Radiofrequency radiation exposure and rat brain tissue samples

Tissue samples for this retrospective study were obtained from animal studies previously conducted at the Ramazzini Institute to assess the carcinogenic hazard related to chronic exposure to radio frequency emissions representative of base station environmental exposure [12]. A 1.8 GHz base station radiation emission system that is similar to GSM cellular telephone base station was used for whole body exposure for the lifetime of Sprague-Dawley (SD) rats as previously described [12]. All the animals received standard feed ad libitum provided by Laboratorio Dottori Piccioni (Milan, Italy) and tap water. The animal experiments were conducted in accordance with the Italian law regulating the use of animals for experimental and other scientific purposes [14] as well as the Ramazzini Institute’s animal care and use committee guidelines [12]. Four groups of 817, 811, 411, 409 male and female SD rats were respectively subjected to a 1.8 GHz GSM of 0, 5, 25, 50 V/m (that correspond to a specific absorption rate (SAR) of 0, 1.001, 0.03, 0.1 W/Kg) with a whole-body exposure for 19 continuous hours per day, 7 days a week. SD rat tissues (gliomas = 12 and cardiac schwannomas = 9 from life-time RFR exposure; interim [1 year] sacrificed non-tumor brain tissues from RFR exposed rats [e.g. Exposed Non-tumor Tissues, “ENTs”] = 30; spontaneous gliomas = 2) from the RI-RFR cancer bioassay were examined in this study (S1 and S2 Tables). We have also included unexposed, age-matched (2 year) controls for brain (n = 10), heart (n = 9) and one-year interim sacrificed rats (n = 10). Matched genomic controls for all animals were taken from kidney tissues. At necropsy after life-time exposures, tissues were fixed in 70% alcohol, processed routinely, embedded in paraffin (EFPE) and 5 μm thick sections were cut and stained with hematoxylin and eosin (H&E). Sections of brain tissues from interim sacrifice animals were flash frozen in liquid nitrogen at the time of necropsy for molecular analysis.

Designing targeted next generation sequencing gene panel.

A targeted next generation sequencing (NGS) panel of 23 glioma-related genes was designed based on curated genetic mutation data from the COSMIC database and published literature representing more than 2000 human glial tumors [5, 15]. The panel was constructed to target all coding regions for 13 genes and known hotspot exons for an additional 10 genes (Table 1). Probes were designed against a 1 kb region upstream of the Tert gene transcription start site, select exons for the additional nine hotspot genes and exons, exon-intron boundaries, 5’UTR and 3’UTR for the remaining full length targeted genes (Table 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Cancer panel design listing targeted genes and exons.

Details of exon amplification success after probe design are included.

https://doi.org/10.1371/journal.pone.0296699.t001

Targeted next generation sequencing data analysis.

Sequencing read quality was assessed with the FastQC software [16]. Trim-Galore v. 0.4.4 was used to trim Illumina adapters and discard reads with phred quality score < 30 and length less than 50 bp after trimming. Trimmed reads were aligned to the Rn6 reference genome with Bowtie2 v.2.3.0 [17]. Reads with a mapping quality score (MapQ) < 30 were discarded with the samtools view function [18].

Strelka2 and deepSNV were used to call raw variants in tumor (Fig 1), ENT and age matched control samples with matched kidney tissue as the matched genomic control [19]. Strelka2 was run using default parameters with flags set to—disableEVS and–targeted—callRegions. For deepSNV, the site-specific background error rate was estimated with the beta-binomial model and the Bonferroni method was applied to p-values to control the family wise error rate.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. NGS panel data analysis pipeline.

https://doi.org/10.1371/journal.pone.0296699.g001

Strelka2 variant filter criteria required site read depth >1% sample coverage, quality score (QSS/QSI) > 25^th percentile, strand bias score < 25^th percentile and allele frequency > = 0.01. Filter criteria for deepSNV variants required Bonferroni corrected p-value < 0.05 and allele frequency > = 0.01. Variants remaining in the control sets after filtering were combined and subsequently removed from the tumor and ENT variants. The final high confidence variant set was comprised of SNVs and small indels that were shared between filtered Strelka2 and deepSNV variants for each sample.

Variants were annotated with the Ensembl Variant Effect Predictor (VEP) and SIFT 4G [20, 21]). The COSMIC v. 94 database was queried for known human variants with homology to the high confidence rat variants (cancer.sanger.ac.uk; [22]).

Comparative pathology of rat tumors

RFR exposure related increased incidences of malignant gliomas in the brain and schwannomas in the heart were observed in rats [12]. The malignant gliomas represented tumors of glial cell origin including astrocytoma, oligodendroglioma and mixed gliomas (specific glial cell populations between 20–80%). Due to the low incidences of glial tumors in the rodent carcinogenicity studies, different types of glial tumors are usually combined into a single category of malignant glioma. However, based on morphological features, three types of brain tumors may be discerned in rats, including astrocytoma, oligodendroglioma, and mixed gliomas. In contrast to human gliomas, rat glial tumors from carcinogenicity studies are rarely graded. Also, the relatively common human glial tumor glioblastoma is not commonly described in rats. Astrocytomas were poorly demarcated, unencapsulated, densely cellular masses composed of spindloid neoplastic cells with fibrillary eosinophilic cytoplasm and hyperchromatic nuclei, which were noted to palisade around necrotic areas and multifocally infiltrated the adjacent neuropil and the meninges (Fig 2A and 2B). Oligodendrogliomas were well-circumscribed and composed of solid cellular areas of small, uniform round to polygonal neoplastic cells with centrally located round nuclei with perinuclear halos (Fig 2C and 2D). Occasionally, clusters of glomeruloid microvascular proliferation were interspersed within the neoplastic oligodendrocytes. Mixed gliomas were poorly circumscribed brain tumors with neoplastic cells with features of astrocytes and oligodendrocytes in variable (20–80%) proportions (Fig 2E and 2F). Most of the rat gliomas in this study, graded based on the WHO human glioma criteria, were classified as low grade.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Histomicrographs of representative rat brain tumors resulting from lifetime exposure to radiofrequency radiation.

All the tissues were fixed in an alcohol-based fixative and stained with hematoxylin and eosin. A-B. Astrocytoma: tumor cells with fibrillary eosinophilic cytoplasm palisading around necrotic areas. C-D. Oligodendroglioma: polygonal to round neoplastic cells with hyperchromatic nuclei. E-F. Mixed Glioma: mixed tumor cells of astrocytic and oligodendrocytic morphologies.

https://doi.org/10.1371/journal.pone.0296699.g002

Based on the location of the heart tumors, rat cardiac schwannomas were classified as intramural schwannoma or endocardial schwannoma. Intramural schwannomas were comprised of a neoplastic proliferation of spindle cells within the myocardium that invaded adjacent muscular layers (Fig 3A and 3B), whereas endocardial schwannomas consisted of proliferating subendocardial spindle cells that often invaded the myocardium and protruded into the ventricular lumen (Fig 3C and 3D).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Histomicrographs of representative rat cardiac schwannomas resulting from lifetime exposure to radiofrequency radiation.

All the tissues were fixed in an alcohol-based fixative and stained with hematoxylin and eosin. A-B. Intramural schwannoma involving the left ventricular wall and interventricular septum. C-D. Endocardial schwannoma.

https://doi.org/10.1371/journal.pone.0296699.g003

Next generation sequencing panel design and data analysis

We developed an amplicon based next generation sequencing panel of cancer genes relevant to human gliomas that would allow detection of commonly occurring orthologous mutations in rat gliomas and cardiac schwannomas. To this end, we selected 23 genes that harbor most common genetic alterations in human gliomas based on the COSMIC database, whole genome sequencing and whole exome sequencing data from human gliomas [22]. The target region spanned a total 64,759 bp across the 23 genes. An exemplary IGV plot displays coverage across the Setd2 gene for one brain tumor (S1 Fig). All target genes were sequenced to at least 1000x coverage in brain, ENT and heart samples (S2 Fig) with the exception of Hras that had a mean coverage of ~100x. Amplification was not successful for some exons due to probe design issues (Table 1).

Analysis of brain tumors

Mutation analysis was performed on 14 rat brain tumors and 9 unexposed age matched non-tumor brain controls. Twelve tumors were derived from the RFR exposed groups and 2 tumors developed spontaneously in the unexposed control group. A total of 1,111 point mutations were detected across all brain tumors of which 1,058 were exonic, 4 were within the Tert promoter and 49 were in either the 3’ or 5’ UTR of full-length target genes (S2 Table). The number of SNVs or indels in each rat glioma ranged from 16–234, and include synonymous, nonsynonymous (NS) and regulatory sequence alterations within the targeted genes. Setd2 and Pik3ca mutations were identified in all tumors while mutations in Nf1, Egfr, Atrx, Cic, Rb1, Pdgfra, Arid1a, Notch1, Tp53, Pik3r1, Check2, Idh2, Pten, Fubp1, and Cdkn2a were present in at least 7/14 (50%) tumors. The remaining genes Erbb2, Kras and Tert, and Braf and Idh1, contained variants in five and two gliomas, respectively.

Ensembl’s Variant Effect Predictor (VEP) was used to annotate SNVs and indels. Variants were first classified as ‘Nonsynonymous’, ‘Synonymous or ‘Regulatory’ (Fig 4A). For this study, the ‘Regulatory’ category included variants found in either the Tert promoter or 3’ UTR and 5’ UTR for genes in which all exons were targeted (Table 1). Most mutations were predicted to be NS while less than half were classified as either synonymous or regulatory. The relatively low fraction of regulatory classification was expected since these regions comprise a small portion of the total target area in the panel design.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Mutation landscape of rat brain tumors and exposed non-tumor (ENT) brain samples resulting from lifetime exposure to radiofrequency radiation profiled using the targeted NGS panel.

A. Venn diagram displaying count of variants unique to or shared between brain tumors and exposed non tumor (ENT) tissues. B.

https://doi.org/10.1371/journal.pone.0296699.g004

Nonsynonymous mutations were further classified according to their predicted functional consequence (S2 Table). Missense variants comprised the majority mutations (633), followed by nonsense (66) and splice region variants (16). Fig 5A displays the most deleterious NS mutations identified in each target gene in brain tumors. Co-occuring NS mutations were ranked with decreasing predicted deleteriousness as follows: Nonsense, Splice site, Missense. The Tert promoter mutations were also included in Fig 5A, although deleteriousness is not predicted in non-coding regions. The predominant deleterious mutations in Notch1, Cdkn2a, Kras, Braf, Idh1 Setd2, Nf1, Egfr, Chek2 and Erbb2 were nonsense, while missense mutations predominated in Pik3ca, Cic, Atrx, Pdgfra, Arid1a, Rb1, Pik3r1, Tp53, Idh2, Pten and Fubp1. Splice site mutations were the most deleterious type in Setd2, Pik3ca, Cic, Pdgfra, Rb1, Egfr, Pik3r1, Chek2, Pten and Erbb2 in some tumors.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5.

Mutation landscape of rat brain tumors (A) and exposed non-tumor (ENT) brain samples (B) resulting from lifetime exposure to radiofrequency radiation. The histograms display the distribution of most deleterious mutation types across samples (upper) and genes (right sidebar).

https://doi.org/10.1371/journal.pone.0296699.g005

The most frequently altered residue in the brain tumors was Setd2 p.1439R (Fig 6, S2 Table). Three separate missense mutations, p.1439R>P/L/H, were detected at this codon in 10/14 tumors classified as either exposed glioma, mixed glioma, oligodendroglioma or unexposed spontaneous oligodendroglioma. A missense mutation at Cic p.1539G>V occurred in 6/14 tumors classified as either exposed glioma, unexposed spontaneous glioma, mixed glioma or oligodendroglioma. The remaining mutations occurred in <6 tumors with the large majority (854/999) occurring in one tumor (S2 Table). Two missense mutations were detected in the Idh1, each occurring once in separate tumors. Eight separate NS mutations occurred in Idh2, with six of those (p.178Q, p.174A, p.150E, p.147V, p.138T, p.130K) occurring each once in single separate tumor and two mutations (p.176G, p.177D) each occurring in more than one tumor.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 6. Lolliplots show distribution of non-synonymous SNVs in selected cancer related genes in rat brain tumors resulting from lifetime exposure to radiofrequency radiation.

Mutations are color-coded according to predicted functional consequence and relevance to known orthologous human mutations are indicated with an upside-down triangle.

https://doi.org/10.1371/journal.pone.0296699.g006

No mutations were detected in rat gliomas that had homology to the human IDH1 p.132R or IDH2 p.172R mutations suggesting that rat gliomas are primarily wild-type for IDH hotspot mutations implicated in human gliomas (S2 Table). Furthermore, the Idh1 and Idh2 mutations identified in rat gliomas did not have homology to known human variants in COSMIC database. Four Tert promoter SNVs were detected in the rat gliomas, three of which occurred once in separate tumors. One SNV at position 32,273,028 in the promoter was detected in four tumors including ‘Glioma 4’, ‘Mixed glioma 2’, ‘Oligodendroglioma 2’ and the spontaneous oligodendroglioma indicating that this mutation could be important in glioma tumorigenesis in rat.

Analysis of non-tumor interim exposure group

We hypothesized that detectable early mutation events occur in non-tumor brain tissue subjected to short term RFR exposure. To explore this hypothesis, variant analysis was performed on non-tumor brain tissue from rats exposed to RFR for one year and subsequently compared to SNVs from brain tumors collected after lifetime exposure. In brief, brain specimens were examined from rats exposed to 50 V/m (high dosage, ‘HD’), 25 V/m (mid dosage, ‘MD’) or 5 V/m (low dosage, ‘LD’) from the interim period with no incurring tumor development (e.g. Exposed Non-tumor Tissues (ENTs)). Ten specimens per dosage group and 10 unexposed normal controls were analyzed as previously described. Overall, there were fewer variants identified in the ENTs (432 total unique, average = 14 variants/ENT) relative to the brain tumors (1,111 total unique, average = 79 variants/tumor) (S3 Table). Nf1, Setd2 and Egfr contained variants in at least 17/30 (50%) of all ENTs (Fig 4A). Mutations were detected in Atrx, Rb1, Pik3ca, Arid1a and Tp53 in at least 13/30 (50%) of ENTs. Cic, Pik3r1, Pten, Pdgfra, Notch1, Fubp1, Kras, Chek2, I dh1, Erbb2, Hras, Cdkn2a and the Tert promoter contained variants in < 30% of samples and no mutations were detected in Braf or Idh2. The average number of variants per ENT for the HD, MD and LD groups was 17.4, 9.3 and 21.5, respectively, suggesting that RFR dosage level may not be associated with frequency of mutation events in the target genes.

There were 397 distinct changes within 390 unique residues. Similar to the brain tumors, the majority of changes were nonsynonymous (298/397) (S3 Table). The most frequent alteration was a missense mutation at Nf1 p.1772I>S that occurred in 7/30 ENTs. It is interesting to note that the most common mutation in brain tumors, Setd2 R1439, was mutated in 4/30 ENTs. Thirty eight mutations were shared between brain tumors and ENTs (Fig 4B). A single Idh1 NS mutation without COSMIC relevance was found in one ENT and no alterations were detected in Idh2 (S3 Table).

Analysis of rat cardiac schwannomas

Mutation analysis was performed on 9 cardiac schwannomas and 9 unexposed age-matched non-tumor hearts from controls. Eight tumors were obtained from rats subjected to lifetime RFR exposure and one tumor developed spontaneously in an unexposed control. A total of 744 point mutations were detected in exons (716), the Tert promoter (1) and 3’UTR or 5’UTR (27) across all tumors (S4 Table). Mutations were found in Arid1a, Setd2, Nf1 and Egfr in all cardiac tumors and in Pdgfra for eight of the tumors (Fig 7A). Rb1, Atrx, Pik3ca, Cic, Pik3r1, Pten, Notch1 and Cdkn2a contained mutations in 5–7 of the tumors while Tp53, Chek2, Fubp1, Idh2, Kras, Tert, Erbb2, Idh2 and Braf mutations were identified in less than half of the tumors. Similar to glioma results, NS mutations occurred more frequently than those classified as synonymous or regulatory. Missense mutations were the most common deleterious NS type detected in Arid1a, Setd2, Pdgfra, Pik3ca, Pik3r1, Cdkn2a, Notch1, Pten, Fubp1, Kras and Braf (Fig 7B). The frequency of the most deleterious missense and nonsense mutations were proportional in Cic, Egfr, Atrx, Tp53 and Erbb2 in the tumors. For 3 tumors, the most deleterious mutation was predicted to impact a splice site in Nf1, Pten and Atrx. A stop loss mutation in Rb1 and a translation start site mutation in Pdgfra were the most deleterious type detected in two separate cardiac tumors. A total of 701 distinct changes occurred in 688 residues. The most frequent (3/9 tumors) alterations occurred at Nf1 p.440R, Nf1 p.448H and Pik3ca p.449P (Fig 8). A stop gain was acquired for Nf1 p.440R while Nf1 p.448H and Pik3ca p.449P acquired missense mutations. The most frequent alteration in brain tumors, Setd2 R1439, was also found in 2/9 heart tumors.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 7. Mutation landscape of rat cardiac schwannomas resulting from lifetime exposure to radiofrequency radiation.

The heatmap displays frequency of all mutations identified in tumors. A. The mutation landscape of most deleterious mutation identified in each gene for cardiac tumors. B.

https://doi.org/10.1371/journal.pone.0296699.g007

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 8. Lolliplots show distribution of non-synonymous rat SNVs in selected cancer related genes in rat cardiac schwannomas resulting from lifetime exposure to radiofrequency radiation.

Mutations are color-coded according to predicted functional consequence and relevance to known orthologous human mutations is indicated with an upside-down triangle.

https://doi.org/10.1371/journal.pone.0296699.g008

Translational relevance of rat mutations to human cancers

We queried the COSMIC database to understand the translation relevance of mutations in rat glioma and cardiac schwannomas to mutations in orthologous genes in human cancers. Greater than 50% of variants in Cdkn2a, Erbb2, Pik3r1 and Tp53 and less than 50% in Arid1a, Braf, Egfr, Nf1, Pdgfra and Pik3ca were described in COSMIC for both brain and heart tumors (S2 and S4 Tables). While Setd2 variant in one cardiac schwannoma was described in COSMIC, none of the Setd2 variants in gliomas were in the database. On the other hand, there were notably more Kras mutations in gliomas (10/12, 83%) that were in COSMIC compared to Kras mutations in cardiac schwannomas (2/4, 50%). All 13 Tp53 NS mutations in gliomas had orthologous alterations annotated in COSMIC and eight (61%) of those were associated with central nervous system tissue (CNS) tumors (Fig 6). Somewhat similarly in cardiac schwannomas, 12/15 mutations were orthologous to human mutations in COSMIC and three (25%) of those were associated with CNS tissue.

Of the 1,111 point mutations in rat gliomas, 282 had homology to mutations in COSMIC and 187 of those were predicted to be NS in rat (S2 Table). Of the 187 NS orthologous mutations, 28 were implicated in human gliomas in COSMIC. Most rat mutations with homology to mutations implicated in human gliomas occurred in one tumor. However, mutations at Cic p.319A. Tp53 [p.71Q, p.124Y, p.142Q] and Nf1 [p.440R, p.1080S] were found in at least 2/14 tumors (Fig 6). Of the 28 NS orthologous mutations implicated in human gliomas, 24 were classified as pathogenic in COSMIC and 17 have conserved amino acid change for rat and human including Tp53 [p.140P>L, p.142Q>*, p.149P>L, p.156R>C, p.260G>V, p.271R>C], Nf1 [p.192R>*, p.379C>Y, p.436S>L, p.440R>*, p.758Q>*, p.1080S>L], Erbb2 p.39P>S, Egfr p.596P>S, Pik3r1 p.464D>N, Cdkn2a p.26A>T and Arid1a p.>E.

Of the 744 cardiac schwannoma point mutations, 205 were orthologous to COSMIC mutations and 146 of those were classified as NS (S4 Table). The orthologous NS mutations were found in Cic, Tp53, Nf1, Erbb2, Chek2, Egfr, Pdgfra, Pik3ca, Fubp1, Pik3r1, Kras, Braf, Cdkn2a, Arid1a, Setd2 and Idh1. Taken together, these results suggest that rat glioma and cardiac schwannoma mutation profiles have 25–27% similarity to known human mutations but most target genes investigated in the panel harbor alterations that are distinct to rat. Genes that appear to be an exception to this are Cdkn2a, Erbb2, Pik3r1 and Tp53 as well as Fubp1 in the heart tumors.

Relevance of mutations in rat gliomas and cardiac schwannomas compared to human cancers

We discovered that rat tumors lack the Idh1 R132 and Idh2 R172 hotspot mutations that are widely implicated in human gliomas (mainly in WHO grade II and III gliomas and secondary glioblastoma). These results are comparable to our preliminary results based on Sanger sequencing of a distinct set of rat gliomas that originated spontaneously or due to chronic chemical exposures (data not presented). Interestingly, the absence or rarity of mutations in IDH1 R132 and IDH2 R172 was also demonstrated in gliomas from dogs, another important model system with significant human relevance [22, 23]. Idh1/2 did harbor other nonsynonymous mutations but only three of their human homologs were found in the COSMIC database and the functional significance of those alterations remains to be tested. This was an unexpected discovery given the prevalence of IDH mutations in human gliomas. Compared to gliomas with wild-type IDH, gliomas with IDH1/2 mutations have better prognosis and outcomes, and these genes have become potential therapeutic targets for gliomas [24, 25]. While tremendous progress has been made towards identifying important IDH mutation profiles, co-occurrence with other genetic alterations can be more predictive for tumor classification and prognosis [24, 26]. For example, prognosis can be stratified based on co-occurrence of IDH mutations with presence or absence of 1p/19q deletion, as well as mutations in the Tert promoter, Tp53 and Egfr. Despite the absence of Idh1/2 mutations in rat and canine gliomas, these model organisms with wild-type IDH gliomas harbor mutations in other genes relevant for human glioma research.

In addition to IDH, mutations in the Tert promoter contribute to glioma classification and may have prognostic clinical value [26]. Furthermore, 70–74% of glioblastomas and oligodendrogliomas contain mutations in the Tert promoter [27]. TERT is the catalytic subunit of telomerase and mutations in this gene lead to telomerase activation in gliomas. We detected remarkably few mutations in the Tert promoter. An interesting discovery was that the Tert g.32,273,028C>T SNV was found in 4/14 gliomas and 3/9 cardiac schwannomas. This mutation is within a region of poor homology to human gene and thus the potential relevance to human cancer is not known at this time. However, this SNV is a candidate for follow up evaluation given its prevalence in the RFR exposed rat tumors.

In humans, two hotspot TERT promoter mutations, C228T and C250T (Hg19), are associated with several cancers. Recently, these mutations were confirmed in tumors as well as in circulating cell-free DNA from glioma patients [28] and are seen in majority (~75%) of IDH wild-type glioblastomas. We did not detect these alterations in the rat gliomas, ENTs or schwannomas. Taken together, these results suggest that there are distinct differences in the relevance of Idh and Tert promoter mutations between rat and human tumors. More clear patterns of mutation association could emerge if the study is extended to include detection of the 1p/19q deletion, for example, performing whole exome or whole genome sequencing if more samples are available for examination. Furthermore, it’s possible genomic alterations incurred due to RFR exposure could be distinct from those examined in the majority of human studies because of the nature of the mutagenic processes leading to tumor development and progression.

Similar to Idh1/2, mutations in Atrx, Notch1, Pten, Rb1 and Setd2 were largely absent from COSMIC. This was particularly surprising for Setd2 given that it was altered in every glioma and cardiac schwannoma in this study, with only one nonsense mutation (SETD2 p158R> *) reported in the COSMIC database. It is notable that this alteration is associated with human tumors of the central nervous system. In humans, SETD2 is responsible for methylation of H3K36 and loss of function (LOF) leads to decreased H3K36me3 levels. Reduced H3K36me3 levels disrupt gene regulation and chromatin stability and is associated with tumor development [29]. In humans, an association was determined between high-grade gliomas and SETD2 LOF, specifically due to alterations that result in protein truncation [30]. In the rat tumors, 20%, 11% and 11% of NS mutations were predicted to be nonsense in the gliomas, ENT brains and cardiac schwannomas, respectively. It is possible that Setd2 LOF, irrespective of precise mutation position, contributes to tumorigenesis in rat as indicated in humans. The Setd2 R1439 residue was one of the most frequently altered amino acids with missense mutations detected in 10/14 gliomas, 4/30 ENT brains and 2/9 cardiac schwannomas. Relevance of this mutation to human cancer is not described in COSMIC or reported in the scientific literature at this time. Investigation revealed the residue lies with a region of Setd2 that is involved in an interaction with α-tubulin and disruption of tubulin methylation by Setd2 contributes to genomic instability [31]. Further examination and validation of this variant should be considered.

In contrast to the above genes with mutations that seemingly have low relevance to human cancers, alterations in Tp53, Cdkn2a, Erbb2, Chek2, Kras and Pik3r1 with homology to COSMIC variants were relatively common in rat gliomas and cardiac schwannomas (e.g. > = 50%). All 13 Tp53 NS mutations in gliomas had orthologous alterations annotated in COSMIC and 61% of those were associated with CNS tumors; somewhat similarly, most Tp53 mutations (12/15) in cardiac schwannomas were orthologous to mutations in human genes in COSMIC. Pik3ca, which was altered in the majority of tumors, harbored NS mutations in gliomas (40%) and cardiac schwannomas (46%) with COSMIC relevance, however in contrast to Tp53, they are not associated with gliomas or CNS tumors at this time.

Less than half of the mutations in Pdfra, Fubp1, Nf1, Egfr, Braf, Pik3ca, Cic and Arid1a were described in COSMIC suggesting that there are some mutations in these genes that are unique to rat or are unknown in human at this time. Pik3ca NS mutations were detected in 13/14 (93%) brain tumors and 7/9 (78%) heart tumors. Of the 52 NS mutations in rat gliomas, 21 (40%) are described in the COSMIC, yet none are implicated in human gliomas or other tumors of the central nervous system. Similarly in cardiac schwannomas, 12/26 (46%) Pik3ca NS variants were described in COSMIC without relevance to central nervous system tumors. Interestingly, we detected a c.3040C>T nonsense mutation in 3/14 gliomas which is not in the COSMIC database but was reported as a human nonsense mutation [32]. These results suggest that Pik3ca mutations might be associated with rat glioma and schwannoma tumor progression but the relationship to tumor type seem to differ between human and rat. It is interesting to note that Arid1a NS mutations were detected in all cardiac schwannomas and 41% of these were orthologous to COSMIC variants associated with a variety of human cancers. Meanwhile, 10/14 of gliomas harbored NS mutations in Arid1a and 43% of those were orthologous to COSMIC variants. An exome study in humans found that 37/125 (29%) of schwannomas have NS mutations in ARID1A or ARID1B [33]. ARID1A and ARID1B are components of the SWI-SNF complex that are involved in chromatin remodeling and are implicated in several human cancers. It is possible that the significance of Arid1a to schwannoma progression may be similar in the rat.

Limitations

Despite lifetime RFR exposure and a large sample size of rats in each exposure group, tumor incidences in our study were low which limited the power of our study design. It is difficult to perform rigorous statistical testing or draw strong conclusions about mutational patterns for a tumor population based on a limited sample size. In general, identification of false positives in somatic variant detection is complicated by the occurrence of low frequency variants that approach known sequencing error rates. Recent advancements in sequencing strategies, such as error corrected duplex sequencing and NanoSeq, have potential to improve the precision of variant identification through reduced error rates in base calling [34, 35]. Further, tumor heterogeneity may contribute to variability in VAF both within tumor and across tumors from different rats. Typically, a large sample size can improve the ability to distinguish between false positives and true variants in the population. In addition to a relatively small sample size, there are no known rat glioma or schwannoma variants described in the literature and cross-species validation is preliminary at this time.

In spite of these challenges, we were able to identify variants that are likely to be important in rat glioma and cardiac schwannoma progression and warrant further examination. The availability of genomic matched controls from kidney tissue of the RFR exposed rats enabled us to exclude germline variants. We also employed two separate variant callers including Strelka2, which is commonly used in the clinical setting, and deepSNV which is designed to detect rare variants in targeted amplicon data. Evaluation of variant quality metrics and sequencing depth revealed variability across samples such that, after much consideration, a single set of cutoffs did not seem appropriate. Thus, in contrast to most mutational studies that apply broad cutoffs to remove false positives, we applied customized, adaptable filters that we were determined separately for each sample. Variants from unexposed non-tumor brain and heart tissues were removed to further reduce false positives. Finally, only variants that were present in both Strelka2 and deepSNV filtered calls were retained to achieve high confidence variant sets.

In summary, our results demonstrate that regardless of their etiology (due to lifetime RFR exposure or arising spontaneously), rat gliomas are primarily Idh1/2 wild type unlike most human gliomas. Histologically, most of the rat gliomas resemble diffuse low-grade gliomas in humans and such gliomas that do not harbor IDH1/2 mutations in humans are known to have poor prognosis. The genetic alterations in other cancer genes evaluated in this panel provide novel insights into tumor progression in rat gliomas and cardiac schwannomas. The relevance of specific mutations to human cancers is variable, with some genes (Tp53, Cdkn2a, Erbb2, Chek2, Kras and Pik3r1) harboring many alterations with COSMIC relevance while the opposite is true for other target genes (Idh1/2, Atrx, Notch1, Pten, Rb1 and Setd2). Several of these conserved mutations in rat tumors do not have comparable alterations in the COSMIC database, suggesting that the orthologous mutations could have different functional consequences in rat carcinogenesis and deserve further study. An important consideration is that molecular differences underlying mutational processes contribute to distinct mutational patterns which could be the result of similar etiology, albeit by different mechanisms.

Several of the variants that were detected in gliomas were also observed in non-tumor brain tissues from interim time point providing an insight into the molecular pathogenesis in rodent carcinogenicity studies and these strategies may be utilized to potentially estimate the cancer hazard risk in shorter term animal studies. Finally, this targeted mutation panel may be refined using data from whole genome or exome sequencing of rat tumors and performing error corrected duplex sequencing to increase the sensitivity to detect rare mutations in exposed non-tumor tissues from early time points.