Improving LC-MS analysis of human milk B-vitamins by lactose removal

Highlights • Lactose in human milk affects B-vitamin analysis by fouling the ion source.• Typical extraction methods are inefficient for lactose removal.• Chromatographic separation from analytes enables lactose mechanical removal.


Introduction
Human milk B-vitamins have been a topic of investigation for decades [1,2]. Analytical methods were usually adapted from those developed for blood, plasma, or urine samples [1], matrices with negligible amounts of sugar and fat, unlike human milk [3,4] (Supplemental Fig. 1); and proper validation for the intricate milk matrix has rarely been described.
B-vitamins differ in their chemical and physical properties. Associations with other milk constituents complicate their simultaneous analysis; some procedures benefit certain B-vitamins while others degrade under the same conditions [1,2]. Their light, temperature, and pH sensitivities demand rapid but gentle handling. We reported the first validated method for simultaneously analyzing several B-vitamins in human milk by ultra-performance liquid-chromatography-tandem massspectrometry (UPLC-MS/MS) [2], revealing severe matrix effects, and recently described methods [5,6] continue to lack focus on matrix effects, recovery, and process efficiency, and how to overcome these limitations.
LC-MS offers superior sensitivity, selectivity, and specificity, but electrospray ionization suppression or enhancement can affect precision and accuracy [7][8][9][10], with the most challenging interferences caused by co-eluting matrix components [9]. Human milk contains about 7% lactose, a polar disaccharide [3], and reported sample preparations [2,5,6] are ineffective in its removal. If not excluded, lactose can rapidly foul the ionization source, decreasing method robustness and impacting detection limits over the course of the analysis. We evaluated solidphase extraction (SPE), liquid-solid extraction (LSE), and protein precipitation (PPT) for lactose removal, a crucial but missing purification step in previously reported methods, to compare matrix effects, process efficiency, and analyte recovery with previously published reports.

Chemicals and materials
Details of chemicals and materials used in this study are listed in Supplemental Table 1. USA) coupled to an API 4000QTRAP-MS (Sciex; Foster City, CA, USA), connected by a two-position switching valve (Valco Instruments, Houston, TX, USA) between UPLC-outlet and MS-inlet. 3-10 µL sample was injected onto an ACQUITY UPLC® HSS T3 column (2.1 × 100 mm, 1.8 µm) protected by an ACQUITY UPLC® HSS T3 VanGuard TM precolumn (2.1 × 5 mm, 1.8 µm; Waters) in reversed-phase (RP) mode (Supplemental Table 3). Starting at 1.8 min, the switch valve diverted the effluent to the detector, enabling a lactose-free detection of the Bvitamins (Supplemental Figures 2 and 3). MS-acquisition was optimized for positive ion mode electrospray ionization (ESI) multiple reaction monitoring (MRM) with Q1/Q3 unit (Supplemental Tables 2 and 3).

Human milk samples
Human milk (HMP) was pooled from healthy donors from Vancouver, BC, Canada, and used as for the all experiments as described. Aliquots were stored at − 70 • C in amber tubes.

Preparation of standards
All external and internal aqueous standard solutions were prepared under dim light on ice and in amber glass vials (Supplemental Table 4). Stable isotope labeled vitamins were used as internal standards for quantitation, 13 C 3 -caffeine was used as a system control to adjust for potential system fluctuation [2]. The stable-isotope internal standards were combined to an aqueous master mix (1 µg/mL; IS) of which 10 µL was added to each sample, test-mix, or lactose solution.

Solid phase extraction
50 µL aqueous standard test-mix (thiamin, riboflavin, FAD, nicotinamide, and pyridoxal: 500 ng/mL) and 150 µL methanol (MeOH) were combined (mock sample) to mimic PPT of milk as previously described [2]. A 5% aqueous lactose solution was treated identically and applied to a separate SPE cartridge. The OASIS® HLB sorbent was chosen due to its all-purpose, strongly hydrophilic, water-wettable polymer with a hydrophilic-lipophilic balance. After conditioning, the mock sample or lactose solution were applied followed by a wash step prior to a stepwise elution ( Table 1). The collected fractions were adjusted to 500 µL for analysis. This served as a preliminary experiment with no replicates.

Liquid-Solid extraction
LSE was evaluated as an additional clean-up step. Acetone, ACN, ethanol (EtOH), ethyl-acetate (EtOAc), 2-propanol (IPA), and MeOH were tested. 40 µL of the standard test-mix were evaporated to dryness and subjected to LSE using 100 µL solvent. The supernatant obtained was evaporated and reconstituted with 100 µL initial mobile phase. The sample was back-extracted with MTBE to mimic the removal of nonpolar matrix. LSE-conditions were tested with 40 µL lactose solution when all tested vitamins were recovered > 50%. LSE using HMP was Table 1 SPE elution profiles for B-vitamins and lactose and analyte recovery after liquid-solid extraction (LSE) of B-vitamins from neat standard mix and human milk.  done with solvents efficiently removing lactose while maintaining analyte recoveries > 50%. Results obtained were compared to those from HMP prepared without LSE (Table 1).

Protein precipitation
Different PPT-solvents were tested using a modified Taguchi method [11]. MeOH, acetone, and IPA were tested as PPT-solvents. ACN caused low FAD and nicotinamide recovery in human milk [2] and was excluded. Additional parameters tested included sample volume and reconstitution solvent ( Table 2). The sample preparation followed the same principle as described above.

Sample preparation
Samples were prepared based on our previous report [2] and optimized based on results obtained here. 10 µL IS and 40 µL sample were combined, and 150 µL MeOH was added for PPT, followed by vortexing for 2 min. After centrifugation (10 min, 4 • C, 19098 × g SORVALL® LEGEND RT, Asheville, NC), the supernatant was transferred into a 4 mL amber glass vial to dry under a gentle nitrogen stream. The residue was reconstituted in 100 µL 10 mM NH 4 -formate aq /ACN 99/1 (v/v) containing 13 C 3 -caffeine-trimethyl (100 ng/mL) and transferred into a fresh 1.5 mL amber centrifuge tube on ice, containing 200 µL MTBE. After mixing for 30sec, the two phases were separated by centrifugation (3 min, 4 • C, 19098 × g). The lower phase (sample extract) was filtered (0.1 µm) and analyzed.

Quality control
An aqueous eight-point standard curve (1-500 µg/L; nicotinamide: 5-2500 µg/L; Ca-pantothenate: 15-7500 µg/L) and a reagent blank were included with each batch of the experiments. The analytes were quantified using the respective trendlines (f(x) = m*x + b; m: slope, b: intercept), as determined by ratio response of analyte to IS.

Analyte recovery
A 3-level standard addition experiment on 5 different days determined recovery rates using HMP with unknown B-vitamin concentrations. 5 replicates for each level (0, 5, 10, and 15 µL of a stock solution [500 µg/L; nicotinamide: 2.5 mg/L, Ca-pantothenate: 7.5 mg/L]) were added to HMP. Recovery rates for each vitamer were calculated as follows:

Analyte stability during sample preparation
Five sequential sets of 24 samples were prepared throughout each day, each set including 2 HMPs. On the first day, an HMP aliquot of 600 µL was thawed with the first sample set, stored on ice, and used throughout the day as a source for the 2 HMPs per set. On day 2, two 40 µL aliquots of frozen HMP were thawed with each sample set prior to sample preparation. Both HMP aliquots were analyzed at the beginning and at the end of the set. The B-vitamin concentrations were compared within and between the sets.

Statistical analysis
Descriptive statistics (mean, standard deviation, coefficient of variation) were carried out using Microsoft Excel 2016 (Microsoft Corporation, Redmond WA, USA). Difference in ME, PE, and RE by injection volume were examined by generalized linear models. Student's t-test was used to evaluate matrix effects to already published data, using the results obtained with the same injection volume (10 µL). P-values below 0.05 were considered significant.

Solid-phase extraction
SPE using NH 4 -formate/ACN resulted in minimal column-retention of all tested analytes (Table 1). Flow-through and column wash contained most amounts for nearly all vitamins, and lactose. Only riboflavin was detected in several fractions (5-15%). Using acetic acid, thiamin, nicotinamide, pyridoxal, and lactose continued to co-elute; riboflavin, FAD, and caffeine eluted mostly at 90/10-75/15 (acetic acid/ACN, v/v). However, none of the tested conditions separated lactose from all analytes.

Liquid-solid extraction
ACN and EtOH were unsuitable for FAD and therefore excluded. Using EtOAc resulted in low recoveries for thiamin, FAD, and pyridoxal, and FAD was not detected using acetone (Table 1). Nicotinamide and 13 C-caffeine were recovered > 70% in all solvents. Only MeOH and IPA produced good overall recoveries and were applied to the lactose solution, which revealed a > 90% recovery with MeOH but < 10% when using IPA. Thus, HMP was only tested using IPA. When compared to HMP without LSE, thiamin, riboflavin, and FAD showed < 20%, nicotinamide ~ 30%, and pyridoxal ~ 60%, recovery. 13 C-caffeine was not affected (99% recovery). Thus, LSE was ineffective in lactose removal while maintaining acceptable analyte recoveries.

Protein precipitation
Thiamin and riboflavin were not affected by choice of PPT-solvent, sample volume, or reconstitution solvent ( Table 2). Nicotinamide was negatively affected by sample matrix, while the opposite was true for pyridoxal. Acetone and IPA were superior for nicotinamide and pyridoxal, while FAD revealed a higher response with MeOH, indicating that optimal conditions are analyte-depended. Analyte recoveries after standard addition employing acetone-and MeOH-PPT showed comparable recoveries for all B-vitamins but FAD, which revealed 8-9x lower concentrations with acetone, indicating only a partial recovery of milk FAD. Hence, MeOH was chosen for PPT due to overall good analyte recoveries and higher volatility and quicker evaporation than IPA.

Chromatography effluent diversion
Post-column lactose removal using a switch-valve was tested as an alternative. We have already described various conditions for optimal analyte resolution [2], which resulted in co-elution of lactose with some B-vitamins, such as pyridoxal, the main B6-vitamer in milk (data not shown). Here, we focused on gradient optimization for chromatographic separation of lactose and B-vitamins. Doubling the column length and changing the initial solvent ratio from 95:5 to 99:1 optimized resolution to the main B-vitamins in milk, but pyridoxamine, a B6-vitamer commonly present in milk in marginal amounts, could not be separated from lactose (Supplemental Figures 2 and 3) and was excluded.

Recovery rates, precision, and accuracy
Shot-to-shot variance in 13 C-caffeine response of 100 injections revealed a CV of 12.6%. The standard addition experiments showed recoveries 92-119% for all vitamins but FAD (81.9%; Table 3). FAD is quantified without the isotope labeled analog, thus the lower recovery is not unexpected. Being the main B2-vitamer, milk-FAD is generally present in a concentration that can tolerate the lower recovery. The nicotinamide recovery was increased (107.5%) compared to previous reports (77-80%) [2,5], likely because the now available isotopelabeled nicotinamide allowed corrections for true losses and matrix effects.
Analyzing 200 HMPs over two months showed an inter-day variability of 4.9-13.7%. All vitamins were quantifiable above their lowest standard concentration; thus all vitamins were analyzable up to 1800fold lower concentration than those used to set the Adequate Intake for infants 0-6 months [12].

Matrix effects
ME, RE, and PE were unaffected by injection volume for thiamin. Biotin's ME, PE, and RE were significantly higher with lower injection volumes (p < 0.05, Table 4), unlike for riboflavin, nicotinamide, and pyridoxal. Pantothenic acid's ME was least affected at lower injection volumes but with increasing effects on PE. Compared to published data [2], we found significant improvements for ME and PE, but lower RE for riboflavin and pyridoxal (p < 0.05, Table 4), possibly due to the differences in sample preparation.

Analyte stability during sample preparation
The ratios of frozen HMPs to those stored on ice revealed a random distribution around 1.0 for all vitamins but nicotinamide (Fig. 1A). The 4-5-fold higher concentrations in the frozen samples suggested analyte degradation in thawed samples on ice. Plotting nicotinamide concentrations against the time of sample preparation (Fig. 1B) and injection (Fig. 1C) showed a linear and stepwise, respectively, decreasing relationship, which occurred with every new sample set prepared, but not between the technical HMP replicates of one set. Thus, nicotinamide degrades most likely on ice, most severely in the first 5 h as sets prepared later revealed comparable concentrations.
The sample extract was stable for at least 15 h at 10 • C; re-injection of all HMP extracts at the end of the analytical run of 5 sets on 2 different days (n = 54) revealed an intra-and inter-day variability < 9.0%. Frozen HMPs showed an inter-day variability for all analytes < 13.7%, including nicotinamide (Table 4). This degradation was specific for endogenous nicotinamide and not found for the added 13 C 6 -nicotinamide-IS, which should undergo similar changes as nicotinamide, e.g. conversion into other B3-vitamers. Thus, the reason for this phenomenon remains unknown. Nonetheless, the linear fashion of this degradation allows for post-run concentrations adjustments, but QC samples should be thawed with the sample set to avoid mathematical adjustments.
The changes described here allowed for the analysis of at least 120 samples, a significant increase compared to our previous report [2], indicative of improved stability and robustness of the method for sample and system.

Conclusions
Previous reports on human milk B-vitamin analysis by LC-MS have Table 3 Recovery rate [%], standard deviations (SD), and mean concentrations [µg/L] of multiple B-vitamins simultaneously analyzed in a pooled human milk sample and signal to noise ratios at 1 µg/L concentrations (15 µg/L for PA) and comparison to the values used to set the Adequate Intake for infants 0-6 months.  not examined the impact of endogenous lactose on the equipment and analysis. This is the first report focusing on targeted removal of known matrix interferences from human milk and on the effects of remaining matrix constituents. We have described the inefficiency of commonly used sample preparation techniques as well as an efficient post-column removal of milk lactose and possibly other polar interferences. This chromatographic separation of matrix from analytes allows for a quicker sample preparation and can be routinely incorporated into the analytical run, resulting in a more robust method with lower matrix effects, which now allows for higher-throughput without loss of sensitivity.

Author Statement
The contribution of all authors are as shown below:

Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Table 4
Matrix effect (ME), process efficiency (PE), and recovery (RE) of the stable isotope labeled internal standards in human milk at different levels of injection volume (mean ± SD).  2 injection volume used: 10 µL. P-values were obtained by general linear models when examining the results obtain with different injection volumes 3, 5, and 10 µL in this study. Labeled means (n = 6) in a row without common letter differ. Differences between results published in reference 2 and the 10 µL injection were tested by Student's t-test, and indicated by asterisks. Statistical differences were observed when p < 0.05. SD: standard deviation.