, USA), and actin (ProteinTech, Rosemont, IL, USA).
4.9. RNA Extraction and Real-Time qPCR
RNA was extracted from cells in vitro using a TRIzol reagent (Thermo Fisher). Tumor cell RNA was extracted using the mirVana RNA extraction kit (Life Technologies). Briefly, tumors were collected and snap frozen, then processed in 300 μL of lysis buffer using homogenization. RNA was then extracted as directed by the kit protocol.
All RNA samples were quantified, and reverse transcription was performed with 2 μg of RNA using qSCRIPT cDNA SuperMix (Quantabio, Beverly, MA, USA). Real-time PCR was performed using primer sets specific for each gene (obtained from Origene, Rockland, MD, USA) and the SYBR® Green PCR Master Mix (Life Technologies). L32 (60S ribosomal gene) was used to normalize expression across samples.
4.10. Statistical Analyses
All analyses were performed using GraphPad Prism software (version 7.01). Unpaired Student’s t tests were used in comparison of two conditions, and one-way ANOVA was used for comparisons of three or more conditions. Tukey’s test was used to correct for multiple comparisons. All analyses were 2-tailed and p-values < 0.05 were regarded as significant.