Effect of storage, temperature, and extraction kit on the phylogenetic composition detected in the human milk microbiota

Abstract Human milk is considered the optimum feeding regime for newborns and is a source of bacteria for the developing infant gastrointestinal tract. However, as with all low biomass samples, standardization across variabilities such as sample collection, storage, and extraction methods is needed to eliminate discrepancies in microbial composition across studies. The aim of this study was to investigate how different storage methods, temperatures, preservatives, and extraction kits influence the human milk microbiome, compared to fresh samples. Breast milk samples were processed via six different methods: fresh (Method 1), frozen at −80°C (Method 2), treated with RNAlater and stored at 4°C or −80°C (Methods 3 and 4), and treated with Milk Preservation Solution at room temperature (Methods 5 and 6). Methods 1‐5 were extracted using PowerFoodTM Microbial DNA Isolation kit (Mobio), and Method 6 was extracted using Milk DNA Preservation and Isolation kit (Norgen BioTek). At genus level, the most abundant genera were shared across Methods 1‐5. Samples frozen at −80°C had fewest significant changes while samples treated and extracted using Milk Preservation and Isolation kit had the most significant changes when compared to fresh samples. Diversity analysis indicated that variation in microbiota composition was related to the method and extraction kit used. This study highlighted that, when extraction from fresh milk samples is not an option, freezing at −80°C is the next best option to preserve the integrity of the milk microbiome. Furthermore, our results demonstrate that choice of extraction kit had a profound impact on the microbiota populations detected in milk.


| INTRODUC TI ON
Human milk was initially considered sterile; however, numerous investigations have since identified milk as an integral source of bacteria for the developing infant (Cabrera-Rubio et al., 2012Hunt et al., 2011;Murphy et al., 2017). Historically, microbiological studies focused on milk contamination due to incorrect collection and storage and subsequent health implications to newborns (Larson et al., 1984;Ryder et al., 1977;West et al., 1979). Culture-based investigations provided the first evidence of a milk microbiome, although this approach was not without its limitations as some bacterial species are not readily cultivable. Advances in next generation sequencing technologies allowed for more detailed insight into the complex and diverse microbial composition of human milk. This has led to many studies seeking to characterize the milk microbiota, resulting in the identification of over several hundred bacterial species (Le Doare et al., 2018;Lyons et al., 2020).
Variations in the microbial composition of human milk are apparent across many studies. Murphy (Chen et al., 2018;Murphy et al., 2017).
These variations in the core genera of human milk may be attributed to many factors such as maternal health, diet and genetics, mode of delivery, and demographic and environmental differences (Browne et al., 2019;Cabrera-Rubio et al., 2012;Hermansson et al.,  However, aside from the influence of maternal and geographical determinants, characterization of the human milk microbiome is subject to many external factors and challenges such as sample collection, storage, and processing which is common with any low biomass sample. Standardization across sample collection, suitable storage conditions, and extraction techniques are essential in order to minimize the risk of contamination from exogenous sources and maintain the integrity of the bacterial community structure in milk (Lackey et al., 2019). Ojo-Okunola et al (2020) recently reported substantial differences in the human milk microbiota based on DNA extraction, emphasizing the need for careful consideration when selecting extraction kits. While it is known that cold storage immediately following sample collection and timely DNA extraction is optimum, in certain circumstances these options are not possible. It has been reported that short term cold storage (refrigeration, 4°C) or longer term freezing (−80°C) are the best methods to preserve microbial communities in many biological samples, with a study by Fouhy et al reporting that there were no significant differences in fecal microbiota when comparing fresh and frozen samples (Fouhy et al., 2015). Moreover, Hill et al documented the success of room temperature transport vials for preserving high diversity microbiota stool samples; however, this storage method is less suitable for low diversity samples such as infant stool samples (Hill et al., 2016).
Similarly, Chen et al noted that in addition to microbial differences arising from preservation solutions, the 16S rRNA gene primer pair chosen are critical determinants affecting gut microbiota composition (Chen et al., 2019).
To the best of our knowledge, limited research has been performed outlining the microbial differences between fresh milk and frozen milk samples (stored at −80°C) via culture-independent approaches. Furthermore, if the option of freezing is not available, the addition of preservatives to preserve the bacterial DNA profile in milk may be an alternative. The use of such preservatives to protect the microbial communities in other biological samples has been investigated with differing levels of success (Choo et al., 2015;Hill et al., 2016;Tap et al., 2019). Thus, knowledge on the impact of different storage and extraction methods are needed to assist planning for future projects. Therefore, the aim of this study is to investigate how different storage methods, temperatures, preservatives, and extraction kits influence the overall human milk microbiome, compared to fresh samples using MiSeq sequencing.

| DNA extractions
For Methods 1-4, 2.5 ml of milk was used for extractions and 0.5 ml used for Methods 5 and 6. Microbial DNA was extracted from Methods 1-5 milk samples using a modified protocol from the PowerFood TM Microbial DNA Isolation kit (MoBio, Carlsbad, CA). Briefly, samples were subject to an initial centrifugation 4000 g x 30 min at 4°C, the fat layer was removed with a sterile cotton swab (Thermo Fisher Scientific, Inc.) and supernatant discarded.
Cell pellets were washed with phosphate-buffered saline (Sigma Aldrich) and centrifuged at 13,000 g × 1 min at room temperature. A second wash step was performed using the same process. Samples were treated with 90 µl of 50 mg/ml lysozyme (Sigma Aldrich) and 50 µl of 5 KU/ml mutanolysin (Sigma Aldrich) followed by incuba-

| Bioinformatic and statistical analysis
Three hundred base pair paired-end reads were assembled using FLASH (FLASH: fast length adjustment of short reads to improve genome assemblies) (Magoč & Salzberg, 2011). QIIME was used for further processing of paired-end reads, and quality filtering was based on a quality score of >25 and removal of mismatched barcodes and sequences below length thresholds . Denoising, chimera detection, and clustering into operational taxonomic units (OTUs) were performed in QIIME using USEARCH v7 (64-bit) 3 (Edgar, 2010). OTUs were assigned using PyNAST (PyNAST: python nearest alignment space termination; a flexible tool for aligning sequences to a template alignment) and taxonomic rank was assigned using BLAST against the SILVA SSU Ref database release v123 Quast et al., 2012). Samples with <15,000 reads were excluded.
Statistical analysis was performed using R (version 3.6.3) and Calypso online software (version 8.84) (Team & R.C., 2013;Zakrzewski et al., 2017). To determine if statistically significant differences occurred in the microbial composition between fresh and each method tested, non-parametric Mann-Whitney analysis was completed using compareGroups package in R (Avramopoulos, 2017;Subirana et al., 2014). Statistical significance was accepted as p < 0.05. In Calypso, cumulative-sum scaling was used to normalize microbial community data and data were log2 transformed to account for the non-normal distribution of taxonomic count data for alpha and beta diversity testing. Alpha diversity was determined using the Shannon, Simpson's diversity, and Chao1 indices. Beta diversity was measured using principal coordinate analysis (PCoA) and adonis variance analysis based on Bray-Curtis distance matrices on data. Multivariate analysis was examined using redundancy analysis (RDA) and canonical correspondence analysis (CCA) method to investigate the associations between microbiota composition and explanatory variables.

| Quantitative polymerase chain reaction (qPCR)
Absolute quantification by qPCR was used to determine total bacterial numbers in milk samples extracted across different methods using the Roche LightCycler 480 II platform. A standard curve was created using 10 9 to 10 2 copies of 16S rRNA/µl to quantify total 16S bacterial counts. Amplification of samples was achieved using the forward primer 5'-ACTCCTACGGGAGGCAGCAG-3' and reverse

| RE SULTS
In this study, DNA was extracted from human milk samples subjected to six different storage, temperature, and extraction methods ( Figure 1). MiSeq sequencing was used to determine the effect of these variables on the milk microbiota when compared to samples extracted from fresh milk.

| MiSeq analysis of milk microbiota diversity following different storage and extraction methods
To examine the impact of storage conditions and extraction methods on beta diversity, PCoA plots were constructed based on the Bray-Curtis distance matrices at OTU level using relative abundance data. No clear separation based on storage and extraction methods were observed for Methods 1-5; however, milk samples stored and extracted using Method 6 appeared to cluster closely together in the same direction ( Figure 2a). There were no obvious clusters of milk samples based on storage temperatures, although separation is apparent in milk samples extracted using different extraction kits ( Figure 2b,c). Samples clustered more closely according to the individual, rather than to other samples in the same storage and extraction group except for Method 6 (Appendix Figure A1).
When all methods were compared to samples extracted from fresh (Method 1), the only significant difference was evident between Method 1 and Method 6 (adonis p = 0.00067) ( Figure 2d).
Although Methods 3 and 4 used the same preservative, RNAlater, F I G U R E 1 Flow diagram displaying the storage, temperature, and extraction kit to which samples were subjected 2.5 mL 0.5 mL 0.5 mL 2.5 mL but were stored at different temperatures (4°C and −80°C), no significant effect was seen between these groups (adonis p = 0.43).
Despite Methods 5 and 6 using the same milk preservation solution and storage temperature, Method 5 was extracted with the Mobio PowerFood TM Microbial DNA Isolation kit and Method 6 with the Norgen BioTek Milk Preservation and Isolation kit and a significant difference was observed between these groups (adonis To explore associations between composition and explanatory variables, redundancy analysis plots (RDA) were used and determined that preservative and extraction kit (p = 0.01, p = 0.001, respectively) had significant impacts on the milk microbiota. In addition, canonical correspondence analysis also determined that extraction kits (p = 0.001) had a significant impact on the milk microbiota, while preservatives did not (p = 0.053) (Figure 3a,b).
To determine if alterations in microbial diversity within samples occurred as a result of different storage and extraction methods alpha diversity was investigated. To estimate microbial richness we used the Chao1 test which showed no significant differences in bacterial richness across the methods (p = 0.83). To estimate microbial diversity, we applied Simpson's diversity index, and to predict microbial evenness, we used the Shannon index and both revealed no significant differences across methods (p = 0.11 and p = 0.052) ( Figure 4).

| MiSeq analysis of milk microbial composition following different storage and extraction methods
With regard to phylum, all methods shared common phyla. The predominant phyla across methods were Proteobacteria, Firmicutes, Bacteroidetes, and Actinobacteria, with Cyanobacteria present in higher levels in Method 6 (3.3%) compared with all other methods (<1%). Firmicutes dominated in Method 1 (47%), Bacteroidetes in Methods 2-5 (58-67%) and Proteobacteria in Method 6 (60%)  were among the most abundant genera with a mean relative abundance of ≥1% (Figure 7) ( Table 2). Streptococcus was significantly lower in Methods 2 and 6 (p = 0.031, p = 0.039, respectively) and The majority of these differences between the groups were among genera present in abundances of <1%. However, Rhizobium and Achromobacter present at relative abundances of 11% and 5%, respectively, were significantly higher in Method 6 (p = 0.001, Table A1).
To determine whether there were bacteria present in the milk microbiota that could discriminate based on method used, a feature selection statistical analysis LDA effect size (LEfSe) which determines the most likely taxa to explain differences between the groups was carried out at genus level. Method 6 had the highest number of discriminative genera when compared to fresh and all other methods.
There were 15 genera that discriminated Method 6, with Rhizobium having the greatest discriminatory power for Method 6 compared to all other methods (Figure 8).
After sequencing, filtering, and quality control, the Mobio Powerfood kit and Norgen Biotek kit negative controls yielded extremely low reads of 89 and 3 respectively. Due to low sequence reads, genera identified in the negative controls did not impact the microbiome of samples in this study (Appendix Table A2).

| Quantitative polymerase chain reaction (qPCR) of bacterial counts across methods
To determine if storage method and extraction kit had an impact on total bacterial counts in the milk microbiota, total 16S rRNA levels were determined by qPCR. Total gene copies were detected at similar levels across all groups and no significant differences were observed when methods were compared to fresh samples (Method 1) (Figure 9). It was established that storage method and extraction kit did not significantly impact total bacterial numbers.  (Douglas et al., 2020;Lackey et al., 2017).

| DISCUSS ION
Our aim was to determine if alterations in the milk microbiome occur There were six methods tested in this study to assess the impact on the microbiota of human milk. DNA extraction from freshly collected samples is considered the gold standard method (Cuív et al., 2011;Hill et al., 2016;Maukonen et al., 2012;Wu et al., 2010), how-   (Fouhy et al., 2015).
Our data show that samples treated with MPS and extracted using the Milk Preservation and Isolation kit (Method 6) were most different in bacterial composition to fresh samples. Gram-negative bacteria (Watson et al., 2019). Moreover, with regard to extraction kits chosen, the different nature of the protocols could explain the differing bacterial composition. It has been documented that the method of cell lysis can significantly influence the microbial communities, with methods using mechanical lysis resulting in higher amounts of DNA at a higher quality. The Mobio Powerfood kit incorporates both mechanical (bead-beating) and enzymatic lysis steps whereas the Norgen Biotek kit employs solely enzymatic lysis steps. It is also worth noting the importance of including negative controls in extractions and sequencing in order to ensure accuracy of results and eliminate any potential contaminants overriding the microbiome especially with low biomass samples. While it has been reported that some taxa are commonly identified in the "kitome" (Olomu et al., 2020;Salter et al., 2014;Weyrich et al., 2019), the low sequencing reads obtained in this study indicate the reagents were not contaminated and did not impact the taxa observed across samples in this study. Furthermore, in addition to extraction kit, multiple negative controls should be included throughout sequencing preparation in order to account for spurious sequences and identify contamination that may occur as reported previously (Hornung et al., 2019;Kim et al., 2017). Failure to identify contamination may lead to unreliable and inaccurate data and results, which is a pitfall of many low biomass studies.
This study also examined microbial diversity and found no significant differences between the methods in terms of alpha diversity.
However, Adonis variance analysis based on Bray-Curtis distance matrices found significance between Methods 1 and 6, and Methods 5 and 6, suggesting that the choice of extraction kit may be driving the separation between samples.
Although our study looked at short term storage (2 weeks) using different temperatures and extraction methods, further investigations are necessary to determine if prolonged storage has any significant effect on the milk microbiome composition. While processing samples from fresh is regarded as optimal, and we have used it as the method of comparison in our study, it was not known what the true microbiome composition of the milk samples were prior to processing. We are aware of the limitations of this study, and future research would benefit from the addition of a mock microbial community in order to determine the exact composition, and subsequent effect of storage temperature and extraction methods on the microbiota populations detected in milk. when planning future large-scale projects, and it will be essential to consider how samples were stored and processed when comparing data from different studies.

ACK N OWLED G EM ENTS
The authors wish to acknowledge the involvement of the women who provided breast milk samples for this study. We thank Prof. Paul

CO N FLI C T O F I NTE R E S T
None declared.