CircRNA biomarker, primer pair and kit for forensic medicine PMI inference and application of circRNA biomarker, primer pair and kit

Abstract

The invention discloses a circRNA biomarker for forensic medicine PMI inference, a primer pair, a kit and application thereof, and relates to the technical field of gene engineering. The circRNA biomarker is human and mouse circFat3 or a variant with sequence homology greater than or equal to 87%; the primer pair comprises a divergent primer pair and a convergent primer pair, wherein an upstream primer sequence and a downstream primer sequence of the divergent primer pair are as shown in SEQ ID NO: 1 and SEQ ID NO: 2; the upstream primer sequence and the downstream primer sequence of the convergence primer pair are as shown in SEQ ID NO: 3 and SEQ ID NO: 4. The invention further provides a kit for specifically detecting the circFat3 gene and application of the kit in human forensic medicine PMI inference. The circRNA circFat3 is determined as a novel biomarker for PMI inference for the first time, the problems that a traditional forensic medicine method is greatly interfered by the environment and is high in subjectivity are solved through matched primers, a kit and an AI model, and high-precision and standardized molecular technical support is provided for criminal case investigation and civil judicial expertise.

Description

Technical Field

[0001] The present invention relates to the field of genetic engineering technology, and specifically relates to a circRNA biomarker, a primer pair, a kit and applications thereof for forensic PMI inference. Background Art

[0002] In forensic medicine, time of death refers to the time elapsed since death (TSD) or postmortem interval (PMI), which is the time interval between the examination of a body and the occurrence of death. Estimating time of death has always been a crucial issue in forensic identification, but accurate methods are still lacking. For decades of research, forensic scientists have primarily used various postmortem changes to estimate time of death. For example, traditional methods commonly used to estimate PMI in autopsy specimens include early postmortem estimation, including coldness, livor mortis, rigor mortis, corneal opacity, and gastric digestion; and late postmortem estimation, including the degree of decomposition, forensic entomology, and forensic anthropology. However, these methods are susceptible to environmental and individual variability and are prone to false positives. Furthermore, the lack of standardized protocols and the reliance on subjective judgment reduce the accuracy and reliability of identification results. For example, using cold body measurements becomes ineffective when the body temperature drops to room temperature.

[0003] In recent years, circular RNA (circRNA) has emerged as a new research focus for PMI due to its unique biological properties. CircRNA is a closed circular structure formed by reverse shearing of precursor messenger RNA (pre-mRNA). It lacks a 5' cap and a 3' poly(A) tail, making it resistant to degradation by the exonuclease RNase R. Its half-life exceeds 48 hours (far exceeding the 10 hours of mRNA) and it remains intact in postmortem tissue, demonstrating its high stability. Notably, circRNA is most abundantly expressed and highly conserved in human brain tissue. The closed intracranial environment reduces external interference from corruption, making it an ideal specimen. Existing studies have shown that circRNA can act as a miRNA sponge to regulate gene expression, and preliminary exploration has been made in forensic fields such as age inference and biological specimen traceability. However, research on its application to PMI inference remains scarce.

[0004] In addition, in the estimation of time of death, it is also very important to discover ideal endogenous reference genes, which should have stable expression levels. However, there are few studies on this topic, and only sporadic reports are available, such as Tu et al. [Tu C, Du T, Shao C, et al. Evaluating the potential of housekeeping genes, rRNAs, snRNAs, microRNAs and circRNAs as reference genes for the estimation of PMI[J]. Forensic Sci MedPathol, 2018, 14(2): 194-201] The stability of tissue-specific reference genes was evaluated by geNorm and NormFinder algorithms, and the reference genes suitable for PMI inference were screened out, including LC-Ogdh, circ-AFF1 and miR-122 in liver tissue and miR-133a, circ-AFF1 and LC-LRP6 in skeletal muscle tissue. Subsequently, LC-Ogdh, circ-AFF1 and miR-122 as internal reference markers for liver tissue, and miR-133a and circ-AFF1 as internal reference markers for skeletal muscle tissue were constructed. Hao et al. [Hao Erwa, Wu Yue, Yang Yingzhong, et al. Degradation pattern of circular RNA after death in mice and its applicability for estimating the time of death [J]. Chinese Journal of Plateau Medicine and Biology, 2019, 40(2): 105-109] found that CDR1ascircRNA is the most suitable internal reference gene, and PMI at 4°C can affect the relative expression of circRNA more than that at 20°C. Dong et al. [Dong Tingting, Xiao Zuorun, Liu Zengjia et al. Preliminary screening and validation study of skeletal muscle circRNA in the estimation of time of death [J]. China Forensic Identification, 2022 (1): 85-92.] used chip hybridization analysis and real-time fluorescence quantitative PCR technology to screen and identify five circRNAs (has_circ_0001727, hsa_circ_0001136, hsa_circ_0001871, hsa_circ_0005251, hsa_circ_0000284) in human skeletal muscle samples that decreased over time. However, the few studies on circRNAs mentioned above have not found effective markers to construct a model for estimating time of death that can be applied in practice. Summary of the Invention

[0005] The present invention aims to overcome the shortcomings of the existing technology and provide a circRNA biomarker, primer pair, kit and application thereof for forensic PMI inference. The invention establishes circRNA circFat3 as a new biomarker for PMI inference for the first time. Through the matching primers, kit and AI model, the problems of traditional forensic methods being greatly affected by environmental interference and strong subjectivity are solved, providing high-precision, standardized molecular technology support for criminal case investigation and civil judicial identification.

[0006] To achieve the above object, the technical solution adopted by the present invention is as follows:

[0007] A circRNA biomarker for forensic PMI inference, wherein the circRNA biomarker is human and mouse circFat3 genes or variants thereof having ≥87% sequence homology.

[0008] The present invention also provides a primer pair for specifically detecting the circFat3 gene, wherein the primer pair comprises a divergent primer pair and a convergent primer pair, wherein:

[0009] The upstream and downstream primer sequences of the divergent primer pair are shown in SEQ ID NO: 1 and SEQ ID NO: 2;

[0010] The upstream and downstream primer sequences of the convergent primer pair are shown in SEQ ID NO: 3 and SEQ ID NO: 4.

[0011] Furthermore, the present invention also provides a kit for specifically detecting the circFat3 gene, the kit comprising the above primer pair and an internal reference gene primer, wherein the internal reference gene is mt-co1 and / or 28S rRNA.

[0012] Furthermore, when the internal reference gene is mt-co1, the upstream sequence of the detection primer is shown in SEQ ID NO: 21, and the downstream sequence of the detection primer is shown in SEQ ID NO: 22; when the internal reference gene is 28S rRNA, the upstream sequence of the detection primer is shown in SEQ ID NO: 11, and the downstream sequence of the detection primer is shown in SEQ ID NO: 12.

[0013] The present invention also provides application of the above kit in forensic PMI inference.

[0014] Furthermore, the forensic PMI inference method includes the following steps:

[0015] S1, total RNA extracted from postmortem brain tissue samples;

[0016] S2, reverse transcription of total RNA into cDNA;

[0017] S3, the expression level of circFat3 in the samples was detected by semiquantitative RT-PCR or RT-qPCR using divergent primer pairs and convergent primer pairs;

[0018] S4, relative expression of circFat3 was calculated using mt-co1 and / or 28S rRNA as internal reference genes;

[0019] S5, based on the relative expression levels, inferring the PMI by constructing a mathematical model or an AI model.

[0020] Furthermore, the processing of semi-quantitative RT-PCR test results included: using image analysis software to calculate the grayscale value of the amplified band and subtracting the background grayscale value; normalizing the grayscale values ​​of different agarose gel plates using standards; after subtracting the background grayscale value, the ratio of the average grayscale value of circFat3 to the average grayscale value of the reference gene was used to determine the relative expression level of circFat3;

[0021] The processing of RT-qPCR detection results includes: calculating the ΔCt value based on the Ct value, where ΔCt is the Ct value of circFat3 minus the Ct value of the internal reference gene, and the ΔCt value is the relative expression level.

[0022] Furthermore, when the mathematical model is a linear equation, a quadratic equation or a cubic equation, its independent variable is PMI and the dependent variable is the relative expression level of circFat3; when the mathematical model is a three-dimensional model, its independent variables are PMI and temperature, and the dependent variable is the relative expression level of circFat3, and the relative expression level of circFat3 is as described above.

[0023] Furthermore, the AI ​​model is a support vector machine (SVM) regression model, and an application software that can be used for PMI inference at multiple temperatures is constructed, wherein the regression model is:

[0024]

[0025] Where x = new input sample with the following parameters: circFat3, expression level, temperature; i = support vector, used to define the training samples of the decision boundary; α i ,αi * = dual coefficient, the Lagrange multiplier from the support vector machine model; b = bias term, i.e. intercept; K(x i ,x) = radial basis function kernel, specifically:

[0026] K(x i , x) = exp(-γ·||x i -x||2 )

[0027] Among them, γ is used to control the influence range of each support vector.

[0028] Compared with the prior art, the present invention has the following beneficial effects:

[0029] (1) This invention is the first to use the circFat3 gene for PMI inference. The circFat3 gene exhibits significant regular degradation characteristics after the death of an organism, and its expression level is highly correlated with the time of death (PMI). Experimental data show that under six different temperature conditions (4°C to 35°C), the expression level of this gene in mouse brain tissue decreases in stages as the PMI prolongs (e.g., stable within 8 days at 4°C and stable within 1 day at 25°C), and the homology between humans and mice reaches 87%, indicating that it can be used as a reliable molecular marker for cross-species PMI inference. This specific degradation overcomes the problems of traditional forensic methods (such as livor mortis and rigor mortis) that are subject to significant environmental interference and strong subjectivity.

[0030] (2) The closed circular structure of circFat3 in the present invention effectively resists degradation by the exonuclease RNase R, degrading slowly in decaying tissues with a half-life significantly longer than that of mRNA. Furthermore, the selected internal reference genes mt-co1 (mitochondrial gene) and 28S rRNA are stably expressed in postmortem tissues and under various temperature conditions, effectively reducing the interference of environmental factors on the results and enhancing the reliability of PMI inference.

[0031] (3) When using the kit provided by the present invention for PMI inference, accurate quantification of circFat3 expression was achieved through standardized procedures (RNA extraction, reverse transcription, semiquantitative / quantitative PCR detection) and data processing methods (grayscale value analysis, ΔCt calculation). The combination of linear, nonlinear, and three-dimensional mathematical models significantly improved the accuracy of PMI inference.

[0032] (4) Based on semi-quantitative RT-PCR and RT-qPCR data, the present invention constructed a mathematical model (linear / nonlinear regression equation and three-dimensional model) and an AI model (SVM regression) covering a wide temperature range (4°C-35°C), realizing the objective quantitative inference of PMI and providing a standardized operation tool for forensic practice.

[0033] (5) The PMI inference method provided by this invention is applicable to autopsy in different temperature environments, overcoming the limitation of traditional methods (such as the cadaver cooling method) that fail after the body temperature drops to room temperature. After verification in a mouse model, it was further successfully applied to human brain tissue, with important application value in criminal case investigation, civil litigation, and other scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 The expression levels of circFat3 in various mouse tissues based on the circAtlas database;

[0035] Figure 2 The expression of circFat3 host genes in various mouse tissues based on the NCBI database;

[0036] Figure 3 The electrophoresis results of circFat3 in 11 different mouse tissues;

[0037] Figure 4 is the relative expression level of circFat3 in 11 different mouse tissues;

[0038] Figure 5 The expression of circFat3 in various human tissues in the circAtlas database;

[0039] Figure 6 The expression levels of the FAT3 gene in various human tissues in the GTEx database;

[0040] Figure 7 A is the electrophoresis analysis of circFat3 convergent primers and divergent primers in cDNA and gDNA samples; Figure 7 B is the analysis results of circFat3, linear Fat3 and Gapdh in the ddH2O group, 0U group and 1U group after RNase R digestion;

[0041] Figure 8 Data analysis of the relative expression levels of circFat3, linear Fat3, and Gapdh after RNase R digestion;

[0042] Figure 9 The structural characteristics and Sanger sequencing results of circFat3;

[0043] Figure 10 is the expression level of candidate reference genes in brain tissue at 35°C at 0, 12, 24, 48, 72, and 96 hours after death;

[0044] Figure 11 is the quantitative analysis result of the relative expression of candidate internal reference genes;

[0045] Figure 12 The expression levels of mt-co1, 28S rRNA, and circFat3 in mouse brain tissue at 4°C at 0, 1, 2, 4, 8, 12, and 16 days after death;

[0046] Figure 13A is the linear, quadratic, and cubic equations at 4 °C established with circFat3 and mt-co1; Figure 13 B is the linear, quadratic, and cubic equations established at 4°C using circFat3 and 28S rRNA;

[0047] Figure 14 The electrophoresis results of the 4°C model validation samples (days 2, 6.25, and 9.25) are shown. The different lanes in the figure are from different areas of the same gel plate and are separated by blank spaces.

[0048] Figure 15 The expression levels of mt-co1, 28S rRNA, and circFat3 in brain tissue at 0, 1, 2, 4, and 8 days after death at 25°C;

[0049] Figure 16 A represents the linear, quadratic, and cubic equations at 25°C established using circFat3 and mt-co1; Figure 16 B is the linear, quadratic, and cubic equations at 25°C established using circFat3 and 28S rRNA;

[0050] Figure 17 The electrophoresis results of the 25°C model validation samples (1.5 days and 4 days) are shown. The different lanes in the figure are from different areas of the same gel plate and are separated by blank spaces.

[0051] Figure 18 The expression levels of mt-co1, 28S rRNA, and circFat3 in mouse brain tissue at 35°C at 0, 12, 24, 48, 72, and 96 hours after death;

[0052] Figure 19 A is the linear, quadratic, and cubic equations at 35°C established with circFat3 and mt-co1; Figure 19 B is the linear, quadratic, and cubic equations at 35°C established using circFat3 and 28S rRNA;

[0053] Figure 20 Electrophoresis results of the 35°C model validation samples (24, 30, and 60 h);

[0054] Figure 21 The expression levels of mt-co1, 28S rRNA, and circFat3 in mouse brain tissue at 15°C at 0, 25, 49, 97, 193, and 289 hours after death;

[0055] Figure 22 A is the linear, quadratic, and cubic equations at 15 °C established with circFat3 and mt-co1; Figure 22B is the linear, quadratic, and cubic equations at 15°C established using circFat3 and 28S rRNA;

[0056] Figure 23 The expression levels of mt-co1, 28S rRNA, and circFat3 in mouse brain tissue at 0, 26, 50, 98, 146, and 241 hours after death under 20°C;

[0057] Figure 24 A is the linear, quadratic, and cubic equations established at 20 °C using circFat3 and mt-co1; Figure 24 B is the linear, quadratic, and cubic equations at 20°C established using circFat3 and 28S rRNA;

[0058] Figure 25 The expression levels of mt-co1, 28S rRNA, and circFat3 in mouse brain tissue at 30°C at 0, 10, 27, 51, 99, and 147 hours after death;

[0059] Figure 26 A is the linear, quadratic, and cubic equations at 30 °C established with circFat3 and mt-co1; Figure 26 B Linear, quadratic, and cubic equations at 30°C established using circFat3 and 28S rRNA;

[0060] Figure 27 A and 27B are the melting curves of circFat3 and mt-co1, respectively;

[0061] Figure 28 A and 28B are the amplification curves of circFat3 and mt-co1, respectively;

[0062] Figure 29 The linear, quadratic, and cubic equations at 35 °C were constructed based on the ΔCt values ​​of circFat3 and mt-co1;

[0063] Figure 30 A is a three-dimensional inference model constructed based on the internal reference gene 28S rRNA and circFat3; Figure 30 B is a three-dimensional inference model constructed based on the internal reference genes mt-co1 and circFat3;

[0064] Figure 31 The expression levels of 28S rRNA and human circFat3 in human brain tissue at 3.5, 18.5, 41.5, 64.5, 113.5, and 161.5 hours after death at 20°C.

[0065] Figure 32The expression levels of 28S rRNA and human circFat3 in human brain tissue at 25°C at 3.5, 19.5, 42.5, 65.5, 114.5, and 162.5 hours after death;

[0066] Figure 33 A represents the linear, quadratic, and cubic equations at 20°C established using human circFat3 and 28S rRNA; Figure 33 B is the linear, quadratic, and cubic equations at 25°C established using human circFat3 and 28S rRNA;

[0067] Figure 34 Application software built for PMI inference;

[0068] Figure 35 A graph showing the prediction results of the human AI model for transfer learning;

[0069] Figure 36 This is a detailed diagram of transfer learning. DETAILED DESCRIPTION

[0070] The present invention will be further described below with reference to the accompanying drawings and examples. The embodiments of the present invention include but are not limited to the following examples.

[0071] This example provides a circRNA biomarker circFat3, specifically as follows:

[0072] 1. Screening of brain tissue-specific human and mouse homologous circRNAs

[0073] CircRNA screening was conducted based on existing circRNA databases, including circBase (http: / / circbase.org / ) and circBank (http: / / www.circbank.cn / ), as well as relevant literature containing human and mouse circRNA sequencing data. First, circRNAs with high expression and strong tissue specificity in human and mouse brain were identified from relevant literature. These circRNAs were then retrieved from circBase, and their circBase IDs were used to retrieve information about the human and mouse homologous circRNAs from circBank, including RNA sequence, genomic location, and length. The parental genes of the identified circular transcripts were queried in the GTEx (https: / / www.gtexportal.org / home / index.html) and NCBI (https: / / www.ncbi.nlm.nih.gov / gene / ) databases to assess the expression of their host genes in various human and mouse tissues. Finally, the chromosomal location of the circRNAs was used to query the expression profiles of the circular transcripts in different human and mouse tissues using the circAtlas (https: / / ngdc.cncb.ac.cn / circatlas / index.php) database.

[0074] 2. circRNA Primer Design

[0075] Primers were designed based on sequences flanking the circular RNA splice site. Convergent primers were designed using the linear transcript corresponding to the circRNA, while divergent primers were designed based on the circRNA backsplicing sequence. The backsplicing sequence typically combines sequences at the 3' end of the linear transcript with upstream sequences at the 5' end, forming a sequence containing the splice site. These primers were designed using Primer 3.0 (https: / / bioinfo.ut.ee / primer3-0.4.0 / ) and the NCBI database (https: / / blast.ncbi.nlm.nih.gov / Blast.cgi). The specificity of the initial primers was verified using the Primer-BLAST tool (https: / / www.ncbi.nlm.nih.gov / tools / primer-blast / ). CircPrimer 2.0 software was used to verify the specificity of the divergent primers and display their sequence locations within the circRNA. The designed primers were synthesized by Beijing Qingke Biotechnology Co., Ltd.

[0076] Based on the sequencing data from the above-mentioned literature, we finally identified the circRNA molecule mmu_circ_0001746 (circFat3) that is highly expressed in brain tissue and has tissue specificity, and designed the corresponding primers (as shown in Table 1). It is worth noting that circFat3 and its host gene showed high expression levels in mouse brain tissue ( Figure 1 、 Figure 2 ). PCR amplification and agarose gel electrophoresis were then performed to verify that circFat3 was highly specific in mouse brain tissue ( Figure 3 The relative expression levels of circFat3 in 11 different tissues were calculated by semi-quantitative RT-PCR ( Figure 4 ), the results showed that circFat3 has high expression and specificity in mouse brain tissue. Database data showed that the human homologous hsa_circ_0000348 (human circFat3) also has high expression characteristics in human brain tissue ( Figure 5 、 Figure 6 In addition, the sequence information of mmu_circ_0001746 and hsa_circ_0000348 was obtained from the circBase database, with full lengths of 3306 bp and 3309 bp, respectively. Homology analysis showed that these circRNAs are highly conserved between humans and mice (87%).

[0077] Table 1. Primer information

[0078]

[0079] 3. Collection of Mouse and Human Samples

[0080] All animal experiments were conducted in accordance with international, national, and institutional animal care guidelines. This experiment has been reviewed and approved by the Ethics Committee of Southwest Medical University, Sichuan, China. The mouse model was constructed using 9-week-old BALB / c male SPF-grade mice (weight: 23-26g). According to the husbandry standards, the housing conditions were a temperature of 20-26°C, a humidity of 40-70%, and a light cycle of 12 hours on and 12 hours dark. All animals were acclimated to the environment after purchase and observed for health before experiments. All animals used were healthy, pathogen-free, and had free access to food and water. Human samples were obtained from brain tissue of a 69-year-old male who died of intracranial hemorrhage.

[0081] Experimental conditions: Mice were anesthetized with intraperitoneal injection of sodium pentobarbital (50 mg / kg) and then sacrificed by cervical dislocation. Mice were grouped according to different temperatures (including 4°C, 15°C, 20°C, 25°C, 30°C, and 35°C), and whole-brain tissue samples were collected at different PMIs. Specifically, brain tissue was extracted at 4°C on days 0, 1, 2, 4, 8, 12, and 16 after death; at 15°C on days 0, 25, 49, 97, 193, and 289 after death; at 20°C on days 0, 26, 50, 98, 146, and 241 after death; at 25°C on days 0, 1, 2, 4, and 8 after death; at 30°C on days 0, 10, 27, 51, 99, and 147 after death; and at 35°C on days 0, 12, 24, 48, 72, and 96 after death. Samples were collected from five mice at each PMI under the same temperature conditions. After preliminary testing, four samples with minimal variability at each PMI were selected for further study. Additionally, tissues were collected at other known PMIs (4°C, 25°C, and 35°C) as validation samples. Known PMI refers to samples collected at a specific time point after death to validate a previously constructed PMI inference model. These samples are different from the samples used to construct the aforementioned model. All mouse experiments adhered to the Guidelines for Euthanasia of Laboratory Animals (GB / T 39760-2021).

[0082] 4. Total RNA and Genomic DNA Extraction

[0083] Total RNA was extracted using an RNA extraction kit (Beijing Tiangen Biochemical Technology Co., Ltd., catalog number: DP419). First, 30-50 mg of mouse or human brain tissue was weighed. The subsequent extraction process was based on the instructions of the RNA extraction kit and the "Concise Guide to Medical Molecular Biology Experiments" (Fu Junjiang, China Medical Science and Technology Press, 2012, p18-20). The extracted total RNA was extracted using a NanoDrop TM The concentration and purity were measured by spectrophotometer, and the integrity of RNA was detected by 1% agarose gel electrophoresis. The integrity and brightness ratio of 28S, 18S, and 5S bands were observed to evaluate the integrity and degradation degree of RNA.

[0084] Extract genomic DNA (gDNA) using the phenol-chloroform method, following the protocol in the "Concise Guide to Medical Molecular Biology Experiments" (Fu Junjiang, China Medical Science and Technology Press, 2012, pp. 5-8). After extraction, aspirate an appropriate amount of sample to determine concentration and purity, then store at 4°C or -20°C.

[0085] 5. Reverse Transcription Experiment

[0086] Total RNA was reverse transcribed into cDNA using the TOYOBO reverse transcription kit (Toyobo Biotech Co., Ltd., Cat. No. FSQ-201) according to the "Concise Guide to Medical Molecular Biology Experiments" (Fu Junjiang, China Medical Science and Technology Press, 2012, pp. 18-20).

[0087] 6. Validation of circular RNA

[0088] Convergent and divergent primers for different circRNAs and the control gene Actb are listed in Table 1. After PCR, PCR products were analyzed by agarose gel electrophoresis. These circRNAs and Actb were detected in cDNA and gDNA using convergent and divergent primers. Agarose gel electrophoresis analysis showed that circFat3 of the expected size was detected in both cDNA and gDNA using convergent primers, whereas it was only detected in cDNA using divergent primers ( Figure 7 A) The product lengths of circFat3 using divergent and convergent primers are 276 and 284 bp, respectively. In the control group, Actb was detected in both cDNA and gDNA using the convergent primers, but no band was amplified using the divergent primers. These results confirm that circFat3 has a circular structure.

[0089] In addition, RNase R (Guangzhou Jisai Biotechnology Co., Ltd., Catalog No.: R0300) was used to treat the samples to detect the expression of circRNA and its corresponding linear RNA. RNase R is a nuclease derived from Escherichia coli that can digest almost all linear RNA molecules, but it is difficult to digest circRNA. The RNase R digestion test was performed according to the manufacturer's instructions. For each sample, 200 ng of total RNA was used for RNase R digestion. The control group added 1 μL of 10× reaction buffer and adjusted to a final volume of 10 μL with RNase-free water. For the first treatment group (0U), 1 μL of 10× reaction buffer was added to the mixed liquid. For the second treatment group (1U), 1U of RNase R (1U / μL) and 1 μL of 10× reaction buffer were added. Both treatment groups were enzymatically digested at 37°C for 15 minutes, and then the enzyme was inactivated at 70°C for 10 minutes.

[0090] As the gold standard for circRNA validation, Sanger sequencing was used to confirm the splicing sites of circRNAs. PCR amplification was performed using divergent primers, followed by sequencing using a 3500Dx Genetic Analyzer.

[0091] After RNase R digestion of total RNA, circFat3 could still be detected, while linear Fat3 was degraded ( Figure 7B). Quantitative analysis further confirmed that circFat3 is more resistant to RNase R digestion than linear RNA ( Figure 8 The experimental results of the 0 activity unit (U) group suggested that the digestion treatment itself may affect the expression level. The structural characteristics of circFat3 were predicted by CircPrimer software combined with sequence data, and it was found that it has a specific reverse splicing site derived from Fat3 exon 1. Sanger sequencing confirmed that the splicing site was consistent with the results of bioinformatics analysis ( Figure 9 ). NCBI gene sequence alignment showed that the circular RNA contained part of the 5' upstream region and the coding sequence of exon 1.

[0092] 7. Screening and determination of internal reference genes

[0093] Based on the ICG (Internal Control Genes) database, https: / / ngdc.cncb.ac.cn / icg / , we analyzed the frequency of mRNA use as internal reference genes in human and mouse studies. We then selected the intersection of the four most commonly used internal reference genes in human and mouse studies (Gapdh, Actb, Hprt1, and Tbp) for analysis. In addition, we included cytochrome c oxidase subunit I (mt-co1), 28S rRNA, and 18S rRNA as candidate internal reference genes. 28S rRNA and 18S rRNA are commonly used for PMI estimation. mt-co1, a mitochondrial mRNA encoded by mitochondrial DNA, has potential as a reference gene due to its long half-life and resistance to ribonucleases. To this end, a total of seven markers were screened as candidate reference genes.

[0094] The stability of these seven candidate internal reference genes was evaluated by measuring the relative expression levels of these genes at different PMIs at 35°C. Electrophoresis analysis showed that mt-co1 and 28S rRNA were more stable than other markers ( Figure 10 ), under the same amplification conditions, the expression level of Tbp was extremely low and almost no band was detected, so its data is not shown. The line graph intuitively presents the postmortem degradation trend of these internal reference genes ( Figure 11 ). mRNA degradation is rapid, with half-lives of no more than 24 hours. Among these reference genes, 28S rRNA is more stable than 18S rRNA, making it a more suitable reference gene. Furthermore, mt-co1 has the slowest degradation rate, maintaining a relatively high abundance even 96 hours after death. Therefore, mt-co1 and 28S rRNA were selected as reference genes for subsequent studies.

[0095] 8. Detection of circFat3 degradation levels after death

[0096] Initially, semi-quantitative RT-PCR was used to detect the level of circRNA degradation. The total reaction volume was 10 μL, including 5 μL 2× Taq PCR Master Mix (Beijing Tiangen Biotechnology Co., Ltd., Catalog No.: DP210831), 3 μL RNase-Free Water, 1 μL dispersed primer pair and 1 μL cDNA. For circRNA biomarkers, the PCR program was set as follows: the first step was pre-denaturation at 95°C / 90s; the second step of denaturation was 30 cycles at 95°C / 30s, annealing at 65°C / 30s, and extension at 72°C / 25s; the third step was final extension at 72°C / 5min. For the internal reference gene, the second step of the PCR program was different from that of the circRNA marker. It was set as follows: denaturation at 95°C / 30s, annealing at 60°C / 30s, extension at 72°C / 25s, 19 cycles for mt-co1 or 20 cycles for 28S rRNA. These amplification reactions were performed using Applied Biosystems. PCR was performed using a 96-Well Thermal Cycler (Thermo Fisher Scientific Inc, USA). After PCR amplification, 1.5% agarose gel was used for electrophoresis and exposed and imaged using a Bio-Rad Universal Hood II (Bio-Rad Laboratories, Inc, USA).

[0097] For RT-qPCR, 2× Fast SYBR Green qPCR Master Mix Kit (ServicebioTechnology Co., China, Cat. No. 3325-05) was used for quantitative analysis of circRNA expression. According to the manufacturer's instructions, a reaction mixture (10 μL) was prepared consisting of 3 μL RNase-free water, 1 μL divergent primer pair, 1 μL cDNA, and 5 μL 2× Fast SYBR Green qPCR mix (High ROX). This experiment used Applied Biosystems StepOneplus TM Real-Time PCR was performed using a Thermo Fisher Scientific Inc., USA. Cycling conditions were as follows: a first PCR denaturation step of 95°C for 2 minutes; a second denaturation step of 40 cycles at 95°C for 5 seconds, followed by annealing / extension at 60°C for 20 seconds; and melting curve analysis at 95°C for 15 seconds, 60°C for 1 minute, and 95°C for 15 seconds. Samples were run in triplicate, and only samples with a standard deviation less than 0.2 were analyzed.

[0098] 9. Construction and Use of Standards

[0099] Standard samples were used for absolute quantification. The cDNA of mouse brain tissue was used as a template, and 28S rRNA was used as a primer, and then PCR amplification was performed. The PCR program was set as follows: the first step was pre-denaturation at 95°C / 90s; the second step was denaturation at 95°C / 30s, annealing at 65°C / 30s, and extension at 72°C / 25s, for a total of 35 cycles; the third step was final extension at 72°C / 5min. The total reaction volume was 50μL, and then 1.5% agarose gel electrophoresis was performed. Gel recovery experiments were performed using an agarose gel DNA recovery kit (Beijing Tiangen Biochemical Technology Co., Ltd., Product No.: DP219). 30-100μL of ddH2O was used for elution, and then NanoDrop TM The concentration was measured by spectrophotometer and diluted with water to 55 ng / μL.

[0100] 10. Establishment and verification of PMI inference model

[0101] For the results of semi-quantitative RT-PCR, the grayscale value of the bands was detected using ImageJ software. In order to eliminate the influence of the image background, the grayscale value of the background was subtracted from the grayscale value of each band. In order to eliminate the influence of exposure conditions, the above-mentioned standards were used. In different agarose gel plates, standards were used to normalize the grayscale values ​​between circRNA markers and internal reference genes. After background subtraction, the average grayscale value of human and mouse circRNA markers was divided by the average grayscale value of the internal reference genes under the same PMIs. This ratio represents the relative expression of circRNAs in different PMIs. Among them, the ratio of the grayscale value of circRNA on day 0 to the corresponding internal reference gene on day 0 was defined as the K value. Construct the equation: Y = (circFat3-n day grayscale - circFat3-n day background grayscale value) / (internal reference n day grayscale value - internal reference n day background grayscale value) / K. According to the data, linear regression equations and nonlinear regression equations (quadratic equations, cubic equations) were constructed, and the p value and R of the corresponding equations were calculated. 2 For RT-qPCR results, the average of the reliable data for circFat3 and the internal reference gene was used to calculate the ΔCt value in each PMI. The formula is as follows: ΔCt = Ct(circFat3) - Ct(internal reference gene mt-co1).

[0102] IBM SPSS Statistics 25 (International Business Machines Corporation, Armonk, USA) was used to establish mathematical models and calculate the corresponding p-values ​​and R values ​​for linear and nonlinear regression equations. 2GraphPad Prism 8 (GraphPad Software, LLC., USA) was used to construct regression curves for linear, quadratic, and cubic equations.

[0103] To validate the PMI mathematical model, we use the error rate to verify the model's accuracy. The formula is as follows: Error rate = (estimated PMI - actual PMI) / actual PMI × 100%. This data can reflect the accuracy of the model.

[0104] In addition, based on the postmortem levels of circFat3 at multiple PMIs at 4°C, 15°C, 20°C, 25°C, 30°C, and 35°C, an AI inference model was constructed using R software (V.4.2.2) packages such as e1071, kernlab, ggplot2, and plotly, which can be used to estimate PMI under different temperature conditions. The PMI inference model of the support vector machine regression model (SVR) with a radial basis kernel is:

[0105]

[0106] Where x = new input sample (with the following parameters: circFat3, expression level, temperature); x i = support vector (training samples used to define the decision boundary); α i ,αi * = dual coefficient (Lagrange multiplier from the support vector machine model); b = bias term (intercept); K(x i ,x) = radial basis function kernel, which is:

[0107] K(x i , x) = exp(-γ·||x i -x|| 2 )

[0108] Among them, γ (gamma) is used to control the influence range of each support vector. The formula can be approximately expressed as: PMI (hours) ≈ SVM RBF (circFat3, expression level, temperature).

[0109] Finally, an application of AI models that can be used for PMI inference in humans and mice was built based on the R language.

[0110] (1) Establishment and validation of a mouse PMI inference model at 4°C

[0111] Total RNA was extracted from mouse brain tissue samples stored at 4°C at 0, 1, 2, 4, 8, 12, and 16 days postmortem, and the postmortem expression levels of circFat3 and the reference genes mt-co1 and 28S rRNA were detected ( Figure 12). The results showed that mt-co1 and 28S rRNA remained stable under low temperature conditions, while circFat3 maintained a relatively stable level within 8 days after death. Based on the grayscale value of the band, parameters such as relative expression and K value were calculated for the establishment of mathematical models. In these models, the relative expression and PMI were set as Y value and X value, respectively. The ratio of the grayscale value of the biomarker to the grayscale value of the reference gene on day 0 was defined as the K value (circFat3 / mt-co1 was 2.587, and circFat3 / 28S rRNA was 1.306). The regression equation was constructed by calculating the grayscale value ratio of circFat3 and the internal reference gene under different PMIs (see Table 2), and the fitting curve is shown in Figure 2. Figure 13 A and 13B.

[0112] When mt-co1 is used as the internal reference, R of linear, quadratic and cubic equations 2 The values ​​and p-values ​​were 0.8604 (p = 0.0026), 0.9208 (p = 0.0063) and 0.9724 (p = 0.0077), respectively. 2 The values ​​are all greater than 0.86, indicating that the model fits well. When 28S rRNA is used as the internal reference, the R 2 The values ​​were 0.9900 (p<0.0001), 0.9906 (p=0.0001) and 0.9986 (p=0.0001), respectively. The equation constructed by combining the two internal reference genes had an R 2 The values ​​were all greater than 0.9700 and the p values ​​were all less than 0.005. 2 Judging from the p-value, the circFat3 / 28S rRNA model is more suitable for low-temperature PMI inference. The details of the mathematical model are shown in Table 2.

[0113] The accuracy of the model under 4°C condition was tested using validation samples with known PMI (2, 6.25 and 9.25 days). Figure 14 In the circFat3 / mt-co1 model, the error rates calculated by linear, quadratic, and cubic equations for samples on day 2 were 225.50%, 143.05%, and 81.52%, respectively; the error rates for samples on day 6.25 were 14.22%, 14.73%, and 35.30%, respectively; and the error rates for samples on day 9.25 were 3.25%, 26.30%, and 38.04%, respectively. In the circFat3 / 28S rRNA model, the error rates for samples on day 2 were 127.12%, 121.09%, and 148.95%, respectively; the error rates for samples on day 6.25 were 1.24%, 1.50%, and 3.40%, respectively; and the error rates for samples on day 9.25 were 35.49%, 37.26%, and 33.29%, respectively.

[0114] When modeling with two internal reference genes, the error rates for day 2 samples were 276.75%, 223.49%, and 211.15%, respectively; for day 6.25 samples, the error rates were 28.96%, 11.34%, and 8.69%, respectively; and for day 9.25 samples, the error rates were 9.70%, 21.73%, and 22.94%, respectively. The error rate for day 2 samples was significantly higher than that for later samples. These results indicate that the circFat3 / 28S rRNA model performs optimally at 4°C. The combined use of internal reference genes did not significantly improve prediction accuracy, but only provided benefits for long-term PMI inference.

[0115] Table 2. Mathematical model for mouse PMI estimation at 4°C established by semi-quantitative RT-PCR

[0116]

[0117] (2) Establishment and validation of a mouse PMI inference model at 25°C

[0118] Total RNA was extracted from brain tissue samples of 9-week-old mice stored at 25°C at 0, 1, 2, 4, and 8 days postmortem, and the amplification results of circFat3, mt-co1, and 28S rRNA were detected by agarose gel electrophoresis ( Figure 15 The results showed that the degradation rate of all markers was faster than that under low temperature conditions, while mt-co1 and 28S rRNA remained stable at room temperature, and circFat3 maintained a relatively stable level within 1 day after death. At 25°C, the K values ​​of circFat3 / mt-co1 and circFat3 / 28S rRNA were 2.702 and 1.259, respectively. The regression equation was constructed by calculating the gray value ratio of circFat3 and the internal reference gene at different PMIs (see Table 3), and the fitting curve is shown in Figure 3. Figure 16 A and 16B.

[0119] When mt-co1 is used as the internal reference, R of linear, quadratic and cubic equations 2 The corresponding values ​​and p-values ​​were 0.6551 (p = 0.0893), 0.9713 (p = 0.0288) and 0.9873 (p = 0.1423), respectively. When 28S rRNA was used as the internal reference, the corresponding values ​​were 0.6014 (p = 0.1233), 0.9399 (p = 0.0601) and 0.9726 (p = 0.2099); when the double internal reference gene was used, the corresponding values ​​were 0.6375 (p = 0.1053), 0.9568 (p = 0.0432) and 0.9806 (p = 0.1766). 2 The values ​​are better than those of the corresponding linear and quadratic equations, but their p values ​​are all greater than 0.05. The details of the mathematical model are shown in Table 3.

[0120] The accuracy of the 25℃ model was tested using validation samples with known PMI (1.5 and 4 days). Figure 17 In the circFat3 / mt-co1 model, the error rates calculated by linear, quadratic, and cubic equations for 1.5-day samples were 58.06%, 38.47%, and 45.75%, respectively, and for 4-day samples were 16.91%, 34.83%, and 39.67%, respectively. In the circFat3 / 28S rRNA model, the error rates for 1.5-day samples were 67.22%, 40.60%, and 49.49%, respectively, and for 4-day samples were 12.31%, 40.12%, and 48.20%, respectively. When using dual internal references for joint modeling, the error rates for 1.5-day samples were 62.54%, 39.56%, and 47.73%, respectively, and for 4-day samples were 14.66%, 37.54%, and 44.09%, respectively. The error rate of 1.5-day samples was significantly higher than that of 4-day samples. The results showed that the circFat3 / mt-co1 model was superior to the circFat3 / 28S rRNA model in 25°C PMI inference, and the combined use of dual internal references could improve the accuracy of long-term PMI inference.

[0121] Table 3. Mathematical model for mouse PMI estimation at 25°C established by semi-quantitative RT-PCR

[0122]

[0123]

[0124] (3) Establishment and validation of a mouse PMI inference model at 35°C

[0125] Total RNA was extracted from brain tissue samples at 0, 12, 24, 48, 72, and 96 hours after death at 35°C, and the amplification results of circFat3, mt-co1, and 28S rRNA were analyzed by agarose gel electrophoresis ( Figure 18 The results showed that circFat3 was degraded faster than at other temperatures, while mt-co1 and 28S rRNA remained stable at high temperatures. At 35°C, the K values ​​of circFat3 / mt-co1 and circFat3 / 28S rRNA were 2.668 and 1.331, respectively. The regression equation was constructed by calculating the gray value ratio of circFat3 and the reference gene at different PMIs (Table 4), and the fitting curve is shown in Figure 4. Figure 19 A and 19B.

[0126] When mt-co1 is used as the internal reference, R of linear, quadratic and cubic equations 2The corresponding values ​​and p-values ​​were 0.8639 (p = 0.0073), 0.9888 (p = 0.0012) and 0.9897 (p = 0.0154), respectively. When 28S rRNA was used as the internal reference, the corresponding values ​​were 0.8496 (p = 0.0090), 0.9196 (p = 0.0228) and 0.9210 (p = 0.1162); when the double internal reference genes were combined, the corresponding values ​​were 0.8667 (p = 0.0070), 0.9634 (p = 0.0070) and 0.9634 (p = 0.0544). 2 Based on the p-values, the circFat3 / mt-co1 model may be more suitable for estimating high-temperature PMI. Although all cubic equations performed better than the linear and quadratic equations, only the cubic equation for the circFat3 / mt-co1 model had a p-value less than 0.05. Details of the mathematical models are shown in Table 4.

[0127] The accuracy of the 35°C model was tested using validation samples with known PMI (24, 30, and 60 hours). Figure 20 In the circFat3 / mt-co1 model, the error rates calculated by linear, quadratic, and cubic equations were 56.54%, 50.21%, and 51.75% for 24-hour samples, 124.26%, 67.60%, and 68.93% for 30-hour samples, and 24.97%, 0.23%, and 3.69% for 60-hour samples. In the circFat3 / 28S rRNA model, the error rates were 233.57%, 216.43%, and 195.56% for 24-hour samples, 145.14%, 111.16%, and 103.37% for 30-hour samples, and 22.46%, 5.41%, and 1.55% for 60-hour samples.

[0128] When modeling with both internal references, the error rates for 24-hour samples were 213.54%, 164.15%, and 162.92%, respectively; for 30-hour samples, they were 134.61%, 87.77%, and 87.26%, respectively; and for 60-hour samples, they were 23.73%, 2.93%, and 2.50%, respectively. The error rates for the 60-hour samples were significantly lower than those for the 24- and 30-hour samples. These results indicate that the circFat3 / mt-co1 model performs optimally at 35°C, and that the combined use of both internal references improves long-term PMI inference only.

[0129] Table 4. Mathematical model for mouse PMI estimation at 35°C established by semi-quantitative RT-PCR

[0130]

[0131]

[0132] (4) Establishment and verification of the PMI inference model for mice at 15°C

[0133] Total RNA was extracted from brain tissue samples of 9-week-old mice stored at 15°C at 0, 25, 49, 97, 193, and 289 hours postmortem, and the amplification results of circFat3, mt-co1, and 28S rRNA were detected by agarose gel electrophoresis ( Figure 21 The results showed that the degradation rate of all markers was faster than that under low temperature conditions, while mt-co1 and 28S rRNA remained stable at 15°C, and circFat3 maintained a relatively stable level within 1 day after death. The K values ​​of circFat3 / mt-co1 and circFat3 / 28S rRNA at 15°C were 1.207 and 0.951, respectively. The regression equation was constructed by calculating the gray value ratio of circFat3 and the internal reference gene at different PMIs (Table 5), and the fitting curve is shown in Figure 5. Figure 22 A and 22B.

[0134] When mt-co1 is used as the internal reference, R of linear, quadratic and cubic equations 2 The corresponding values ​​and p-values ​​were 0.6282 (p = 0.0601), 0.7959 (p = 0.0922) and 0.8535 (p = 0.2115), respectively. When 28S rRNA was used as the internal reference, the corresponding values ​​were 0.6038 (p = 0.0690), 0.8796 (p = 0.0418) and 0.9071 (p = 0.1360). When the double internal reference genes were combined, the corresponding values ​​were 0.6198 (p = 0.0631), 0.8431 (p = 0.0622) and 0.8837 (p = 0.1693). 2 The circFat3 / 28S rRNA model was superior, as judged by the p-value. All cubic equations had better goodness-of-fit than the linear and quadratic equations, but only the quadratic equation for the circFat3 / mt-co1 model had a p-value less than 0.05. Details of the mathematical models are shown in Table 5.

[0135] Table 5. Mathematical model for mouse PMI estimation at 15°C established by semi-quantitative RT-PCR

[0136]

[0137]

[0138] (5) Establishment and verification of the PMI inference model for mice at 20°C

[0139] Total RNA was extracted from brain tissue samples of 9-week-old mice stored at 20°C at 0, 26, 50, 98, 146, and 241 hours postmortem, and the amplification results of circFat3, mt-co1, and 28S rRNA were detected by agarose gel electrophoresis ( Figure 23 The results showed that mt-co1 and 28S rRNA remained relatively stable within 98 hours, and circFat3 maintained a relatively stable level within 26 hours after death. The K values ​​of circFat3 / mt-co1 and circFat3 / 28S rRNA at 20°C were 1.135 and 0.959, respectively. The regression equation was constructed by calculating the gray value ratio of circFat3 and the internal reference gene at different PMIs (Table 6), and the fitting curve is shown in Figure 6. Figure 24 A and 24B.

[0140] When mt-co1 is used as the internal reference, R of linear, quadratic and cubic equations 2 The corresponding values ​​and p-values ​​were 0.8559 (p = 0.0082), 0.9864 (p = 0.0016) and 0.9880 (p = 0.0180), respectively. When 28S rRNA was used as the internal reference, the corresponding values ​​were 0.7207 (p = 0.0325), 0.9328 (p = 0.0174) and 0.9777 (p = 0.0333); when the double internal reference genes were combined, the corresponding values ​​were 0.7993 (p = 0.0162), 0.9690 (p = 0.0055) and 0.9843 (p = 0.0234). 2 Judging from the p-value, the circFat3 / mt-co1 model is better, and the goodness of fit of the nonlinear equation is better than that of the linear equation. However, the p-values ​​of all models are less than 0.05. The details of the mathematical model are shown in Table 6.

[0141] Table 6. Mathematical model for mouse PMI estimation at 20°C established by semi-quantitative RT-PCR

[0142]

[0143] (6) Establishment and verification of the PMI inference model for mice at 30°C

[0144] Total RNA was extracted from brain tissue samples of 9-week-old mice stored at 30°C at 0, 10, 27, 51, 99, and 147 hours postmortem, and the amplification results of circFat3, mt-co1, and 28S rRNA were detected by agarose gel electrophoresis ( Figure 25The results showed that mt-co1 and 28S rRNA remained relatively stable within 27 hours, while circFat3 levels decreased significantly 10 hours after death. The K values ​​of circFat3 / mt-co1 and circFat3 / 28S rRNA at 30°C were 1.201 and 0.848, respectively. The regression equation was constructed by calculating the grayscale value ratio of circFat3 and the reference gene at different PMIs (Table 7), and the fitting curve is shown in Figure 7. Figure 26 A and 26B.

[0145] When mt-co1 is used as the internal reference, R of linear, quadratic and cubic equations 2 The corresponding values ​​and p-values ​​were 0.7301 (p = 0.0302), 0.9166 (p = 0.0241) and 0.9351 (p = 0.0957), respectively. When 28S rRNA was used as the internal reference, the corresponding values ​​were 0.6127 (p = 0.0657), 0.9231 (p = 0.0213) and 0.9816 (p = 0.0275). When the double internal reference genes were used, the corresponding values ​​were 0.6836 (p = 0.0424), 0.9312 (p = 0.0180) and 0.9671 (p = 0.0489). 2 Judging by the p-value, the model with dual internal parameters is better, in which the goodness of fit of all nonlinear equations is better than that of linear equations, but the p-values ​​of all equations with dual internal parameters are less than 0.05. The details of the mathematical model are shown in Table 7.

[0146] Table 7. Mathematical model for mouse PMI inference at 30°C established by semi-quantitative RT-PCR

[0147]

[0148] 11. Further construct a high temperature PMI inference model based on RT-qPCR data

[0149] According to the semi-quantitative RT-PCR results, circFat3 and mt-co1 were more suitable for PMI inference at 35°C. The primer specificity was detected by fluorescence quantitative melting curve, such as Figure 27 As shown in Figures A and 27B, the results showed that both circFat3 and mt-co1 presented a single peak, indicating that they were suitable for subsequent experiments. Figure 28 As shown in A and 28B, the sample error is minimal and the data is reliable. The circFat3 / mt-co1 model constructed based on the ΔCt value is shown in Figure 29 As shown, the R of its linear, quadratic and cubic equations 2The p-values ​​were 0.8998 (p=0.0039), 0.9740 (p=0.0042) and 0.9797 (p=0.0304), respectively (Table 8), indicating that the data fitted the equation well and the curve was reliable.

[0150] The accuracy of the equations was tested using validation samples with known PMIs (24, 30, and 60 hours). The error rates calculated using the linear, quadratic, and cubic equations for the 24-hour sample were 100.10%, 51.99%, and 58.08%, respectively; for the 30-hour sample, the errors were 53.56%, 16.43%, and 21.82%, respectively; and for the 60-hour sample, the errors were 32.02%, 22.95%, and 11.11%, respectively. The results showed that the nonlinear equations were significantly more accurate than the linear equations. This RT-qPCR experiment further validated the semi-quantitative RT-PCR results, confirming that these equations are more suitable for long-term PMI inference.

[0151] Table 8. Mathematical model for mouse PMI estimation at 35°C established by qPCR

[0152]

[0153]

[0154] 12. Establishment of Mouse 3D Model

[0155] To construct a PMI inference model applicable to multiple temperature conditions, a three-dimensional inference model ( ) was established based on the relative expression data at 4°C, 15°C, 20°C, 25°C, 30°C, and 35°C, with 28S rRNA and mt-co1 as internal references. Figure 30 A and 30B). The upper right corner of the figure contains a color scale bar indicating the mapping between surface color and relative circRNA expression levels. Warmer colors (red) represent higher expression levels, while cooler tones (gray) correspond to lower expression levels. A three-dimensional model was constructed by standardizing the PMI units at different temperatures to hours. The results showed that these three-dimensional equations had a high degree of fit, with p-values ​​far less than 0.0001 (Table 9).

[0156] The three-dimensional model was validated using validation samples at different temperatures. For circFat3, the model demonstrated higher accuracy when using 28S rRNA as an internal reference, particularly for PMI inference at 25°C. Overall, under most environmental conditions, the model provided superior accuracy for estimating longer-term PMI (Table 10).

[0157] It should be noted that in the future, by incorporating more postmortem change data of circRNA under temperature conditions (especially low temperature), the accuracy of the model is expected to be further improved.

[0158] Table 9. Three-dimensional model of mouse PMI inference established by semi-quantitative RT-PCR

[0159]

[0160] Table 10. Validation of the PMI 3D model

[0161]

[0162] Establishment of a mathematical model for estimating human PMI at 20°C and 25°C

[0163] Total RNA was extracted from human brain tissue samples stored at 20°C at 3.5, 18.5, 41.5, 64.5, 113.5, and 161.5 hours postmortem, and the amplification results of human circFat3 and 28S rRNA were detected by agarose gel electrophoresis ( Figure 31 ). The results showed that 28S rRNA remained relatively stable within 113.5 hours, and the level of human circFat3 decreased significantly 3.5 hours after death. Absolute quantification was performed using standard products, and a regression equation was constructed by calculating the gray value ratio of human circFat3 and the internal reference gene 28S rRNA at different PMIs (Table 11). In addition, total RNA was extracted from human brain tissue samples stored at 25°C at 3.5, 19.5, 42.5, 65.5, 114.5 and 162.5 hours after death, and the amplification results of human circFat3 and 28S rRNA were detected by agarose gel electrophoresis ( Figure 32 The results showed that 28S rRNA remained relatively stable within 114.5 hours, and the level of human circFat3 decreased significantly 3.5 hours after death. Absolute quantification was performed using standard products, and a regression equation was constructed by calculating the gray value ratio of human circFat3 and the internal reference gene 28S rRNA at different PMIs (Table 11). The fitting curve is shown in Figure 33 A and 33B.

[0164] Table 11 Human mathematical model for PMI inference established by semi-quantitative RT-PCR

[0165]

[0166] 28S rRNA was used as the internal reference, and the R values ​​of the linear, quadratic, and cubic equations at 20°C were calculated. 2 The R values ​​and p values ​​were 0.2936 (p = 0.2668), 0.5933 (p = 0.2594), and 0.7740 (p = 0.3191), respectively; the R values ​​of the linear, quadratic, and cubic equations at 25 °C were 2The values ​​and p-values ​​were 0.3311 (p = 0.2340), 0.5587 (p = 0.2932) and 0.8219 (p = 0.2549), respectively. 2 Judging by the p-value, the goodness of fit of all nonlinear equations is better than that of linear equations, but the p-values ​​of all equations are greater than 0.05. The details of the mathematical model are shown in Table 11.

[0167] 14. Establishment of AI Model for Human PMI Inference

[0168] The AI ​​model was constructed through machine learning algorithms, and the support vector machine (SVM) was selected to construct the mouse model. Then, based on human data and using transfer learning, the migration construction from the mouse model to the human model was completed. Among them, the process of selecting the best SVM model was achieved by comparing the cross-validation root mean square error (RMSE) of the radial basis kernel model and the polynomial kernel model. Based on the data of various mouse temperatures and human 20°C and 25°C, the R language packages readxl, ggplot2, plotly, caret, e1071, lmtest and openxlsx were used to build an SVM inference model, and transfer learning was used to complete the migration from the mouse model to the human model, successfully building an application software that can be applied to PMI inference. The relevant information and operation interface of the software are displayed in Figure 34 It can be used to infer PMI in mice and humans. According to the cross-validation results, among all combinations, the RMSE (0.1081203) is the smallest when C=100 and sigma=1, and the corresponding R-squared is 0.6962008, which is also a relatively high value; the mean absolute error (MAE) is 0.07416523, which performs well. The result interface of the inference using this software shows the three-dimensional pattern diagram, fit and accuracy of the SVM machine learning model ( Figure 35 ). The details of the AI ​​model built based on transfer learning are also shown in Figure 36 The code is as follows, where the mouse data file is named "Mouse_PMI" and the human data file is named

[0169]

[0170]

[0171] The present invention uses semi-quantitative RT-PCR and RT-qPCR to detect the postmortem degradation levels of different PMIs and finds that circFat3 is highly specifically expressed in mouse brain tissue, highly homologous in humans and mice, and expressed at high levels. Postmortem levels are significantly correlated with PMI at multiple temperatures. In addition, mt-co1 and 28S rRNA exhibit stability under various temperature conditions, making them suitable as reference genes for PMI models. Validation results show that the model provided by the present invention is more accurate for long-term PMI estimation. The present invention establishes circRNA circFat3 as a new biomarker for PMI inference for the first time. Through supporting primers, kits, and AI models, it solves the problems of traditional forensic methods being subject to significant environmental interference and strong subjectivity, providing high-precision, standardized molecular technology support for criminal case investigation and civil forensic identification.

[0172] The above embodiment is only one of the preferred implementation methods of the present invention and should not be used to limit the scope of protection of the present invention. Any changes or modifications that have no substantive meaning made to the main design concept and spirit of the present invention, as long as the technical problems solved are still consistent with the present invention, should be included in the scope of protection of the present invention.

Claims

1. A circRNA biomarker for forensic PMI inference, characterized in that: The circRNA biomarkers are human and mouse circFat3 genes or variants thereof with ≥87% sequence homology.

2. A primer pair for specifically detecting the circFat3 gene according to claim 1, characterized in that: The primer pairs include divergent primer pairs and convergent primer pairs, wherein: The upstream and downstream primer sequences of the divergent primer pair are shown in SEQ ID NO: 1 and SEQ ID NO: 2; The upstream and downstream primer sequences of the convergent primer pair are shown in SEQ ID NO: 3 and SEQ ID NO:

4.

3. A kit for specifically detecting the circFat3 gene according to claim 1, characterized in that: The kit comprises the primer pair according to claim 2 and an internal reference gene primer, wherein the internal reference gene is mt-co1 and / or 28SrRNA.

4. The kit according to claim 3, wherein When the internal reference gene is mt-co1, the upstream sequence of the detection primer is shown in SEQ ID NO: 21, and the downstream sequence of the detection primer is shown in SEQ ID NO: 22; when the internal reference gene is 28S rRNA, the upstream sequence of the detection primer is shown in SEQ ID NO: 11, and the downstream sequence of the detection primer is shown in SEQ ID NO:

12.

5. Use of the kit according to claim 4 in forensic PMI inference.

6. The use according to claim 5, characterized in that The forensic PMI inference method includes the following steps: S1, total RNA extracted from brain tissue samples; S2, reverse transcription of total RNA into cDNA; S3, the expression level of circFat3 in the samples was detected by semiquantitative RT-PCR or RT-qPCR using divergent primer pairs and convergent primer pairs; S4, relative expression of circFat3 was calculated using mt-co1 and / or 28S rRNA as internal reference genes; S5, based on the relative expression levels, inferring the PMI by constructing a mathematical model or an AI model.

7. The use according to claim 6, characterized in that The mathematical model is a linear equation, a quadratic equation or a cubic equation, wherein the independent variable is PMI and the dependent variable is the relative expression level of circFat3; the relative expression level of circFat3 is the ratio of the expression level of circFat3 to the expression level of the internal reference gene.

8. The use according to claim 6 or 7, characterized in that Processing of semi-quantitative RT-PCR results included: using image analysis software to calculate the grayscale value of the amplified band and subtracting the background grayscale value; normalizing the grayscale values ​​of different agarose gel plates using standards; after subtracting the background grayscale value, the ratio of the average grayscale value of circFat3 to the average grayscale value of the reference gene was used to determine the relative expression level of circFat3; The processing of RT-qPCR detection results includes: calculating the ΔCt value based on the Ct value, where ΔCt is the Ct value of circFat3 minus the Ct value of the internal reference gene. The ΔCt value is the relative expression level.

9. The use according to claim 6, characterized in that The AI ​​model is a support vector machine (SVM) regression model, and an application software that can be used for PMI inference at multiple temperatures is constructed, where the regression model is: Where x = new input sample with the following parameters: circFat3, expression level, temperature; i = support vector, used to define the training samples of the decision boundary; α i ,αi * = dual coefficient, the Lagrange multiplier from the support vector machine model; b = bias term, i.e. intercept; K(x i ,x) = radial basis function kernel, specifically: K(x i ,x)=exp(-γ·||x i -x|| 2 ) Among them, γ is used to control the influence range of each support vector.

Images