Loss of miR-29a/b1 cluster reprograms the tumor microenvironment and contributes to immunosuppression in lung cancer.

Introduction
Lung cancer remains the leading cause of cancer-related deaths worldwide, with 5-year survival across all stages of 26%(1, 2). While survival trends have improved, likely due to improved targeted and systemic therapies such as EGFR- and KRAS-specific inhibitors and immune checkpoint inhibitors (ICI) respectively, it remains a deadly disease. ICI therapy, which blocks suppressive T-cell receptors including the PD-1/PD-L1 axis and/or CTLA-4, among others, has generated promising and durable responses, but only in a subset of patients (3, 4). There remains a critical need to examine ICI response and resistance mechanisms to extend response for patients.
To identify ICI resistance mechanisms, experimental models of lung adenocarcinoma serve as valuable discovery tools, particularly those driven by mutant KRAS and p53 (Kras/p53 mutant; KP). Patients with KP mutant tumors represent a large clinical cohort and are partially responsive to ICI therapy(5, 6). We previously developed tumor models using the 344SQ KP murine lung cancer cell line, passaged in vivo with anti-PD-1, which generated upfront resistance to therapy(7). Using these models, we found the phosphodiesterase enzyme autotaxin (ATX) was upregulated in resistant tumors, but we did not determine the mechanism underlying this(7). ATX converts lysophosphatidylcholine into the bioactive metabolite, lysophosphatidic acid (LPA), which in turn activates G-protein coupled receptors called LPA receptors (LPARs) that signal downstream and influence a myriad of biological functions(8). While historically studied as an autocrine factor promoting tumor cell invasion and metastasis(9, 10), more recent studies report it has paracrine effects on tumor-infiltrating immune cells. Specifically, LPA modulates T-cell chemotaxis, metabolic fitness, T-cell receptor signaling, and cytokine production(11, 12, 13, 14, 15). Our data in lung cancer verified an immunosuppressive role for the ATX/LPA axis(7). However, while co-targeting ATX with ICI could delay resistance and even promote tumor rejection(7), complete tumor regressions occurred in only a subset of mice, suggesting other resistance mechanisms. We conjectured ATX upregulation may occur during the tumor response phase, as inflammatory cytokines such as TNFα and IFNγ are known promoters of ATX expression(16, 17). However, follow up studies using our KP models revealed no direct upregulation in ATX levels due to cytokine stimulation, suggesting an alternative mechanism of aberrant ATX upregulation.
In the current study, we used an unbiased approach to reveal mechanisms driving ICI resistance. Single-cell RNA sequencing (scRNA-seq) was performed on tumors from a representative anti-PD-1 resistant model (termed PD1R) and compared to tumors from the parental 344SQ model. Focusing on differential gene expression specifically within malignant cells, predicted regulators of this gene set were identified, revealing the micro-RNA miR-29 as an upstream regulator likely to be suppressed within the PD1R model.
The miR-29 family consists of miR-29a, -29b, and -29c, expressed as 2 genomic clusters, miR-29a/b-1 and miR-29c/b-2(18). Loss of both miR-29a and -29b is implicated in fibrosis of multiple organs, including lung, as these miRs regulate 20+ collagen genes and extracellular matrix (ECM) molecules (18). Additionally, miR-29 loss is documented in several cancers, including non-small cell lung cancer (NSCLC), where it is associated with worse overall survival (OS) and progression-free survival (PFS)(19, 20). However, the role of miR-29 loss in the tumor immune microenvironment is poorly understood. Previous literature indicates miR-29 can regulate the immune checkpoint molecule CD276, also known as B7-H3(21), for which inhibitors are currently under investigation in patients with NSCLC and melanoma, among other cancer types(22). miR-29a/B7-H3 dysregulation can influence NK-cell activation and cytotoxicity in neuroblastoma(23). The studies to date have focused only on a single gene target, underscoring the need to more broadly analyze the impact of miR-29 loss on transcriptional profiles which may alter antitumor immunity, with implications for ICI therapy response.
To investigate whether miR-29 loss contributed to immune dysfunction, we generated a miR-29a/b1 re-expression model in KP PD1R cells. We confirmed ATX as a bona fide miR-29a/b target gene and determined that re-expression of miR-29a/b1 was sufficient to increase T-cell infiltration into PD1R tumors and improve response to PD-1 blockade. These data link our previous findings highlighting ATX as an immunosuppressive molecule to the current work aimed at analyzing a broader mechanism of ICI resistance due to miR-29 loss. Finally, we used publicly available lung adenocarcinoma datasets to analyze miR-29 expression in patient tumors. These data revealed a strong correlation between miR-29 expression and positive regulation of the immune system, CD8 expression, and improved OS in patients. Together, our data indicate a new direction for miR-29 biology as a post-transcriptional regulator of a network of genes with potential for altering antitumor immunity in lung cancer.

Materials and Methods

Cell culture
All cell lines were cultured at 37°C in a humidified incubator at 5% CO2 and maintained in RPMI 1640 (Gibco) + 10% fetal bovine serum (Gibco). Human lung cancer cells were gifted to the Konen laboratory by the Gibbons laboratory in 2024 (H2122, RRID: CVCL_1531; HOP62, RRID: CVCL_1285; A549, RRID: CVCL_0023; H1299, RRID: CVCL_0060; H2073, RRID: CVCL_1521; HCC193, RRID: CVCL_5130; HCC827 RRID: CVCL_2063; HCC461 RRID: CVCL_5135; H157, RRID: CVCL_2458; H647, CVCL_1574; H2444, RRID: CVCL_1552). Cell lines have not been authenticated since gifted to the lab.
Anti-PD-1 sensitive (PD1S) and resistant (PD1R) models were derived from 344SQ Kras/p53 mutant murine cell lines, which was originally derived from the autochthonous KP model(24). Briefly, 344SQ tumor-bearing mice were treated with either IgG or anti-PD-1 until the timepoint of resistance, whereby tumors were excised and cultured over time, as previously described(7). These cells were obtained from the Gibbons laboratory in 2022. For most experiments, cell lines were used within one month from thawing and verified mycoplasma negative via the MycoAlert PLUS detection kit (Lonza). The date of last mycoplasma test was December 01, 2025.

In silico miRNA analysis
The 3’UTR of murine Enpp2 (1,391 bp) was assessed for predicted miRNA binding sites using Targetscan.org. Those miRNAs broadly conserved among vertebrates were selected as those of interest. The same analysis was completed on the human ENPP2 3’UTR (444 bp), and overlapping miRNAs between the species were compared.

Plasmid cloning
For the luciferase assays, a portion of the murine 3’UTR of Enpp2 (chr15:54702297–54702761) was cloned into the pmir-GLO plasmid (Promega) using the SacI/XbaI sites. This wildtype plasmid was then mutated using QuikChange ligation protocol using the primers outlined in Supplementary Table S1 to mutate 4 of the 7 nucleotides within the predicted miR-29 binding site. Sanger sequencing was used to confirm mutation using the primer also in Supplementary Table S1. For the miR-29 OE cell line, the miR-29a/b1 cluster was cloned into the pCW57-MCS1-2A-MSC2 plasmid (Addgene, RRID:Addgene_71782) between the NheI/BamHI sites. The resulting construct was sequenced with the primers listed in Supplementary Table S1. For the ATX rescue cells, the pCW57-MCS1-2A-MSC2 (GFP) plasmid was purchased from Addgene (RRID:Addgene_80924) and miR-29a/b1 was cloned into MCS1 as described above.

Generation of miR-29 overexpressing/knockout cells
The pCW57-miR-29a/b1, the pCW57-ctrl, or the GFP versions of these plasmids (4 μg) were transfected into HEK293T cells (RRID: CVCL_0063) with the pMD2G and psPAX2 packaging plasmids to create lentiviral particles that were then added to the PD1R1 or 344SQ cell lines. For the non-GFP plasmids, puromycin (Fisher Scientific) was used to select for transduced cells (8 μg/ml). For the GFP-containing plasmids, FACS was used to sort GFP+ cells and expand in culture. These were used to then generate the ATX rescue cells. GFP-control or GFP-miR-29a/b1 cells were transduced with either pLenti-control (RRID: Addgene_39481) or pLenti-ATX lentivirus (cloning previously described(7)), and cells were then selected via puromycin (8 μg/ml). To induce miR-29a/b1 cluster expression, doxycycline (Sigma) was added to the cells at 2 μg/ml and RNA was extracted at 24-, 48-, and 72-hour time points. The expression of miR-29a and miR-29b was confirmed using real-time PCR.
The miR-29a knockout (KO) cells were generated using the CRISPR-Cas9 gene editing system. The single guide RNAs (sgRNAs) were created and ordered (Synthego), and the sequences are included in Supplementary Table S1. Three different sgRNAs (1.2 μg) or a control (Catalog number 063-1010-000-000) were transfected separately into the 344SQ parental cells with Cas9 using Lipofectamine CRISPRMAX transfection reagent (Thermo Fisher) and following the manufacturer guidelines. A second round of transfection was performed after 24 hours. Cells were then trypsinized and plated at ~400 cells in a 10cm dish. Clones were then selected from these dishes after ~1 week of growth. Clones were screened for miR-29a KO by standard PCR amplification of genomic DNA using Phusion Plus (Thermo Fisher) and the primers listed in Supplementary Table S1. Genomic DNA was extracted using the Quick-DNA extraction kit (Zymo Research). The clones with the best KO were then also confirmed for low miR-29a expression by qPCR as described below.

Animal studies
All animal studies were completed under approval of Emory University Institutional Animal Care and Use Committee (IACUC) (protocol number 201700322) or the University of Texas MD Anderson Cancer Center IACUC (protocol #1271).
Murine lung cancer cells were implanted subcutaneously (1 × 106 cells in 100 μl serum-free RPMI media) into the right flank of 129/Sv male or female mice (Charles River; RRID: MGI:6294278) aged 3–6 months. Tumor growth was measured 2–3 times weekly via calipers and volume calculated as (width2 × length)/2, where the width is the smaller measurement. After ~1–2 weeks of tumor growth, normal chow was replaced with doxycycline-containing chow (Envigo; 625 mg/kg) to induce miR-29a/b1 cluster expression in PD1R1 tumor cells. For antibody treatments, mice were randomized into treatment groups and treated with anti-PD-1 (clone RMP1–14; BioXCell) or IgG control (clone 2A3; BioXCell) by i.p. injection weekly (200 μg/treatment in 100 μl).

Sex as a biological variable
Our study examined male and female animals, and similar findings are reported for both sexes.

Single-cell RNA sequencing (scRNA-seq)
The 344SQ and PD1R1 lines were implanted into wildtype mice. After 4 weeks (timepoint 1) or 6 weeks (timepoint 2) of growth, tumors were excised, processed into single cells and submitted for sequencing. One tumor per condition (tumor model and timepoint) was submitted for sequencing. The scRNA-seq libraries were generated using the 10X Genomics Chromium Single Cell platform (Single Cell 5′ chemistry) and were prepared using the standard 10X Genomics Single Cell 5′ Gene Expression workflow, with the specific kit version (v2). No custom sequencing adapter sequences were used. Sequencing was performed using an Illumina NovaSeq6000 S1, 100-cycle flow cell. The run format used was Read1 26 nucleotide (nt), Read2 91 nt and single index of 8 nt, and paired-end reading was used. Data were processed using Cell Ranger software v5.0.0 (RRID:SCR_017344) from 10X Genomics(25), aligned to the mm10 reference genome with UMI-based quantification for transcript counting. High quality cells were extracted and batch correction performed using the Seurat integration as previously described (26, 27). Briefly, read quality filtering was performed automatically by Cell Ranger (v5.0.0) using default settings for all four samples. Adapter trimming, barcode correction, and UMI filtering were handled internally by the pipeline. Only reads passing standard Illumina quality filters were retained for downstream analysis. Across the four samples, Q30 scores for barcode and UMI bases exceeded 90%, and RNA read Q30 scores exceeded 90%, indicating high sequencing quality. For downstream analysis in Seurat, cells expressing fewer than 200 genes or more than 5,000 genes were excluded to remove low-quality cells and potential doublets. Cells with mitochondrial transcript content greater than 10% were also removed prior to downstream analyses. Uniform Manifold Approximation and Projection (UMAP) was generated using the top principal components derived from principal component analysis (PCA) to reduce dimensionality. Differentially expressed genes (DEGs) were assessed using the FindMarkers and FindAllMarkers functions in Seurat, which involved the default two-sided non-parametric Wilcoxon rank-sum test with a Bonferroni correction. Genes with an adjusted p value < 0.05 and log2FC > 0.5 were considered significant and subjected to cell type identification and pathway enrichment analysis. Cell type identities were based on expression of established marker genes as previously described(28). Briefly, major cell type clusters were identified by markers specific to each population as follows: Epcam, Clu, and Krt7/Krt8/Krt18 for epithelial cells; Cd3e, Cd2, Cd4, and Cd8a for T cells, Ncr1 for NK cells, Cd79a/b, Cd19, Ms4a1 for B cells, and Csf1r, Fcgr2b, Lyz2, Mafb for monocyte/macrophage/dendritic cells; neutrophil markers like Csf3r, S100a8, S100a9, and G0s2; Siglech, Ly6c2, Mpeg1, and Cd209d for plasmacytoid dendritic cell (pDC); Dcn, Col1a1, Fap, and Cldn4 for fibroblasts. The robust expression of these marker genes in specific clusters supported the identification of cell types within the dataset.

miRNA mimic/antagomir transfection
344SQ PD1R1 and PD1R2 murine lung cancer cells or H2122 and HOP62 human lung cancer cells were plated in a 6-well culture dish. After 24 hours, 30 pmol of miR-29a or miR-29b mimics or a negative control miRNA were diluted in OPTI-MEM with Lipofectamine RNAiMAX (all reagents from Invitrogen; Catalog numbers/Assay IDs: 4464066 – MC12499 and MC10103 for miR-29a and miR-29b respectively; 4464058 for negative control). This was transfected into the cell lines listed above. After another 24-hour incubation, cells were then collected for semi-quantitative PCR or Western blot to assay Enpp2/ATX expression changes with miR-29 transfection relative to the negative control. The same procedure as described above was completed for the antagomirs at 60 pmol (Catalog number/Assay IDs: 4464084 – MH12499 and MH10103 for miR-29a and miR-29b respectively), transfected into the 344SQ and PD1S1 murine lung cancer cells.

In vitro co-culture assay
Naïve immune cells were collected from wildtype mouse spleens, which were processed via mechanical digestion by grinding the organ through a 40 μm strainer, followed by red blood cell (RBC) lysis. Splenocytes were cryopreserved in 90% FBS + 10% DMSO until needed. For co-cultures, splenocytes were thawed, washed, and stimulated with 1 μg/mL of anti-CD3 (clone OKT3; RRID:AB_468855) and 3 μg/mL of anti-CD28 (clone 37.51; RRID:AB_468922) (Fisher Scientific). Activated cells were then plated on top of tumor cells at a 10:1 (Spleno:Tumor) ratio in 24-well plates in replicates. The cells were cultured together for 4 days in RPMI + 10% FBS + 50 μM beta-mercaptoethanol (Sigma), after which the immune cells were collected from the media and stained for multiple markers for flow cytometry analysis of immune populations, as described below.
Bone-marrow derived progenitors (BMDMs) were isolated from four 8–10-week-old mice by flushing femurs with PBS, filtering through a 70 μm cell strainer, and plating at 3.5 × 105 cells/mL in RPMI supplemented with 10% FBS + 10 ng/mL M-CSF (Sino Biological). BMDMs were then co-cultured with miR-29 overexpressing (OE) cancer cells or control cancer cells, which were plated at 1.5 × 105 cells/well in a 1 μm permeable membrane transwell insert to ensure contact-independent conditions. Doxycycline was added to induce miR-29 expression, and tumor cells were replated every other day. After 3 and 7 days, tumor cells in the inserts were discarded and both non-adherent and adherent bone marrow-derived cells were collected and stained with antibodies for flow cytometry (see below).

Flow cytometry analysis
Excised tumors were processed mechanically and enzymatically (with collagenase I, hyaluronidase IV, and DNAse IV; Sigma) into single cells using a gentleMACS dissociator (Miltenyi), and RBCs were lysed. Spleens were processed as described above. Single tumor and spleen cells were stained with cell surface antibodies in FACS buffer (1x PBS + 2% FBS + 1mM EDTA), fixed in 1% paraformaldehyde, permeabilized, and then stained with intracellular antibodies. The antibodies and their dilutions are listed in Supplementary Table S2 (31-color panel in total). The samples were acquired on a 5-laser Cytek Aurora machine. Using FlowJo software (RRID: SCR_008520; v10.10.0), downsampling was performed on the live/CD45+ gated population with a 10,000-cell cut-off. This process excluded one tumor, which had too few CD45+ cells. The downsampled files were then concatenated into a single file, and FlowSOM (RRID:SCR_016899) was utilized to generate 8 clusters with the following markers: CD3, CD4, CD8, CD11b, CD11c, CD14, CD49b, F480, LAG3, Ly6C, MHCII, and PD1. Clusters were defined based on the expression patterning of these markers to the best of our ability, and the percentage of each cluster was then compared between groups. Canonical gating strategy was also used to specifically pull out individual CD8+ and myeloid populations of interest. Specifically, CD44, PD1, CD62L, Ki67, and Granzyme B were used to measure various CD8+ subsets, and F4/80, CD11b, GR-1, MHCII, and iNOS were used to identify total and inflammatory macrophages.
For the T-cell co-culture assay samples, a smaller 17-color panel was used to demarcate CD8 and CD4 T cell subpopulations (no myeloid markers included). FlowJo was used to analyze the processed data. Briefly, CD45+ live cells were then separated into CD3+CD8+ or CD3+CD4+. The markers CD44+PD1+ were used for effector CD8+ T cells, and Ki67+ was used for proliferation. For BMDM co-culture samples, live cells were analyzed for the frequency of CD11b+F480+CD14−Ly6C− populations with FlowJo software. These samples were acquired on a 5-laser Cytek Aurora machine.

Malignant cell FACS
To isolate malignant cells from bulk tumor, tumors were minced and processed into single cells as described above. Tumors were stained with a panel to remove dead and non-malignant cells, including the following markers: Live/dead, CD45, CD31, CD90 and CD19. Cells negative for these markers were then collected into Trizol and processed for RNA extraction. Genes of interest were analyzed by semi-quantitative real-time PCR, as described below.

miRNAscope HD Red assay
Tumors were formalin-fixed, then paraffin embedded and sectioned by the Cancer Tissue Pathology Shared Resource at Emory University following a standard protocol. Using the miRNAscope HD Red Assay and following the manufacturer protocol (ACD), tumor sections were specifically stained for miR-29a (Catalog number 888461-S1). A scrambled, non-targeting probe (Catalog number 727881-S1; ACD) was used as a negative control. After coverslip mounting, slides were scanned using an Olympus NanoZoomer, and brightfield whole-slide images were analyzed in QuPath (v0.5.0). Images were set to Brightfield H-DAB with color deconvolution stain vectors for hematoxylin (0.65111, 0.70119, 0.29049) and DAB (0.26917, 0.56824, 0.77759) with background 255, 255, 255. Tissue regions were automatically annotated using a pixel classifier and then constrained by a 50 μm inward dilation to exclude slide edges. Necrotic regions were then excluded by applying a secondary pixel classifier that applied a mean RGB color transformation followed by a fixed intensity threshold (120) to distinguish necrotic from viable tumor, leaving only non-necrotic tumor for analysis. Within the viable tumor, cells were segmented using Watershed Cell Detection with a pixel size of 0.2299 μm, background radius 8 μm, sigma 1.5 μm, cell expansion 5 μm, and a cell area range of 10–400 μm2. Per-cell DAB optical density (OD) mean intensity values were recorded. Thresholds for DAB positivity were determined using histograms of negative control slides, applying the 99th percentile of background distribution and verifying by threshold sweeping in R (v4.3.2). A final cutoff of 0.15 DAB OD mean was selected and applied to images in QuPath. Results were exported for downstream statistical analysis.

Immunohistochemistry
Paraffin-embedded tumor sections were processed following a standard protocol. Briefly, following heat-mediated antigen retrieval and blocking with 5% goat serum, the primary antibodies were added to tissue at concentrations outlined in Supplementary Table S2. The collagen stain was performed using a Masson’s Trichrome staining kit (Abcam). Images were taken on a Keyence BZ-X810 widefield microscope and quantified using ImageJ software. Images were deconvoluted, converted to black and white images, and a threshold was applied. The positive pixels were measured as a percentage of the whole image. For the CD8 stain, ImageJ software was used to quantify the number of positive cells per field of view (FOV), avoiding areas of necrosis. Image deconvolution was used to view DAB staining separately from the nuclear stain, and a threshold was applied to all images. The “analyze particles” feature was used to quantify the number of positively stained cells per FOV. All FOVs captured for the analysis were intratumoral regions, not peripheral, and we also excluded any areas of obvious necrosis.

Western blot
For the conditioned media sample collection, cells were plated in a 6 well plate at equal cell numbers, grown overnight to ~95% confluency, then the media was changed from complete media to serum free media and cells were incubated for an additional 24 hours. The media was then collected, and 1 mL was processed using cold acetone precipitation (4 mL acetone to 1 mL media). Samples were stored at −80°C overnight. The samples were then centrifuged at 14,000 rpm for 10 min at 4°C and precipitated protein was resuspended in RIPA buffer. Protein from adherent cells was collected in 1X RIPA buffer (10x from Cell Signaling) containing protease cocktail at 1X (100x from Cell Signaling, #5871S) and phosphatase inhibitors at 1X (100X cocktails II and III from Sigma, # P5726 and P0044, respectively). Protein from snap frozen tumors was extracted by dropping a small chunk of tumor into a pre-chilled glass tissue grinder containing 1X RIPA buffer plus the protease and phosphatase inhibitors described above and grinding until the chunk was broken down. Tumor and cell culture samples were sonicated and centrifuged, the supernatant was collected, and these lysates were quantified using a Bradford assay (Thermo Fisher). Lysates (between 30–50 μg total) and conditioned media samples were boiled for 5 min at 95°C in 1x Laemmli buffer. The samples were run on polyacrylamide gels (10%) and transferred to nitrocellulose membranes. Membranes were blocked in 5% non-fat dry milk and antibodies were diluted in 1%BSA/TBST/Sodium azide solution and incubated overnight at 4°C (antibodies and dilutions listed in Supplementary Table S2). Anti-Rabbit secondary antibody (Table S2) was added the next day after washes, and signal was observed using ECL reagents (Prometheus) and a BioRad Chemidoc machine. Densitometry analyses were performed using ImageJ software (RRID: SCR_003070).

Cytokine/Chemokine analyses
Soluble factors present in tumor lysates were initially analyzed by the Proteome Profiler Mouse XL Cytokine Array (R&D Systems), following the manufacturer protocol. 200 μg of tumor lysate was used, and the membranes were imaged on a BioRad Chemidoc machine. To follow up on potential hits from the cytokine array on multiple tumors per group, we utilized commercially available ELISAs specific for mouse CXCL9, CXCL10, CXCL11, CD93, IL1ra, CD105 (Catalog numbers in order: DY492, DY466, DY572, MCD930, MRA00, MNDG00; R&D Systems). 400 μg/ml of lysate was used for each sample. The concentration of protein in each sample was calculated from the standard curve for each individual ELISA, as outlined by the manufacturer instructions. The concentration of LPA was also measured using ELISA (Catalog number LS-F25111, LSBio). Chemiluminescent dot blots were converted to 8-bit grayscale images and background-subtracted in ImageJ/Fiji using a 50-pixel rolling-ball radius. A custom Fiji macro automated spot quantification by recentering user-defined spot locations to local intensity maxima within an 8-pixel radius, generating 17-pixel diameter circular ROIs, and measuring mean intensity and integrated density. Duplicate spots were averaged, negative control signal was subtracted, and normalized intensities were used to calculate fold changes between the two conditions.

WST-1 assay
The 344SQ PD1R1-miR-29a/b1 and control cells were plated at a density of 1,000 cells/well in multiple 96-well plates (one per timepoint). Half of the cells were plated in media containing doxycycline (2 μg/ml). After 24 hours of incubation, WST-1 reagent (Catalog number 102963–852, VWR) was prepared by diluting 1 to 10 in normal growth media. One 96 well plate was removed and 100 μl of the diluted WST-1 reagent was added to each well. Wells containing only growth media (no cells) were used as a negative control. The cells + WST-1 mixture was then incubated for 1 hour at 37°C, at which point the absorbance was read at 450nm on a BioTek H1M plate reader. These steps were repeated for 48 and 72 hour timepoints. The average absorbance value for the 344SQ-control cells at 24 hours (no dox) was used to normalize across all groups over time. This experiment was repeated with the 344SQ parental, non-targeting control (NTC), and miR-29 KO cell lines, and these values were normalized to the 344SQ 24 hour absorbance value.

3-D assay
Single cells were seeded on a matrix comprised of a Matrigel (BD-Biosciences) plus collagen type 1 (Corning) 50:50 mixture, with a final collagen concentration of 1.5 mg/ml. Media was replenished every other day with doxycycline for the 344SQ PD1R1-miR-29a/b1 and control cell lines. Sphere structures were imaged every other day using an inverted Olympus IX51 microscope, and ImageJ was used for the analysis. 3-D structures were outlined, and morphological analysis features were used to measure the area of the structure as well as the circularity, which is a measure for the degree of invasiveness (a perfect circle = 1, whereas a line is 0). The more invasive protrusions in a structure, the smaller the circularity value.

Semi-quantitative real-time PCR
RNA was extracted from cells in culture using Trizol Reagent (Thermo Fisher) following the recommended protocol. From tumor samples, RNA was extracted using the mirVana RNA Isolation Kit (Thermo Fisher). All RNA samples were quantified, and reverse transcription was performed with 2 μg of RNA using qSCRIPT cDNA SuperMix (Quantabio). Real-time PCR was performed using primer sets specific for each gene (sequences listed in Supplementary Table S1) and the SYBR® Green PCR Master Mix (Life Technologies). L32 (60S ribosomal gene) was used to normalize expression across samples. Fold change was calculated using the 2^−ΔΔCT method. For analysis of miRs, the Taqman MicroRNA Reverse Transcription Kit was used (Thermo Fisher) following manufacturer guidelines. The primer–probe pairs for miR-29a-3p, miR-29b-3p, and miR-16-3p were also purchased from Thermo Fisher. Real-time PCR was performed using these specific sets and Taqman Universal PCR Master Mix (Thermo Fisher), and expression of miR-29a/b was normalized to miR-16 for each sample, and fold change was calculated using the 2^−ΔΔCT method. All samples were run on a 7500 Fast qPCR machine (Applied Biosystems).

Luciferase assays
The pmir-GLO plasmid was used to clone the wildtype and mutant versions of the 3’UTR of ATX containing the miR-29 predicted binding site as described above (see Plasmid cloning). These plasmids (0.8 μg) were co-transfected into 344SQ or PD1R1 cells, plated within a white 96-well plate, with 30pmol of miR-29a, miR-29b, or negative control miR mimics (see miRNA mimic/antagomir transfection) using Lipofectamine 3000. The manufacturer recommended protocol was used. After 24 hours, the luciferase activity was measured using the Dual-Glo Luciferase kit (Promega). Firefly and Renilla luciferase readings were measured on a BioTek H1M plate reader, and the Firefly measurement was normalized to the background for each replicate (Renilla).

TCGA data analyses
Micro-RNA and RNA expression data from The Cancer Genome Association (TCGA: https://www.cancer.gov/tcga, RRID:SCR_003193) Lung Adenocarcinoma project was retrieved using TCGAbiolinks(29, 30). Cases with publicly available miRNA and RNAseq were retained, for a total of 510 patients. For one set of analyses, miR-29a and miR-29b expression was correlated with ENPP2/ATX and broadly with all other transcripts in the RNAseq dataset. These data were analyzed using Spearman correlation between miR-29 and each mRNA. For analysis of miR-29 high versus low tumors, a composite score was calculated for miR-29a-3p and miR-29c, assigning miR-29a-3p a weight of 1 and miR-29c a weight of 2 (2*miR-29c + miR-29a-3p). Other family members, notably miR-29b and isoform miR-29a-5p, had overall low expression and were excluded from the composite score. A composite score greater than or equal to 20,000 was considered high and a composite score less than or equal to 14,000 was considered low. The high group contained 190 samples, while the low group contained 177 samples. The middle range, containing 143 samples, was excluded. Expression of CD8A and CD14 were individually compared across the groups, with significance determined using a two-sided Wilcoxon rank sum test. Analysis was run in R (v4.3.3, RRID:SCR_001905). CIBERSORT (v0.1.0, RRID:SCR_016955) was run in R using the leukocyte gene signature matrix (LM22) on the transcripts per million (TPM) normalized TCGA lung adenocarcinoma (LUAD) expression data, split into high and low by composite miR-29 score (as above). LM22 was developed with CIBERSORT and includes 22 immune cell types, the fractions of which are estimated for each sample(31). One immune population (naïve CD4 T cells) is excluded as the normalized expression was 0 in all samples. Immune cell fractions between miR-29_high and miR-29_low were compared using the Wilcoxon rank-sum test. The results were corrected for multiple testing with the Benjamini–Hochberg FDR, and the log2 fold-change was calculated. For survival analyses, the miRNA pan-cancer plotter on KMplot.com was used(32–34), with lung adenocarcinoma specifically chosen (n = 513). A composite score of miR-29a, -29b, and -29c was used with each equally weighted, and the patients were split into 2 groups (high and low) using the auto select cutoff based on percentile. The CD8 enriched or CD8 decreased patients were specifically analyzed for overall survival in miR-29a/b/c low versus high groups.

Statistical considerations
Unpaired students two-tailed t-tests were performed with two comparisons and one-way ANOVA for comparisons with 3 or more groups with a Dunnett correction for multiple comparisons, unless stated in the legends. For tumor growth curves with two groups, multiple t tests (1 per timepoint) were used with a Holm-Sidak correction method for multiple comparisons. For tumor growth curves with multiple comparisons, a mixed-effects model with Geisser-Greenhouse correction compared across tumor groups at each timepoint. A Tukey’s correction was used for multiple comparisons. A p-value of <0.05 was considered statistically significant. Error bars represent standard deviation around the mean unless noted. All analyses were performed in GraphPad Prism (RRID: SCR_002798; v.10.5.0) unless noted.

Data availability
The scRNA-seq files have been deposited to the NCBI Gene Expression Omnibus (GEO) database (RRID:SCR_005012), accession number GSE285606. Additional raw data is available upon request from the corresponding author.

Results

Single-cell transcriptomics reveal miR-29 as a potential predictor of ICI resistance.
To discover underlying regulators of ICI resistance, we performed scRNA-seq on tumors generated from KP mutant 344SQ parental and 344SQPD1R1 (anti-PD-1 resistant) models (referred to as PD1R1 from here on) (Fig 1A). The PD1R1 model was previously defined as intrinsically resistant to anti-PD-1 treatment in vivo (7). Tumors were collected for sequencing at two timepoints: timepoint 1 (~week 4) represents an early phase of tumor development, whereas timepoint 2 (~week 6) represents a later stage and is typically when 344SQ tumors have developed resistance to anti-PD-1(35). We combined these timepoints for the analysis to increase the total cell number, providing analysis of 26,110 high-quality cells, which revealed seven major cell subsets (Fig 1B; Supplementary Fig S1A–C).
We focused on the malignant-cell cluster and analyzed DEGs between 344SQ and PD1R1 tumors (Fig 1C). A total of 759 and 623 significant DEGs were found between models at the early and late timepoints, respectively (Supplementary Table S3). Upstream regulator analysis of malignant-cell DEGs at the early timepoint provided model-dependent tumor-promoting pathways versus highly necrotic and stressed tumors often observed at a later timepoint. This analysis identified Hlx, N-myc, Hoxa10, and Foxo1 as activated, whereas miR-29, Hmgb1, Tp73, Ncor1, Irf2, and Cepba were predicted to be suppressed (Supplementary Table S4). The expression of these predicted malignant-cell regulators was analyzed in 344SQ and PD1R1 cell lines and only Hlx was found upregulated, whereas miR-29a, Trp73, Irf2, and Cepba were downregulated, as predicted (Supplementary Fig S1D). However, when assayed across a broader panel of PD1R models, only miR-29 (both miR-29a and miR-29b) and Hlx trends remained consistent (Fig 1D; Supplementary Fig S1E). The other family member, miR-29c, was not expressed in these models, so it was not included in functional analyses. As disruption of single miRNAs can modulate hundreds to thousands of genes, we explored miR-29a/b moving forward. Specific miR-29 target genes in the scRNA-seq malignant-cell DEGs dataset were annotated with name and denoted with red circles (Fig 1C). We also confirmed miR-29a downregulation in 344SQ tumors treated with anti-PD-1 over time (Supplementary Fig S1F–G), supporting findings from the PD1R models.
The miR-29 family controls hundreds of transcripts, which regulate pathways involved in tumor progression via diverse mechanisms (18, 36). To demonstrate functional loss of miR-29 in the PD1R models, we analyzed expression of known miR-29 target genes versus the 344SQ parental line. Of the 30 miR-29 target genes analyzed, half were upregulated in the PD1R1 model (Fig 1E), and most were also upregulated in one or two additional PD1R models (Fig 1F), providing evidence the miR-29 axis was partially disrupted with anti-PD-1 resistance. While Hlx levels were significantly different between PD1S and PD1R models, as predicted by the scRNA-seq analysis, we found no similar trend in Hlx target gene expression in these models. Therefore, we further studied the role of the miR-29 transcriptional axis in ICI resistance.

The immunosuppressive enzyme ATX is a miR-29 target gene.
Our group has demonstrated upregulation of the ATX phosphodiesterase enzyme with acquisition of anti-PD-1 resistance(7). However, mechanisms driving this were previously undetermined. In silico analysis of the 3’UTR of both human and mouse ATX revealed a predicted miR-29 binding site, which was the only conserved miRNA predicted site between both species (Supplementary Table S5). This was also confirmed in a recent publication(37). To determine if miR-29 regulates ATX in lung cancer, we assayed expression of miR-29a/b and ATX in human lung cancer cell lines, finding a significant inverse correlation (Fig 2A). This trend was also shown in a similar analysis on TCGA LUAD dataset with ~500 patients (Supplementary Fig S2A). Together, these data support an inverse association between expression of miR-29 and ATX in human lung cancer cell lines and patient tumors. To demonstrate a causal effect between miR-29 expression and ATX transcriptional repression, we transiently transfected miR-29a/b mimics into cells and found that either miR-29a or miR-29b downregulated ATX RNA and protein in two PD1R models (Fig 2B–C; Supplementary Fig S2B), although in human cell lines, only the miR-29a mimic achieved ATX repression (Supplementary Fig S3). Conversely, treatment with miR-29a or miR-29b antagomirs increased ATX expression in parental 344SQ and the PD1S1 models (Fig 2D–E, Supplementary Fig S2C).
While the data were supportive of miR-29 regulating ATX, we wanted to confirm a direct interaction via the 3’ UTR of ATX. Using a portion of the 3’UTR of ATX containing the predicted miR-29 binding site, we performed a luciferase assay with co-transfection of miR-29a or miR-29b mimics, observing significant downregulation of luciferase activity as compared to the negative control mimic (Fig 3A). To further support these findings, we mutated 4 of the 7 residues within the predicted miR-29 binding site in the ATX 3’UTR (Fig 3B) and transfected cells expressing either the wildtype or mutant version with the miR-29 mimics. While the mimics downregulated luciferase activity when co-transfected with the wildtype version of the 3’UTR, the mutant version completely abrogated this effect (Fig 3C). These data provide evidence that ATX is a bona fide miR-29 target gene, and this regulation occurs via a single site within the 3’UTR.

Manipulation of miR-29 expression in KP lung cancer cells impacts cellular invasion and growth in vitro.
Next, we genetically manipulated miR-29 expression to determine if it directly modulated ATX expression, other miR-29 gene targets in vivo, and the tumor immune microenvironment. Re-expression of the miR-29a/b1 cluster in the PD1R1 tumor model via a doxycycline-inducible vector was confirmed (Fig 4A), restoring levels to those observed in PD1S models (Supplementary Fig S4A), and resulted in reduced ATX mRNA expression (Fig 4B). Transcriptional downregulation of ATX also decreased secreted ATX protein and LPA levels in the conditioned media of the PD1R1-miR-29 versus control cells (Supplementary Fig S4B–C). Several additional miR-29 targets were also repressed in PD1R1 cells with miR-29 re-expression (Supplementary Fig S4D), supporting functional expression of miR-29 in these cells. Conversely, loss of miR-29a in parental 344SQ cells resulted in increased expression of ATX, upregulation of additional miR-29 target genes compared with control cells, and elevated LPA levels (Supplementary Fig S4E–H).
Using these isogenic models, we queried the impact of miR-29 expression on cellular growth and invasion. Using 2-D and 3-D in vitro cultures, we found a small but significant effect of miR-29 expression on cellular growth, with its expression negatively correlating with growth (Supplementary Fig S4I–J). Additionally, in the 3-D model, we assayed invasiveness of structures over time by measuring morphological features, specifically structure circularity. miR-29a/b1 re-expression significantly increased structure circularity, indicative of a less invasive phenotype versus control (Supplementary Fig S4K); however, KO lines had no effect on invasion (Supplementary Fig S4L), suggesting its loss alone is not sufficient to induce an invasive phenotype.

miR-29 re-expression reprograms the tumor microenvironment in PD1R tumors.
Using PD1R1-miR-29a/b1 re-expression cells, we performed an in vivo study to determine miR-29-dependent effects on tumor growth, metastasis and the tumor microenvironment (TME). Our data indicated little overall effect on primary tumor growth due to miR-29 re-expression, though there were 2 complete regressions in this group versus none in controls (Fig 4C–D). Additionally, there was a significant repression in the metastatic propensity of these tumors (Fig 4E), supporting previous findings that miR-29 can control transcripts involved in tumor cell motility and invasion(38). Tumors from this experiment were then analyzed for expression of ATX and other miR-29 targets. Consistent with the in vitro data, ATX was repressed with miR-29 re-expression at the RNA and protein level (Fig 4F–G; Supplementary Fig S5A), with a trend toward reduced LPA production (Supplementary Fig S5B). Several additional miR-29 target genes were also repressed in these tumors, including Cd276, Hdac4, Snail, and Pdgfa, whereas mRNA levels of the ECM-related genes Col1a1 and Lamc1 showed no trend or were increased in miR-29 tumors when examining bulk tumor RNA (Fig 4H), although tumor cell–specific transcript levels of these same genes were reduced, as we expected (Supplementary Fig S5C). This suggests additional cell types within heterogenous tumors contribute to transcript levels of these genes and overcompensate for miR-29 reduction in tumor cells. To further define the ECM signature within these tumors at the functional level, we analyzed collagen deposition by Trichrome staining. In contrast to bulk tumor mRNA levels, there was a significant repression in collagen fibrils within those tumors with miR-29 re-expressed compared to control tumors (Fig 4I). Conversely, the miR-29 KO tumors when implanted in vivo did not show a significant difference in tumor volume compared with controls but demonstrated an increase in collagen deposition as shown by Trichrome staining (Supplementary Fig S5D–E). RNAscope staining confirmed miR-29 loss was stable in KO tumors (Supplementary Fig S5F).
Lastly, we generated a rescue of ATX expression in miR-29 cells lacking the 3’UTR and thus was resistant to miR-29 degradation (Supplementary Fig S5G). An in vivo study revealed ATX rescue was sufficient to increase tumor growth of miR-29 expressing cells, although due to tumor regression with the miR-29_Ctrl model, the replicates were limited (Supplementary Fig S5H–I). However, this suggests ATX may be a primary transcriptional change downstream of miR-29 loss in our models. Together, these data indicate the manipulation of miR-29 has functional consequences within the TME of PD1R tumors, with its loss derepressing expression of key target genes, including ATX, and regulating fibrotic deposition within the tumor.

miR-29 expression correlates positively with CD8+ T cells and pro-inflammatory cytokine signatures.
We next queried whether miR-29 expression impacts the immune microenvironment in PD1R tumors. Because ATX negatively regulates infiltration and CD8+ T-cell activation within tumors(7), we questioned whether miR-29 re-expression would have the converse effect, altering the tumor immune microenvironment to promote antitumor immunity, in part by diminishing ATX levels. We first analyzed tumor RNA extracted from miR-29 re-expressing or control tumors for CD8 expression, finding a ~2-fold increase in CD8A transcript levels in tumors with induced miR-29 expression versus controls (Fig 5A). These findings were corroborated via IHC staining for CD8+ T-cell infiltration (Fig 5B). We also found a trend towards higher Granzyme B staining in miR-29 overexpressing (OE) tumors versus the control tumors (Fig 5C). Thus, in tumors with higher miR-29/lower ATX expression, there was an influx of cytotoxic CD8+ T cells that contribute to a pro-inflammatory microenvironment.
To further characterize broad changes occurring within the TME due to miR-29 re-expression, we analyzed cytokines/chemokines using a proteome array that detects 111 soluble factors. Proteomic profiling identified increased CXCL9/10/11 and decreased IL-1ra, CD93, and CD105, in miR-29 expressing tumors (Fig 5D; Supplementary Table S6), with selected findings validated by ELISA (Fig 5E). Each of these factors has complex roles, with many different cell types contributing to the extracellular pool of soluble factors within a heterogeneous TME. To evaluate tumor cell–intrinsic effects on expression of these factors, we assayed transcript levels of each of the 4 hits from the array. Two transcripts, Cd93 and Cd105, were downregulated within miR-29 expressing tumor cells in vitro, suggesting these may be altered within tumors in a tumor cell–intrinsic manner (Fig 5F). However, Cxcl9 was downregulated, whereas there was no change in Il1ra. Together, these data suggest that CD93 and CD105 may be modulated directly by miR-29 expression in tumor cells, whereas CXCL9 and IL1ra changes within the TME may occur in an indirect manner via recruitment and/or activation of other cell types.

miR-29 re-expression coupled with ICI extends survival by promoting an antitumor immune microenvironment.
Because miR-29a/b1 re-expression promoted pro-inflammatory chemokine/cytokine profiles and CD8+ T-cell infiltration into ICI-resistant tumors(7), we tested whether miR-29a/b1 expression was sufficient to improve response to ICI in these tumors. Using PD1R1 miR-29a/b1 versus control cells, we began anti-PD-1 treatment after doxycycline-mediated initiation of miR-29 expression and monitored tumor growth over time (Fig 6A). We observed the control cells were responsive to PD-1 blockade therapy, unlike their unmanipulated parental cells (as previously demonstrated(7)). Despite this, miR-29a/b1 re-expression plus anti-PD-1 improved tumor growth control (Fig 6A), with a significant extension of OS in mice bearing miR-29 tumors, which was further enhanced by anti-PD-1 therapy (Fig 6B). The median survival reached 73 days in miR-29-expressing tumor-bearing mice treated with anti-PD-1, compared to 45 days in controls. In fact, 3 of 6 mice demonstrated complete tumor regressions with anti-PD-1 treatment in the miR-29 group.
Because the in vivo phenotype of the PD1R model was impacted by lentiviral manipulation, we repeated the above study in the parental 344SQ model, which initially responds, but later acquires resistance to PD-1 therapy (35). With these isogenic models, we investigated the immune landscape in ICI-treated tumors as a function of miR-29. After 1 week of treatment, we found a reduction in tumor size with PD-1 blockade, particularly in miR-29-expressing tumors, with 1 complete regression (Fig 6C–D). Remaining tumors were immunophenotyped using multiparameter flow cytometry. Comparing bulk CD45+ compartments revealed miR-29a/b1 tumors had a robust increase in total immune infiltrate especially in the face of ICI, with CD45+ cells encompassing ~85% of all live cells within these tumors (Fig 6E). To delineate specific subpopulations, we performed cluster analysis of all CD45+ cells using a panel of lineage and phenotypic markers (Supplementary Fig S6A). This analysis revealed a significant increase in CD8+ T cells in miR-29 tumors at baseline, which was further increased with anti-PD-1 (Fig 6F, Supplementary Fig S6B), corroborating data from PD1R1-miR-29a/b1 tumors (Fig 5A–B). Conversely, clustering analysis also demonstrated that the major cluster of CD45+ cells in PD1R1 control tumors was a monocytic cluster, characterized by CD14, CD11b, and Ly6C expression, and this cluster increased in frequency with anti-PD-1 (Fig 6F, Supplementary Fig S6C). This result aligns with previous data from our group implicating monocyte infiltration as a potential ICI-resistance mechanism(39). Moreover, this monocytic cluster was significantly reduced within miR-29 re-expression tumors, which was compounded with ICI therapy (Fig 6I).
Because several myeloid clusters had transitional state markers and thus were difficult to classify, we also analyzed the myeloid subset with additional markers using flow cytometry gating. Among myeloid populations examined, including dendritic cells and myeloid-derived suppressor cells (MDSCs), no significant differences were observed. In contrast, miR-29-expressing tumors treated with anti-PD-1 displayed ~2-fold increase in total macrophages, with a significant enrichment of macrophages exhibiting a pro-inflammatory phenotype (Fig 6G–H). To determine whether miR-29–expressing cancer cells influenced macrophage differentiation, we co-cultured BMDMs using cancer cells that either overexpressed miR-29 or not under contactless conditions. At early stages of differentiation (day 3), there was no significant difference in frequency of F4/80+CD11b+Ly6C−CD14− macrophages, although a slight trend was noted (Supplementary Fig S6D). Yet, by day 7, miR-29–overexpressing cancer cells promoted more differentiated macrophages compared with the control, consistent with the increased macrophage abundance observed in vivo.
We also further characterized the CD8+ T-cell cluster to determine functional status. Flow cytometry analysis revealed these cells were more proliferative in miR-29 tumors, which was further potentiated with anti-PD-1 treatment (Supplementary Fig S6E). Additionally, while most CD8+ T cells differentiated into effector cells at this timepoint across groups, this population was further increased with miR-29 expression plus anti-PD-1, with decreased expression of LAG3 and PD-1. Together, these results mirror our previous data demonstrating that targeting the ATX/LPA axis improved cytotoxic T-cell phenotypes and increased pro-inflammatory macrophages (7).
To support these data, we also performed an in vitro co-culture assay with 344SQ miR-29 KO cells and naïve immune cells to determine whether loss of miR-29 was sufficient to impact immune-cell differentiation and activation. Compared to non-targeting controls, miR-29 KO cells caused significant changes to immune phenotypes. There was a significant decrease in total CD45+ immune cells, particularly the CD8+ T-cell population, and additionally, these cells were less proliferative and effector differentiated (Supplementary Fig S6F). These data indicate miR-29 loss in an anti-PD1-sensitive model is sufficient to re-wire the immune microenvironment towards a pro-tumor phenotype. This is supported by the lack of response to anti-PD-1 treatment in the miR-29 KO tumors versus the control tumors in vivo (Supplementary Fig S6G).

Transcriptional correlates with miR-29 reveal an enrichment in immunoregulatory pathways.
To extend the translational relevance of these findings, we performed analyses on LUAD patients from TCGA database for which both miRNA and mRNA-sequencing data were available. Patients were categorized into high and low expression of tumoral miR-29a plus miR-29c, which were highly correlated with each other (Supplementary Fig S7A). The level of miR-29b expression was significantly lower than that of the other 2 family members in these samples and was thus excluded from this analysis. We then assessed CD8A expression in miR-29a/c-low versus -high tumors and found significantly higher expression of CD8A in the patients with high miR-29a/c (Fig 7A). We also analyzed the monocyte marker CD14 and found a significant reduction in CD14 expression in tumors with high miR-29a/c expression (Fig 7B). These findings support our flow cytometry data demonstrating a positive correlation between CD8 and miR-29, whereas miR-29 repressed monocytic cells.
To further expand our understanding of miR-29 transcriptional regulation in lung cancer patients in an unbiased manner, we broadly correlated gene expression with miR-29a expression, applying a cutoff of ±0.2 correlation coefficient. This provided a gene set of ~1,400 mapped genes that positively or negatively correlated with miR-29a (Supplementary Fig S7B; Supplementary Table S7). Gene set enrichment analysis (GSEA) then examined gene ontology (GO) biological processes enriched in this dataset. The top 10 of these GO terms were involved in immune pathways, including adaptive immune response, leukocyte migration, and alpha-beta T-cell activation, while negatively enriched pathways included several involved in cell cycle regulation (Fig 7C)(40). Specific immunoregulatory genes involved in positively enriched pathways included NFATC1, IL12B, and CD40LG, among others (Supplementary Fig S7C). miR-29 and CD276/B7-H3 were also inversely correlated in these LUAD patients, supporting a previous study demonstrating B7-H3 is a miR-29 target (21). Along these lines, miR-29-high versus low tumors had significant changes to several immune populations, as predicted by CIBERSORT, including an increase in plasma cells and decrease in macrophages (Supplementary Fig S7D). Finally, patients with low miR-29 expression had worse OS than patients with high expression (Fig 7D). Although this trend was significant in tumors with enrichment in CD8+ T cells, it was more pronounced in tumors with a decreased CD8+ T-cell signature. Together, these data indicate miR-29 loss contributes to worse OS in patients with lung cancer via broad dysregulation of the antitumor immune response.

Discussion
The identification of ICI resistance mechanisms is a critical need for lung cancer research. Our previous data indicated ATX is aberrantly upregulated in models with anti-PD-1 resistance(7), although the mechanism remained unclear at the time. Herein, based on our scRNA-seq data comparing DEGs between a sensitive line (344SQ) and an anti-PD-1 resistant line (PD1R1), both HLX and miR-29 were predicted as upstream regulators for genes in our dataset, although analysis of a single timepoint is a limitation of this study. In silico analysis indicated that miR-29 has a predicted binding site within the 3’UTR of both human and mouse ATX transcripts, which made this hit of particular interest. We confirmed that miR-29 and ATX expression were inversely correlated in human lung cancer samples and that both miR-29a and -29b directly bound to the 3’UTR of ATX, suppressing its expression post-transcriptionally. These data are supportive of a recent publication demonstrating ATX is regulated by miR-29c, although that study identified a trans regulation by macrophage-derived exosomal miR-29c taken up by and suppressing ATX in melanoma cells(37). Our study differs in that regulation of ATX by miR-29a/b occurred in cis within lung cancer cells, although we cannot discount that microenvironmental exosome uptake may also influence gene expression in PD1R models.
Because dysfunction of a single miR impacts expression of hundreds of genes, we explored additional miR-29 target genes that were aberrantly activated within PD1R versus PD1S models. Both miR-29a and miR-29b had ~1/2 the expression in resistant models; thus, this loss would cause a multitude of downstream alterations with both tumor cell–intrinsic and –extrinsic consequences. Indeed, many miR-29 target genes were upregulated in PD1R models besides ATX, including ECM-related genes like Col6a1, Col6a2, and Lamc1, which in turn was associated with increased matrix deposition. This increased deposition could have an indirect effect on immune-cell infiltration into tumors by acting as a physical barrier. Additionally, other miR-29 targets were dysregulated in PD1R models with potentially pleiotropic effects within the TME including IGF-1, which has previously been shown to promote expansion of regulatory T cells to promote immune tolerance(41). Several soluble factors were also differentially expressed in miR-29–expressing tumors, including upregulation of CXCL9 and downregulation of LPA, IL-1ra, CD93, and CD105. Each of these have complicated effects on the TME and can be produced by multiple cell types including stromal cells. We hypothesize that downregulation of LPA caused directly by miR-29 activity, and/or other factors like CD93 and CD105, promote pro-inflammatory macrophage differentiation, which in turn secrete CXCL9 into the TME. Together, these direct and indirect effects of miR-29 expression facilitate CD8+ T-cell recruitment and proliferation, ultimately improving response to PD-1 blockade. However, how miR-29 may regulate these factors, and which factors are primarily mediating the effects on macrophages and CD8+ T cells, is unknown and the focus of future studies. We will continue to probe the effects of these secreted factors on the heterogeneous TME, the mechanism of miR-29–dependent regulation, and how they may cooperate to promote immune dysfunction and PD-1 therapy resistance.
While the data presented herein indicate dysfunction of the miR-29 axis in PD-1 blockade resistant models, the mechanism(s) of miR-29 repression have yet to be elucidated in this context. It will be crucial to understand how miR-29 is lost within PD-1 resistant models, as this may indicate an upstream targetable factor. Data in the literature indicates that miR-29 can be negatively regulated by TGF-β(42), c-Myc(43), Yy1/NF-κB(44), and positively regulated by Cepba(45). Our preliminary data indicate that TGF-β is in fact upregulated in PD1R cells at the transcriptional level; therefore, we hypothesize that an aberrantly active TGF-β cascade is causing miR-29 repression among other effects, although future studies will need to explore this hypothesis in greater detail.
To interrogate miR-29 in human samples beyond NSCLC cell lines, we analyzed publicly available LUAD datasets through TCGA, which revealed a negative enrichment in cell cycle–related pathways in samples with high miR-29 expression. These included pathways involved in chromosome segregation and DNA replication. We have yet to explore the relevance of cell cycle pathways in the context of PD1R models; however, we found no growth disadvantage with miR-29 re-expression in PD1R models, as evidenced by the similar primary tumor growth between these and control cells in vivo. However, further examination of the cellular fitness of the PD1R models due to miR-29 loss and its impact on response to ICI is still needed. On the other hand, we found a positive enrichment in pathways related to immunoregulatory mechanisms with miR-29 expression in LUAD patients, including the adaptive immune response, leukocyte migration, and T-cell activation. These data suggest tumors with high miR-29 expression and low ATX expression have enhanced antitumor immunity; however, these correlative analyses need to be supported by future work that interrogates a miR-29–dependent immunoregulatory gene signature in ICI-responders versus non-responders to determine if this axis influences and/or predicts response to ICI in lung cancer patients.

Supplementary Material
1234567891011121314