Case Study Drug #1: NIT is a nitrofuran derivative antibiotic that is used to treat urinary tract infections (UTIs) and has been used for over 60 years (Muller et al., 2017). Common brand names include Furadantin, Macrobid and Macrodantin (Prescribers’ Digital Reference Network, 2021). NIT has been shown to be mutagenic in the Ames assay due to activation by bacterial nitroreductases, which is not relevant for mammalian cells (Olive and McCalla, 1977; Ni et al., 1987; Mokdad-Bzeouich et al., 2015). NIT has also been shown to be weakly mutagenic in mammalian cells (Gao et al., 1989; Slapsyte et al., 2002), and in the kidneys of Big Blue transgenic mice (Quillardet et al., 2006). Additionally, NIT can induce chromosome damage (Parodi et al., 1983; Thompson, 1986). Despite experimental evidence of genotoxicity, the long-term antimicrobial treatment with NIT is generally considered safe (Uhari et al., 1996); thus, NIT provides an excellent model compound to evaluate the human relevance of the conflicting genotoxicity findings using the TGx-DDI biomarker.

Case Study Drug #2: MTZ is an antibacterial and antiprotozoal agent used to treat anaerobic bacterial infections and protozoal infections in human and veterinary applications. It is also prescribed off-label in the treatment of rosacea and Crohn’s disease, as a post-surgical prophylactic agent, and in the treatment of Helicobacter pylori infection (National Center for Biotechnology Information, 2022). Common brand names include Flagyl, MetroCream, and Vandazole (Prescribers’ Digital Reference Network, 2021). MTZ has demonstrated mutagenic activity in the Ames bacterial reverse mutation assay, both induced by the drug itself (Melo and Ferreira, 1990) and by the urine of treated patients (Speck et al., 1976; Connor et al., 1977). In mammalian cells and in laboratory animals, conflicting evidence exists regarding the genotoxicity of MTZ (Buschini et al., 2009). Human studies have failed to demonstrate the potential for genetic damage. Due to the conflicting experimental data for the mutagenicity and genotoxicity of MTZ, this agent was selected as a case study chemical to provide further clarity on its genetic safety.

Case Study Drug #3: NOV, also known as albamycin, streptonivcin, or cathomycin, is a narrow-spectrum antibiotic drug. It is used in veterinary applications for the treatment of bovine mastitis in lactating and dry-off cows. NOV was also widely used as a human therapeutic (Committee for Veterinary Medicinal Products, 1999). NOV interacts directly with the B subunit (GyrB) of the bacterial DNA gyrase enzyme (Maxwell, 1993; Heide, 2009a) and it is a non-specific inhibitor of DNA topoisomerase II (DNA topoII) enzyme (Savoldi-Barbosa et al., 1999). NOV has been shown to cause concentration-dependent DNA fragmentation that preceded apoptosis in primary cultures of mouse thymocytes, by inhibiting DNA-rejoining activity of the enzyme (Onishi et al., 1993). While NOV administration initially demonstrated little toxicity (Kirby et al., 1956), it has since been withdrawn from the market for human therapeutic use due to poor pharmacological properties and liver toxicity (Shirude and Hameed, 2012). As a result, NOV was selected as a case study chemical to investigate its potential genotoxic hazard to clarify its genetic safety.

Nitrofurantoin (NIT; CAS no. 67-20-9), metronidazole (MTZ; CAS no. 443-48-1), and novobiocin sodium salt (NOV; CAS no. 1476-53-5) were purchased from Sigma-Aldrich, Inc. (St. Louis, Missouri, United States). NIT (lot no. MKCC9622), MTZ (lot no. MKBZ3056V), and NOV (lot no. 2998909) were 100%, >99%, and 96.0% pure, respectively. Identical lot numbers were used across all laboratories. Dimethyl sulfoxide (DMSO; 0.125% for 2-250 μM, 0.25% for 500 μM, and 0.5% for 1,000 µM), acetic acid (HOAc; 0.0625% for 2-250 μM, 0.25% for 500 μM, and 0.5% for 1,000 µM) and water (H₂O; 0.125% for 2-250 μM, 0.25% for 500 μM, and 0.5% for 1,000 µM) were used as vehicle controls for NIT, MTZ, and NOV, respectively. Caffeine (CA; 2 mM) was used as a negative control. Bleomycin (BL; 10 µM) and ionizing radiation (IR; 4 Gy) were used as positive controls for the NanoString nCounter^® and BioSpyder TempO-Seq^® experiments. Hydrogen peroxide (H₂O₂; 25-100 µM) was used as a positive control for the CometChip^® experiments.

Human TK6 cells are the recommended cell line for TGx-DDI biomarker analysis. Extensive validation of TGx-DDI performance has been completed using TK6 cells, which were selected because they are widely used in genotoxicity testing and have an intact p53 response pathway (Li et al., 2019). In this study, TK6 cells (ATCC, Manassas, Virginia, United States) were exposed to ten concentrations of each test chemical in triplicate, as follows: 2.0, 3.9, 7.8, 15.6, 31.3, 62.5, 125, 250, 500, and 1,000 μM, in addition to matched vehicle controls. The maximum concentrations were determined based on guidance for top concentration selection for mammalian cell assays in the ICH Guideline S2 (R1) on Genotoxicity Testing and Data Interpretation for Pharmaceuticals Intended for Human Use (ICH, 2012). Cells were exposed to the extended concentration range of each chemical for 4 h in parallel 96-well plates for cytotoxicity analysis and 6-well plates for the collection of cells for gene expression analysis using nCounter^® and TempO-Seq^® for TGx-DDI classification at Georgetown University (GU). For the evaluation of DNA damage by CometChip^®, a sterile 12-well reagent reservoir was used for cellular exposures at the Massachusetts Institute of Technology (MIT). One 12-well reservoir was used for each test chemical, and each well in this reservoir contained cells exposed to one of the ten chemical concentrations or matched vehicle controls. Then, treated cells were loaded into duplicate wells of the CometChip^® for DNA damage assessment. TK6 cells, a spontaneously transformed human lymphoblastoid cell line, were cultured in suspension in RPMI 1640 supplemented with 10% fetal bovine serum and treated with case study chemicals, as described previously (Li et al., 2015; Li et al., 2017). Specifically, exponentially growing cells were treated at a density of 4–5 × 10⁵ cells per mL in parallel 96-well plates. Following the 4 h exposure, cells were washed and collected for RNA extraction for subsequent gene expression analysis using NanoString nCounter^® and BioSpyder TempO-Seq^® gene expression analysis.

The MTT assay (Cayman Chemical, Ann Arbor, MI, United States) was used to quantify cellular metabolic activity as an indicator of cell viability and was conducted at GU. TK6 cells were exposed for 4 h, as described in the previous section, then cells were washed and re-incubated in fresh media for a 20 h recovery period for cytotoxicity assessment. The cytotoxicity of the vehicle was tested and the concentration of vehicle that caused minimal (less than 10%) cytotoxicity was used. In the cases that a more concentrated vehicle was needed due to the solubility of the chemical, the corresponding vehicle controls were included as the reference. The MTT assay was performed in triplicate at the 24 h time point (4 h exposure, plus 20 h recovery) for all 10 concentrations plus the solvent controls, following the manufacturer’s instructions. Absorbance was measured at 570 nm using a microplate reader. The cytotoxicity threshold was 50%, in line with the guidance provided in OECD Test Guideline 487 for In Vitro Mammalian Cell MN Test and the ICH Guideline S2 (R1) on Genotoxicity Testing and Data Interpretation for Pharmaceuticals Intended for Human Use (ICH, 2012; OECD, 2016). Chemical conditions that surpassed the cytotoxicity threshold were excluded from the nCounter^® and TempO-Seq^® analyses.

RNA extraction was performed using the protocol accompanying the TRIzol™ reagent (Invitrogen, part of ThermoFisher Scientific, Waltham, MA United States), which was then purified via an RNA cleanup step using an RNeasy column according to the manufacturer’s instructions (Qiagen, Germantown, MD United States). Purified RNA was quantified and analyzed for quality using a NanoDrop™ spectrophotometer and an Agilent BioAnalyzer, respectively.

The TGx-DDI nCounter^® assay was conducted at GU using 100 ng of high quality, purified RNA. All non-cytotoxic concentrations of each test chemical were included in this analysis, in addition to the corresponding vehicle controls for each treatment. The highest two concentrations of MTZ and NIT were excluded (i.e., 500 μM and 1,000 µM), in addition to the highest concentration of NOV (i.e., 1,000 µM) due to levels of cytotoxicity exceeding the threshold of 50% cell death. The nCounter^® assay was performed in triplicate for each treatment and matched control. Caffeine (CA; 2 mM) served as a negative control, while bleomycin (BL; 10 µM) and ionizing radiation (IR; 4 Gy) were used as positive controls. nCounter^® experiments were conducted as previously reported (Li et al., 2017) and full details of the nCounter^® methods have been previously reported (Geiss et al., 2008). The nCounter^® experiments were conducted following the instructions outlined in the NanoString nCounter^® XT Assay User Manual (MAN-10023-11 July 2016). Briefly, optimized sequences for the TGx-DDI biomarker genes, plus eight housekeeping genes, including G6PD, GUSB, HPRT1, LDHA, NONO, PGK1, PPIH, and TFRC, were chosen based on stability and detectable expression levels and were included in a custom-designed CodeSet manufactured by NanoString. Unique barcodes were counted for each target, and the data were exported for analysis. The counts for each target were analyzed for quality control and normalization using nSolver™ Analysis version 4.0. Normalized data were exported and subjected to further analysis, which is described below in the Statistical and Bioinformatic Analyses for TGx-DDI Classification section.

TempO-Seq^® gene expression analysis was conducted on all non-cytotoxic concentrations of exposed and control TK6 cell lysates at Health Canada. The 500 μM and 1,000 μM concentrations of MTZ and NIT were omitted, in addition to the 1,000 μM concentration of NOV, as these concentrations surpassed the cytotoxicity threshold of 50% cell death. The 2 μM concentrations were also eliminated in order to run the positive and negative TempO-Seq^® assay controls. Libraries were prepared in a 96-well plate format using the TempO-Seq^® Human Tox + Surrogate Panel reagent kit (BioSpyder Technologies, Carlsbad, CA, United States), following the company’s protocol. Human Brain Total RNA and qPCR Human Reference Total RNA (Takara Bio, CA, United States) were included as positive assay controls and a no-lysate control (1X TempO-Seq^® Lysis Buffer only) was included as a negative assay control (two replicates per control). The same positive and negative controls were included for TempO-Seq^® analysis as for nCounter^® analysis (i.e., 2 mM CA as a negative control; 10 μM BL and 4 Gy IR as positive controls). In brief, 100 ng of total RNA (in a 2 μl volume) from exposed and control cells were hybridized with the targeted Human S1500+ Tox Panel detector oligo (DO) probe mix (v1.1; 2,977 probes) in 1X TempO-Seq^® Lysis Buffer for 10 min at 70°C followed by a temperature gradient with a ramp rate of 0.5^°C/min to 45°C over a 50 min period, followed by a nuclease digestion at 37°C for 90 min to remove excess, unbound, or incorrectly bound DOs enzymatically. Ligation of the DO pairs bound to adjacent target sequences was then completed with a 60 min incubation at 37°C, immediately followed with a 15 min enzyme denaturation at 80°C to generate a pool of amplification templates. All amplification templates (i.e., 10 μl of ligated DOs) were transferred to its corresponding well of the 96-well PCR Pre-Mix and Primers plate. Amplification proceeded with a CFX96 Real-Time PCR Detection System (Bio-Rad, Mississauga, ON, Canada) to attach a unique sequence tag and the sequencing adaptors to each sample using the following PCR program settings: 37°C for 10 min, 95°C for 2 min; 6 cycles of 95°C for 30 s, 54°C for 30 s, 72°C for 120 s; 16 cycles of 95°C for 30 s; 72°C for 2 min; 72°C for 1 min. The TempO-Seq^® libraries were pooled (5 μl of all 96 samples) and purified using the Macherey-Nagel NucleoSpin^® Gel and PCR Clean-Up kit (Clontech Laboratories Inc., Bethlehem, PA, United States), according to the manufacturer’s instructions for PCR cleanup with three modifications outlined in the TempO-Seq^® Assay User Guide (v2.1, 2017). The pooled, purified TempO-Seq^® libraries were sequenced using a NextSeq^® 500/550 High Output flow cell (v2 kit, 75 cycles) on an Illumina NextSeq^® 500 Sequencing platform (Illumina, San Diego, CA, United States).

Sequencing data have been archived in the National Center for Biotechnology Information (NCBI) Gene Expression Omnibus (GEO) database under accession number GSE196373. Raw sequencing data were analyzed at Health Canada and assigned to respective sample files by demultiplexing them with blc2fastq v2.20.0.422. They were then trimmed for quality control using fastp (v0.20.0). The subsequent FASTQ files were aligned to the TempO-Seq^® Human Tox + Surrogate Panel reference sequences (2,977 probes) from BioSpyder using their purpose-built analysis pipeline (TempO-SeqR, v3.0) to generate a table of counts (one per gene per sample). Briefly, the TempO-SeqR pipeline used STAR v2.7.8a to perform alignment of raw reads to the reference sequences. Subsequently, the qCount function of the QuasR package (v1.30.0) was used to produce a gene X sample count matrix of raw counts from the BAM files output by STAR.

Quality control of all study samples was performed on the count matrix using several methods to measure consistency and remove low quality samples, using criteria outlined in Harrill et al. (2021) as a guideline. Samples were removed from the study if they clustered as singletons at a dissimilarity of 0.1 using 1-Spearman correlation using complete linkage. We used a cut-off of 10% of the median number of reads to remove samples that had insufficient sequencing depth. We eliminated any samples outside of Tukey’s Outer Fence (3X interquartile range) for: 1) the number of probes capturing the top 80% of the signal; 2) the Gini coefficient (which measures inequality in distributions); and 3) the number of active probes (those with at least 5 mapped reads). Based on these metrics and suboptimal sequencing depth, nine experimental samples were removed (NIT: one replicate each of DMSO solvent control, 2 μM, 3.9 μM, 15.6 μM; MTZ: one replicate of 2 μM and two replicates of 250 μM; NOV: one replicate each of 2 μM and 125 μM).

To account for sequence-to-sequence variability in read depth between the samples, read counts were normalized using DESeq2 (v1.30.1) (Love et al., 2014) using the counts () function in R (R, 2020). Samples with poor data quality were identified through visualization using boxplots and hierarchical cluster analysis. This resulted in the exclusion of nine samples from the TGx-DDI classification analysis due to suboptimal sequencing depth (identified in previous section), or the fact that they clustered as singletons. The TGx-DDI genomic biomarker was used to classify chemicals as DDI or non-DDI using statistical modeling and bioinformatics tools. Detailed analytical information can be found in Yauk et al. (2016) and Buick et al. (2017). Gene Symbols for TGx-DDI biomarker genes that had multiple probes were averaged. Hierarchical clustering was accomplished using the hclust () function in R (www.r-project.org). Agglomerative clustering was based on average linkage with Euclidean distances (Becker et al., 1988). DDI vs. non-DDI classifications were determined using the Nearest Shrunken Centroids (NSC) method (Tibshirani et al., 2002) in the pamr function of R (www.bioconductor.org), as previously described (Yauk et al., 2016; Buick et al., 2017; Li et al., 2017). Briefly, the standardized centroid (SC) was calculated by applying the NSC method for DDI and non-DDI test chemicals in the training set and is the mean expression level for each gene in a class divided by its within-class standard deviation. For each DDI and non-DDI chemical, the SC is shrunken in the direction of the overall centroid to create the NSC. Experimental samples (i.e., exposed and control) were then classified by comparing their gene expression profile to the class of NSCs and then assigned to a class closest to it in squared distance so that the probability of class membership was >0.90 (Li et al., 2015; Li et al., 2017).

Three distinct analyses were performed to classify the compounds as DDI and non-DDI using the TGx-DDI genomic biomarker, including NSC probability analysis (PA; visualized using heatmaps), principal component analysis (PCA), and hierarchical clustering (HC), as outlined in Yauk et al. (2016) and Buick et al. (2017). PCA was completed using the prcomp () function in R (Venables and Ripley, 2002), where the training set data (Li et al., 2015) was used to estimate the principal components (PC). These PC loadings were applied to the data generated with the three test agents. Samples with PC1 <0 were classified as DDI; otherwise, the sample was classified as non-DDI. For the HC analysis, samples clustering with the DDI training compounds were classified as DDI and, similarly, samples clustering with the non-DDI agents were classified as non-DDI. Samples that clustered as singletons (i.e., that cluster on their own) were classified as unknown. A scatterplot generated using data from the TGx-DDI training set and experimental chemicals was generated to visualize the outcomes. Classification was determined as follows: a chemical was classified as DDI if it resulted in a positive call in any one of three classification analyses (NSC PA heatmaps, PCA, or HC); whereas, a chemical was classified as non-DDI if it was negative in all of the three analyses (Yauk et al., 2016; Buick et al., 2017; Buick et al., 2020).

The HT alkaline CometChip^® assay was conducted at MIT and was used to assess DNA damage (i.e., single-strand breaks) across the concentration range for each chemical in 96-well plates following a 4 h chemical exposure (Wood et al., 2010). For all CometChip^® data, results reflect three independent experiments. Analysis was done at 4 h to align with the gene expression analysis experimental design. Control and exposed TK6 cells were loaded into the CometChip^® wells and were allowed to settle by gravitational forces into the microwells, as previously described (Wood et al., 2010; Buick et al., 2021). Briefly, the cells were encased in a 1% low melting point agarose overlay and then lysed in cold lysis buffer (2.5 M NaCl, 100 mM EDTA, 10 mM Tris, pH 10 with 1% Triton X-100 and 10% DMSO) overnight at 4°C. After cell lysis, the CometChip^® was equilibrated in alkaline electrophoresis buffer (300 mM NaOH and 1 mM EDTA) for 40 min and then separated for 50 min with a 300 mA current at 4°C. After electrophoresis, the chip was neutralized twice for 15 min in 0.4 M Tris, pH 7.4 at 4°C and then equilibrated overnight in 20 mM Tris, pH 7.4 at 4°C. Following equilibration, the chip was stained in 0.1X SYBR Gold for 30 min and then destained for >1 h in 20 mM Tris, pH 7.4 at 4°C. Once destained, comet images were captured using a Nikon 80i microscope at 4× magnification for all 96 wells on the chip. The tiff images were taken and analyzed using in-house software, as previously described (Wood et al., 2010). All chemical conditions and controls were run in parallel.

Median % tail DNA was analysed using one-way analysis of variance (ANOVA). The normality assumption was tested using the Anderson-Darling statistic (Stephens, 1986) and the common variance assumption was verified using the Fligner-Killeen test of homogeneity of variances (Conover et al., 1981). If either assumption was not met, the rank transformation was applied and the nonparametric one-way ANOVA was performed (Conover and Iman, 1981). Pairwise contrasts to corresponding vehicle controls were conducted using the t-test. The subsequent p-values were then FWER (Family-Wise Error Rate) adjusted using the Dunnett’s method.

To select the appropriate concentration range for detection of the DNA damage response using the TGx-DDI biomarker and DNA damage using the CometChip^® assay, we first evaluated cytotoxicity using the colorimetric MTT assay following a 4 h exposure + 20 h recovery to ten concentrations of each drug in triplicate (2 μM–1,000 μM). In this assay, cellular metabolic activity provides an indication of cell viability and cytotoxicity in treated TK6 cells (Figure 1). The two highest concentrations of NIT and MTZ (500 and 1,000 μM) caused declines in viability beyond the threshold of 50% cytotoxicity (Figures 1A,B, respectively). Only the highest concentration of NOV tested (1,000 μM) caused a reduction in cell viability in excess of the cytotoxicity cut-off of 50% (Figure 1C). The 250 μM concentration of NIT, in addition to the 250 μM and the 500 μM concentrations of NOV, caused declines in cell viability but did not surpass the 50% threshold. The remaining concentrations tested for all three pharmaceutical agents did not cause any notable declines in cell viability (Figure 1). All concentrations of test chemicals were used for CometChip^® analysis, including those causing excessive cytotoxicity. The drug treatments exceeding the cytotoxicity limits were excluded from TGx-DDI analysis by nCounter^® and TempO-Seq^® (i.e., 500 and 1,000 μM for NIT and MTZ, and 1,000 μM for NOV) (Ellinger-Ziegelbauer et al., 2009).

We used two different gene expression methods to measure the TGx-DDI biomarker genes: NanoString nCounter^® digital quantification and BioSpyder TempO-Seq^® S1500+ sequencing (Figure 2). The negative assay controls (1X TempO-Seq^® Lysis buffer, no lysates) had low mapped read counts (i.e., <1,200) as expected and the positive assay controls (Human Reference Total RNA and Human Brain Total RNA) showed Pearson correlation coefficients between the replicates that were >0.98 for all pairwise comparisons (data not shown). The negative control (2 mM CA) classified as non-DDI using both gene expression technologies (Figure 2, Supplementary Figures 1D,2A, while the two positive controls (10 μM BL and 4 Gy IR) classified as DDI in both assays (Figure 2, Supplementary Figures 1D,2B-C). After outlier analysis, the final sample size was n = 3, except for NIT DMSO solvent control, 2 μM NIT, 3.9 μM NIT, 15.6 μM NIT, 2 μM MTZ, 2 μM NOV, and 125 μM NOV that had an n = 2 and 250 μM MTZ that had an n = 1.

A three-pronged analysis that includes NSC PA, PCA, and HC was used for classification (Li et al., 2017). If a drug had a positive outcome in one or more analyses, it was predicted to be DDI; whereas, a chemical that had a negative outcome in all three analyses was classified as non-DDI. The heatmap (Figure 2) shows the gene expression profiles and the TGx-DDI classifications for the three test compounds using nCounter^® and TempO-Seq^® gene expression analysis. Supplementary Figure 1A–C depict the PC analyses for NIT, MTZ, and NOV, respectively; whereas, the HC analyses are shown in Supplementary Figures 3A–H for NIT, Supplementary Figures 4A–H for MTZ, and Supplementary Figures 5A–I for NOV. The TGx-DDI biomarker classified all concentrations of NIT and MTZ as non-DDI using both gene expression technologies. NOV was also classified as non-DDI at all concentrations except 250, which was DDI using both nCounter^® and TempO-Seq^® technologies. The 500 μM concentration could not be classified (probability of class membership <0.9 for both DDI and non-DDI calls) using the TGx-DDI with nCounter^® data but was identified as non-DDI using TempO-Seq^® data. Overall, the TGx-DDI classification results were highly comparable across laboratories and technologies.

DNA damage (i.e., single-strand breaks) was quantified in human TK6 cells following a 4 h exposure to three antibiotic drugs using the alkaline CometChip^® assay (Figure 3). Hydrogen peroxide (25–100 μM), used as a positive control for the CometChip^® assay, showed a positive concentration-response (data not shown). Chemicals were considered positive if there was an increase in mean % tail DNA that was statistically significant compared to matched solvent control for non-cytotoxic test concentrations (p < 0.05). There was no accumulation of DNA damage observed at non-cytotoxic concentrations for NIT, MTZ, or NOV, compared to their matched vehicle controls (Figure 3A for NIT, Figure 3B for MTZ and Figure 3C for NOV, respectively). DNA damage was only observed for each test compound at overtly cytotoxic concentrations shown in red in Figure 3 (i.e., 500 μM and 1,000 μM for NIT and MTZ, and 1,000 μM for NOV). As cytotoxicity is associated with potential DNA damage and/or degradation, it is critical that the highest concentration tested using the Comet assay does not induce excessive cell death (Tice et al., 2000). As such, the positive CometChip^® results at overtly cytotoxic concentrations were not considered relevant in the final analysis. Overall, there was no evidence for single strand breaks following exposure to NIT, MTZ, or NOV, except at highly cytotoxic concentrations (>50% cell death).