A theoretic approach to the breast tissue microbiomes under tumor influences with near-neutral, neutral and niche-neutral hybrid models

Background : The structure and dynamics of breast tissue microbiomes can have far-reaching influences on women’s health, particularly on breast tumor development. However, there is currently little understanding on the ecological processes that shape the structure and dynamics of breast tissue microbiomes. Methods : Here we fill the gap by applying three metacommunity models for investigating the community assembly and diversity maintenance, including Sloan near neutral model, Harris et al . multisite neutral and Tang & Zhou niche-neutral hybrid models to reanalyze the 16s-rRNA sequencing datasets of 23 healthy, 12 benign tumor, and 33 malignant tumor tissue samples. To remedy the practical difficulty in collecting tissue microbiome samples, we adopted a sophisticated random re-sampling (up to 1000 times) scheme in applying the three metacommunity models for analyzing the patterns in the breast tissue microbiomes. Results : First, we found that, at the community/metacommunity levels, the mechanisms of microbiome assembly and diversity maintenance of breast tissue microbiomes were predominantly driven by stochastic drifts of bacteria demography (division, death and dispersal of bacterial cells), whereas the deterministic selection forces such as tumor progression were insignificant. However, at species level, on average, approximately 10% and 5% species were above (positively-selected) and below (negatively-selected) neutral, respectively. Furthermore, malignant tumor may raise the positively selected species up to 17%. Second, malignant tumor appears to inhibit microbial dispersal as evidenced by lowered migration rates, compared with the migration in normal and benign tumor tissues. Conclusions : The mechanisms of microbiome assembly and diversity maintenance of breast tissue microbiomes were predominantly driven by stochastic drifts, and malignant tumor may inhibit microbial dispersal. These theoretic findings can be inspirational for further investigating the relationships between tissue microbiomes and breast tumor progression/development.


Introduction
With the rapid development of 16S amplicon sequencing, the microbiome of breast milk has become an area of interest for research on the health of infants and mothers. The microbiome in human milk may be the seed source of gut microbiota in infants, with its composition and diversity found to be correlated with body-mass index (BMI), parity, mode of delivery, breastfeeding practices, and infant oral cavities. For example, breastfeeding mode is a key determining factor of milk microbiota composition (Moossavi et al., 2019). Furthermore, microbiomes can influence susceptibility to cancers and the response to therapeutics (Helmink et al., 2019). Many researchers have focused on the relationship between microbial profiles and disease development. Human milk and breast tissue contain microbial communities that are thought to be sterile (Hunt et al., 2011;Urbaniak et al., 2014Urbaniak et al., , 2016Hieken et al., 2016). Evidence indicates that distinct microbial communities exist among healthy, benign, and malignant breast tissue. Hunt et al. (2011) explored the microbial profiles in human milk based on pyrosequencing of the 16S ribosomal RNA gene and found that the milk microbiome is relatively stable over time within an individual. However, this is not always true. Ma et al. (2015) re-analyzed bacterial interactions using the same datasets as Hunt et al. (2011) and found that dysbiosis of the milk microbiome following a shift in the balance between potential opportunistic pathogens and harmless bacteria is likely responsible for some breast diseases. The gut microbiome can also affect breast cancer due to the estrobolome, i.e., bacterial genes capable of metabolizing estrogens, and thus can affect the emergence of estrogen-driven breast cancers (Kwa et al., 2016;Goedert et al., 2015).
As described previously, breastfeeding and estrogen can impact the breast microbiome profile and diversity. It has been reported that diversity in the milk microbial community is decreased in Hodgkin's lymphoma cohorts compared with healthy controls, and is positively correlated with milk metabolites that benefit children and negatively correlated with harmful metabolites related to mastitis and breast cancer (Ma et al., 2016). Obviously, the interaction between microbes and breast disease can change the composition and diversity of the microbiome. However, the mechanisms of breast microbiome construction in healthy and diseased groups remain unclear.
To explore the mechanisms of microbial community construction and diversity maintenance, niche theory and neutral theory can be applied to illustrate the role of stochastic and deterministic forces. Hubbell (2001) initially applied the unified neutral theory of biodiversity and biogeography (UNTB) to explain community construction mechanisms, with many researchers subsequently extending and challenging traditional neutral and niche theories (Etienne, 2005(Etienne, , 2007Hubbell 2006;Volkov, 2003Volkov, , 2007Sloan et al., 2006Sloan et al., , 2007McGill, 2003;Harris et al., 2017;Tang & Zhou, 2013, Li & Ma 2016Burns 2016). For example, Sloan et al. (2006Sloan et al. ( , 2007 extended Hubbell's discrete neutral theory to a continuous version to test large microbiomes. This model can define whether a species is neutral or not beyond the community level and can be applied to identify important microbes in the community construction process. In the current study, we applied Sloan's neutral community model to define neutral and non-neutral species in healthy (control), benign, and malignant tumor cohorts and to explain the dynamics of neutral species.
Compared with the original Hubbell (2001) UNTB model, Sloan (2006, 2007 model is actually a near neutral model since it allows for the existence of competitive advantages or disadvantages. Thanks to this advance, all species in a community can be categorized as three types: neutral species, negatively selected (under neutral) and positively selected (above neutral). We take advantage of this feature to detect the potential correlation between tumor development and bacterial species competitiveness. We also used another pair of models, i.e., multi-site neutral (MSN) (Harris et al., 2017) and niche-neutral hybrid models (NNH) (Tang & Zhou, 2013), to evaluate the relative significance of stochastic neutral drifts vs. deterministic niche selection in driving community assembly and shaping the diversity patterns of the breast tissue microbiome. The MSN by Harris et al. (2017) is a major computational advance to Hubbell's classic UNTB because it allows for simultaneously estimation of the migrations rates among large number of sites (local communities), which was a significant computational challenge until recently. Obviously, this simultaneous estimation of migration parameters is closer to reality. While the MSN model is an orthodox implementation of Hubbell's UNTB model, and Sloan model is a near neutral model, the NNH model is a mixture (hybrid) of neutral and niche mechanisms. Therefore, the three models we choose to apply in this study span the whole spectrum of the so-termed niche-neutral continuum, which postulates that different types of metacommunities are likely fall in different locations of the continuum. One end of the spectrum is occupied by the completely neutral assemblages, and another end is by completely nichedifferentiated assemblages. Therefore, the objective of this article is to examine the 'position' of the breast tissue microbiome on the niche-neutral continuum, particularly when tumor development occurs with the breast tissue. The integrated analysis with the three models allows us to present a relatively complete and reliable 'picture' of the process (mechanism) underlying the tissue microbiome distribution and dispersal patterns. In perspective, our study, if successful, is likely to offer important insights for investigating the relationship between breast tumor development and breast tissue microbiomes (Nejman et al 2020;Poore et al 2020). Urbaniak et al. (2016) collected samples from women following lumpectomies or mastectomies and from healthy individuals. The breast tissue microbiome datasets consisted of three groups: i.e., healthy (23 samples), benign tumor (12 samples), and malignant tumor tissues (33 samples). Samples were taken from normal tissue adjacent to the tumor, rather than from the tumor tissue itself. The 16S rRNA sequencing datasets were analyzed by QIIME to obtain the operational taxonomic unit (OTU) tables. Sloan et al. (2006Sloan et al. ( , 2007 derived a neutral model to explain the assembly mechanisms of prokaryotic communities. As a continuous version of Hubbell's discrete neutral community model, Sloan's model does not require observed species abundance distributions or patterns and can test very large prokaryotic communities. The model contains source and local communities, similar to "mainland" and "island" in the theory of island biogeography. We can first assume that the local community is saturated with NT individuals. One individual dies or leaves the local community and is replaced by another individual immigrating from a source community with probability m or offspring of a random individual within local community with probability 1-m. Thus, the probability that the abundance of the i-th OTU increases by one individual, decreases by one individual, or shows no change can be given by:

Sloan (2006, 2007) near neutral model
where pi is the occurrence frequency of the ith OTU in the source community and Ni is the abundance of ith OTU in the local community. Let xi=Ni/NT be the occurrence frequency of the ith OTU in the local community. The prediction abundance (fi) of community is the beta distribution: (4) where, .
From Sloan's model, we can judge whether each species is neutral or not. According to Burns et al. (2016), the process for testing Sloan's neutral model can be summarized as follows: (1) Compute pi and xi, fitting beta distribution and obtaining the estimation of m.
(2) Compute the theoretical occurrence frequency of species i across all local community samples with m and the beta distribution.
(3) Judge whether the observed xi of species i falls within its 95% theoretical interval predicted from the neutral community model, and obtain a list of neutral, below neutral, and above neutral species. Harris et al. (2017) is an implementation of Hubbell (2001) Harris et al. (2017).

Multi-site neutral (MSN) model
Niche-neutral hybrid model Tang & Zhou (2013) proposed a hybrid niche-neutral model for multiple discrete communities developed by Volkov et al. (2007). Volkov et al. (2007) assumed that interspecies interactions in a steady-state community can be ignored and all species in the community become functionally equivalent. Here, based on the datasets of the healthy and breast tumor microbiomes, we treated each sample as a niche occupied by a local microbial community and fit the neutral model for each local community. We used the p-value of the Chi-squared test to determine whether the metacommunity fit the NNH model. At the metacommunity level, if p > 0.05, the metacommunity satisfies the NNH and its assembly is co-driven by both niche and neutral processes, implying that the metacommunity itself does not satisfy the neutral theory, but within each niche, the local community is neutral; if p < 0.05, the metacommunity does not satisfy the NNH, implying that within each niche, the local community is not neutral either, and the metacommunity assembly is solely influenced by the niche process.

The overall modeling design and computational implementations
Summarizing previous sub-sections for the breast tissue microbiome datasets, as well as the three metacommunity models, we still need to design computational implementations to apply the models to the datasets for achieving our objectives-assessing and interpreting the relative important of stochastic neutral drifts and deterministic selection (specifically tumor progression in this study). Table 1 below outlines our study design. An important step in our design was the adoption of random re-sampling (reallocations) of the samples for 1000 times from each of the three tissue categories (normal, benign and malignant) or 100 times from each of the pair-wise two categories (to evaluate the directional changes). The re-sampling of 1000 (100) times was used build metacommunities, and therefore 1000(100) metacommunity models (Sloan, MSN or NNH) were built for the whole datasets. The random re-sampling was used to raise the robustness of the modeling, in particular to compensate for the relative small sample sizes. Obviously, the tissue microbiome samples are much more difficult to obtain than those in noninvasive studies such as stool or oral microbiome sample collections.

near neutral model
Using Sloan's model, all species in a metacommunity can be classified into three groups: neutral, below neutral (negatively selected) and above neutral (positively selected). As shown in Table 2,  Table S5A and Figure 1, when we treated the normal tissue microbiomes as source community and the benign tumor ones as destination community, 84.8% of the species belonged to neutral category, 5.6% were negatively selected species, and 9.6% were positively selected species. Given that the normal tissue microbiomes were treated as source community, when we treated the malignant tumor ones as destination community, 76.5% of the species were driven by stochastic neutral forces, 8% were driven by negatively selected, and 15.5% were driven by positively selected (Table 2, Table S5B & Figure 2). We also tested Sloan's model by treating the benign tumor tissue microbiomes as the source community and the malignant tumor ones as the destination community, in which 75.2% of the species satisfy neutral theory, 8% belonged to negatively selected species, and 16.8% belonged to positively selected species (Table 2, Table  S5C & Figure 3).
From these results and the results of significance tests of difference, we found that the percentage of neutral species in tissue microbiome significantly decreased with the transformation of tumor from benign to malignant; meanwhile, the percentage of non-neutral species, especially positively selected species, significantly increased when malignant transformation occurred (Fisher test: p-values < 0.001; Figure 4). Specifically, in the breast tissue microbiome, approximately 10% of the species is positively selected by the progress of benign tumorigenesis, but the percentage of the species positively selected by malignant transformation is up to about 17%.   The percentage of neutral, below neutral (negatively selected) and above neutral (positively selected) species in the three types of metacommunity settings: "normal to benign", "normal to malignant" and "benign to malignant". The "benign to malignant" group shows the lowest neutral-species percentage but highest percentage of the above-neutral species.  Table S1. Each of other three groups consisted of two microbiomes from two types of breast tissues. The group names of these three meta-communities were "normal & benign", "normal & malignant" and "benign & malignant". The full results of fitting to these three groups were listed in Table S2. Table 3 listed the mean parameters and corresponding standard errors of MSN models fitted to the breast microbiomes, which were summarized from Tables S1 and S2. Figure 5 shows an example of fitting the MSN model with the dataset of "normal & benign & malignant" group.

Determining the neutrality at local community and metacommunity levels with Harris et al
From these results, we found that all tested datasets passed the neutrality test with the MSN model (p-values > 0.9). It suggested that, at community level, the community assembly and diversity maintenance of breast tissue microbiome were driven predominantly by stochastic neutral forces, including dispersal and drift. The selections from tumor formation and malignant transformation were too weak to be detected. Table 3 also lists the average migration rates (Mvalues) between two types of breast tissue microbiomes. The microbial migration rates between normal and two types of tumor tissue were similar without significant difference (Wilcoxon test: p-value = 0.348 > 0.05), and the average M of them was 74.65. The migration rate between microbiomes of two tumor tissues was 54.321, and was significantly smaller than M-values of "normal & benign" and "normal & malignant" (Wilcoxon test: p-value < 0.001).     Table S3, and to other three groups were listed in Table S4. Table 4 listed the average NNH parameters and corresponding standard errors, which are summarized from Table S3 and Table S4.
From these results, we found that there was no tested dataset passing the test with the NNH model (p-values < 0.001). It further verified the previous finding, i.e., stochastic forces or neutral drift plays a predominant role in shaping the structure and diversity of breast tissue microbiome, while niche-differentiations or deterministic selection forces from tumor formation and malignant transformation plays little role.

Conclusions
The dispersal (migration) patterns of breast tissue microbiome In previous sections, based on Sloan near neural model (Table 2), we have demonstrated that while neutral species constitute, on average, approximately 85% species in the breast tissue microbiome, the selection effect from tumor progression does lead to certain percentage of above-neutral or positively selected species. The percentage of positively selected species is about 10% on average, but could be up to about 17% in the progression to malignant tumor. The MSN/NNH models, nevertheless, showed that the selection is not sufficiently strong to lead to community/meta-community level dominance of stochastic neutral forces. In other words, whereas the breast tissue microbiomes are driven predominantly by stochastic neutral forces, at the species level, up to 17% approximately of species can be positively selected by tumor progression.
With MSN model, we can further infer the migration rate between breast tissue microbiomes. Fig  6 shows the fundamental dispersal number or the migration rate (M) of pair-wise microbial migration between two types of breast tissue microbiomes (normal tissue and benign tumor; normal and benign tumor, benign and malignant tumor). Average M-values were previously displayed in Table 3 and detailed M-values were displayed in Table S1 & S2. The non-parametric Wilcoxon tests revealed no significant difference in the M-value between "normal & benign" and "normal & malignant" (p=0.348), but the difference in M was significant between "benign to malignant" and the two other groups mentioned above (P-value<0.001). This may indicate a significant difference between the benign and malignant tumor tissues in their microbial dispersal.

Ethical Approval and Consent to participate
Not applicable