Methylmalonic acidemia (MMA) is a serious, fatal, and multisystem damage of the most common organic acidemia (Chen et al., 2023). With neonatal screening, the pooled prevalence of MMA increased approximately 3.5 times between 2011 and 2020 (Jin et al., 2022). In China’s Shandong province, Henan province, and eastern areas, the incidence of live births due to MMA is 1:3,920, 1:4,714, and 1:5,590, respectively (Han et al., 2016; Wang, 2019; Yang et al., 2020). More significantly, early identification has improved the prognosis of MMA, and the mortality rate has decreased from 60%–80% to approximately 40% (Fabie et al., 2019; Zhou et al., 2019). Therefore, accurate and rapid diagnosis of MMA may be a key research topic in the field of inherited metabolic diseases (IMD).

MMA is a single-gene autosomal recessive disease; Zhou et al. (2018) reported that approximately 30% of Chinese patients have isolated MMA, and 70% have MMA with homocystinuria. MMA currently includes more than ten related genes and has high genetic heterogeneity. Among these, the variants in MUT and MMACHC are the most common in China (Zhou et al., 2018; Kang et al., 2020). According to the typical variant evidence types (such as population database, computational data, and functional data), ACMG classifies these variants as “pathogenic”, “likely pathogenic”, “uncertain significance”, “likely benign”, or “benign” (Richards et al., 2015). With the rapid development of gene sequencing techniques, several variants have been identified. Surprisingly, we screened the genes of 1773 newborns in Guangdong Province in 2021 and found that among the MMA-related variant sites, variants of uncertain significance (VUS) accounted for 85% of the variant sites, and approximately 78% were high-frequency variant sites. However, rapid and accurate analysis of the clinical effects of variants other than nonsense and frameshift variants has always been difficult in gene function analyses (Frederiksen et al., 2021). This is particularly true for inherited metabolic diseases requiring early diagnosis and treatment. Alternatively, VUS may cause serious adverse effects. Our research team has previously shown that one patient with homozygous VUS (c.1159A>C) and two patients with compound heterozygous (c.1159A>C and c.1280G>A, c.1159A>C, and c.1106G>A) variants in MMUT died because of severe metabolic disorders in the neonatal period, which is why the c.1159A>C variant was classified as likely pathogenic in the ClinVar database (Hui et al., 2019). However, owing to the limited number of cases, the pathogenicity of the c.1159A>C variant and other VUS in MMUT and MMACHC is worth further verification.

Currently, genetic screening, such as case reports, determination of protease activity in vitro fibroblast cultures, and determination of metabolite content using tandem mass spectrometry, is one of the most common methods for identifying the pathogenicity of variants (Kolhouse et al., 1981; Jansen and Ledley, 1990; Wang et al., 2022). However, this approach is often limited by the small sample sizes and complex operating technologies. Previous studies have found that enzyme activity in some hereditary metabolic disease carriers fluctuates between 25% and 52%, and the corresponding metabolites increase accordingly (Tucci et al., 2021). Therefore, interpreting the relationship between variants and the functions of MMUT or MMACHC variant sites using carriers with a large population base may be a new research strategy. In addition, artificial intelligence applications in the medical field have become a popular topic in modern science and technology. The accuracy of natural language processing (NLP) algorithms for the auxiliary diagnosis of rare diseases has reached 88%–98% (Hens et al., 2022). Artificial intelligence can be used as an effective and accurate tool to promote rapid diagnosis and early intervention in rare diseases. Therefore, our research team planned to take the lead in gene sequencing, protein structure analysis, and tandem mass spectrometry data (within 7 days of birth) to evaluate the pathogenic efficacy of VUS in MMUT or MMACHC gene carriers from 35 participating units in China and establish an MMA-related high-risk database to quickly identify and improve the accuracy and efficiency of early screening for MMA patients.

Our study unveils the cross-sectional data of carrying MMUT and MMACHC gene variations from 35 participating units in China. Using gene sequencing, protein structure analysis, and metabolic data, our team was the first to reveal the change in the metabolic level of those carrying the P/LP variant group and to evaluate the pathogenicity of VUS, boasting the largest sample size of MMUT and MMACHC variant carriers reported to date. The new strategy identified the opportunity to reclassify these variants in MMUT and MMACHC in the current database and may enable us to prioritize and target the pathogenicity of variants in other inherited metabolic diseases for evaluation.

In the NIH database, the VUS of the most common MMUT and MMACHC genes in MMA account for 38.34% and 34.57%, respectively. We found that 68.15% of the examiners carried VUS by studying the variant of MMUT or MMACHC gene with only carried and an incidence rate greater than >1/11,000. Among them, the first three variant sites with the highest constituent ratios were all VUS variant; this potential health problem imposes a psychological burden on asymptomatic individuals and affects clinical management. To further classify VUS, the ClinGen expert group adopted a Bayesian framework to facilitate the automatic quantitative evaluation of pathogenicity (Tavtigian et al., 2018). In addition, multidisciplinary cooperation such as clinical phenotypic assessment, in-depth literature research including artificial intelligence, and protein structure analysis have become important tools for VUS classification (Caswell et al., 2022; Casaletto et al., 2023; Farman et al., 2024). However, the pathogenicity of rare diseases still lacks sufficient samples, and artificial intelligence is easily affected by parameter changes. Therefore, accurately evaluating the relationship between variants and pathogenicity or disease clinical phenotypes is a challenge for molecular diagnosis and the implementation of individualized precision therapy.

Liquid chromatography-tandem mass spectrometry (LC-MS/MS) is a highly sensitive and accurate technique for detecting small-molecule metabolites (<1,500 kDa). Since the mid-20th century, it has been widely used in the diagnosis and treatment of genetic and metabolic diseases such as amino acid disorders, organic acidemia, and fatty acid oxidation defects, and has shown considerable effects and prospects in IEM characterization (Vernon, 2015; Piraud et al., 2018; Wang et al., 2019; Zhang et al., 2021). According to the human metabolome database (HMDB), more than 700 metabolites are associated with IEM (Mandal et al., 2018). For example, plasma biomarkers such as oxidized sterols, bile acids, and lysosphingolipids are used to detect sphingolipids and related diseases (such as Niemann-Pick type C) (Piraud et al., 2018). Detection propionylcarnitine using liquid and gas mass spectrometry is the most reliable method for screening prenatal MMA (Chen et al., 2021). Metabolites, as the final functional products of the genome, can be used to analyze genetic variation and explore gene–phenotype correlations. Heterozygous familial hypercholesterolemia with an 1/500 estimated prevalence may lead to metabolic abnormalities (cholesterol, low-density lipoproten, etc) and a 13-fold increased risk of coronary heart disease (Nordestgaard et al., 2013). We identified 83 newborns carrying P/LP variants in MMUT and 225 patients carrying P/LP variants in MMACHC using genome sequencing. It was revealed the C3, C3/C2, and C3/C0 levels in MMUT gene variant carriers were higher than those in the non-variant group, which could be used as an important rationale for exploring the pathogenicity of MMUT gene variants. This finding is similar to those observed in other genetic metabolic diseases. Van Acker et al. (1977) found abnormal adenine, 8- hydroxy −9, and 2,8-dihydroxyadenine levels in the urine of six carriers from a male family with complete adenine phosphoribosyltransferase deficiency. Pyrimidine, orotic acid, uridine, and uracil levels also increased in the plasma and urine of heterozygous women with ornithine carbamoyltransferase (OCT) deficiency (Webster et al., 1981) and OTC enzyme activity was 0%–60% of the normal value in a cohort of 281 heterozygous females (Gobin-Limballe et al., 2021). Neonatal screening data were collected. Our research team found that carriers of other metabolic diseases also show metabolic changes such as medium-chain acyl-CoA dehydrogenase deficiency and primary carnitine deficiencies. No significant changes were found in the metabolites carrying MMACHC variant sites. It is likely that adequate intake of vitamin B12 during pregnancy and maternal vitamin B12 levels are associated with a better neonatal vitamin B12 status (Aparicio et al., 2020; Li et al., 2021). To analyze the variant types more accurately, we showed that the conservative scores for all three missense variants (c.323G>A, c.1663G>A, and c.1280G>A) were greater than 7. Compared to the wild-type, the free energy of the combination caused by the above three genetic variations was significantly increased. It is speculated that the increased metabolite levels might be related to the activity of mutant methylmalonate coenzyme A transferase (MCM). However, quantitative analysis of enzyme activity and protein structure stability of heterozygous variants in MMUT needs further verification in future studies in view of MMA expression intensity in the liver, kidney, and brain of the Mut animal model (Forny et al., 2016). Combined with the results of the quantitative analysis of enzyme activity changes in MMUT variants, we expect to establish a relative quantitative prediction of metabolites in lesion tissues using a genome metabolism model as the structure in the future.

Based on neonatal metabolic data, artificial intelligence (Random Forest Machine Learning Class Fiber) trained by the Yale University School of Medicine and Stanford Genome Technology reduced the false positive rate of MMA by more than 51%, making it an efficient and cheap auxiliary screening tool (Peng et al., 2019). Therefore, we comprehensively analyzed the metabolic biomarkers, gene sequencing data, and silico analysis of 967 MMUT and MMACHC gene carriers during the neonatal period to predict VUS pathogenicity. The details are as follows. 1) Mass spectrometry data showed that the C3, C3/C0, and C3/C2 levels of newborns with VUS (c.1159A>C and c.1286A>G) were close to those of newborns carrying P/LP variant group and significantly higher than those of newborns with non-variants and the remaining VUS newborns in our study. 2) According to PANTHER, PolyPhen-2, SIFT, FATHMM, and DeepDGG biological tools, these two variants affect protein structure stability and damage biological functions. The conservative scores used to verify the degree of damage caused by VUS (c.1159A>C and c.1286A>G) were 9 and 7, respectively. Molecular docking showed that the free energy of variant ligand binding was significantly higher than that of wild-type ligand binding. In addition, the C3, C3/C0, C3/C2, and MMA levels of three patients with c.1159A>C variation in our clinical database were significantly higher. All the patients died from severe metabolic disorders during the neonatal period. This provides a theoretical basis for the pathogenic properties of c.1159A>C discovered previously by our research group ^[10]. However, the c.1286A>G variants have not been previously reported, and experimental evidence to identify their biological functions is not available. Nucleotide site-directed mutagenesis screening, which can accurately locate and induce point variants in any genome and identify cell clones with target variants, has been used to explore the functional analysis of Lynch syndrome-related mismatch repair gene variations with 97.6% specificity and 92.3% sensitivity (Houlleberghs et al., 2016). This study provides technical support for further clarification of the pathogenicity of MMA-related VUS, which is beneficial for high-throughput research on its pathogenesis and drug targeting.

We found for the first time that the C3, C3/C0, and C3/C2 levels in pathogenic/likely pathogenic MMUT gene variant carriers were higher than those in non-variant carriers. The c.1159A>C and c.1286A>G variants may be pathogenic based on genomic, tandem mass spectrometry data, and bioinformatics analyses. Therefore, a comprehensive analysis based on genomic, carrier metabolite, and bioinformatics analyses may be a new method for interpreting variant pathogenicity, which is important for early MMA diagnosis and precise treatment and can thus improve patient prognosis and the health of the whole population.

Our study had several limitations. 1) The data were mostly obtained from southern China and may not be representative of the general population. Therefore, we could not establish effective critical values for MMUT and MMACHC pathogenic variant carrier-related metabolic screening indicators in newborns. 2) We did not verify the physiological status of these carriers after identifying the pathogenicity of the variant site. 3) Considering that sillico analysis is primarily designed for missense variants, we were unable to predict the structural and functional outcomes following frameshift and nonsense variants. Moreover, it is difficult for this tool to predict the interactions with partner proteins, multimers, and glycosylation. Consequently, in future research we will add an in vitro experiment of primary cells and an in vivo experiment of transgenic mouse models to verify the pathogenicity of the variants.