A gene methylation prognosis evaluation model for differentiated thyroid cancer and a construction method thereof

Abstract

This invention discloses a method for constructing a gene methylation prognostic assessment model for differentiated thyroid cancer, characterized by the following steps: S1, obtaining a test sample; S2, extracting and storing DNA from the test sample; S3, performing methylation analysis; S4, constructing a prognostic classification model based on DNA methylation, calculating the risk value of the prognostic model as Risk Score = 0.15411928*cg03190661 - 0.10405129*cg15676916 + 0.06108015; S5, cross-validating to evaluate performance and obtain the gene methylation prognostic assessment model. This model uses the Risk Score to assess the prognosis of differentiated thyroid cancer: low-risk group, normal follow-up is recommended; medium-risk group, follow-up time can be reduced to half; high-risk group, close follow-up is required.

Description

Technical Field

[0001] This invention belongs to the field of medical modeling, specifically a gene methylation prognostic assessment model for differentiated thyroid cancer. Background Technology

[0002] Thyroid cancer is the most common endocrine malignancy, and its incidence has been gradually increasing worldwide in recent years. Thyroid cancer can be classified into several types, including papillary thyroid cancer (PTC), medullary thyroid carcinoma (MTC), and anatomyeloid thyroid cancer (ATC). PTC is the most common type, also known as differentiated thyroid cancer, and has the best prognosis. Anatomyeloid thyroid cancer has the highest malignancy, but because it accounts for less than 2% of all thyroid cancers, it receives the most attention and is the most closely watched among thyroid cancers.

[0003] Peripherally tracheal cancer (PTC) occurs frequently in thyroid cancers. Generally, PTC patients with standard treatments, including surgery and radiotherapy, have a good prognosis. However, some PTC patients experience lymph node metastasis, in situ metastasis, or recurrence after surgical resection some time after treatment. These patients' thyroid cancers are characterized by multiple lesions, metastasis, and recurrence. Patients with these characteristics have low five-year survival rates, high mortality rates, and significantly worse prognoses than other thyroid cancer patients. Therefore, there is an urgent clinical need for tools or methods to help clinicians determine whether a patient's tumor has these related risk factors. This would aid in monitoring and treating thyroid cancer recurrence and metastasis, enabling more refined management, treatment, and monitoring of thyroid cancer patients, thereby prolonging their survival and improving their quality of life. However, currently, there is no effective method for assessing the prognosis of thyroid cancer. Besides, PTC generally has a good prognosis. Identifying high-risk groups could significantly reduce the psychological stress of low-risk PTC patients and improve their quality of life. Therefore, there is an urgent need for methods to help clinicians identify this high-risk group among PTC patients. Summary of the Invention

[0004] In response to the problems existing in the background art, this invention proposes a gene methylation prognostic assessment model for differentiated thyroid cancer and its construction method.

[0005] Technical solution:

[0006] A gene methylation prognostic assessment model for differentiated thyroid cancer is constructed through the following steps:

[0007] S1. Obtain test samples and manually grade them;

[0008] S2. Extract and store DNA from the test samples;

[0009] S3. DNA from PTC (experimental group) and normal thyroid nodules (control group) was used to construct an RRBS library and methylation analysis was performed.

[0010] S4. Construction of a prognostic classification model based on DNA methylation. The prognostic assessment model shows that most PTC patients with cg15676916 have a poor prognosis, while patients with cg03190661 have a relatively better prognosis. Therefore, the calculated risk value of the prognostic model is Risk Score = 0.15411928*cg03190661 - 0.10405129*cg15676916 + 0.06108015 (Gene methylation sequencing is a current technology, and after sequencing, all methylation sites of the sample will be available (i.e., the RRBS result). If it is necessary to specifically detect these two sites, probes and methylation panels related to these two sites can be designed to specifically detect the status of these two sites). Prognostic assessment is performed based on the Risk Score value; the higher the Risk Score value, the more dangerous the condition.

[0011] S5. Cross-validation (leave-one-out cross-validation) is used to evaluate performance and obtain a gene methylation prognostic assessment model.

[0012] Preferably, in S1, a fine-needle aspiration biopsy (FNAC) under ultrasound guidance is used to obtain the test sample.

[0013] Preferably, in S2, the extracted DNA is stored in a test tube at -80°C.

[0014] Preferably, in S3, the RRBS library is sequenced on Genome Analyzer II based on the established single-end sequencing operation steps; the raw sequencing data is filtered and evaluated; methylation-related information of cytosine is obtained, including coverage analysis, methylation analysis and DMRs analysis; the amount of methylated cytosine with a sequence depth coverage of at least 10 and covered by at least four readings is selected to determine the level of gene methylation.

[0015] Preferably, in S4, the specific steps are as follows:

[0016] S4-1. Obtain methylation data from the TCGA database of cancer genome maps, including clinical features of thyroid cancer;

[0017] S4-2. The PTC dataset (DMG-2) in the TCGA database is split into two independent datasets: one is used to merge with the RRBS dataset from the FNAC samples to build a prognostic model (DMG-3); the other dataset is used as a validation dataset (DMG-4) to validate the built model.

[0018] S4-3. Map the differentially methylated probes (DMPs) between FNAC PTC samples and normal controls to their corresponding DMGs using a string database;

[0019] S4-4. Compare these DMGs from the TCGA database with the DMGs from RRBS sequencing, and name the DMGs from the two datasets DMG-5 for building a PTC prognostic model.

[0020] S4-5. Receiver operating characteristic (ROC) curve analysis was performed on the DMP-4 data to identify statistically significant DMPs for univariate regression analysis, with survival as the dependent variable.

[0021] S4-6. The DMPs selected by univariate regression analysis are further used for multivariate regression analysis;

[0022] S4-7. Based on this, construct a prognostic classification model.

[0023] Preferably, in S5, performance is evaluated using leave-one-out cross-validation (LOOCV).

[0024] Preferably, in S4, a Risk Score in the range of [-0.04, 0) indicates a good prognosis for the patient, belonging to the low-risk factor group, and can be followed up normally; a Risk Score in the range of [0, 0.3) indicates a higher risk for the patient, belonging to the medium-risk factor group, and the follow-up period can be shortened to half; a Risk Score in the range of [0.3, 0.6] indicates a high risk of poor prognosis for the patient, belonging to the high-risk factor group, and should be closely followed up.

[0025] Beneficial effects of the present invention

[0026] This model utilizes ultrasound-guided biopsy samples from thyroid cancer patients for sequencing. The sequencing data is then combined with extensive data from thyroid cancer patients in the TCGA and GEO databases. Methylation data of thyroid cancer patients with clinical data such as recurrence, metastasis, and survival are selected for analysis and modeling to construct a prognostic model that can be used to clinically assist in assessing patient prognosis. This model uses a risk score to assess the prognosis of differentiated thyroid carcinoma and categorizes patients into three groups based on the specific risk score: low-risk, intermediate-risk, and high-risk groups. According to the model, different follow-up methods are used for different risk groups: The low-risk group has a lower risk of thyroid cancer recurrence and metastasis, a better prognosis, and can be followed up normally; the intermediate-risk group has a higher risk of thyroid cancer recurrence and metastasis, a poorer prognosis, and the follow-up period can be shortened by half; the high-risk group has a high risk of thyroid cancer recurrence and metastasis, a poor prognosis, and requires close follow-up.

[0027] This invention, through sequencing of clinical samples, reveals that the prognosis of thyroid cancer is related to abnormal gene methylation in patients, with decreased abnormal gene methylation closely associated with the occurrence and development of thyroid cancer. Currently, it is believed that although surgery can remove the tumor area, the patient's cells, especially those near the tumor, still retain epigenetic memory, and these epigenetic imprints are key factors leading to tumor recurrence, metastasis, and poor prognosis. This model was trained on a large number of real-world cases in clinical settings and databases, and validated in an independent database, demonstrating a good ability to predict the prognosis of thyroid cancer patients. This allows high-risk PTC patients to receive closer attention and more timely treatment. For low-risk individuals, this model can also reduce their psychological burden and stress, while avoiding other burdens such as financial pressure from overtreatment. Psychological factors have a very complex and profound impact on physical health. Based on the prognostic assessment results of this model, it can reduce anxiety about the prognosis of low-risk thyroid differentiated carcinoma patients, thereby improving their quality of life and promoting their physical recovery. For patients with intermediate- to high-risk thyroid differentiated carcinoma, closer follow-up can be conducted to minimize the risk of disease progression due to delayed diagnosis.

[0028] The mechanisms of tumor development and progression are complex, and the treatment and prognosis management of cancer patients should also be comprehensive. Clinicians can combine the scores from this model with other clinical data to comprehensively assess the patient's condition, thereby identifying the most suitable treatment and follow-up approach, minimizing patient stress, improving quality of life, strengthening postoperative prognosis monitoring, preventing tumor recurrence, and ensuring timely detection and treatment to improve patient survival. Attached Figure Description

[0029] Figure 1 This is a schematic diagram of the application process of the prognostic model.

[0030] Figure 2 Bar chart showing enrichment analysis of PTC vs Normal.

[0031] Figure 3 This is a graph showing the results of an analysis of DMG based on DisGeNET in the Metascape database.

[0032] Figure 4 A graph showing the difference between FNAC samples from thyroid cancer and normal tissue.

[0033] Figure 5 This is a graph showing the difference between thyroid cancer samples and the normal control group in the training set data of the TCGA database.

[0034] Figure 6 This is a schematic diagram of the intersection of DMG-3 and DMG-1 corresponding to the differentially methylated sites in the training set of TCGA.

[0035] Figure 7 A box plot showing the methylation levels of genes in the test set.

[0036] Figure 8 This is a schematic diagram of the survival analysis for the test set.

[0037] Figure 9 This is the ROC curve for the test set.

[0038] Figure 10 A bounding box plot to validate the methylation levels of genes in the dataset.

[0039] Figure 11 A schematic diagram illustrating the survival analysis used to validate the dataset.

[0040] Figure 12 To verify the ROC curve of the dataset. Detailed Implementation

[0041] The present invention will be further described below with reference to embodiments, but the scope of protection of the present invention is not limited thereto:

[0042] Combination Figure 1 The entire process mainly includes ultrasound-guided thyroid nodule biopsy, pyrosequencing, and calculation of risk values ​​and classification based on the sequencing results, as detailed in the technical solution.

[0043] Example 1

[0044] 1. Experimental Methods

[0045] (1) After approval by the hospital's ethics committee and with the patient's consent, samples were collected from patients who underwent ultrasound-guided fine-needle aspiration biopsies, hereinafter referred to as biopsy samples. DNA was extracted from the biopsy samples using a Qiagen DNA kit and stored frozen at -80°C. Patients were followed up for five years, and detailed clinical data, including thyroid cancer recurrence, metastasis, mode and location of metastasis, and pathology after the second surgery, were recorded for future analysis. Biopsy samples from patients who experienced thyroid cancer recurrence or metastasis during the later follow-up period and samples from thyroid cancer patients who did not experience recurrence or metastasis and had a good prognosis during the follow-up period were selected as the research subjects. At the same time, the DNA concentration in the samples should reach 30 ng / ul, and the purity OD260 / 280 ≥ 1.8 to ensure the quality of the sequencing samples and the accuracy of the results.

[0046] (2) The samples selected in the above steps were subjected to methylation sequencing using pyrosequencing, the current gold standard for methylation sequencing. The sequenced samples were analyzed using a Qiagen Q48 pyrosequencing system for highly accurate quantitative methylation analysis, detecting the methylation level of gene promoter regions (transcription start site ± 1K). The sequencing reference was version hg38 (Homo sapiens genome assembly GRCh38-NCBI-NLM (nih.gov)). The screening threshold for differentially methylated genes was |logFC|>1&p.value<0.05. The DMGs obtained from the sequencing of the experimental samples were labeled DMG-1.

[0047] (3) Methylation-related data, clinical data, and survival data of thyroid cancer were downloaded from the TCGA database from UCSC Xena (https: / / xenabrowser.net / datapages / ). Methylation-related data of tumor samples with clinical and prognostic information were retained. After selection, data from 498 samples (DMP-2) were included in the analysis. Prognosis was performed on tumor samples from cancer patients. The thyroid cancer tumor data downloaded from TCGA were randomly divided into a training set DMP-3 (249) and a validation set DMP-4 (249) in a 5:5 ratio. The training set and normal samples (64) were subjected to differential analysis again according to step 2.4, and the DMP screening threshold was the same. The intersection of the DMG corresponding to the DMP annotations selected in the training set and the DMG-1 obtained from the methylation sequencing data was taken to obtain DMG-5, which was used to construct the prognostic model.

[0048] (5) Survival analysis, univariate regression, multivariate regression, and stepwise regression analysis. Perform survival analysis on the methylation sites DMP corresponding to DMG-5 in the training set separately. For each DMP, first use the survminer package to calculate the optimal cutoff. Those with >optimal cutoff are regarded as hypermethylation, and those with <optimal cutoff are regarded as hypomethylation. After grouping, perform survival analysis, and select the DMPs with significant survival analysis (p<0.05) for subsequent analysis. Then perform univariate regression analysis on the DMPs with significant survival analysis results (taking the survival outcome as the dependent variable) to screen the DMPs significantly related to prognosis. After univariate regression analysis, further perform multivariate regression on the DMPs significant in univariate regression analysis (taking the survival outcome as the dependent variable); then perform stepwise regression analysis to obtain the independent variables and coefficients when the final AIC value is the smallest, which is used as the optimal and simplest prognostic model we obtained. After obtaining the prognostic model, calculate the Risk Score using the model, and use survival analysis, Risk Score grouped box plot, and ROC methods to verify the prognostic model respectively.

[0049] (6) Construct a prognostic risk model based on the optimal DMP combination. We use the model to recalculate the Risk Score in the training set in the validation set, and then use survival analysis, Risk Score grouped box plot, and ROC methods in the validation set to verify the prognostic model respectively, to see if the corresponding results obtained in the training set are basically the same, and complete the verification of the prognostic model by the validation set.

[0050] (7) Finally, calculate the risk coefficient of the patient according to the Risk score value, and judge the interval where the risk coefficient is located and the risk level of the risk time of recurrence, metastasis, etc. after surgery.

[0051] 2. Experimental results

[0052] After pyrosequencing, compared with the DNA methylation levels of thyroid cancer and normal control groups in the samples of fine needle aspiration biopsy under ultrasound guidance, it is related to a variety of thyroid diseases ( Figure 2 showing the sequencing results, that is, the differentially methylated sites are closely related to many thyroid function activities), and too low methylation levels of specific genes may be related to thyroid cancer ( Figure 3 showing that the differentially methylated sites are associated with many thyroid diseases and highly related to thyroid cancer). The sequencing data show that the abnormal methylation of genes is closely related to thyroid cancer, and the gene methylation level in the thyroid cancer group is significantly lower than that in the normal group ( Figure 4The figure shows differences in gene methylation levels across the entire genome, with the tumor group exhibiting lower levels of gene methylation compared to the normal control group. Analysis of methylation levels in thyroid cancers with recurrent and metastatic characteristics from the TCGA database, compared to the normal control group, also showed that abnormal methylation of specific genes is closely related to the occurrence and development of thyroid cancer. Figure 5 The results of differential analysis between thyroid cancer samples and normal controls in the training set data of the TCGA database are consistent with the results of sequencing samples, showing a decrease in methylation levels in the tumor group. The DMG (dimethylation gene) obtained by intersecting the data used to build the prognostic model in the TCGA database with the pyrosequencing data is also shown. Figure 6 The figure shows the intersection of DMG-3 and DMG-1 corresponding to differentially methylated sites in the TCGA training set. As can be seen from the figure, DMG-1 has 749 sites, and DMG-3 has 974 sites. The intersection yields 31 DMG sites, meaning these 31 DMG sites can accurately distinguish between tumors with high-risk factors and specimens from tumor patients with better prognoses. The prognostic effect of the prognostic model on the test set is shown in a box plot. Figure 7 In the test dataset, the gene methylation level bounding box plot showed that the prognostic model could distinguish between patients with poor prognosis and those with good prognosis who did not die, demonstrating good prognostic efficacy. Survival plot ( Figure 8 In the test dataset, survival analysis showed that the prognostic model significantly differentiated between patients with poor and good prognoses in thyroid cancer patients, demonstrating a good prognostic effect. (ROC curves were also analyzed.) Figure 9 The ROC curves show that the prognostic model demonstrated good sensitivity and specificity in distinguishing patients with poor prognosis in the test dataset for thyroid cancer, exhibiting a good ability to differentiate patients with clinical characteristics such as recurrence and metastasis. The prognostic effect of the prognostic model on the validation dataset in the TCGA database is illustrated by a box plot. Figure 10 In the validation dataset, the methylation level bounding box plot shows that the prognostic model can also distinguish between patients with poor prognosis and those with good prognosis who do not die in thyroid cancer patients, validating the prognostic effect of the model. Survival plot ( Figure 11 Survival analysis showed that the prognostic model could effectively group patients with poor prognosis and those with good prognosis in the validation dataset, validating the prognostic effect of the model. (ROC curves were also observed.) Figure 12 The ROC curves show that the prognostic model demonstrated good sensitivity and specificity in distinguishing poor prognoses among thyroid cancer patients on the validation dataset, validating the good sensitivity and specificity of the prognostic model.

[0053] Example 2

[0054] Based on the above research findings, this invention employs machine learning algorithms to provide a relatively objective and cost-effective method for evaluating the prognosis of thyroid cancer by calculating the numerical values ​​of differentially methylated gene sites. This method can be used as a reference for follow-up surgery and treatment selection in differentiated thyroid cancer. Specifically, a thyroid cancer prognostic assessment model based on pyrosequencing and machine learning algorithms calculates the patient's gene methylation status according to the patient's sequencing results using the Risk Score formula: Risk Score = 0.15411928*cg03190661 - 0.10405129*cg15676916 + 0.06108015. In the formula, Risk Score represents the risk value; 0.15411928 is the weight of the differentially methylated site cg03190661 in the model; cg03190661 represents the value of this differentially methylated site; -0.10405129 is the weight of the differentially methylated site cg15676916 in the model; cg15676916 represents the value of this differentially methylated site; and 0.06108015 is a constant. The corresponding risk score for each patient is calculated using this formula, and then patients are classified according to the risk score interval. A Risk Score in the range [-0.04, 0) indicates a good prognosis, requiring only normal follow-up; a score in the range [0, 0.3) indicates a higher risk of recurrence and metastasis, requiring a reduction in follow-up days to half; and a score in the range [0.3, 0.6] indicates the highest risk of poor prognosis, requiring close follow-up. Clinicians can combine the prognostic model scores with other clinical data of the patient, such as imaging images and treatment status, to jointly determine the patient's monitoring and treatment plan.

[0055] Example 3

[0056] Example of thyroid cancer detection

[0057] Following the method described in Example 1, the patient's gene methylation data were obtained. The risk of poor prognosis, such as recurrence and metastasis of thyroid cancer, was calculated using the formula: Risk Score = 0.15411928*cg03190661 - 0.10405129*cg15676916 + 0.06108015. An example calculation is shown below.

[0058] Sample 1 (thyroid cancer biopsy sample with no recurrence or metastasis observed during post-operative follow-up and a good five-year survival): cg03190661 = -0.34562; cg15676916 = 0.255776, therefore, Risk Score = 0.15411928*cg03190661 - 0.10405129*cg15676916 + 0.06108015 = 0.15411928*-0.34562 - 0.10405129*0.255776 + 0.06108015 = -0.018800378. The diagnostic model Risk for patients in Sample 1 is... The score was -0.018800378, which falls within the range of [-0.04, 0), indicating a low-risk group. This means that the patient's thyroid cancer has a good prognosis after treatment and only requires normal follow-up, consistent with clinical results.

[0059] Sample 2 (Thyroid biopsy sample from a patient with recurrent thyroid cancer before the first surgery):

[0060] cg03190661=0.253861; cg15676916=-0.29211, therefore Risk Score=0.15411928*0.253861-0.10405129*(-0.29211)+0.06108015=0.15411928*(-0.34562)-0.10405129*0.255776+0.06108015=0.130599447. The prognostic model risk value of patients in Sample 2 is 0.130599447, which falls within the [0, 0.3) interval, indicating a medium-risk group. This suggests that the patients have a high risk of recurrence and metastasis. Shortening the follow-up period and timely diagnosis and treatment for any discomfort is consistent with the clinical outcome.

[0061] Sample 3 (Thyroid cancer biopsy sample from a patient with postoperative thyroid cancer lymph node metastasis before the first surgery):

[0062] cg03190661=-0.30545; cg15676916=-0.27328, therefore the Risk Score=0.15411928*cg03190661-0.10405129*cg15676916+0.06108015=0.15411928*(-0.30545)-0.10405129*(-0.27328)+0.06108015=0.042439552. The risk value of the prognostic model for patients in Sample 3 is 0.042439552, which falls within the [0, 0.3) interval, indicating a medium-risk group. This suggests that the patients have a high risk of recurrence and metastasis. When the follow-up period is shortened, the predictive results of the prognostic model are consistent with the clinical outcome.

[0063] Table 1

[0064]

[0065] ※ Risk Score=0.15411928*cg03190661-0.10405129*cg15676916+0.06108015

[0066] The specific embodiments described herein are merely illustrative of the spirit of the invention. Those skilled in the art to which this invention pertains may make various modifications or additions to the described specific embodiments or use similar methods to substitute them, without departing from the spirit of the invention or exceeding the scope defined by the appended claims.

Claims

1. A method for constructing a gene methylation prognostic assessment model for differentiated thyroid cancer, characterized in that... It includes the following steps: S1. Obtain test samples and manually grade them; S2. Extract and store DNA from the test samples; S3. DNA containing PTC and normal thyroid nodules was used to construct an RRBS library and methylation analysis was performed. S4. Construction of a prognostic classification model based on DNA methylation. The calculated risk score of the prognostic model is Risk Score = 0.15411928*cg03190661 - 0.10405129*cg15676916 + 0.06108015. Prognostic assessment is performed based on the Risk Score, with higher Risk Score indicating greater risk. S5. Cross-validation was used to evaluate the performance and obtain a gene methylation prognostic assessment model.

2. The method according to claim 1, characterized in that... In S1, a fine-needle aspiration biopsy (FNAC) under ultrasound guidance was used to obtain the test sample.

3. The method according to claim 1, characterized in that... In S2, the extracted DNA was stored in a test tube at -80°C.

4. The method according to claim 1, characterized in that... In S3, the RRBS library was sequenced on Genome Analyzer II based on the established single-end sequencing procedure; the raw sequencing data was filtered and evaluated; methylation-related information of cytosine was obtained, including coverage analysis, methylation analysis and DMRs analysis; the amount of methylated cytosine with a sequence depth coverage of at least 10 and covered by at least four reads was selected to determine the level of gene methylation.

5. The method according to claim 1, characterized in that... In S4, the specific steps are as follows: S4-1. Obtain methylation data from the TCGA database of cancer genome maps, including clinical features of thyroid cancer; S4-2. The PTC dataset in the TCGA database is split into two independent datasets: one is used to merge with the RRBS dataset from the FNAC samples to build a prognostic model; the other dataset is used as a validation dataset to validate the built model. S4-3. The differentially methylated probe DMP between FNAC PTC samples and normal controls was mapped to its corresponding DMG through a string database and labeled as DMG-1. S4-4. Label all DMGs corresponding to methylation data from the TCGA database as DMP-2, with the DMGs corresponding to the validation set as DMG-3 and DMGs corresponding to the validation set as DMG-4. Compare DMG-3 with DMG-1 from RRBS sequencing, and name the DMGs that are present in both datasets as DMG-5, which will be used to construct the PTC prognostic model. S4-5. Perform receiver operating characteristic (ROC) curve analysis on the DMP-5 data to identify statistically significant DMPs for univariate regression analysis, with survival as the dependent variable. S4-6. The DMPs selected by univariate regression analysis are further used for multivariate regression analysis; S4-7. Based on this, construct a prognostic classification model.

6. The method according to claim 1, characterized in that... In S5, performance is evaluated using leave-one-out cross-validation (LOOCV).

7. The method according to claim 1, characterized in that... In S4, a Risk Score in the range of [-0.04, 0) indicates a good prognosis for the patient, belonging to the low-risk group, and can be followed up normally; a Risk Score in the range of [0, 0.3) indicates a higher risk for the patient, belonging to the medium-risk group, and the follow-up period can be shortened to half; a Risk Score in the range of [0.3, 0.6] indicates a high risk of poor prognosis for the patient, belonging to the high-risk group, and should be closely followed up.

8. A gene methylation prognostic assessment model for differentiated thyroid cancer, characterized in that... The model is constructed by the method described in any one of claims 1-7.

Images