Somatic expression of piRNA and associated machinery in the mouse identifies short, tissue-specific piRNA

ABSTRACT

Piwi-interacting RNAs (piRNAs) are small non-coding RNAs that associate with PIWI proteins for transposon silencing via DNA methylation and are highly expressed and extensively studied in the germline. Mature germline piRNAs typically consist of 24–32 nucleotides, with a strong preference for a 5ʹ uridine signature, an adenosine signature at position 10, and a 2ʹ-O-methylation signature at the 3ʹ end. piRNA presence in somatic tissues, however, is not well characterized and requires further systematic evaluation. In the current study, we identified piRNAs and associated machinery from mouse somatic tissues representing the three germ layers. piRNA specificity was improved by combining small RNA size selection, sodium periodate treatment enrichment for piRNA over other small RNA, and small RNA next-generation sequencing. We identify PIWIL1, PIWIL2, and PIWIL4 expression in brain, liver, kidney, and heart. Of note, somatic piRNAs are shorter in length and tissue-specific, with increased occurrence of unique piRNAs in hippocampus and liver, compared to the germline. Hippocampus contains 5,494 piRNA-like peaks, the highest expression among all tested somatic tissues, followed by cortex (1,963), kidney (580), and liver (406). The study identifies 26 piRNA sequence species and 40 piRNA locations exclusive to all examined somatic tissues. Although piRNA expression has long been considered exclusive to the germline, our results support that piRNAs are expressed in several somatic tissues that may influence piRNA functions in the soma. Once confirmed, the PIWI/piRNA system may serve as a potential tool for future research in epigenome editing to improve human health by manipulating DNA methylation.

Introduction

Piwi-interacting RNAs (piRNA) are a class of small non-coding RNAs that associate with a subset of the Argonaute protein family that directs the silencing of transposons. In recent studies, piRNAs have also shown important functional roles in epigenetic reprogramming, and regulation of transcription, translation, and development [1–3]. As a class, piRNA is ubiquitous and diverse in animal germlines, with over 500,000 unique piRNA-derived sequences in web-accessible databases [4–9]. Most of the piRNA discoveries have been deposited in several databases including piRBase [6] piRNAQuest [7], piRNABank [8], and piRNA cluster database [9]. Overwhelmingly, these databases survey the piRNA expression from the germline at various developmental stages of the mouse.

Structurally, currently characterized germline piRNAs are roughly 24–32 nucleotides in length and have a strong preference for a 5ʹ uridine signature and an adenosine signature at position 10 [10,11]. Unlike other small non-coding RNAs such as microRNA (miRNA), mature piRNAs are modified by 2ʹO-methylation at the 3ʹ end by Hen1, a conserved piRNA methyltransferase, to ensure RNA stability [1,12]. In mice, the genomic regions encoding piRNAs are often clustered in groups with average length of 20–90 kb. Approximately 20% of recovered germline piRNAs match transposons and the remainder match unique genomic sites [13]. Many of the piRNAs overlap with transposons or their remnants, thereby enabling the targeting and silencing via DNA methylation of the RNAs transcribed from the transposons in the germline [14].

The Argonaute protein, PIWI (p-element induced wimpy testis), was originally discovered in Drosophila melanogaster by Haifan Lin’s laboratory, who first identified the destructive effect on the testis observed upon mutating the PIWI gene [15,16]. Subsequently, several groups demonstrated that piRNAs bind to PIWI-interacting proteins in mouse testis [17,18]. The Piwi-group homologs in mice are named MIWI, MILI, and MIWI2, orthologous to PIWIL1, PIWIL2, and PWIL4 in humans, respectively. Mutations in any of these paralogs cause male sterility due to unchecked transposon activation [13]. Together, the PIWIL proteins form the necessary machinery critical for piRNA synthesis [1]. Initially, piRNAs and their associated proteins were thought to be generated and amplified only in germ cells via a complex ‘ping-pong’ cycle, distinct from the miRNA dicer mechanism [19]. Some researchers debate whether piRNAs have somatic function; however, Haifan Lin recently noted in a 2014 Nature review that somatic functions of PIWI proteins were documented, in spite of the common perception that PIWIL proteins and piRNAs are germline-specific [12]. There is increasing evidence for piRNA expression in somatic tissues, albeit at lower levels than in germline, from diverse lineages such as the brain and kidney, which are conserved among sea slug, mouse, and macaque [11,20,21]. Recent research indicates that piRNAs and PIWI proteins are expressed and functional in the hippocampus through immunofluorescence and inhibition of specific piRNA, and DNA methylation defects noted in PIWIL2 null mouse brains [3,22]. It is still unclear, however, how these non-germline piRNAs are generated and processed [19].

We now provide comprehensive evidence for the presence of piRNA synthesizing machinery and piRNAs in somatic tissues, albeit at significantly lower levels than germline, from the three germ layers (ectoderm, endoderm, and mesoderm). We utilized sodium periodate treatment to specifically select for small RNAs containing the 2ʹ O-methylation signature [23,24]. We also incorporated novel bioinformatics techniques and quality controls to select for piRNAs based on their characteristics. The results indicate that PIWIL machinery and piRNAs are present in mouse soma, and that somatic piRNAs tend to be shorter in length, with tissue-specific expression patterns. This information is critical for the advancement of future somatic cell epigenome editing studies, where piRNA and its associated proteins may serve as a potential tool for in vivo and in vitro epigenetic manipulation.

Results

piRNA synthesizing machinery is expressed in somatic tissues

To characterize the somatic expression of piRNA, we analysed mRNA expression levels of the piRNA machinery (PIWIL proteins) responsible for synthesizing piRNA in the mouse: PIWIL1, PIWIL2, and PIWIL4. The RNA integrity scores (RIS) are summarized in Table 1, to demonstrate the quality of RNA used for qRT-PCR experiments. PIWIL1 mRNA expression was observed in the germline (testis and ovary) as well as several somatic tissues including the brain, hippocampus, and heart (Figure 1 (a)). The testis showed the highest PIWIL1 mRNA expression, consistent with previous results [10]. The brain, hippocampus, and ovary showed the second highest expression while the heart indicated very low levels of PIWIL1 mRNA expression. Kidney and liver tissues showed no expression of PIWIL1 mRNA in any of the tested animals. PIWIL2 mRNA expression was observed in all tested tissues (Figure 1 (b)). Similar to PIWIL1 observations, PIWIL2 mRNA expression was highest in the testis and lowest in the kidney and liver; the second highest levels were observed in ovaries. In contrast to the low levels detected in PIWIL1, the highest somatic PIWIL2 mRNA expression was detected from the heart, followed by the brain and hippocampus. PIWIL4 mRNA expression was also observed in all tissues (Figure 1 (c)). The testis and liver showed the highest and lowest levels of PIWIL4 mRNA expression, respectively, with ovaries being the second highest. Interestingly, the kidney showed the highest average somatic expression for PIWIL4. Brain, hippocampus, and heart showed low levels of PIWIL4 mRNA expression. Thus, PIWIL1, 2, and 4 mRNA were present in somatic tissues, in addition to the tissues specific for the germline.

Table 1.

Summary of RNA integrity scores (RIS).

Sample	Male 1	Male 2	Male 3	Female 1	Female 2	Female 3
Brain	7.8	7.7	8.8	7.5	7.6	7.1
Hippocampus	8.2	7.3	8.3	7.1	7.0	7.1
Heart	7.2	8.8	5.6*	8.3	N/A*	7.9
Kidney	7.6	8.0	7.7	7.3	7.5	4.4*
Liver	8.6	7.1	8.4	8.3	7.5	9.1
Testis/Ovary	8.3	8.7	8.3	7.8	8.8	9.1

*RIS scores lower than 7.

Figure 1.

mRNA expression of piRNA machinery in mouse tissues. Quantitative RT-PCR analysis using the cDNA generated from three male (blue) and three female (red) sets of two-month-old adult brain, hippocampus, heart, kidney, liver, and testis/ovaries. The Ct values for each promoter was first normalized with an internal control (β-actin) to generate ΔCt and subsequently used for calculating gene expression levels (ΔΔCt). A range of average Ct values for each tissue is also presented below the graph to indicate the absolute expression levels of: (a) PIWIL1 mRNA expression (b) PIWIL2 mRNA expression and (c) PIWIL4 mRNA expression. The error bars represent the average from standard deviations calculated by triplicates.

Detection of piRNAs using sodium periodate treatment

Sodium periodate treatment is selective for piRNAs, which are the predominant small RNA species that contain the 2ʹ-O-methylation 3ʹ modification. The 2ʹ-O-methylation prevents base extension during sequencing reactions [12,23]. To identify piRNAs in each tissue, we tested for read pile-ups (peaks) that were significantly higher in the sodium periodate-treated samples compared to the untreated, using all mapped reads (Figure 2). Many of the piRNA-like transcripts appeared multiple times across the genome with 100% sequence identity, thus we quantified piRNAs both by number of unique sequences and number of genomic sites.

Figure 2.

Computational pipeline for small RNA sequencing analysis. Small RNAs (<200bp) were isolated from cortex (brain), hippocampus (brain), liver, kidney, and testis tissues from a pair of 2-month-old male mice. A subset of the small RNAs were treated with sodium periodate to identify 2ʹ-O-methylation signature, while the remaining untreated small RNA served as a control. All samples were library prepped, multi-plexed 10 per lane, and sequenced on an Illumina v4 HiSeq2500 with 50-bp single-end reads. The resulting sequence reads were first trimmed for adapters and size selected for 10–45 bp reads for quality control purposes (¹Cutadpter). The sequences were then aligned to the mm10 mouse genome assembly (²STAR aligner) to filter out GENCOE exons, miRNAs, other small RNAs and tRNA species from analysis. The resulting reads were compared with periodate treated and control samples to identify peaks (³ Pepr peakcalling.pl), which were defined as piRNA-like transcripts. The piRNA expression levels (⁴BedTools multicov) and sequence motifs were generated, allowing for ±3bp flanking region for each tissue (⁵WebLogo). Somatic piRNA tissues were compared based on location, sequence composition, and relative expression. The total piRNA peaks/reads were compared with existing piRNA databases: piRBase, piRNAbank, piRNA cluster database, and piRNAQuest.

Detection of piRNAs in testis

We first sought to assess the quality of our results in testes before characterizing the properties of the piRNAs identified in somatic tissues. To do this, we compared the piRNAs identified using our pre-processing and peak detection pipeline in testes to the piRNAs found in mouse testes in four other public piRNA databases (piRBase, piRNAbank, piRNA cluster DB, and piRNAQuest) [6–9]. Overall, we identified 24,777 unique sequences of piRNAs (40,808 genomic locations) in testis. Additional information about the piRNA sequences and locations are reported in the following sections. We observed that the piRNAs we detected in testes are strongly enriched with those found by all four of the other databases. In total, 40,581 of the 40,805 periodate-treated testis piRNA sequences from two biological replicates were collectively found in the four publicly available, germline piRNA databases, validating that the piRNA sequences generated through this study are genuine (Supplemental Figure 1).

Detection of piRNAs in somatic tissues

In depth investigation of several features of mature piRNAs were used to further confirm the presence of piRNA in somatic tissues (Figure 3). A mature germline piRNA has been defined as being 24–32 bp in length, with a bias towards having a 5ʹ uridine signature, an adenine signature at the 10th position, and a 2ʹ-O-methylation signature at its 3ʹ end (Figure 3 (a)). However, it is unclear whether all features are equally important for piRNA structure and function [1]. Small RNA size selection, together with sodium periodate treatment is involved in eliminating small RNA fractions that a 2ʹ-O-methylation modification at the 3ʹ end, resulting in small RNA fractions enriched in piRNA. The peak length distribution of piRNA-like transcripts in testis confirms that the highest frequency of piRNAs lies in the expected 24-32 bp range, and is consistent with previous studies (Figure 3 (b)) [10,11]. In contrast, the somatic tissues contain a high frequency of piRNA-like transcripts at a lower range of distribution despite having consistent 2ʹ-O-methylation signatures at the 3ʹ end. Hippocampus piRNA-like transcripts in particular demonstrate a similar frequency distribution to that of testis, given the slight piRNA plateau around 24-32bp, setting it apart from other somatic tissues that shows a declining slope (Figure 3 (b)). Furthermore, in testis, 22,734 out of 39,229 of the piRNA-like transcripts contained an adenosine signature at the 10th position as well as a uridine signature at the 5ʹ end (Figure 3 (c)). The somatic tissues (cortex, hippocampus, kidney, and liver) also displayed incidence of adenosine enrichment at the 10th position (Figure 3 (c)). Overall, these results support that somatic piRNAs are structurally similar to germline piRNAs as they include a 5ʹ uridine signature, an adenine signature at the 10th position and a similar 3ʹ modification, despite their slightly shorter length.

Figure 3.

Somatic piRNA characteristics. (a) Schematic representation of a typical germline piRNA. The ‘N’ symbolizes an RNA nucleotide (A- adenine, G- guanine, C- cytosine, or U-uracil). The subscripts signify the nucleotide number to infer the length of a piRNA, which may range from 24-32bp as indicated by the bracketed area. Germline piRNA characteristics include 24-32bp in length shown in blue, 5ʹ uridine signature (red arrow), adenosine signature at the 10^th position (red arrow), and a 2ʹ -O-methylation signature at the 3ʹ end (red arrow). (b) Length distribution of peaks identified by our computational pipeline. The y-axis of histogram is represented in a logarithmic scale. The yellow bars represent peaks in which length ≥20bp and were defined as piRNA-like transcripts in this study. The blue bars represent shorter peaks (<20bp), which were excluded from further analyses. (c) Sequence logos of somatic piRNA-like transcripts. The sequence motifs were generated using the first 20bp from the 5ʹ end of piRNA-like transcripts, allowing for ±3bp flanking region to determine the strand direction. The number of peaks indicating a uridine signature at the 5ʹ end (U1) and an adenine at the 10^th position (A10) are shown directly above the sequence motifs to its corresponding tissue.

Somatic piRNAs are shorter in length

Sequenced piRNAs were analysed to better understand the length distributions for somatic tissues in comparison with the germline. The total number of all mapped peaks was calculated using three different minimum length limitations (20, 22, and 24 bp) in an effort to evaluate piRNA length based on piRNA-like peaks in somatic tissues (Table 2). In testis, where the highest number of total piRNA-like peaks was detected, the minimum length made almost no difference, with 40,808, 40,564, and 40,270 piRNAs detected when using peak lengths greater than 20 bp, 22 bp, and 24 bp, respectively. Thus, only a 538 piRNA-like peak difference (1.3% relative increase in piRNA-like peaks) was observed reducing stringency from 24 bp to 20 bp. Somatic tissues including the cortex, hippocampus, kidney, and liver also indicated the presence of piRNA-like peaks that were several magnitudes fewer than in testis (175 to 5,494 peaks). In contrast to testis, the somatic tissues indicated higher levels of piRNA-like peak differences by reducing the length limit from 24 bp to 20 bp, with 114.5% relative increase in piRNA-like peaks for cortex, 42.8% for hippocampus, 133.9% for kidney, and 132.0% for liver. The somatic tissues also displayed an incidence of adenine at the 10th position for cortex (55%), hippocampus (57%), kidney (55%), and liver (54%), similar to that of testis (56%). Since the initial analysis suggested the existence of several piRNA characteristics at peaks greater than 20 bp in somatic tissues, this criterion was used for downstream applications in identifying piRNA-like transcripts, which will be referred to as piRNA transcripts hereafter.

Table 2.

Summary of piRNA characteristics in somatic and germline piRNA-like peaks with varying minimum peak lengths from 20, 22, and 24bp and ±3bp flanking for both unique and multi-mapped peaks.

Tissue	Minimum peak length (bp)	Total number of peaks	Difference in peak number relative to 24bp (%)	Number of peaks with 1st position Uridine (U1)	Number of peaks with 10th position Adenine (A10)	U1 (%)	A10 (%)
Cortex	20	1963	114.5	1856	1082	95%	55%
	22	1225	33.9	1160	718	95%	59%
	24	915	-	875	503	96%	55%
Hippocampus	20	5494	42.8	5109	3138	93%	57%
	22	4581	19.1	4274	2679	93%	58%
	24	3846	-	3574	2252	93%	59%
Kidney	20	580	133.9	529	317	91%	55%
	22	395	59.3	361	216	91%	55%
	24	248	-	226	133	91%	54%
Liver	20	406	132.0	370	218	91%	54%
	22	280	60.0	252	161	90%	58%
	24	175	-	150	107	86%	61%
Testis	20	40,808	1.3	39,229	22,734	96%	56%
	22	40,564	0.7	37,729	23,043	93%	57%
	24	40,270	-	37,320	22,794	93%	57%

piRNA expression based on sequence similarity and position

We quantified piRNAs both by number of unique piRNA sequences and number of piRNA transcripts based on locations/positions in the mouse genome (mm10). These two quantifications were used to examine somatic piRNA properties in comparison with the germline. Based on Table 3, testis resulted in the highest average number of total reads for the two sodium periodate treated samples (51.3 million reads), followed by liver (20.6 million), cortex (17.2 million), hippocampus (6.2 million), and kidney (2.5 million), for periodate-treated samples. However, the somatic piRNA showed a higher percentage of alignment with the genome ranging from 73.8% to 80.8% compared to testis, which resulted in a 66.9% alignment rate without any mismatches (Table 3). This result indicates that perfectly mapped piRNA transcripts may be more abundant in the somatic tissues as compared to testis. Consistently, the testis showed the highest total number of piRNAs, with 11,821 uniquely mapped and 6,149 multi-mapped piRNA sequences and 11,824 uniquely mapped and 22,625 multi-mapped piRNA transcripts based on position/location (Supplemental Table 1). The somatic tissues revealed 1,681 multi-mapped piRNA transcripts for cortex, 2,466 uniquely mapped and 3,123 multi-mapped piRNA transcripts for hippocampus, 146 unique and 379 multi-mapped piRNA transcripts for kidney, and 611 unique and 231 multi-mapped piRNA transcripts for liver (Supplemental Table 1).

Table 3.

Summary of piRNA expression based on location and sequence similarity for two, sodium periodate treated samples.

Tissue	Average total reads (Millions)	Alignment rate (%)	piRNA sequences	piRNA transcripts	Median counts per million reads per peak
Cortex	17.2	77.9	816	1,963	8.8
Hippocampus	6.2	80.8	3,147	5,494	4.1
Kidney	2.5	77.9	369	580	27.7
Liver	20.6	73.8	323	406	4.4
Testis	51.3	66.9	24,777	40,808	1.7

Somatic piRNAs are depleted in repetitive sequences compared to testis

We investigated both the unique and multi-mapped piRNAs based on location to determine the extent to which piRNAs in various tissues overlap with repetitive sequences, similar to the germline. Testis contains the highest number of multi-mapped piRNA transcripts, which also indicated the highest percent of bases that overlapped with repetitive regions (masked) in the mouse genome (mm10) at 24.4% through RepeatMasker (Figure 4). Remarkably, in testis, 79% of the repeats masked were LTRs, which includes 68% of Class II ERV repeats. According to this survey, testis also consisted of 1% small RNAs. In contrast to testis, the majority of the normalized multi-mapped somatic piRNA transcripts were small RNAs representing 90% for cortex, 15% for hippocampus, 100% for kidney, and 82% for liver, respectively (Figure 4, Supplemental Table 2). RepeatMasker identified simple repeats that overlapped with piRNA transcripts (4%) exclusively in the cortex, while SINE elements were found to overlap piRNA transcripts in the cortex (5%), hippocampus (22%), and liver (18%). Interestingly, hippocampus also showed 24% of multi-mapped piRNAs matched to LINE elements and 33% to LTRs, which also included 3% to class II ERVs. Overall, these data indicate that somatic tissues contain an increased number of uniquely-mapped piRNAs, while the hippocampus includes a unique repeat element composition in comparison with other somatic tissues.

Figure 4.

The piRNA transcripts represented by various repeat classes. The pie charts represent the distribution of repeats for both somatic piRNAs (cortex, hippocampus, kidney, and liver) and germline piRNAs (from testis) detected from periodate treated samples in two biological replicates. The RepeatMasker software was used to annotate the repeats. The number of bases from the piRNA transcripts that overlapped with repetitive regions (‘masked’ through RepeatMasker) at each tissue and the mouse genome (mm10) are represented as a percentage. The repeat classifications include short interspersed elements (SINE), long interspersed elements (LINE), long terminal repeat elements (LTR), DNA elements, small RNA elements, and unclassified elements. The normalization is relative to the mouse genome mm10 RepeatMasker information.

piRNA comparisons among somatic tissues and testis

As a higher percent of somatic piRNA transcripts mapped uniquely to the genome, we determined how many expressed piRNAs were shared between testis and the other tissues based on sequence identity. We found that testis shared 63 piRNAs with cortex, 45 with hippocampus, 49 with kidney, and 32 with liver (Figure 5 (a)). We also found that a higher percent of piRNAs were expressed among the somatic tissues as compared to testis, with results indicating that at least 45 piRNA sequences are shared across cortex, hippocampus, kidney, and liver (Figure 5 (b)). Only 19 of the 45 shared somatic piRNA sequences overlap with testis (Supplemental figure 2), presenting 26 unique piRNAs exclusive to all four somatic tissues (Supplemental Table 3). To understand the stability of the piRNA products identified, an RT-PCR-based approach validates the presence of piRNA in testis, cortex, and liver (Supplemental figure 2). Hippocampus showed the highest number of unique piRNA sequences (89.8%) that were not shared with any other somatic tissue, while the kidney displayed the fewest number of tissue-specific piRNA sequences (38.8%). The brain (hippocampus and cortex) indicated the highest overlap of 144 piRNA sequences shared between any two of the somatic tissues (Figure 5 (b)). Similar observations were made upon surveying the unique piRNAs based on their location/position; testis shared 202 piRNA genomic locations with cortex, 166 with hippocampus, 157 with kidney, and 54 with liver (Supplemental figure 3 (a)). At least 70 piRNA locations were shared among cortex, hippocampus, kidney, and liver, of which 40 are exclusive to somatic tissues (Supplemental figure 3 (b) and Supplemental Table 4). Interestingly, 14 out of 40 (35%) of the piRNA position overlap with the X chromosome (Supplemental Table 4). Taken together, these results indicate the tissue-specificity of piRNAs, with hippocampus showing the highest number of piRNAs expressed from the tested somatic tissues.

Figure 5.

piRNA sequence comparisons between somatic and germline tissues. (a) Tissue-specific piRNA sequences. The four two-way Venn diagrams represent the unique piRNA sequence comparisons between testis (pink) and cortex (gray), hippocampus (blue), kidney [47], and liver (orange), respectively. (b) Somatic piRNA sequence comparison between the three germ layers. The four-way Venn diagram indicates the piRNA sequences that are unique to somatic tissues including the brain and hippocampus for ectoderm, kidney for mesoderm, and liver for endoderm. It also indicates the piRNA sequences that are common between two or more tissue types.

Somatic piRNA comparison with existing piRNA databases

We compared periodate-treated somatic piRNA data with the four publicly available databases to further confirm the validity of the piRNA sequences generated from our findings. It is important to note that the four databases contain samples derived predominantly from the germline and a few from the hippocampus and were identified by PIWIL co-immunoprecipitation and/or size selection, but not with sodium periodate treatment [6–9]. The piRBase contains the highest number of total piRNAs with 10,211,084 annotations, followed by piRNAQuest (3,646,281), piRNAbank (659,639), and piRNA cluster database (29,349), with only 2,591 piRNA annotations shared between all four databases (Figure 6 (a)). Moreover, some of them are longer sequences consisting of multiple piRNAs. Interestingly, 1,629 of the 1,963 total somatic piRNA transcripts were found in at least one of the public databases for the cortex, with 22 sequences common to all four databases. In hippocampus, 3,330 of 5,494 piRNA transcripts were found in at least one of the databases, with 10 sequences common to all databases (Figure 6 (b)). Similar results were observed from the remaining somatic tissues, with kidney showing 522 of 580 and liver with 358 out of 406 total piRNA transcripts overlapping with at least one database. The majority of the piRNA sequences were found in the piRBase and piRNAQuest databases for all the somatic tissues. These observations, together with the testis piRNA analysis (Supplemental Figure 1) and high rate of the observed 10th position uridine, indicate that germline and somatic piRNAs detected via periodate treatment are authentic.

Figure 6.

Surveys of public piRNA databases and periodate treated piRNA for brain tissues in the current study. (a) Comparisons of currently available piRNA databases. The four-way Venn diagram represents the piRNAs annotated by the four available piRNA databases: piRBase (10,211,139), piRNAbank (659,639), piRNA cluster database (29,349), piRNAQuest (3,646,281). (b) The cortex (grey) and hippocampus (blue) piRNA transcripts comparison with the four databases. Based on the database comparison, 1,629 of 1,963 and 3,330 of 5,494 piRNA transcripts were also annotated in at least one of the four databases for cortex and hippocampus, respectively. Each Venn diagram represents periodate treated samples derived from two-month-old adult mice.

The piRNAs are clustered based on their proximity to each other. To do so, the distance between each piRNA within a cluster was calculated to be less than or equal to 200 bp (Supplemental table 5). Compared to the piRNA cluster database (top of Supplemental table 5), 50% of the testis piRNA formed clusters, followed by kidney (8%), cortex (6%), liver (5.6%), and hippocampus (2%). Based on our data (bottom of Supplemental table 5), the highest number of total piRNA clusters was observed from testis with an average of 4.6 piRNAs per cluster, while the somatic tissues averaged around 2 piRNAs per cluster. The numbers of short piRNAs (20–25 nt) in these clusters was higher in somatic tissues than in testes (an average of 1.18–2 short piRNAs per cluster for somatic tissues versus 0.19 in testes). In contrast, the highly expressed piRNA clusters contained relatively similar average piRNAs per cluster for all tissues with 2.7 piRNAs for testis, followed by cortex (1.78), kidney (1.5), liver (1.43), and hippocampus (1.29) (Supplemental table 5).

Discussion

The current study identifies the presence of piRNA and its associated machinery in somatic tissues from the three germ layers – brain and hippocampus (ectoderm), heart and kidney (mesoderm), and liver (endoderm) – in addition to the germline. The results indicate that PIWIL1, 2, and 4 mRNA are highly expressed in testis, while detectable but lower mRNA levels were identified in somatic tissues. PIWI-interacting machinery have long been considered components exclusively expressed from the germline [16–18], but low levels of PIWIL1 mRNA expression in the brain and hippocampus, PIWIL2 mRNA expression in the heart, brain, and hippocampus, and PIWIL4 mRNA expression in the kidney, heart, brain, and hippocampus indicate that PIWI-interacting machinery are also transcribed in the soma in adult mice (Figure 1). It is important to note the significance of using RNA with high RNA integrity scores for mRNA analysis, as the high cDNA quality is required for detecting low abundant PIWIL transcripts (Table 1). These results are consistent with recent studies indicating that PIWIL1, 2, and 4 are expressed and functional in mouse embryonic fibroblasts and stem cells (MESCs), human embryonic stem cells (HESc), and adult human somatic tissues [25,26]. Human PIWIL transcription has also been associated with cancer and may induce age-related regulatory activity [27,28]. Since PIWI protein expression of PIWIL1, PIWIL2, and PIWIL4 in mouse testis is tightly regulated during late embryogenesis and early postnatal development at a stage-specific manner, a similar protein expression may be responsible for somatic tissues [29]. Nevertheless, future studies of PIWIL expression in somatic tissues during development are necessary to fully understand PIWI protein regulation in the soma.

Germline piRNAs are highly expressed and have been extensively studied for several decades [1,10,16]. A mature (germline) piRNA contains several characteristics: 24-32 bp length, preference for a uridine at the 5ʹ end, preference for an adenine at the 10th position, and 2ʹ-O-methylation at the 3ʹ end (Figure 3 (a)) [1]. The 2ʹ-O-methylation signature is also present in tRNA, rRNA, and snRNA species, and is rarely found on mRNA [30]. We have eliminated the possibility of rRNA contamination through small RNA size selection prior to periodate treatment. Sodium periodate treatment prevents base extension during sequencing and is selective for piRNA, which are the predominant small RNA species that exhibit this modification. As sodium periodate treatment does not exclusively select piRNAs, we filtered known small RNA species during analysis to ensure high selectivity of piRNAs (Figure 2). Our method of piRNA detection is different from recent reports of somatic piRNA expression as we have included sodium periodate treatment prior to sequencing small RNAs [31,32]. Small RNA selection during small RNA library preparation (post periodate treatment) has also eliminated the possibility of tRNA and snRNA contamination, increasing the selectivity of piRNA. We are able to validate our method of piRNA detection since 40,581 of the 40,805 periodate-treated testis piRNAs were collectively found in the four publicly available piRNA databases (Supplemental Figure 1). These databases consist of piRNAs that were obtained through experimental methods such as chromatography, PIWI protein crosslinking-immunoprecipitation, small RNA sequencing (including size selection without sodium periodate treatment), and immunoprecipitation of PIWI or PIWI-associated proteins [6–9]. Interestingly, these piRNAs are classified primarily on length and association to PIWI interacting proteins. It is important to note that none of the piRNA databases employed sodium periodate treatment for analysis, likely overestimating the number of mature piRNAs and increasing the introduction of false positive results (e.g., miRNAs and siRNAs) [6–9].

We were successful in detecting somatic piRNAs from the cortex (1,963 peaks), hippocampus (5,494 peaks), kidney (580 peaks), and liver (406 peaks), with 55%, 57%, 55%, and 54%, of the tissues representing characteristics of piRNAs, respectively (Table 2). Our results are also consistent with the germline piRNA observations as testis showed the highest frequency distribution of piRNAs at 24-32bp, with 56% (22,734) of the total piRNA transcripts portraying all five characteristics of piRNA (Table 2 and Figure 3). Interestingly, several of these piRNA characteristics were detected from the soma, strongly suggesting that somatic piRNAs share structural similarities with the germline (Figure 3). This is supported by several reports indicating that somatic piRNAs prefer a uridine at the 5ʹ end and an adenine at the 10th position [11,12,20,21]. We report that somatic piRNAs are shorter in length when compared to the germline based on the higher levels of somatic piRNA-like peak differences detected from 24bp to 20bp length limitations, while presenting all other piRNA characteristics (Table 2). This is consistent with our results from surveying short piRNA (20-25bp) clusters, where somatic tissues on average contained 1.18–2 short piRNAs per cluster, while testis only contained 0.19 short piRNAs per cluster (Supplemental table 5). Although rare, there are a handful of instances where other groups have suggested that certain piRNAs are smaller in length in mice [11]. Thus, the current criterion for piRNA length should be revised to include both germline and somatic piRNA. Furthermore, somatic piRNAs may be functionally different from the germline due to this newly found characteristic.

One of the main functions of piRNAs in the germline is transposon silencing via induced DNA methylation [1]. Therefore, we expected testis to have an abundance of piRNAs that mapped to multiple genomic locations. Indeed, upon surveying the piRNA transcript locations and piRNA sequence types, testis showed the least amount of alignment with the genome at 66.9%, with 11,824 unique and 22,625 multi-mapped piRNA transcripts based on position/location, as expected (Table 3 and Supplemental Table 1). The high read counts observed in somatic tissues indicate confidence that data are unlikely to be false positives (Table 3). The kidney in particular indicated the lowest sequencing depth, while showing high median read counts per peak (per million mapped reads) among all tested tissues, suggesting that the kidney has very high piRNA expression despite its low abundance (Table 3). Furthermore, the cortex and kidney displayed the highest fraction of multi-mapped piRNA transcripts among the somatic tissues, suggesting that these tissues most likely share similar transposon-silencing functions with testis (Supplemental Table 1). Interestingly, cortex was devoid of uniquely mapped piRNAs, which may be due to poor detection, sequencing depth, and sample variability. On the other hand, the liver displayed the highest fraction of uniquely mapped piRNA transcripts, implying that there may be another function for somatic piRNAs (Supplemental Table 1). According to RepeatMasker analysis, 79% of the masked, multi-mapped piRNA transcripts for testis aligned with LTRs, which is consistent with previous studies (Figure 4) [33,34]. This includes 68% of Class II ERV repeats (4% of the mouse genome), which correspond to IAP transposons [35]. Moreover, testis also contained the least amount of small RNAs that overlapped with multi-mapped piRNA transcripts, which is consistent with previous observations [6–9]. In contrast, over 82% of all the multi-mapped somatic piRNA transcripts revealed a high number of small RNAs with the exception of hippocampus (Figure 4). Remarkably, multi-mapped piRNA transcripts from the hippocampus overlapped to multiple repeat classes including LTRs (class II ERVs), small RNAs, SINEs, and LINE elements, and hippocampus was the only somatic tissue that shared similar RepeatMasker components with testis (Figure 4). Thus, additional research is necessary to explore the importance in function for multi-mapped piRNAs in the soma, which may be different from the germline.

We also surveyed the tissue-specificity of piRNA sequences and transcript locations for soma versus the germline (Figure 5–6). Although the cortex and hippocampus are two different areas of the brain, it is important to note that hippocampus alone showed the highest number of piRNA sequences that were tissue-specific (Figure 5). This suggests that piRNAs are highly tissue-specific, which may result in piRNA functions that are catered towards a specific tissue or cell-type [11,20]. Although a few groups have uncovered piRNA expression in certain somatic cells/tissues, none have systematically characterized somatic tissues from the three germ layers [3]. Of note, 35% of the somatic piRNAs collectively overlap with the X chromosome, suggesting that the somatic piRNAs may be functionally important for the roles associated with the X chromosome in the surveyed animals (Supplemental Table 4) [36]. However, we uncovered 26 piRNA sequences and 40 of the piRNA transcript locations specific to the cortex, hippocampus, kidney, and liver (Supplemental Tables 3 and 4), providing strong evidence for the presence of piRNA in the soma, which likely result in tissue- and cell-type specific function. The stability and consistency of identified piRNA from the study is validated through an RT-PCR-based experiment that confirms its authenticity (Supplemental figure 2).

By comparing periodate-treated somatic piRNAs with those in publicly available databases, we were able to determine the extent of tissue-specificity relative to the germline (Figure 6). When taking a closer look at each individual database, piRBase (10,211,139) and piRNAQuest (3,646,281) contained the most piRNA sequences with a mere 2,591 sequences shared among all four databases. This suggests that the databases likely contain non-specific sequences (Figure 6 (a)). Since the public databases contained mouse brain tissues in addition to the germline tissues, we expected a high overlap with cortex and hippocampus piRNAs from our study [6–9]. Concordantly, the cortex indicated an 82.9% overlap with piRNAs found in at least one database, while the hippocampus showed approximately 60.6% sequence similarity (Figure 6(b)). Much to our surprise, we found that 90% of piRNA from the kidney overlap with at least one other database, and 88.2% overlap for liver, while both tissues only contained a handful of piRNA sequences that were common to all four databases (Supplemental figure 4).

Hippocampus piRNA distribution was particularly interesting as it showed the highest piRNA expression among all somatic tissues, and is the tissue that is most similar to testis based on frequency distribution and high repetitive sequence content (Figure 3–4). This suggests that a subset of piRNAs from hippocampus may pose similar functions as in testis [37]. Hippocampus also shares similarities with other somatic tissues based on its short piRNA length, and its overlap with other somatic tissues, suggesting its functional relevance to other somatic tissues as well (Figure 5). Thus, somatic and germline piRNAs may portray functions that are cell- and tissue-specific and independent in nature. We hypothesize that the hippocampus most likely consists of two types of piRNAs that correspond to at least two different functions.

Conclusions

Overall, our findings support the presence of piRNAs and PIWI-interacting machinery in the soma. Since germline PIWIL mRNA is highly regulated during development, PIWIL expression in the soma may also be regulated based on developmental stage and sex [29]. Hence, further research is required to understand the somatic piRNA expression and function with respect to sex, developmental stage, and cell-type specificity. The shorter piRNA lengths detected in somatic tissues along with the expected 5ʹ uridine signatures, 2ʹ-O-methylation signatures at the 3ʹends, and adenine signatures at the 10th position, indicates an unknown PIWI-interacting mechanism that is likely specific to somatic tissues during synthesis, using the ping-pong cycle. More information and future research is necessary, however, to elucidate the role of smaller length piRNAs. Together, these observations suggest that piRNAs are readily available in the soma and can be classified into several categories based on its specificity: 1) piRNAs common to the germline and soma; 2) piRNAs exclusive to the soma; and 3) piRNAs exclusive to each tissue type. These piRNA categories may correspond to one or more functions respective to each tissue, in addition to known transposon silencing functions similar to the germline. Further investigation into the role of piRNA in association with epigenetic marks including DNA methylation and histone modifications in the soma is necessary. Thus, the comprehensive somatic tissue piRNA findings from the current study are likely to influence future scientific inquiry into the use of the PIWI/piRNA system as a potential epigenome editing technology for novel therapeutics.

Materials and methods

Mouse breeding experiments

The current study used a mouse colony carrying the agouti viable yellow (A^vy) epigenetic allele and the non-agouti (a) allele maintained through forced heterozygosity for more than 250 generations [38]. The animals used in this study are derived from the Agouti colony, but are wild-type non-agouti, with 93% genetic identity to the C57BL/6J background [39]. All mice were housed at the ULAM (Unit for Laboratory Animal Medicine) of University of Michigan and at RAR (Research Animal Resources) of University of Minnesota on regular 12–12 dark-light cycles at 70–73°F and 50% humidity. Animals housed at the University of Michigan were given Mouse Diet 5 LOD for maintenance and Mouse Diet 5008 for breeding, and animals at the University of Minnesota were given the Mouse Diet 2018 (Envigo Tekled) for maintenance and Mouse Diet 2019 for breeding. All the mice were euthanized by CO₂ asphyxiation followed by a bilateral pneumothorax procedure as a secondary form of euthanasia in accordance with the rules and regulations set forth by the IACUC. The following tissues were collected for RNA analysis and stored in RNAlater (Thermo Fisher) from adult male and female animals that were 2 months of age: whole brain, hippocampus (brain), heart, liver, kidney, and testis/ovary tissues for a total of six samples per animal. Tissues from two, 2-month-old adult males that were used for piRNA-seq analysis included those from tissues representative of the three germ layers: brain and hippocampus (ectoderm), heart and kidney (mesoderm), and liver (endoderm). All of the mouse experiments were performed in accordance with National Institutes of Health guidelines for care and use of animals and also approved by the University of Michigan and the University of Minnesota Institutional Animal Care and Use Committee (IACUC), protocols PRO00005953 and 1510-33114A, respectively.

RNA isolation, qRT-PCR, and RT-PCR analysis

The commercially available RNeasy kit (QIAGEN, cat: 74,104) was used for RNA isolation according to the manufacturer’s protocol. Since RNA treated with DNAse I yielded results similar to untreated RNA, DNAse I was not used in the current experiments. We are confident that our experiments contain no genomic DNA contamination as the control primers span both introns and exons, serving to measure any DNA contamination. Additionally, RNA was treated with genomic DNA wipe out buffer before reverse transcription to remove any possibility of DNA contamination. RNA concentration was initially measured using a NanoPhotometer N50 system, and the RNA quality and integrity (RIN score) calculated by a QIAxcel Advanced Instrument and summarized in Table 1.

The total RNA of samples exhibiting a RIN score >7 (whole brain, hippocampus, heart, kidney, liver, testes, ovary) were then reverse-transcribed to cDNA using QuantiTech Reverse Transcription Kit (cat: #205,311). The resulting cDNA was used to quantify PIWIL mRNA expression in somatic mouse tissues (brain, hippocampus, heart, liver, kidney) and germline tissues (ovary, testis), as measured by qPCR on the Applied Biosystems StepOne Real-Time PCR system using Taqman (Applied Biosystems). Representative cDNA samples were assayed in duplicate against multiple probe sets (Taqman, cat # 4,369,016) for each of the PIWIL genes (PIWIL1, PIWIL2, PIWIL4), housekeeping genes, purchased from ThermoFisher (Mm02619580_g1 for β-actin and Mm02524711_g1 for RPL41), and blank controls. The primer efficiency was further confirmed by creating a serial dilution of cDNA for subsequent PCR experiments. The qPCR information and amplicon sizes are reported in Supplemental table 6. All qRT-PCR reactions were carried out for 40 cycles under standard PCR conditions. The ΔCt value was initially calculated by subtracting Ct value of a testing replicate of a given gene from the average Ct value of the internal control (RPL41 or β-actin). The fold difference for each replicate was then calculated by raising the ΔΔCt value as a power of 2 [40].The average and standard deviation for each sample was calculated using the data from triplicates. Each qRT-PCR result was confirmed using at least two independent experiments for a total of three males and three females. The information regarding individual primer sequences is available in Supplemental Table 6.

An RT-PCR-based assay was used to confirm piRNA expression based on sequence (Supplemental Figure 2). Small RNA was isolated from testis, cortex, and liver tissues from two biological replicates using a small RNA isolation kit (QIAGEN, cat: 74,104). The BioRad iScript select cDNA Synthesis kit (cat: #1,708,896) was used to reverse transcribe cDNA from 200 ng of small RNA for each sample, using gene-specific primers for a piRNA based on sequence (Supplemental table 6). The gene-specific primer includes an adapter sequence during reverse transcription, which is later used for PCR amplification [41]. In preparation for RT-PCR, 2 uL of cDNA samples were mixed with 10 µM forward/reverse primers, nuclease-free water, and Qiagen HotStarTaq master mix kit (Cat. #203,446). RT-PCR was then performed using a Bio-rad CFX96 C1000 Series Thermal Cycler. The following RT-PCR conditions were used for the reaction – 95°C for 10 min, [95°C for 15 s, 60°C for 1 min] x45. The resulting RT-PCR was visualized using the QIAxcel using 2uL of product (Supplemental figure 2). This RT-PCR experiment represents the average results obtained from three independent RT-PCR experiments.

Sodium periodate treatment and piRNA sequencing

Mouse piRNAs in adult testes have demonstrated resistance to sodium periodate treatment due to their unique 3ʹ modification – 2ʹ-O-methylation [12,23]. Thus, we treated small RNA samples from a pair of 2-month-old male littermates (matched biological replicates) with sodium periodate to identify somatic piRNA species in mouse. Cortex (brain), hippocampus (brain), liver, kidney, and testis tissues were used for small RNA extractions consisting <200 bp fractions (including piRNA) using the AllPrep (Qiagen) and RNeasy kits (Qiagen), respectively. The AllPrep kit selects for total RNA, which includes RNAs that are >200 bp, while the RNeasy kit was used to further select for small RNAs that are <200 bp. RNA quality check was performed using the QIAxcel to determine sample size prior to sodium periodate treatment.

All samples were put into two categories based on the treatment: group A served as the control group (sodium periodate untreated), while group B was subject to sodium periodate treatment to eliminate the small RNA fragments that do not contain a 3ʹ modification (2ʹ O-methylation) specific for piRNA. Both groups A (periodate untreated) and B (periodate untreated) were prepared under similar conditions. For group B, 27 μL of the small RNA (a total of 400 ng of RNA) fractions were treated with 5 μL of freshly made sodium (meta) periodate (Sigma Aldrich, lot: #BCBS5360V) and 8 μL of 5X borate buffer generated with 150 mM borax (Alfa Aesar, lot: #T29C533) and 150 mM boric acid (Fluka Analytics, lot: #SZBG1380V) at pH 8.6, generating a total of 40 μL reactions. Three such reactions were made for each sample tissue. For group A, one set of reactions was made substituting 5 μL of RNase free water for sodium periodate. Both groups were then incubated at room temperature for 30 min, followed by ethanol precipitation. The resulting RNA was dissolved with RNase free water before library preparation. The NEB small RNA kit was used for library preparation, where the RNA amplification step selected for piRNA containing the 2ʹ O-methylation. All of the samples were size selected, analysed through Bioanalyser, multiplexed 10 per lane, and sequenced on an Illumina v4 HiSeq2500 with 50bp single-end reads.

Identification of piRNA transcripts and computational analysis

Raw sequencing reads were first checked for quality using FastQC and MultiQC, and then processed by adapter trimming, size selection (10–45 bp), and quality trimming (phred quality score >10) using cutadapt v1.15 software. The processed reads were then aligned to the mm10 mouse genome assembly by Bowtie 2 version 2.3.2 with -end-to-end align mode without mismatch, and at most ten distinct alignments were reported [42]. Aligned reads that overlapped with exons, tRNAs, or non-coding RNAs including rRNAs, snRNAs, and miRNAs annotated in the GENCODE project (version M14), were filtered out to exclude any non-piRNA fragments that may have escaped the sodium periodate treatment [30]. The filtered alignments from sodium periodate treated and control samples were then compared pairwise for each tissue type to identify regions overrepresented in periodate-treated RNA reads. This step was performed using the peak-calling software PePr (version 1.1.21 with parameters – shiftsize 0 – windowsize 20 – threshold 1E-3 – peaktype sharp) to ensure that we only identify regions significantly more resistant to the sodium periodate [43]. Peaks shorter than 20bp were excluded in subsequent analyses in order to avoid transcription noise. In addition to 20bp, we also assessed 22bp and 24bp as the minimum size selection. Finally, the identified peaks were defined as piRNA-like transcripts. The strand of piRNA-like transcript was defined by the strand having the majority of aligned reads; for 97.6% piRNA-like transcripts in all tissues, greater than 95% of reads aligned to the same strand. A brief summary of the computational pipeline can be found in Figure 2. We also applied the pipeline to uniquely mapped and multi-mapped alignments separately (i.e., based on whether a read aligned uniquely or to multiple genomic regions) and defined these transcripts as uniquely mapped piRNA and multi-mapped piRNA, respectively.

Identification of piRNA expression and motif signature

To characterize the nucleotide composition of piRNA-like transcripts, we generated motif logos for the first 20 nucleotides of transcripts using WebLogo web server version 2.8.2 (http://weblogo.berkeley.edu/logo.cgi). We searched the ±3bp flanking regions of the 5ʹ end of the piRNA-like transcripts to determine the exact position of the nucleotide with the 5ʹ uracil. If no uracil was present, we used the peak edge as the start position. We then calculated the proportion of adenine in the 10^th position downstream from the 5ʹ end. The average read-count in treated samples per piRNA transcript was calculated using the multicov function in BEDTools and then converted into reads per million mapped reads for the expression level of the piRNA transcript [44].

Comparisons across tissues and with available piRNA databases

Identified piRNA transcripts in this study were compared across tissues and also with public piRNA databases. The first comparisons were based on their genomic locations, i.e., piRNA transcripts having at least 1 bp overlap were grouped together. Custom Perl scripts using merge and intersect functions in BEDTools (v2.27.0) [44] were developed to conduct comparisons and to generate Venn diagrams. Multiple piRNA transcripts overlapping with a single piRNA transcript from another tissue were counted only once in a Venn diagram. For example, 40,808 piRNA transcripts in testis were merged to 40,805 locations in the Venn diagram. We also conducted comparisons based on sequence similarity. The CD-HIT software [45] was utilized to cluster sequences of piRNA transcripts with a sequence similarity threshold of 99%. Majority of the piRNAs resulted in requiring 100% identity. The piRNA transcripts in the same sequence cluster were treated as identical. In addition, we defined unique piRNA sequences as the nucleotide arrangement shared by all sequences in the cluster. Four public piRNA databases were queried to evaluate piRNA transcripts: piRBase database, piRNAQuest, piRNABank, and piRNA cluster database [6–9]. Mouse piRNA annotations from the four databases were downloaded and converted into mm10 by the UCSC LiftOver tool (http://genome.ucsc.edu/cgi-bin/hgLiftOver) as needed. We collected all the mouse annotations and did not specify particular tissue, gender, or age. Most of these piRNA annotations were based on germline samples (i.e., testes, oocytes, and spermatids) or without tissue information. Only 20 annotations in the piRBase database were from hippocampus samples. For the piRNA cluster database, we downloaded pooled/generic testis piRNA clusters annotation. Duplicated or overlapped (with at least 1bp) annotations in a database were integrated as one piRNA using the merge function in BEDTools. After processing, there were 10,211,139 piRNAs from piRBase database, 659,639 piRNAs from piRNABank database, 29,349 piRNAs from piRNA cluster database, and 3,646,281 piRNAs from piRNAQuest database.

In addition, piRNAs identified from this study was mapped to pre-defined piRNA clusters downloaded from the piRNA cluster database. An R package, GenomicRanges [46] (version 1.34.0), was used to find overlaps between the newly found piRNAs and piRNA clusters from the database. The results are listed in Supplemental Table 5. Furthermore, piRNA identified from this study were clustered to examine how many piRNAs are clustered together in somatic tissues compared to testis. A piRNA cluster was defined based on a genomic location that contains at least two piRNAs, where the distance between each piRNAs within a cluster is less than or equal to 200 bp. The percentage of clustered piRNAs in the total identified piRNAs of each tissue was calculated. Since somatic piRNAs are shorter than testis piRNAs, the percentage of clustered short piRNAs (20–25 nt) in total identified short piRNAs of each tissue was also calculated. Same calculations were performed to identify highly expressed piRNAs of each tissue as well. Details and results are listed in Supplemental Table 5.

Funding Statement

This work was supported by the National Institute of Health (NIH) Director’s Transformative Award program and the National Institute of Environmental Health Sciences (NIEHS) [ES026877], with additional support from the Michigan Lifestage Environmental Exposures and Disease (M-LEEaD) NIEHS Core Center [P30 ES017885] and a NIEHS Pathways to Independence Award to CF [R00 ES022221];National Institute of Environmental Health Sciences [ES017885];National Institute of Environmental Health Sciences [ES026877];National Institute of Environmental Health Sciences [ES022221];

Acknowledgments

We thank the University of Michigan DNA Sequencing Core for their assistance with the library prep and sequencing the samples generated by the study, and the Zamore lab from the University of Massachusetts Medical School, for sharing the sodium periodate protocol with us.

Disclosure statement

No potential conflict of interest was reported by the authors.

Authors’ contributions

CF, DD, MS, BP, JG, and ZT participated in the study design and wrote the manuscript; BP, MC, JG, and TJ carried out the molecular studies and interpreted results; ZT and MS performed the bioinformatics analysis, data interpretation, and methodology; MS and CF supervised bioinformatics analysis and interpreted the results; BP, DD, and CF participated in the coordination of the study. BP and ZT drafted the manuscript with input from DD, CF, MS, JG, MC, TJ; all authors read and approved the final manuscript.

Ethics declaration

All animal work was performed in accordance with National Institutes of Health guidelines for care and use of animals and also approved by the University of Michigan and the University of Minnesota Institutional Animal Care and Use Committee (IACUC), protocols PRO00005953 and 1510-33114A, respectively.

Data availability

The authors declare that all data supporting the findings of this study are available within the paper and its supplementary information files. Additional information is available from the corresponding authors upon reasonable request.

Supplementary material

The supplementary data for this article can be accessed here.