Tumor samples were obtained from the Children’s Hospital of Philadelphia (CHOP) and Hospital for Sick Children, Toronto, Ontario, Canada. This study included archived Formalin-Fixed Paraffin-Embedded (FFPE) samples from 20 non-neoplastic thyroids, 8 adenomatous nodules and 60 sporadic well-differentiated follicular derived thyroid cancers (DTC): 47 PTCs and 13 FTCs. Demographic information including age and sex was collected from each patient in addition to histopathologic results and tumor staging. Ethics approvals for collection and use of the patient samples were obtained from the CHOP Institutional Review Board as part of the Child and Adolescent Thyroid Consortium (CATC) Biorepository study (IRB# 20-018240).

Tissue was cut into 5µm serial sections and mounted on glass microscope slides. Diagnosis was confirmed by a board-certified pathologist specialized in thyroid disease and regions of interest representing tumor foci on an H&E slide were selected and circled. Samples were sent to HTG Molecular Diagnostics in Tucson, AZ, and regions of interest were microdissected under a Leica LMD6500 laser capture microscope (LCM). Each sample was immediately suspended in EdgeSeq lysis buffer (HTG Molecular Diagnostics, Tucson, AZ) and miRNA/mRNA profiling was carried out using one FFPE tissue slide from each sample for each assay as previously described (25). miRNA and mRNA expression were profiled using HTG Edgeseq miRNA whole transcriptome assay and HTG EdgeSeq Oncology Biomarker Panel (OBP), respectively. HTG EdgeSeq system is a NGS application that measures gene expression without the need for extracting RNA. HTG Edgeseq miRNA whole transcriptome assay measures the expression of 2,083 human miRNAs described in the miRBase v20 database. HTG EdgeSeq OBP measures the expression of 2,559 genes associated with tumor biology, including 15 housekeeping genes. All miRNAs and mRNAs screened in this study are listed in Supplemental Tables 1 , 2 , respectively. Briefly, the workflow entailed automated, extraction-free sample preparation, quantitative nuclease protection using the EdgeSeq processor, and library generation and sequencing. The HTG EdgeSeq Parser (HTG Molecular, Tucson, AZ, USA) was used to align the FASTQ files to the probe list to collate the data. Data were provided as data tables of raw, quality control (QC) raw, counts per million (CPM), and median normalized counts.

The DESeq2 data normalization, analyses, and statistical comparisons between benign thyroid tissue and DICER1-mutant thyroid tumors were performed using the HTG EdgeSeq Reveal software version: 4.0.1. DESeq2 normalized data were logarithmically scaled for data visualization. Plots were created using GraphPad prism 9 (GraphPad software Inc., La Jolla, CA). TCGA RNA-Seq expression data for adult PTCs and non-neoplastic thyroid tissues were obtained from http://tcga-data.nci.nih.gov and http://gdac.broadinstitute.org/in September 2022. Functional enrichment analysis was performed using g:Profiler to search for genes significantly over-represented in the list of DEG, as compared to all background genes included in the HTG OBP panel (26). Predicted targets for different miRNAs were obtained from TargetScan Human release 8.0 (https://www.targetscan.org/vert_80/). Only predicted targets with a total context score <−0.2 were included in the analysis.

Differential expression was calculated using DESeq2 and only differentially expressed mRNAs and miRNAs with a false-discovery rate (FDR) < 0.05 were included. T-test analysis and One-Way ANOVA were performed with GraphPad prism 9. A two-sided p value < 0.05 was considered statistically significant.

In thyroid, DICER1-driven PTCs are infrequent in adult (0-0.4%) but comparatively more prevalent in pediatric PTCs and FTCs (5-10%) (15, 17, 20, 27, 28). Despite its overall low prevalence, we identified 12 cases of DTC with DICER1 hotspot mutations as part of the Child and Adolescent Thyroid Consortium (CATC) multi-institutional collaborative research program (nine from CHOP and three from Sick Kids). Detailed histopathologic features and molecular alterations of these 12 cases are summarized in Table 1 . Four of these 12 have been previously reported and published: cases ID# 6, 7, 11, and 12 (20, 29). The 12 cancer cases with mutations in the RNase IIIb domain of DICER1 included nine females (75%) with mean age of 13.7 years (SD = 2.0) at surgery. DTC harboring DICER1 mutations were all of the follicular type: four FTCs and six follicular variant PTCs (fvPTC). None had extrathyroidal extension, lymph node or distant metastasis. Loss of function defects were found in six cases (three loss of heterozygosity (LOH) and three inactivating variants). To determine whether DICER1 mutations were somatic or germline, DNA from non-neoplastic thyroid tissue from the same cases was analyzed by conventional Sanger sequencing and somatic alterations are highlighted in bold in Table 1 . All the RNase IIIb mutations tested and the three inactivating variants reported here (S125*, L777fs, G661Vfs*24) were found to be somatic by Sanger sequencing ( Supplemental Figure 1 ). In 3 cases, DICER1 RNase IIIb alteration was revealed due to the miRNA signature (further discussed in the next section), but no tissue was available to confirm the mutational status. In one case (case 9), DICER1 positivity was detected by a commercially available somatic thyroid oncogene panel but the specific codon was not reported.

FFPE tissue specimens were available for eight of 12 cases with DTCs harboring DICER1 RNase IIIb domain mutations. To determine the impact of DICER1 hotspot mutations on miRNA synthesis, we analyzed the differential miRNA expression in DICER1-mut DTC (six fvPTC and two FTC) vs non-neoplastic/adenomatous thyroids (n=26) using the HTG Edgeseq miRNA whole transcriptome assay which evaluates the expression of 2,083 human mature miRNAs. The DICER1 hotspot mutations in the RNase IIIb domain (E1705, D1709, D1810, and E1813) have been shown to impair the ability of DICER1 to generate mature 5p miRNAs as illustrated in Figure 1A (8). On par with its role in miRNA biogenesis, we show that tumors harboring these mutations demonstrate clear reduction of 5p miRNAs (regardless of the specific mutation site of the RNase IIIb domain), including those abundantly expressed in the non-neoplastic thyroid tissue and associated with tumor suppressor function in thyroid cancer such as let-7 and mir-30 families and mir-125b (30–32)( Figures 1B, C ). Previous mechanistic studies have shown that RNase IIIa and IIIb activities are distinct and can be uncoupled (2, 33). We found that levels of 3p miRNAs were markedly increased when compared to non-neoplastic thyroid tissue and benign hyperplastic lesions ( Figure 1B ). This striking decrease in 5p:3p strand ratios caused by the RNase IIIb hotspot mutations is further illustrated by the differential processing of miR-20a pre-miRNA in DICER1-mut tumors ( Figure 1D ). While benign thyroids and DICER1-wt DTC show predominance of miR-20a-5p, DICER1-mut DTC have a prominent reduction in 5p:3p strand ratios and express mostly the 3p strand (miR-20a-3p). Moreover, we report a patient with bilateral FTC, with the carcinoma on the right lobe harboring biallelic DICER1 alterations (p.G661Vfs*24, p.E1813K) while the carcinoma on the left lobe was DICER1-wt ( Figure 1E ). Predictably, the DICER1-mutant FTC showed reduced levels of 5p miRNAs and increased levels of 3p miRNAs compared to the DICER1-wt FTC ( Figure 1F ). miR-451a is the only known miRNA for which maturation has been shown to be DICER1-independent (34). Indeed, expression levels of miR-451a were similar between the two tumors, on par with the non-requirement for DICER1 for its processing, and emphasizing the causal effect of the DICER1 on the downstream miRNA changes.

We next sought to determine whether this unique miRNA signature of tumors with RNase IIIb mutations could be used to identify thyroid cancers harboring DICER1 hotspot mutations. Principal component analysis (PCA) revealed that a classifier with as few as four 3p-miRNA markers (miR-20a-3p, miR-30d-3p, miR-99b-3p, and miR-450a-1-3p) could efficiently distinguish DICER1-mut DTC (regardless of histology: PTC or FTC) from non-neoplastic/benign hyperplastic and DICER1-wt PTCs/FTCs ( Figure 2A ). Two cases of benign hyperplastic nodules (adenomatous follicular hyperplasia) harboring DICER1 variants, one case with biallelic DICER1 alterations (p.E1813G, p.T847delinsFHKHS) and one with a variant of unknown significance (p.Q7R) showed an interesting pattern of miRNA expression. Both cases showed decreased expression of 5p miRNAs, comparable to that observed in DICER1-mut DTCs ( Supplemental Figure 2 ). However, these DICER1-mut benign nodules had no increase of 3p miRNAs as observed for DICER1-mut DTCs ( Figure 2B ), indicating these 3p miRNAs are specifically upregulated in malignant cases with RNase IIIb mutation. One case of DTC harboring the DICER1 p.E1420del (VUS) mutation had no significant differences in miRNA levels. Therefore, 3p miRNAs are not only valuable to classify RNase IIIb mutants as benign or malignant but could also be causally involved in DICER1-driven malignant progression. The validity of these four miRNAs biomarkers was subsequently tested using the TCGA cohort comprising 496 adult PTC patients for which we know there are two cases with DICER1 hotspot mutations: p.E1813G and p.D1810H (17). The expression of these four miRNAs is also dramatically increased in two adult PTCs harboring DICER1 RNase IIIb mutation when compared to 59 non-neoplastic and 496 DICER1-wt PTCs ( Figure 2C ). One PTC case driven by NRAS p.Q61R and with a co-occurring DICER1 p.R1906S mutation (outside of the RNase IIIb domain) had no increase in the expression of these markers. We next sought to determine whether DICER1 mRNA expression was altered in any of the tumors. In both cohorts (CHOP and TCGA), we found DICER1 mRNA expression to be downregulated in DICER1-wt DTC compared to non-neoplastic cases. On the other hand, DTC with DICER1 RNase IIIb mutations were associated with increased expression of DICER1 mRNA and could explain, at least in part, the increased levels of 3p miRNAs observed in these cases ( Figure 2D ). The differential expression of selected 5p and 3p miRNAs (let-7i-5p, miR-30a-5p, miR-20a-3p and miR-99b-3p) was further validated by quantitative PCR in four DICER1-mut DTC and matched normals – cases ID# 5, 6, 7 and 8 from Table 1 ( Supplemental Figure 3 ).

In thyroid cancer, genetic alterations activating the MAPK pathway are highly prevalent and mutually exclusive to each other, indicating that harboring more than one of these mutations confers no clonal advantage (17, 29, 35). In this study, all cases with DICER1 RNase IIIb mutations lacked genetic alterations activating the MAPK pathway. A literature review of 42 thyroid cancers (including cases from this study) harboring DICER1 RNase IIIb mutations reinforced the mutually exclusive nature of these alterations in thyroid cancer ( Figure 3A ). The DICER1 mutations reviewed here - all in the RNase IIIb domain - are asymmetrically distributed between codons E1705, D1709, D1810, and E1813. E1813 is the most common alteration (20/42; 48%) overall and in each thyroid cancer type (PTC, FTC, PDTC) ( Figures 3A, B ). Even in more advanced forms of the disease such as PDTC, there was no overlap between DICER1 hotspot mutations and genetic alterations activating the MAPK pathway (18, 27). This supports the hypothesis that these alterations are associated with MAPK signaling activation. Indeed, mRNA expression of established MAPK output markers HMGA2, PPAT, EGR1 and CCND1 (from TCGA ERK-score (17)) is increased in DICER1-mut tumors when compared to non-neoplastic/hyperplastic thyroids ( Figure 3C ). Although it is not clear how DICER1 hotspot mutations activate MAPK, we found several positive regulators of the pathway to be upregulated in DICER1-mut tumors: NRG1, RTKs (FGFR3, ERBB2/3), BRAF and RAF1. We also found decreased expression of TPO, TFF3, LRP2 and increased expression of FN1, genes positively and negatively correlated with differentiation, respectively ( Figure 3D ). Although patients with DICER1 hotspot mutations have well differentiated thyroid tumors of follicular cell origin and are at low risk for lymph node metastasis, a fraction of cases are associated with high-risk PTC or PDTC (16). Our literature review of DICER-mutant cases shows that the one case of PTC with N1b disease harbored a TP53 truncating mutation (p.E343Afs*3) ( Figure 3A ). Moreover, five out of 12 (41%) DICER1-driven PDTC also had a TP53 mutation (4/12, 33%) or a TERT promoter mutation (1/12, 8%), alterations associated with progression of thyroid cancer ( Figure 3A ).

To understand the molecular mechanisms (gene networks and signaling pathways) underlying tumorigenesis of follicular cells harboring DICER1 mutations, we analyzed the mRNA expression profiles of the same pediatric thyroid tumors using the HTG Edgeseq OBP panel, which evaluates the mRNA expression of 2,559 genes related to cancer ( Supplemental Table 2 ). We performed a differential gene expression analysis between eight DICER1-mut tumors and 26 non-neoplastic/hyperplastic thyroids and observed 569 differentially expressed genes (DEG) between the two groups (247 up and 322 down, FDR<0.05) ( Figure 4A ). To evaluate the pathways enriched among the DEGs from the attained dataset, we used g:Profiler web-based tool (v11.5) (26). Functional enrichment analysis of these 247 upregulated genes exposed a significant enrichment for cell-cycle genes in four different databases: Gene Ontology (GO), Reactome, KEGG and WikiPathways ( Figure 4B ). Among the upregulated genes in DICER1-mutant compared to non-neoplastic and hyperplastic lesions (FC≥2, and FDR≤.05), were genes related to cell-cycle, such as the transcription factors E2F1 and E2F5, different cyclins (CCNB1, CCND1, CCND2, CCNE2, CCNF), SKP2, TOP2A, MCM2, and the established clinical marker of cell proliferation, MKI67 ( Figure 4C ). E2F transcription factor binding sites were also found to be significantly overrepresented in the collection of upregulated genes based on TRANSFAC analysis, pointing to these transcription factors as important hubs in the cell-cycle regulation of DICER1-mut thyroid tumors ( Figure 4B ). While E2F1 was upregulated in all tumors (DICER1-wt and DICER1-mut), E2F5 was specifically upregulated in DICER1-mut tumors. Remarkably, we found that let-7, mir-17-5p, mir-96-5p and mir-181-5p, all of which were found to be downregulated in DICER1-mut tumors, have predicted binding sites that are conserved among vertebrates in the 3’ UTR of E2F5 ( Figures 4D, E ). Indeed, the regulation of E2F5 mRNA expression by some of these miRNAs has already been previously validated in other cancer types (36–38). The differential expression of selected mRNAs associated with cell-cycle (E2F5, SKP2), MAPK signaling output (HMGA2, CCND1) and thyroid differentiation (TPO) was further validated by quantitative PCR in four DICER1-mut DTC and matched normals – cases ID# 5, 6, 7 and 8 from Table 1 ( Supplemental Figure 4 ).