Human Milk Oligosaccharide Profile Variation Throughout Postpartum in Healthy Women in a Brazilian Cohort

Human milk oligosaccharide (HMO) composition varies throughout lactation and can be influenced by maternal characteristics. This study describes HMO variation up to three months postpartum and explores the influences of maternal sociodemographic and anthropometric characteristics in a Brazilian prospective cohort. We followed 101 subjects from 28–35 gestational weeks (baseline) and throughout lactation at 2–8 (visit 1), 28–50 (visit 2) and 88–119 days postpartum (visit 3). Milk samples were collected at visits 1, 2 and 3, and 19 HMOs were quantified usinghigh-performance liquid chromatography with fluorescence detection (HPLC-FL). Friedman post-hoc test, Spearman rank correlation for maternal characteristics and HMOs and non-negative matrix factorization (NMF) were used to define the HMO profile. Most women were secretors (89.1%) and presented high proportion of 2′-fucosyllactose (2′FL) at all three sample times, while lacto-N-tetraose (LNT, 2–8 days) and lacto-N-fucopentaose II (LNFPII, 28–50 and 88–119 days) were the most abundant HMOs in non-secretor women. Over the course of lactation, total HMO weight concentrations (g/L) decreased, but total HMO molar concentrations (mmol/L) increased, highlighting differential changes in HMO composition over time. In addition, maternal pre-pregnancy body mass index (BMI) and parity influence the HMO composition in healthy women in this Brazilian cohort.


Introduction
Human milk is a conditionally complete food, allowing infants to reach adequate growth and development [1][2][3][4]. The paramount benefits for both mother and child health include reduced risk of breast cancer and diabetes for mothers and lower infectious morbidity and mortality for children [5]. The World Health Organization (WHO) has recommended since 2001 the adoption of exclusive breastfeeding for the first six months of life [6,7]. These health benefits have been related to the composition of human milk [8,9].
Human milk is composed of macronutrients (carbohydrates, proteins and fat), micronutrients (vitamins and minerals) and bioactive compounds (cytokines, hormones, growth factors, among others) [8]. Oligosaccharides, part of the carbohydrate fraction, are the 3rd most abundant solid component of human milk (after lactose and fat) [10]. Human milk oligosaccharides (HMOs) are complex unconjugated glycans that are synthesized from lactose, can be by elongated by lacto-Nbiose or N-acetyllactosamine disaccharide units, and modified by fucose or sialic acid [10]. More than 150 structurally distinct HMOs have been identified, however, less than 20% of HMOs by number comprise approximately 90% of total HMO composition by concentration [11,12]. HMO concentrations range from 20-25 g/L in colostrum to 5-15 g/L in mature milk [10,13,14].
Maternal genetic features play an important role in HMO composition and variation. Single nucleotide polymorphisms (SNPs) in the secretor gene, encoding for the enzyme fucosyltransferase-2 (FUT2), and the Lewis gene, encoding for fucosyltransferase-3 (FUT3), contribute a large part to the variation in the HMO composition between different women [15,16]. Women with an active FUT2 enzyme are secretors, whose milk contains a high amount of α1-2-fucosylated HMOs like 2'fucosyllactose (2'FL) and lacto-N-fucopentaose I (LNFP I). In contrast, women with specific FUT2 SNPs that introduce a premature stop-codon and abolish the expression of the FUT2 enzyme, are non-secretors and their milk has almost no 2'FL or LNFP1 [10,[17][18][19]. In addition to these genetically determined factors, studies have shown that HMO concentrations vary throughout lactation [14,20,21], but also by geographic location, parity, maternal age, weight, body mass index (BMI), mode of delivery and other environmental factors such as seasonality [21][22][23][24][25].
HMOs play important roles in infant health and development outcomes [26][27][28]. Higher HMO diversity is associated with lower total body weight and lower body fat percentage in children [29]. HMOs can act as substrate for specific and potentially health-promoting bacteria and modulate the infant intestinal microbiota composition [30][31][32][33].
Several studies have described the concentrations and variations of HMOs during lactation [20,21,34,35]. However, few investigations have considered a longitudinal design during the first months postpartum and investigated the association between modifiable maternal characteristics and HMO composition. Furthermore, to the best of our knowledge, there are no longitudinal studies during this period in healthy Brazilian or Latin-American women.
This study describes the variation of HMO concentrations at 2-8, 28-50 and 88-119 days postpartum and explores the role of maternal sociodemographic and anthropometric characteristics on HMO composition among healthy women followed in a cohort study in Rio de Janeiro, Brazil.

Study Design and Eligibility Criteria
This study is part of a prospective cohort conducted in a Public Health Care Center in Rio de Janeiro, Brazil. Pregnant women were invited to participate if they met the following eligibility criteria: 18 to 40 years, 28th to 35th weeks of gestation, live in the programmatic area (neighborhood), intended to remain living within the catchment area after the child's birth, free of infectious and chronic diseases (except obesity) and without twin pregnancy.
The Research Electronic Data Capture (REDCap), an online platform with robust data entry tool, that allows the management of studies with different types of design, was used to collect information and to create a database [36].

Human Milk Data and Oligosaccharides Analysis
Human milk samples were collected at 2-8 days, 28-50 and 88-119 days according to the Brazilian Network of Human Milk Banks protocol [37]. The milk was collected preferably in the morning and after breakfast. A general guideline on massage techniques and hand expression of breast milk to prevent breast engorgement and mastitis was provided. Afterwards, the participant was counseled for personal protective equipment use (face protection masks and head cover). The investigators performed the same procedures and used nitrile gloves without talcum powder, which reduces the possibility of samples contamination [38]. Investigators were available to assist participants at the time of collection whenever needed.
The mother performed manual milk extractions directly in 50 mL sterile, ribonuclease (RNAse) and deoxyribonuclease (DNAse)-free and non-pyrogenic falcon tubes. At 2-8 days, 5 mL of milk were collected, and were later divided into 5 aliquots of 1 mL. At 28-50 and 88-119 days, 17 mL were collected, which were subdivided into 1 aliquot of 5 mL and 12 aliquots of 1 mL. The samples were processed immediately after collection. Human milk was stored in sterile, RNAse and DNAse-free and non-pyrogenic tubes; refrigerated at −20 °C, transported in a temperature-controlled box (−1 °C to −5 °C), and finally stored in a freezer at −80 °C until analysis.
Gestational weight gain (GWG) was calculated by means of the difference between weight at the last prenatal visit and pre-pregnancy weight (self-reported or retrieved from the prenatal care booklet). The supplements used during pregnancy, and sociodemographic information (maternal age, education and parity) were self-reported and obtained using structured questionnaires. The breastfeeding status was classified according to the WHO definitions [43].
The study was approved by the Research Ethics Committees of the Municipal Secretariat of Health and Civil Defense of the State of Rio de Janeiro (Protocol number: 49218115.0.0000.5275) and of Maternity School of Rio de Janeiro Federal University (Protocol number: 49218115.0.0000.5275), and was carried out following the rules of the Declaration of Helsinki of 1975. Written consent from all participants was obtained freely and spontaneously, after all necessary clarifications were provided.

Statistical Analysis
Statistical analyses were performed with R software, version 3.6 [44]. The variables distribution was evaluated by histograms, skewness and kurtosis measures and a Shapiro-Wilk test (stats package). Outlier values of HMOs concentration were not excluded. Median and interquartile ranges were used to describe maternal sociodemographic and anthropometric characteristics, HMO concentrations and alpha diversity. A Mann-Whitney test was performed in order to evaluate differences between women with and without milk samples. For HMO's components and variations in HMO group throughout lactation, Friedman's test and post-hoc analysis for multiple comparisons were performed (using a pairwise multiple comparison of mean ranks-PMCMR package), and a significance level <0.05.
For exploratory analysis, a Spearman rank correlation with heat maps (Corrplot Package) was performed to investigate the correlation between maternal characteristics (age, education, parity, gestational age at birth, GWG, pregnancy iron and folic acid supplement and pre-pregnancy weight, pre-pregnancy BMI) and HMO individuals' concentrations and alpha diversity index at the different time points. For these analyses, a significance level of <0.01 was adopted and correlations levels were interpreted as: 0.00 to 0.19-very weak correlation, 0.20 to 0.39-weak correlation, 0.40 to 0.69moderate correlation, 0.70 to 0.89-strong correlation and 0.90 to 1.00-very strong correlation [44].
A stacked bar chart plot (ggplot2 package) was derived to present an absolute concentration and relative abundance of HMOs. The ɑ-diversity index was calculated using Shannon diversity and evenness, inverse Simpson and Pielou evenness (biodiversity R package and vegan package). For index of diversity and evenness comparison in follow-up visits, a Friedman post-hoc test for multiple comparisons was performed.
Non-metric multidimensional scaling (NMDS) was used to define the number of dimensions, preserving the differences to explore patterns of HMO profiles, using the Bray-Curtis dissimilarity matrix (metaMDS in the vegan and ggplot2 package). Subsequently, an analysis of the main components of HMO profiles was performed using the Brunet method, which identified three basic components retained based on the classification estimate. The non-negative matrix factorization (NMF) was used to define potential patterns in the HMO profile data according to postpartum period, BMI nutritional status, parity, and maternal age. The HMO with the highest value on NMF represents the highest probability of that compound contributing to the HMO profile.
Sensitivity analyses were performed using only a subset of 15 women with samples in all time periods.

Participants and HMOs Concentrations
A total of 322 women were screened and 147 were recruited. However, 46 women did not collect breast milk samples for the following reasons: 31 missed the visit, 5 failed to collect the samples, 4 women did not breastfeed, 2 collected samples after the predefined time frame of the visit, 3 were excluded (gestational diabetes, pre-eclampsia and delivered a stillborn) and 1 refused to collect. Thus, 101 participants collected milk samples: 52 at 2-8 days (visit 1), 75 at 28-50 days (visit 2) and 46 at 88-119 days (visit 3). Overall, 15 participants had samples for all time points, 42 had samples at two visits and 44 for one visit ( Figure 1). Overall, 89.1% (n = 89) of the women were identified as secretors and 10.9% (n = 11) as nonsecretors. At 2-8 days, 90.4% of mothers were exclusively breastfeeding (n = 47) and 9.6 were using infant feeding formulas (n = 5). At 28-50 days 67.6% of mothers were exclusively breastfeeding (n = 49), 20.3% used infant feeding formulas (n = 15), 9.5% were predominantly breastfeeding (n = 7) and 2.7% (n = 2) were in complementary breastfeeding. At 88-119, 55.8% of mothers were exclusively breastfeeding (n = 24), 30.2% used infant formula feeding (n = 13) and 14.0% were predominantly breastfeeding (n = 6). No differences were observed when comparing women with and without milk samples (Table 1). IQR: interquartile range. Data of the participants with and without samples were not different according to Mann-Whitney, p < 0.05. 1 Baseline information. 2 Pre-pregnancy weight was self-reported. 3 Pre-pregnancy body mass index (BMI) was calculated with measured height at 28-50 days postpartum and pre-pregnancy weight. 4 Due to missing data, n = 95. 5 Due to missing data, n = 44. 6 Due to missing data, n = 35. 7 Due to missing data, n = 41. 8 Due to missing data, n = 15. 9 During pregnancy.

Relations between Maternal Anthropometric, Demographic and Reproductive Characteristics and HMO Compositions Concentrations
Spearman correlations varied between low and moderate values. Parity was directly correlated with LNFP II (0.

HMO Diversity and Evenness
The milk produced at 28-50 days presented the highest diversity (Shannon entropy index) and evenness (Shannon and Pielou evenness indexes), with higher median values of indexes compared to the other visits. Additionally, the milk produced in the last visit presented the lowest median for the same indexes of diversity and evenness (Table 4). All values are median and interquartile range. HMO: human milk oligosaccharide. a indicates that the HMO concentration at visit 1 is significantly different (p < 0.05) from the concentration at visit 2, b indicates that the HMO concentration at visit 1 is significantly different (p < 0.05) from the concentration at visit 3 and c indicates that the HMO concentration at visit 2 is significantly different (p < 0.05) from the concentration at visit 3 according to Friedman post-hoc test.

HMO Profiles (NMF)
The components that most contributed to HMO profiles at 2-8 days, 28-50 days, and 88-119 days were LNFP II, 3FL and 2ꞌFL, respectively. For all women, the major contribution derived from LNnT (Table 5) and also for age, except the third quartile (Table 6).    3ꞌSL was the HMO that most contributed to the profile of women with pre-pregnancy underweight, while LSTc contributed more for those with normal weight and 2ꞌFL represented the highest contribution for overweight and obese women (Table 7). Overall, 2'FL contributed the most to the HMO profile in women with ≤1 child while LNnT contributed the most to the HMO profile in women with two or more children, according to baseline information (Table 8).

Sensitivity Analyses with Women with Samples in All Time Periods (n = 15)
The majority of the results derived from the sensitivity analyses confirm the findings observed on the complete dataset, such as the smaller total HMO concentration at last visit in g/L, obese women with 2'fucosylactose as the main contributor to the HMO profile and major HMO diversity and evenness at visit 2 (data not shown). However, some minor differences were found. The most abundant HMO on non-secretor women in this subset on the last visit was different from the complete dataset (3syalillactose). Spearman rank significant correlations were observed between maternal age and FLNH at visit 2 and 3. Finally, maternal characteristics, such as primiparity and normal weight and excess weight, showed a different main contributor HMO in the subgroup, when compared with the complete dataset results.

Discussion
The present study has several interesting findings. First, differences of HMO concentrations over time were observed. Similar to previous studies [21,29], HMO concentrations in g/L decrease over time postpartum. However, when measured in molar concentrations, HMOs increase over time. This is due to a relative increase in low-molecular HMOs and a relative decrease in high-molecular HMOs over time. The data indicate that looking at weight concentrations in g/L alone provides only a limited picture of HMO composition changes over time. Second, we found that 89.1% of the women were secretors and that 2ꞌFL was the predominant HMO in their milk at all three collection periods. However, for nonsecretor women, LNT (at 2-8 days) and LNFPII (at 28-50 days and 88-119 days) were the most abundant HMOs. Third, differences between secretor and non-secretor women were found for 2ꞌFL, 3FL, 3ꞌSL, DFLac, DFLNH, DFLNT, FDSLNH, LNFP I, LNFP II, LNT, LSTb and LSTc-once again highlighting that single nucleotide polymorphisms in a single fucosyltransferase gene affect the synthesis of almost all HMOs, not only those directly fucosylated by the impaired enzyme.
McGuire et al. [22] studied the same 19 HMOs using our same analytical platform and found total concentrations of 10.1 g/L for Peruvian women at 15-150 days postpartum compared to 12.5 g/L for 2-8 days, 11.5 g/L for 28-50 days, and 11.3 g/L for 88-119 days in the present study. The only other Brazilian study detected 5.57 g/L of HMO at 17-76 days postpartum [46], which is quite different to the present results. Different analytical platforms (quantitative HPLC-FL vs. semi-quantitative LC-MS) and identification and quantification of different HMOs may contribute to the observed differences in HMO data between the two Brazilian cohorts. Results from a systematic review concluded that there is not a standard method to quantify HMOs yet, but also highlighted that mass spectrometry does not provide absolute quantitative measures, which is why we opted to use HPLC-FL with non-HMO internal standard combined with established HMO standard response curves to allow for absolute quantification [47].
We observed that concentrations of DSLNT, LSTc and LNFP I in secretor women decreased over time, which aligns with previously described data in longitudinal studies [48] and in cross-sectional studies with milk samples at different stages of breastfeeding [20,24].
The NMF procedure indicated that different HMOs were responsible for the major contribution of the HMO profiles at different post-partum time points. At 2-8 days, LNFP II presented the major contribution, while, at 28-50 and 88-119 days, 2ꞌFL and 3FL were the major contributors, respectively. Our findings partially agree with McGuire et al. [22] results, that found 2ꞌFL as the main contributor of HMO composition at 20-46 and 79-41 days postpartum.
In this study, pre-pregnancy BMI and pre-pregnancy weight were directly correlated with LNnT at 2-8 days. Unlike our study, McGuire et al. [22] observed that maternal weight and BMI were positively correlated with 2ꞌFL and FLNH, and Samuel et al. [21] described that overweight women presented a lower concentration of LNnT, when compared with normal weight women. Our data revealed that pre-pregnancy BMI and pre-pregnancy weight are inversely correlated with LNFP III at 2-8 days. In contrast, McGuire et al. [22] found maternal weight to be positively correlated with LNFP III. While these studies used the same HMO analytical platform, these differences could indicate a variation between the studied populations and by geographic location. Overall, these discrepancies between cohorts warrant further investigations on how maternal pre-pregnancy BMI, pregnancy weight gain, as well as overall maternal nutrition affect HMO composition and how these changes in HMO composition potentially impact infant health and development.
Azad et al. [23] observed higher concentrations of LNT and LNnT and lower concentrations of 3FL in multiparous women. Likewise, Tonon et al. [45] found a positive correlation between parity and LNT and LNnT and a negative correlation with 3FL. In contrast, Samuel et al. [21] observed higher concentrations of LNnT at 17 days postpartum in primiparous women. We observed that mothers with two children had 2ꞌFL as the main component that contributed to the HMO profile. It is possible that interpregnancy interval may play a role on this association, because this interval was previously described as being associated with changes in breastfeeding practices and could also be linked with human milk composition [49]. However, the mechanism that explains the relationship between parity and HMOs is still unclear and new investigations are necessary.
The sensitivity analyses results were mostly in line with the findings observed for the complete dataset and reaffirm the main conclusions of the study. However, it is important to highlight that some minor differences were observed. It is very likely that these minor differences can probably be attributable to the small sample size of the subset analyzed.
There are some limitations that pertain to this study. The first is that the small sample size for nonsecretors throughout postpartum limits statistical analysis and the comparison of our results with other studies. Additionally, stratifications according to the Lewis blood group were not possible due to this small sample size. Further, during milk collection the breast was not emptied completely. However, some studies have shown small or no variations of HMOs during the feeding. Thus, samples collected in the middle of feeding can be representative of the whole concentration [50,51]. Another limitation was the loss of follow-up during all study visits. Despite these limitations, the present study has important strengths. First, the use of a rigorous analytical platform to quantify HMO concentrations instead of merely measuring relative abundancies. Second, the use of robust statistical method to identify the HMO profile of Brazilian women, with NMF performed to identify patterns that together explain the data set. Third, the longitudinal design used allowed the assessment of prospective changes of HMO concentrations throughout the postpartum period. Last, to the best of our knowledge, this is the first study that has investigated HMOs concentrations at different time points during lactation in Latin American women.

Conclusions
The results found in the present study suggest that HMO composition varied throughout lactation in Brazilian women, with lower concentrations in g/L at 88-119 days, but an increase in concentrations of low-molecular HMOs at the same period. In addition, maternal pre-pregnancy BMI and parity were associated with HMO composition in healthy women in this Brazilian cohort. These study findings represent an important step toward knowing the profile and variation on HMOs composition in Brazilian women and the characteristics associated with this profile, mainly considering the geographic variation of the HMO concentration in different countries of the world. Additionally, this will be important for future studies, to evaluate the impact of HMO in infants' outcomes.
Supplementary Materials: The following are available online at www.mdpi.com/xxx/s1, Table S1: Time variation of HMOs concentrations in total and according to secretor status of women followed on Brazilian cohort. Rio de Janeiro, Brazil (n = 101).