Xylose is the second most abundant sugar in lignocellulose and can be used as a feedstock for nextgeneration biofuels by industry. Saccharomyces cerevisiae, one of the main workhorses in biotechnology, is unable to metabolize xylose natively but has been engineered to ferment xylose to ethanol with the xylose reductase (XR) and xylitol dehydrogenase (XDH) genes from Scheffersoymces stipitis. In the scientific literature, the yield and volumetric productivity of xylose fermentation to ethanol in engineered S. cerevisiae still lags S. stipitis, despite expressing of the same XR-XDH genes. These contrasting phenotypes can be due to differences in S. cerevisiae's redox metabolism that hinders xylose fermentation, differences in S. stipitis' redox metabolism that promotes xylose fermentation, or both. To help elucidate how S. stipitis ferments xylose, we used flux balance analysis to test various redox balancing mechanisms, reviewed published omics datasets, and studied the phylogeny of key genes in xylose fermentation. In vivo and in silico xylose fermentation cannot be reconciled without NADP phosphatase (NADPase) and NADH kinase. We identified eight candidate genes for NADPase. PHO3.2 was the sole candidate showing evidence of expression during xylose fermentation. Pho3.2p and Pho3p, a recent paralog, were purified and characterized for their substrate preferences. Only Pho3.2p was found to have NADPase activity. Both NADPase and NAD(P)H-dependent XR emerged from recent duplications in a common ancestor of Scheffersoymces and Spathaspora to enable efficient xylose fermentation to ethanol. This study demonstrates the advantages of using metabolic simulations, omics data, bioinformatics, and enzymology to reverse engineer metabolism. METHODSGenome-scale network reconstruction and analysis. The S. stipitis GENRE was obtained from a panfungal GENRE that combined the GENRE's of S. stipitis (Balagurunathan et al., 2012;Caspeta et al., 2012;Li, 2012;Liu et al., 2012), Schizosaccharomyces pombe (Sohn et al., 2012), Aspergillus niger (Andersen et al., 2008), Yarrowia lipolytica (Pan and Hua, 2012; Loira et al., 2012), Komagataella phaffii (Caspeta 3/28 et al., 2012), Kluyveromyces lactis (Dias et al., 2014), and S. cerevisiae (Heavner et al., 2013). Model simulations were carried out using COnstraints Based Reconstruction and Analysis for Python version 0.9.1 (COBRApy) (Ebrahim et al., 2013). The xylose uptake rate was set to 10 mmol · gDCW -1 · h -1 .The growth associated maintenance (GAM) and non-growth associated maintenance (NGAM) were set to 60 mmol · gDCW -1 and 0 mmol · gDCW -1 · h -1 , respectively. Flux variability analysis (FVA) was used to evaluate alternative optimum solutions (Mahadevan and Schilling, 2003). The exchange bounds for erythritol, ribitol, arabitol, sorbitol, and glycerol were all set to zero to simplify the solution space for polyols. The cofactor selectivities of NADH and NADPH were varied in a single XR reaction (Balagurunathan et al., 2012). XR solely driven by NADH or NADPH were blocked. The e...