Transcriptomic characteristics according to tumor size and SUVmax in papillary thyroid cancer patients

The SUVmax is a measure of FDG uptake and is related with tumor aggressiveness in thyroid cancer, however, its association with molecular pathways is unclear. Here, we investigated the relationship between SUVmax and gene expression profiles in 80 papillary thyroid cancer (PTC) patients. We conducted an analysis of DEGs and enriched pathways in relation to SUVmax and tumor size. SUVmax showed a positive correlation with tumor size and correlated with glucose metabolic process. The genes that indicate thyroid differentiation, such as SLC5A5 and TPO, were negatively correlated with SUVmax. Unsupervised analysis revealed that SUVmax positively correlated with DNA replication(r = 0.29, p = 0.009), pyrimidine metabolism(r = 0.50, p < 0.0001) and purine metabolism (r = 0.42, p = 0.0001). Based on subgroups analysis, we identified that PSG5, TFF3, SOX2, SL5A5, SLC5A7, HOXD10, FER1L6, and IFNA1 genes were found to be significantly associated with tumor aggressiveness. Both high SUVmax PTMC and macro-PTC are enriched in pathways of DNA replication and cell cycle, however, gene sets for purine metabolic pathways are enriched only in high SUVmax macro-PTC but not in high SUVmax PTMC. Our findings demonstrate the molecular characteristics of high SUVmax tumor and metabolism involved in tumor growth in differentiated thyroid cancer.

F-fluorodeoxyglucose (FDG) PET/CT is increasingly performed for the staging or localization of metastatic disease in patients with various kinds of malignancies 16,17 .In thyroid cancer, iodine-131-WBS has been useful for determining the differentiation of a tumor on the basis of its avidity to iodine, identifying remnant thyroid tissue, and assessing patients for distant metastatic disease 18 ; most well-differentiated thyroid carcinomas are relatively slow growing and can be FDG-negative 19 .Recently, however, a "flip-flop phenomenon" has been observed, in that radioiodine uptake in differentiated thyroid carcinoma (DTC) cells decreases when they dedifferentiate while their glucose metabolism generally increases, and 18 F-FDG PET/CT has emerged as a powerful tool for predicting recurrence in DTC patients 20 .Moreover, preoperative SUV max was found to be related to postoperative recurrence-free survival (RFS) 21 .It is commonly accepted that a high 18 F-FDG uptake reflects the dedifferentiation of thyroid tumors; however, there is little data on the role of 18 F-FDG PET/CT in thyroid cancer, and the underlying molecular glucose metabolism mechanisms are not entirely understood.
This study is the first to report findings on 18 F-FDG PET/CT in thyroid cancer in conjunction with a gene expression analysis in an attempt to examine the molecular characteristics related to metabolism in thyroid cancer.We investigated the relationship between SUV max and clinical features of PTC; we also analyzed differentially expressed genes (DEGs) and activated pathways related to SUV max .The significance of genes and pathways were validated using The Cancer Genome Atlas (TCGA) database.

Enriched DNA replication, pyrimidine and one-carbon metabolism, and cell cycle signaling with reduced thyroid differentiation scores (TDS) in PTCs with high SUV max
All 80 patients underwent preoperative 18 F-FDG-PET/CT (Fig. 1A).As previously reported 22,23 , SUV max correlated with tumor size (r = 0.54, p < 0.0001) (Fig. 1B).Since the predictive preoperative SUV max value for tumor recurrence derived from the ROC curves was 10.15, we divided the subjects into PTC SUV-high (SUV max > 10) and PTC SUV-low SUV max ≤ 10) group (Supplementary Fig. 1) and PTC SUV-high revealed the worse prognosis in recurrence free survival analysis (Fig. 1C).
Since glucose metabolism in thyroid cancer cells is reprogrammed to enhance glucose uptake, glycolysis, and lactate synthesis 24 and glucose uptake via glucose transporters (GLUTs) is the first step in producing energy and nucleic acids for cancer survival, we investigated the expressions of GLUTs in relation to SUV max .Previously, thyroid cancer cells show overexpression of hypoxia-responsive GLUT1 and GLUT3 proteins compared to normal cells 25 , however, it is not fully understood the relation of GLUTs and PET-CT SUV max in PTC.We found that GLUT3, GLUT5, and GLUT8-10 were negatively correlated with SUV max and othter GLUTs were not significantly associated with SUV max (Fig. 1D).Since many studies have focused on the impact of TDS, and tumor differentiation rate has been shown to correlate with GLUT expression 26,27 , we also analyzed TDS, GLUTs, and glucolysis according to SUV max and tumor upon the scores calculated using GSVA (Fig. 1D,E).Gene score of GLUTs family was not significantly associated with SUV max (r = − 0.20, p = 0.078) (Fig. 1F), however, gene score of glucose metabolic process was significantly associated with SUV max (r = 0.24, p = 0.036) (Fig. 1G).TDS was not significantly correlated with SUV max (r = − 0.17, p = 0.138) (Fig. 1H) , but The TDS expression heatmap derived from SUV max showed a tendency for some of TDS gene expression were decreased in the PTC SUV-high group (Fig. 1I).Although the gene score of glucose metabolic process was not significantly changed between PTC SUV-high group and PTC SUV-low group (Fig. 1J), PTC SUV-high group revealed significantly decreased TDS score compared to PTC SUV-low group (Fig. 1K).Although the levels of GLUTs genes, such as SLC2A1, SLC2A2, SLC2A3, and SLC2A4 were not changed, several TDS genes, SLC5A5, TPO, DIO2, and TG, were significantly lower in the PTC SUV-high than in the PTC SUV-low group (Fig. 1L and Supplementary Table 1).Moreover, SUV max was negatively correlated with several TDS genes, DIO1 (r = − 0.33, p = 0.003), SLC5A5 (r = − 0.33, p = 0.003), and TPO (r = − 0.30, p = 0.007) expression (Fig. 1M).
Next, we performed unsupervised analysis based on GSEA to identify enriched pathways in each subtype.The PTC SUV-high group exhibited enriched DNA replication, ribosome assembly, pyrimidine metabolism, onecarbon pool by folate, purine metabolism, tight junction, adherens junction, and cell cycle processes in the KEGG database (Fig. 2A and Supplementary Table 2).Ribosome biogenesis, DNA replication initiation, DNAdependent DNA replication, and DNA replication from Gene Ontology Biological Process (GOBP) gene sets were also enriched in the PTC SUV-high group (Supplementary Fig. 2 and Supplementary Table 3).We calculated the scores using GSVA related to various signaling pathways for each tumor sample based on the meaningful GSEA results.We confirmed a positive correlation between SUV max and DNA replication (r = 0.29, p = 0.009), pyrimidine metabolism signaling (r = 0.50, p < 0.0001), cell cycle (r = 0.24, p = 0.029), and purine metabolism (r = 0.42, p = 0.0001) (Fig. 2B-E).In addition, the expression of genes related to DNA replication, one-carbon metabolism, and cell cycle was higher in the PTC SUV-high group (Fig. 2F).Collectively, our data revealed that SUV max of thyroid tumor was not correlated with glycolysis, but was significantly related with several molecular pathways, including DNA replication, cell cycle, pyrimidine metabolism and purine metabolism.Heatmap of TDS genes using log 10 TPM in our cohort.Tumors were sorted according to SUV max .(J,K) Comparison of GSVA for glycolysis or tumor differentiation score (TDS) between PTC SUV-low and PTC SUV-high tumors.(L) Expression of genes encoding SLC2A1, SLC2A2, SLC2A3, SLC2A4, SLC5A5, and TPO between PTC SUV-low and PTC SUV-high tumors.(M) Expression of TDS genes, DIO1, DUOX2, SLC26A4, SLC5A5 and TPO, between PTC SUV-low and PTC SUV-high tumors, and scatter plots of correlation between SUV max and either SLC5A5 or TPO.In the scatter plots, blue line is drawn using simple linear regression and the gray colored area indicate 95% confidence band.
Vol:.( 1234567890 1).In comparison of PTC SUV-high and PTC SUV-low , PTC SUV-high has more extracapsular invasion (ECI, p = 0.003), gross extrathyroidal extension (gross ETE, p = 0.013), and recurrence (p = 0.026) than PTC SUV-low .Although the PTC SUV-high group had more invasive features, the distribution of subtypes and pathological stage were comparable to those in the PTC SUV-low group (subtype, p = 0.331; pathological stage, p = 0.776).Next, we compared the various clinic-pathologic findings after tumor-size adjustments using ANCOVA or Mantel-Haenszel Chi-square test (Table 1).After adjustment for tumor size, PTC SUV-high has also more extracapsular invasion (ECI, p = 0.013), gross extrathyroidal extension (gross ETE, p = 0.028), and recurrence (p = 0.047) than PTC SUV-low group.These data suggested the need for establishment of the molecular features of SUV max high tumors and the importance of SUV max values for predicting cancer progression independent of tumor size.In addition, a comparison of the factors between PTMC and macro-PTC after adjustment to SUV max revealed that lateral lymph node metastasis (L-LNM, p = 0.019) rate was significantly higher in macro-PTC than PTMC (Supplementary Table 4).Collectively, our clinical data suggested that large tumor was associated with lymph node metastasis after adjustment to SUV max , and high SUV max tumor was associated with invasion or extrathyroidal extension after adjustment to tumor size.
To identify the major contributory molecular characteristics according to tumor size and SUV max in PTC, we divided tumors into PTMC and macro-PTC by 1 cm of the tumor size and further subgrouped using 10 of SUV max : PTMC SUV-low , PTMC SUV-high , macro-PTC SUV-low , and macro-PTC SUV-high (Supplementary Fig. 1B and Table 2).Consistently with the findings of Table 1, the PTMC SUV-high group exhibited more ECI and gross ETE than the PTMC SUV-low group (Table 2).Moreover, the PTMC SUV-high group exhibited even more gross ETE than the macro-PTC SUV-low group and the ECI and gross ETE rates did not significantly differ between the two SUV-high groups (PTMC SUV-high and macro-PTC SUV-high ), suggesting the importance of SUV max on gross ETE, independently on tumor size.In contrast, significantly less lateral LNM (L-LNM) was observed in the PTMC SUV-high group than that in the macro-PTC SUV-high group (Table 2).Taken together, SUV max and tumor size were independently correlated with different clinical factors, such as ETE or lymph node metastasis.We conducted an analysis of DEGs and enriched pathways between subgroups: (1) PTMC SUV-low vs. PTMC SUV-high , (2) macro-PTC SUV-low vs. macro-PTC SUV-high , and (3) all SUV max -low vs. SUV max -high (Supplementary Tables 5-7, 8-10, and 1-3, respectively).The relationship between up-and down-regulated DEGs in each comparison is shown in Supplementary Fig. 3A,B.To identify DEGs related to high SUV max independent of tumor size, we intersected all three comparisons and found 28 common DEGs (6 upregulated and 22 downregulated) (Supplementary Fig. 3A,B, Supplementary Table 11).Among these, one upregulated DEG (PSG5) and four downregulated DEGs (TFF3, SOX2, SLC5A5, SLC5A7) showed a significant difference in RFS in the GEPIA 2 database (Table 3, Fig. 3A-E).In addition, DEGs from comparison (3) that did not belong to those from comparison (1) or (2) were defined as DEGs related to SUV max but not tumor size.We identified 16 upregulated DEGs and 60 downregulated DEGs (Supplementary Fig. 3A,B).Among these DEGs, one upregulated (HOXD10) and two downregulated (IFNA1 and FER1L6) DEGs exhibited a significant difference in RFS in the GEPIA 2 database (Table 3, Fig. 3F-H).Based on these results, we identified the up-and down-regulated DEGs related to SUV max or tumor size and further selected some genes associated with RFS.
Our gene ontology analyses identified several gene sets that were shared across the comparisons.The distribution of upregulated KEGG and GOBP genes is shown in the Venn diagrams in Supplementary Fig. 3C,D.In KEGG analyses, genes pivotal for tumor survival and progression were enriched in PTC SUV-high compared to PTC SUV-low (Table 4).Our subgroup analyses revealed enriched DNA replication, cell cycle processes, and ribosome assembly in the SUV max -high groups compared to the SUV max -low groups.Interestingly, the gene sets associated with glucose metabolism, such as glycolysis, the citrate cycle (TCA cycle), and the glycolysis offshoot pathway including purine metabolism and the pentose phosphate pathway were enriched in high SUV max tumors of macro-PTCs but not in those of PTMCs (Table 4).GOBP analyses also revealed that gene sets for DNA replication, cell cycle processes, and ribosome assembly were enriched in the high SUV max subgroups of both PTMC and macro-PTC.In contrast, gene sets for purine metabolism, glucose import, metabolic processes, and ribose  5).These results suggest that purine metabolism, the ribose phosphate pathway, and glucose import may be related to tumor growth rather than SUV max .

Discussion
In this study, we analyzed the clinicopathological characteristics of PTC SUV-low and PTC SUV-high patient groups and investigated their metabolic features using transcriptomic analysis.We found that SUV max was positively correlated with tumor size, and the PTC SUV-high group exhibited higher ECI and gross ETE rates than the PTC SUV-low   www.nature.com/scientificreports/ group after adjustments of tumor size.Transcriptomic analysis revealed lower expression of TDS genes in the PTC SUV-high compared to the PTC SUV-low group, and SUV max was significantly associated with various gene signatures, including DNA replication, pyrimidine metabolism, purine metabolism, and Cell cycle.To determine the molecular characteristics that are independent of tumor size, a DEG analysis of the four tumor size and SUV max subgroups identified five shared DEGs (upregulated PSG5, and downregulated TFF3, SOX2, SLC5A5, and SLC5A7) that were related to SUV max and RFS, and three DEGs (upregulated HOXD10 and downregulated IFNA1 and FER1L6) related to tumor size and RFS that were unrelated to SUV max .PET/CT plays an important role in the diagnosis, staging, and treatment response assessment of various solid cancers.In thyroid cancer, PET/CT scans are not routinely performed; instead, they are recommended in patients with an aggressive subtype and poorly differentiated thyroid cancer at initial staging and follow-up.Clinical practice is based on the inverse relationship between RAI-avidity and FDG-avidity 28 , however there was no study focused on the dissection of PET-CT imaging and transcriptomics in thyroid cancer.Tumors with high SUV max are generally considered to have high glycolytic activity and the Warburg effect explains that aggressive tumors gain energy from aerobic glycolysis rather than from the TCA cycle, producing lactate than pyruvate 29 .In contrast to the previous study reporting a positive correlation between GLUT3 and GLUT4 protein expression and SUV max in PTC 23 , our transcriptomics shows a negative or neutral relationship between GLUT family gene expression and SUV max, although glycolysis tended to show a positive correlation with SUV max , suggesting GLUTs gene expression is not directly aligned with glycolysis.
The clinicopathological impact of high SUV max tumors has been extensively investigated in various tumors 30,31 .In thyroid cancer, a previous retrospective study of a relatively small number of patients (N = 88) failed to show a difference in SUV max between the recurrent and non-recurrent group 32 .However, another retrospective study with www.nature.com/scientificreports/an 8-year follow-up period with a large patient size (N = 400) showed a significant survival decrease in patients with high-SUV max determined on the initial PET/CT scan 21 .In agreement with a previous study, the PTC SUV-high group in our cohort exhibited lower expression levels of TDS genes, and RFS was shorter in this group than in the PTC SUV-low group.When we focused on PTMC, although ECI and gross ETE rates were higher in patients with PTMC SUV-high than in those with PTMC SUV-low , we could not find a difference in recurrence, primarily due to the small number of patients in these subgroups.Therefore, further studies with larger numbers of patients are required to validate the efficacy of PET/CT scans in predicting recurrence in PTMC.Tumor progression is accompanied by both the physical growth of the tumor and concomitant metabolic changes, making it challenging to identify the genes and pathways responsible for both these changes separately.Our study revealed five SUV max -related DEGs (PSG5, TFF3, SOX2, SLC5A5, and SLC5A7) contributing to RFS differences.The PSG5 gene, upregulated in PTC SUV-high regardless of tumor size, was previously reported as a prognostic marker for laryngeal cancer and is known to interact with prognostic lncRNAs in gastric cancer 33,34 .However, its prognostic role in thyroid cancer has not been determined.Another SUV max -related gene, TFF3, which is downregulated in PTC SUV-high tumors, plays a role in angiogenesis and tumorigenesis in breast, stomach, and colon cancers.In thyroid cancer, low expression of TFF3 can increase cell proliferation, migration, and invasion via activation of the IL-6/JAK/STAT3 signaling pathway 35 .These inflammatory pathways lead to high immune cell infiltration around thyroid cancer cells and could serve as a source of increased SUV max 36 .High expression of the stemness marker SOX2 is associated with poor prognosis in several solid tumors and is a regulator of GLUT1 expression 37,38 .In our cohort, the PTC SUV-high group also exhibited low SOX2 expression and relatively low levels of GLUT1 37,38 .SLC5A5 is a well-known marker of thyroid differentiation, and PTC with low expression of SLC5A5 is iodine non-avid and has a poor prognosis 39 .SLC5A7, which encodes a choline transporter, is downregulated in various solid cancers, and its expression is markedly suppressed in PTC  .In colorectal cancer, promoter methylation and the resultant low expression of SLC5A7 are poor prognostic factors as our results.
We also identified three DEGs (HOXD10, IFNA1, and FER1L6) related to tumor size.The expression of HOXD10 is low in PTC, and the HOXD10 gene is hypermethylated in BRAF V600E mutants 41 .However, PTC SUV-high in our study exhibited high expression levels of HOXD10, which might be related to large tumor size.HOXD10 overexpression has been reported to induce cancer cell proliferation, while low expression induced invasion and metastases in head and neck cancer cell lines, supporting the proliferative role of HOXD10 in cancer 42 .A previous study showed low expression of FER1L6 in PTC; however, its prognostic significance and mechanism are not fully understood 43 .IFNA1 has an antitumor effect that inhibits proliferation; thus, low expression of IFNA1 could lead to cancer cell proliferation 44 .
Several studies have been conducted to understand the metabolic features of high SUV max tumors using transcriptomic analyses.In breast cancer, the SUV-high-cluster was associated with frequent TP53 mutations and enhanced the expression of downstream glycolysis genes through FOXM1-LDHA 45 .In multiple myeloma, a negative 18 F-FDG PET/CT scan was associated with low expression of hexokinase-2, whereas a positive scan is accompanied by high expression of proliferation genes or GLUT5 46 .In intrahepatic cholangiocarcinoma, cell cycle processes, cell division, and mitosis gene sets were enriched in high SUV max tumors 47 .Similarly, in this study, the PTC SUV-high group exhibited enriched gene sets for DNA replication, cell cycle processes, and ribosome assembly, regardless of tumor size, in both KEGG and GOBP analyses.Notably, some gene sets showed differences in SUV max in macro-PTC but not in PTMC: these were the gene sets for glucose import, glycolysis, citrate cycle TCA cycle, purine metabolism, one-carbon pool by folate, and the pentose phosphate pathway.For tumor growth or proliferation, new macromolecules, such as nucleic acids, lipids, and proteins, are essential, and macro-PTC reprograms and exploits cellular pathways to obtain the materials necessary for proliferation.These cellular pathways could be targets for anticancer therapy, and further studies are needed to assess the precise manipulation of key steps.
In summary, we investigated the clinicopathological and transcriptomic features of PTC based on the SUV max and tumor size.In PTMC and PTC, tumors with high SUV max exhibited more capsular invasion and gross ETE than low SUV max tumors.DEG analyses revealed the genes contributing to RFS and related to SUV max (PSG5, TFF3, SOX2, SLC5A5, and SLC5A7) and tumor size (HOXD10, IFNA1, and FER1L6).GSEA revealed that gene sets for DNA replication, cell cycle processes, and ribosome assembly were enriched in high SUV max tumors regardless of tumor size, whereas gene sets for glucose import, glucose metabolic process, purine metabolism, and the pentose phosphate pathway were related to large tumor size.
Our research provides insight into metabolic reprogramming of PTC related to SUV max , as well as markers to account for SUV max and tumor size-related RFS.Going beyond the current method of evaluating tumors only by size, using suggested gene biomarkers as well as SUV max to classify tumors into more diverse subgroups will help predict patient prognosis and pave the way for tailor-made treatment protocols in the future.

Study population
We retrospectively reviewed 80 patients postoperatively diagnosed with PTC who underwent preoperative 18 F-FDG-PET/CT and provided informed consent for collection of fresh frozen thyroid tissue from January 2003 to December 2010.Prior to the 2015 ATA guidelines, total thyroidectomy was performed in patients with the tumor size of 1 cm or more, bilateral multifocality, aggressive variant type, ETE in preoperative radiology (except ETE to only the strap muscle), or N1b lymph node metastasis.70 patients were received total thyroidectomy and 10 patients were received lobectomy.Among the 10 patients, none underwent recurrence or completion thyroidectomy.All patients who underwent total thyroidectomy were received radioactive iodine (RAI) treatment according to 2009 ATA guideline.All patients were denied the history of diabetes related with glucose signaling.Data were retrospectively collected, including demographic information, laboratory findings, SUV max , www.nature.com/scientificreports/and pathology data.This study was approved by the Institutional Research and Ethics Committee at Chungnam National University Hospital (CNUH-2022-11-004-001) and conducted in accordance with the Declaration of Helsinki.Informed consent was obtained from all individual participants involved in the study.All personal identifiers were removed or disguised to ensure participant anonymity, in line with HIPAA guidelines.

Postoperative follow-up and recurrence
Patients were followed up for 8.8 ± 0.5 years (mean ± SEM).After primary treatment, all patients underwent TSH suppression therapy with thyroid hormone supplementation according to the American Thyroid Association guidelines 2 .Patients were assessed every 3 months in the first year after surgery, every 6 months for the next 2 years, and annually thereafter.Thyroid ultrasound imaging and thyroid function tests (including thyroglobulin and antithyroglobulin antibodies), were routinely performed at each follow-up consultation.Indeterminate or suspicious thyroid nodules or lymph nodes (LNs) were evaluated by fine-needle aspiration.All six structural recurrence cases were confirmed by cytological analysis, two from the operation bed and four from the lateral LNs.One patient died due to airway obstruction due to the tumor extending significantly into the mediastinum, while the remaining five patients were cured and maintained stable disease after treatment with I-131 100-150 mCi RAI or neck dissection.

RNA extraction for sequencing
To analyze the transcriptome and identify DEGs, RNA was extracted from tumor and paired non-tumor tissue samples.Thyroid samples were isolated from specimens frozen at − 80 °C immediately after thyroidectomy and homogenized using a mortar and pestle; total RNA was extracted using an RNA extraction kit (QIAGEN, Germantown, MD, USA) following the manufacturer's protocol.All experiments were conducted under clean conditions and equipment was pre-autoclaved.The quality of the extracted RNA was evaluated using the Agilent 2100 Bioanalyzer RNA Nano Chip (Agilent, Santa Clara, CA, USA).The extracted RNA was used to construct RNA libraries using the TruSeq access library or the stranded mRNA LT Sample Preparation Kit (Illumina, San Diego, CA, USA), according to the manufacturer's protocols.Library quality was analyzed using an Agilent 2100 Bioanalyzer and an Agilent DNA 1000 kit (Agilent, Santa Clara, CA, USA).Samples were sequenced on the Illumina HiSeq 2500 platform (Illumina, San Diego, CA, USA), yielding an average of 38 million paired-end 100 bp reads.

Bioinformatic transcriptome analysis
To analyze the relationship between the thyroid differentiation score (TDS) and SUV max , we used the "Com-plexHeatmap" and "corrplot" R packages with log10 transcript per million (TPM).We calculated the TDS by combining the gene set variation analysis (GSVA) package with the TDS gene list 15 .To confirm survival probability, we used the "survival" and "survminer" R packages.Additionally, we identified DEGs by subdividing our cohort into two groups based on tumor size and SUV max .Differential expression analysis was carried out in R using "DESeq2" and enrichment analysis was performed using the "fgsea" R package.Gene sets used for Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway and Gene Ontology enrichment analyses were downloaded from the Gene Set Enrichment Analysis (GSEA) website (https:// www.gsea-msigdb.org).DEGs and KEGG pathways with corrected p values < 0.05 were considered statistically significant.We assessed RFS by specific gene expression level using open database from GEPIA 2 which is based on TCGA database.

Statistical analysis
Group data for continuous variables are presented as mean ± standard deviation and, in some cases, as mean ± standard error of the mean, as noted in the footnotes.Categorical variables were presented as numbers and percentages.To compare the means of continuous variables, we used the unpaired Student's t-tests or Mann-Whitney U test.Chi-square tests or Fisher's exact test were used to compare the distributions of categorical variables, and Pearson's correlation analysis was used to evaluate the associations between tumor size and SUV max .Kaplan-Meier (K-M) survival curves were created to evaluate differences in RFS between the groups, and receiver operating characteristic (ROC) curves were used to determine the cutoff SUV max value for predicting recurrence.For tumor size-or SUV max -adjusted statistical analyses, we used ANCOVA, Mantel-Haenszel Chi-square test, and Cox regression analysis.Statistical significance for all analyses was established with a twotailed p value < 0.05.Statistical analyses/graphs were performed/created using SPSS Version 26.0.(IBM corp., Armonk, NY, USA), R, GraphPad Prism 9.

Figure 1 .
Figure 1.SUV max were not significantly correlated with GLUTs or glycolysis, however, PTCs with high SUV max are enriched low TDS score than those with low SUV max .(A) The design of our studies to verify the metabolic features by SUV max in PTC (N = 80).(B) The correlation plot between tumor size and SUV max using Pearson correlation method (r = 0.54, p < 0.0001).(C) The recurrence-free survival probability of low SUV max (n = 50, blue) and high SUV max group (n = 30, red).The p = 0.013 is from log-rank test.(D) Correlation plots of SUV max with GLUTs family and TDS.(E-H) Correlation plots of SUV max with tumor size, TDS, GSVA score of GLUTs family, and GSVA score Glycolysis.(I)Heatmap of TDS genes using log 10 TPM in our cohort.Tumors were sorted according to SUV max .(J,K) Comparison of GSVA for glycolysis or tumor differentiation score (TDS) between PTC SUV-low and PTC SUV-high tumors.(L) Expression of genes encoding SLC2A1, SLC2A2, SLC2A3, SLC2A4, SLC5A5, and TPO between PTC SUV-low and PTC SUV-high tumors.(M) Expression of TDS genes, DIO1, DUOX2, SLC26A4, SLC5A5 and TPO, between PTC SUV-low and PTC SUV-high tumors, and scatter plots of correlation between SUV max and either SLC5A5 or TPO.In the scatter plots, blue line is drawn using simple linear regression and the gray colored area indicate 95% confidence band.

Figure 2 .
Figure 2. PTCs with high SUV max are enriched with several pathways, such as DNA replication, cell cycle, pyrimidine metabolism, and one carbon pool by folate.(A) Comparison of GSVA based on unsupervised analysis using KEGG analysis between PTC SUV-low and PTC SUV-high tumors.(B-E) Scatter plots of correlation between SUV max and GSVA score of DNA replication, pyrimidine metabolism, cell cycle, or purine metabolism.In the scatter plots, blue line is drawn using simple linear regression and the gray colored area indicate 95% confidence band.(F) Heatmap of genes related upregulated pathway using log 10 TPM in our cohort.Log2 fold change and p value were calculated by the comparison of gene expressions between PTC SUV-low and PTC SUV-high tumors.

Figure 3 .
Figure 3. Survival plots of genes contributing RFS difference by SUV max and tumor size.(A-E) Kaplan-Meier plots showing RFS difference from TCGA dataset by expression level of PSG5 (A), TFF3 (B), SOX2 (C), SLC5A5 (D), and SLC5A7 (E) that contribute RFS difference by SUV max .The high expression of PSG5 (A) and low expression of TFF3 (B), SOX2 (C), SLC5A5 (D), and SLC5A7 (E) have shorter RFS.(F-H) Kaplan-Meier plots showing RFS difference from TCGA dataset by expression level of HOXD10 (F), FER1L6 (G), IFNA1 (H) which were found to contribute to RFS difference by tumor size.The high expression of HOXD10 (F) and low expression of FER1L6 (G) and IFNA1 (H) have shorter RFS.

Importance of SUV max in predicting tumor aggressiveness after adjustment of tumor size Since our data revealed the positive correlation of tumor size and SUV max , we compared to various clinic- pathologic features in relation to tumor size or SUV max to investigate the pivotal gene signature in relation to SUV max , independently on tumor size (Table
)

Table 1 .
Comparisons of clinicopathologic characteristics by SUV max group, unadjusted and tumor sizeadjusted.BMI body-mass index, DM diabetes mellitus, ECI extracapsular invasion, ETE extrathyroidal extension, C-LNM central lymph node metastasis, L-LNM lateral lymph node metastasis, LVI lymphovascular invasion, RFS recurrence-free survival, F/U follow-up.Stage was determined based on eighth edition AJCC cancer staging manual.# Not adjusted tumor size is shown.*p value < 0.05.In continuous variables, unpaired t-test is used for unadjusted comparisons and analysis of covariance (ANCOVA) is used for tumor-size adjusted comparisons, and data are presented as mean ± SEM. § Covariates of ANCOVA model are evaluated by tumor size = 1.3362.In categorical variables, Chi-square test and Mantel-Haenszel Chi-square test were used for unadjusted and tumor size-adjusted RFS comparison, respectively, and the data are presented as number (proportion in %).

Table 2 .
Comparison of clinicopathologic characteristics of PTC by SUV max and tumor size.PTMC papillary thyroid microcarcinoma, Macro-PTC macro papillary thyroid carcinoma, BMI body-mass index, DM diabetes mellitus, ECI extracapsular invasion, ETE extrathyroidal extension, C-LNM central lymph node metastasis, L-LNM lateral lymph node metastasis, LVI lymphovascular invasion, RFS recurrence-free survival, F/U follow-up.*pvalue< 0.05 in comparison of PTMCSUV-highand macro-PTC SUV-low .† p value < 0.05 in comparison of PTMC SUV-low and PTMC SUV-high .‡ p value < 0.05 in comparison of PTMC SUV-high and macro-PTC SUV-high .Data are presented as mean ± SD for continuous variables and number (proportion in %) for categorical variables.Significant values are in bold.

Table 4 .
Up -regulated KEGG in relation to SUV max and tumor size.KEGG KEGG Kyoto Encyclopedia of Genes and Genomes, SUV max maximum standardized uptake value, PTMC papillary thyroid microcarcinoma, Macro-PTC macro papillary thyroid carcinoma.*pvalue < 0.05.KEGG ID

Table 5 .
Up-regulated gene ontology (GO) biological process (GOBP) in relation to SUV max and tumor size.SUV max maximum standardized uptake value, PTMC papillary thyroid microcarcinoma, Macro-PTC macro papillary thyroid carcinoma.*p value < 0.05.