DNA Extracted From Aged Formalin‐Fixed Paraffin‐Embedded Specimens of Patients With Papillary Thyroid Carcinomas Can Be Used for Whole Exome Sequencing Authors: Kousuke Tanimoto, Tomonori Hayashi, Hidewaki Nakagawa, Kazuhiro Maejima, Koji Arihiro, Megumu Fujihara, Takashi Nishisaka, Masahiro Ito, Masahiro Nakashima, Naohiro Kato, Zhenqiu Liu, Misa Imaizumi, Alina Brenner, Stephen J Chanock, Osamu Tanabe, Kotaro Ozasa Published: 2026 ABSTRACT ABSTRACTBackgroundIn Hiroshima, thyroid cancer tissue samples from A‐bomb survivors have been preserved since 1958. However, the feasibility of using these old formalin‐fixed paraffin‐embedded (FFPE) specimens for genetic analysis has not been fully evaluated. Here, we aimed to assess the quality of DNA extracted from long‐term stored FFPE specimens obtained from patients with papillary thyroid carcinoma (PTC) in Hiroshima and to evaluate their applicability to whole exome sequencing (WES).MethodsWe extracted DNA from tumor and adjacent normal regions of FFPE thyroid tumor specimens from 57 cases stored for up to 55 years at three hospitals in Hiroshima and confirmed the quality of the extracted DNA. From this cohort, tumor and paired normal samples from 24 patients with PTC were selected for WES, and sequencing metrics were evaluated.ResultsBoth DNA yield and DNA integrity number were negatively correlated with storage duration. WES analysis showed that storage duration was also negatively correlated with mean insert size and normalized read depth. In contrast, the uniformity score, which reflects the evenness of sequence coverage across targeted regions, did not show a significant correlation with storage duration. Among the 22 BRAF V600E‐positive cases identified by droplet digital PCR, 18 cases (82%) had at least one sequencing read supporting the mutation.ConclusionsThese findings suggest that FFPE samples preserved for up to 55 years may be amenable to sequencing with increased read depth. However, the progressive decline in DNA quality with longer storage periods indicates that careful interpretation of sequencing results is warranted.Keywords: BRAF mutation, digital droplet PCR, formalin‐fixed paraffin‐embedded, papillary thyroid cancer, whole exome sequencing Introduction Thyroid cancer among A‐bomb survivors is one of the first solid cancers shown to be associated with ionizing radiation, and its incidence has been reported to increase more than a decade after exposure. The Life Span Study (LSS) conducted by the Radiation Effects Research Foundation (RERF) reported a strong linear dose–response relationship between radiation exposure and thyroid cancer incidence among A‐bomb survivors [1]. Radiation‐related risk decreased rapidly with increasing age at exposure, with the highest risk observed among those exposed at younger ages [2]. It has been estimated that approximately 36% of thyroid cancer cases among survivors exposed before the age of 20 years are radiation‐induced, many of which occurred early during follow‐up. However, excess radiation risk has been shown to persist for more than 50 years after exposure. Candidate gene studies of somatic mutations in thyroid cancer cases from the LSS have demonstrated a radiation dose‐dependent increase in RET fusions, and a dose‐dependent decrease in the frequency of BRAF point mutations [3, 4]. These findings are consistent with those reported in individuals who developed thyroid cancer following I‐131 exposure due to the Chernobyl nuclear reactor accident [5, 6, 7]. Similar to the Chernobyl cohort with I‐131 exposure, the LSS cohort—with a large number of thyroid cancer cases (over 370 cases), a high attributable risk and well characterized radiation dose estimates—provides a rare opportunity to investigate genetic alterations associated with radiation exposure. DNA extracted from formalin‐fixed paraffin‐embedded (FFPE) specimens can be informative for genomic analyses using massively parallel sequencing platforms, and its utility has been demonstrated in numerous studies. Several reports have shown that DNA extracted from stored FFPE specimens can be successfully sequenced [8, 9, 10, 11, 12]. However, the use of older FFPE samples remains controversial because specimen age may affect the accurate detection of genetic alterations, including potential driver mutations in papillary thyroid carcinoma (PTC). Although the LSS and other epidemiological studies with long‐term follow‐up necessarily rely on archived FFPE specimens to investigate the effects of radiation on the human genome, few studies have evaluated the usability of DNA from FFPE specimens stored for more than 30 years for genome sequencing. The aim of this pilot study was to evaluate the quality and suitability of FFPE thyroid tissue specimens collected from the general population and stored for up to 55 years at several hospitals in Hiroshima, Japan. The results of this study will inform the design and feasibility of future genomic studies using archived FFPE specimens of thyroid cancer cases in long‐term follow‐up studies, including the LSS. Materials and Methods Sample Collection Thyroid tumor specimens from the general population were collected from the Hiroshima Red Cross Hospital & Atomic Bomb Survivors Hospital (RC), Hiroshima Prefectural Hospital (HP), and Hiroshima University Hospital (HU) with the plan to collect 7 cases from each of the three different storage periods (pre‐1975, 1976–2000, and post‐2001), for a total of 21 cases per institution. As a result, a total of 21 cases each were collected from RC and HU, but only 3 cases were collected from HP before 1975 and 5 cases were collected from HP after 2001, for a total of 57 cases. The pathological diagnosis of the obtained FFPE specimens ultimately resulted in 20 available cases from RC (13 tumor/normal pair, 7 tumor only, Table 1), 12 from HP (10 tumor/normal pair, 2 tumor only), and 18 from HU (15 tumor/normal pair, 3 tumor only). The tumor and adjacent normal regions were macrodissected by pathologists. DNA was extracted from a pooled sample of ten 10 μm‐thick slices using an FFPE tissue DNA isolation kit (Qiagen) in 2017 and sequenced shortly thereafter. All samples were anonymized, and the study protocol was reviewed and approved by the internal review board of the RERF. Because the pathological tissue specimens had been archived more than 30 years ago and it was impracticable to obtain informed consent from the original patients, the requirement for informed consent to participate was waived by the Institutional Review Board of the RERF. TABLE 1Number of cases collected in this study.Storage periodCancer/NormalRCHPHUTotal−1975Cancer and Normal33410Cancer only30251976–2000Cancer and Normal44513Cancer only30032001—Cancer and Normal63615Cancer only1214TotalCancer and Normal13101538Cancer only72312Abbreviations: HP, Hiroshima Prefectural Hospital; HU, Hiroshima University Hospital; RC, Hiroshima Red Cross Hospital & Atomic Bomb Survivors Hospital. Number of cases collected in this study. Abbreviations: HP, Hiroshima Prefectural Hospital; HU, Hiroshima University Hospital; RC, Hiroshima Red Cross Hospital & Atomic Bomb Survivors Hospital. Library Preparation and Analysis of Whole Exome Sequencing (WES) Data WES of tumor samples and paired normal DNA samples from 24 patients with PTC was performed. The library was prepared using the KAPA HyprePlus kit, following the manufacturer's protocol (KAPA Biosystems, South Africa). The coding regions were enriched using SureSelect XT Human All Exon v6 (Agilent Technologies). The enriched DNA was sequenced using HiSeq2000 (Illumina), yielding 150 base‐paired reads. The sequence reads obtained were analyzed using the Genomon2 pipeline (https://genomon‐project.github.io/GenomonPagesR/). Briefly, the following procedure was used. The obtained sequence reads were aligned to the human genome reference GRCh37 using Burrows‐Wheeler Aligner (BWA) [13]. The sequencing read depth was calculated using samtools [14]. Single nucleotide variant (SNV), insertion/deletion (InDel), and somatic variant identification were performed using Genomon2 proprietary pipelines. The identified somatic variants were annotated using ANNOVAR [15]. Annotation Somatic variant annotation was performed using data from The Cancer Genome Atlas (TCGA) papillary thyroid cancer [16], the Tohoku Medical Megabank Organization (ToMMo) 60KJPN database [17], and the Genome Aggregation Database (gnomAD) v4.1.0. TCGA data were accessed through cBioPortal (https://www.cbioportal.org/). Calculations and Data Visualization The read depth at each genomic position targeted by SureSelect was calculated using samtools [14]. The read depth was normalized per 50 million mapped reads and was used as the normalized read depth. Normalized read‐depth calculations and visualizations with a bin size of 1000 bp were performed using in‐house Python scripts. In this study, the uniformity score was defined as the percentage of regions with a depth of more than 75% of the average read depth on each chromosome. Validation of BRAF Mutation Detected BRAF V600E mutations (chr7:140453136 A>T), the most common point mutation driver in PTC, were validated using Sanger sequencing and droplet digital polymerase chain reaction (ddPCR) [4]. Briefly, PCR was performed in a 25 μL mixture containing 10 pmoles of each primer, 200 μM of each dNTP, 0.5 U of FastStart High Fidelity DNA polymerase (Roche, Basel, Switzerland), 20–50 ng genomic DNA, and 1× reaction buffer supplied by the manufacturer. The PCR conditions consisted of initial denaturation (95°C for 2 min), followed by 40 cycles of denaturation at 94°C for 30 s, annealing at 54°C for 60 s, and extension at 72°C for 30 s. The following primers were used: 5′‐tcatgaagacctcacagtaaaaat‐3′ and 5′‐tggatccagacaactgttcaa‐3′. The BRAF V600E mutation was initially screened using restriction fragment length polymorphism (RFLP) with TspRI (New England Biolabs, Ipswich, MA) and subsequently sequenced using the forward primer and Big Dye Terminator (ABI Systems, Applied Biosystems, Foster City, CA, USA). The DNA sequences and BRAF mutations were determined using an ABI‐PRISM 3100 automatic sequencer (Applied Biosystems). The ddPCR mixtures, 20 μL in volume, contained 20 ng DNA, 1× ddPCR supermix for the probe (no dUTP), 0.5 μmol/L of each primer (forward primer: 5′‐CATGAAGACCTCACAGTAAAAATAGGTGAT‐3′; reverse primer: 5′‐TGGGACCCACTCCATCGA‐3′), and 0.25 μmol/L of each probe (wild type: 5′‐VIC‐CGAGATTTCACTGTAGCT‐MGB‐3′, mutated: 5′‐FAM‐CGAGATTTCTCTGTAGCT‐MGB‐3′). Droplets were generated and analyzed using a QX200 system (Bio‐Rad). Amplification was performed as follows: 95°C for 10 min (1 cycle), 94°C for 30 s and 55°C for 1 min (40 cycles), and 98°C for 10 min (1 cycle) with a ramp rate of 2°C/s; the reaction was then held at 4°C with a ramp rate of 1°C/s. The absolute quantification of mutant alleles and wild‐type alleles using ddPCR was performed by modeling as a Poisson distribution using the QuantaSoft v1.6.6 analysis software (Bio‐Rad). The threshold was defined as that described in the “Droplet Digital Application Guide.” Samples with droplet numbers exceeding three in the positive area were considered positive in the QuantaSoft analysis software. Statistical Analysis The Mann–Whitney U test and 95% confidence interval calculations were performed using the Python package, statannotations [18]. Results Quality Assessment of DNA Extracted From FFPE Samples PTC and paired normal tissue samples were obtained from three hospitals in Hiroshima, Japan. DNA was extracted from 50 tumor samples and 38 normal tissue samples derived from 50 patients (Table 1). A summary of the DNA quality is shown in Figure 1. Samples were classified into three groups according to the year in which the histological slides were prepared: 1962–1975, 1976–2000, and 2001. In RC and HP, DNA yield (μg), yield per unit area (μg/cm2), and yield per slide glass (μg/slide) from the oldest period were significantly lower than those from the most recent period; however, no significant differences were observed in HU (Figure 1a–c). DNA integrity was assessed the DNA integrity Number (DIN) measured by TapeStation (Agilent Technologies), revealing a clear decline in DIN with increasing storage period (Figure S1d). DNA yield, yield per unit area, yield per slide glass, and DIN were all negatively correlated with storage period, with Pearson correlation coefficients of −0.38, −0.38, −0.39, and −0.44, respectively (Figure S1a–d). Librares for WES were prepared from the extracted DNA, and library yields are shown in Figure 1d. For samples obtained from HP and HU, library yield increased with more recent storage periods. In contrast, samples from RC showed an inverse trend. No significant differences in DNA quality metrics were observed between tumor and normal tissues. Detailed DNA quality metrics for all samples are provided in Table S1. Quality check of the DNA extracted from the FFPE samples. (a–d) The bar plots show the DNA yield (μg) (a), DNA yield per unit area (μg/cm2) (b), DNA yield per slide glass (μg/slide glass) (c), or WES library yield (d) from FFPE slide glasses in each hospital. Each bar color indicates a storage period: Blue, until 1975; orange, from 1976 to 2000; green, from 2001. Asterisks indicate p‐values calculated using the two‐sided Mann–Whitney U test. *0.01 < p ≤ 0.05, **0.001 < p ≤ 0.01, ***0.0001 < p ≤ 0.001, ****p ≤ 0.0001, ns P > 0.05. Quality Assessment of WES WES was performed on tumor and paired normal DNA from 24 PTC patients. Samples were selected to minimize bias with respect to storage period and hospital while ensuring availability of paired tumor‐normal DNA. The trends in DNA quality observed in these samples were consistent with those observed in the full sample set. The mean number of mapped reads was 76,993,589 (range: 56,019,973–92,978,954). Detailed sequencing statistics are summarized in Table S2. Distributions of normalized read depth, mean insert size and percentage of duplicate reads stratified by storage period and hospital are shown in Figure 2a,b. Across the three storage periods (1962–1975, 1976–2000, and 2001), the median normalized read depths were 15.2, 23.1, and 29.0, respectively, while median mean insert sizes were 81.0, 103.5, and 141.8 bp, respectively. Median duplicate read percentages were 6.2%, 9.6%, and 18.9%, respectively. When stratified by hospital, median normalized read depths were 21.0 for HU, 35.0 for HP, and 20.5 for RC. Median mean insert sizes were 110.6 bp (HU), 142.8 bp (HP), and 87.1 bp (RC), while median duplicate read percentages were 11.6%, 18.7%, and 5.6%, respectively. In the overall samples, no correlation was observed between the number of mapped reads and normalized read depth (Pearson correlation coefficient = 0.29; Figure 2c). When samples were stratified by storage period, a positive correlation was observed in the two more recent storage periods (2001–, 1976–2000). In contrast, increasing the number of mapped reads did not improve normalized read depth in the oldest storage period. A positive correlation was also observed between normalized read depth and mean insert size across all samples, although this correlation was weaker in the oldest storage period (Pearson's correlation coefficient = 0.75; Figure 2d). Storage period showed strong negative correlations with both mean insert size and normalized read depth, with Pearson correlation coefficients of −0.69 and −0.66, respectively (Figure 2e,f). Quality check of WES results. (a, b) Box plots show the distribution of the normalized read depth (left), mean insert size (center), and percentage of duplicate reads (right) in each storage period (a) and hospital (b). Asterisks indicate p‐values calculated using the two‐sided Mann–Whitney U test. *0.01 < p ≤ 0.05, **0.001 < p ≤ 0.01, ***0.0001 < p ≤ 0.001, ****p ≤ 0.0001, ns p > 0.05. (c, d) Scatter plots show the correlation between the number of mapped reads and the normalized read depth (c), mean insert size, and normalized read depth (d). Each dot indicates a library constructed from DNA of cancerous or normal tissue. Each dot color indicates the storage period: Blue, until 1975; orange, from 1976 to 2000; green, from 2001. The line indicates the regression line and the transparent area indicates the 95% confidence interval for each hospital. (e, f) The scatter plots show the correlation between storage period and mean insert size (e), and storage period and normalized read depth (f). Each dot color indicates a hospital: Purple, Hiroshima University Hospital (HU); brown, Hiroshima Prefectural Hospital (HP); light green, Hiroshima Red Cross Hospital and Atomic‐bomb Survivors Hospital (RC). The line indicates the regression line and the transparent area indicates the 95% confidence interval for each hospital. To further evaluate associations among exome sequencing quality metrics while accounting for hospital specific effects, we performed a linear mixed‐effects analysis using the lme4 package in R, allowing both intercepts and slopes to vary by hospital. The estimated coefficients for the fixed effects are summarized in Table 2. TABLE 2Pairwise associations of exome sequencing quality metrics using a mixed‐effects model.The fixed effect of mixed‐effects modelXYCoeff95% CI p Mapped readsNormalized read depth2.857−1.643, 7.3570.214Mean insert sizeNormalized read depth0.2360.144, 0.328< 0.00001Storage time (year)Mean insert size−1.471−2.320, −0.6210.00069Storage time (year)Normalized read depth−0.483−0.732, −0.2340.00014Abbreviations: CI, Confident interval; Coeff, Coefficient of the fixed effect. Pairwise associations of exome sequencing quality metrics using a mixed‐effects model. Abbreviations: CI, Confident interval; Coeff, Coefficient of the fixed effect. Consistent with the Pearson correlation analysis, no statistically significant association was observed between mapped reads and normalized read depth (p = 0.214). In contrast, mean insert size was significantly and positively associated with normalized read depth (p < 0.00001). Storage time was significantly and negatively associated with both mean insert size (p = 0.00069) and normalized read depth (p = 0.00014). Evaluation of Exome Sequencing Uniformity In addition to sequencing depth, read depth uniformity is an important metric for evaluating exome sequencing quality. We therefore assessed uniformity as an additional quality control parameter. Uniformity was defined as the percentage of targeted regions with read depth exceeding 75% of the average read depth for each chromosome. The distributions of normalized read depth across chromosomes for the three storage periods are shown in Figure 3a. Samples from the oldest period (1962–1975) exhibited lower read depth across all autosomal chromosomes compared with those from the two more recent periods. However, the distributions of the uniformity scores did not differ among the three periods (Figure 3b). When stratified by hospital, normalized read depth was lower for RC compared with HU and HP (Figure 3c), whereas uniformity scores were largely comparable across hospitals, with the exception of chromosome 8 (Figure 3d). Normalized read depth across the BRAF locus is shown in Figure 3e. Average read depth and uniformity scores for all chromosomes are summarized in Table S3. Although normalized read depth was lower in the oldest samples, the overall read depth distribution patterns were similar across all three storage periods, indicating that even the oldest FFPE samples exhibited relatively uniform coverage. Similar patterns were observed for chromosome 7, which harbors BRAF (Figure 3f), as well as for across all other autosomal chromosomes (Figure S2). Distribution of the uniformity score of exome sequencing. (a‐d) The box plots show the distributions of the normalized read depth (a, c) or the uniformity score (b, d) of WES libraries in each autosomal chromosome. The X‐axis and X‐axis indicate the chromosome and normalized read depth (a, c) or the uniformity score (b, d), respectively. Each bar color indicates a storage period: Blue, until 1975; orange, from 1976 to 2000; green, from 2001 or a hospital; purple, Hiroshima University Hospital (HU); brown, Hiroshima Prefectural Hospital (HP); light green, Hiroshima Red Cross Hospital & Atomic‐bomb Survivors Hospital (RC). (e, f) Normalized read depth in the BRAF coding region (e) or across chromosome 7 (f) with a bin size of 1000 bp was plotted. Each line color indicates the storage period: Blue, until 1975; orange, from 1976 to 2000; green, from 2001. Each line indicates the average normalized read depth for each storage period. The transparent area indicates one standard deviation around the mean for each storage period. Results of Somatic Variant Detection A total of 290 somatic variants, including synonymous alterations, were identified, of which 200 were non‐synonymous. These variants comprised 281 single‐nucleotide variants (SNVs) and 9 insertions/deletions (InDels). Variant allele frequency (VAF) distributions are shown in Figure 4a. Median VAF values ranged from 10.0 (case 13061C) to 33.9 (case 12023C). No association was observed between VAF distribution and storage period. To evaluate potential effects of long‐term storage on base substitution patterns, mutational signature analysis was performed using the identified SNVs. Across all storage periods, C>T transitions predominated, consistent with known formalin‐induced artifacts [19, 20, 21, 22], and no differences in mutational signature patterns were observed among the three groups (Figure 4b). Characteristics of the identified somatic variants. (a) The allele frequency of each identified somatic variant was plotted. The X and Y axes indicate sample ID and allele frequency (%), respectively. Each dot indicates an identified somatic variant. The color of each dot indicates a storage period: Blue, until 1975; orange, from 1976 to 2000; green, from 2001. Each red horizontal line indicates the median of each sample. (b) The mutation signatures of each storage period are shown. Base substitution patterns are divided into 96 patterns defined by the substituted base with flanking 5′ and 3′ nucleotides. (c) The scatter plots show the correlation between the variant allele frequency in WES and mutation percentage in ddPCR. Each dot indicates a case. The color of each dot indicates a storage period: Blue, until 1975; orange, from 1976 to 2000; green, from 2001. The line indicates the regression line and the transparent area indicates the 95% confidence intervals for each hospital. Somatic variants were identified most frequently in BRAF (10/24, 42%), followed by MUC17 (2/24, 8%), RP1L1 (2/24, 8%), FLG (2/24, 8%), GSG2 (2/24, 8%), and TP53 (2/24, 8%). All the identified BRAF variants harbored the V600E mutation. Only these six genes harbored somatic variants in two or more individuals, whereas the remaining 256 genes were identified in only one individual. Of the genes harboring somatic variants in our study, 72 were also represented in the TCGA papillary thyroid cancer dataset. With the exception of BRAF, the frequencies of cases harboring somatic variants in a given gene were below 1% for all genes in our cohort. A similar pattern was observed in the TCGA dataset, where the frequencies were also low for most genes except the major driver genes BRAF, NRAS, and HRAS [16]. We assessed the frequencies of the identified somatic variants in ToMMo Japanese and gnomAD East Asian population datasets. A total of 55 somatic variants were observed across the two datasets, all in a single individual. Of these, the allele frequency of 22 variants was more than 0.0001. Detailed information on all identified somatic variants is provided in Table S4. To assess the accuracy of WES‐based variant detection, BRAF V600E mutations were validated using Sanger sequencing and droplet PCR (ddPCR). Validation revealed that 22 patients (92%) harbored the BRAF V600E mutation. Detailed validation results are presented in Table 3 and Figure S3. Although WES‐based variant calling and ddPCR results were discordant in 12 cases, at least one read harboring the BRAF V600E mutation was detected in 18 cases (82%). Concordance appeared to be lower in samples with low mutant allele fractions and limited sequencing depth. Among the four cases without detectable mutant reads by WES, two exhibited mutation fractions below 6% by ddPCR, suggesting extremely low tumor cellularity. One additional case had fewer than 10 reads mapped to the BRAF V600E locus (chr7: 140453136) in WES. A weak positive correlation was observed between VAF estimated by WES and mutation fraction measured by ddPCR (Pearson correlation coefficient = 0.36, Figure 4c). TABLE 3Summary of BRAF mutation validation results.HospitalPeriodSample IDExomeddPCRConcordantDepthVAF (%)VariantCopies/μLMutant %Variant callRef (A)Alt (T)DepthCallWildMutantRC−197511016C92111802.30.7231No11021C8080011.32.3171No1976–200012023C614207012.20.5191Yes12024C17017000.77000Yes12026C26026003.3000Yes2001—13031C266321912.30.61211Yes13034C219303017.53.3311Yes13061C30030006.82.3251NoHP−197521001C710175902.90.59171No1976–200022004C104142918.31.7171Yes22008C2743113117.74.3201Yes22010C671354111.34.6291Yes2001—23011C231244018.70.6531No23013C2302300170.841No23014C342366030.50.5121NoHU–197531041C15015003.70.2561No31042C102121702.51.5381No31043C108184412.70.58181Yes1976–200032047C26123832110.75.5341Yes32048C14115704.31.9311No32052C53838011.92141No2001—33054C931225016.67.3311No33059C2182928117.65.4231Yes33060C1742119115.38.5361YesAbbreviations: Alt, number of reads supporting the alternative allele (i.e., differing from the reference sequence); HP, Hiroshima Prefectural Hospital; HU, Hiroshima University Hospital; RC, Hiroshima Red Cross Hospital & Atomic‐Bomb Survivors Hospital; Ref, number of reads supporting the reference allele (i.e., matching the reference sequence); VAF, variant allele frequency calculated as Alt/(Ref + Alt) × 100; Variant called, Variant call: 0 = not detected, 1 = detected. Concordant indicates agreement between WES‐based variant calling and ddPCR results (Yes = concordant, No = discordant). Summary of BRAF mutation validation results. Abbreviations: Alt, number of reads supporting the alternative allele (i.e., differing from the reference sequence); HP, Hiroshima Prefectural Hospital; HU, Hiroshima University Hospital; RC, Hiroshima Red Cross Hospital & Atomic‐Bomb Survivors Hospital; Ref, number of reads supporting the reference allele (i.e., matching the reference sequence); VAF, variant allele frequency calculated as Alt/(Ref + Alt) × 100; Variant called, Variant call: 0 = not detected, 1 = detected. Concordant indicates agreement between WES‐based variant calling and ddPCR results (Yes = concordant, No = discordant). Discussion In this study, we examined the quality of DNA extracted from FFPE samples of patients with PTC stored in three hospitals in Hiroshima. We observed that older samples yielded less DNA and showed increased fragmentation (Figure 1), which was expected because DNA extracted from long‐stored FFPE samples is known to be of poor quality [23]. To evaluate the feasibility of genome sequencing, we performed WES using tumorous and paired normal DNA extracted from FFPE slides of 24 cases. As expected, the average read depth decreased with increasing sample age (Figure 2f). Read depth showed a strong positive correlation with library insert size (reflecting the degree of DNA fragmentation) and a negative correlation with storage period (Figure 2d,e), indicating that the major limitation in sequencing older FFPE samples was insufficient read depth caused by fragmented genomic DNA. Therefore, increasing sequencing depth alone may not fully compensate for the effects of severe DNA fragmentation in very old samples. Although we were unable to define a specific storage‐duration threshold because of the limited sample size, our findings suggest that FFPE specimens stored for several decades remain suitable for WES analysis. In contrast, specimens preserved for more than approximately 50 years may require careful assessment of DNA quality and greater sequencing effort to achieve adequate exome coverage. On the other hand, the uniformity score in the oldest storage period was comparable to those in the two more recent storage periods (Figure 3b). This observation suggests that increasing the number of sequencing reads could improve sequencing performance, provided the libraries retain sufficient complexity. Although comparable uniformity scores do not necessarily indicate comparable library complexity, uniformity is commonly used as an indicator of library complexity. Therefore, within a given library, increasing the number of sequencing reads may improve read depth as long as sufficient library complexity is preserved. Nevertheless, further investigation is required to determine the feasibility of whole‐exome sequencing using specimens stored for such extended periods. In comparing institutions, WES derived from FFPE samples obtained from the RC hospital tended to exhibit poorer quality metrics than those from the other two institutions (Figure 2b,e,f). This difference may not be explained solely by storage duration but could also be attributable to long‐term storage under environmental conditions less conducive to molecular biological analyses. Specifically, for several decades, these samples were reportedly stored in facilities without air conditioning and were potentially exposed to extreme summer temperatures exceeding 40°C. Such prolonged exposure to high temperatures may have accelerated DNA degradation and fragmentation, thereby contributing to the observed reduction in DNA quality. In general, WES for somatic variant detection requires a read depth of approximately 80–150× [24, 25, 26, 27]. Based on the approximate relationship between normalized read depth and mean insert size derived from Figure 2d (y = 0.2737x − 4.9533), we estimated that approximately 270 million mapped reads would be required for a library with an insert size of 120 bp when using an Illumina sequencer and SureSelect exon enrichment reagent, although this requirement may vary depending on cancer content. To assess the impact of long‐term FFPE storage on base substitution patterns, we performed mutational signature analysis and found no significant differences among storage periods (Figure 4a). Formalin fixation is known to cause hydrolytic deamination of cytosine, resulting in C>T (G > A) base substitutions in sequencing data [19, 20, 21, 22]. The mutation signature of unrepaired FFPE samples has been reported to resemble SBS30 in the COSMIC Single Base Substitution (SBS) database [19], which is characterized by dominant C>T. Consistent with this, our results showed a pattern similar to SBS30, particularly in samples stored before 1975 and between 1976 and 2000. Although comparison of substitution pattern excluding C>T would be ideal, we were unable to perform such an analysis because the fragmented DNA yielded an insufficient number of SNVs due to limited read depth. Our whole‐exome analysis identified 290 somatic variants, and 42% of individuals harbored BRAF variants. BRAF V600E is the most common mutation in PTC, occurring in approximately 45%–70% of cases [28, 29]. Other known driver genes, such as NRAS and HRAS, were not observed in our cohort due to their low frequencies compared with BRAF and the limited sample size. A total of 22 (8%) somatic variants were observed in the ToMMo Japanese or gnomAD East Asian populations at allele frequencies greater than 0.0001, suggesting that most somatic variants identified through our tumor‐normal paired analysis are rare in the general population. However, further functional studies are required to determine the biological significance and pathogenic potential of these variants. To evaluate the accuracy of somatic variant identification using WES, we validated BRAF V600E mutation using Sanger sequencing and ddPCR. The specificity of WES was 100% (2/2) (Table 3). Although the sensitivity based on the variant caller implemented in the analysis pipeline was relatively low (10/22) (Table 3), this disconcordance was not solely attributable to sample quality. In general, ddPCR has much higher sensitivity than WES, even when high quality DNA and sufficient read depth are available. Current clinical cancer gene panel tests typically require a variant allele fraction (VAF) of at least 20% [30] and read depths of several hundred reads [31], with higher read depths required as cancer content decreases. Among the 12 cases not detected by the variant caller, four had VAFs below 20% as estimated by ddPCR, and the remaining cases also exhibited read depth below 40. Nevertheless, 18 of 22 BRAF V600E‐positive cases contained at least one read supporting the mutation, suggesting that the potential sensitivity could reach 82% if sufficient read depth were achieved. In addition, a weak positive correlation was observed between VAFs estimated by WES and mutation percentages measured by ddPCR (Figure 4c), indicating that false negative calls likely resulted from insufficient numbers of mutant reads falling below the variant call threshold. These findings further support the notion that increasing read depth can improve variant call accuracy in old and highly fragmented DNA samples. In conclusion, we assessed the quality and quantity of DNA extracted from FFPE slides stored for an extended period of 55 years and demonstrated that fragmented genomic DNA can still be used for somatic variant analysis using short‐read massively parallel sequencing. However, since DNA quality declined with storage period, careful interpretation of genotyping results is required. We anticipate that our findings will provide valuable insights for studies using archival FFPE samples and facilitate investigations into the relationship between genomic profiles and various diseases. Author Contributions T.H., H.N., K.M., and K.O. planned the original study design. K.A., M.F., and T.N. provided the pathological materials and information, and M.I., M.N., M.I., and A.B. gave pathological, clinical, and epidemiological suggestions. K.T., T.H., and H.N. conducted the experiments, and K.T., T.H., H.N., K.M., M.N., and N.K. conducted data analyses. M.I., A.B.G., and S.J.C. provided technical and supervision support. K.T. and T.H. wrote the manuscript. All authors contributed to the writing of the manuscript and its final approval. Funding This study was based on the RERF Research Protocol #P1‐16 and was also partially funded by the DOE award no. DE‐HS0000031 from the National Academy of Sciences, and contract no. HHSN261201400009C from the U.S. National Cancer Institute (NCI), with additional support from the Division of Cancer Epidemiology and Genetics of the NCI Intramural Research Program. The funding bodies did not participate in the study design, collection, analysis, and interpretation of data, or in the writing of the report. The authors' views do not necessarily reflect those of the two governments. Ethics Statement This study was reviewed and approved by the Research Protocol Review Committee (RPRC) of the Radiation Effects Research Foundation (RERF) (Approval No. RP‐P1‐16). The study was conducted in accordance with the ethical standards of the responsible committee and with the Declaration of Helsinki and its later amendments. Consent The requirement for informed consent was formally waived by the Committee because the archival formalin‐fixed paraffin‐embedded specimens analyzed in this study were collected more than 30 years ago, and the donors are deceased. Conflicts of Interest The authors declare no conflicts of interest. Supporting information Figure S1: Correlation between FFPE storage time and the quality of extracted DNA. Figure S2: Distribution of normalized read depth across each chromosome. Figure S3: Validation of the BRAF V600E mutation in all samples. Table S1: DNA quality metrics for all samples. Table S2: Whole‐exome sequencing statistics. Table S3: Average read depth and uniformity scores for all autosomal chromosomes. Table S4: All identified somatic variants.