mir-21 is associated with inactive low molecular weight Argonaute complexes in thyroid cancer cell lines

Thyroid cancer is the most prevalent endocrine malignancy. We and others have shown that several microRNAs, which are post-transcriptional gene regulators, are aberrantly expressed in anaplastic thyroid cancer (ATC) and papillary thyroid cancer (PTC) tissues, as well as cell lines derived from these cancers. In the cell, miRNAs are bound to Argonaute (AGO) proteins as what could be termed low molecular weight RNA-Induced Silencing Complexes (LMW-RISCs) that can assemble with additional proteins, mRNA, and translation machinery into high molecular weight RISCs (HMW-RISCs) that exert regulatory function. In this study, we sought to analyze the association of miRNAs with RISC complexes in ATC and PTC. For ATC and PTC lines, miRNA species were enriched in both HMW-RISC and LMW-RISC cellular fractions, compared with intermediate molecular weight fractions and very low molecular weight (AGO-poor) fractions. Furthermore, 60% of all miRNAs were slightly more abundant in LMW-RISC versus HMW-RISC fractions by ~2-4 fold. Surprisingly, miR-21-5p, one of the most abundant miRNAs in both ATC and PTC lines and one of the most widely studied oncogenic miRNAs in many solid tumors, was consistently one the least abundant miRNAs in HMW-RISC and the most enriched miRNA in LMW-RISC fractions. These findings may suggest that miR-21 has a role or roles distinct from canonical post-transcriptional regulation in cancer. Furthermore, the methodology described here is a useful way to assess the distribution of miR-21 between HMW and LMW-RISCs and may help to reveal the true roles of this miRNA in thyroid cancer development, progression, and treatment.


INTRODUCTION
Thyroid cancer is the most prevalent endocrine malignancy in the United States [1], [2]. Welldifferentiated thyroid cancers (WDTC) such as papillary thyroid cancer (PTC) are the most common and typically have an excellent prognosis [3]. In contrast, undifferentiated tumors such as anaplastic thyroid cancer (ATC), are rare and almost invariably fatal [4], [5]. ATC can arise de novo, or more commonly, via dedifferentiation of WDTCs. Nevertheless, the precise mechanism whereby WDTC transitions to ATC is unknown [6] [7].
In cancers, including thyroid cancers, aberrant miRNA expression is commonly presumed to contribute to alterations in mRNA repression, which may promote pathogenesis [23] [24] [25] [26]. That is, miRNA and target RNA expression are presumed to be negatively correlated. However, this paradigm has been challenged in recent years [27]. Prominently, La Rocca et al. [28] reported that, in various adult mouse tissues in vivo and resting T-cells in vitro, and regardless of expression level, the vast majority of miRNAs are associated with AGO only. With few exceptions, just a small portion of AGO-bound miRNAs assembled into mRNAassociated RISCs [28]. These investigators also showed that RISC tended to be in a more active state in transformed cell lines, especially in the exponential growth phase and in mitogenstimulated T-cells [28]. Because the activity state of RISC may be context-dependent, and miRNA activity could thus be uncoupled from overall abundance, we reasoned that a closer examination of active versus inactive miRNA species in thyroid cancer might provide valuable insights into pathogenesis.
In this study, we therefore separated cellular AGO-miRNA containing complexes using size exclusion chromatography (SEC) and profiled active and inactive AGO-miRNA in SW1736 (ATC) and TPC-1 (PTC) cell lines [28]. Active complexes, which comprise miRNA, AGO, GW182, and mRNA, have a molecular weight that is normally > 2MDa [28]. Inactive complexes, with only AGO and miRNA, have a lower mass ~150kDa [28]. Hence, active and inactive complexes are referred to as high molecular weight RISC (HMW-RISC) and low molecular weight RISC (LMW-RISC), respectively ( Figure.1).

Figure. 1 Illustration of RISC separation.
Cell lysate is separated over a Superose 6 column. High molecular weight complexes including high molecular weight RISC (HMW-RISC) elute in early fractions, while low molecular weight fractions including Argonaute/miRNA complexes (LMW-RISC) elute later.

PBMC Isolation
Peripheral blood mononuclear cells (PBMCs) were isolated from fresh blood obtained from a human donor under a university IRB-approved protocol (JHU IRB #CR00011400). Up to 180mL of blood was collected into 3 60 mL syringes pre-loaded with 6 mL of anticoagulant acid citrate dextrose (ACD) (Sigma C3821 Lot#SLBW6172). Within 15 min of collection, whole blood was diluted 1:1 with 1X PBS containing 2% FBS, layered (appr. 35 mL per tube) onto 15 mL Ficoll-Paque PLUS (GE Healthcare 17-1440-02) using SepMate-50 tubes (Stem Cell Technologies 85450), and centrifuged for 15 min at 1200 x g. The plasma and buffy coat fractions were transferred to 50 mL conical tubes, diluted 1:1 with PBS, 2% FBS, and centrifuged for 8 min at 300 x g. The supernatant was decanted, and the cell pellet was suspended in red blood cell lysis buffer (155mM NH4Cl, 12mM KHCO3, 1mM EDTA, pH 7.3) and incubated for 10 min at 37°C. The suspension was diluted 1:1 with HBSS and centrifuged at 400 x g for 10 min. Cells were washed with PBS at 400 x g for 10 min. The cell pellet was snap frozen on dry ice and stored at -80°C.

AGO2-RNA Immunoprecipitation
Unfractionated input and pooled HMW-RISC, MMW-RISC, LMW-RISC, and RISC-Poor SEC fractions were incubated with 48µg of anti-AGO2 antibody (Sigma SAB4200085 Lot#087M4841V) for 16 hrs at 4°C with rotation. The mixture was then incubated with 18mg Protein G Dynabeads (Thermo Fisher 10003D) for 2 hrs at room temperature with rotation. Beads were washed 3X with ice-cold Sup6-150 and once with PBST, with beads retained by magnet at the end of each wash step. Bound protein was eluted with 240µL 50 mM glycine, pH 2.8. One volume of 1M Tris, pH 7.5 was added to the eluate. AGO2 was detected by Western blot and miR-21-5p by qPCR, using half of each sample for each assay.

Small Library Preparation and Sequencing
Unfractionated input, HMW-RISC, MMW-RISC, LMW-RISC, and RISC-Poor small sequencing libraries were constructed with CATS RNA-seq kit (Diagenode C05010041 Lot#4). Library primer dimers were removed by AMPureXP beads (Beckman). Quality control was assessed by Bioanalyzer high sensitivity assay (Agilent). Libraries of 160-180 bp were size-selected by bluepippin (Sage Science). Libraries were spiked with 20% PhiX Control v3 (illumina 1501766) and sequenced using Next Seq with 1 X 50 Mid output of 130M reads by the Johns Hopkins Microarray and Deep Sequencing Core. Sequencing quality control was performed by FASTQC (Babraham Bioinformatics) followed by adapter and polyA trimming with Cutadapt 1.17. Reads were sequentially aligned to reference transcriptomes rRNA, tRNA, RN7S, snRNA, snoRNA, scaRNA, VT-RNA and Y-RNA. All reads that did not map to the aforementioned RNAs were aligned to pre-miRNA, protein-coding mRNA, and long non-coding RNAs (lncRNAs). Species with low read counts were filtered out such that average number of reads for a gene in all samples were >=5. For each fraction, all reads that did not map to the human transcriptome were aligned to the hg38 genome reference (rest_hg38). Hg38_reads were used to calculate log2 [counts-per-million+1] for each gene. The log2[cpm+] difference between different conditions was used as a final metric to determine relative abundance. Sequencing data were deposited with the Gene Expression Omnibus (GEO; database accession number GSE146015). See supplementary section for bash scripts.

RISC and Associated Proteins Across Molecular Weight Fractions
SEC was performed with protein lysates from cell lines SW1736 (ATC) and TPC-1 (PTC) to separate active and inactive RISCs by molecular weight (Figure.1). Enrichment of AGO2, a key component of RISC, as assessed by Western blot (WB), was used to define the SEC elution profiles of active and inactive complexes [28]. For both cell lines, AGO2 was detected in elution fractions 15-37, but was highly enriched in high molecular weight (HMW) fractions 15&16 (>2MDa) and low molecular weight (LMW) fractions 27-35 (~150kDa-100kDa) ( Figure.2A). Additionally, presence of fully-assembled, presumably functional RISC was further determined by enrichment of GW182, PABP, and RPL26. As previously demonstrated by La Rocca et al., each RISC-associated protein was present chiefly in HMW fractions 15-16 ( Figure.

Representative Small RNA Distribution Across Molecular Weight Fractions
miR-16, an abundant miRNA in most cell types and often used as a normalization control, was evaluated in each SW1736 and TPC-1 SEC fraction by RT-qPCR. miR-16 was detected in all fractions but was present predominantly in high molecular weight fractions 15 and 16 (>2MDa) and low molecular weight fractions 27-35 (~150kDa), respectively ( Figure.2B). This distribution closely resembled the elution profile of AGO2. In contrast, U6 small nuclear RNA (U6 snRNA), which is not known to complex specifically with RISC proteins, was enriched chiefly in SEC fractions 20-30, rather than in the AGO/miRNA-rich fractions (Supplementary Figure.2). This result suggests that miR-16 elution across molecular weight fractions mirrors that of AGO and may be AGO-dependent.

HMW-RISC Fractions are rRNA-and mRNA-Enriched
miRNAs associated with HMW-RISC and LMW-RISC fractions in SW1736 and TPC-1 cells were next profiled more broadly by sequencing pooled HMW and LMW AGO2/miR-16-rich fractions. Additionally, fractions 17-26, which are presumably "intermediate" or "medium" RISCs (MMW-RISC), and fractions 36-47, which have very low levels of AGO2 and miR-16 (RISC-poor), were also pooled and sequenced. Ligation-independent small RNA sequencing, which enables analysis of RNA fragments of all classes regardless of end modifications, was performed using each of the pools. Relative to other RNA classes, pre-miRNA and mature miRNA comprised <1% of total RNA in each fraction, in agreement with a previous report [32]. Ribosomal RNA (rRNA), which was well represented in each fraction, had the highest mapping percentage in HMW-RISC and MMW-RISC fractions ( Figure.3). This is consistent with the association of active RISC with polysomes. We corroborated this by analyzing ribosome marker (RPL26) enrichment in HMW-RISC and MMW-RISC fractions (Supplementary Figure.1) [28]. HMW-RISC and MMW-RISC fractions, as expected, also showed significant mapping to protein coding mRNA (Figure.3). The mRNA transcripts present in LMW-RISC and RISC-poor fractions were also shorter than fragments detected in HMW-RISC and MMW-RISC. For instance, mRNA transcripts such as CTC1, TTC39B and PDPR had a much lower mapping coverage in LMW-RISC fractions compared to transcripts associated with the heavier RISCs (Supplementary Figure.3-5). In contrast, tRNA fragments (tRNAFs), which may be associated with AGO, were highly enriched in LMW-RISC and RISC-poor fractions ( Figure.3) (Supplementary Figure.6) [33]. Taken together, the HMW-RISC fractions associate preferentially with rRNA and mRNA compared with LMW-RISC fractions.

SW1736 and TPC-1 miRNAs Exist Predominantly in Low Molecular Weight Fractions
As seen with miR-16 enrichment across the individual SEC fractions ( Figure.2B), SW1736 and TPC-1 miRNAs detected by sequencing were most abundant in the pooled HMW-RISC and LMW-RISC fractions when compared with MMW-RISC and RISC-poor fractions (Supplementary figure.7). Of 101 miRNAs that were abundantly expressed in both SW1736 and TPC-1 cells, approximately 60% of total miRNA species had greater enrichment in LMW-RISC fractions relative to HMW-RISC ( Figure.4 A). This included miRNA miR-21-5p, which was the most abundant miRNA in LMW-RISC fractions in each cell line [34] (Figure. 4 B).
The HMW-RISC and LMW-RISC distribution of miR-21 compared with other dysregulated ATC and PTC miRNAs, including miR-146b, miR-221 and miR-222, was also measured by qPCR analysis of pooled RISC fractions (Figure.5A). Of the four miRNAs, differential enrichment between SW1736 and TPC-1 was observed only for miR-21-5p, which was more enriched in LMW-RISC fractions by > 2-fold ( Figure.5A).
Additionally, the relative expression levels of these miRNAs between SW1736 and TPC-1 were compared by qPCR. Confirming the results of RNA-seq, qPCR showed that the profound enrichment of miR-21-5p in LMW-RISC cannot be explained simply by overall abundance of miR-21-5p compared with other miRNAs, as miR-21-5p is not the most abundant miRNA in these cell lines. Furthermore, when each miRNA was normalized to miR-16 or U6-snRNA, all differences between SW1736 and TPC-1 were < 2-fold ( Figure.5C). This implies that the relative distribution of a given miRNA may be cell type-specific and independent of expression levels.

Primary PBMCs Show Greater miR-21-5p Enrichment in LMW-RISC Than ATC or PTC Cell lines
Because miR-21 is one of the most extensively studied miRNAs and is reportedly overexpressed and oncogenic in various solid tumors, including thyroid cancer, the enrichment of this miRNA in inactive, LMW-RISC fractions was surprising [35]. Therefore, we assessed if the accumulation of this miRNA in mRNA-depleted fractions was ATC/PTC specific or also occurred in other cell types.
Analysis of mouse primary resting T-cell data from La Rocca et al. 2015, showed that miR-21a, which is comparable to miR-21 in humans, was present mostly in LMW-RISC fractions [28]. We observed a similar distribution of miR-21 in freshly obtained human primary PBMCs. Interestingly, however, when compared with the elution profile of miR-21-5p in SW1736, miR-21-5p LMW-RISC enrichment in human PBMCs is greater by 10-fold (Figure7. A-B). Therefore, the size of the miR-21-5p LMW-RISC reservoir may be dependent on the cell type and may actually be smaller in thyroid cancer cell lines than in some primary cell types. Figure. 7 miR-21-5p across SEC fractions of SW1736 cells and peripheral blood mononuclear cells. Abundance of miR-21-5p (qPCR) across SEC fractions 14-47 relative to fraction 14 for SW1736 cells (Panel A) and peripheral blood mononuclear cells (PBMCs, Panel B).

miR-21-5p Inhibitors Increase LMW-RISC Reservoir in SW1736
Hypothesizing that antisense inhibitors of miR-21-5p might compete with components of HMW-RISC complexes (including mRNA) and result in even greater enrichment of miR-21-5p in LMW-RISC fractions, we transfected SW1736 cells with miR-21-5p inhibitors [36]. Indeed, miR-21-5p inhibitors increased the abundance of miR-21-5p in LMW-RISC fractions and decreased HMW-RISC enrichment by greater than ten-fold ( Figure8.A-B). This result supports the conclusion that the large reservoir of miR-21-5p in LMW fractions is inactive and can be increased further by outcompeting the presumed HMW-RISC targets of miR-21-5p.

Decreased Expression of miR-21-5p in Hypoxia Lowers LMW-RISC Enrichment
In previous experiments, we observed a >2-fold decrease of miR-21-5p in SW1736 cells cultured at 2% oxygen (hypoxia) rather than in near-atmospheric levels of oxygen (unpublished results). Therefore, we investigated how lowering miR-21-5p expression levels by growth in hypoxic conditions might impact its distribution across HMW-RISC and LMW-RISC fractions. A similar distribution of miR-21-5p, which was largely enriched in LMW-RISC fractions, was observed in normoxic and hypoxic SW1736 lysates. However, the magnitude of enrichment was noticeably lower (by ~2-fold) in hypoxic SW1736. In contrast, there was only a slight decrease in HMW-RISC enrichment (<2-fold).

miR-21-5p HMW-RISC Enrichment Increases with Mimic Transfection
Furthermore, to see if large quantities of miR-21 could shift enrichment from LMW-RISC to HMW-RISC fractions, SW1736 cells were transfected with single-stranded or duplex miR-21-5p mimics. With the single-stranded mimics, an overall increase in miR-21-5p abundance was observed across all SEC fractions relative to the negative control. Interestingly, miR-21 remained substantially enriched in LMW-RISC, but was also now enriched in HMW-RISC fractions by >300-fold. In contrast, miR-21-5p duplex mimics were predominantly enriched in LMW-RISC fractions.

DISCUSSION
Recently, it has become evident that miRNA expression differences may not be perfectly correlated with canonical mRNA repression, in part due to the influence of intercellular signaling pathways on incorporation of miRNAs into active or inactive complexes [37] [38]. In this study, our results support the preferential enrichment of specific miRNAs into active, targeting RISCs or inactive small complexes. miR-21-5p is the most striking example. It is moderately abundant in the thyroid cell lines we studied, is minimally associated with target RISCs relative to other miRNAs, and is the most highly abundant miRNA in low molecular weight fractions that are not associated with mRNA.
MiR-21 is one of the most studied miRNAs [35][39] [40], with numerous validated targets. It is frequently deregulated in tumors, including thyroid, and is often described as an oncogenic miRNA or oncomiR [41] [42]. Numerous studies have demonstrated that aberrantly high levels of miR-21 promote tumor growth by negatively regulating the expression of tumor suppressor genes such as PTEN [43] [44]. Additionally, tumors may become reliant on miR-21 in a phenomenon known as oncogene addiction, and down-regulation of miR-21 often results in tumor regression [45] [46]. Effects of miR-21 have been demonstrated in both overexpression/knockout mouse models and in vitro transcription profiling studies [47] [46]. Because of the extensive evidence for regulatory roles of miR-21 in cancers and elsewhere, the high level of miR-21 within mRNA-poor fractions was unexpected. Indeed, our results would tend to suggest that most miR-21 copies are inactive in the cell types we examined. Because HMW-RISCs are associated with polysomes actively translating mRNA [48][49] [50], it could be that the large LMW-RISC miR-21 reservoir is due to a paucity of cognate miR-21 target transcripts within HMW-RISC fractions relative to those of other miRNAs.
We also predicted that miR-21-5p might utilize a mechanism of action distinct from other miRNAs: one that does not require direct binding to a target mRNA. However, because miR-21-5p inhibitors increased the magnitude of miR-21-5p within LMW-RISC fractions and lowered levels in target complex fractions, it is evident that miR-21-5p does interact with mRNA to some degree but mostly accumulates within LMW-RISCs. The large accumulation of LMW miR-21-5p might be explained by its known low target binding affinity. Relative to other miRNAs such as let-7a, miR-21 tenuously binds to targets due to the low GC% of its seed sequence [51] [52]. Hence, miR-21 may transiently interact with targets to mostly degrade mRNA through the recruitment of deadenylases or other factors, rather than remaining bound to a target and blocking the ribosome before returning to a low molecular weight state [53] [54]. This might explain the pronounced biological effects on tumor growth when miR-21 expression is modulated and why it mostly exists in LMW-RISC fractions.
Although mir-21-5p is relatively abundant in both cell lines examined here, we were curious to see how exogenous miR-21-5p would be distributed between HMW-RISC and LMW-RISCs in cells using single-stranded or duplex mimics. Interestingly, increased miR-21-5p enrichment was observed in both HMW-RISC and LMW-RISC fractions with each type of mimic; yet, the single-stranded miR-21 mimics largely accumulated in HMW-RISC fractions, and duplex mimics mostly accumulated in LMW-RISC fractions. We reasoned that the single stranded mimics may not efficiently load into Argonaute proteins and that both types of mimics are likely non-specifically bound to targets, as it is known that low binding affinity of miR-21 can be supplemented by the combination of 3'end supplemental base pairing and higher expression levels [51].
Taken together, our results suggest that miR-21 may be uniquely distributed in both cancer and normal cells, with molecular and biochemical features distinct from other miRNAs. Because miR-21-5p is a promising therapeutic target for cancer, further functional characterization of this miRNA in its molecular context is crucial. The methodology described here is a useful way to assess the relative magnitude by which miR-21 is associated with HMW and LMW-RISCs and should be combined with the analysis of post pharmacological inactivation of miR-21functional effects. Since our observations have been restricted to thyroid cell lines and PBMCs, we plan to extend our studies to normal, benign and malignant thyroid tissues in the future. Figure. 1 Thyroid cell line SEC elution profile. Western blot of SEC fractions 14-38. RPL26 and PABP enrichment in each fraction was analyzed to map HMW-RISCs (>2MDa). Actin complexes were also analyzed as a non-RISC associated control to confirm fractionation of lysate by SEC. GFP was spiked-in to each fraction post SEC and prior concentrating down for gel