Genomic and cellular responses to aspirin in colonic organoids from African- and European-Americans

Abstract

Background:

Aspirin (ASA) is a proven chemoprotective agent for colorectal cancer (CRC), though inter-individual responses and cellular mechanisms are not well characterized. Human organoids are ideal to study treatment responses across individuals. Here, colonic organoids from African-Americans (AA) and European-Americans (EA) were used to profile genomic and cellular ASA responses.

Methods:

Colonic organoids from 67 participants, 33 AA and 34 EA, were treated with 3mM ASA or vehicle control for 24h. Gene expression was assessed by RNA-seq, and differentially responsive genes analyzed by condition, population and for gene set enrichment. Top differentially responsive genes were assessed by time and ASA doses in independent organoids. Expression quantitative trait loci (eQTL) mapping was performed to identify variants associated with condition-specific responses. Apoptosis and necrosis assays were performed, and apoptosis gene expression measured in organoids.

Results:

Overall, 8343 genes were differentially responsive to ASA with differences between AA and EA. Significant enrichment for fatty acid oxidation (FAO) and PPAR signaling was found. Significant treatment eQTLs were identified for relevant genes involved in FAO, apoptosis and prostaglandin metabolism. ASA-induced apoptosis and secondary necrosis were confirmed with identification of significant differential responses of apoptotic genes to ASA.

Conclusions:

Results demonstrate large transcriptional responses to ASA treatment with differences in responses between individuals. Genomic and cellular results suggest that ASA effects on the mitochondria are key mechanisms of action that could underlie clinical effects. These results could be used to assess clinical treatment responses for chemoprevention in the future.

Graphical Abstract

New & Noteworthy:

Aspirin treatment in colonic organoids from diverse individuals revealed significant transcriptome-wide responses, especially for genes in lipid and apoptosis signaling pathways. In normal organoids, apoptosis was induced by aspirin, providing one possible mechanism of colorectal cancer chemoprevention. Our results are a first step toward implementation of personalized medicine for aspirin in colorectal cancer prevention.

Introduction

A major challenge in studying treatment responses in humans is that they are complex and difficult to measure and control. Traditionally, transformed cell lines and animal models have been employed, but these do not fully recapitulate human biology and cannot assess how interactions vary across individuals and across populations. Modeling of cellular environments ex vivo using a paired experimental design allows for a more controlled way to study treatment effects on sub-cellular phenotypes such as gene expression especially because this design controls for a number of confounding factors. Stem cell-derived organoids offer several advantages for assessment of treatment effects because they recapitulate human cell composition and function (1) and they enable comparisons of treatment effects across individuals and populations not possible with traditional models. Furthermore, organoids could be used clinically to assess individualized responses in order to personalize treatments to maximize benefits while minimizing harms.

Colorectal cancer (CRC) is a leading cause of cancer and cancer deaths globally (2) with high incidence and mortality in the United States among African-American (AA) individuals (3, 4). Experimental, observational and clinical studies have demonstrated chemoprotective effects of aspirin (acetylsalicylic acid, ASA) with significant reductions in CRC incidence and mortality (5). Several mechanisms of ASA’s chemoprotective effects are proposed but remain incompletely understood with both cyclooxygenase 2 (COX2, encoded by the gene PTGS2) and non-COX2 mechanisms implicated (6). Comprehensive evaluation of ASA genomic and cellular responses across individuals is lacking but could help elucidate mechanistic insights into chemoprevention as well as to identify relevant biomarkers of individualized ASA response for CRC chemoprevention across diverse populations.

In this study, human colonic organoids from a diverse cohort of individuals were used to assess transcriptional and cellular treatment responses to ASA in order to provide further insights into ASA primary mechanisms of action in the normal human colon and to explore whether inter-individual and inter-ethnic differences in responses to ASA treatment exist that could inform future clinical precision prevention strategies.

Methods

Study participants.

For transcriptional profiling, 67 participants undergoing screening colonoscopy were included from whom colonic biopsies were obtained for organoid cultures. For validation and cellular studies, organoid lines from independent participants were used. For validation experiments, organoid lines included participants with and without the a common cancer predisposition syndrome, Lynch syndrome, in order to assess for potential differences in this high-risk group. Since response differences were not found between Lynch and non-Lynch lines (Supplemental_ Figure_S1), results for the combined group are shown throughout the manuscript. For all participants, research biopsies were obtained from normal-appearing mucosa at the rectosigmoid junction approximately 20cm from the anal verge using standard size biopsy forceps. Demographic data including age, sex, self-identified race/ethnicity were recorded. The study was approved by the University of Chicago Institutional Review Board.

Organoid cultures.

Organoids were derived from colonic biopsies using a protocol adapted from Sato et al., (14) as previously described (15). Organoids were cultured on 24-well CytoOne plates (USA Scientific, FL), embedded in a 30 µL droplet of Matrigel, cultured in 500 µL of the corresponding organoid media in an incubator at 37°C and 5% CO₂. A summary of media formulations used in this study is shown in Supplemental_Table_S1. Organoids were incubated in basal media for 24 hours to enable differentiation.

YAMC cultures.

Young adult mouse colonic (YAMC) cells were obtained from Dr. Candace Cham (University of Chicago). Cells were cultured in RPMI 1640 (Corning, Glendale, AZ) supplemented with 1× L-glutamine, 1× ITS (insulin transferrin and sodium selenite), 5% fetal bovine serum (FBS), 1% penicillin and streptomycin, and 5 U/L interferon gamma from Thermo Fisher Scientific (Waltham, MA) at 33°C and 10% CO₂. All treatments were performed in RPMI 1640 supplemented with 0.5% charcoal-stripped FBS after overnight serum starvation to synchronize the cells. Before performing assays, cells were trypsinized and counted using trypan blue exclusion assay (16).

Treatments.

For transcriptional profiling, after 24 hours of culturing in the basal media (which did not contain ASA), organoids were treated with either 3mM ASA or vehicle control (0.1% DMSO) for 24 hours. The 3mM ASA dose was based on previous studies (17–20). For qPCR and cellular assays, organoids and YAMC cells were cultured and treated with 0.5 mM ASA, 3mM ASA or vehicle control for various time points.

RNA-Sequencing.

Organoids were harvested using cold Advanced DMEM, pipetted up and down 10 times, moved to centrifuge tube, and then spun at 400 g for 5 min at 4°C. Upper media and Basement Membrane Extract (BME) were carefully aspirated and discarded. mRNA was then isolated from cells using the RNeasy^® Plus Mini Kit (Qiagen, Germantown, MD) according to the manufacturer’s protocol. RNA quality and quantity were assessed using an Agilent Bio-analyzer with RNA integrity numbers (RIN) of >9 for all samples. RNA-seq libraries were prepared using Illumina mRNA TruSeq Kits as protocolled by Illumina. Library quality and quantity were checked using an Agilent Bio-analyzer, and the pool of libraries was sequenced using an Illumina NovaSeq6000 (paired end 100 bp) using Illumina reagents and protocols at the University of Chicago Genomics Facility.

qPCR.

Organoids were harvested and mRNA was extracted using the RNeasy^® Plus Mini Kit (Qiagen, Germantown, MD) according to the manufacturer’s protocol. RNA concentration was measured using a nanodrop and an appropriate concentration of cDNA was made using the High Capacity cDNA Reverse Transcription kit (Applied Biosystems, Waltham, MA) with 1 U/uL of RNase inhibitor (Applied Biosystems, Waltham, MA) added to the reaction mix. qPCR was performed using the Fast SYBR^® Green Master Mix Protocol (Applied Biosystems, Waltham, MA) and ran on the real time PCR system (QuantStudio). In total, 19 genes: ACSL1, APAF1, ANGPTL4, BCL2, BCL2L2, BAK1, BAX, CASP3, CASP9, CYCS, FABP1, FADS2, GAPDH, HMGCS2, PPARA, PPARD, PPARG, PLIN2, and SCD1 were examined (primers listed in Supplemental_Table_S2). GAPDH was used as an endogenous control. Taqman assays were used for APAF1, BCL2, BCL2L2, BAK1, BAX, CASP3, CASP9 and CYCS. Relative changes in expression of genes were analyzed by using 2^−∆∆CT method (21). The results were expressed as the log fold change (LFC) in gene expression normalized to endogenous reference gene (GAPDH) as well as with the expression of untreated control sample (DMSO) at the threshold cycle (Ct). Statistical analysis was performed by means of two-tailed paired t-test with p<0.05 considered as significant.

Genotyping.

Genomic DNA was extracted from blood samples obtained from 64 participants (3 participants did not have available blood samples) and genotyped on the InfiniumOmniExpress-24v1-3_A1 microarray (Illumina; San Diego, CA) which included 714,238 SNPs. Genotype data were used to ascertain genetic ancestry in order to validate self-reported ancestry. To increase the density of sites for the purpose of eQTL mapping, genotype imputation was performed with IMPUTE2 (22) using data from 1000 Genomes Project (23) haplotypes as a reference. Weighted means (dosages) of IMPUTE2’s estimated posterior genotype probabilities were calculated and sites were then filtered by minor allele frequency (>0.05), leaving approximately 8.5 million loci.

Apoptosis and necrosis assays.

The RealTime-Glo^™ Annexin V Apoptosis and Necrosis Assay (Promega, Madison, WI) was performed in 6 organoid lines using two doses of ASA (0.5mM & 3mM). For this assay, organoids were grown and differentiated in basal media for 24 hours before treatment. Organoids were then dissociated into single cell suspension using repeated pipetting and pre-warmed TrypLE^™ Express dissociation reagent (Thermo Fisher Scientific, Waltham, MA) for 10–15 minutes at 37°C before seeding. Organoids were counted and 150 organoids per well were plated on a 96-well plate. cells derived from organoids were treated with ASA and vehicle control DMSO from 1–48 hours. Apoptosis and necrosis 2X detection reagent was added at the same time with ASA starting from 0 hour. After incubation, luminescence (apoptosis; exposure of phosphatidylserine (PS) on the outer leaflet of the cell membrane) and fluorescence (necrosis; loss of membrane integrity) were recorded at initial hours from 1–8 hours, 24h and 48h using a plate reader. The assay was performed in triplicate.

Bioinformatic analyses

Ancestry estimates.

Genetic ancestry proportions were estimated with the program ADMIXTURE (v1.3.0) (24) using approximately 255,000 imputed SNPs that were pruned for linkage disequilibrium and filtered for a minor allele frequency greater than 0.05 using PLINK (v1.90) (25) commands –indep-pairwise 50kb 10 0.2 and –maf 0.05. The genotype data were then merged with the genotype data from 10 CEU, 10 YRI, and 10 CHB from the 1000 Genome Project. ADMIXTURE was run with k=3 to capture African, European, and possibly Native American or Asian ancestry components of individuals who self-identified as either non-Hispanic White or Black/African American.

Transcriptional responses.

Sequence alignment and gene expression value estimation was performed using the rsem-calculate-expression function of the RSEM v.1.3.1 (26) software package using the STAR aligner (27). The STAR transcriptome reference was generated from the 1000 genomes Phase2 Reference Genome Sequence (hs37d5) and transcript annotations from the Gencode comprehensive gene annotation GTF (Release 29). Differential expression analysis was performed on count data using the R package DESeq2 (v1.30.1) (28). Transcript-level count data was first summarized at the gene-level using the R package tximport (v1.18.0) (29). Reads counts were then successively adjusted for sex and self-reported ancestry using the Combat_seq function of the R sva package (v3.38.0) (30) while using the function’s group argument to preserve the signal for the treatment condition covariate. Genes tested for differential expression were initially filtered for protein coding biotype and by applying a minimum total count threshold of 10 across all samples. Further gene filtering was achieved through the default independent filtering method of DESeq2. False discovery rate (FDR) of 5% or less was considered significant unless otherwise indicated. Gene expression is reported as the transformed and normalized counts after applying the variance stabilizing transformation (VST) of DESeq2. Mean differential gene expression is reported as the DESeq2 log₂ fold change of ASA/DMSO. Overlap analysis with a previous study (31) was performed using a hypergeometric test in R. To test for differences in response effect sizes between the two populations, the mixed effects model: Model 1: individual + age + sex + race + treatment + race x treatment was considered, where the individual term was treated as a random effect and the other covariate terms were treated as fixed effects. The interaction term effect size represents the difference in response to treatment between the EA and AA populations. This mixed model was tested using the dream statistical package, part of the variancePartition R package (32, 33) which is built on top of the standard limma workflow.

Gene set enrichment analysis.

Gene set enrichment analysis (GSEA) was performed using the R package SetRank (34). This method was designed to minimize false positives by taking gene set overlap into account. Briefly, the algorithm inputs a gene set collection from the KEGG, GOBP and REACTOME databases and the list of differentially expressed genes in response to treatment (not filtered by a p-value cut-off). The output includes a setRank value (reflects the importance of the gene set in the gene set network; the higher the value, the more important the gene set), a p-value associated with the SetRank value (probability of observing a gene set with the same SetRank value in a random network), a corrected p-value (account for overlap with other gene sets) and adjusted p-value (correction of multiple testing). Pathways with the highest SetRank values and associated SetRank p-values<0.05 are shown in the results.

eQTL analyses.

cis-eQTL mapping was performed on the expression data of protein coding genes after DESeq2 VST transformation and quantile normalization, and using imputed bi-allelic variants that fall within 500 kb of a gene’s transcription start site and have a MAF>0.05. After applying these filters, there were approximately 49.6 million gene-variant pairs for analysis. To find condition-specific eQTLs, the program BRIdGE (35) was run, using the documentation suggested grid of effect sizes. A posterior probability threshold of 0.70 was used to assign significance to an eQTL that had either an ASA effect only, a DMSO effect only, equal effects in both treatment conditions, or unequal, nonzero effects in both conditions.

Results

Ancestry and demographics of participants.

For transcriptional response experiments, organoid lines were established from rectosigmoid biopsies from 67 healthy participants undergoing screening colonoscopy of which 33 individuals self-reported their race as AA and 34 as non-Hispanic White. Participants were 56.7% female and average age of 56 years. A summary of participant demographics is provided in Supplemental_Table_S3. Using ADMIXTURE with k=3, ancestry proportions were estimated for the 64 of the 67 individuals with genotype data. For the 32 genotyped individuals self-reporting as non-Hispanic White, the ancestry proportions ranged between 98% and 100% European ancestry. For the 32 genotyped individuals self-reporting as AA, ADMIXTURE estimates ranged between 68% and 96% African ancestry. The remaining 3 individuals without genotype data were classified by self-reported race. For gene expression validation and cellular assays, organoid lines selected from an additional 9 cancer-free participants were used.

Large number differentially responsive genes to ASA treatment in colonic organoids with evidence of inter-individual differences in responses.

Organoids from the same individual were treated with 3mM ASA or vehicle control (0.1% DMSO) for 24 hours. Across the 134 samples, after rsem filtering was applied, an average of 72957.4 reads mapped uniquely to a gene. In total, 8343 protein-coding genes showed significant differential expression in response to ASA: 4079 up- and 4264 down-regulated genes at FDR <5% (Figure 1A; Supplemental_Table_4). Among differentially responsive genes, top significantly upregulated genes included XPO1, HMGCS2, SEMA5A, VLDLR, RAB30, SLC16A9 and PLIN2 and top significantly downregulated genes included IVNS1ABP, AGR2, ATP2C1, and PD1A4 (Figure 1B). Genes in CRC-related pathways from Kyoto Encyclopedia of Genes and Genomes (KEGG) showed variable differential responses to ASA treatment (significant genes in Supplemental_Figure S2).

Figure 1. ASA-induced genomic responses in normal human colonic organoids.

Organoids from 67 individuals were treated with 3mM ASA or vehicle control (0.1% DMSO) for 24 hours (h) for RNA-seq. (A) In total, 8343 protein-coding genes showed significant differential expression in response to ASA at FDR <5% of which 4079 and 4264 genes were up- and down-regulated, respectively. Several top significant genes are labeled in the volcano plot. (B) Top 10 up- and top down-regulated differentially expressed genes in response to ASA based on adjusted p-value. (C) Difference in the absolute LFC values between EA and AA compared to permuted values. Permutations of the ancestry labels among the individuals shown in light blue and the actual difference shown in the light orange (p=0.02). (D, E, F) Genetic variants associated with condition-specific responses to ASA treatment. Assessment for a genetic contribution of ASA treatment responses by eQTL mapping. For each gene-variant pair, we show the LFC by eQTL genotype (Left) and expression data in each treatment condition across genotypes (Right). Top ASA-only eQTLs include variants associated with responses of PTGES2 (D). A significant eQTL for PTGS2 was noted in both ASA and DMSO conditions (E). Top DMSO control-only eQTLs include variants associated with responses of PLIN2 (F).

Abbreviations: AA, African American; ASA, aspirin; DMSO, dimethyl sulphoxide; EA, European American; FDR, false discovery rate; LFC, log fold change; log₁₀padj; Log10 adjusted p-values; padj; adjusted p-values; vst(Exp), variance stabilizing transformation of expression.

Next, we determined the overlap of differentially responsive genes in the present study with a previous study of transcriptional responses of 0.5mM ASA treatment for 72 hours in human organoids from 38 participants (no information on race or ethnicity) by Devall et al (31). At a threshold of FDR<10%, the Devall et al (31) study found 1154 differentially responsive genes, while the present study identified 8976 differentially responsive genes at the same FDR threshold. Out of the 1143 significantly differentially expressed genes in the Devall et al (31) study that were also analyzed in the present study, a total of 910 (80%) were overlapping which was significant using a hypergeometric test (p = 2.36 × 10⁻⁶); of these, 78.7% of genes had concordant expression, while 21.3% were discordant between the two studies. There was significant correlation between log fold change (LFC) in both studies (r=0.336; p=2.2×10⁻¹⁶), though larger average fold changes in differentially responsive genes were found in the present study at 24 hours (Supplemental_Figure_S3).

Given that the present study included organoid lines from nearly equal numbers of AA and EA participants, we asked whether there were genes that showed population-specific responses to ASA treatment. Overall, significantly differentially responsive genes showed similar patterns between populations using the DESeq method, though the top significant genes varied slightly in effect sizes and p-values (Supplemental_Figure_S4). Moreover, significant correlations between effect sizes (r=0.808; p=2.2×10⁻¹⁶) and z-scores (r=0.944; p=2.2×10⁻¹⁶) were noted between populations with minimal deviation (Supplemental_Figure_S4). Due to the linear dependency that exists between variables in models with race-specific effects, we applied a random effects model as implemented in the dream method. Using the dream method, no individual genes were significantly differentially responsive between populations at a significant adjusted p-value (Supplemental_Table_S5). Gene set enrichment of population-specific response genes using SetRank, an advanced analysis algorithm designed to minimize false positive results (34), revealed pathways related to membrane lipid metabolic processes, lysosome and glycolipid metabolic processes among others (full results in Supplemental_Table_S6). To assess for differences in genome-wide ASA responses between individuals of African and European ancestry, we calculated the differences in the absolute value of effect sizes between the 2 populations and found significantly greater ASA responses in organoids from individuals of European vs. African ancestry when compared to the null distribution (p=0.02) (Figure 1C).

In order to assess for a genetic contribution in treatment responses, we performed eQTL mapping using BRIdGE, a Bayesian approach for determining gene-environment interactions for paired measurements under two conditions (35). In total, 57, 100, and 2333 variants were associated with ASA-only, DMSO control-only or equal effect sizes in both conditions, respectively, at a posterior probability cut-off of >0.70 (Supplemental_Table_S7). No variants showed unequal, non-zero effect sizes in the two treatment conditions. Examples of top ASA-only gene targets included PTGES2 (Figure 1D) as well as CCDC93 and BCL2L2 (Supplemental_Figure_S5). Examples of DMSO control-only gene targets included PLIN2 (Figure 1E) as well as GPR89A, and CAPN8 Supplemental_Figure_S5). Of note, the eQTL associated with PLIN2 response, rs10118790, had large allele frequency differences between individuals of African and European descent (e.g., G allele frequency of 66% in Africans vs. 10% in Europeans). Finally, a significant eQTL for PTGS2 was noted in both treatment conditions (Figure 1F).

Fatty acid oxidation (FAO) and PPAR signaling pathways enriched among ASA-differentially responsive genes that were validated in independent organoids.

SetRank identified a number of significantly enriched pathways using the gene ontology, KEGG and Reactome databases (Figure 2A & B; full results in Supplemental_Table_S8). The top enriched pathways included PPAR signaling as well as processes related to fatty acid oxidation (FAO) such as lipid biosynthetic, fatty acid metabolic and monocarboxylic acid metabolic processes among others. These enriched pathways were similar to top enriched pathways identified in our previous study using an organoid line from 1 individual (36) and were also enriched using differentially responsive genes that overlapped between the present study and Devall et al (31).

Figure 2. Gene set enrichment of ASA transcriptional responses identifies fatty acid oxidation (FAO) and PPAR signaling pathways validated in independent organoid lines.

Gene set enrichment analysis was performed using the R package SetRank (34). This method was designed to minimize false positives by taking gene set overlap into account. (A) Network figure of top significantly enriched pathways (pSetRank<0.05) showing relationships between pathways. The size of the node corresponds to size of the gene set, and the node border color reflects the −log10(corrected p-value). Solid arrows depict overlap between gene sets, while the broken arrows refers to gene subsets. The arrow color reflects the −log10(Intersection p-value). (B) Significant SetRank results detailing pathway name, database, gene set size, setRank value (reflects the importance of the gene set in the gene set network; the higher the value, the more important the gene set), a p-value associated with the SetRank value (probability of observing a gene set with the same setRank value in a random network), a corrected p-value (account for overlap with other gene sets) and adjusted p-value (correction of multiple testing). (C, D) Validation of top differentially expressed genes from RNA-seq (n=67 lines) data by qPCR in an independent set of organoid lines (n=6–14 lines) with 3mM ASA treated for 24h. All tested FAO genes (ACSL1, ANGPTL4, FABP1, HMGCS2, PLIN2 and SCD1) and PPAR signaling genes (PPARA, PPARD, and PPARG) were validated except for FADS2 that showed overall upregulation in the RNA-seq experiment, while there was downregulation with ASA treatment in the lines tested by qPCR.

p<0.05*, p≥0.05=ns

Abbreviations: ASA, aspirin; FAO, fatty acid oxidation; h, hours

In order to validate expression of top genes in these enriched pathways, we performed qPCR using an independent set of organoid lines for FAO (ACSL1, ANGPTL4, FABP1, FADS2, HMGCS2, PLIN2 and SCD1) and PPAR signaling (PPARA, PPARD, and PPARG) genes. All of the genes in FAO and PPAR signaling pathways tested by qPCR showed significant differential expression in 3mM ASA compared to DMSO, respectively. For all but 1 gene, FADS2, we found consistent effect sizes and direction compared to results from the multi-individual RNA-sequencing experiment for these pathways, respectively (Figure 2C & D). For FADS2, we noted overall upregulation with ASA treatment averaged across the 67 organoid lines, while, in the qPCR validation experiment, there was significant FADS2 downregulation with ASA. These discordant results did not appear to be due to different responses by FADS2 isoforms (Supplemental_Figure_S6).

Time and dose responses of top differentially responsive FAO and PPAR signaling genes in independent organoid lines.

Given that genes and pathways involved in FAO and PPAR signaling showed significant enrichment and showed overlap with previous ASA studies (31, 36), we determined whether treatment responses differed by dose and time. Responses of top FAO and PPAR signaling genes were evaluated at 3h, 6h and 24h with 0.5mM and 3mM ASA treatment by qPCR in 6 independent organoid lines.

Overall, at both doses, significant increased expression of genes was noted over time except for FADS2 which showed significant downregulation between 3h and 24h with 3mM ASA treatment (Figure 3A–D). Differences in treatment responses by dose were investigated for the same top FAO and PPAR genes. At the 3mM ASA dose, ANGPTL4, HMGCS2, PLIN2 and PPARA showed greater significant upregulation, while FADS2 showed greater downregulation (Figure 3E & F). The remaining genes showed no significant dose-related differences across the 3 time points. No significant differences by time or dose were noted between organoids from Lynch and non-Lynch syndrome patients (Supplemental_Figure_S1).

Figure 3. Time- and dose-related ASA responses of top differentially responsive FAO and PPAR signaling genes.

FAO genes tested include: ACSL1, ANGPTL4, FABP1, FADS2, HMGCS2, PLIN2, and SCD1. PPAR signaling genes tested include: PPARA, PPARD and PPARG. An independent set of colonic organoid lines (n=6) was used to evaluate ASA responses at 3h, 6h and 24h; (A) Expression of genes involved in FAO with 0.5mM ASA over time. (B) Expression of genes involved in PPAR signaling with 0.5mM ASA over time. (C) Expression of genes involved in FAO with 3mM ASA over time. (D) Expression of genes involved in PPAR signaling with 3mM ASA over time. (E) Dose response in genes involved in FAO measured as expression in 3mM ASA/0.5mM ASA at each time point. (E) Dose response in genes involved in PPAR signaling measured as expression in 3mM ASA/0.5mM ASA at each time point.

*p<0.05, **p<0.01, ***p<0.001, ****p<0.0001, ns=p≥0.05

Abbreviations: ASA, aspirin; FAO, fatty acid oxidation; h, hours; LFC, log fold change.

ASA treatment increases apoptosis and secondary necrosis of colonic organoids.

Given that gene set enrichment results showed significant induction of FAO and PPAR signaling, processes localized to the mitochondria, we focused our cellular assays on assessment of apoptosis and secondary necrosis in responses to ASA treatment since these are processes that also affect the mitochondria and have been previously implicated in ASA effects. To test for these cellular responses, a phosphatidylserine (PS)-based apoptosis and secondary necrosis assay was performed in 6 normal colonic organoid lines with 2 doses of ASA (3mM and 0.5mM) at several time points. As shown in results from a representative organoid line (Figure 4A), we observed significant increase in apoptosis at early time points that appeared to peak around 8h with subsequent decrease over time. A significant dose response for apoptosis was found with greater apoptosis with 3mM vs. 0.5mM. Secondary necrosis was noted to increase over time with a steeper slope of increase between 8–24h and plateauing after 24h (Figure 4B). Dose differences were also noted for secondary necrosis.

Figure 4. Apoptosis and necrosis in ASA-treated normal human colonic organoids.

Apoptosis and secondary necrosis assays and qPCR of key apoptosis genes were performed in normal colonic organoid lines (n=6). (A) A representative organoid line is shown with significant increase in apoptosis at early time points peaking around 8h with subsequent decrease over time. A significant dose response with greater apoptosis was found with 3mM vs. 0.5mM. (B) Secondary necrosis in a representative organoid line was noted to increase over time with a steeper slope of increase between 8–24h and plateauing after 24h. Dose differences were also noted for secondary necrosis. (C) Across the 6 organoid lines, greater % apoptosis in 3mM compared with 0.5mM was noted at early time points (2h, 3h & 7h). (D) Greater % necrosis was found in 3mM versus 0.5mM for all time points except at 1h. Some variability in % apoptosis and % necrosis was noted across organoid lines especially with the 3mM dose and later time points. (E) ASA effects on apoptosis genes in normal human colonic organoids. Gene expression of key apoptosis genes (APAF1, BCL2, BCL2L2, BAK1, BAX, CASP3, CASP9, and CYCS) at 3h, 6h and 24h was evaluated with 3mM ASA or DMSO. Significant upregulation of anti-apoptotic BCL2L2 gene at all time points and significant downregulation of pro-apoptotic BAK1 and BAX genes at 24h was noted with 3mM ASA dose.

Experiments performed in triplicate.

*p<0.05, **p<0.01, ***p<0.001.

Abbreviations: ASA, aspirin; DMSO, dimethyl sulphoxide; h, hour; %, percentage; LFC, log fold change.

In order to compare apoptosis and necrosis across 6 organoid lines, we calculated percent apoptosis and necrosis as effects with ASA treatment (0.5mM and 3mM) relative to DMSO vehicle control in each line. When combining all 6 lines, we noted greater percent apoptosis in 3mM compared with 0.5mM at early time points (2h, 3h & 7h) (Figure 4C & E), while percent necrosis was greater in 3mM vs. 0.5mM for all time points except at 1h (Figure 4D & F). Some variability in apoptosis and necrosis was noted across organoid lines especially with the 3mM dose and later time points (Supplemental_Figure_S7 & S8). We validated similar ASA-related apoptosis and necrosis results in YAMC cells, a normal murine colonic cell line (Supplemental_Figure_S9).

Given that ASA increased apoptosis and necrosis in colonic organoids that, in turn, could explain increased FAO, we asked whether key genes involved in regulation of apoptosis were also affected by ASA treatment to including APAF1, BCL2, BCL2L2, BAK1, BAX, CASP3, CASP9, and CYCS. Organoid lines were treated with 0.5mM and 3mM ASA or DMSO and gene expression assessed by qPCR at 3h, 6h and 24h (Figure 4G). Overall, more apoptosis genes were differentially expressed with 3mM compared with 0.5mM ASA treatment. BCL2L2 was significantly upregulated at all but one time point in both ASA doses. With the 3mM ASA dose, we noted significant upregulation of APAF1 and CYCS at 3h and 6h but not at 24h; CASP9 was also significantly upregulated at 6h. In contrast, we noted significant downregulation of BAK1 and BAX at 24h. No significant differential expression was found at either dose or time points for BCL2 or CASP3.

Discussion

This study leveraged colonic organoids from diverse individuals to assess transcriptional and cellular responses to ASA, a proven CRC chemoprotective treatment whose mechanisms of action remain incompletely understood. Our results showed large genome-wide transcriptional responses to ASA in epithelial-derived colonic organoids with enrichment of FAO and PPAR signaling pathways among differentially response genes, of which top genes in these pathways were validated and showed response differences by ASA dose and time. Given that FAO and PPAR signaling and effects are largely localized to the mitochondria, we further assessed ASA effects on apoptosis, a process similarly localized to the mitochondria, and found dose- and time-dependent effects with early induction of pro-apoptotic genes, CYCS and APAF1. These results underscore ASA’s effects on the mitochondria in normal colonic epithelium that could contribute to clinical responses. Our exploratory investigations highlighted potential ASA response differences between individuals and populations, though further studies are needed to confirm these preliminary findings. These results demonstrate how human organoids can be used to gain new knowledge about ASA transcriptional and cellular responses across diverse individuals.

Transcriptional and cellular responses identified in this study specifically highlight ASA’s mitochondrial effects on FAO, PPAR signaling and apoptosis that could contribute to both beneficial and deleterious biological effects such as chemoprevention or toxicity (e.g., ulceration, Reye’s syndrome and ASA poisoning) (42–45). Previous studies have shown that ASA and its metabolite, salicylate, as well as other non-steroidal anti-inflammatory medications activate mitochondria that leads to uncoupling of oxidative phosphorylation, increased permeability as well as activation of FAO, PPAR signaling and apoptosis (46–49). Results from the present study are in line with previous studies that showed early induction of mitochondrial cytochrome c release by ASA doses of 5–10mM beginning within 6h (50, 51). Similar to previous studies, we also noted significant upregulation of the cytochrome c binding partner, APAF1, and caspase 9, CASP9, but we did not find induction of caspase 3, CASP3, by ASA treatment as previously described (52). Contrary to some previous studies (41, 53–56), we did not find ASA-mediated suppression of the anti-apoptotic gene, BCL2 at any of the doses or time points. Interestingly, we found consistent upregulation of the anti-apoptotic gene, BCL2L2, at both ASA doses at all but one of the time points which is similar to a previous study (56) but different from results in other studies (57). An ASA-specific eQTL for BCL2L2 was identified that could play a role in inter-individual treatment effects. While we did not find large differences in apoptosis or gene expression in response to ASA treatment across tested organoid lines, a larger sample size would be required to assess inter-individual variation in apoptosis.

FAO activation by ASA also has been previously linked to the mitochondria through increased phosphorylation of AMP-activated protein kinase (AMPK) and subsequent reduction of malonyl-coA, an inhibitor of CPTI, a membrane protein that shuttles acyl carnitine into the mitochondria (58). While we did not find that PRKAA2, the gene that encodes the catalytic subunit of AMPK, was differentially expressed, we did find that CPT1 was significantly upregulated suggesting that ASA induction of CPT1 could mediate FAO activation in organoids. In addition, PPAR signaling, mediated by transcription factors PPARα, PPARδ, and PPARγ that regulate FAO, has been studied in the context of ASA response. In the present study, PPARA, previously shown to be protective against colorectal neoplasia (59, 60) and a direct ASA receptor (61), was induced by ASA treatment at both doses across all time points. PPARG was also upregulated in response to ASA, which is in line with results from a study in fatty liver disease (62) but different from a study in colon cancer (63). Further work is needed to fully dissect the roles of FAO and PPAR signaling in ASA treatment including apoptosis-dependent and -independent effects.

Beyond ASA-mediated effects on apoptosis, FAO and PPAR signaling, ASA was also noted to impact previously studied chemoprotective mechanisms such as COX inhibition (11). Specifically, ASA significantly downregulated PTGS2, the gene that encodes COX2, in normal colonic organoids, which could reduce cancer initiation by reduction of PGE2 levels. An eQTL for PTGS2 was noted in both conditions that potentially could impact inter-individual COX2-mediated effects. In addition, we found an ASA-specific eQTL for PTGES2, the gene encoding prostaglandin E synthase that catalyzes conversion of PGH2 to PGE2. It is possible that this variation could impact levels of PGE2 across different individuals and thereby mediate chemoprotective effects. Additional genes important in colonic carcinogenesis, such as FOS, JUN, KRAS, MAPK8, MAPK9, MAPK10, MYC, NFKB1, RAC1, TGFB1 and TP53 were also found to be significantly differentially responsive with ASA treatment and could further mediate chemoprotective effects.

Strengths of the present study include application of human-derived organoids, paired study design to control for confounders, and inclusion of diverse individuals to elucidate genome-wide ASA treatment effects in the colon. We also note a few limitations. First, the dose of ASA used for transcriptomic profiling is on the higher end of physiological ranges in humans, though this higher dose is consistent with anti-inflammatory doses and doses used for chemoprotection in Lynch syndrome (13). Furthermore, we previously reported no overt cytotoxicity at this dose that could confound genomic responses (36) and validated top differentially responsive genes with a lower ASA dose in line with results from previous studies (31). Second, this study investigated ASA effects on normal colonic organoids, not colonic neoplasia; however, these results are still relevant for elucidating potential protective effects given that the normal colon is an optimal target tissue for chemoprevention. Future studies could compare responses in differentiated vs. undifferentiated organoids as well as organoids from colonic neoplasia (adenomas and cancer) to further dissect ASA effects along the continuum from homeostasis to carcinogenesis. Third, ASA treatments were applied directly to the basolateral surface of organoids which might not entirely mimic systemic effects in vivo given that organoids represent epithelial cell types and do not capture ASA effects on immune, stromal or other non-epithelial cell types. Moreover, all organoids were derived from the left colon, and a previous study suggested possible anatomic site-specific ASA responses (31). Given the large number of differentially responsive protein-coding genes, we would anticipate similar treatment effects on protein expression but this was not directly tested here.

In conclusion, this is the first study to assess ASA-mediated transcriptomic and cellular responses in normal colonic organoids from diverse individuals. We demonstrate ASA regulation of genes and pathways involved in FAO, PPAR signaling and apoptosis. Understanding how these ASA responses differ between individuals including individuals of diverse ancestry will be important and could help personalize ASA treatment in the future.

Supplementary Material

Supplemental Figures: S1–9: 10.6084/m9.figshare.26196980

Supplemental Tables: S1–8: 10.6084/m9.figshare.26196983

Supplemental Figure legends:

Supplemental_Figure_S1. Gene expression validation by qPCR in independent Lynch and non-Lynch organoid lines (n=6). Independent set of colonic organoid lines used to study gene expression of top FAO genes (ACSL1, ANGPTL4, FABP1, FADS2, HMGCS2, PLIN2 and SCD1) and PPAR signaling genes (PPARA, PPARD, and PPARG). Lines from cancer-free participants with Lynch syndrome and non-Lynch lines were analyzed at 3 different time points (3h, 6h & 24h) or at 24h with 3mM ASA.

*p<0.05, ns=p≥0.05

Abbreviations: ASA, aspirin; DMSO, dimethyl sulphoxide; FAO, fatty acid oxidation; h, hours; LFC, log fold change.

Supplemental_Figure_S2. ASA responses to genes involved in CRC-related pathways including prostaglandin synthesis, apoptosis, cell cycle, TGFB signaling, MAPK signaling, MSI pathway, proto-oncogenes and inflammation.

Abbreviations: LFC, log fold change; padj, adjusted p-values.

Supplemental_Figure_S3. Comparison of transcriptional responses to ASA (FDR<10%) at 24h (present multi-individual study) and 72h (1) in colonic organoids.

Abbreviations: ASA, aspirin; FDR, false discovery rate; LFC, log fold change; padj, adjusted p-value.

Supplemental_Figure_S4. Ancestry differences in genome-wide responses to ASA treatment in normal human colonic organoids. Significantly differentially responsive genes showed similar patterns between AA and EA populations using the DESeq method, albeit the top significant genes varied slightly in effect sizes and p-values. (A) 6267 genes showed significant differential gene expression in AA in response to ASA of which 2992 and 3275 were up- and downregulated, respectively (FDR <5%). (B) 7089 genes showed significant differential gene expression in EA in response to ASA of which 3557 and 3532 were up- and downregulated, respectively (FDR <5%). (C) Significant correlation was seen between effect sizes of AA and EA populations with minimal deviation. (D) Significant correlation was found between z-scores of both AA and EA populations with minimal deviation.

Abbreviations: AA, African American; ASA, aspirin; EA, European American; FDR, false discovery rate; LFC, log fold change; padj, adjusted p-value.

Supplemental_Figure_S5. Genetic variants associated with condition-specific responses to ASA treatment. Assessment for a genetic contribution of ASA treatment responses by eQTL mapping. For each gene-variant pair, we show the LFC by eQTL genotype (Left) and expression data in each treatment condition across genotypes (Right). (A & B) Top ASA-only eQTLs include variants associated with responses of CCDC93 and BCL2L2. (C & D) Top DMSO control-only eQTLs include variants associated with responses of GPR89A and CAPN8.

Abbreviations: ASA, aspirin; DMSO, dimethyl sulphoxide; LFC, log fold change; vst(Exp), variance stabilizing transformation of expression.

Supplemental_Figure_S6. FADS2 responses by isoforms. (A) Expression of FADS2 by isoform and treatment (ASA and DMSO). (B) Log-fold change of ASA responses of FADS2 by isoforms. Of note, the qPCR primers used in the validation experiments corresponds to isoform ENST00000278840.

Abbreviations: ASA, aspirin; DMSO, dimethyl sulphoxide

Supplemental_Figure_S7. Apoptosis and necrosis with and without ASA treatment in 6 normal colonic organoid lines. Experiments were performed in triplicate.

Abbreviations: ASA, aspirin; DMSO, dimethyl sulphoxide; RFU, relative fluorescence unit; RLU, relative luminescence unit.

Supplemental_Figure_S8. Interline % variation in apoptosis and necrosis. % variation calculated as 0.5mM or 3mM ASA divided by DMSO vehicle control x100% in each line. (A & C) Apoptosis at 0.5mM and 3mM, respectively (B & D) Necrosis at 0.5mM and 3mM, respectively.

Abbreviations: ASA, aspirin; DMSO, dimethyl sulphoxide; %, percentage.

Supplemental_Figure_S9. ASA-induced apoptosis and necrosis in YAMC cells. (A) Relative luminescence (RLU) for apoptosis, (B) Relative fluorescence unit (RFU) for necrosis, (C) % apoptosis (RLU in ASA/DMSO x 100%), and (D) % necrosis (RFU in ASA/DMSO x 100%). Experiments performed in triplicate.

Abbreviations: ASA, aspirin; DMSO, dimethyl sulphoxide; RFU, relative fluorescence unit; RLU, relative luminescence unit; YAMC, young adult mouse colon; %, percentage.

Supplemental Tables:

Supplemental_Table_S1. Media formulations used in present study.

Supplemental_Table_S2. List of primers used for real-time PCR in the present study. Primer sequences of the selected candidate genes.

F; forward, R; reverse.

Supplemental_Table_S3. Participant characteristics in the overall study cohort and independent validation cohort.

AA, African American; BMI, body mass index; EA, European American; n, number of individuals; SD, standard deviation.

Supplemental_Table_S4. DESeq2 output for differential expression from all protein-coding genes to 3mM ASA treatment at 24 hours.

Supplemental_Table_S5. Dream method output for inter-ethnic response of genes to 3mM ASA treatment at 24 hours.

Supplemental_Table_S6. SetRank output for inter-ethnic gene set enrichment analyses using the Gene Ontology, KEGG and Reactome databases.

Supplemental_Table_S7. BRIdGE eQTL output for the gene-environment interactions under ASA and DMSO conditions. Mod1=ASA-only eQTL, Mod2=DMSO-only eQTL, Mod3= eQTL with unequal effects, Mod4=eQTL with equal effects.

Supplemental_Table_S8. SetRank output showing all significantly enriched pathways for the differential expression analysis using the Gene Ontology, KEGG and Reactome databases.

Data Access

All raw and processed sequencing data generated in this study have been submitted to the NCBI Gene Expression Omnibus under accession number GSE239425.