https://orcid.org/0000-0002-4498-0915
Abstract
Purpose
Benchmark dose (BMD) modeling is used to determine the dose of a stressor at which a predefined increase in any biological effect above background occurs (e.g. 10% increase from control values). BMD analytical tools have the capacity to model transcriptional dose-response data to derive BMDs for genes, pathways and gene ontologies. We recently demonstrated the value of this approach to support various areas of radiation research using predominately ‘in-house’ generated datasets.
Materials and methods
As a continuation of this work, transcriptomic studies of relevance to ionizing radiation were retrieved through the Gene Expression Omnibus (GEO). The datasets were compiled and filtered, then analyzed using BMDExpress. The objective was to determine the reproducibility of BMD values in relation to pathways and genes across different exposure scenarios and compare to those derived using cytogenetic endpoints. A number of graphic visualization approaches were used to determine if BMD outputs could be correlated to parameters such as dose-rate, radiation quality and cell type.
Results
Curated studies were diverse and derived from experiments with varied design and intent. Despite this, common genes and pathways were identified with low and high dose thresholds. The higher BMD values were associated with immune response and cell death, while transcripts with lower BMD values were generally related to the classic DNA damage response/repair processes, centered on TP53 signaling. Analysis of datasets with relatively similar dose-ranges under comparable experimental conditions showed a bi-modal distribution with a high degree of consistency in BMD values across shared genes and pathways, particularly for those below the 25th percentile of total distribution by dose. The median BMD values were noted to be approximately 0.5 Gy for genes/pathways that comprised mode 1. Furthermore, transcriptional BMD values derived from a subset of genes using in vivo and in vitro datasets were in accord to those using cytogenetic endpoints.
Conclusion
Overall, the results from this work highlight the value of the BMD methodology to derive meaningful outputs that are consistent across different models, provided the studies are conducted using a similar dose-range.
Introduction
Over the decades, new ‘omic’ technologies have evolved that enable the identification of early molecular changes that may be precursors of more profound biological effects observed at higher or cumulative doses (Aardema & MacGregor, Citation2002; Collins et al. Citation2003; Borgert Citation2007; Kyrtopoulos, Citation2013; National Toxicology Program Citation2018). As a result, considerable transcriptomic data have been generated that provide rich mechanistic information relevant to radiation exposures (reviewed in UNSCEAR, Citation2012). Recently, global interest in the use of transcriptional profiling for risk assessment in the field of chemical toxicology has led to the development of new high-throughput analytical approaches to assess dose-response data (Chepelev et al. Citation2015; Buesen et al., Citation2017). Nonetheless, application of this type of data to risk assessment has remained limited, especially in the field of radiation protection.
Traditionally, the No-Observed Adverse Effect Level (NOAEL) methodology has been used to analyze transcriptional data to determine the points of departure (PODs) for chemical stressors. As an effective and more modern approach to derive PODs, benchmark dose (BMD) modeling has been developed and serves as an alternative to the NOAEL methodology. Studies in the past decade suggest that the application of BMD modeling to transcriptional data (reviewed in Filipsson et al. Citation2003; Davis et al. Citation2011) provides more precise quantitative estimates of the thresholds of gene and pathway responses, simpler interpretation of large datasets, and allows direct comparison to other data with similar endpoints without the need to have matched doses (Sand et al., Citation2008). BMD modeling is currently being employed in chemical risk assessment to estimate acceptable levels of exposure and, although typically applied to conventional endpoints, newer software has enabled this application to be extended to transcriptomic datasets (Yang et al. Citation2007; Bourdon et al. Citation2013; Thomas et al. Citation2013; Clewell et al. Citation2014; Jackson et al. Citation2014; Chepelev et al., Citation2015; Webster, Chepelev, et al. Citation2015; Webster, Zumbo, et al., Citation2015; Labib et al. Citation2016; Philips et al. Citation2019). Today, BMD analytical tools can readily be used to model transcriptional dose-response data to derive meaningful BMD values for genes, pathways, and gene ontologies. The approach has been used for assessing non-cancer risks in chemical toxicology and is well supported by U.S National Toxicology Program (Tatusova et al. Citation2016; Williams et al. Citation2020).
Although widely applied in chemical toxicology, BMD modeling has not been leveraged to the benefit of radiation–specific risk assessment. It was recently shown, using ‘in-house’ generated transcriptional datasets that the BMD methodology can identify pathway/gene sensitivities across a range of radiation doses (0–8 Gy) and experimental conditions (time-points, cell types) (Chauhan et al. Citation2016) to derive meaningful and comparable values to current human exposure limits derived from gross morphological changes in animals. This was demonstrated for radiation-induced cataract formation in the lens of the eye (Chauhan et al. Citation2018) and UV-induced skin cancer (Qutob et al. Citation2018). To expand on these studies, publicly available transcriptomic repositories were used to retrieve datasets from dose-response radiation studies for BMD analysis. The datasets were compiled, grouped on exposure parameters and analyzed to identify the distribution and reproducibility of the BMD outputs. In addition, the most consistent grouping of genes and pathways across studies were identified and these BMD values were correlated to cytogenetic endpoints.
Materials and methods
Datasets
High quality transcriptional response data from radiation bioeffects studies were downloaded from Gene Expression Omnibus (GEO) (www.ncbi.nlm.nih.gov/geo). Study quality for inclusion was based on filtering to ensure: (1) The study included ≥3 unique doses of radiation not including controls and (2) each dose was paired with ≥3 unique replicates. Downloaded datasets were imported into BMDExpress version 2.30 (Phillips et al. Citation2019) for gene and pathway BMD derivation. A summary of all datasets that met the filtering criteria is provided in .
Table 1. (a) Summary of the doses, dose rate (where applicable), cell type and species, time points, radiation source and experiment type. (b) Summary of studies and their BMD results. Median BMD values for genes and pathways are quoted with their lower and upper fit results ‘[U,L]’. The fraction of best-fit models for linear (Lin.), polynomial (Pol.), exponential (Exp), Hill and Power is also provided.
BMD modeling
Datasets were formatted according to the BMDExpress software requirements, imported into BMDExpress (version 2.30) and pre-filtered using the built-in Williams’ Trend Test, with p-value ≤ .05 and |fold change| value ≥ 1.5. Genes that passed pre-filtering were modeled using the following built-in mathematical functions: Exponential 2, Exponential 3, Exponential 4, Exponential 5, Linear, Polynomial 2, Polynomial 3, Hill, and Power to identify best dose-response relationships. For each dataset, the maximum degree of polynomial models applied was dependent upon the number of radiation doses (excluding control) minus one. The best model was chosen based on the following criteria: (1) a nested chi-square test was performed with a cutoff of 0.05 to choose between Linear and Polynomial models; (2) between the Hill, Power, Linear and Polynomial models, the least complex model was selected with the lowest Akaike Information Criterion; (3) goodness-of-fit p-value ≥ .1; (4) Hill models were flagged if the k-parameter was < 1/3 of the lowest non-zero dose, at which point the next best model with goodness-of-fit p-value > .05 was selected. If such a model was not available, the flagged Hill model was retained and modified to 0.5 of the lowest non-flagged Hill BMD value; (5) Power was restricted to ≥1; (6) maximum iterations was set to 250; (7) confidence interval (CI) was set at 0.95; (8) a benchmark response (BMR) value of 1.349% risk was chosen as this approximates a 10% shift in the gene expression levels relative to background and is commonly used in chemical risk assessment (US EPA. Citation2012). For downstream analysis, genes were considered responsive only if they passed the following filtering criteria: (1) Best BMD ≤ highest dose; (2) Best BMD upper (U)/BMD lower (L) ≤40; (3) Best BMDU/BMD ≤20; (4) Best BMD/BMDL ≤20; (5) Best fit p value ≥.1. US Environmental Protection Agency recommends to report the BMD confidence interval rather than the value of the BMD. The lower bound and the upper bound allow for determining the uncertainty in the BMD estimate.
Using the Defined Category Analysis feature, responding genes were mapped to REACTOME, a curated pathway database (downloaded on December 18, 2018). Promiscuous probes (i.e. those that mapped to >1 gene), those with BMD values greater than the highest tested dose, and those with a goodness-of-fit p-value ≤.1 were removed. Pathways were considered as responsive only if they passed the following filtering criteria: 1) comprised ≥3 genes that passed all post-modeling filters; and 2) ≥5% of the pathway’s genes could be modeled ().
Figure 1. Workflow of steps taken for all datasets subjected to BMD modeling analysis. All normalized data was input into BMD Express, prefiltered, subjected to quality control measures, and run through category analysis. Resulting data was then grouped into probe and pathway data. Reactome (REAC) analysis was used for all pathway analysis. ProbeALL/REACALL = the entire probe/pathway distribution, 25th = first 25 percent of distribution, MODE = first mode of distribution.
Confidence intervals
CIs were computed for BMDs across studies. For each dataset, genes that passed the above described filters were bootstrapped (random sampling with replacements) and a median BMD value was computed. This process was repeated 50,000 times/dataset and the 95% CIs were obtained at 2.5% of the lower end and 97.5% of the upper end. The same process was performed for REACTOME pathways that passed their respective filters on the BMD median values.
Reproducibility assessment across relatively similar studies
A set of studies conducted under similar experimental parameters were selected for reproducibility assessment in terms of the pathway and gene benchmark dose response (). These data were from studies on transcriptomic analysis of blood (in vivo and ex-vivo analyzed at two time-points post-exposure) following irradiation with x-rays or Cs-137; details can be found in . The BMDs were compared across studies at both the pathway and gene level. InteractiVenn (http://www.interactivenn.net/) was used to identify genes and pathways that were common between studies. The analysis was conducted across the total BMD distribution of pathways/genes, the top 25th percentile of responses based on dose and using mode 1 as it is representative of low dose response in comparison to mode 2.
Table 2. (a) Summary of the doses, dose rate (where applicable), species, time points, radiation source and experiment type. (b): Summary of gene BMD results for 100%, 25% and 5% response selections.
Comparison of transcriptional BMD outputs to cytogenetic outputs
For BMD modeling of cytogenetic datasets, previously reported dose-response curves were imported into BMD Express and analyzed as described above. The data were generated ‘in-house’ as per standard biodosimetry practices. Briefly, whole blood was irradiated with 250 kVp X-rays (0–5 Gy) or Cs-137 (0–4 Gy), cultured and mitogen-stimulated 2 h post exposure to radiation. For the dicentric chromosome assay (DCA) cells were fixed after 48 h of incubation, as described previously (Flegal et al. Citation2012). Data collected represent the frequency of dicentric chromosomes per cell. For the cytokinesis block micronucleus (CBMN) assay, cells were blocked in cytokinesis after 44 hours of culture time and fixed after an additional 28 hours, as described previously (McNamee et al. Citation2009). Data collected represent the number of micronuclei per binucleated cell.
Results
Global BMD outputs
Radiation transcriptomic datasets were retrieved from GEO and subjected to BMD analysis. A total of 50 studies were identified that met the selection criteria. These studies were comprised of human, animal, and cell models exposed to ionizing radiation at different dose-ranges, radiation qualities and analyzed at different time-points. Datasets were downloaded and individually analyzed using BMDExpress v2.3 software. The information extracted from the analysis included parameters related to total number of genes modeled, best-fit model, BMD median values including upper and lower limits (BMDU/BMDL) for pathways and genes (). A global assessment of all the datasets showed anywhere from 40 to 7348 genes were modeled depending on the study, reflecting the quality of the dose-response relationship produced from the experimental design and conduct. The best-fit model tended toward polynomial or linear curve fitting (). The BMD median values for genes and pathways across the studies averaged to 2.1 ± 2.1 Gy and 2.6 ± 2.1 Gy respectively (, ). A 2D histogram of the BMD measurement for genes and pathways identified the frequency in the BMD values for a given set of genes/pathways, showing some to be more readily modeled than others (). Common genes and pathways across all studies were identified and BMD values were ranked. The lower BMD values were associated with six pathways related to: TP53 regulation, transcription of cell cycle genes, PRC2 methylation histones and DNA condensation of prophase chromosomes, SIRT1 negatively regulates rRNA expression, activated pkn1 stimulates transcription of androgen receptor regulated genes KLK2 KLK3, and RNA polymerase 1 promoter opening. Genes that comprised these pathways included: IER5, BAX, TRIAP1, MDM2, AEN, ZNF79, XPC, PCNA, DDB2, and POLH. The higher BMD values were generally associated with cell death, apoptosis and immune function pathways such as: immunoregulatory interactions between a lymphoid and a non-lymphoid cell, DNA damage bypass, g1/s transition, translesion synthesis by Y family DNA polymerases bypasses lesions on DNA template, apoptotic execution phase, apoptosis, programed cell death, nucleotide-binding domain and leucine rich repeat containing receptor (NLR) signaling pathways. Genes that were consistently expressed at the higher BMD values included: CDKN1A, CCNG1, MDM2, DDB2, SESN2, PCNA, SESN1, MYC, GDF15, and MAP4K4.
Figure 2. (A) Log10 of the BMD value for genes and pathways (B). Genes and pathways are numbered along the y-axis in order of appearance throughout datasets. The red lines denote the Log10 of the average (median) BMD value across a given group of genes/pathways. The 25-75% quartile range is denoted by the green band, (C) 2D histogram of the Log10 BMD measurement across genes and pathways (D). The color of each bin denotes the frequency the BMD value appeared for a given set of genes/pathways.
Global assessment of BMD values and relation to exposure parameters
A global assessment of gene/pathway responses allowed for the determination of common pathways and genes; however, clear trends in BMD values as a function of exposure conditions (e.g. cell type, dose-rate, dose range, radiation quality and time-point) were not visible. A number of approaches were used to identify if the datasets could be compiled in a way to generate more meaningful clusters based on BMD outputs. A two-dimensional analysis of BMD distribution as a function of the type of radiation exposure (photons, neutrons, alpha particles (Astatine-211(At-211)) across common genes and pathways (), showed some trends as a function of radiation quality, irrespective of the model (e.g. in vitro, in vivo, human study). It should be noted that although trends were observed these may also be related to the range of doses chosen for these studies. The BMD values were similar for Cs-137 and x-rays exposures (except for one Cs-137 study that had a lower dose range), while the alpha-particle exposures had slightly higher BMD values. A further restricted analysis using a simplified heat map facilitated interpretation of the data (). In this analysis, studies were grouped based on radiation quality (x-rays, gamma, neutrons and alpha particles) and tissue type (lung, blood and thyroid), irrespective of any other experimental conditions. The twelve most frequent radiation-responsive pathways across these studies were used to visualize how the BMD outputs varied as a function of exposure variables. All twelve pathways were present in the lung and blood tissue type studies whereas only eight were present in the thyroid tissue study. Generally, across shared pathways blood exposures conducted using animal and in vitro models using x-rays displayed lower BMD values relative to the other tissue types (lung and thyroid). However, it was noted that the average dose of exposure was higher in the thyroid studies (9 Gy) and the radiation was derived from an alpha particle insult using At-211. Although some delineation of BMD values was observed as a function of exposure parameter (e.g. type of radiation and the tissue sensitivity), this analysis highlights that the dose-ranges used in the studies may be driving the BMD values, partially confounding the interpretation of the data. Therefore, a further analysis using studies conducted under relatively similar dose-ranges and experimental conditions was undertaken to see if consistent BMD responses for pathways and genes were observed.
Figure 3. Distribution of BMD values across common genes (A) and pathways (B) for the different radiation qualities.
Figure 4. (A) Average BMD values for different tissues across common responding pathways. (B) Doses across common studies were averaged and plotted as a function of the associated median pathway BMD valuesfor the same pathways as shown in (A). Different marker colors and styles denote cell type and radiation source respectively.
Figure 5. Gene BMD histograms (A) and pathway accumulation plots (B) of studies conducted under similar dose ranges for 6 hr time-point.
Reproducibility of BMD values across similar exposure conditions
Studies conducted for the purposes of supporting biodosimetry offered an opportunity to assess reproducibility of BMD response with respect to gene and pathway activation. These studies were performed across different research centers using Agilent, Operon, or Affymetrix microarray platforms () under relatively similar dose-ranges. Isolated blood cells or cell-lines (ex-vivo) from human or mouse models (in vivo) were exposed to low linear energy transfer radiation at comparable dose-rates. Two time-points were chosen (6 and 24 h) for BMD output comparison (). Across all studies and for the two time-points, the distribution of BMD median gene values was bimodal ( and ). The median BMD values were relatively consistent across the studies, despite differences in the type of model (in vitro vs in vivo) being used (). However, differences in pathway activation were clearly observed from an accumulation plot ( and ). It was noted that dataset GSE10640 identified more pathways relative to the other datasets, possibly due to a high number of modeled probes (, ). Restricting the analysis to genes that represented mode 1 and 25th percentile of the total BMD gene distribution allowed for more reproducible BMD median values, particularly at the 25th percentile range which generated a suitable number of modeled genes (). Shared genes and pathways across these studies showed highly comparable BMD outputs at both time-points irrespective of the study type (in vivo and ex vivo) ( and ; and ).
Figure 6. Gene BMD histograms (A) and pathway accumulation plots (B) of studies conducted under similar dose ranges for 24 hr time-point.
Figure 7. Summary of BMD outputs for studies conducted under similar exposure conditions. Refer to for details on the figure legend. Mean pathway BMD is the mean of the median BMD value for each pathway. Error bars represent BMDL/BMDU values.
Figure 8. Probe BMD distributions for individual probes common to mode 1 across studies conducted under similar exposure conditions at (A) 6- and (B) 24-hour time point. The x-axis represents the median BMD (Gy) and the error bars represent BMDL/BMDU values.
Figure 9. Pathway BMD distributions for pathways common to mode 1 across studies conducted under similar exposure conditions with a 6-hour (A) and 24-hour (B) time point. Error bars represent BMDL/BMDU values.
Figure 10. A comparison of transcriptional BMD values to cytogenetic-derived BMD values.
Table 3. (a) Summary of the gene and pathway BMD results for mode 1. (b) Summary of the gene and pathway BMD results for mode 2.
Table 4. (a) Summary of frequent genes, the studies from which they were measured and the respective gene BMD values. (b) Summary of frequent pathways, the studies from which they were measured and the respective gene BMD values.
Transcriptional BMD vs cytogenetic BMDs
To determine how BMD values compare with cytogenetic endpoints such as dicentric and micronucleus formation, dose-response data previously published by Flegal et al. (Citation2012) and McNamee et al. (Citation2009) were extracted and analyzed using the BMD approach. The data were derived from human blood ex vivo irradiated and exposed to ionizing radiation using X-rays or Cs-137 at 2 h. This exposure scenario was comparable to the transcriptional studies represented in the reproducibility analysis. Cytogenetic-derived BMD values were identified from the dose-response curves to be 0.5 Gy across the two endpoints (). These values corresponded to the median BMD values of mode 1 and the 25th percentile of the total BMD response from the transcriptional studies.
Discussion
BMD modeling provides a means to quantify transcriptional dose-response data by identifying the lowest dose at which a predefined change occurs (e.g. 10%) in a molecular response (Crump Citation1984; Yang et al. Citation2007). Conceptually, the BMD approach should allow comparison of diverse datasets that have used varied platforms and experimental designs to help derive thresholds of gene and pathway activation that can be interpreted with confidence (Haber et al. Citation2018). Previous work in our laboratory provided some evidence of a correlation between transcriptional and apical endpoint BMD values in the area of ocular damage to the lens of the eye and UV-induced erythema (Chauhan et al. Citation2018; Qutob et al. Citation2018). In addition, the methodology was shown to produce comparable transcriptional BMDs across studies using similar dose-ranges. To further evaluate the scope and limits of the BMD approach in the context of diversified radiation studies, herein we conducted a meta-analysis using datasets deposited in a publicly available repository (GEO). The selected studies varied experimentally in exposure conditions with respect to radiation qualities, time-points, cell-types, dose ranges and dose-rates. This offered the opportunity to address the following questions: (1) How reproducible are BMD values for pathways and genes across similar and different studies? (2) What radiation exposure parameters are key drivers of BMD values? (3) How do transcriptional BMD values compare with BMDs for cytogenetic endpoints produced under similar exposure scenarios?
The US Environmental Protection Agency Benchmark Dose Technical Guidance (US EPA. Citation2012) recommends design parameters for BMD studies that include: (1) a broad dose-range be used for BMD analysis, (2) a dose that is expected to induce effects for the endpoints of interest near that of the BMR, and (3) at least three non-zero doses. While none of the radiation datasets used in our experiment were designed for BMD modeling, these guidelines were followed when we selected the studies for inclusion to ensure dependable data quality. From over 100 studies identified, 50% were discarded as they did not meet the dose criteria or did not identify dose-response modeled genes. The remaining datasets were then subjected to extensive evaluation using the BMD analysis approach. Within this study, datasets were used that were purposefully diverse and derived from experiments with different purposes. Theoretically, this is the strength of the BMD approach as it should provide consistency in the BMDs derived from studies with varied design within an experimental model, allowing for comparison of thresholds across common genes and pathways for different exposure types (Haber et al. Citation2018). The results of the analysis showed a broad distribution of BMD values for individual genes and pathways across datasets; this is not unexpected given the highly varied exposure parameters of each study. Furthermore, in contrast to expectations, it was also noted that dose ranges, which were variable across studies, impacted the BMD values. For example, a study conducted at high-centric doses (e.g. radiotherapy applications) had relatively higher BMD values compared to those that used a broader dose-range. This is perhaps reflective of ineffective modeling of the dose-response relationship (e.g. poorly fitting curves) highlighting the importance of dose-ranges in data interpretation, particularly in comparing BMD values across diverse studies.
A number of data visualization approaches were used to delineate trends associated with the BMD outputs with respect to dose-rate, radiation quality and cell-type. It was a challenge to identify studies that only had one variable altered between them that could be reliably compared. Studies differed among each other in at least two factors (e.g. dose-rate, radiation quality or tissue type and dose-rate or radiation quality and tissue type). A global assessment of BMD values at the pathway and gene level showed broad variability across studies. However, it was noted that consistent genes and pathways were discernable that could be classified as being activated at low BMD values or high BMD values. In general, the results were suggestive of a continuous activation of pathways across doses in a modal fashion. An analysis of the responding genes showed that those with higher BMD values were associated with immune response and cell death, while transcripts with lower BMD values were generally related to the classic DNA damage response/repair processes, centered on TP53 signaling. These results were comparable with our previous observations using a limited dataset of two studies (Chauhan et al. Citation2016). It shows the approach to produce meaningful data which is well-aligned with our current understanding of radiation biology (Boss et al. Citation2014). It highlights molecular processes that are enriched at different doses and which may facilitate the prediction of a specific type of radiogenic exposure.
A focused analysis of studies conducted under relatively comparable exposure conditions was anticipated to display reproducible BMD outputs. Six studies were identified that used either animal or in vitro model systems exposed to photon radiation that had harvested blood at two common time-points. It was of interest to see if in vitro and in vivo responses were distinguishable using the BMD approach. The gene BMDs produced a bimodal response and the distributions at each mode occurred at very similar dose-ranges across the studies. Genes and pathways that were shared had very comparable BMD values for both the animal and in vitro studies. This provided some level of confidence that under similar exposure conditions, BMD values are reproducible, irrespective of the model system. Furthermore, genes comprising mode 1 had BMD values similar to cytogenetic-derived BMD values. We showed that a threshold dose of 0.5 Gy initiated damage in the form of translocations, micronuclei formation and dicentric chromosomes. This was shown to parallel DNA response and repair pathways that are initiated in mode 1 of the transcriptional BMD distribution. Chemical toxicity studies have also revealed that in vivo transcriptomic BMD values can predict apical BMDs (Thomas et al. Citation2013; Buick et al. Citation2017). However, toxicogenomic- derived POD can be within tenfold of those derived from apical endpoints (Farmahin et al., Citation2017; Johnson et al. Citation2020). Therefore, work is underway in the chemical field to identify a subset of genes that produce BMDs that approximate the doses at which phenotypic effects occur (Farmahin et al., Citation2017). Conversely, it may also be valuable in the radiation field to identify a grouping of BMD values derived from pathways and genes that may be predictive of an exposure scenario. For example, high BMD values may be indicative of an alpha radiation exposure compared to low centric BMD values. In addition, the BMD approach would provide avenues to address questions related to tissue/organ sensitivities (). Future work will assess the feasibility of binning BMD values based on the exposure parameters.
In conclusion, transcriptional BMD analysis can be used to fully exploit dose-response data to derive a useful metric that facilitates the comparison of experiments providing meaningful dose thresholds for pathways and gene activation across different exposure scenarios. Broadly, this information can be harnessed to identify tissue sensitivities, radiation quality differences and linkages to phenotypic changes as demonstrated in the toxicological fields and using radiation-relevant datasets (Chauhan et al. Citation2018; Qutob et al. Citation2018). Some challenges to the approach were identified, specifically with respect and dose-ranges for studies needing to be broad in order to obtain reproducible BMD values. In addition, although the BMD modeling tools provide computationally efficient methods to examine the shape of the dose-response curves for individual genes, the meaning of these in the context of phenotypic dose-response curves will need further investigation. To advance this area such focused studies should be undertaken using datasets conducted under similar exposure conditions that are compatible with the BMD methodology and in which relevant parameters are varied independently and linked to an adverse outcome. This will be highly informative in determining the ability to derive more meaningful BMD outputs. Future work will analyze selected studies under matched dose conditions, to better understand if parameters such as dose-rates, time-points, radiation quality and confounders (smoking status, species, gender) can yield informative gene/pathway BMD values that are exposure-predictive.
| Abbreviations | ||
| BMD | = | Benchmark Dose |
| GEO | = | Gene Expression Omnibus |
| LET | = | Linear Energy Transfer |
| LOAEL | = | Lowest Observed Adverse Effects Levels |
| NOAEL | = | No Observed Adverse Effects Levels |
| POD | = | Point of Departure |
| BMR | = | Benchmark Response |
| BMDU | = | Benchmark dose upper confidence limit |
| BMDL | = | Benchmark dose lower confidence limit |
| CBMN | = | Cytokinesis block micronuclei assay |
| DCA | = | Dicentric chromosome assay |
Acknowledgments
The authors would like to acknowledge Lindsay Beaton and Ngoc Vuong for insightful comments and edits to the manuscript.
Disclosure statement
No potential conflict of interest was reported by the author(s).
Additional information
Notes on contributors
Vinita Chauhan
Vinita Chauhan, Ruth Wilkins and Carole Yauk are research scientist at Health Canada.
Nadine Adam
Nadine Adam is a research assistant at Health Canada.
Andrew Williams
Byron Ko and Andrew Williams are Bioinformatic/Statisticians at Health Canada
Robert Stainforth
Robert Stainforth, is a post-doctoral fellow at Health Canada.
References
- Aardema MJ, MacGregor JT. 2002. Toxicology and genetic toxicology in the new era of “toxicogenomics”: impact of “-omics” technologies. Mutat Res. 499(1):13–25.
- Borgert CJ. 2007. Predicting interactions from mechanistic information: can omic data validate theories? Toxicol Appl Pharmacol. 223(2):114–120.
- Boss MK, Bristow R, Dewhirst MW. 2014. Linking the history of radiation biology to the hallmarks of cancer. Radiat Res. 181(6):561–577.
- Bourdon JA, Williams A, Kuo B, Moffat I, White PA, Halappanavar S, Vogel U, Wallin H, Yauk CL. 2013. Gene expression profiling to identify potentially relevant disease outcomes and support human health risk assessment for carbon black nanoparticle exposure. Toxicology. 303:83–93.
- Buesen R, Chorley BN, da Silva Lima B, Daston G, Deferme L, Ebbels T, Gant TW, Goetz A, Greally J, Gribaldo L, et al. 2017. Applying ‘omics’ technologies in chemicals risk assessment: report of an ECETOC workshop. Regul Toxicol Pharmacol. 91(Suppl 1):S3–S13.
- Buick JK, Williams A, Kuo B, Wills JW, Swartz CD, Recio L, Li HH, Fornace AJ, Jr, Aubrecht J, Yauk CL. 2017. Integration of the TGx-28.65 genomic biomarker with the flow cytometry micronucleus test to assess the genotoxicity of disperse orange and 1,2,4-benzenetriol in human TK6 cells. Mutat Res. 806:51–62. Epub 2017 Sep 13.
- Chauhan V, Kuo B, McNamee JP, Wilkins RC, Yauk CL. 2016. Transcriptional benchmark dose modeling: exploring how advances in chemical risk assessment may be applied to the radiation field. Environ Mol Mutagen. 57(8):589–604.
- Chauhan V, Rowan-Carroll A, Gagné R, Kuo B, Williams A, Yauk CL. 2018. The use of in vitro Transcriptional data to Identify Thresholds of Effects in a Human Lens Epithelial Cell-line Exposed to Ionizing Radiation. Int J Radiat Biol. 5:1–40.
- Chepelev NL, Moffat ID, Labib S, Bourdon-Lacombe J, Kuo B, Buick JK, Lemieux F, Malik AI, Halappanavar S, Williams A, Yauk CL. 2015. Integrating toxicogenomics into human health risk assessment: lessons learned from the benzo[a]pyrene case study. Crit Rev Toxicol. 45(1):44–52.
- Clewell HJ, Efremenko A, Campbell JL, Dodd DE, Thomas RS. 2014. Transcriptional responses in the rat nasal epithelium following subchronic inhalation of naphthalene vapor. Toxicol Appl Pharmacol. 280(1):78–85.
- Collins FS, Green ED, Guttmacher A, Guyer MS. 2003. A vision for the future of genomics research. Nature. 422(6934):835–847.
- Crump K. 1984. A new method for determining allowable daily intakes. Fundam Appl Toxicol. 4(5):854–871.
- Davis JA, Gift JS, Zhao QJ. 2011. Introduction to benchmark dose methods and U.S. EPA’s benchmark dose software (BMDS) version 2.1.1. Toxicol Appl Pharmacol. 254(2):181–191.
- Farmahin R, Williams A, Kuo B, Chepelev NL, Thomas RS, Barton-Maclaren TS, Curran IH, Nong A, Wade MG, Yauk CL, et al. 2017. Recommended approaches in the application of toxicogenomics to derive points of departure for chemical risk assessment. Arch Toxicol. 91(5):2045–2065.
- Filipsson AF, Sand S, Nilsson J, Victorin K. 2003. The benchmark dose method—review of available models, and recommendations for application in health risk assessment. Crit Rev Toxicol. 33(5):505–542.
- Flegal FN, Devantier Y, Marro L, Wilkins RC. 2012. Validation of QuickScan dicentric chromosome analysis for high throughput radiation biological dosimetry. Health Phys. 102(2):143–153.
- Haber LT, Dourson ML, Allen BC, Hertzberg RC, Parker A, Vincent MJ, Maier A, Boobis AR. 2018. Benchmark dose (BMD) modeling: current practice, issues, and challenges. Crit Rev Toxicol. 48(5):387–415.
- Jackson AF, Williams A, Recio L, Waters MD, Lambert IB, Yauk CL. 2014. Case study on the utility of hepatic global gene expression profiling in the risk assessment of the carcinogen furan. Toxicol Appl Pharmacol. 274(1):63–77.
- Johnson KJ, Auerbach SS, Costa E. 2020. A rat liver transcriptomic point of departure predicts a prospective liver or non-liver apical point of departure. Toxicol Sci. 2020;176(1):86-102.
- Kyrtopoulos SA. 2013. Making sense of OMICS data in population-based environmental health studies. Environ Mol Mutagen. 54(7):468–479.
- Labib S, Williams A, Yauk CL, Nikota JK, Wallin H, Vogel U, Halappanavar S. 2016. Nano-risk Science: application of toxicogenomics in an adverse outcome pathway framework for risk assessment of multi-walled carbon nanotubes. Part Fibre Toxicol. 13:15–15.
- McNamee JP, Flegal FN, Greene HB, Marro L, Wilkins RC. 2009. Validation of the cytokinesis-block micronucleus (CBMN) assay for use as a triage biological dosimetry tool. Radiat Prot Dosimetry. 135(4):232–242.
- National Toxicology Program. 2018. NTP Research Report on National Toxicology Program Approach to Genomic Dose-Response Modeling: Research Report 5 [Internet]. Durham (NC): National Toxicology Program; [accessed 2020 July 22]. http://www.ncbi.nlm.nih.gov/books/NBK531562/.
- Phillips JR, Svoboda DL, Tandon A, Patel S, Sedykh A, Mav D, Kuo B, Yauk CL, Yang L, Thomas RS, et al. BMDExpress 2: enhanced transcriptomic dose-response analysis workflow. Bioinformatics. 2019;35(10):1780–1782.
- Qutob SS, Chauhan V, Kuo B, Williams A, Yauk CL, McNamee JP, Gollapudi B. 2018. The application of transcriptional benchmark dose modeling for deriving thresholds of effects associated with solar-simulated ultraviolet radiation exposure. Environ Mol Mutagen. 59(6):502–515.
- Sand S, Victorin K, Filipsson AF. 2008. The current state of knowledge on the use of the benchmark dose concept in risk assessment. J Appl Toxicol. 28(4):405–421.
- Tatusova T, DiCuccio M, Badretdin A, Chetvernin V, Nawrocki EP, Zaslavsky L, Lomsadze A, Pruitt KD, Borodovsky M, Ostell J, et al. 2016. NCBI prokaryotic genome annotation pipeline. Nucleic Acids Res. 44(14):6614–6624.
- Thomas RS, Philbert MA, Auerbach SS, Wetmore BA, Devito MJ, Cote I, Rowlands JC, Whelan MP, Hays SM, Andersen ME, Meek ME, Reiter LW, et al. 2013. Incorporating new technologies into toxicity testing and risk assessment: moving from 21st century vision to a data-driven framework. Toxicol Sci. 136(1):4–18.
- Thomas RS, Wesselkamper SC, Wang NC, Zhao QJ, Petersen DD, Lambert JC, Cote I, Yang L, Healy E, Black MB, Clewell HJ, 3rd, et al. 2013. Temporal concordance between apical and transcriptional points of departure for chemical risk assessment. Toxicol Sci. 134(1):180–194.
- United Nations Scientific Committee on Atomic Radiation (UNSCEAR). 2012. The 2012 white paper to guide the scientific committee’s white paper future programme or work. Biological mechanisms of radiation actions at low doses. New York: United Nations Publications.
- US EPA. 2012. Benchmark Dose Technical Guidance (EPA/100/R-12/001); [accessed 2015 Oct 7]. http://www2.epa.gov/sites/production/files/2015-01/documents/benchmark_dose_guidance.pdf.
- Webster AF, Chepelev N, Gagne R, Kuo B, Recio L, Williams A, Yauk CL. 2015. Impact of genomics platform and statistical filtering on transcriptional Benchmark Doses (BMD) and multiple approaches for selection of chemical Point of Departure (PoD). PLoS One. 10(8):e0136764.
- Webster AF, Zumbo P, Fostel J, Gandara J, Hester SD, Recio L, Williams A, Wood CE, Yauk CL, Mason CE. 2015. Mining the archives: a cross-platform analysis of gene expression profiles in archival formalin-fixed paraffin-embedded tissues. Toxicol Sci. 148(2):460–472.
- William MG, Auerbach SS, Parham F, Stout MD, Waidyanatha S, Mutlu E, Collins B, Paules RS, Merrick BA, Ferguson S, et al. Evaluation of 5-day in vivo rat liver and kidney with high-throughput transcriptomics for estimating benchmark doses of apical outcomes. Toxicol Sci. 2020:kfaa081. DOI:https://doi.org/10.1093/toxsci/kfaa081.
- Yang L, Allen BC, Thomas RS. 2007. BMDExpress: a software tool for the benchmark dose analyses of genomic data. BMC Genomics. 8:350–387.