Unravelling the structural complexity of glycolipids with cryogenic infrared spectroscopy

Glycolipids are complex glycoconjugates composed of a glycan headgroup and a lipid moiety. Their modular biosynthesis creates a vast amount of diverse and often isomeric structures, which fulfill highly specific biological functions. To date, no gold-standard analytical technique can provide a comprehensive structural elucidation of complex glycolipids, and insufficient tools for isomer distinction can lead to wrong assignments. Herein we use cryogenic gas-phase infrared spectroscopy to systematically investigate different kinds of isomerism in immunologically relevant glycolipids. We show that all structural features, including isomeric glycan headgroups, anomeric configurations and different lipid moieties, can be unambiguously resolved by diagnostic spectroscopic fingerprints in a narrow spectral range. The results allow for the characterization of isomeric glycolipid mixtures and biological applications.


Ion Mobility-Mass Spectrometry and Tandem Mass Spectrometry
Supplementary Table 2 CCS of glycolipid adducts obtained from DT-IM-MS measurements (Helium,2.2 Torr). The values are given in Å 2 (± 1 %). Only sodium adducts of β-Gb3-and β-iGb3 sphingosine and silver adducts of glycosylphytosphingosines exhibit distinguishable CCS. Silver adducts of glycosylphytosphingosines allow for the distinction of α-and β-glycosidic bonds.  in the fingerprint region and the amide region. The fingerprint region of the sodiated species was employed for deconvolution of biological lipid extracts (cf. Supplementary Fig. 20). Source data are provided as a Source Data file.

Synthetic Mixtures of α-and β-GalCer
Supplementary Fig. 9 IR spectra of protonated α-and β-GalCer in the fingerprint region (1000-1150 cm -1 ). The main absorption bands of the pure isomers do not overlap (a, b). The decrease of α-GalCer from 50 % to 5 % is accompanied by a roughly linear decrease of intensity of the absorption band within the set boundaries (1059-1066 cm -1 ). At 1 %, α-GalCer cannot be detected anymore (c). The simulated mixture spectra generated by mathematically averaging the spectra of the pure isomers match the experimental spectra reasonably well (d-h). All spectra except for (c) are normalized. Source data are provided as a Source Data file.

Supplementary Fig
. 10 * Graphical representation of non-negative matrix factorization (NMF) to deconvolute IR spectra of α-and β-GalCer. The (76 x 7) input matrix contains the binned experimental spectra from 1000-1150 cm -1 . The expected ratio of α-GalCer is indicated on top. The output contains the two spectra of pure α-and β-GalCer in a (76 x 2) matrix multiplied by their relative contributions to each of the input spectra.
Supplementary Fig. 11: Comparison of NMF-simulated spectra of α-and β-GalCer (red) with the experimental IR spectra (black, input spectra from Supplementary Table 3). The spectra were simulated by matrix factorization of the output matrix (Supplementary Table 3) and the weighting factors depicted in Supplementary Fig. 10. Source data are provided as a Source Data file.
* The graphical representation of the input-and output IR spectra is based on the data in Supplementary

Synthetic Mixtures of Monoglycosyl Phytosphingosines (4 Components)
Supplementary Fig. 12 * Graphical representation of NMF to deconvolute IR spectra of α-and β-Glc/Gal phytosphingosine. The (76 x 15) input matrix contains the experimental spectra. The expected ratio of the four isomers is indicated on top. The output contains the four spectra of pure α-and β-Glc/Gal phytosphingosine in a (76 x 4) matrix multiplied by their relative contributions to each of the input spectra.
Supplementary Fig. 13 Heatmap representation of the weighting factors predicted by the mixing ratios (top) and calculated by NMF from the 15 experimental spectra (bottom). The weighting factors in the (4 x 15) matrix from Supplementary Fig. 12 were converted into the corresponding ratio of black and white (white = 0 %, black = 100 % glycolipid).
* The graphical representation of the input-and output IR spectra is based on the data in Supplementary Tables 4-5.
Supplementary Fig. 14 Comparison of NMF-simulated spectra of Glc/Gal phytosphingosines (red) with the experimental IR spectra (black, input spectra from Supplementary Table 4). The spectra were simulated by matrix factorization of the output matrix (Supplementary Table 5) and the weighting factors given in Supplementary Fig. 12. Source data are provided as a Source Data file.    Supplementary Fig. 19 IR spectra of sodiated glycosylceramides (m/z 832) from Folch extracts 1 (red) and 2 (blue) after hydrolysis, compared with reference spectra of α-and β-Glc/GalCer. The spectra were recorded in the diagnostic fingerprint region and the amide region. The best spectral match is provided by the reference spectrum of β-GlcCer. Source data are provided as a Source Data file.

Supplementary
Supplementary Fig. 20 * Graphical representation of NMF to deconvolute the IR spectra of of sodiated glycosylceramides (d18:1/24:1) from Folch extracts 1 and 2. The (100 x 6) input matrix contains experimental spectra of the two biological samples and of the four standards (α-and β-Glc/GalCer) in the fingerprint region (950-1150 cm -1 ). The output contains the four spectra of αand β-Glc/GalCer in a (100 x 4) matrix multiplied by their relative contributions to each of the input spectra. The weighting factors were converted into the corresponding ratio of black and white in the adjacent heatmap representation (white = 0 %, black = 100 % glycolipid).
Supplementary Fig. 21 Comparison of NMF-simulated spectra of synthetic and biological glycosylceramides (red) with the experimental IR spectra (black, input spectra from Supplementary Table 6). The spectra were simulated by matrix factorization of the output matrix (Supplementary Table 6) and the weighting factors depicted in Supplementary Fig. 20. Source data are provided as a Source Data file.  Fig. 23 Comparison of the theoretical spectrum of a low-energy conformer of [α-GalCer+Na] + (a) with lipid chains truncated to include only relevant functional groups of the lipid chain and calculated within harmonic approximation; (b) with full lipid chains attached to the same core-structure of the ion and calculated with harmonic approximation and (c) with lipid chains truncated and calculated including anharmonic effects. Attachment of the full lipid chains results in only minor changes to the diagnostic region (average blue-shift of 5 cm -1 ) of the IR spectrum between 1000 and 1150 cm -1 . Anharmonicity of the potential has a larger impact on the IR spectrum.