Host metabolite-cytokine correlation landscape in SARS-CoV-2 infection


 The systemic cytokine release syndrome (CRS) is a major cause of the multi-organ injury and fatal outcome induced by SARS-CoV-2 infection in severe COVID-19 patients. It has been well-known that metabolism plays a role in modulating the immune responses in infectious diseases. Yet, how the host metabolism correlates with CRS in COVID-19 patients and how the perturbed metabolites affect the cytokine release remains unclear. Here, we performed both metabolomics and cytokine/chemokine profiling on serum samples from the same cohort of healthy controls, mild and severe COVID-19 patients and delineated the global metabolic and immune response landscape along disease progression. Intriguingly, the correlation analysis revealed the tight link between metabolites and proinflammatory cytokines and chemokines, such as IL-6, M-CSF, IL-1α, IL-1β, implying the potential regulatory role of arginine metabolism, tryptophan metabolism, and purine metabolism in hyperinflammation. Importantly, we demonstrated that targeting metabolism markedly modulated the proinflammatory cytokines release by PBMCs isolated from SARS-CoV-2-infected rhesus macaques ex vivo. Beyond providing a comprehensive resource of metabolism and immunology data of SARS-CoV-2 infection, our study showed that metabolic alterations can be potentially exploited to develop novel strategy for the treatment of fatal CRS in COVID-19.

The systemic cytokine release syndrome (CRS) is a major cause of the multi-organ injury 24 and fatal outcome induced by SARS-CoV-2 infection in severe COVID-19 patients. It has 25 been well-known that metabolism plays a role in modulating the immune responses in 26 infectious diseases. Yet, how the host metabolism correlates with CRS in  patients and how the perturbed metabolites affect the cytokine release remains unclear. 28 Here, we performed both metabolomics and cytokine/chemokine profiling on serum 29 samples from the same cohort of healthy controls, mild and severe COVID-19 patients and 30 delineated the global metabolic and immune response landscape along disease progression. 31 Intriguingly, the correlation analysis revealed the tight link between metabolites and 32 proinflammatory cytokines and chemokines, such as IL-6, M-CSF, IL-1α, IL-1β, implying  Fig. 4a, b). Molecules 175 enriched in cluster 1 increased at symptoms onset but gradually deceased during 176 hospitalization; molecules in cluster 2 exhibited a sharp decreased at symptoms onset and 177 sustained stable levels in later time-points. However, molecules in cluster 3 sustained 178 steady levels but presented a delayed elevation in the very late events; cluster 4 contained 179 molecules that elevated gradually and declined in late phases (Fig. 3a). 180 Three CRS-related cytokines including IL-6, IP-10 and M-CSF belonged to cluster 1 (Fig.   181   3b). Notably, the levels of IL-6, which is highly correlative to CRS 30 , decreased during the 182 first two weeks of symptom onset and remained at low level in later phases (Fig. 3b, 183 Extended Data Fig. 4c). The IFN-γ inducible protein, IP-10/CXCL-10, is considered as a 184 member of CXC chemokine family with proinflammatory and severity-related properties 185 in COVID-19 8 . The levels of IP-10 showed a sharp decline in the initial phase of treatment 186 and sustained at a relatively low levels during hospitalization (Fig. 3b, Extended Data Fig.   187 4c). Consistently, the myeloid cytokine M-CSF also showed a downward trend over the 188 course of mild disease similar to IP-10 (Fig. 3b, Extended Data Fig. 4c). In addition, 189 10 proinflammatory cytokines in cluster 2, including G-CSF, also 190 showed decreased levels in mild patients compared to healthy controls and remained steady 191 levels during hospitalization (Fig. 3b, Extended Data Fig. 4c). However,proinflammatory 192 cytokines in cluster 3 (i.e., IL-17A and TNF-β) and cluster 4 (i.e., IL-1α, IL-1β, IL-18 and We next characterized metabolites that were enriched in four clusters and specific 210 metabolite-cytokine correlations. Interestingly, metabolites associated with arginine 211 metabolism were enriched in cluster 2 and cluster 3 (Fig. 3c). We observed the upward 212 trend of arginine, ornithine, glutamate and proline, whereas the decrease in citrulline , and showed the negative correlation with most proinflammation cytokines (Fig. 3e). 227 Taken together, our data identify a panel of metabolite-cytokine correlation, which may 228 provide an unbiased way to determine the potential metabolic regulators of 229 proinflammatory cytokine secretion.  231 Our data presented above delineated the strong correlation between cytokines and 232 metabolites, and identified metabolic pathways that are potentially crucial for 233 proinflammatory cytokine production. We next evaluated whether intervening arginine 234 metabolism, tryptophan metabolism and purine metabolism could regulate cytokine 235 induction. To this end, we isolated PBMCs from SARS-CoV-2-infected and mock-infected 236 rhesus macaques, and measured the cytokine concentrations after treatment with 237 metabolites or compounds interfering these key metabolic pathways (Fig. 4a). Interestingly, 238 we observed that supplementation of arginine markedly inhibited the SARS-CoV-2-239 induced proinflammatory cytokines production by PBMCs, most of which linked to CRS 240 including IL-1α, IL-1β, IL-2, IL-6, TNF-α, GM-CSF, G-CSF and MIP-1α (Fig. 4b, 241 Extended Data Fig. 6). Notably, the elevated level of IL-10, which is responsible for 242 inhibition of proinflammation cytokines production from macrophage and dendritic 243 cell (DC) populations 31 , was also suppressed (Fig. 4b). These observations suggest that 244 serum arginine metabolism may play a modulatory role in the hyperinflammation, thus 245 could be exploited as a potential therapeutic target for CRS in  The conversion of tryptophan into kynurenine in immune cells is finely regulated by the 247 enzyme indoleamine 2,3 dioxygenase 1 (IDO1), which is reportedly involved in regulating 248 hyperinflammatory responses 24 . The elevated ratio between circulating kynurenine and 249 tryptophan (Kyn/Trp) in patient's serum described above suggested an increased activity 250 13 of IDO1. Addition of epacadostat, an IDO1 inhibitor, suppressed the SARS- CoV-2-251   induced proinflammation cytokine release including IL-1α, IL-1β, IL-6, TNF-α, GM-CSF,   252 G-CSF, IL-17A and MIP-1α, which confirmed an essential role of tryptophan metabolism 253 in exaggerated cytokine release upon SARS-CoV-2 infection (Fig. 4c, Extended Data Fig.   254   7)). In addition, direct inhibition of purine metabolism with mycophenolic acid (MPA), 255 which blocks the rate-limiting enzyme inosine monophosphate dehydrogenase (IMPDH) 256 in de novo synthesis of guanosine nucleotides, significantly reduced levels of IL-10, IFN-257 γ, IL-15, IL-12 p40, IL-17A and TNF-α induced by SARS-CoV-2 infection. However, a 258 profound increase in proinflammation cytokines of IL-6, GM-CSF, IL-1α and IL-1β was 259 also observed, which suggests the exacerbated hyperinflammation upon interfering with 260 purine metabolism (Fig. 4d, Extended Data Fig. 8). Finally, we evaluated the effect of 261 interfering these metabolic pathways on the SARS-CoV-2 replication in PBMCs. Our 262 results showed that arginine supplementation, IDO1 or IMPDH inhibitors did not affect the 263 SARS-CoV-2 replication (Extended Data Fig. 9). Taken together, these data suggest that 264 targeting dysregulated host metabolism may serve as a viable approach to suppressing 265 SARS-CoV-2-induced inflammatory cytokine secretion.  The cytokine profiling of COVID-19 patients in our study provide further evidence that 284 most CRS-associated cytokines, such as IL-6, IL-1β, IL-10, IL-18 and IFN-γ, are 285 dramatically elevated in severe patients. Conversely, the increased inflammatory responses

299
Our metabolomics data identified alterations of circulating metabolite levels in patient's 300 serum and determined dysregulated metabolic pathways upon SARS-CoV-2 infection.

301
Consistent with recent studies 18,41 , the metabolites that are associated with arginine 302 metabolism, tryptophan metabolism, TCA cycle as well as purine and pyrimidine 303 metabolism changed remarkedly. Interestingly, correlation network analysis between 304 metabolites and cytokines in mild and severe patients revealed that circulating cytokine 305 levels were highly correlated with arginine metabolism, tryptophan metabolism, nucleic              This served as an additional quality control measure of analytical performance and a 573 reference for normalizing raw metabolomic data across samples.

574
To remove potential inter-batch variations, the mean peak area of each metabolite from all 575 the QC samples in a given batch (QCall), as well as the mean peak area of each metabolite 576 from the QC samples that are the most adjacent to a given group of test samples (QCadj) 577 were first calculated. The ratio between these two mean peak areas for each metabolite was 578 computed by dividing the same QCall by each QCadj and used as the normalization factor 579 for each given group of test samples. The peak area of each metabolite from each test 580 sample were normalized by multiplying their corresponding normalization ratio to obtain 581 the normalized peak areas. In addition, to effectively correct the sample to sample variation 582 in biomass that may contribute to systematic differences in metabolite abundance detected 583 by LC-MS, we generated the scaled data by comparing the normalized peak area of each 584 metabolite to the sum of the normalized peak area from all the detected (for targeted 585 metabolomics) or identified metabolites (for untargeted metabolomics) in that given 586 sample.

587
Our validation analyses suggested that these normalization and scaling steps could 588 effectively correct both the inter-sample artificial differences in sample biomass and inter-589 batch systematic variations in detected metabolite abundance.    Fisher's exact test (e) followed by Benjamini-Hochberg multiple comparison test.