In the course of our continuing studies on the biosynthesis of riboflavin, we have encountered an apparently novel biochemical carbon skeletal rearrangement. Riboflavin (3, Scheme I) has been shown to form by a disproportionation of two molecules of 6,7 dimethyl-8-ribityllumazine (2), with the eight carbon atoms of its xylene ring originating from two biochemically identical four-carbon units (carbons 1a, 6, 7, and 7a of 2). The pyrimidinedione 1 is known to be the direct precursor of 2, but the origin of the four carbon atoms that eventually form the pyrazine ring of 2 is not clear, with several authors suggesting the involvement of a pentose (hence requiring the elimination of one carbon atom from the pentose precursor). Earlier studies using 13C-labeled precursors into riboflavin followed by 13C NMR spectroscopic analysis suggested an intramolecular skeletal rearrangement, and the present paper reports proof for such a process. [1,3-13C2]Glycerol (0.2 g, 90% 13C) was fed to the flavinogenic fungus Ashbya gossypii (ATCC 10859), and the resulting riboflavin was isolated, purified, and analyzed by 50.3-MHz proton-decoupled 13C NMR. The spectrum showed extensive coupling between the 3* and 4* positions (approximate ratio of coupled to uncoupled signals 3:1), indicating the terminal carbons of glycerol become directly connected via an intramolecular process during conversion into the pyrazine ring of 2, while remaining separated in the C-3'-to-C-5' portion of the ribityl side chain. The present experiment gives no indication of whether the rearrangement occurs at the triose or pentose level or at some other stage, and further work is required to settle this issue and unravel the mechanism of this intriguing rearrangement. The bicyclic geometry of P(OCH2)3P (1) permits complexation of each of the bridgehead phosphorus atoms to separate metal carbonyl moieties. Although 1 is prone to forming insoluble polymers with most metal carbonyl substrates, the last step in the reactions shown in Scheme I provides 40-86% yields of the novel tetramers 5-9, each of which are 20-membered rings. Comparison of the 31P NMR spectra of the mononuclear metal complexes 2-4 and the dinuclear compounds 10-12 with those of 5-9 reveals two equi-intensity peaks in all three sets of compounds, with closely corresponding 31P δ values for a given metal. The single 1H NMR resonance for all three types of compounds gives credence to the postulated cyclic nature of 5-9 (chemically equivalent protons in contrast to an acyclic oligomer), with broad (~25 Hz) symmetrical multiplets stemming from the "virtual" coupling. Efforts to obtain molecular weight data by osmometry failed owing to facile decomposition in pyridine, but the parent ion of 6 was clearly observed by FAB/MS. The 1:1 stoichiometry of the reactants in the reactions forming 5-7 coupled with the high yields are consistent with cyclic tetramer formation, likely arising from the unligated phosphorus in the final intermediate D cyclizing into a strainless macrocyclic ring structure by reacting with a nearby reactive metal center.