Cancer cells restrict immunogenicity of retrotransposon expression via distinct mechanisms

Abstract

To thrive, cancer cells must navigate acute inflammatory signaling accompanying oncogenic transformation, such as via overexpression of repeat elements. We examined the relationship between immunostimulatory repeat expression, tumor evolution, and the tumor-immune microenvironment. Integration of multimodal data from a cohort of pancreatic ductal adenocarcinoma (PDAC) patients revealed expression of specific Alu repeats predicted to form double-stranded RNAs (dsRNAs) and trigger retinoic-acid-inducible gene I (RIG-I)-like-receptor (RLR)-associated type-I interferon (IFN) signaling. Such Alu-derived dsRNAs also anti-correlated with pro-tumorigenic macrophage infiltration in late stage tumors. We defined two complementary pathways whereby PDAC may adapt to such anti-tumorigenic signaling. In mutant TP53 tumors, ORF1p from long interspersed nuclear element (LINE)-1 preferentially binds Alus and decreases their expression, whereas adenosine deaminases acting on RNA 1 (ADAR1) editing primarily reduces dsRNA formation in wild-type TP53 tumors. Depletion of either LINE-1 ORF1p or ADAR1 reduced tumor growth in vitro. The fact that tumors utilize multiple pathways to mitigate immunostimulatory repeats implies the stress from their expression is a fundamental phenomenon to which PDAC, and likely other tumors, adapt.

Keywords: ADAR1; Tp53; cancer evolution; cancer immunity; inverted Alus; retrotransposons; tumor-immune microenvironment.

Figure 1.. LINE-1 (L1) retrotransposition activity is associated with mutant TP53 status and tracked with phylogenetic tumor evolutionary branching

A) Multiregional sampling and multi-omics overview of autopsy PDAC cohort. B) Sample composition: 148 total RNAseq (left) - 117 with matched WES samples. 67 WGS samples - 59 with matched RNAseq that passed QC (right). C) L1 Insertion loci overview of 31 samples from 5 patients with non-zero somatic insertions detected. Track background colored by patient; each dot represents one insertion locus that passed QC (Methods). Position of each dot is proportional to number of samples sharing the same insertion within a patient: dots on the inner margin of each patient indicate the insertion is only found in one sample, dots on the outer margin indicate that the insertion happened in all samples for that patient. D) Boxplot enumerating L1 insertions in samples with different TP53 mutations. Multiple dots per box indicate multiregional samples from the same patient. Predicted TP53 fitness for each patient labeled in the parathesis. E) Oncoprint of SNPs with frequency over 13% across samples from the 5 patients. MPAM01 & MPAM04 have TP53 missense mutations of c.406C>G and c.734G>A, respectively. MPAM05 has a frameshift deletion in TP53 at c.78delT. MPAM06 carries splice site mutation c.560–1G>A, and PAM03 is WT. F) Boxplot of number of somatic L1 insertion in samples with wild type (n=30) or mutant TP53 (n=196) in the PDAC biopsy cohort from the COMPASS trial (n=201). Benjamini-Hochberg corrected p-value of t-test is labeled. G) Integrative analysis of L1 insertion phenotypes and tumor evolution of patient MPAM04. Detected target site L1 integration features are coded as red (target site duplication) and green (target site deletion). Columns are labeled with the chromosome and the genomic coordinates of the L1 target site (e.g. 12_ 86527687_86527678 represents the insertion detected on chr12 with the left breakpoint at 86,527,687, and right breakpoint at 86,527,678).

Figure 2.. SINE elements are co-expressed with RIG-I like receptor associated IFN signaling in PDAC autopsy cohort

A) Pairwise correlation analysis of the expression of 140 interferon related genes (Table S1) in 214 RNA samples. B) KEGG pathway annotation of genes identified in each cluster. Cluster 4 (C4) exclusive pathways are highlighted in yellow. C) Scheme representing the subfamily-based quantification of repeats expression (Methods). Pearson correlation coefficient between median expression of genes in C4 (hereafter RLR signature) and subfamily based repeat expression, stratified by repeat classes: retrotransposons – SINE, LINE, ERV; transposons – DNA; and satellite repeats – SAT across 214 samples. D) Pearson correlation distribution between RLR signature and repeats within different classes stratified by TP53 mutation status. Benjamini-Hochberg corrected p-value of t-test are labeled. E) Pearson correlation between RLR signature and different SINE subtypes. SINE subtypes are colored based on evolutionary age. Evolutionary timeline of SINEs: SINE2/SINE3, SINE1 (Monomeric and dimeric AluJ, AluS and AluY, and SVA). F) Scheme representing the locus-based quantification of repeats expression: rep1_1, rep1_2, rep1_3. Pearson correlation coefficients between RLR signature and locus specific expression of SINEs stratified by evolutionary age groups. G) dsRNA forces calculated for the human genome (hg38) encompassing SINEs (Methods). (*** p.value < 0.0001, **** p.value < 0.00001).

Figure 3.. Elevated LINE1 ORF1 protein expression is associated with high M2 macrophage infiltration

A) Density plots colored by quartiles of samples (n=27, excluding 3 samples from MPAM05 with an extreme number of somatic L1 insertions) with different numbers of somatic L1 insertions. Samples in quartile 4 (Q4) and quartile 1 (Q1) are selected to represent L1 insertion high and low samples. The Volcano plot below shows the differential expression of repeats between samples with high (Q4) or low number (Q1) of somatic L1 insertions. The calculated Log₂FoldChanges have been corrected for tumor purity (Methods). B) Pie chart depicting the patient compositions in terms of tumor grand and clinical stage from two different PDAC cohorts: autopsy samples and resected samples. C) Representative TMA core stained for L1 ORF1p (green), RNA probe for SINEs (Pink), and hematoxylin stain for nuclear (Purple). D) Left - Quantified Alu intensity in pathologically defined tumor glands (n=59) with L1 ORF1p+ and ORF1p- regions (Table S2) (Methods). Right - Scatter plot of quantified L1 ORF1p positive cells and Alu intensity in pathologically defined tumor glands. Each dot represents one core. Pearson correlations are calculated, and a linear regression line is plotted. E) Violin plot of Pearson correlation coefficient between absolute M2 macrophage infiltration estimated using CIBERSORTx from RNAseq data (Table S4) and different repeat expression from 214 samples in the autopsy cohort. Benjamini-Hochberg corrected p-value are labeled (**** p.adj < 0.00001). F) Spearman correlation between the number of somatic L1 insertions and CIBERSORTx inferred myeloid cell infiltration. Red dot indicates a significant correlation between L1 insertion and M2 infiltration. G) Representative images of immunofluorescent staining in the autopsy PDAC cohort stained with the indicated markers (CD163 labels macrophage M2 cells, CD8 labels T cells). H) Quantification of CD163 signals, in 17 autopsy cohort samples with 8–11 scans per sample (Table S4).

Figure 4.. L1 ORF1p suppresses immunostimulatory Alus expression in vitro

A) Western blot to quantify L1 ORF1p expression across 11 PDAC cell lines. High and low L1 ORF1p expressing cell lines were determined based on actin normalization (short e. – short exposure, long e – long exposure). B) Comparison of change in SINEs expression between PDAC patient samples Log₂FoldChanges (high L1 insertions/low L1 insertions) and in vitro cell lines Log₂ FoldChanges (high ORF1p expression/low ORF1p expression). C) RT-qPCR analysis of three Alu subfamilies in three high L1 ORF1p expressing cell lines: PDAC8, 9, 6 and one low L1 ORF1p expressing cell lines PANC-1. D) Immunofluorescent staining with probes against dsRNA (Jr2), L1 ORF1p, and DAPI in L1 ORF1p high (PDAC6, 8, 9) and L1 ORF1p low (PANC-1, KP-4) cell lines (Methods). Intensities of L1 ORF1p green signal and dsRNA red signal are quantified for each cell line in the violin plots. p-values of Wilcoxon test are labeled (*** p < 0.00001, **** p < 0.00001). E) Construction of dox-inducible ORF1p using gate-way cloning method in PANC-1 cell line (Methods). pDONR: Donor vector (pCR^™8/GA/TOPO^™TA), pDEST: Destination vector (pCW57.1). Western blot to verify ORF1p expression upon Dox treatment for 48hrs. Lysate from PDAC6 cell line used as a positive control. Similar expression level of ORF1p was reached in the PANC-1 line when dox concentration is around 0.1 ug/mL. RT-qPCR analysis of three Alu subfamilies in PDAC-1 cell line with and without Dox treatment for 48hrs. F) Short hairpin (sh)RNA sequences targeting for L1 ORF1p. L1 ORF1p knockdown efficiency was verified by western blot and ORF1p intensity was normalized by GAPDH (n=3, biological replicates). Expression of representative Alus that previously confirmed by qPCR in different cell lines, in non-targeting control (NTC) and shORF1–2 KD in PDAC3 and PDAC6 cell lines quantified by RNA-seq. Statistical analysis was performed using t-test in biological triplicates. (G) Phase-contrast images depict 3D tumorspheres derived from the specified cell lines (shNTC or shORF1) cultured for 7 days. Cell viability was assessed using CellTiterGlo assay. Statistical significance was evaluated using a two-way ANOVA followed by Tukey’s multiple comparison test. H) Graphical model shows the repressive effect of L1 ORF1p on immunostimulatory SINEs.

Figure 5.. L1 retrotransposition activity suppresses immunostimulatory SINE expression likely via in-cis binding

A) RNA ImmunoPrecipitation sequencing (RIPseq) experimental setup and computational pipeline for detecting ORF1p and ORF2p binding transcripts in ECC line (ECC) – N2102p. B) Violin plots show the distribution of dsRNA force assigned for each repeat copy (Methods) grouped by whether it is bound by proteins of interest – ORF1p and ORF2p. p-values of Wilcoxon test are labeled (**** p < 0.00001). C) Composition of SINEs bound by both L1 ORF1p and ORF2p. SINE subtypes are colored based on their evolutionary age as in Figure 2D) Heatmap shows the normalized and scaled expression (Z-score) of L1 ORF1p and ORF2p-binding SINEs (colored by evolutionary age and locus on columns) in the RIPseq experiment. E) Left: Fluorescent staining with dsRNA antibody (Jr2), L1 RNA FISH, and DAPI in ECC lines transfected with scrambled control vector or shORF1p vector for KD (Methods & Table S6). Right: Violin plots show intensity of J2 signals in the indicated number of fields are quantified for each cell line. F) Left: Immunofluorescent staining with ORF1p antibody, IRF3p antibody, and DAPI in ECC lines that are transfected with scrambled control vector and shORF1p vector for KD (Methods). Right: Violin plots show intensity of nuclear IRF3p signal quantified for each cell line. G) IL6 RNA FISH in ECC lines that are transfected with scrambled control vector and shORF1p vector for KD (Methods). Number of IL6 mRNAs per cell was quantified for each cell line and shown in violin plots on the right.

Figure 6.. ADAR1 expression and function are dampened in TP53 mutation

Violin plots show the expression of ADAR1 and MDM2 in different TP53 backgrounds in the autopsy PDAC cohort. B) Violin plots show the expression of SINE subgroups in different TP53 backgrounds in the autopsy PDAC cohort. C) A-I editing index for four patients with different somatic mutations of TP53 and number of somatic L1 insertions. Inverted SINEs are SINE copies with an adjacent copy inserted in the opposite direction within 3kb that can potentially form dsRNA when expressed; non-inverted Alu are Alu copies that do not meet these criteria (Methods). D) A-I editing index for the PDAC biopsy samples in the COMPASS trial stratified by different TP53 mutation backgrounds. E) Scatter plot of A-I editing index as a function of predicted TP53 missense mutation fitness. F) Left: Western plot showing ADAR1 and ORF1p expression. Middle: Phase-contrast images depict tumorspheres derived from the specified cell lines treated with the indicated negative controls (shNTC) or shRNAs (shADAR-1 & shORF1p) cultured for 7 days. Right: Cell viability was assessed by CellTiterGlo assay. Statistical significance was evaluated using a two-way ANOVA followed by Tukey’s multiple comparison test. p-values of Wilcoxon test are labeled: * p < 0.05, ** p < 0.01, *** p< 0.001, **** p < 0.0001.

Figure 7.. ADAR1 expression and function are dampened in TP53 mutation.

A) Schematic of strategies tumors leverage to maintain anti-tumorigenic SINE expression in TP53 WT (via ADAR1 editing) and TP53 mutant (via L1 binding) backgrounds. Thickness of the errors representing the activity of the indicated pathways. B) Rank correlation (y-axis) between predicted IFN (Methods) and the real IFN signature measured from RNAseq with simulations of ADAR1 ($w_{AI}\bullet x_{AI}$) and ORF1p ($w_{L1}\bullet x_{L1}$) contributions at different weights ($w_{L1}$) (x-axis). Yellow – weight of ADAR1 ($w_{AI}$), purple – weight of ORF1p ($w_{L1}$), sum of which equals to 1 (details see Methods). C) The z-scores of IFN signature profiles were assessed across various autopsy patient samples (n=37) that were used in simulating the models (minimum of three quantified samples). D) Integrative analysis of IFN signatures and the calculated relative suppression from ADAR1 and ORF1p along tumor evolution in MPAM04, a TP53 mutated patient. Samples are boxed and connected to IFN signature by whiskers with colors representing the tissue type. Pie chart representing the relative suppression of ADAR1 and L1 calculated by the model (Methods) in each sample. E) Integrative analysis of IFN signatures and the calculated relative contribution from ADAR1 and ORF1p along tumor evolution in PAM03 - a TP53 wildtype patient. Samples are boxed with colors representing the tissue type. Pie chart represents relative contributions of ADAR1 and L1 in each sample.