Sample collection
We collected milk samples from healthy human volunteers representing Russian (n = 9) and Chinese (n = 10), cows (n = 4), goats (n = 4), pigs (n = 4), yaks (n = 2), rhesus monkeys (n = 2), and crab-eating monkeys (n = 2) (Additional file 5: Table S1). Informed consent for the use of milk in this study was obtained from each of the human volunteers. In all species, milk samples were collected at matched lactation stage, several weeks after infants’ birth. In each case, milk was sampled at the end of breastfeeding or milking event into the same type of 10-ml plastic container and immediately placed into -20°C freezer for short-term storage not exceeding two weeks. Samples were then transported on dry ice without de-freezing into -80°C freezer for long-term storage, not exceeding four months.
Ethic statement
All animals were treated in accordance with good animal practice as defined by the local welfare authorities; all human volunteers have signed an Informed Consent Form, confirming that they understand the purpose of the research and their participation is entirely gratuitous.
Lipid extraction
The milk aliquots were thawed at 0°C mixed, and 16 µl of milk were transferred to a 2.0 ml Eppendorf safe-lock tube and resuspended in 34 µl of LC-MS grade water. Prior to extraction, samples were randomized with regard to species’ identity. Furthermore, for each species, we prepared a pooled sample containing equal volumes of milk from each individual. For lipid extraction, a modified two-phase protocol was used as described in [20]. All manipulations with samples were performed on ice. Briefly, 750 µl of MeOH:MTBE (1:3) solution containing internal standards in concentration of 1 mg/L were added to each sample, vortexed for 1 min, sonicated for 15 min in an ice-cooled sonication bath, incubated for 30 min at 4 °C, and sonicated for the second time in a pre-cooled sonication bath. Then, 560 µl of MeOH:H2O (1:3) solution was added to each sample, vortexed for 10 sec and centrifuged for 10 min at 14.000 x g at 4 °C. The 400 µl of the upper-phase, containing organic fraction, was transferred to a new 2.0 ml Eppendorf tube and dried in a Speedvac for 1 h at 30 °C.
Mass-spectrometry
Dried lipid pellets were resuspended in 400 µl of acetonitrile:isopropanol (1:3) solution. Samples were rigorously vortexed for 10 sec, shaken for 10 min at 4℃ and sonicated in an ice bath. Then, 5 µl of each sample was transferred to the autosampler glass vial and diluted 1:20 with 95 µl of acetonitrile:isopropanol (1:3) solution. A pool of all samples was prepared by mixing 5 µl from each sample in a separate Eppendorf tube, transferred to glass vials and diluted 1:20 with acetonitrile:isopropanol (1:3) solution to get quality control (QC) samples. Sample pools for each species were made by mixing 10 µl of each sample of the corresponding species and diluted 1:20 with acetonitrile:isopropanol (1:3) solution. From each diluted sample, 3 µl were injected to a reversed-phase Bridged Ethylene Hybrid (BEH) C8 reverse column (100 mm x 2.1 mm, containing 1.7 µm diameter particles, Waters) coupled to a Vanguard pre-column with the same dimensions, using a Waters Acquity UPLC system (Waters, Manchester, UK). The mobile phases used for the chromatographic separation were: water, containing 10 mM ammonium acetate, 0.1% formic acid (Buffer A) and acetonitrile:isopropanol (7:3 (v:v)), containing 10 mM ammonium acetate, 0.1 % formic acid (Buffer B). The gradient separation was: 1 min 55 % B, 3 min linear gradient from 55 % to 80 % B, 8 min linear gradient from 80 % B to 85 % B, and 3 min linear gradient from 85 % A to 100 % A. After 4.5 min washing with 100 % B the column was re-equilibrated with 55 % B. The flow rate was set to 400 µl/min. The mass spectra were acquired in a positive mode using a heated electrospray ionization source in combination with a Bruker Impact II QTOF (quadrupole-Time-of-Flight) mass spectrometer (Bruker Daltonics, Bremen, Germany).
Four blank samples were run at the beginning of the queue, followed by four QC samples to equilibrate the column. After them, 38 samples were queued in the same random order used for extraction with all samples randomized by species, interleaved with seven pooled samples, one QC preceding the first sample and then a QCs after every 9th sample. At the end of the queue, we performed two injections containing 100% acetonitrile to wash the column, followed by blank samples. Blank samples were prepared as usual samples, but contained only extraction buffers to reveal all contaminants that could come from the extraction and other technical steps, and not from the sample itself.
Data preprocessing
After the acquisition, Bruker raw data .d files were automatically calibrated using the internal calibration and converted into mzXML format using a custom DataAnalysis script (Bruker, Version 4.3). The mzXML files were then subjected to the standard alignment and peak picking procedure using xcms software [21]. We then filtered from the output table all lipid features with the coefficient of variation (CoV, calculated as the standard deviation over the mean across QC samples) > 30%, and peaks with zero values in > 50% of the individual samples. Lipid features’ intensities were then normalized using the upper quartile normalization and base-two log-transformed. Raw data is uploaded to the metabolomics study data repository MetaboLights [22].
Phylogenetic distances calculation
The lipid intensity-based distances between species were calculated as the Euclidean distance between the vectors containing the intensities of 472 lipids detected in each pair of species. The phylogenetic distances between the species’ pairs were obtained from the TimeTree database [23].
Statistical analysis
Species-dependent lipids were defined with ANOVA and BH-corrected p-value cutoff of 0.05. The correlation matrix of species-dependent lipids was calculated as (1-cor) Pearson’s distances between all lipids. Unsupervised clustering of the species-dependent lipids was performed using hierarchical clustering with complete linkage in the R statistical environment.
TAGs annotation
All detected lipid features were annotated against the theoretical list with all possible masses of NH4+ adducts of TAGs. The theoretical list of masses was generated using ALEX Target List Generator with the ALEX lipid database (5.2) [24]. Among the detected lipids, 76 matched the theoretical masses with < 10 ppm and were considered for further analysis.
Species-dependent TAG intensity differences
We calculated the mean intensities of TAGs in each species using the raw intensities of the annotated TAGs. To assess TAGs concentrations in each species, the mean intensity of a particular TAG was divided by the sum of the mean intensities of all TAGs in that species. For comparison between species, the 100% intensity value of the TAG was defined as the maximal intensity of the TAG across the species.