Taurine/chenodeoxycholic acid ratio as a circulating biomarker of insidious vitamin B12 deficiency in humans

26 Deficiency of vitamin B 12 (B 12), an essential water-soluble vitamin, leads to irreversible 27 neurological damage, osteoporosis, cardiovascular diseases, and anemia. Clinical tests to detect B12 28 deficiency lack specificity and sensitivity. B 12 deficiency is thus insidious because progressive 29 decline in organ functions may go unnoticed until the damage is advanced or irreversible. Here, 30 using targeted unbiased metabolomic profiling in the sera of B 12-deficient versus control 31 individuals, we set out to identify biomarker(s) of B12 deficiency. Metabolomic profiling identified 32 77 metabolites, and Partial least squares discriminant-analysis (PLS-DA) and hierarchical 33 clustering analysis (HCA) showed a differential abundance in B 12-deficient sera of taurine, 34 xanthine, hypoxanthine, chenodeoxycholic acid, neopterin, and glycocholic acid. Random forest 35 (RF) multivariate analysis identified a taurine/chenodeoxycholic acid ratio, with an AUC score of 36 1, to be the best biomarker to predict B 12 deficiency. Mechanistically, B12 deficiency reshaped the 37 transcriptomic and metabolomic landscape of the cell identifying a downregulation of methionine, 38 taurine, urea cycle, and nucleotide metabolism, and an upregulation of Krebs cycle. Thus, we 39 propose taurine/chenodeoxycholic acid ratio in serum as a potential biomarker of B12 deficiency in 40 humans and elucidate cellular metabolic pathways regulated by B12 deficiency. 41 42 43 44 45 46 47 48 49 50 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 19, 2024. ; https://doi.org/10.1101/2024.02.15.580499doi: bioRxiv preprint 3

51 Vitamin B 12 (B 12) is an essential water-soluble vitamin derived from animal-based diets that 52 regulates a multitude of cellular processes in humans such as one-carbon metabolism and Krebs 53 cycle.(1-4) The absorption of dietary B12 requires gastric intrinsic factor (GIF), a stomach-specific 54 protein.(4) Gif binds to B 12 in the small intestine forming the GIF-B 12 complex. This complex is 55 endocytosed by the intestinal epithelial cells and B 12 is released into the bloodstream.(4) In the 56 bloodstream, B 12 binds to the protein transcobalamin 2, which then carries it to the liver, the 57 primary storage and recycling organ for B 12 in mammals.(5) Once acquired, humans, for instance, 58 can recycle B 12 to maintain B 12-dependent cellular processes for up to a decade.(2) In the cells, 59 B12–derivatives function as cofactors for only two known enzymes: methylmalonyl-CoA mutase 60 and methionine synthase, and through them affect a variety of downstream metabolic pathways 61 such as Krebs cycle, amino acid synthesis, and DNA and histone methylation. (1, 6) In humans, 62 decreased production of functional GIF protein or non-consumption of animal products causes B 12 63 deficiency and results in various abnormalities, such as anemia, osteoporosis, and cognitive 64 defects. (7-10) 65 66 In clinical practice, the diagnosis of B 12 deficiency is typically established by the measurement of 67 serum cobalamin (Cbl) levels.(11) Although B 12 deficiency can be reflected by elevated 68 methylmalonic acid (MMA) and homocysteine (Hcy) levels, these tests are not routinely used 69 unless the initial Cbl levels are equivocal because MMA and Hcy can be elevated in conditions 70 independent of B 12 levels.(12-16) Despite the importance of B 12 and its association with many 71 physiological functions, many issues remain unresolved in the diagnosis of B12 deficiency, leading 72 to poor diagnosis and irreversible consequences on the body.(17, 18) First, B 12 is a very stable 73 molecule and because 95-97% of B12 is stored in the liver, its serum levels do not accurately reflect 74 its actual functional levels i.e., the amount of B 12 required for maintaining body functions.(19) 75 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 19, 2024. ; https://doi.org/10.1101/2024.02.15.580499doi: bioRxiv preprint 4 Second, the cost of measurement of B 12 in patient samples, despite being not able to accurately 76 predict a B 12-deficient state, remains high and therefore is not the first line of measurement; 77 clinicians measure B 12 only when a patient presents signs of B 12 deficiency such as anemia to 78 confirm a deficient state. (20-23) These facts necessitate the need to identify molecules regulated 79 by B12, which can provide a functional readout of B12 deficiency in humans. 80 81 We recently created a transgenic mouse model of B 12 deficiency by deleting the gene essential for 82 B12 absorption from the gut, Gif, to understand the molecular consequences of B 12 deficiency. 83 These studies led to the identification that B 12 stored in the liver regulates the production of 84 taurine. Taurine is a semi-essential micronutrient that has recently been shown to be a driver of 85 aging as its supplementation increases healthy lifespan in diverse species from worms to mice, and 86 low taurine levels are associated with poor health in aged humans(24). In the B 12 mode of action, 87 taurine plays an important role as the reversal of taurine deficiency through daily oral taurine 88 administration was shown to fully rescue the consequences of B 12 deficiency(25). More 89 importantly, the targeted metabolomics analysis of liver tissue collected from control and B 12-90 deficient mice showed changes in a multitude of metabolites besides taurine that are secreted from 91 cells and could be detected in the serum(25). These studies suggested a plausible and testable 92 hypothesis that certain metabolites or sets of metabolites may exist which could serve as a readout 93 of, difficult to detect, B12-deficient state in humans. 94 95 The present study was initiated to test the above hypothesis by performing a metabolomic analysis 96 on serum samples collected from control and B 12-deficient individuals to identify which factor(s) 97 could serve as a biomarker of B 12-deficient state. Results showed that serum levels of certain 98 metabolites such as taurine, xanthine and hypoxanthine were dramatically downregulated in the 99 B12-deficient individuals. Using various downstream analyses, we suggest that taurine in 100 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 19, 2024. ; https://doi.org/10.1101/2024.02.15.580499doi: bioRxiv preprint 5 conjugation with chenodeoxycholic acid can serve as a biomarker of B12-deficient state in humans. 101 Furthermore, using mouse B 12-deficient tissues, we elucidate how despite only needed for 2 102 enzyme functions, B 12 deficiency alters the metabolic and transcriptomic landscape in the cells, 103 which will facilitate advances in further understanding biology of B12. 104 105

106 Study population, sample classification, acquisition, pre-processing, and normalization of 107 metabolomic data 108 A schematic diagram illustrating different steps of this study is presented in Figure 1. The samples 109 utilized in this study are from the Kuopio Ischaemic Heart Disease Risk Factor (KIHD) study 110 aimed at identifying the risk factors for coronary heart diseases, atherosclerosis, and other related 111 conditions in the Eastern Finnish population.(26) Sera were classified in accordance with 112 internationally established criterion into control subjects (n=13) with B 12 levels >250 pmol/L, and 113 into deficient subjects (n=8) with B12 levels <150 pmol/L.(1, 11, 17, 27) Samples were randomized 114 before metabolite extraction and quantified using a ACQUITY UPLC-MS/MS system. Ninety-four 115 metabolites could be detected in the sera, out of which 77 that passed quality control were selected 116 for further downstream analysis. Imputation of one missing value with the minimum value in that 117 cohort was done, and data was pre-processed by generalized log transformation (glog) and auto-118 scaling of metabolite concentration peaks in each sample to represent uniform distribution. 119 120 Identification of differentially expressed serum metabolites following B12 deficiency 121 We first performed a principal component analysis (PCA), an unsupervised multivariate analysis, 122 to group/classify samples without any consideration of prior classification to detect any outliers in 123 the two cohorts. The principal component 1 (PC1) accounted for 22.6% of the variance and PC2 124 accounted for 13.6% of the variance ( Figure 2A ). To identify differential concentration of each 125 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 19, 2024. ; https://doi.org/10.1101/2024.02.15.580499doi: bioRxiv preprint 6 metabolite between the control and B 12-deficient groups, we calculated the mean fold change and 126 performed t-tests to compare the mean of each metabolite. A metabolite was considered 127 significantly different between each group when the value of p ≤ 0.05 and log2 fold change ±0.5. 128 In the colvano plot the 3 blue dots in the upper left and 3 red dots in upper right quadrants 129 represent the most significantly altered metabolites in B 12-deficient subjects compared to that in 130 controls (Figure 2B). A hierarchical clustering analysis (HCA) of the metabolomic data using the 131 top 3 downregulated and top 3 upregulated metabolites showed well-defined clustering of thirteen 132 healthy subjects (pink, left cluster) versus eight B12-deficient subjects (green, right cluster) (Figure 133 2C). The control group showed high abundance (shades of red colour) of taurine, hypoxanthine 134 and xanthine compared to the B 12-deficient group, whereas the abundance of glycocholic acid, 135 neopterin and chenodeoxycholic acid was significantly higher in the B 12-deficient group as 136 compared to healthy controls (Figure 2C). Following the identification of differentially expressed 137 metabolites (DEMs), we did Metabolite Set Enrichment Analysis (MSEA) and Metabolomic 138 Pathway Analysis (MetPA) to determine the metabolic pathways that are associated with 139 differences in the abundance of identified metabolites, and perturbations of which is associated 140 with the B12 deficiency. The MSEA classified the 77 DEMs into 50 different metabolic pathways 141 (Figure 2D ) that include divergent cellular metabolism pathways such as bile acid biosynthesis, 142 amino acid biosynthesis, glucose metabolism, and nucleic acid synthesis, which are listed in the 143 order of descending fold enrichment ( Figure 2D). Out of the 50 listed pathways, the taurine and 144 hypotaurine metabolism pathway was the most enriched pathway with highest fold enrichment 145 value (-logP value ~6). MetPA results revealed that taurine and hypotaurine metabolism pathway 146 had the highest pathway impact value between the controls and B 12-deficient subjects, further 147 validating the importance of this pathway (Figure 2E). 148 149 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 19, 2024. ; https://doi.org/10.1101/2024.02.15.580499doi: bioRxiv preprint 7 Once we identified the most significant DEMs and major pathways to which these DEMs belonged 150 to, we wanted to check the consistency of identified DEMs as most discriminant variables for 151 classifying healthy controls versus B 12-deficient subjects. For this purpose, we performed a PLS-152 DA analysis that helps in highlighting whether a metabolite is upregulated or downregulated in a 153 group/sample by creating a latent structure, and the values of variable importance projection (VIP) 154 score which represent the importance of the metabolite in the PLS-DA model ( Figure 2F). The 155 VIP score plot (threshold of >1.0) revealed that taurine had the maximum score with low 156 abundance in B 12-deficient samples versus controls ( Figure 2F). The other metabolites that were 157 identified in volcano plot i.e., xanthine, hypoxanthine, chenodeoxycholic acid, neopterin, and 158 glycocholic acid also came up in PLS-DA plot, suggesting the consistency of these metabolites as 159 important DEMs in controls versus B 12-deficient subjects. Further, we performed univariate 160 analysis (t-test) on individual DEMs to determine the significant difference in the abundance of 161 each metabolite between the two groups. Based on the analysis, abundance of taurine ( p=0.002), 162 xanthine ( p=0.019) and hypoxanthine ( p=0.000) was significantly lower whereas the levels of 163 chenodeoxycholic acid (p = 0.063), neopterin ( p= 0.023), and glycocholic acid ( p= 0.027) was 164 significantly higher in sera of B 12-deficient subjects (green bars) compared to healthy controls 165 (pink bars) (Figure 2G). 166 167 Metabolites that belong to the same pathway tend to work in coherence. To this end, we subjected 168 the metabolite data to Pearson’s correlation matrix analysis to reveal any correlation that might 169 exist between the 77 identified metabolites or between 21 study subjects ( Figure S1A-B ). 170 Between the two cohorts, metabolites such as taurine, xanthine, and hypoxanthine were positively 171 correlated (red color) to each other and negatively correlated (blue color) to chenodeoxycholic 172 acid, neopterin, and glycocholic acid ( Figure S1A ). Moreover, there was a high positive 173 correlation observed between all the essential amino acids. This suggests a strong inter-relationship 174 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 19, 2024. ; https://doi.org/10.1101/2024.02.15.580499doi: bioRxiv preprint 8 between these metabolites which could be expected as these belong to same metabolic pathway 175 such as amino acid biosynthesis. Pearson’s correlation matrix analysis on the different cohort 176 subjects, however, revealed no significant trends (Figure S1B), suggesting no inter-relationship or 177 correlation between the samples, which negates the possibility of any biases in the sample 178 workflow. 179 180 Taken together, these multiple lines of evidence suggest that taurine, hypoxanthine, xanthine, 181 chenodeoxycholic acid, neopterin, and glycocholic acid are the most significant DEMs in the sera 182 of healthy controls versus B 12-deficient subjects. Pathway enrichment analysis further confirmed 183 that the alteration in taurine and hypoxanthine metabolic pathway is strongly associated with B 12 184 deficiency. 185 186 Selection and identification of metabolite and/or metabolite ratio as biomarker 187 To identify the best metabolite and/or metabolites ratio that could serve as a sensitive biomarker 188 for prediction of B12 deficiency, we subjected the data to two statistical analysis tools: Partial least 189 squares discriminant-analysis (PLS-DA) ( Figure 3A and 3E ) and Random forest (RF) analysis 190 (Figure 3C and 3G ). Multiple statistical models generated by these analyses were validated and 191 compared for their ability to identify the metabolite or metabolites ratio which can serve as the best 192 biomarker to predict B12 deficiency. All models generated by PLS-DA or RF were validated using 193 Receiver Operating Characteristic (ROC) analysis, in which Area Under the Curve (AUC) score 194 was used to monitor the sensitivity and specificity of a model (variable) in predicting the B 12 195 deficiency. Although both are predictive modelling tools, PLS-DA analysis has a tendency to 196 overfit even on completely random data as compared to RF analysis. Thus, the quality of the 197 models was further assessed using Monte-Carlo cross validation (MCCV) to create ROC curve for 198 every model generated from both PLS-DA and RF analysis. These models use a combination of 199 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 19, 2024. ; https://doi.org/10.1101/2024.02.15.580499doi: bioRxiv preprint 9 the most important features to build classification models, ranging from a minimum of 2 to a 200 maximum of 100. Since MCCV uses defined sub-sampling, 2/3 of the samples were used to 201 evaluate the feature importance and 1/3 of the samples were used for validation. This iterative 202 procedure was used to calculate the performance (AUC) and confidence interval of each model 203 and the one with AUC closest to 1 with low variability (CI) was considered to be the best model. 204 The software gave output in the form of ROC curves of top 6 models, referred to as variables, 205 based on the CV performance. we used the most significant DEMs (Figure 3A & C) or metabolite 206 ratio (Figure 3E & G) as top features to generate best 6 models for prediction of B 12 deficiency. 207 Note that the nomenclature of models (referred to as variables, hereinafter) is representative of the 208 number of features used to create the model. Figure 3B, D, F, and H represent the ROC curve for 209 the top 6 models obtained following PLS-DA and RF analysis, whereas the model numbers 1, 2, 3, 210 4, 5 and 6 represent the variables (Var.) 3 (red), 5 (green), 10 (blue), 20 (cyan), 28 (pink) and 77 211 (yellow), respectively, signifying that model 1 was created using 2 metabolites of top importance, 212 whereas model 6 used top 77 metabolites. 213 214 Both PLS-DA ( Figure 3A) and RF (Figure 3C ) analysis, using singular metabolites as features, 215 showed that models with more than 20 metabolites (38 and 77) have high AUC (> 7) and tight CI, 216 suggesting their potential to be better models, compared to those with fewer than 20 metabolites. A 217 higher score suggests better predictive ability of a model to identify the B 12-deficient state. The 218 feature ranking plot for both PLS-DA (Figure 3B) and RF (Figure 3D) analysis showed the top 15 219 metabolites arranged in descending order of average importance scores contributing to the model 220 accuracy. The average importance scores of hypoxanthine and taurine were among the top three 221 metabolites in both analyses, with hypoxanthine having the maximum score. Both models showed 222 lower (blue) abundance of taurine in B 12-deficient cohort, but the same was not true for 223 hypoxanthine. This was consistent with PLS-DA analysis done in Figure 2F . It is important to 224 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 19, 2024. ; https://doi.org/10.1101/2024.02.15.580499doi: bioRxiv preprint 10 note that (a) 7 of 15 top metabolites were different between the models generated by PLS-DA and 225 RF and (b) the individual average importance score for the 8 identical metabolites varied in the 226 two analyses. This suggested that both analyses work on independent algorithms and there was no 227 bias in the selection of hypoxanthine and taurine as top metabolite biomarkers for predicting B 12 228 deficiency. 229 Next, we investigated whether abundance ratios of metabolite pairs could increase the sensitivity 230 o f P L S - D A a n d R F m o d e l s t o d e t e c t B12 deficiency (Figure 3C, 3D ). Ratios of all possible 231 metabolite pairs were computed, and top ranked ratios (based on p values) and top 20 were 232 included for biomarker analysis. Using abundance ratios of metabolite pair as a feature, both PLS-233 DA (Figure 3E) and RF (Figure 3G) models showed that all the top 6 models have high AUC (> 234 9) and high CI which were comparable, suggesting any model with more than 3 features was a 235 good model with high specificity and sensitively but high variability (scattered CI) as well. One-to-236 one comparison of AUC and CI scores for both the PLS-DA and RF models based on the 237 abundance ratios of metabolite pair versus singular metabolites revealed that the former can serve 238 as better biomarkers in predicting B12 deficiency. The feature ranking plot for models in Figure 3F 239 and Figure 3H listed 13 identical sets of metabolite pairs with taurine/chenodeoxycholic acid 240 gaining the highest average importance score in both ( Figure 3G-H ). The abundance for 241 taurine/chenodeoxycholic acid ratio however was reversed in the two models, being low (blue) in 242 PLS-DA and high (red) in RF for B 12-deficient group (Figure 3E, 3G). It is important to note that 243 this analysis was consistent with the previous analysis shown in Figure 2 (PCA, volcano plot, 244 PLS-DA and univariate analysis). 245 246 Together, results suggest that out of the metabolites identified to be differentially expressed 247 between healthy controls and B 12-deficient group taurine, hypoxanthine and the ratio of 248 taurine/chenodeoxycholic acid could serve as biomarkers for B12 deficiency. 249 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 19, 2024. ; https://doi.org/10.1101/2024.02.15.580499doi: bioRxiv preprint 11 250 Comparison of the abilities of taurine, hypoxanthine and taurine/chenodeoxycholic acid ratio 251 to predict B12-deficient state 252 We performed ROC analysis to further characterise the predictive ability of taurine alone, 253 hypoxanthine and taurine/chenodeoxycholic acid ratio, which were shortlisted from previous PLS-254 DA and RF analysis. The sensitivity and significance of taurine, hypoxanthine and 255 taurine/chenodeoxycholic acid in predicting B 12 deficiency is represented using AUC score from 256 ROC analysis (Figures 4A-C). The scaled concentration of the indicated metabolites are shown in 257 Figures 4D-F. This analysis showed that AUC for taurine/chenodeoxycholic abundance ratio was 258 1, which is equivalent to being a perfect diagnostic biomarker (Figure 4C). Furthermore, the AUC 259 and p-values for taurine/chenodeoxycholic acid ratio were the lowest ( p-value=5.3193E-7) in 260 comparison to hypoxanthine (AUC = 0.885, p-value =7.0513E-4) and taurine alone (AUC = 0.885, 261 p-value =0.002), suggesting that taurine/chenodeoxycholic ratio was the best variable as a 262 biomarker to predict B 12 deficiency compared to others. Between taurine and hypoxanthine, the 263 AUC scores were comparable, but hypoxanthine was significant in differentiating the two groups 264 because of lower p-value. 265 266 These results suggest that serum taurine/chenodeoxycholic acid abundance ratio can serve as a 267 diagnostic biomarker for predicting B12 deficiency with high specificity and sensitivity. 268 269 270 To further test the ability of RF using taurine alone or and in combination with other metabolites as 271 biomarker to predict B12 deficiency, we trained a RF model on train data using cross validation and 272 predicted on the test data. For unbiased assessment, equal number of samples (n=4/group) were 273 randomly selected from control and B 12-deficient group as hold-out samples. These samples were 274 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 19, 2024. ; https://doi.org/10.1101/2024.02.15.580499doi: bioRxiv preprint 12 not used for fitting process in the model but used as testing samples. The rest of the samples were 275 used as training samples to predict B12 deficiency. We compared predictive ability of taurine alone, 276 taurine and hypoxanthine, and ratio of taurine/chenodeoxycholic acid using AUC score (ROC 277 analysis), predicted class probabilities, and cross validation (CV) prediction ( Figure 5). Amongst 278 these model (Figure 5A, 5C, 5E) comparisons, taurine/chenodeoxycholic acid showed the highest 279 margin of separation between the control (empty grey circles, left edge of x-axis) and B12-deficient 280 (filled grey circles, right edge of x-axis) group in training set, ( Figure 5E ). Also, the hold-out 281 samples from both groups (control = empty red circles, B 12-deficient = red filled circles) fit 282 perfectly well with the corresponding group in testing data set. Moreover the ROC-AUC curve 283 showed that taurine/chenodeoxycholic abundance ratio had the highest accuracy (AUC CV=1, 284 AUC holdout =1, Figure 5F) in predicting B 12 deficiency compared to taurine alone (AUC CV = 285 0.665, AUC holdout=0.938, Figure 5B ) or hypoxanthine (AUC CV= 0.809, holdout=0.938, 286 Figure 5D). Overall, this analysis was consistent with previous RF analysis, suggesting towards 287 great potential of taurine/chenodeoxycholic acid to serve as serum biomarker for predicting B 12 288 deficiency. 289 290 Metabo-transcriptomic network analysis linked B 12-dependent reactions with 291 taurine/chenodeoxycholic acid. 292 We performed a network analysis of differentially expressed genes and metabolites 293 between controls and B12-deficient livers in a mouse model of B 12 deficiency reported previously 294 by us.(28) Liver is a suitable tissue to investigate effects of B 12 deficiency since it is one of the 295 principal site of B 12 storage, and we demonstrated earlier that B 12 deficiency compromises its 296 functions.(28) In the cells, B 12 is known thus far to be converted into two cofactors (methyl-B 12 297 and adenosyl-B 12), which are required for the functioning of two known enzymes, methionine 298 synthase and methyl-malonyl CoA mutase.(29, 30) Thus, we focused our attention on metabolic 299 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 19, 2024. ; https://doi.org/10.1101/2024.02.15.580499doi: bioRxiv preprint 13 pathways that are interconnected with the B 12-derived cofactor-dependent reactions such as Krebs 300 cycle, amino acid metabolism, urea cycle, and nucleotide metabolism. 301 302 The network visualization of differentially expressed transcriptome showed that transcripts 303 encoding the enzymes that catalyze metabolite conversions in these pathways were overall 304 downregulated (in blue), except for the Krebs cycle, in which expression of 5 out of 9 enzymes 305 was upregulated (in red) ( Figure 6 ). This upregulation in the expression levels of Krebs cycle 306 enzymes could be linked to decreased activity of methyl-malonyl CoA mutase (Mut), which is 307 dependent on the adenosyl-B 12 for its activity. Mut catalyzes the synthesis of Succinyl-CoA, an 308 intermediate in the Krebs cycle that plays a critical role in providing protons for the OXPHOS 309 system, and thus, energy production in the cells. B 12 deficiency leads to an energy deficit in the 310 cells, and consequently likely, a compensatory increase in the expression levels of enzymes in the 311 Krebs cycle. However, no react ions surrounding the adenosyl-B 12-dependent Mut enzyme and 312 Krebs cycle could relate to known taurine biosynthetic machinery in B12-deficient cells. 313 314 An analysis of reactions surrounding methionine synthase (Mtr), the second enzyme that is 315 dependent on the methyl-B 12 as a cofactor, showed that the concentrations of methionine, the 316 downstream product, were decreased while concentrations of its precursor, homocysteine, were 317 increased (Figure 6 ). Expression levels of the enzymes in the methionine cycle were either not 318 affected or were decreased. The methionine cycle is linked to cysteine synthesis in the cells and 319 through a relay of changes, to taurine biosynthesis. Most of the enzymes and their downstream 320 products in this pathway were downregulated, consequently leading to deficiency of multiple 321 metabolites in taurine metabolic pathway (taurine, taurocholate, tauro-chenodeoxycholate) (Figure 322 6). The expression levels of the enzyme, Csad, that catalyzes the rate limiting step in taurine 323 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 19, 2024. ; https://doi.org/10.1101/2024.02.15.580499doi: bioRxiv preprint 14 biosynthesis, was increased likely as a compensatory mechanism due to deficiency of taurine 324 (Figure 6). 325 326 Further analysis of gene-metabolite networks interconnected with B12-dependent reactions showed 327 that gene expression of enzymes and metabolite intermediates in the urea cycle were 328 downregulated. In the amino acid metabolism pathway, barring tryptophan metabolite, HIAA and 329 NAD+ pathways, all enzyme expressions and metabolite intermediates were downregulated. In the 330 nucleotide metabolism pathways, metabolite intermediates were either downregulated or not 331 affected, and apart from a few enzymes, most of the enzyme expressions were downregulated. 332 333 Together, these integrated metabolomic and transcriptomic analyses in the WT and B 12-deficient 334 liver samples revealed global downregulation of metabolic networks upon B 12 deficiency and 335 identified a hitherto unanticipated connectivity between B 12-dependent reactions and taurine 336 metabolism. 337 338

339 By using metabolomic analysis of serum from controls and B 12-deficient subjects, we were able to 340 identify that a ratio of taurine/chenodeoxycholic acid levels can serve as a biomarker of, difficult 341 to detect, B 12 deficiency. The quantitative metabolomic analysis of 77 relevant metabolites in the 342 sera of B 12-deficient patients revealed that most of the metabolites were downregulated and are 343 involved in metabolism of amino acids, betaine, glutathione, bile acid, and purines ( Figure 2 ). 344 Metabolite set enrichment analysis on the perturbed metabolite profiles showed alterations in the 345 metabolic pathways associated with amino acid and methionine metabolism ( Figure 1 ). 346 Downregulation in methionine levels in this metabolome is consistent with the role of B 12 as an 347 essential cofactor of methionine synthase, while homocysteine accumulated from the dysfunction 348 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 19, 2024. ; https://doi.org/10.1101/2024.02.15.580499doi: bioRxiv preprint 15 of methionine synthase was 1.8-fold elevated. Furthermore, univariate analysis of the B12-deficient 349 metabolome identified a differential abundance of taurine, hypoxanthine, and xanthine between the 350 two groups. The multivariate random forest (RF) analysis aimed towards identifying which 351 metabolite(s) contributed to the separation of the two groups with higher specificity and sensitivity 352 showed taurine/chenodeoxycholic ratio as the metabolic parameter that could separate the two 353 groups with 99% accuracy. Thus, we propose taurine/chenodeoxycholic acid ratio as a potential 354 biomarker of a B12-deficient state in humans. 355 Previous studies have characterized the human serum metabolome in B 12-deficient subjects in an 356 attempt to reveal connections between B 12-deficient state and serum metabolic markers. Alex et 357 al., performed metabolomic profiles in sera of Chilean older adults with subclinical borderline B 12 358 deficiency (defined by serum B 12 <148 pmol/L, holotranscobalamin 15 359 μ mol/L, or MMA >271 nmol/L).(31) Although, this study showed perturbations in multiple 360 metabolite such as acylcarnitine and plasmalogens Authors did not subject their data to 361 downstream algorithms to identify potential biomarkers of B 12 levels. Moreover, the previous 362 study did not include a control group, whereas our study has a well-defined control group. 363 Although, these studies provide evidence that serum metabolome is altered by B 12 deficiency it 364 was unknown whether any of the metabolites of set of metabolites could serve as a biomarker of 365 B12-deficient state. Our study fills this gap in our knowledge and elucidates the effect of B 12 366 deficiency on the cellular, metabolic and transcriptomic landscape of the cell using liver biopsies 367 from a B 12-deficient mouse model. Together, these studies pave a way towards better 368 understanding of the cellular defects caused by B12 deficiency. 369 370 We acknowledge that our study has certain limitations. Firstly, the small sample size limits the 371 statistical power of the RF models. Repeating the same study in a larger sample size may allow a 372 greater number of metabolites to pass quality control for downstream analysis. Secondly, the 373 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 19, 2024. ; https://doi.org/10.1101/2024.02.15.580499doi: bioRxiv preprint 16 current study population was only tested for B 12 deficiency, which does not rule out the possibility 374 of deficiency of other vitamins or nutrients in the study population. These and other questions will 375 need to be addressed in future studies. 376 377 Vitamin B 12 deficiency leads to perturbed levels of taurine, hypoxanthine, xanthine, 378 chenodeoxycholic acid, neopterin, and glycocholic acid. We show that taurine levels alone and 379 taurine/chenodeoxycholic acid ratio are promising candidates for serum metabolite-based 380 biomarkers to identify B 12 deficiency. The two critical metabolites identified in this study 381 regulated by B 12, taurine and chenodeoxycholic acid, belong to the taurine metabolic pathway. 382 Taurine metabolism gets compromised with age and leads to taurine deficiency in humans, 383 however, the cause of this deficiency is unknown(24). The present study identifies vitamin B 12 as 384 the very first upstream regulator of taurine metabolism in aged humans and illustrates the 385 transcriptomic and metabolomic changes through which B 12 regulates this process. These results 386 are significant given that taurine deficiency has recently been shown to be a driver of aging in 387 diverse species, and is associated with poor health in humans. This study paves a way for future 388 clinical work to streamline diagnostic tools to detect B12 deficiency through a simple blood test and 389 perhaps other age-associated diseases. 390 391 DISCLOSURE STATEMENT 392

We thank Research Support Facility staff at Sanger Institute and 393 National Institute of Immunology especially Dr. P. Nagarajan for assistance with animal 394 experiments. 395 Financial Support: This work was supported by Wellcome Trust grant (09851) to VKY 396 and and a core Grant from National Institute of Immunology to VKY. 397 Conflict of Interest: Authors declare no conflict of interest 398 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 19, 2024. ; https://doi.org/10.1101/2024.02.15.580499doi: bioRxiv preprint 17 Authorship: All authors have seen and approved the manuscript 399 400

401 Chemicals and reagents 402 All the metabolite standards, ammonium formate, ammonium acetate and ammonium hydroxide 403 were obtained from Sigma-Aldrich (Helsinki, Finland). Formic acid (FA), 2-proponol, acetonitrile 404 (ACN), and methanol (all HiPerSolv CHROMANORM, HPLC grade, BDH Prolabo) were 405 purchased from VWR International (Helsinki, Finland). Isotopically labelled internal standards 406 were obtained from Cambridge Isotope Laboratory. Inc., USA (Ordered from Euriso-Top, France). 407 Deionized Milli-Q water up to a resistivity of 18 MΩ /i3 cm was purified with a purification system 408 (Barnstead EASYpure RoDi ultrapure water purification system, Thermo scientific, Ohio, USA). 409 410 Metabolite extraction protocol 411 The working calibration solutions were prepared in 96-well plate by serial dilution of the stock 412 calibration mix using Hamilton’s MICROLAB® STAR line (Hamilton, Bonaduz AG, 413 Switzerland) liquid handling robot system. Starting from a stock solution mix, 10 additional lower 414 working solutions were prepared using water as the diluent to build the calibration curves. 415 416 Clinical serum samples: 417 Clinical samples used for assessing the changes in vitamin B 12 levels and metabolites in blood 418 were obtained from the Kuopio Ischaemic Heart Disease Risk Factor Study (KIHD study), a 419 population-based cohort study described previously (25, 32), and were donated by J. Kauhanen 420 and T. Nurmi (University of Eastern Finland, Kuopio, Finland). Ten microliters of labelled internal 421 standard mixture was added to 100 μ L of serum sample. Metabolites were extracted by adding 4 422 parts (1:4, sample: extraction solvent) of the 100% ACN + 1% FA solvent. The collected extracts 423 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 19, 2024. ; https://doi.org/10.1101/2024.02.15.580499doi: bioRxiv preprint 18 were dispensed in OstroTM 96-well plate (Waters Corporation, Milford, USA) and filtered by 424 applying vacuum at a delta pressure of 300-400 mbar for 2.5 min on robot’s vacuum station. This 425 resulted a cleaner extract to the 96-well collection plate, which was placed under the OstroTM 426 plate. The collection plate was sealed with the cap map and placed in auto-sampler of the LC 427 system for the injection. 428 429 Instrumentation and analytical conditions 430 Sample analysis was performed on an ACQUITY UPLC-MS/MS system (Waters Corporation, 431 Milford, MA, USA). The auto-sampler was set at 5°C, and the column, 2.1 × 100 mm Acquity 432 1.7um BEH amide HILIC column (Waters Corporation, Milford, MA, USA), temperature was 433 maintained at 45°C. The total run time is 14.5 min including 2.5 min of equilibration step at a flow 434 rate of 600 μ L/min. Initially the gradient started with a 2.5 min isocratic step at 100% mobile 435 phase B (ACN/ H2O, 90/10 (v/v), 20 mM ammonium formate, pH at 3), and then rising to 100% 436 mobile phase A (ACN/H2O, 50/50 (v/v), ammonium formate, pH at 3) over the next 10 min and 437 maintained for 2min at 100% A and finally equilibrated to the initial conditions for 2.5 min. An 438 injection volume of 5 μ L of sample extract was used and two cycles of 300 μ L of strong wash 439 (methanol/isopropanol/ACN/H2O, 25/25/25/25, 0.5% FA) and 900 μ L of weak wash 440 (methanol/isopropanol/ACN/H2O, 25/25/25/25, 0.5% ammonium hydroxide) and in addition 2 441 min of seal wash (90/10, methanol/H2O) were carried out. The auto-sampler was used to perform 442 partial loop with needle overfill injections for the samples and standards. 443 444 The detection system, a Xevo® TQ-S tandem triple quadrupole mass spectrometer (Waters, 445 Milford, MA, USA), was operated in both positive and negative polarities with a polarity 446 switching time of 20 msec. Electro spray ionization (ESI) was chosen as the ionization mode with 447 a capillary voltage at 0.6 KV in both polarities. The source temperature and desolvation 448 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 19, 2024. ; https://doi.org/10.1101/2024.02.15.580499doi: bioRxiv preprint 19 temperature of 120°C and 650°C, respectively, were maintained constantly throughout the 449 experiment. Declustering potential (DP) and collision energy (CE) were optimized for each 450 compound. High pure nitrogen and argon gas were used as desolvation gas (1000 L/hr) and 451 collision gas (0.15 ml/min), respectively. Multiple Reaction Monitoring (MRM) acquisition mode 452 was selected for quantification of metabolites with individual span time of 0.1 sec given in their 453 individual MRM channels. The dwell time was calculated automatically by the software based on 454 the region of the retention time window, number of MRM functions and depending on the number 455 of data points required to form the peak. MassLynx 4.1 software was used for data acquisition, 456 data handling and instrument control. Data processing was done using TargetLynx software and 457 metabolites were quantified by using labelled internal standards and external calibration curves. 458 459 Data analysis using MetaboAnalyst 5.0 software and downstream analysis. 460 The raw data was analyzed using MetaboAnalyst 5.0 software (https://www.metaboanalyst.ca/). 461 (33, 34) Metabolite raw values were generalized log (glog) transformed and auto-scaled (mean-462 centered and divided by the standard deviation of each variable).(35) Missing values for any 463 metabolites in the sample below the limit of detection were inputted with 1/5 of the minimum 464 positive value for each variable. Unsupervised Principal component analysis (PCA) was done to 465 differentially cluster the two groups.(36, 37) Hierarchical clustering and Pearson’s correlation 466 analysis were also performed to cluster the metabolite and sample data in the form of a heatmap to 467 easily identify patterns in metabolite concentrations across samples. Metabolite Set Enrichment 468 Analyses (MSEA)(38) were performed on all metabolites with a VIP ≥ 1.5 that matched the 469 database using the “Pathway-associated metabolite sets (SMPDB)” database in the MetaboAnalyst 470 software . Pathway analysis was performed using the “Homo sapiens (KEGG(39, 40))” database in 471 the MetaboAnalyst software. Interactive scatter plot with ‘Enrichment Factor’ as x axis and 472 ‘−log10(P)’ as y axis was generated for functional analysis to show the significance of top 50 473 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 19, 2024. ; https://doi.org/10.1101/2024.02.15.580499doi: bioRxiv preprint 20 metabolic pathways involving the metabolites identified. The variable importance to projection 474 (VIP) score for each metabolite was calculated to quantitatively represent metabolite feature 475 importance in the model. A volcano plot scatterplot that shows statistical significance (-log10(p-476 value) versus magnitude of change (log 2-fold change) of metabolites. Metabolites that show 477 significant (p ≤ 0.05) change (log 2-fold change ±0.5) are highlighted. Multivariate supervised 478 Partial least squares discriminant analysis (PLS‐ DA) and Random-forest (RF) analysis were 479 performed to assess the difference between the abundance of top metabolites or metabolite ratio 480 between the two groups. The area under the curve (AUC) of the receiver operating characteristic 481 (ROC) curve was also calculated for each metabolite to determine its predictive ability as a 482 biomarker. The ROC curve is a plot of false positive rate (FPR) vs the true positive rate (TPR). 483 The higher the AUC value, the better the measurements are at classifying between the two groups. 484 485 486

487 1. Stabler SP. Vitamin B12 deficiency. New England Journal of Medicine. 488 2013;368(2):149-60. 489 2. Lazar MA, Birnbaum MJ. De-meaning of metabolism. Science. 490 2012;336(6089):1651-2. 491 3. Hunt A, Harrington D, Robinson S. Vitamin B12 deficiency. Bmj. 2014;349. 492 4. Nielsen MJ, Rasmussen MR, Andersen CB, et al. Vitamin B12 transport from food 493 to the body's cells—a sophisticated, multistep pathway. Nature reviews Gastroenterology & 494 hepatology. 2012;9(6):345-54. 495 5. Quadros EV, Regec AL, Khan KF, et al. Transcobalamin II synthesized in the 496 intestinal villi facilitates transfer of cobalamin to the portal blood. American Journal of 497 Physiology-Gastrointestinal and Liver Physiology. 1999;277(1):G161-G6. 498 6. Kalhan SC, Marczewski SE. Methionine, homocysteine, one carbon metabolism 499 and fetal growth. Reviews in Endocrine and Metabolic Disorders. 2012;13:109-19. 500 7. Stabler SP, Allen RH. Vitamin B12 deficiency as a worldwide problem. Annu Rev 501 Nutr. 2004;24:299-326. 502 8. Van Meurs JB, Dhonukshe-Rutten RA, Pluijm SM, et al. Homocysteine levels and 503 the risk of osteoporotic fracture. New England Journal of Medicine. 2004;350(20):2033-41. 504 9. McLean RR, Jacques PF, Selhub J, et al. Homocysteine as a predictive factor for hip 505 fracture in older persons. New England Journal of Medicine. 2004;350(20):2042-9. 506 10. Toh B-H, van Driel IR, Gleeson PA. Pernicious anemia. New England Journal of 507 Medicine. 1997;337(20):1441-8. 508 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 19, 2024. ; https://doi.org/10.1101/2024.02.15.580499doi: bioRxiv preprint 21 11. Stabler SP, Allen RH, Savage DG, et al. Clinical spectrum and diagnosis of 509 cobalamin deficiency. Blood. 1990;76(5):871-81. 510 12. Stabler SP, Marcell PD, Podell ER, et al. Elevation of total homocysteine in the 511 serum of patients with cobalamin or folate deficiency detected by capillary gas 512 chromatography-mass spectrometry. The Journal of clinical investigation. 1988;81(2):466-513 74. 514 13. Stabler S. Megaloblastic anemias: pernicious anemia and folate deficiency. Clinical 515 hematology Philadelphia: Mosby. 2006:242-51. 516 14. Klee GG. Cobalamin and folate evaluation: measurement of methylmalonic acid 517 and homocysteine vs vitamin B(12) and folate. Clin Chem. 2000;46(8 Pt 2):1277-83. 518 15. Lindenbaum J, Savage DG, Stabler SP, et al Diagnosis of cobalamin deficiency: II. 519 Relative sensitivities of serum cobalamin, methylmalonic acid, and total homocysteine 520 concentrations. Am J Hematol. 1990;34(2):99-107. 521 16. Nygard O, Nordrehaug JE, Refsum H, et al. Plasma homocysteine levels and 522 mortality in patients with coronary artery disease. N Engl J Med. 1997;337(4):230-6. 523 17. Lindenbaum J, Healton EB, Savage DG, et al. Neuropsychiatric disorders caused by 524 cobalamin deficiency in the absence of anemia or macrocytosis. New England journal of 525 medicine. 1988;318(26):1720-8. 526 18. Savage DG, Lindenbaum J, Stabler SP, et al. Sensitivity of serum methylmalonic 527 acid and total homocysteine determinations for diagnosing cobalamin and folate 528 deficiencies. The American journal of medicine. 1994;96(3):239-46. 529 19. Lloyd-Wright Z, Hvas A-M, Møller J, et al. Holotranscobalamin as an indicator of 530 dietary vitamin B12 deficiency. Clinical Chemistry. 2003;49(12):2076-8. 531 20. Dhonukshe /i3 Rutten RA, Pluijm SM, De Groot LC, et al. Van Staveren WA. 532 Homocysteine and vitamin B12 status relate to bone turnover markers, broadband 533 ultrasound attenuation, and fractures in healthy elderly people. Journal of Bone and Mineral 534 Research. 2005;20(6):921-9. 535 21. Goerss JB, Kim CH, Atkinson EJ, et al. Risk of fractures in patients with pernicious 536 anemia. Journal of Bone and Mineral Research. 1992;7(5):573-9. 537 22. Tucker KL, Hannan MT, Qiao N, et al. Low plasma vitamin B12 is associated with 538 lower BMD: the Framingham Osteoporosis Study. Journal of Bone and Mineral Research. 539 2005;20(1):152-8. 540 23. Morris MC, Evans DA, Bienias JL, et al. Dietary folate and vitamin B12 intake and 541 cognitive decline among community-dwelling older persons. Archives of neurology. 542 2005;62(4):641-5. 543 24. Singh P, Gollapalli K, Mangiola S, et al. Taurine deficiency as a driver of aging. 544 Science. 2023;380(6649):eabn9257. 545 25. Roman-Garcia P, Quiros-Gonzalez I, Mottram L, et al. Vitamin B(1)(2)-dependent 546 taurine synthesis regulates growth and bone mass. J Clin Invest. 2014;124(7):2988-3002. 547 26. Jussi K. Kuopio Ischemic Heart Disease Risk Factor Study. Encyclopedia of 548 Behavioral Medicine. 2020:1271-2. 549 27. Pennypacker LC, Allen RH, Kelly JP, et al. High prevalence of cobalamin 550 deficiency in elderly outpatients. J Am Geriatr Soc. 1992;40(12):1197-204. 551 28. Roman-Garcia P, Quiros-Gonzalez I, Mottram L, et al. Vitamin B 12–dependent 552 taurine synthesis regulates growth and bone mass. The Journal of clinical investigation. 553 2014;124(7):2988-3002. 554 29. Nielsen MJ, Rasmussen MR, Andersen CB, et al. Vitamin B12 transport from food 555 to the body's cells--a sophisticated, multistep pathway. Nat Rev Gastroenterol Hepatol. 556 2012;9(6):345-54. 557 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 19, 2024. ; https://doi.org/10.1101/2024.02.15.580499doi: bioRxiv preprint 22 30. Clemens TL. Vitamin B12 deficiency and bone health. N Engl J Med. 558 2014;371(10):963-4. 559 31. Brito A, Grapov D, Fahrmann J, et al. The human serum metabolome of vitamin B-560 12 deficiency and repletion, and associations with neurological function in elderly adults. 561 The Journal of Nutrition. 2017;147(10):1839-49. 562 32. Vanharanta M, Voutilainen S, Rissanen TH, et al. Risk of cardiovascular disease-563 related and all-cause death according to serum concentrations of enterolactone: Kuopio 564 Ischaemic Heart Disease Risk Factor Study. Arch Intern Med. 2003;163(9):1099-104. 565 33. Pang Z, Zhou G, Ewald J, et al. Using MetaboAnalyst 5.0 for LC-HRMS spectra 566 processing, multi-omics integration and covariate adjustment of global metabolomics data. 567 Nat Protoc. 2022;17(8):1735-61. 568 34. Pang Z, Chong J, Zhou G, et al. MetaboAnalyst 5.0: narrowing the gap between raw 569 spectra and functional insights. Nucleic acids research. 2021;49(W1):W388-W96. 570 35. Saccenti E, Hoefsloot HC, Smilde AK, et al. Reflections on univariate and 571 multivariate analysis of metabolomics data. Metabolomics. 2014;10:361-74. 572 36. Worley B, Powers R. Multivariate analysis in metabolomics. Current metabolomics. 573 2013;1(1):92-107. 574 37. Jolliffe IT. Principal component analysis for special types of data: Springer; 2002. 575 38. Xia J, Wishart DS. Metabolomic data processing, analysis, and interpretation using 576 MetaboAnalyst. Current protocols in bioinformatics. 2011;34(1):14.0. 1-.0. 48. 577 39. Ogata H, Goto S, Sato K, et al. KEGG: Kyoto encyclopedia of genes and genomes. 578 Nucleic acids research. 1999;27(1):29-34. 579 40. Kanehisa M, Goto S. KEGG: kyoto encyclopedia of genes and genomes. Nucleic 580 acids research. 2000;28(1):27-30. 581 582 583 FIGURE LEGENDS: 584 FIGURE 1: 585 Study population, sample classification, acquisition, pre-processing, and normalization of 586 metabolomic data. Schematic diagram illustrating the steps for metabolomic analysis of serum 587 samples from B 12-deficient (B12 levels <150 pmol/L) versus the healthy control group. (1) In this 588 study, 8 and 13 subjects were grouped in B 12-deficient and control groups (age- and gender-589 matched), respectively, (2) blood samples were collected and processed, (3) metabolomics data 590 was acquired from serum samples using ACQUITY UPLC-MS/MS system (Waters Corporation, 591 Milford, MA, USA), data was pre-processed and analyzed using MetaboAnalyst 5.0 to identify (4) 592 differentially expressed metabolites between 2 study groups, (5) serum metabolic biomarker for 593 Vitamin B12 deficiency followed by (6) pathway analysis. 594 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 19, 2024. ; https://doi.org/10.1101/2024.02.15.580499doi: bioRxiv preprint 23 595 596 597 FIGURE 2: 598 Identification of differentially expressed serum metabolites following B 12 deficiency . (A) 599 Unsupervised multivariate PCA plot showing the spread of control (pink dots) versus B12-deficient 600 (green dots) cohort based on the serum metabolic profile. The horizontal and vertical coordinates 601 are the first and second principal components, respectively. Each dot represents a sample. (B) 602 Volcano plot showing six (blue and red dots) most significant differentially expressed metabolites 603 between the B 12-deficient patients versus controls, with a p-value < 0.05 and a log2 fold change 604 ±0.5. X-axis corresponds to log2(Fold Change) and Y-axis to −log10(p-value). (C) Hierarchical 605 clustering analysis sorted the control (pink) versus B 12-deficient (green) group based on 606 differential abundance of six metabolites (taurine, hypoxanthine, xanthine, glycocholic acid, 607 neopterin, and chenodeoxycholic acid). Relative abundance scored from 4 (highest, red color) to -4 608 (lowest, blue). (D) MSEA plot with top 50 enriched metabolic pathways (vertical-axis) to which 609 the 77 identified metabolites belong. The pathways are arranged in descending order of fold 610 enrichment score (horizontal axis) where the highest is 6 (red color) and lowest is 0 (yellow color) 611 (E) MetPA plot showing most enriched pathways with significance (-logP) values for each of the 612 pathway as dots of red (high significance) or yellow (low significance). X-axis corresponds to 613 pathway impact and Y-axis to -logP values. The size of the dot represents its impact value. (F) VIP 614 score plot from PLS-DA analysis showing the top 20 differentially expressed metabolites in serum 615 of control versus B 12-deficient group scored from 1 to 2. Relative abundance is depicted with red 616 (highest) and green (lowest) color. (G) Box plots showing normalized concentrations of individual 617 metabolites following univariate analysis: taurine (p=0.002), xanthine (p=0.019) and hypoxanthine 618 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 19, 2024. ; https://doi.org/10.1101/2024.02.15.580499doi: bioRxiv preprint 24 (p=0.000), chenodeoxycholic acid (p=0.063), neopterin (p=0.023), and glycocholic acid (p=0.027) 619 in the sera of control (red) versus B12-deficient (green) groups. 620 621 622 FIGURE 3: 623 Selection and identification of metabolite and/or metabolite ratio as a biomarker . The top 6 624 predictive models (Var.) generated by various multivariant analyses were compared for their 625 performance as metabolite biomarker predictors for B 12 deficiency using ROC-AUC curves based 626 on the MCCV method. ROC-AUC curve for (A) PLS-DA and (C) RF models using singular 627 metabolites as features. ROC-AUC curve for € PLS-DA and (G) RF models using abundance ratio 628 of metabolite pairs as features. Feature ranking plot for (B) PLS-DA and (D) RF models 629 representing the top 15 metabolites arranged in descending value of average importance score. The 630 average importance scores range from 1 to 2 for PLS-DA and 0 to 2 for RF. Feature ranking plot 631 for (F) PLS-DA and (H) RF models representing top 15 abundance ratio of metabolite pairs 632 arranged in descending value of average importance score. The average importance score ranges 633 from 1 to 2 for PLS-DA and 1 to 4 for RF. In all the feature ranking plots the relative abundance of 634 each feature between the control and B 12-deficient group was graded with red and blue colors 635 representing high and low abundance, respectively. 636 637 FIGURE 4: 638 Comparison of the abilities of taurine, hypoxanthine and taurine/chenodeoxycholic acid ratio 639 to predict B 12-deficient state. ROC-AUC curve showing performance of (A) taurine, (B) 640 hypoxanthine and (C) taurine/chenodeoxycholic acid ratio as biomarker to predict B 12 deficiency 641 based on AUC (sensitivity, specificity) and CI (variability) values. Each ROC curve is a plot 642 between false positive rate (x-axis) and true positive rate (y-axis). Box plots showing normalized 643 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 19, 2024. ; https://doi.org/10.1101/2024.02.15.580499doi: bioRxiv preprint 25 concentration of (D) taurin€(E) hypoxanthine and (F) taurine/chenodeoxycholic acid ratio between 644 control (pink) versus B 12-deficient (green) group. Each dot represents a sample. Y-axis represents 645 fold change values. P value <0.05. 646 647 FIGURE 5: 648 Statistical Model to test predictive ability of taurine alone and in combination as biomarker. 649 Random forest was used as a model to test the predictive abilities of taurine, taurine and 650 hypoxanthine together, and taurine/chenodeoxycholic acid ratio to predict B 12 deficiency. 651 Predicted class probability plot for (A) taurine, (B) taurine and hypoxanthine together, and (C) 652 taurine/ chenodeoxycholic acid ratio showing the classification accuracy of each factor to 653 differentiate between control (grey dots) and B 12-deficient (red dots) samples. The solid dots are 654 training data sets and the empty dots are test data sets. ROC-AUC curve analysis showing cross-655 validation (pink) and hold-out (blue) scores to determine the performance of (D) tau €e, (E) taurine 656 and hypoxanthine, and (F) taurine/chenodeoxycholic acid ratio as a biomarker to predict B 12 657 deficiency. Each ROC curve is a plot between the false positive rate (specificity) on the x-axis and 658 true positive rate (sensitivity) on the y-axis. 659 660 FIGURE 6: 661 Metabo-transcriptomic network analysis links B 12 dependent reactions with 662 taurine/chenodeoxycholic acid. Network analysis showing the differentially expressed genes and 663 metabolites between controls and B12-deficient livers in a mouse model of B 12 deficiency reported 664 previously(25). The network shows interactions between enzymes (italics font) and metabolites 665 (normal font) across various metabolic pathways in the liver such as Krebs cycle, urea cycle, 666 amino acid metabolism, nucleotide metabolism, etc. The arrows represent the direction of the 667 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 19, 2024. ; https://doi.org/10.1101/2024.02.15.580499doi: bioRxiv preprint 26 reaction. The downregulation and upregulation of enzyme transcript or metabolite concentrations 668 are represented by blue and red color, respectively. 669 670 FIGURE S1: 671 Correlation analysis between metabolites and samples. Pearson’s correlation matrix to identify 672 highly correlated (A) metabolites and (B) samples in two groups. Correlation score ranged from 1 673 (highest, red) to -1 (lowest, blue). 674 675 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 19, 2024. ; https://doi.org/10.1101/2024.02.15.580499doi: bioRxiv preprint Figure 1 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 19, 2024. ; https://doi.org/10.1101/2024.02.15.580499doi: bioRxiv preprint Component 2 (13.6%) 100-10 -10 0 10 Controls B12 Deficient A -2 -1 Taurine 0 1 Xanthine -2 -1 0 1 Hypoxanthine -2 -1 0 1 -2 -1 Chenodeoxycholic acid 0 1 Neopterin -1 0 2 Glycocholic acid -1 0 1 2 2 1 3 VIP Score 1.2 1.4 1.6 2.01.8 Control B12 deficient Taurine Neopterin 3-OH- Anthranilic Acid 2-Aminodipic Acid Threonine Cotinine Cytidine Isoleucine Cholic Acid Serine Leucine NAD Allantoin Glychocholic Acid Xanthine Chenodeoxycholic Acid Hypoxanthine Glutamic Acid Ornithine Inositol0 1 2 3 4 5 6 0.0 0.2 0.4 0.6 0.8 Pathway Impact -log(p) High Low Component 1 (22.6%) Group Taurine Hypoxanthine Xanthine Glycocholic Acid Neopterin Chenodeoxycholic Acid 4 0 2 -2 -4 KAUH IIVA MARI KOTT KAIN POHJ HEIN SORM VAIT BERG SUTI TUOM PUUR ROIV VAIN ALAN TEPP TURP VEHV KART KUOK 0-2 -1 0 1 2 0.5 1.0 1.5 2.0 2.5 3.0 3.5 P-value<0.05 FC=1.4 C D E F B G -log10(p) 0 2 4 6 Fold Enrichment Enrichment overview (top 50) Subjects Top differentially expressed metabolites Normalized concentration Figure 2 Log2FC Controls B12 Deficient Subjects .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 19, 2024. ; https://doi.org/10.1101/2024.02.15.580499doi: bioRxiv preprint A C ControlDeficient Hypoxanthine Taurine 3-Hydroxyanthranilic Chenodeoxycholic acid Neopterin Xanthine Aminoadipic acid Glycocholic acid Spermidine Cholic acid Allantoin L-Threonine NAD L-Isoleucine Ornithine Average Importance 1.2 1.4 1.6 1.8 Var. AUC CI 3 0.647 0.195-0.981 5 0.682 0.121-0.958 10 0.709 0.384-0.958 20 0.766 0.5-1 38 0.839 0.593-1 77 0.869 0.625-1 0.0 0.2 0.4 0.6 0.8 1.0 1-Specificity (False positive rate) 0.0 0.2 0.4 0.6 0.8 1.0 Sensitivity (True positive rate) ControlDeficient Hypoxanthine Chenodeoxycholic acid Taurine L-Serine 3-Hydroxyanthranilic Xanthine Cytosine Aminoadipic acid Cotinine Inosine Neopterin Sorbitol Glycocholic acid Inositol Phosphate L-Aspartic acid Average Importance 0.5 1.0 1.5 1-Specificity (False positive rate) 0.0 0.2 0.4 0.6 0.8 1.0 Sensitivity (True positive rate) Taurine/Chenodeoxych Neopterin/Allantoin Hypoxyanthine/NAD Taurine/Neopterin Taurine/Succinic aci Hypoxanthine/Chenode Hypoxanthine/4-Pyrid Taurine/L-Isoleucine Taurine/L-Leucine Taurine/Aminoadipic Hypoxanthine/Neopter Xanthine/Chenodeoxyc Hypoxanthine/Succini Hypoxanthine/Inosini Hypoxanthine/Sorbito 1.5 1.6 1.7 1.8 Taurine/Chenodeoxych Taurine/Neopterin Taurine/Succinic aci Hypoxanthine/NAD Taurine/Aminoadipic Neopterin/Allantoin Creatine/Hypoxanthin Hypoxanthine/Neopter Xanthine/Chenodeoxyc Hypoxanthine/Pantoth Hypoxanthine/Succini Hypoxanthine/4-Pyrid Taurine/L-Leucine Hypoxanthine/Chenode Hypoxanthine/Sorbito 1.5 2.0 2.5 3.0 3.5 1-Specificity (False positive rate) 0.0 0.2 0.4 0.6 0.8 1.0 Sensitivity (True positive rate) 1-Specificity (False positive rate) 0.0 0.2 0.4 0.6 0.8 1.0 Sensitivity (True positive rate) High Low PLSDA Random Forest PLSDA Random Forest ControlDeficient ControlDeficient Average Importance Average Importance 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 Var. AUC CI 3 0.662 0.227-0.912 5 0.689 0.292-0.949 10 0.693 0.246-0.991 20 0.731 0.38-0.991 38 0.736 0.31-1 77 0.77 0.371-0.991 Var. AUC CI 3 0.93 0.75-1 5 0.927 0.667-1 10 0.948 0.768-1 20 0.958 0.769-1 48 0.959 0.801-1 96 0.947 0.759-1 Var. AUC CI 3 0.929 0.718-1 5 0.952 0.843-1 10 0.955 0.859-1 20 0.958 0.843-1 48 0.960 0.875-1 96 0.956 0.833-1 E G B D F H Figure 3 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 19, 2024. ; https://doi.org/10.1101/2024.02.15.580499doi: bioRxiv preprint 0.0 0.2 0.4 0.6 0.8 1.0 1.00.80.60.40.20.0 AUC: 0.885 (0.721-0.981) -0.0403(0.7, 1) Taurine True Positive Rate 0.0 0.2 0.4 0.6 0.8 1.0 1.00.80.60.40.20.0 Hypoxanthine False Positive Rate 0.351(0.8, 0.9) 0.0 0.2 0.4 0.6 0.8 1.0 1.00.80.60.40.20.0 Taurine/Chenodeoxycholic acid -0.054(1,1) -2 -1 Taurine 0 1 Control B12 deficient -2 -1 Hypoxanthine 0 1 -2 -1 Taurine/Chenodeoxycholic acid 0 1 2 A D E F B C AUC: 0.885 (0.654-1) AUC: 1 (1-1) Figure 4 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 19, 2024. ; https://doi.org/10.1101/2024.02.15.580499doi: bioRxiv preprint 0.0 0.2 0.4 0.6 0.8 1.0 10-1 Taurine Samples Predicted Class Probabilities Control Deficient KAUH KART SORM TEPP BERG SUTI HEIN IIVA 0.0 0.2 0.4 0.6 0.8 1.0 10-1 Hypoxanthine Samples Predicted Class Probabilities IIVA POHJ HEIN BERG SUTI 0.0 0.2 0.4 0.6 0.8 1.0 10-1 Taurine/Chenodeoxycholic acid Samples Predicted Class Probabilities 0.0 0.2 0.4 0.6 0.8 1.0 0.0 Taurine Sensitivity (True positive rate) 1-Specificity (False positive rate) 0.2 0.4 0.6 0.6 1.0 Type AUC CV 0.665 Holdout 0.938 0.0 0.2 0.4 0.6 0.8 1.0 0.0 Hypoxanthine Sensitivity (True positive rate) 1-Specificity (False positive rate) 0.2 0.4 0.6 0.6 1.0 Type AUC CV 0.809 Holdout 0.938 0.0 0.2 0.4 0.6 0.8 1.0 0.0 Taurine/Chenodeoxycholic acid Sensitivity (True positive rate) 1-Specificity (False positive rate) 0.2 0.4 0.6 0.6 1.0 Type AUC CV 1 Holdout 1 Holdout VAIN ROIV KART ALAN VAIT KOTT IIVA POHJ Control Deficient Control Deficient A C E B D F Figure 5 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 19, 2024. ; https://doi.org/10.1101/2024.02.15.580499doi: bioRxiv preprint Propionyl-CoA D-Methylmalonyl-CoA L-Methylmalonyl-CoA Methyl-B12 Kreb’s Cycle Succinate Succinyl-CoA α-Ketoglutarate Isocitrate Citrate Oxaloacetate Malate Fumarate Acetyl-CoA Pyruvate Methionine Homocysteine S-Adenosyl- Methionine (SAM) S-Adenosyl- Homocysteine Mtr Adenosyl-B12 Serine Cystathione L-Cysteine L-Serine Aminoacrylate L-Cysteate Taurine Taurocheno- deoxycholateTaurocholate Cholate Glycine Sarcosine Dimethyl-Glycine Betaine Betaine-aldehyde CholineAcetoacetyl-CoA Acetoacetate Isoleucine Threonine Oxaloacetate Aspartate Arginine Ornithine Citruline Asparagine Guanidinoacetate Creatine Creatine-P Creatinine Tryptophan Lysine Ketone bodies Quinolinate NAD+ Leucine, Tyrosine Phenylalanine Tyrosine 4-HPP Homogentistic Acid Glutamate Glutamine Orotate UMPUridineUracil Cytidine dCMPDeoxycytidine Deoxyuridine AICAR FAICAR IMP Inosine Hypoxanthine Xanthine Xanthosine XMP Adenine Adenosine AMP 3’5’cAMP L-Histidine Guanine Guanosine GMP GDP GTP 3’5’cGMP Carnosine β Alanine Mut Isoleucine Urea Cycle Mat1a Dnmt1 Ahcyl1 Cth Cbs Gnmt Shmt1 Dmgdh Bhmt Aldh7a1 Chdh Csad Ggt6Baat Glutamyltaurine Cth Sds Sds Baat CslAco2 Idh2 Ogdh Suclg1 Sdha Fhl Mdh1,2 Pcx Pdha1 Arg1 Arginosuccinate Ass1Asl Otc1 Bhmt Acat1,2 Acat3 Aminoadipate Pyrroline 5-Carboxylate Proline Hydroxyproline Pycr1 Pycr2 P4ha1 Lysine Carnitine Pdha1 Tha1 Sorbitol Glucose Glucose6P Akr1b7 Akr1b8 Hk1, 2, 3 Gatm Gamt Ckb Adcy6, Adcy4 Nme3,4 Nt5c3 Nt5e Pnp Pnp2 Nt5c3 Nt5e Nt5c3 Nt5e Pnp Pnp2 Xdh Pnp Pnp2 Pnp Pnp2 Nt5c3 Nt5e Pde4b, 9a, 4c, 6c Pde4b, 4c Impdh1 Atic Cndp1 Cndp2Carns1 Upp1, 2 Guk1 Pklr Nt5c3 Nt5e Uck1 Umps Dhodh Cad Mcce Pcca, b Asns IL4l1 Pah Hmgcs2 Hmgcl Tat Hpd Hgd Gstz1 Fah Enpp1 Enpp3 Cndp1 Cndp2 Valine Nt5c3 Nt5e Cda Glud1 Ahcyl2 Pnp Pnp2 Succinate Semialdehyde GABA Aldh5a1 Abat 5-HIAA Figure 6 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 19, 2024. ; https://doi.org/10.1101/2024.02.15.580499doi: bioRxiv preprint