77067613886GeneGeneGene0378-11191879-003827866042519195710.1016/j.gene.2016.11.028NIHMS831157ArticleCarcinogenic Effects of Oil Dispersants: a KEGG Pathway-based RNA-seq Study of Human Airway Epithelial CellsLiuYao-Zhong1ZhangLei2§Roy-EngelAstrid M3SaitoShigeki4LaskyJoseph A4WangGuangdi5WangHe6Dept. of Global Biostatistics and Data Science, Tulane University School of Public Health and Tropical Medicine, New Orleans, LA, USACenter for Genetic Epidemiology and Genomics, School of Public Health, Medical College, Soochow University, Su Zhou, ChinaDept. of Epidemiology, Tulane University School of Public Health and Tropical Medicine, New Orleans, LA, USADept. of Medicine, Tulane University School of Medicine, New Orleans, LA, USADept. of Chemistry, RCMI Cancer Research Center, Xavier University of Louisiana, New Orleans, LA, USADept. of Chronic Respiratory Diseases, School of Health Sciences, University of Newcastle, Callaghan, AustraliaCorresponding author: Yao-Zhong Liu, M.D., Ph.D., Dept. of Global Biostatistics and Data Science, Tulane University School of Public Health and Tropical Medicine, 1440 Canal Street, Suite 2001, New Orleans, LA 70112, USA, yliu8@tulane.edu, phone: 504-988-1888
co-first author
2311201616112016202201720220186021623
The health impacts of the BP oil spill are yet to be further revealed as the toxicological effects of oil products and dispersants on human respiratory system may be latent and complex, and hence difficult to study and follow up. Here we performed RNA-seq analyses of a system of human airway epithelial cells treated with the BP crude oil and/or dispersants Corexit 9500 and Corexit 9527 that were used to help break up the oil spill. Based on the RNA-seq data, we then systemically analyzed the transcriptomic perturbations of the cells at the KEGG pathway level using two pathway-based analysis tools, GAGE (generally applicable gene set enrichment) and GSNCA (Gene Sets Net Correlations Analysis). Our results suggested a pattern of change towards carcinogenesis for the treated cells marked by upregulation of ribosomal biosynthesis (hsa03008) (p = 1.97e-13), protein processing (hsa04141) (p = 4.09e-7), Wnt signaling (hsa04310) (p = 6.76e-3), neurotrophin signaling (hsa04722) (p = 7.73e-3) and insulin signaling (hsa04910) (p = 1.16e-2) pathways under the dispersant Corexit 9527 treatment, as identified by GAGE analysis. Furthermore, through GSNCA analysis, we identified gene co-expression changes for several KEGG cancer pathways, including small cell lung cancer pathway (hsa05222, p = 9.99e-5), under various treatments of oil/dispersant, especially the mixture of oil and Corexit 9527. Overall, our results suggested carcinogenic effects of dispersants (in particular Corexit 9527) and their mixtures with the BP crude oil, and provided further support for more stringent safety precautions and regulations for operations involving long-term respiratory exposure to oil and dispersants.
BP oil spillRNA-seqairway epithelial cellsdispersantsoilcarcinogenesisGAGEGSNCAKEGGIntroduction
The year 2016 marked the sixth anniversary of the Deepwater Horizon oil spill, the largest man-made disaster in the history of petroleum industry. The sheer scale of this disaster is marked by its long duration (lasting for ~3 months), the gigantic volume of crude oil (210 million gallons) spilled and dispersants (1.8 million gallons) applied and the large number of workers (>50,000) involved in the cleaning operation (Hayworth and Clement 2012; Kujawinski et al. 2011).
The full impacts of the disaster to the environment and marine and human lives have yet to be fully unveiled. Specifically, the long term health impacts of the BP oil spill to the >50,000 workers involved in the cleaning operation have not been well characterized and followed up, although limited data on other smaller scale oil spills (e.g., the Prestige oil spill) did suggest that involvement in oil spill cleaning operations may cause persistent respiratory symptoms (Zock et al. 2012), long-lasting airway oxidative stress (Rodriguez-Trigo et al. 2010), and systemic genetic effects (Laffon et al. 2006; Perez-Cadahia et al. 2007; Perez-Cadahia et al. 2008a; Perez-Cadahia et al. 2008b). In addition, the oil-dispersant mixtures may contain potentially mutagenic/carcinogenic chemicals including PAH, benzene, and benzene derivatives (Rodrigues et al. 2010; Saeed and Al-Mutairi 1999). More importantly, chemical components in the mixtures may enhance each other to induce harmful effects synergistically. The mechanism for inhalation of hazardous substances during the oil spill was also proposed through models where inhalable aerosols that contain dispersed oil can be formed on the sea surface (Ehrenhauser et al. 2014; Middlebrook et al. 2012).
To characterize the effects and mechanisms of oil spill to human lung health at the molecular level, we hypothesize that oil spill chemicals (i.e., oil, dispersant, or their mixtures) may have significant effects on respiratory cells, which can be detected at the transcriptomic level. To test this hypothesis, we performed an RNA-seq study of human airway epithelial cells treated with BP crude oil, oil dispersants (Corexit 9500 and 9527) and their mixtures (Liu et al. 2016). Through that study we identified a large number of genes differentially expressed due to the treatments, suggesting significant transcriptomic perturbations of the cells caused by the toxicological effects of the oil and oil cleaning chemicals. The findings provided a strong support to our hypothesis. Furthermore, by annotating the differentially expressed genes using DAVID analysis (Dennis, Jr. et al. 2003), our study (Liu et al. 2016) suggested several key biological processes affected by the chemicals, including degradation of the cell junction, enhanced immune response, decreased local steroid biosynthesis and enhanced angiogenesis. These identified biological processes are consistent with some of the pathological features for several common lung diseases, such as COPD (Faner et al. 2013; Holtzman et al. 2014), asthma (Ribatti et al. 2009) and cystic fibrosis (Georas and Rezaee 2014; Heijink et al. 2014; Rezaee and Georas 2014). Therefore our study (Liu et al. 2016) not only detected the existence but also preliminarily characterized potential molecular mechanisms for the toxicological effects of oil and oil cleaning chemicals. Overall our findings (Liu et al. 2016) provided compelling evidence for the potential lung health impact of the BP oil spill on those workers involved in the cleaning operation.
One of the key limitations of our previous study (Liu et al. 2016) is that the analysis was largely performed at single gene level. Although we did perform analysis at the level of GO functional terms, the analysis was still based on the results from single gene differential expression analysis. For example, only those genes that achieved a p value of <0.05 in single gene differential expression analysis were submitted to GO analysis. Such a p value threshold, although commonly used, may be arbitrary in a genomic study as those genes that did not achieve the p value <0.05 may also contribute to the effects of biological significance. Hence annotation based only on those genes with a small p value may lose some sensitivity to capture the key signatures of the toxicological effects on the cells. Furthermore, genes often work correlatively and collaboratively in pathways and functional modules. Differential expression analysis at the single gene level as in our previous study (Liu et al. 2016) ignored such correlative relationship, which again may have missed some important transcriptomics signatures and failed to capture some key functional variations of the transcriptome.
To alleviate the problems/limitations of our previous study and take full advantage of this valuable RNA-seq dataset (Liu et al. 2016), here we performed a KEGG pathway-based analysis using the two well-developed software packages, GAGE (generally applicable gene set enrichment) (Luo et al. 2009) and GSNCA (Gene Sets Net Correlations Analysis) (Rahmatallah et al. 2014). While the findings from our new analysis agree with some of the previous study (Liu et al. 2016), the findings provided distinct new clues. A number of pathways, including ribosomal biosynthesis, protein processing, Wnt signaling, neurotrophin signaling and insulin signaling pathways, were all upregulated at the whole pathway level (mainly by the dispersant 9527). Importantly, upregulation of these pathways was all closely related with cancers, including lung cancer, in previous molecular and epidemiological studies (Mazieres et al. 2005; Poloz and Stambolic 2015; Prakash et al. 2010; Zhou et al. 2015). Moreover, by GSNCA analysis (Rahmatallah et al. 2014), we further identified a number of cancer-related pathways, including the small cell lung cancer pathway, whose gene co-expression was changed due to the exposure to the oil/dispersant treatment, especially the mixture of dispersant 9527 and oil. Overall, our study here by analyzing the transcriptomics signals at the KEGG pathway level has revealed potential carcinogenic effects of dispersant 9527 and its mixture with crude oil. The findings provided further evidence for the health hazards of oil and oil dispersants to the respiratory system.
MethodsExperimental methods
The methods for generating the RNA-seq data were detailed in our previous study (Liu et al. 2016). Briefly, human airway epithelial cells (BEAS-2B cells, ATCC® CRL-9609™) were grown under six treatments using WAF (water accommodated fraction) of the following, i.e., the BP crude oil (abbreviated as “oil”), dispersant Corexit 9500 (abbreviated as “9500”), dispersant Corexit 9527 (abbreviated as “9527”), the mixture of oil + 9500 (abbreviated as “oil+9500”), the mixture of oil + 9527 (abbreviated as “oil+9527”), and water (abbreviated as “control”). There were three cell samples (biological replicates) for each treatment and total RNA extraction followed by RNA-seq experiments was performed on each sample. The RNA-seq experiments were performed at Omega Bioservices (Norcross, GA). The RNA-seq data was submitted to GEO under the accession number (GSE70909). Differential expression analysis was performed between each of the first five treatment groups vs. the control group.
The BP crude oil was kindly provided by The Architecture, Engineering, Consulting, Operations and Management Company (AECOM, Los Angeles, CA). This oil was obtained from the site of the Macondo well during the BP Oil Spill disaster. Commercially available Corexit EC9500A and EC9527 dispersants were kindly provided by a contract between Nalco/Exxon Energy Chemicals, L.P. (Sugar Land, TX, USA) and Tulane University (New Orleans, USA). The dispersants were provided as liquid solutions ready for use.
GAGE analysis
Following the Bioconductor workflow http://www.bioconductor.org/help/workflows/rnaseqGene/, the raw RNA-seq data (GSE70909) was analyzed to generate the raw count matrix, based on which, we used DESeq2 (Love et al. 2014) to perform differential expression analysis between a treatment (e.g., oil) and the control.
DESeq2 package requires “raw” counts of sequencing reads as the starting point for differential expression analysis (Love et al. 2014). Therefore, before submitted to the program for analysis, the count matrix was not normalized (which is explicitly required by the software) (Love et al. 2014). However, during the analysis procedures of DESeq2, normalization did occur in the modeling process, where the read count for gene i and sample j was modeled as a negative binomial distribution with mean μij and dispersion αj, and μij = sijqij where qij is the raw read count and sij is a size factor that normalizes differences in sequencing depth between samples and other sources of technical biases, such as GC content and gene length (Love et al. 2014). The size factor sij was estimated with the “estimateSizeFactor()” function and the dispersion αj estimated with the “DESeq()” function.
We ran the “DESeqDataSetFromMatrix” function, with the count matrix for all 18 samples (3 for each of the 5 treatments and 3 for the control) and the design matrix as input data, which produced an R object for the downstream differential expression analysis. Including count data for all the 18 samples (rather than only 3 samples from a treatment and the 3 control samples) is more robust for estimating parameters (such as the size factor and dispersion). For differential expression analysis to compare a certain treatment with the control, we used “contrast()” function, e.g., by defining “contrast = c (“treatment”, “oil”, “control”)”. The result file that contain p values and fold changes for each gene was generated with the “results()” function.
The log2 fold changes for all the genes from the differential expression analysis (which is a column in the result file) were submitted as an input file for GAGE analysis (Luo et al. 2009). Note that we submitted the log2 fold changes for all the genes, not just those genes that achieved a certain significant p values (such as p < 0.05). GAGE (Luo et al. 2009) then uses the information of fold change for each gene to obtain mean and standard deviation of fold changes for a gene set (pathway) as well as for the background (the whole transcriptome) and generate a t test statistic and p value for a comparison in fold change between a gene set and the background. Essentially, if there is significant difference in fold change between a gene set and the background, an extreme t statistic and a small p value will be achieved. As the fold change involves information on up- or downregulation (i.e., fold change >1 for up- and <1 for downregulation), a pathway identified by GAGE will also be indicated as up- or downregulated. As a nice feature of GAGE, a KEGG pathway identified as significantly differentially expressed can be visualized in a KEGG pathway plot using PathView (Luo and Brouwer 2013), where an upregulated gene will be shown in red and a downregulated gene in green and an unregulated gene in grey (Figure 1). Therefore, if, for example, in a pathway plot, the majority of genes are shown in red, the pathway’s upregulation can be visually assessed (Figure 1).
GSNCA analysis
GSNCA (Rahmatallah et al. 2014) offers multivariate nonparametric statistical methods testing difference in correlation structure of a gene set (pathway) between two conditions. As a key feature, GSNCA quantitatively characterizes the importance of each gene in a correlation network, assigning each gene a weight (wi for gene i) that is proportional to the gene’s cross-correlation with all the other genes. Thus, genes with high cross-correlations (e.g., the hub gene) will normally have a high weight that may indicate their regulatory importance. Based on cross-correlations among different genes, correlation structure of a pathway can be constructed and a statistical test can be performed to test difference of the structure in different conditions (e.g., treatment vs. control). The correlation structure for a pathway under a specific condition usually features a group of high weight genes (including the hub gene), which might suggest the central regulatory roles exerted by these genes.
For GSNCA analysis, we first downloaded gene annotation file from ENSEMBL and extracted gene length information. The raw count matrix was filtered to remove rows of zero counts and rows of no gene annotations. For each element of the count matrix (each gene from each sample), count per million reads (cpm) was calculated to standardize for library size and those rows (genes) with a cpm < 0.1 were further removed, followed by calculation of rpkm (reads per kilobase per million reads) to further standardize for gene length. KEGG pathway list was downloaded and then further partitioned into single KEGG pathway file.
Using the GSAR package (Rahmatallah et al. 2012) we performed GSNCA analysis for each of the KEGG pathways (as a single gene set). The analysis is a multivariate differential co-expression test that accounts for the correlation structure between genes. The test assigns weight factors ws to genes under one condition and adjusts these weights simultaneously such that equality is achieved between each genes' weight and the sum of its weighted absolute correlations (rij)with other genes in a gene set of p genes
wi=∑j≠iwj∣rij∣where1≤i≤p.
The test statistic WGSNCA is given by the 1st norm of difference between the scaled weight vectors w(1) and w(2) (each vector is multiplied by its norm) for the two conditions.
WGSNCA=∑i=1p|wi(1)-wi(2)|
This test statistic was used to test the hypothesis:H0: WGSNCA = 0 vs. H0: WGSNCA ≠ 0. The P-values for the test statistic are obtained by comparing the observed value of the test statistic to its null distribution, which is estimated using a permutation approach (through permuting samples’ class labels). Specifically, the p value is calculated as
p=b+1nperm+1, where b is the number of permutations with a more extreme statistic WGSNCA than the observed statistic and nperm is the total number of permutations. For each specific test, we permute for 10,000 times, the most significant empirical p value that can be possibly achieved is 9.99e-5 (meaning that among 10,000 permutations no permutation has achieved a more extreme test statistic than the observed one).
Results
From GAGE analysis, we identified 4 KEGG pathways that were downregulated by the oil treatment, 7 pathways significantly regulated (including 2 up- and 5 downregulated) by 9500 treatment, 27 pathways upregulated by the 9527 treatment, and 8 pathways upregulated by the 9500 + oil treatment. The results are presented in Table 1. For the statistical criterion of significant regulation, we used an FDR q value <0.10 as a cutoff threshold.
With oil treatment, among the most significantly regulated pathways are downregulation of actin cytoskeletion (hsa04810) (p = 6.53e-5) and adherens junction (hsa045020) (p = 9.69e-5) pathways. Treatment with 9500 caused upregulation of ribosome (hsa03010) (p = 2.22e-11) and antigen processing and presentation (hsa04612) (p = 1.22e-3) and downregulation of adherens junction (hsa045020) (p = 1.06e-4) pathways, among the most significantly regulated pathways. (Table 1)
The 9527 treatment leads to the largest number of pathways that are significantly regulated (with a total of 27 pathways, all upregulated). At the top of the list are the ribosome biogenesis (hsa03008) (p = 1.97E-13) and protein processing (hsa04141) pathways. In addition, Wnt signaling (hsa04310) (p = 6.76E-3), neurotrophin signaling (hsa04722) (p = 7.73e-3) and insulin signaling (hsa04910) (p = 1.16E-02) are also among the upregulated pathways.
Overall, in terms of statistical significance, the most significantly upregulated pathways are ribosome biogenesis (hsa03008) and ribosome (hsa03010) under 9527, 9500 or 9500+oil treatments. The most significantly downregulated pathways are regulation of actin cytoskeleton (hsa04810) and adherens junction (hsa04520) induced by oil or 9500 treatments. As a key feature, a large number of pathways were upregulated under 9527 or 9500+oil treatments, with the majority of upregulated pathways related to RNA metabolism (Table 1).
To illustrate the significance of the two most regulated pathways (hsa03008 and hsa04810) mentioned above, we used PathView to visualize the up- (in red color) and downregulation (in green color) of each gene in the context of the whole KEGG pathway plots (Figure 1). Unregulated genes are shown in grey. If the majority of the colored genes are red/green, then that pathway is up/downregulated as a whole. As examples, note the prevailing red color for the ribosome biogenesis pathway upregulated by the 9527 treatment (Figure 1B) and the prevailing green color for the actin cytoskeleton pathway downregulated by the oil treatment (Figure 1A).
Using GAGE (Luo et al. 2009), we also performed differential expression analysis at the GO level to corroborate the KEGG pathway results. The detailed GO terms significantly regulated are not presented here due to the large number of terms involved and the space limitation. Instead we present the number of regulated terms and some example GO terms in Table 2.
The GO results are generally consistent with the KEGG results. The impact of oil treatment is featured by downregulation of anchoring junction (GO:0070161) and adherens junction (GO:0005912). The 9500 treatment effect is characterized by upregulation of inflammatory response (GO:0006954) and innate immune response (GO:0045087), in addition to simultaneous downregulation of anchoring junction (GO:0070161) and adhereins junction (GO:0005912). 9527 treatment induced the largest significantly regulated GO terms, with the vast majority of the terms being upregulated, especially the upregulation of the BP (biological process) terms (n = 159), which again features ribosome biogenesis (GO:0042254), translation (GO:0006412) and the related terms.
Using GSNCA analysis (Rahmatallah et al. 2014), we identified 64 KEGG pathways that have differential co-expression patterns (i.e., the between-gene correlation structure) in treatment and control groups (Table 3), which achieved a p value of 9.99e-5, the smallest p value that can be possibly achieved in 10,000 permutations. This p value is significant even after Bonferroni correction for testing a total of 187 KEGG pathways (as the Bonferroni corrected significance level is 0.05/186 = 2.67e-4). Many of these pathways overlap with those pathways identified by GAGE analysis, such as regulation of actin cytoskeleton (hsa04810), endocytosis (hsa04144), antigen processing and presentation (hsa04612) and insulin signaling pathway (hsa04910). The most interesting pathways identified that are distinct from the GAGE analysis are a number of KEGG pathways related to cancer. These pathways (bolded and italicized in Table 3) include small cell lung cancer (hsa05222), pathways in cancer (hsa05200), prostate cancer (hsa05215), endometrial cancer (hsa05213), and basal cell carcinoma (hsa05217). Among the treatments, oil+9527 appears to be the one most frequently associated with these cancer-related pathways as oil+9527 is either the only one or one of the conditions associated with these cancer pathways.
As an example (shown in Figure 2), the major change in the small lung cancer pathway due to the oil+9527 treatment is that a number of genes in this pathway that have high or medium weights in controls become genes with a very low weight, suggesting the loss of their correlative relationship with other genes and their regulatory roles. Under the oil+9527 treatment, the pathway has hence become functionally dominated by a smaller subset of higher weight genes as compared with controls. Using DAVID (Dennis, Jr. et al. 2003) to annotate the subset of the higher weight genes produces a top annotation cluster (with an enrichment score of 7.27) that contains 8 KEGG cancer pathway terms, e.g., hsa05215:Prostate cancer (p = 5.60e-14), hsa05220:Chronic myeloid leukemia (p = 7.65e-12), hsa05212:Pancreatic cancer (p = 1.61e-10), hsa05223:Non-small cell lung cancer (p = 2.85e-07), etc. (note: p values are Bonferroni corrected p values). In contrast, if we annotate those lower weight genes with DAVID (Dennis, Jr. et al. 2003), the top annotation cluster (with an enrichment score of 4.95) does not contain cancer terms. These results suggest that under the oil+9527 treatment the pathway is now functionally shifted to a smaller set of genes that are further more enriched with cancer pathways.
Overall, at the whole KEGG pathway level, the most pronounced regulated pathways are ribosomal biosynthesis and its related pathways. Gene co-expression changes were also observed for a number of cancer pathways, most relevantly, the small cell lung cancer pathway. While 9527 treatment has the largest number of pathways significantly regulated, including not only ribosomal biosynthesis but also several signaling pathways, such as Wnt signaling, 9527 + oil appears to be the most common treatment to change gene co-expression patterns of cancer related pathways. Therefore, the overall effects appear to be most pronounced and enriched for 9527 and 9527+oil treatments.
Discussion
In this study, we performed pathway-based analysis on RNA-seq data of the human airway epithelial cells treated with oil and dispersants 9500 and 9527. Pathway-based analysis of gene expression data is more informative and biologically meaningful than individual gene based analysis since very often genes work correlatively and orchestrally as a “team” (a functional module or a functional pathway). For some genes, their moderate changes of expression at the individual gene level may be insignificant under single gene statistical analysis framework yet they may be part of a biologically significant process by contributing to the process in an orchestrated manner. In such case, it is more reasonable that the analytical unit for gene expression analysis be a group of (functionally related) genes rather than individual transcripts. Statistically speaking, it can also alleviate effectively the common problem of multiple testing and “winner’s curse” (Ioannidis 2008) in a genomic study. Therefore, here we used GAGE (Luo et al. 2009) to identify differential expression at the whole pathway level due to oil and/or dispersant treatments.
Given the intrinsic correlation among the genes in a pathway, sometime a pathway is not up or down-regulated at the whole pathway level but may change in a more delicate manner, which is the alteration of the inner correlative structure or the co-expression patterns. For example, under certain conditions, some genes (e.g., the so called hub genes) that have a large number of correlative relationships with other members of a pathway may lose the “network hub” status and be replaced by other genes as new hubs. If that happens, the pathway’s function may also changes as the regulatory roles asserted by different hub genes may be different. Here we used GSNCA (Rahmatallah et al. 2014) to capture such delicate changes at the whole pathway level and identified a large number of pathways with changed co-expression patterns under the oil and/or dispersant treatments.
Some of the results from the present study are consistent with what our previous study reported (Liu et al. 2016) based on individual genes. For instance, downregulation of the cell junction (e.g., adherens junction and anchoring junction) and upregulation of inflammatory response and immune response due to oil and/or dispersants treatment were observed in both studies. These findings again indicated the potential detrimental effects on cellular structure integrity and activation of immune response by the oil/dispersant treatment, lending mechanistic support for the observed respiratory symptoms in workers involved in previous oil cleaning operations (Zock et al. 2012).
However, the key distinct features observed in this study are upregulated ribosomal biosynthesis, protein processing, Wnt signaling, neurotrophin signaling and insulin signaling pathways upon treatment with the dispersant 9527. Importantly the observation of such a coordinated upregulation of these pathways may indicate a tendency towards cancer development for the cells.
Increased ribosomal biosynthesis has been linked to oncogenic signals. For example, a study showed that c-Myc oncogenic activity required robust ribosome biogenesis and protein synthesis and when the upregulated protein synthesis was reduced to normal level by deleting one allele of RPL24 (Ribosomal Protein L24), the oncogenic potential of c-Myc was markedly abolished (Barna et al. 2008). Several tumor suppressors, such as RB and p130, inhibit tumor cell growth and proliferation by interfering with ribosome biogenesis and thus restricting global protein synthesis (Cavanaugh et al. 1995; Ciarmatori et al. 2001; Hannan et al. 2000). Ribosomal biosynthesis pathway has been proposed as a potential target for cancer treatment (Zhou et al. 2015).
Activation of Wnt signaling pathway was associated with lung cancer development (Mazieres et al. 2005; Stewart 2014) and inhibition of Wnt signaling was shown to inhibit lung cancer cell growth both in vitro and in vivo (Kim et al. 2007). Neutrotropin signaling has also been shown to promote lung cancer development (Prakash et al. 2010) as high levels of TrkA and TrkB have been found in human lung adenocarcinoma and squamous cell carcinomas (Ricci et al. 2001), and TrkB has been found in small cell lung cancers and atypical carcinoids (Ricci et al. 2005).
Insulin signaling has been well known for its crucial role in cancer development (as reviewed in (Poloz and Stambolic 2015)), including lung cancer (Dziadziuszko et al. 2008; Ferguson et al. 2012; Frisch et al. 2015). Hyperinsulinemia is linked to the development of lung cancer (Ferguson et al. 2012) as well as other types of cancers (Calle et al. 2003; Goodwin et al. 2002; Renehan et al. 2008; Stewart et al. 2009). Insulin signaling through insulin receptor A has direct oncogenic and anti-apoptotic effects on cancer cells (Cox et al. 2009; Frasca et al. 1999; Kalli et al. 2002; Law et al. 2008; Novosyadlyy et al. 2010; Papa et al. 1990). Insulin signaling has also been a target for cancer treatment as hyperinsulinemia drugs, such as metformin, have proven to be valuable for cancer therapy (Bosco et al. 2011; Evans et al. 2005; Goodwin et al. 2011; Vissers et al. 2015).
Through GSNCA analysis we also identified several cancer-related pathways, including small cell lung cancer pathway, which have gene co-expression changes due to oil/dispersant treatment. For example, under oil+9527 treatment, the higher weight genes (those with more important regulatory roles) were reduced to a smaller subset of genes that are further enriched for cancer development (Figure 2).
Not surprisingly, our findings indicate that exposure to Corexit 9527-oil mixture has the most nocive effect on cells with significant upregulation of cancer pathways. The toxic effect of Corexit 9527 has been previously established. In particular, Corexit 9527 contains the toxin 2-Butoxyethanol (NIH US National Library of Medicine Toxicology Data Network https://toxnet.nlm.nih.gov/cgi-bin/sis/search/a?dbs+hsdb:@term+@DOCNO+7838) that was shown to cause liver, kidney, lung, nervous system, and blood disorders among cleanup crews in Alaska following the 1989 Exxon Valdez oil spill (http://www.takepart.com/article/2013/04/17/corexit-deepwater-horizon-oil-spill). Although approved by the EPA, formulations of Corexit 9500 and 9527 were banned from use in the United Kingdom in 1998 because laboratory tests found them harmful to marine life that inhabits rocky shores (http://www.biologicaldiversity.org/programs/public_lands/energy/dirty_energy_development/oil_and_ga s/gulf_oil_spill/dispersants.html). Regardless of the risks, these dispersants were still used in the BP oil spill site, and by June 1 2010, 800,000 to 900,000 gallons of Corexit 9527 had already been sprayed (https://www.whistleblower.org/gulftruth).
The above findings, both from differential expression analysis at the pathway level using GAGE (Luo et al. 2009) and from gene co-expression analysis using GSNCA (Rahmatallah et al. 2014), suggest a significant pattern of change towards cancer development in human airway epithelial cells under oil/dispersant treatment, especially the 9527 treatment. This, together with our previous study (Liu et al. 2016), provides compelling evidence for the toxicological effects, especially carcinogenic effects, of oil and dispersants and further reveals the potential health impacts of the recent BP oil spill on the lung health for those workers involved in the rescue mission. Accordingly, more strict regulations should be established to protect the personnel routinely working on the sites of oil spill and even those working in the refineries who may be exposed to oil products on a daily basis. Precautions should also be exercised on using oil dispersants, especially the Corexit 9527, to contain an oil spill in the future.
This study was supported by NIOSH (National Institute for Occupational Safety and Health) Pilot Projects Research Training Program of Southwest Center for Occupational and Environmental Health. AMR-E is also supported by the by National Institutes of Health (NIH) P20GM103518/ P20RR020152. GW is supported by 2G12MD007595. Its contents are solely the responsibility of the authors and do not necessarily represent the official views of NIH.
This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
Abbreviation listGAGE
generally applicable gene set enrichment
GSNCA
Gene Sets Net Correlations Analysis
WAF
water accommodated fraction
DAVID
Database for Annotation, Visualization, and Integrated Discovery
Reference ListBarnaMPusicAZolloOCostaMKondrashovNRegoERaoPHRuggeroD2008Suppression of Myc oncogenic activity by ribosomal protein haploinsufficiencyNature456722497197519011615BoscoJLAntonsenSSorensenHTPedersenLLashTL2011Metformin and incident breast cancer among diabetic women: a population-based case-control study in DenmarkCancer Epidemiol Biomarkers Prev20110111121119073CalleEERodriguezCWalker-ThurmondKThunMJ2003Overweight, obesity, and mortality from cancer in a prospectively studied cohort of U.S. adultsN Engl J Med348171625163812711737CavanaughAHHempelWMTaylorLJRogalskyVTodorovGRothblumLI1995Activity of RNA polymerase I transcription factor UBF blocked by Rb gene productNature37465181771807877691CiarmatoriSScottPHSutcliffeJEMcLeesAAlzuherriHMDannenbergJHteRHGrummtIVoitRWhiteRJ2001Overlapping functions of the pRb family in the regulation of rRNA synthesisMol Cell Biol21175806581411486020CoxMEGleaveMEZakikhaniMBellRHPiuraEVickersECunninghamMLarssonOFazliLPollakM2009Insulin receptor expression by human prostate cancersProstate691334018785179DennisGJrShermanBTHosackDAYangJGaoWLaneHCLempickiRA2003DAVID: Database for Annotation, Visualization, and Integrated DiscoveryGenome Biol453DziadziuszkoRCamidgeDRHirschFR2008The insulin-like growth factor pathway in lung cancerJ Thorac Oncol3881581818670298EhrenhauserFSAvijPShuXDugasVWoodsonILiyana-ArachchiTZhangZHungFRValsarajKT2014Bubble bursting as an aerosol generation mechanism during an oil spill in the deep-sea environment: laboratory experimental demonstration of the transport pathwayEnviron Sci Process Impacts161657324296745EvansJMDonnellyLAEmslie-SmithAMAlessiDRMorrisAD2005Metformin and reduced risk of cancer in diabetic patientsBMJ33075031304130515849206FanerRCruzTAgustiA2013Immune response in chronic obstructive pulmonary diseaseExpert Rev Clin Immunol9982183324070046FergusonRDNovosyadlyyRFierzYAlikhaniNSunHYakarSLeroithD2012Hyperinsulinemia enhances c-Myc-mediated mammary tumor development and advances metastatic progression to the lung in a mouse model of type 2 diabetesBreast Cancer Res141R822226054FrascaFPandiniGScaliaPSciaccaLMineoRCostantinoAGoldfineIDBelfioreAVigneriR1999Insulin receptor isoform A, a newly recognized, high-affinity insulin-like growth factor II receptor in fetal and cancer cellsMol Cell Biol1953278328810207053FrischCMZimmermannKZillessenPPfeiferARackeKMayerP2015Non-small cell lung cancer cell survival crucially depends on functional insulin receptorsEndocr Relat Cancer22460962126113601GeorasSNRezaeeF2014Epithelial barrier function: at the front line of asthma immunology and allergic airway inflammationJ Allergy Clin Immunol134350952025085341GoodwinPJEnnisMPritchardKITrudeauMEKooJMadarnasYHartwickWHoffmanBHoodN2002Fasting insulin and outcome in early-stage breast cancer: results of a prospective cohort studyJ Clin Oncol201425111773152GoodwinPJStambolicVLemieuxJChenBEParulekarWRGelmonKAHershmanDLHobdayTJLigibelJAMayerIAPritchardKIWhelanTJRastogiPShepherdLE2011Evaluation of metformin in early breast cancer: a modification of the traditional paradigm for clinical testing of anti-cancer agentsBreast Cancer Res Treat126121522020976543HannanKMHannanRDSmithSDJeffersonLSLunMRothblumLI2000Rb and p130 regulate RNA polymerase I transcription: Rb disrupts the interaction between UBF and SL-1Oncogene19434988499911042686HayworthJSClementTP2012Provenance of Corexit-related chemical constituents found in nearshore and inland Gulf Coast watersMar Pollut Bull64102005201422959174HeijinkIvanOAKliphuisNJonkerMHoffmannRTelengaEKloosterKSlebosDJtenHNPostmaDvan den BergeM2014Oxidant-induced corticosteroid unresponsiveness in human bronchial epithelial cellsThorax69151323980116HoltzmanMJByersDEAlexander-BrettJWangX2014The role of airway epithelial cells and innate immune cells in chronic respiratory diseaseNat Rev Immunol141068669825234144IoannidisJP2008Why most discovered true associations are inflatedEpidemiology19564064818633328KalliKRFalowoOIBaleLKZschunkeMARochePCConoverCA2002Functional insulin receptors on human epithelial ovarian carcinoma cells: implications for IGF-II mitogenic signalingEndocrinology14393259326712193537KimJYouLXuZKuchenbeckerKRazDHeBJablonsD2007Wnt inhibitory factor inhibits lung cancer cell growthJ Thorac Cardiovasc Surg133373373717320573KujawinskiEBKido SouleMCValentineDLBoysenAKLongneckerKRedmondMC2011Fate of dispersants associated with the deepwater horizon oil spillEnviron Sci Technol4541298130621265576LaffonBFraga-IrisoRPerez-CadahiaBMendezJ2006Genotoxicity associated to exposure to Prestige oil during autopsies and cleaning of oil-contaminated birdsFood Chem Toxicol44101714172316814914LawJHHabibiGHuKMasoudiHWangMYStratfordALParkEGeeJMFinlayPJonesHENicholsonRICarboniJGottardisMPollakMDunnSE2008Phosphorylated insulin-like growth factor-i/insulin receptor is present in all breast cancer subtypes and is related to poor survivalCancer Res6824102381024619074892LiuYZRoy-EngelAMBaddooMCFlemingtonEKWangGWangH2016The impact of oil spill to lung health-Insights from an RNA-seq study of human airway epithelial cellsGene5781385126692141LoveMIHuberWAndersS2014Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2Genome Biol151255025516281LuoWBrouwerC2013Pathview: an R/Bioconductor package for pathway-based data integration and visualizationBioinformatics29141830183123740750LuoWFriedmanMSSheddenKHankensonKDWoolfPJ2009GAGE: generally applicable gene set enrichment for pathway analysisBMC Bioinformatics1016119473525MazieresJHeBYouLXuZJablonsDM2005Wnt signaling in lung cancerCancer Lett222111015837535MiddlebrookAMMurphyDMAhmadovRAtlasELBahreiniRBlakeDRBrioudeJde GouwJAFehsenfeldFCFrostGJHollowayJSLackDALangridgeJMLuebRAMcKeenSAMeagherJFMeinardiSNeumanJANowakJBParrishDDPeischlJPerringAEPollackIBRobertsJMRyersonTBSchwarzJPSpackmanJRWarnekeCRavishankaraAR2012Air quality implications of the Deepwater Horizon oil spillProc Natl Acad Sci U S A10950202802028522205764NovosyadlyyRLannDEVijayakumarARowzeeALazzarinoDAFierzYCarboniJMGottardisMMPennisiPAMolinoloAAKurshanNMejiaWSantopietroSYakarSWoodTLLeroithD2010Insulin-mediated acceleration of breast cancer development and progression in a nonobese model of type 2 diabetesCancer Res70274175120068149PapaVPezzinoVCostantinoABelfioreAGiuffridaDFrittittaLVannelliGBBrandRGoldfineIDVigneriR1990Elevated insulin receptor content in human breast cancerJ Clin Invest865150315102243127Perez-CadahiaBLaffonBValdiglesiasVPasaroEMendezJ2008aCytogenetic effects induced by Prestige oil on human populations: the role of polymorphisms in genes involved in metabolism and DNA repairMutat Res6531–211712318495522Perez-CadahiaBLafuenteACabaleiroTPasaroEMendezJLaffonB2007Initial study on the effects of Prestige oil on human healthEnviron Int33217618517055056Perez-CadahiaBMendezJPasaroELafuenteACabaleiroTLaffonB2008bBiomonitoring of human exposure to prestige oil: effects on DNA and endocrine parametersEnviron Health Insights2839221572833PolozYStambolicV2015Obesity and cancer, a case for insulin signalingCell Death Dis6e203726720346PrakashYThompsonMAMeuchelLPabelickCMMantillaCBZaidiSMartinRJ2010Neurotrophins in lung health and diseaseExpert Rev Respir Med4339541120524922RahmatallahYEmmert-StreibFGlazkoG2012Gene set analysis for self-contained tests: complex null and specific alternative hypothesesBioinformatics28233073308023044539RahmatallahYEmmert-StreibFGlazkoG2014Gene Sets Net Correlations Analysis (GSNCA): a multivariate differential coexpression test for gene setsBioinformatics30336036824292935RenehanAGTysonMEggerMHellerRFZwahlenM2008Body-mass index and incidence of cancer: a systematic review and meta-analysis of prospective observational studiesLancet371961256957818280327RezaeeFGeorasSN2014Breaking barriers. New insights into airway epithelial barrier function in health and diseaseAm J Respir Cell Mol Biol50585786924467704RibattiDPuxedduICrivellatoENicoBVaccaALevi-SchafferF2009Angiogenesis in asthmaClin Exp Allergy39121815182120085597RicciAGrazianoPMariottaSCardilloGSposatoBTerzanoCBronzettiE2005Neurotrophin system expression in human pulmonary carcinoid tumorsGrowth Factors23430331216338793RicciAGrecoSMariottaSFeliciLBronzettiECavazzanaACardilloGAmentaFBisettiABarboliniG2001Neurotrophins and neurotrophin receptors in human lung cancerAm J Respir Cell Mol Biol25443944611694449RodriguesRVMiranda-FilhoKCGusmaoEPMoreiraCBRomanoLASampaioLA2010Deleterious effects of water-soluble fraction of petroleum, diesel and gasoline on marine pejerrey Odontesthes argentinensis larvaeSci Total Environ40892054205920167351Rodriguez-TrigoGZockJPPozo-RodriguezFGomezFPMonyarchGBousoLCollMDVereaHAntoJMFusterCBarberaJA2010Health changes in fishermen 2 years after clean-up of the Prestige oil spillAnn Intern Med153848949820733177SaeedTAl-MutairiM1999Chemical composition of the water-soluble fraction of the leaded gasolines in seawaterEnviron Int25117129StewartDJ2014Wnt signaling pathway in non-small cell lung cancerJ Natl Cancer Inst1061djt35624309006StewartSTCutlerDMRosenAB2009Forecasting the effects of obesity and smoking on U.S. life expectancyN Engl J Med361232252226019955525VissersPACardwellCRvan de Poll-FranseLVYoungISPouwerFMurrayLJ2015The association between glucose-lowering drug use and mortality among breast cancer patients with type 2 diabetesBreast Cancer Res Treat150242743725762476ZhouXLiaoWJLiaoJMLiaoPLuH2015Ribosomal proteins: functions beyond the ribosomeJ Mol Cell Biol729210425735597ZockJPRodriguez-TrigoGRodriguez-RodriguezEEspinosaAPozo-RodriguezFGomezFFusterCCastano-VinyalsGAntoJMBarberaJA2012Persistent respiratory symptoms in clean-up workers 5 years after the Prestige oil spillOccup Environ Med69750851322539655
Downregulation of actin cytoskeleton by oil treatment and upregulation of ribosome biogenesis by 9527 treatment
Note:
A: Downregulation of actin cytoskeleton by oil treatment
B: Upregulation of ribosome biogenesis by 9527 treatment
Gene co-expression change of small cell lung cancer pathway under oil+9527 treatment
Note:
DAVID (Dennis, Jr. et al. 2003) annotation for the high weight genes resulted in the top annotation cluster (with an enrichment score of 7.27) that contains 8 KEGG cancer pathways, e.g., hsa05215:Prostate cancer (p = 5.60E-14), hsa05220:Chronic myeloid leukemia (p = 7.65E-12), hsa05212:Pancreatic cancer (p = 1.61E-10), hsa05223:Non-small cell lung cancer (p = 2.85E-07), etc. (note: p values are Bonferroni corrected p values), suggesting that the high weight genes are highly enriched for these cancer pathways.
In contrast, DAVID (Dennis, Jr. et al. 2003) annotation for the low weight genes resulted in the top annotation cluster (with an enrichment score of 4.95) that does not contain cancer terms.
The difference in annotation results may indicate that under treatment of oil+9527 this pathway’s regulatory genes (as manifested by those high weight genes) are more concentrated to a subset of genes highly enriched for cancer development.
KEGG pathways significantly regulated by oil and/or dispersants
KEGG pathway name
P value
FDR
Treatment group
Up or down-regulated by treatment
hsa04810 Regulation of actin cytoskeleton
6.53E-05
5.97E-03
Oil
Downregulated
hsa04520 Adherens junction
9.69E-05
5.97E-03
Downregulated
hsa04144 Endocytosis
1.11E-04
5.97E-03
Downregulated
hsa03015 mRNA surveillance pathway
1.61E-03
6.51E-02
Downregulated
hsa03010 Ribosome
2.22E-11
3.60E-09
9500
Upregulated
hsa04612 Antigen processing and presentation
1.22E-03
9.91E-02
Upregulated
hsa04520 Adherens junction
1.06E-04
1.72E-02
Downregulated
hsa00900 Terpenoid backbone biosynthesis
1.84E-03
9.55E-02
Downregulated
hsa04144 Endocytosis
1.95E-03
9.55E-02
Downregulated
hsa03015 mRNA surveillance pathway
2.37E-03
9.55E-02
Downregulated
hsa04810 Regulation of actin cytoskeleton
2.95E-03
9.55E-02
Downregulated
hsa03008 Ribosome biogenesis in eukaryotes
1.97E-13
3.20E-11
9527
Upregulated
hsa03013 RNA transport
2.03E-09
1.64E-07
Upregulated
hsa00970 Aminoacyl-tRNA biosynthesis
3.42E-07
1.66E-05
Upregulated
hsa04141 Protein processing in endoplasmic reticulum
4.09E-07
1.66E-05
Upregulated
hsa03040 Spliceosome
6.09E-07
1.97E-05
Upregulated
hsa03050 Proteasome
4.66E-05
1.26E-03
Upregulated
hsa00240 Pyrimidine metabolism
6.59E-05
1.53E-03
Upregulated
hsa04120 Ubiquitin mediated proteolysis
1.01E-04
2.05E-03
Upregulated
hsa03018 RNA degradation
1.64E-04
2.95E-03
Upregulated
hsa00020 Citrate cycle (TCA cycle)
2.22E-04
3.59E-03
Upregulated
hsa03015 mRNA surveillance pathway
4.09E-04
5.53E-03
Upregulated
hsa04110 Cell cycle
4.10E-04
5.53E-03
Upregulated
hsa04114 Oocyte meiosis
6.65E-04
8.29E-03
Upregulated
hsa00230 Purine metabolism
7.65E-04
8.85E-03
Upregulated
hsa03010 Ribosome
1.07E-03
1.15E-02
Upregulated
hsa03022 Basal transcription factors
1.34E-03
1.35E-02
Upregulated
hsa03020 RNA polymerase
1.75E-03
1.67E-02
Upregulated
hsa00520 Amino sugar and nucleotide sugar metabolism
6.48E-03
5.77E-02
Upregulated
hsa04310 Wnt signaling pathway
6.76E-03
5.77E-02
Upregulated
hsa04722 Neurotrophin signaling pathway
7.73E-03
6.27E-02
Upregulated
hsa04914 Progesterone-mediated oocyte maturation
8.47E-03
6.54E-02
Upregulated
hsa00010 Glycolysis / Gluconeogenesis
1.05E-02
7.71E-02
Upregulated
hsa04910 Insulin signaling pathway
1.16E-02
8.16E-02
Upregulated
hsa00290 Valine, leucine and isoleucine biosynthesis
1.37E-02
9.22E-02
Upregulated
hsa03008 Ribosome biogenesis in eukaryotes
6.17E-11
1.00E-08
9500+oil
Upregulated
hsa00970 Aminoacyl-tRNA biosynthesis
1.11E-07
8.98E-06
Upregulated
hsa03013 RNA transport
3.02E-07
1.63E-05
Upregulated
hsa03010 Ribosome
1.53E-05
6.18E-04
Upregulated
hsa03040 Spliceosome
1.81E-04
5.86E-03
Upregulated
hsa03018 RNA degradation
3.12E-04
8.41E-03
Upregulated
hsa03050 Proteasome
9.20E-04
2.13E-02
Upregulated
hsa04141 Protein processing in endoplasmic reticulum