Ecology

C-STABILITY description
Organic matter representation
The description of SOM in C-STABILITY consists of several subdivisions. First, organic matter is separated in two main pools, one for living microbes (noted Cmic) and one for the substrate (noted Csub) (Fig. 1a). Several groups of living microbes can be considered simultaneously (e.g., bacteria, fungi, etc.) and C-STABILITY classes them into functional communities. Second, SOM is also separated between several biochemical classes, e.g., cellulose (or plant sugar), lignin, lipid, protein, and microbial sugar in this study (Fig. 1a). Third for each biochemical class, substrate accessible to its enzymes (noted ac) is separated from substrate which is inaccessible (noted in) due to specific physicochemical conditions, e.g., interaction between different molecules, inclusion in aggregates, sorption on mineral surfaces, etc.
Polymerization is a driver of interactions between substrate and living microbes and a continuous description of the degree of organic matter polymerization (noted p) is provided for each of these pools, as a distribution (Fig. 1b). The polymerization axis is oriented from the lowest to the highest degree of polymerization. A right-sided distribution corresponds to a highly polymerized substrate whereas a left-sided distribution corresponds to monomer or small oligomer forms. For each biochemical class ∗ (∗ = cellulose, lignin, lipid, etc.), the polymerization range is identical for both accessible and inaccessible pools. The total amount of C (in gC) in the accessible and the inaccessible pools of any biochemical class is as follows:

$${C}_{* }^{rm{ac}}=int_{{p}_{* }^{rm{min}}}^{{p}_{* }^{rm{max}}}{chi }_{* }^{rm{ac}}(p)dp,$$
(1)

$${C}_{* }^{rm{in}}=int_{{p}_{* }^{rm{min}}}^{{p}_{* }^{rm{max}}}{chi }_{* }^{rm{in}}(p)dp,$$
(2)

where ({p}_{* }^{rm{min}}) and ({p}_{* }^{rm{max}}) are the minimum and maximum degrees of polymerization of the biochemical class ∗ and ({chi }_{* }^{rm{ac}}), ({chi }_{* }^{rm{in}}) (gC.p−1) are the polymerization distributions. Finally, the total substrate C pool is defined as the sum of all biochemical pools,

$${C}_{rm{sub}}=sum _{* }left({C}_{* }^{rm{in}}+{C}_{* }^{rm{ac}}right).$$
(3)

Accessibility to microbe uptake is described by the interval (also called domain) ({{mathcal{D}}}_{u}), which corresponds to small substrate compounds, monomers, dimers or trimers smaller than 600 Daltons1, that microbes are able to take up (in red in Fig. 1b). Besides, accessibility to enzymes occurs in the ({{mathcal{D}}}_{rm{enz}}) domain (in blue in Fig. 1b). Over time the substrate accessible to enzymes is depolymerized and its distribution shifts toward ({{mathcal{D}}}_{u}) where it eventually becomes accessible to microbe uptake.
The numerical rules chosen to represent polymerization are as simple as possible in the context of theoretical simulations. Each pool is associated with a polymerization interval ([{p}_{* }^{rm{min}},{p}_{* }^{rm{max}}]) of length two. Initial substrate distributions are represented by Gaussian distributions centered at a relative distance of 25% from ({p}_{* }^{rm{max}}) (here 0.5), with the standard deviation set at 5% of polymerization interval length (here 0.1). In the accessible pool, the microbial uptake domains ({{mathcal{D}}}_{u}) are positioned at the left of the interval with a relative length of 20% (here 0.4), and enzymatic domains ({{mathcal{D}}}_{rm{enz}}) overlap the entire polymerization intervals.
Organic matter dynamics
As described in Fig. 1, three processes drive OM dynamics: (i) enzymatic activity, (ii) microbial uptake, biotransformation, and mortality, and (iii) changes in local physicochemical conditions. First, enzymes have a depolymerization role, which enables the transformation of highly polymerized substrate into fragments accessible to microbes. Second, microbial uptake of substrate is only possible for molecules having a very small degree of polymerization. When C is taken up, a fraction is respired and the remaining is metabolized, and biotransformed into microbial molecules that return to the substrate upon microbe death. Each microbial group has a specific signature that describes its composition in terms of biochemistry and polymerization. Third, changes in local substrate conditions drive exchanges between substrate accessible and inaccessible to enzymes (e.g., aggregate formation and break). All of these processes are considered with a daily time step (noted d).
Enzymatic activity Enzymes are specific to biochemical classes. They are not individually reported, but rather as a family of enzymes contributing to the depolymerization of a biochemical substrate (e.g., combined action of endoglucanase, exoglucanase, betaglucosidase, etc., on cellulose will be reported as cellulolytic action). Figure 2 describes how substrate polymerization distributions are impacted by enzymes. The overall functioning of each enzyme family (noted enz) is described by two parameters: a depolymerization rate ({tau }_{rm{enz}}^{0}) providing the number of broken bonds per time unit and a factor accounting for the type of substrate cleavage αenz. The term ({F}_{rm{enz}}^{rm{act}}) (gC.p−1.d−1) represents the change in polymerization of ({chi }_{* }^{rm{ac}}) due to enzyme activity for all (pin {{mathcal{D}}}_{rm{enz}}),

$${F}_{rm{enz}}^{rm{act}}({chi }_{* }^{rm{ac}},p,t)= -{tau }_{rm{enz}}(t){chi }_{* }^{rm{ac}}(p,t)\ +int_{{{mathcal{D}}}_{rm{enz}}}{{mathcal{K}}}_{rm{enz}}(p,p^{prime} ){tau }_{rm{enz}}(t){chi }_{* }^{rm{ac}}(p^{prime} ,t)dp^{prime}.$$
(4)

The depolymerization rate, τenz (d−1), is expressed as a linear function of microbial C biomass Cmic (gC),

$${tau }_{rm{enz}}(t)={tau }_{rm{enz}}^{0}{C}_{rm{mic}}(t),$$
(5)

where ({tau }_{rm{enz}}^{0}) (g({,}_{C}^{-1}).d−1) is the action rate of a given enzyme per amount of microbial C. If several microbial communities are associated with the same enzyme family, we replace the Cmic term by a weighted sum of the C mass of all communities involved in Eq. (5). The ({{mathcal{K}}}_{rm{enz}}) (p−1) kernel provides the polymerization change from (p^{prime}) to p,

$${{mathcal{K}}}_{rm{enz}}(p,p^{prime} )={{mathbb{1}}}_{ple p^{prime} }({alpha }_{rm{enz}}+1)frac{{(p-{p}_{* }^{rm{min}})}^{{alpha }_{rm{enz}}}}{{(p^{prime} -{p}_{* }^{rm{min}})}^{{alpha }_{rm{enz}}+1}},$$
(6)

where ({{mathbb{1}}}_{ple p^{prime} }) equals 1 if (ple p^{prime}) and 0 otherwise. The αenz cleavage factor denotes the enzyme efficiency to generate a large amount of small fragments and to shift the substrate polymerization distribution toward the microbe uptake domain ({{mathcal{D}}}_{u}) (Fig. 2). αenz = 1 is typical of the action of endo-cleaving enzymes, which randomly disrupts any bond of its polymeric substrate and generates oligomers. The shift toward ({{mathcal{D}}}_{u}) is slower if αenz increases. This is characteristic of exo-cleaving enzymes, which attack the end-members of their polymeric substrate, generate small fragments, and preserve highly polymerized compounds. To satisfy the mass balance, the kernel verifies (int {{mathcal{K}}}_{rm{enz}}(p,p^{prime} )dp=1). Then ({F}_{rm{enz}}^{rm{act}}) does not change the total C mass but only the polymerization distribution (i.e., (int {F}_{rm{enz}}^{rm{act}}(chi ,p,t)dp=0)).
Microbial biotransformation Each microbial group (denoted mic) produces new organic compounds from the assimilated C. After death, the composition of the necromass returning to each biochemical pool ∗ of SOM is assumed to be constant, accessible and is depicted with a set of distributions smic,∗, named signature. Each distribution smic,* (p–1) describes the polymerization of the dead microbial compounds returning to the pool ∗. The signature is normalized and unitless to ensure mass conservation, i.e., if we note that,

$${S}_{rm{mic},* }=mathop{int}nolimits_{{p}_{* }^{rm{min}}}^{{p}_{* }^{rm{max}}}{s}_{rm{mic},* }(p)dp,$$
(7)

then we have ∑*Smic,* = 1.
For each accessible pool of substrate, the term ({F}_{rm{mic},* }^{rm{upt}}) describes how the microbes utilize the substrate available in the microbial uptake ({{mathcal{D}}}_{u}) domain (Fig. 1b). For all (pin {{mathcal{D}}}_{u}),

$${F}_{rm{mic},* }^{rm{upt}}({chi }^{rm{ac}},p,t)={u}_{rm{mic},* }^{0}{C}_{rm{mic}}(t){chi }_{* }^{rm{ac}}(p,t),$$
(8)

where ({u}_{rm{mic},* }^{0}) (g({,}_{C}^{-1}).d−1) is the uptake rate per amount of microbe C. The substrate uptake rate linearly depends on the microbial C quantity.
Depending on a carbon use efficiency parameter ({e}_{rm{mic},* }^{0}) (ratio between microbe assimilated C and taken up C), taken up C is respired or assimilated and biotransformed into microbial metabolites. This induces a change in the biochemistry and polymerization (Fig. 1a).
Finally, microbial necromass returns to the substrate pools with a specific mortality, which linearly depends on the microbial C quantity,

$${F}_{rm{mic},* }^{rm{nec}}(p,t)={m}_{rm{mic}}^{0}{C}_{rm{mic}}(t){s}_{rm{mic},* }(p),$$
(9)

where ({m}_{rm{mic}}^{0}) (d−1) is the mortality rate of the microbe (Fig. 1a).
Change in local physicochemical conditions The polymerization of a substrate inaccessible to its enzymes remains unchanged over time. A specific event changing the accessibility to enzymes (e.g., aggregate disruption or desorption from mineral surfaces) is modeled with a flux from the inaccessible to the accessible pool. Transfer between these pools is described by the ({F}_{rm{ac},* }^{rm{loc}}) term for each biochemistry ∗,

$${F}_{rm{ac},* }^{rm{loc}}(p,t)={tau }_{rm{tr}}^{rm{ac}}{chi }_{* }^{rm{in}}(p,t)$$
(10)

where ({tau }_{rm{tr},* }^{rm{ac}}) (d−1) is rate of local condition change toward accessibility.
Transfer in the opposite way (e.g., aggregate formation, association with mineral surfaces) is described by the ({F}_{rm{in},* }^{rm{loc}}) term,

$${F}_{rm{in},* }^{rm{loc}}(p,t)={tau }_{rm{tr}}^{rm{in}}{chi }_{* }^{rm{ac}}(p,t)$$
(11)

where ({tau }_{rm{tr},* }^{rm{in}}) (d−1) is rate of local condition change toward inaccessibility.
Organic matter input We defined time dependent distributions for carbon input fluxes. There are denoted ({i}_{* }^{rm{ac}}) and ({i}_{* }^{rm{in}}) (gC.p−1.d−1) for both accessible and inaccessible pools of biochemical classes ∗. The total carbon input flux, expressed in gC.d−1, is:

$$I(t)=sum _{* }int_{{p}_{* }^{rm{min}}}^{{p}_{* }^{rm{max}}}left({i}_{* }^{rm{in}}(p,t)+{i}_{* }^{rm{ac}}(p,t)right)dp.$$
(12)

General dynamics equations The distribution dynamics for each biochemical class * is obtained from Eqs. (4)–(6) and (8)–(11),

$$frac{partial {chi }_{* }^{rm{ac}}}{partial t}(p,t)= , {F}_{rm{ac},* }^{loc}(p,t)-{F}_{rm{in},* }^{rm{loc}}(p,t)\ +{F}_{rm{enz}}^{rm{act}}({chi }_{* }^{rm{ac}},p,t) +sum _{{rm{mic}}}left({F}_{rm{mic},* }^{rm{nec}}(p,t)-{F}_{rm{mic},* }^{rm{upt}}({chi }^{rm{ac}},p,t)right) +{i}_{* }^{rm{ac}}(p,t),$$
(13)

$$frac{partial {chi }_{* }^{rm{in}}}{partial t}(p,t)={F}_{rm{in},* }^{rm{loc}}(p,t)-{F}_{rm{ac},* }^{rm{loc}}(p,t) +{i}_{* }^{rm{in}}(p,t).$$
(14)

Then, the expended equations are,

$$frac{partial {chi }_{* }^{rm{ac}}}{partial t}(p,t)= , {tau }_{rm{tr},* }^{rm{ac}}{chi }_{* }^{rm{in}}(p,t)-{tau }_{rm{tr},* }^{rm{in}}{chi }_{* }^{rm{ac}}(p,t)\ -{tau }_{rm{enz}}^{0}sum _{,text{mic},}{C}_{rm{mic}}(t){chi }_{* }^{rm{ac}}(p,t)\ +{tau }_{rm{enz}}^{0}sum _{,text{mic},}{C}_{rm{mic}}(t)({alpha }_{rm{enz}}+1)\ int_{p}^{{p}_{* }^{rm{max}}}frac{{(p-{p}_{* }^{rm{min}})}^{{alpha }_{rm{enz}}}}{{(p^{prime} -{p}_{* }^{rm{min}})}^{{alpha }_{rm{enz}}+1}}{chi }_{* }^{rm{ac}}(p^{prime} ,t)dp^{prime} \ +sum _{,text{mic},}{C}_{rm{mic}}(t){m}_{rm{mic}}^{0}{s}_{rm{mic},* }(p)\ -sum _{,text{mic},}{C}_{rm{mic}}(t){{mathbb{1}}}_{{{mathcal{D}}}_{u}}(p){u}_{rm{mic},* }^{0}{chi }_{* }^{rm{ac}}(p,t)\ +{i}_{* }^{rm{ac}}(p,t),$$
(15)

$$frac{partial {chi }_{* }^{rm{in}}}{partial t}(p,t)={tau }_{rm{tr},* }^{rm{in}}{chi }_{* }^{rm{ac}}(p,t)-{tau }_{rm{tr},* }^{rm{ac}}{chi }_{* }^{rm{in}}(p,t)+{i}_{* }^{rm{in}}(p,t).$$
(16)

where ({{mathbb{1}}}_{{{mathcal{D}}}_{u}}(p)) equals 1 if (pin {{mathcal{D}}}_{u}) and 0 otherwise.
The dynamics of the total substrate is ruled by,

$$frac{d{C}_{rm{sub}}(t)}{dt}=sum _{* }int_{{p}_{* }^{rm{min}}}^{{p}_{* }^{rm{max}}}left(frac{partial {chi }_{* }^{rm{ac}}}{partial t}(p,t)+frac{partial {chi }_{* }^{rm{in}}}{partial t}(p,t)right)dp.$$
(17)

The dynamics of microbial Cmic is obtained by,

$$frac{d{C}_{rm{mic}}}{dt}(t)=-{m}_{rm{mic}}^{0}{C}_{rm{mic}}(t) +sum _{* }{u}_{rm{mic},* }^{0}{e}_{rm{mic},* }^{0}{C}_{rm{mic}}(t)int_{{{mathcal{D}}}_{u}}{chi }_{* }^{rm{ac}}(p,t)dp,$$
(18)

and the CO2 flux (gC.d−1) produced by the microbes is given by,

$${F}_{rm{CO}_{2}}(t)={C}_{rm{mic}}(t)sum _{* }{u}_{rm{mic},* }^{0}left(1-{e}_{rm{mic},* }^{0}right){int}_{{{mathcal{D}}}_{u}}{chi }_{* }^{rm{ac}}(p,t)dp.$$
(19)

Model implementation
The model was implemented in the Julia© language63,64. An explicit finite difference scheme approximates the solutions of integro-differential equations with a Δt = 0.1d time step and a Δp = 0.01p polymerization step. Differential equations were solved with a Runge–Kutta method.
Scenarios
Scenario 1: cellulose decomposition kinetics and model sensitivity
A first simulation was run to depict cellulose depolymerization and uptake by a decomposer community over one year (see parameters in Table 1). A global sensitivity analysis focusing on the residual cellulose variable was made to determine (i) the relative influence of parameters, and (ii) how parameters influence varies over time (Fig. 3c). A specific attention was given on enzymatic parameters (especially α) to verify the pertinence of their introduction in the model.
We considered a specific method defined by Sobol for calculating sensitivity indices65. It provides the relative contribution of the model parameters to the total model variance, here at different times of the simulation. The method relies on the same principle as the analysis of variance. It was designed to decompose the variance of a model output according to the various degrees of interaction between the n uncertain parameters ({({x}_{i})}_{iin {1,n}}). Formally, by assuming that the parameter uncertainties are independent, the model output, denoted y, could be expressed as a sum of functions that take parameter interactions into account,

$$y={f}_{0}+mathop{sum }limits_{i=1}^{n}{f}_{i}({x}_{i})+mathop{sum }limits_{{i,j=1}atop {ine j}}^{n}{f}_{i,j}({x}_{i},{x}_{j})+…+{f}_{1, , …, , n}({x}_{1},…,{x}_{n}).$$
(20)

Under independence assumptions between models parameters variability, model variance is:

$${{mathbb{V}}ar}(y)= , mathop{sum }limits_{i=1}^{n}{{mathbb{V}}ar}({f}_{i}({x}_{i}))\ +mathop{sum }limits_{{i,j=1}atop {ine j}}^{n}{{mathbb{V}}ar}({f}_{i,j}({x}_{i},{x}_{j}))\ +…+{{mathbb{V}}ar}({f}_{1, , …, , n}({x}_{1},…,{x}_{n})).$$
(21)

This variance decomposition leads to the definition of several sensitivity indices. The first-order Sobol’s index of each parameter is,

$${S}_{i}=frac{{{mathbb{V}}ar}({f}_{i}({x}_{i}))}{{{mathbb{V}}ar}(y)},$$
(22)

and higher order indices are defined by:

$${S}_{i,j}=frac{{{mathbb{V}}ar}({f}_{i,j}({x}_{i},{x}_{j}))}{{{mathbb{V}}ar}(y)},$$
(23)

and so on. These indices are unique, with a value of 0–1 and their sum equals 1. Here we focused on Sobol’s first-order indices as they are usually sufficient to give a straightforward interpretation of the actual influence of different parameters66,67. We computed the sensitivity of the model outputs at several times of the simulation to highlight the role of model’s parameters at different phases. Figure 3c shows the normalized Sobol’s first-order indices to illustrate the relative influence of the model parameters on residual cellulose-C amount. Sobol’s indices were estimated using a Monte Carlo estimator of the variance68. This was performed for a small variation in parameter values (±5% uniform variability), by running 12,000 model simulations for the Monte Carlo sampling.
Scenario 2: effect of substrate inaccessibility to enzyme on litter decomposition kinetics
A simulation of lignocellulose (76% cellulose, 24% lignin) degradation was performed by taking into account peroxidases, which deconstruct the lignin polymer, and cellulases, which hydrolyze cellulose. The cellulose was initially embedded in lignin and inaccessible to cellulase. The lignolytic activity (peroxidases) induces a disentanglement of the cellulose from the lignocellulosic complex. Therefore, the action of peroxidases was seen as a change of cellulose physicochemical local conditions resulting in a progressive transfer to the accessible pool. This transfer was assumed to be linearly related to the activity of lignolytic enzymes in Eq. (10),

$${tau }_{rm{tr,cell.}}^{rm{ac}}={tau }_{rm{tr,cell.}}^{rm{ac},0}int_{{{mathcal{D}}}_{rm{lig.}}}{tau }_{rm{lig.}}^{0}{C}_{rm{mic}}(t){chi }_{rm{lig.}}^{rm{ac}}(p,t)dp,$$
(24)

where ({{mathcal{D}}}_{rm{lig.}}) is the domain of lignolytic activity and where the ({tau }_{rm{tr,cell.}}^{ac,0}) coefficient is set at 13 g({,}_{C}^{-1}) for the current illustration.
The simulation was performed over one year. Enzymatic and microbial parameters given in Table 1 were chosen to be closely in line with the litter decomposition and enzyme action observation16,37,69.
Scenario 3: effect of community succession on C fluxes and substrate biochemistry
We simulated the succession of two microbial functional communities, on the same previous lignocellulose, considering microbial residue recycling. The parameters (Table 1) were chosen according to the microbial community succession observations43,45,70. The first microbial community was specialized in plant substrate degradation, the second was specialized in the degradation of microbial residues. We referred to them as plant decomposers and microbial residue decomposers. Microbial residue decomposers were more competitive than plant decomposers because of their higher carbon use efficiency and lower mortality rate (Table 1). Both communities had the same biochemical signature, i.e., 50% polysaccharides, 30% lipids, and 20% proteins. We tested the impact of cheating as follows. Either uptake was impossible, i.e., u0 equaled 0 for the community not involved in enzyme production, or uptake was possible but at a lower rate than the enzyme producers because the substrate fragments were released in the vicinity of the enzyme producers (Table 1).
Scenario 4: soil organic matter composition at steady state
We resolved the analytic formulation of the C stock and chemistry at steady state under several assumptions. We only considered one microbial community, a continuous constant plant input I at a rate of 2.74.10−4 gC.cm−2.d−1 and microbial recycling47. To be able to explicitly calculate the steady state, we only considered accessible pools, then cellulose substrate was not embedded in lignin but directly accessible to cellulolytic enzymes (Fig. 6). Finally, we considered that C use efficiency (({e}_{rm{mic}}^{0})) and uptake (({u}_{rm{mic}}^{0})) parameters were identical for all biochemical classes. A full mathematical proof is given in Supplementary Note 3.
At steady state, the amount of microbial carbon is,

$${C}_{rm{mic}}=frac{{e}_{rm{mic}}^{0}I}{(1-{e}_{rm{mic}}^{0}){m}_{rm{mic}}^{0}}.$$
(25)

For each biochemical class ∗, we define ({p}_{* }^{u}) which verifies ({p}_{* }^{rm{min}} More

Many precautions should be taken to limit the modification of the commensal communities studied and the increase of interindividual variation not attributable to the experimental variables. The following factors can influence the human microbiota and should be considered when designing studies targeting the lung microbiota: the administration of antibiotics or neoadjuvant25,26,27,28, the size of the lesion, the type of surgical procedure, the type of pulmonary pathology under study, and living habits of patients (e.g., smoking status, physical exercise, buccal hygiene, alcohol consumption)29,30,31,32,33,34.
A more exhaustive list of concomitant factors was pointed out by Carney et al.35. However, as the different fields of microbiota studies expand, it is likely that additional variables that can alter its composition will be uncovered. The molecular tools currently used to analyze the human microbiota do not have the power to discriminate the impact of that many factors over the microbial profiles. Whenever possible, patients selected for lung microbiota studies should be extensively screened so that they can be as similar as possible. Longitudinal studies could also minimize the impact of those variables, as the same patient, with similar concomitant factors through the study, would be compared to himself overtime.
Tissue management steps should consider the contamination possibilities. In addition to the selection of a less contamination-prone procedure, such as thoracoscopic lobectomy, the manipulations and the instrument used in subsampling the excised organ should be taken into account. A combination of bleach and humid heat was chosen to sterilize the instruments used to sample the cancerous and healthy tissue as it was considered the most easily accessible method. The use of humid heat itself (autoclave) lacks the power to completely neutralize bacterial genomic DNA in solutions and on surfaces36. On the other hand, the utilization of bleach, or a chlorinated detergent, leads to the complete degradation of contaminating DNA on surfaces, such as benches and instruments37,38, but requires rinsing to avoid corrosion. Hence, combining both methods, soaking the instruments in bleach 1.6% for 10 min before rinsing with distilled water and autoclaving in a sterilization pouches, ensures a minimal amount of DNA has to be degraded by moist heat. The rest of the single-use equipment used was commercially sterilized with ionizing radiation.
Healthy lung tissue was subsampled from the pulmonary lobe containing the tumor to ensure that the developed method could be used on a variety of lung tissue samples. It could also act as a control of non-pathologic microbiota to allow comparisons of cancerous and non-cancerous samples within the same subject, hence minimizing the impact on inter-individual microbiota variations. In fact, Riquelme et al. found that the gut microbiota has the capacity to specifically colonize pancreatic tissue8. Correspondingly, the use of adjacent pulmonary tissue to the tumor could help get better insights at a specific colonization of the tumor by lung bacteria. A 5 cm distance between the tumor and the healthy sample was ensured to minimize the potential effect of increased inflammation surrounding the tumor. Furthermore, the lung microbiota composition seems to vary dependently on the position and depth of the respiratory tract, even inside a same lobe39. The healthy tissue was collected in the same tierce of pulmonary depth (Supplementary Fig. 4) in an attempt to sample a microbial community that it would be as representative of non-pathologic microbiota in the tumoral region as possible.
The homogenization of frozen and thawed pulmonary tissues was attempted and was unsuccessful, both with the use of only a 2.8 mm tungsten bead in the Retsch – MM301 mixer mill (30 beats/s, 20 min) or of the Fisherbrand 150 homogenizer with plastic probes (Fisher scientific, Pittsburg). The elasticity of the tissue or its frozen state make the mass nearly unbreakable. The use of the Liberase™ TM enzymatic cocktail (collagenase I & II, thermolysin) prior to the mechanical homogenization proved successful and a homogeneous suspension was obtained using the two-step homogenization protocol (Supplementary Fig. 3). Multiple ratios of liquid to mass of tissue were tested and 3 mL/g was found optimal, as it facilitates the homogenization without overly diluting to sample. A similar ratio of liquid to tissue was used in breast tissue microbiota study40. The samples were first thawed at 4 °C to reduce potential growth or degradation of microorganisms. The digestion was performed directly in the 50 mL collection tube to limit the tissue manipulation and ensure possible contaminant tracking.
Our team was also unable to replicate the results obtain by Yu et al. on larger tissue samples using 0.2 mg/mL of Proteinase K for 24 h13. The samples remained firm and turned brown. Using the Liberase™ cocktail enabled a much faster digestion (75 min) and broke down specifically the lung component responsible for its elasticity, the collagen.
Three commercially available DNA extraction kits were tested. They were selected for their previous successful use in the study of pulmonary or gut microbiota and their intended application as described by the manufacturer. The extraction kits were first tested on homogenized lung tissue spiked with whole-cell bacterial community to assess the efficiency of DNA extraction and recuperation of the commercial kits. The three kits were able to recover more than 88% of the genera added to the samples. All the genera that were not detected by the Microbial and Powersoil (Cutibacterium acnes, Bacteroides vulgatus, Bifidobacterium adolescentis, D. radiodurans, Clostridium beijerinckii, L. gasseri), with the exception of H. pylori, were Gram-positive bacteria. This type of bacteria has been reported to require more aggressive extraction methods to break their tougher cell walls19. However, the bacterial community did not go through the enzymatic and physical homogenization that usually takes place before DNA extraction since we needed to obtain a homogenous tissue sample that could be processed with or without spiked bacteria. These hard to lyse Gram-positive bacteria could have been fragilized by these processes, rendering them easier to break down during the extraction protocol. Furthermore, the detection of the artificially incorporated bacteria does not account for the natural physical association that may occur between the human tissue and microbial cells. Nonetheless, these high percentages of recovery were promising and lead us to continue with the characterization of the extraction kits in a real-life context, meaning the analysis of the base-level microbiota in pulmonary samples collected and processed through the entire pipeline.
Every measurement of the efficiency of extraction, including DNA yield (Supplementary Fig. 5), DNA purity (Supplementary Figs. 6 and 7), and alpha diversity (Fig. 1), pointed in the same direction. In fact, they all showed that the Blood extraction kit was the best option out of the three kits. Therefore, using the Blood kit is recommended as one of the pieces of a complete study design. Additionally, the presence of a high concentration of host DNA in tissue samples might tend to saturate the purification column, which could reduce to amount of bacterial DNA recovered. The superior DNA binding capacity of the affinity column of the Blood kit compared to the two others could explain its better performance and its higher yield in most cases. The samples extracted with the Blood kit were also associated with higher alpha diversity (Shannon index). Therefore, this extraction method was able to recover a higher number of different bacterial organisms (richness) and proportionality in the different OTUs (evenness). The absence of PCR inhibitors and a higher recuperation rate of bacterial DNA in the Blood extracted samples could have led to a more proper amplification in the sequencing process and to the recuperation of very low abundance bacterial DNA in the extraction eluate. For further research, it is advised to take the additional precaution of working under a biosafety cabinet or in the sterile field when analyzing the microbiota of lung tissues to reduce the risk of incorporation of airborne contaminants.
The Illumina Miseq sequencing platform with the use of dual-index strategy has become the dominant technology used in microbial ecology studies for its cost efficiency, low error rate, and user-friendliness41,42,43. Most studies interested in the pulmonary microbiota have also used this technology11,13,14. The sequencing of the 16S rRNA gene amplicon was favored over a shotgun sequencing method because of the overwhelming quantity of human DNA joining bacterial genomes in the pulmonary tissue. The 16S rRNA gene is the most used marker of bacterial identification. No consensus has been reached on the selection of the 16S rRNA gene variable region (V) to sequence for human microbiota18,44. However, it should be kept consistent across studies to allow comparisons. Targeting the V3–V4 regions was suggested using the universal primers developed by Klindworth et al.45. Several microbiota studies, including lung microbiota, have also used these regions7,13,46,47,48.
In the context of this study, genomic mock-community was spiked in DNA extracted from the pulmonary tissue at a biological meaningful concentration. Every genus added to the samples was successfully detected. Consequently, the high ratio of human DNA to bacterial DNA did not interfere with the amplification and detection steps of the sequencing procedure. The sequencing method in place seems adequate for its application in the characterization of pulmonary microbiota.
Contaminating bacteria or DNA can have an important impact of the microbial profile observed in very low biomass samples such as pulmonary tissue23. Consequentially, in addition to proper protocol selection, methodological design that attempts to follow, detect, and account for contamination was proposed. Its main features include the incorporation of a single negative control that monitors the incorporation of contaminants at every step of the experimental method (Supplementary Fig. 3). Since every step of the protocol prior to the extraction is meant to be executed in a single tube and only by the addition of reagents, it is possible to carry and detect the contaminants introduced throughout the procedure. On the contrary, microbiota study methodologies usually dictate for the incorporation of multiple controls at every step of the procedure (e.g. DNA extraction kit, PCR controls, etc.)18. Although more informative as to which step leads to contamination, it makes data analysis harder since the presence of contamination in the multiple controls cannot by added.
No bioinformatics standard operating procedure is available and what should be done with controls sequencing data is still under debate18. Some research groups tried to use a neutral community model49, additional qPCR data50, amplicon DNA yield, or prevalence algorithms51 to assess the influence of methodological contaminants. The removal of every bacterial OTU found in controls from the samples is often not appropriate as these OTUs might also be naturally present in the samples22. We propose using relative abundance ratio between samples and controls to remove contaminating OTUs. Since controls have much lower richness than extracted lung samples and that the total number of reads (sequencing depth) is distributed across every OTU, the relative abundance of reads for each OTU tend to be much higher in the control than the same OTU in samples. Therefore, if the relative abundance of an OTU is greatly superior in the sample than in the control, it is reasonable to think that the same OTU was also in the sample in a substantial quantity. To ensure that OTUs that were present in very low absolute abundance (e.g., from only 1–2 reads) do not lead to the removal of the highly abundant corresponding OTU in samples, only the OTUs with a ratio of 1000 (relative abundance of sample/relative abundance of sample) were kept. The rest of the OTUs found in controls were completely removed from the related samples, since the influence of contaminating DNA could not be differentiated from the pulmonary microbiota. This method would theoretically tolerate no more than 20 reads (0.1%) before removing the entire OTU from the sample if only one OTU was present in the samples (20,000 reads, 100%). The use of relative abundance helps reduce the absolute abundance bias induced by the divergence in sequencing depth. The OTUs were removed from both tissues at the same time or not at all to avoid adding artificial intraindividual variation. The authors acknowledge that the proposed contaminant management method does not have the in-dept validation of other methods, such as described by Davis et al. with the decontam package51. However, it does not share its limitations regarding the lack of consideration for OTU abundance and need of high number of controls to ensure sensitivity while using prevalence-based detection. Further research focused on the development of statistical methods to detect contaminant OTUs in the cases of lung microbiota is needed. This work is to be a starting point toward methodological standardization and its modular nature makes the bioinformatic contaminant management method proposed here interchangeable once a more robust one is uncovered.
Pearson’s correlation tests were performed on the number of reads per OTU between the samples and their respective controls. Although these values were not normally distributed (Shapiro-Wilk, p  More

Sprouting behaviour of potato tubers
In this study, the stored tubers exhibited clear variety-specific differences in sprouting behaviour, with the Agata variety being the earliest and the Hermes variety the latest to sprout. However, within each variety, there were differences in sprouting time depending on the soil in which the tubers had grown (Suppl. Table 1). These soil-dependent differences in sprouting were greatest in the Fabiola variety (sprouting time ranged from 135 to 169 days after harvest) and least significant in the Hermes variety (158 to 168 days after harvest). By examining the storage stability of tubers broken down by soil type, we found that tubers grown in soil from the field site in Karnabrunn sprouted on average up to 9 days later than the tubers from other soils. The chemical profile of the soil collected in Karnabrunn was similar to that of the other farmland soils, especially those from Kettlasbrunn. The soil collected in Tulln showed reduced K and P contents compared to those of the other farmland soils. As expected, the potting soil differed most significantly from the other soils, i.e., there was no clay, and the exchange capacity was substantially higher than that of the farmland soils (Suppl. Table S2). Statistical analysis revealed that the chemical parameters of the soils did not correlate with the sprouting behaviour of tubers (data not shown).
Sequencing results
The sequencing of 16S rRNA gene amplicons of tubers, sprouts and soil samples yielded 9,158,550 high-quality merged reads, corresponding to an average of 33,920 reads per sample. All sequences were cut to a read length of 360 bp. To ensure sufficient diversity by maintaining an adequate sequencing depth, samples with a low read number were excluded from the analysis. Therefore, the data of four samples were excluded from further analysis (LC_T3_PS; LC_T7_K; H_T6_K: LC_T7_T; H_T6_K; F_T7_K). A detailed explanation for the sample naming can be found in Table 1. Sequencing reads from all 270 samples were clustered into operational taxonomic units (OTUs), which resulted in 11,485 OTUs with an average of 42,537 OTUs per sample.
Table 1 Abbreviation of sample names.
Full size table

The bacterial community of potato tubers during storage
To test whether the bacterial community in potato tubers changes during storage, the 16S rRNA gene amplicon data of the potato tubers of the Agata, Lady Claire and Hermes varieties, cultivated in five different soil types (T2), stored until dormancy break (T6) and until sprouting (T7) were analysed as well as samples of the sprouts (T7_Sprouts) (Fig. 1). Cultivation of the cultivar Fabiola in the soil Kettlasbrunn B did not yield sufficient tubers for analysis throughout the whole storage period, i.e., no data were available for T7 and the sprouts. Therefore, community data for the Fabiola variety were not included in the following statistical analysis.
Figure 1

Overview of potato cultivars, sampling time points and corresponding BBCH stages investigated in this study. After harvesting four tuber varieties (Agata, Fabiola, Lady Claire and Hermes) from five different soil types (T2), tubers were stored at 8–10 °C in darkness. After 2 (T3), 5 (T4) and 10 weeks (T5), tubers were sampled. To consider the individual dormancy break (T6) of each variety, tubers were sampled according to the BBCH scale at stage 03. The same procedure was performed for samples that were taken at sprouting at stage 05 (T7). Additionally, at T7, sprout samples were taken. At each sampling time point, tuber/sprout samples were used for 16S rRNA gene amplicon sequencing. The red circles mark sites on the tubers that show visible signs of sprouting.

Full size image

For the statistical analysis, we selected only OTUs that showed at least 0.01% relative abundance (1425 OTUs) and that were present in at least two of three replicates (1395 rOTUs). We considered them “reproducibly occurring OTUs” (rOTUs)2.
For calculation of alpha diversity values, read numbers were rarefied to 6785 reads in each sample. The permutation ANOVA of rOTU richness (9999 permutations) revealed significant differences in the bacterial community of tubers and sprouts between cultivars, time points and soil types (observed for cultivar (F value = 3.963, P value = 0.0211*), for time point (F value = 4.762, P value = 0.0021*) and for soil type (F value = 3.779, P value = 0.0057**). The same results were obtained with Simpson’s index for the factors cultivar (F value = 4.594, P value = 0.01*), time point (F value = 6.741, P value = 0.0004***) and soil type (F value = 5.123, P value = 0.0008***) (Suppl. Table S3). Overall, the statistical analysis showed that the soil type had the strongest impact on the microbial community in potato tubers, followed by time point and cultivar.
Similarly, the bacterial community composition (beta diversity) in potato tubers was significantly affected by the storage time, soil type and plant variety. The CAP ordination plot indicated a shift in the bacterial community of tubers from harvest to dormancy break and sprouting, which could be proven by PERMANOVA on the Bray–Curtis dissimilarity distance matrix (R2 = 0.113, P value = 0.0004***) (Fig. 2A). Additionally, the bacterial communities in tubers grown in different soil types (R2 = 0.255, P value = 0.0004***) and of different varieties were significantly different from one other (R2 = 0.027, P value = 0.0021**) (Suppl. Table S4). The results for the factor time point (P value = 0.01**), genotype (P value = 0.01**) and soil type (P value = 0.01**) could be confirmed by the multivariate generalized linear model for multivariate abundance data on the Bray–Curtis dissimilarity distance matrix (999 permutations) (Suppl. Table S4).
Figure 2

Constrained analysis of principal coordinates (CAP) of Bray–Curtis dissimilarities. CAP based on the V5–V7 regions of the 16S rRNA gene investigated for (A) tubers of the varieties (Agata, Lady Claire and Hermes) sampled at T2, T6 and T7 and (B) tubers of the varieties (Agata, Lady Claire and Hermes) sampled at T2-7 as well as sprout samples at T7. An overview of potato cultivars, soil types and sampling time points is shown in Fig. 1.

Full size image

The dynamics of bacterial communities in potato tubers during storage
To obtain more in-depth insight into the changes in community composition during storage, we compared the bacterial communities in tubers of the Agata, Lady Claire and Hermes varieties grown in potting soil at six different timepoints: T2 (harvest), T3 (2 weeks after harvesting), T4 (5 weeks after harvesting), T5 (10 weeks after harvesting), T6 (dormancy break), T7 (sprouting) and in the corresponding sprouts (T7_Sprouts) (Fig. 1). We focused here on tubers grown in potting soil, since the cultivation of potatoes in farmland soil did not yield sufficient tubers to be analysed throughout the entire storage period.
Again, we filtered data for OTUs with at least 0.01% relative abundance and “reproducibly occurring OTUs”. After both filtering steps, 1108 rOTUs remained. Before calculating alpha values, read numbers were rarefied to 6761 reads in each sample. The permutation ANOVA of richness and evenness (9999 permutations) did not reveal differences in the alpha diversity of the tuber microbiome between cultivars but between time points (observed species F value = 6.159, P value = 0.0004*** and Simpson’s index F value = 2.216, P value = 0.0263*) (Suppl. Table S3). The bacterial richness of each cultivar declined significantly during the period between harvest (T2) and the dormancy break (T6) (Suppl. Figure S1B). The CAP scaling plot of six different time points during storage and the corresponding sprouts indicated a shift from harvest to sprouting (Fig. 2B). The PERMANOVA on the Bray–Curtis dissimilarity distance (9999 permutations) revealed that the microbiomes of tuber samples differed significantly between the cultivars (R2 = 0.069, P value = 0.0004***) and time points (R2 = 0.221, P value = 0.0001***) (Suppl. Table S4). The test results for cultivar (P value = 0.01**) and time point (P value = 0.01**) were confirmed by the multivariate generalized linear model for multivariate abundance data on the Bray–Curtis dissimilarity distance matrix (999 permutations) (Suppl. Table S4). To identify the bacterial taxa that changed over time during storage, differentially abundant rOTUs were calculated with the random forest function and visualized in Fig. 3 at the genus level. The relative abundance of Staphylococcus sp. increased during the period from harvesting (T2) to dormancy breaking (T6) from approximately 0% to 11% relative abundance. A similar result was visible for Propionibacterium sp. and Acinetobacter sp. The relative abundance of both was approximately 2% in tubers after harvesting (T2) and increased during storage to 8% and 9% in dormancy broken tubers (T6), respectively, whereas the relative abundance of the taxa Iamia sp. and Nocardioides sp. decreased from approximately 10% after harvesting to 4% and 6% in already sprouted tubers (T7), respectively (Suppl. Tables S5 and S6).
Figure 3

Differentially abundant taxa at the genus level. Visualization of differentially abundant taxonomic groups at the genus level of the bacterial community in tubers of three different potato cultivars (Agata, Hermes and Lady Claire) at all sampling time points T2-7 as well as in sprout samples at T7. Differentially abundant taxa were calculated with the function varSelRF48 and visualized in barplot with the function group.abundant.taxa of the R package RAM47.

Full size image

Taxa associated with short, medium or long storage stability
To identify the bacterial taxa in the potato tuber microbiome associated with early or late potato tuber sprouting, sample data were grouped into short, medium and long storage abilities based on the individual storage time (in days from harvest until sprouting) of the samples. The data of all samples, including those of cultivar Fabiola, were considered for the following analysis (Fig. 1). After filtering for the OTUs with at least 0.01% relative abundance and presence in at least two of three replicates, 813 rOTUs were obtained. To identify rOTUs associated with short, medium or long storage stability at the beginning and end of the storage period of the potato tubers, two different analysis steps were performed. In the first step, differentially abundant rOTUs depending on storage stability (short, medium or long) and storage time points (T2, T6 and T7) were calculated with the random forest function (Suppl. Table S7). In a second step, a correlation matrix based on Spearman’s rank correlation was calculated based on the previously described factor storage stability and storage time points to identify positive or negative interactions between rOTUs (Suppl. Table S8). The same rOTUs obtained with random forest analysis, as well as with the correlation matrix, based on Spearman analysis, were filtered and named key OTUs (Suppl. Table 9). Even if both analysis steps do not provide the same evidence, they provide a meaningful indication of which OTUs are related to the factors, storage stability and storage time point. In total, we identified 24 key OTUs. The relative abundance of nine key OTUs was associated with long storage stability, meaning that the OTUs were significantly increased in samples where dormancy break (T6) and sprouting (T7) started late. These key OTUs are members of the orders Flavobacteriales, Cytophagales, Sphingobacteriales, Gaiellales, Corynebacteriales, Caulobacterales, Methylophilales and Solirubrobacterales. Additionally, the relative abundance of eight rOTUs was associated with short storage stability. These rOTUs are members of the orders Enterobacteriales, Pseudomonadales, Myxococcales, Rhizobiales, Bacillales and Burkholderiales. The relative abundance of seven key OTUs was associated with medium storage stability.
Testing the effect of selected bacterial taxa on potato tuber sprouting
In the next step, we tested whether the bacterial taxa that correlated with longer storability can directly affect the sprouting of potato tubers. Therefore, we screened a collection of bacteria isolated from seed potatoes18,20 for isolates that are homologous in the 16S rRNA gene to the key OTUs identified in the statistical analysis. We identified two isolates that were homologous to OTU_14 (Flavobacterium sp.), which correlated with late sprouting. The selected isolates were tested in an in vitro sprouting assay adapted from Hartmann et al.21. Tuber discs treated with cultures of Flavobacterium sp. isolates (AIT1165 and AIT1181) resulted in sprout growth inhibition compared to control discs treated with sterile tryptic soy broth (Fig. 4). Furthermore, we observed significant individual differences in the sprouting behaviour of the tested buds. Differences in sprouting behaviour are only partly genotype-dependent, but these differences might also be due to the natural developmental variability between buds. The apical eye on a tuber usually begins to sprout first, marking the start of the apical dominance stage22. For the sprouting assay performed in this study, we took several buds from one tuber independently of their position.
Figure 4

In vitro potato tuber sprouting assay results. Tuber buds of three different varieties (Lady Claire, Ditta and Agata) were treated with two different isolates of Flavobacterium sp. (AIT1165 and AIT1181). For evaluation, bud growth was assessed according to the first principal growth stages of the BBCH scale. Hereby, the first stage 00 is considered innate or enforced dormancy with no sprouting at all, followed by stages 01 and 02, which represent the beginning of sprouting when sprouts are visible with sizes up to  More

Phylogenetic analysis
A 1444 bp fragment of the 16S rRNA gene was amplified from the P. eucalypticola strain NP-1 T, sequenced and the sequence deposited in GenBank under accession number MN 238,862. A similarity search with this sequence was performed using EzBioCloud. Thirty valid species belonging to P. fluorescens intrageneric group (IG) proposed by Mulet et al.15 exhibited at least 97% similarity with NP-1 T, and these include P. vancouverensis ATCC 700688 T (98.8% similarity), P. moorei DSM12647T (98.8% similarity), P. koreensis Ps9-14 T (98.8% similarity), P. parafulva NBRC16636T (98.5% similarity) and P. reinekei Mt-1 T (98.5% similarity). The similarities with the other 25 species are provided in Supplementary Table S1. A phylogenetic tree based on the 16S rRNA sequence was constructed and is shown in Fig. 1. Strain NP-1 T forms a weakly supported cluster with P. kuykendallii NRRL B-59562 T, but both strains are situated on separate branches. Strain NP-1 T grouped in none known group or subgroup within P. fluorescens lineage, and it clusters of the outer edge of a much larger group containing several Pseudomonas groups/subgroups. However, Pseudomonas species cannot be identified based only on 16S rRNA analysis.
Figure 1

Neighbor-joining phylogenetic tree based on the 16S rRNA gene of Pseudomonas eucalypticola NP-1T and phylogenetically close members of Pseudomonas. The evolutionary distances were computed using the Jukes-Cantor method. The optimal tree with a sum of branch length = 0.23535266 is shown. The percentage of replicate trees in which the associated taxa clustered together in the bootstrap test (1000 replicates) is shown next to the branches. Cellvibrio japonicus Ueda107T was used as outgroup.

Full size image

The MLSA approach based on the concatenated sequences of the partial 16S rRNA, gyrB, rpoB and rpoD genes, has been demonstrated to greatly facilitate the identification of new Pseudomonas strains16. According to the 16S rRNA alignment, 33 species from P. fluorescens IG and one species from P. pertucinogena IG were selected for MLSA. The concatenated sequences of the type strains of each selected species comprised a total of 3813 bp (Supplementary Table S2) and were used for phylogenetic tree construction. The analysis of concatenated gene sequences indicated that strain NP-1 T belongs to the P. fluorescens lineage, and this finding was supported by a bootstrap value of 91% (Fig. 2).However, NP-1 T still cannot be determined which group belongs to17.
Figure 2

Neighbor-joining phylogenetic tree based on concatenated 16S rRNA, gyrB, rpoB and rpoD gene partial of Pseudomonas eucalypticola NP-1T and the type strains of other Pseudomonas species. The evolutionary distances were computed using the Jukes-Cantor method. The evolutionary distances were computed using the Jukes-Cantor method e optimal tree with the sum of branch length = 1.37677586 is shown. The percentage of replicate trees in which the associated taxa clustered together in the bootstrap test (1000 replicates) is shown next to the branches.

Full size image

For further identification of NP-1 T, a phylogenomic tree inferred with GBDP was constructed by using Type (Strain) Genome Server (TYGS)18, and all reference type strains and their genome sources are listed in Supplementary Table S3. The result showed the presence of an independent branch supported by a bootstrap value of 88% that can be differentiated from the other Pseudomonas species type strains (Fig. 3) and revealed that NP-1 T clustered with P. coleopterorum LMG 28558 T and P. rhizosphaerae LMG 21640 T which affiliated with P. fluorescens IG, but does not belong to any group. Strain NP-1 T was not be affiliated with any previously described Pseudomonas species and can thus be considered to represent a novel species. Based on above-described the results, P. coleopterorum, P. rhizosphaerae, P. graminis and P. lutea were selected for further analysis with NP-1 T.
Figure 3

Phylogenomic tree of strain NP-1T and related type strains of the genus Pseudomonas available on the TYGS database. The tree inferred with FastME 2.1.6.1 based on GBDP distances calculated from the genome sequences. The branch lengths are scaled in terms of the GBDP distance formula d5. The numbers above the branches show the GBDP pseudo-bootstrap support values  > 60% from 100 replications, and the average branch support is 94.6%.

Full size image

General taxonomic genome feature
The draft genome assembly of strain NP-1 T contains 6,401,699 bp. The genome of NP-1 T, which consists of one chromosome and one plasmid, has been deposited in GenBank under the accession numbers CP056030 and CP056031, respectively. The genome has a G + C content of 63.96 mol%, as determined from the complete genome sequence, and 83.45% of the genome is coding and consists of 5,788 genes. The similarity of the genome of P. eucalypticola NP-1 T to other publicly available genomes of closely related Pseudomonas species was determined using ANI, digital DDH and G + C mol %5,6,7,8,9. Each of these comparisons yielded different ANIm and ANIb values, but the highest ANIb and ANIm values of 78.7 and 86.5 were obtained for NP-1 T and P. rhizosphaerae LMG 21640 T. The similarity between P. coleopterorum LMG 28558 T and NP-1 T was higher than that between P. graminis DSM 11363 T and P. lutea LMG 21974 T (Table 1). All ANIb and ANIm values obtained from the comparisons of NP-1 T with the other tested species were below 95%, which confirmed that strain NP-1 T belongs to an independent species. The TETRA frequencies between NP-1 T and the other tested type strains were lower than 0.99, which is the recommended cutoff value for species (Table 2). The digital DNA-DNA hybridization (dDDH) comparison with the draft genome of the type strain NP-1 T yielded low percentages ( More

1.
Seilacher, A. Begriff und Bedeutung der Fossil-Lagerstätten. N. Jahrb. Geol. Paläont. Monatsh. 1970, 34–39 (1970).
Google Scholar 
2.
Zhuravlev, A. & Riding, R. The Ecology of the Cambrian Radiation, ix + 525 pp. (Columbia University Press, New York, 2000).

3.
Erwin, D. H. & Valentine, J. W. The Cambrian Explosion: The Construction of Animal Biodiversity. X + 406 pp. (Roberts and Company Publishers Inc., Greenwood Village, CO, 2013).

4.
Landing, E. et al. Early evolution of colonial animals (Ediacaran Evolutionary Revolution–Cambrian Evolutionary Radiation–Great Ordovician Diversification Interval). Earth-Sci. Rev. 178, 105–135 (2018).
ADS  Article  Google Scholar 

5.
Briggs, D. E. G., Erwin, D. H. & Collier, F. J. The Fossils of the Burgess Shale. xvii + 238 pp. (Smithsonian Institution Press, Washington DC, 1994).

6.
Hou X. G., Siveter, D. J., Siveter, D. J., Aldridge, R. J., Cong, P. Y., Gabbott, S. E., Ma, X. Y., Purnell. M. A. & Williams, M. The Cambrian Fossils of Chengjiang, China: The Flowering of Early Animal Life. 2nd ed. 328 pp. (Wiley-Blackwell, Oxford, 2017).

7.
Peel, J. S. & Ineson, J. R. The Sirius Passet Lagerstätte (early Cambrian) of  North Greenland. Palaeontographica Canadiana 31, 109–118 (2011).
Google Scholar 

8.
Paterson, J. R. et al. The Emu Bay Shale Konservat-Lagerstätte: a view of Cambrian life from East Gondwana. J. Geol. Soc. Lond. 173, 1–11 (2016).
Article  Google Scholar 

9.
Eriksson, M. E. & Waloszek, D. Half-a-billion-year-old microscopic treasures—the Cambrian ‘Orsten’ fossils of Sweden. Geol. Today 32, 115–120 (2016).
Article  Google Scholar 

10.
Fu, D. J., Tong, G. H., Dai, T., Liu, W., Yang, Y. N., Zhang, Y., Cui, L. H., Li, L. Y., Yun, H., Wu, Y., Sun, A., Liu, C., Pei, W. R., Gaines, R. R. & Zhang X. L. The Qingjiang biota—A Burgess Shale–type fossil Lagerstätte from the early Cambrian of South China. Science 363, 1338–1342.

11.
Hu, S. X. et al. Biodiversity and taphonomy of the Early Cambrian Guanshan Biota, eastern Yunnan. Sci. China Earth Sci. 53, 1765–1773 (2010).
ADS  Article  Google Scholar 

12
Zhao, Y. L. et al. Kaili biota: a taphonomic window on diversification of metazoans from the basal Middle Cambrian, Guizhou China. Acta Geol. Sin. Engl. Ed. 79, 751–765 (2005).
Google Scholar 

13.
Robison, R. A., Babcock, L. E. & Gunther, V. G. Exceptional Cambrian fossils from Utah: a window into the age of trilobites. Utah Geol. Surv. Misc. Publ. 15(1), 97 (2015).
Google Scholar 

14.
Lerosey-Aubril, R. et al. The Weeks Formation Konservat-Lagerstätte and the evolutionary transition of Cambrian marine life. J. Geol. Soc. Lond. 175, 705–715 (2018).
Article  Google Scholar 

15.
Kimmig, J., Strotz, L. C., Kimmig, S. R., Egenhoff, S. O. & Lieberman, B. S. The Spence Shale Lagerstätte: an important window into Cambrian biodiversity. J. Geol. Soc. Lond. 176, 609–619 (2019).
Article  Google Scholar 

16.
Conway Morris, S. Cambrian Lagerstätten: their distribution and significance. Philos. Trans. R. Soc. Lond. Ser. B. Biol. Sci. 311, 49–65 (1985).
ADS  Google Scholar 

17.
García-Bellido, D. & Conway Morris, S. New fossil worms from the Lower Cambrian of the Kinzers Formation, Pennsylvania, with some comments on Burgess Shale−type preservation. J. Paleont. 73, 394–402 (1999).

18.
Müller, K. J., Walossek, D. & Zakharov, A. ‘Orsten’ type phosphatized soft-integument preservation and a new record from the Middle Cambrian Kuonamka Formation in Siberia. N. Jahrb. Geol. Paläont. Abh. 191, 101–118 (1995).
Article  Google Scholar 

19.
Martin, E. et al. The Lower Ordovician Fezouata Konservat-Lagerstätte from Morocco: age, environment and evolutionary perspectives. Gondwana Res. 34, 274–283 (2016).
ADS  Article  Google Scholar 

20.
Van Roy, P. et al. Ordovician faunas of Burgess Shale type. Nature 465, 215–218 (2010).
ADS  Article  Google Scholar 

21.
Van Roy, P., Briggs, D. E. G. & Gaines, R. R. The Fezouata fossils of Morocco; an extraordinary record of marine life in the Early Ordovician. J. Geol. Soc. Lond. 172, 541–549 (2015).
Article  Google Scholar 

22.
Gutierrez-Marco, J. C., Pereira, S., García-Bellido, D. C. & Rábano, I. Ordovician trilobites from the Tafilalt Lagerstätte: new data and reappraisal of the Bou Nemrou assemblage. Geol. Soc. Lond., Spec. Publ. https://doi.org/10.1144/SP485-2018-126 (2019).

23.
Ortega-Hernández, J. et al. A xandarellid artiopodan from Morocco – a middle Cambrian link between soft-bodied euarthropod communities in North Africa and South China. Nat. Sci. Rep. 7, 42616. https://doi.org/10.1038/srep42616 (2017).
ADS  CAS  Article  Google Scholar 

24.
Geyer, G. Late Precambrian to early Middle Cambrian lithostratigraphy of southern Morocco. Beringeria 1, 115–143 (1989).
ADS  Google Scholar 

25.
Geyer, G. Revised Lower to lower Middle Cambrian biostratigraphy of Morocco. Newsl. Stratigraphy 22, 53–70 (1990).
Article  Google Scholar 

26.
Geyer, G. & Landing, E. Latest Ediacaran and Cambrian of the Moroccan Atlas regions. Beringeria Spec. Issue 6, 7–46 (2006).
Google Scholar 

27.
Geyer, G. The earliest known West Gondwanan trilobites from the Anti-Atlas of Morocco, with a revision of the Family Bigotinidae Hupé, 1953. Fossils Strata 64, 55–153 (2019).
Article  Google Scholar 

28.
Geyer, G. A critical evaluation of the Resserops clade (Trilobita: Despujolsiidae, early Cambrian) with remarks on related redlichiacean families. Freib. Forsch. H. C 558, 1–107 (2020).
Google Scholar 

29.
Geyer, G. A new obolellid brachiopod from the Lower Cambrian of Morocco. J. Paleont. 68, 995–1002 (1994).
Article  Google Scholar 

30.
Geyer, G. An enigmatic bilateral fossil from the Lower Cambrian of Morocco. J. Paleont. 68, 710–716 (1994).
Article  Google Scholar 

31.
Geyer, G. A comprehensive Cambrian correlation chart. Episodes 42, 321–332. https://doi.org/10.18814/epiiugs/2019/019026 (2019).
Article  Google Scholar 

32.
Chen, L. et al. Paleoceanographic indicators for early Cambrian black shales from the Yangtze Platform, South China: Evidence from biomarkers and carbon isotopes. Acta Geol. Sinica 86, 1143–1153 (2012).
CAS  Article  Google Scholar 

33.
Steiner, M., Zhu, M. Y., Weber, B. & Geyer, G. The Lower Cambrian of eastern Yunnan: trilobite-based biostratigraphy and related faunas. Acta Palaeont. Sinica 40, Suppl., 63–79 (2001).

34.
Landing, E. Time-specific black mudstones and global hyperwarming on the Cambrian–Ordovician slope and shelf of the Laurentia palaeocontinent. Palaeogeogr. Palaeoclimatol. Palaeoecol. 367–368, 256–272 (2012).
Article  Google Scholar 

35
Maloof, A. C., Schrag, D. P., Crowley, J. L. & Bowring, S. A. An expanded record of Early Cambrian carbon cycling from the Anti-Atlas Margin Morocco. Can. J. Earth Sci. 42, 2195–2216 (2005).
Google Scholar 

36.
Brasier, M. D., Rozanov, A, Yu., Zhuravlev, A. Yu., Corfield, R. M. & Derry, L. A. A carbon isotope reference scale for the Lower Cambrian succession in Siberia: report of IGCP Project 303. Geol. Mag. 131, 767–783 (1994).
ADS  CAS  Article  Google Scholar 

37.
Zhu, M. Y., Babcock, L. E. & Peng, S. C. Advances in Cambrian stratigraphy and paleontology: integrating correlation techniques, paleobiology, taphonomy and paleoenvironmental reconstruction. Palaeoworld 15, 217–222 (2006).
Article  Google Scholar  More

Study area
The study was conducted across the Loess Plateau (33°43′–41°16′N, 100°54′–114°33′E) (Fig. 1a), which represents approximately 6.5% of the total area of China6. The study area is dominated by temperate, arid, and semiarid continental monsoon climates. The annual evaporation is 1400–2000 mm, and the annual temperature ranges from 3.6 °C in the northwest to 14.3 °C in the southeast on the Loess Plateau7, while the annual precipitation ranges from 150 to 800 mm, where 55–78% of the precipitation falls from June to September7. The annual solar radiation ranges from 5.0 × 109 to 6.7 × 109 J m−2. The vegetation zones are forest, forest-steppe, typical-steppe, desert-steppe, and steppe-desert zones8 from southeast to northwest.
Figure 1

Locations of the Loess Plateau region in China (a) and the sampling sites (b); image data processed by ArcGIS 10.5 http://developers.arcgis.com.

Full size image

Field sampling
According to the different climate zones and vegetation types, five classic sampling sites were selected (Fig. 1b) on the Loess Plateau, which were Yangling, Changwu, Fuxian, Ansai, and Shenmu from south to north. Drilling equipment (assembled by Xi’an Qinyan Drilling Co. Ltd, China) was used to collect soil samples from soil surface down to bedrock. At each sampling site, disturbed soil samples were collected to determine the SAP and SAK concentrations, pH, soil particle composition, and soil organic matter contents. In addition, disturbed soil samples were collected from the middle of the soil column at 1-m intervals (i.e., 0.5 m, 1.5 m, 2.5 m, 3.5 m, etc.). The drilling and sampling work was carried out from April 28 to June 28, 2016. The total numbers of disturbed soil samples collected from Yangling, Changwu, Fuxian, Ansai, and Shenmu were 103, 205, 181, 161, and 58, respectively, and the corresponding soil drilling depths were 103.5 m, 204.5 m, 187.5 m, 161.6 m, and 56.6 m, respectively.
Laboratory analyses
Undisturbed soil samples were air-dried, separated, and passed through 0.25-mm or 2-mm sieves. SAP and SAK were extracted with ammonium lactate solution and detected by spectrophotometry and flame photometry. Soil total nitrogen (STN) concentrations were determined by the Kjeldahl digestion procedure9. Soil total phosphorus (STP) concentrations were determined by molybdenum antimony blue colorimetry10. The soil organic carbon (SOC) contents were analyzed by dichromate oxidation method11. The soil particle composition was determined by laser diffraction (Mastersizer 2000, Malvern Instruments, Malvern, UK)12. According to the mixture of soil and water mass ratio of 1:1, the pH value was determined with a pH meter equipped with a calibrated combined glass electrode. The soil water content (SWC) was determined by the mass loss after drying to constant mass in an oven at 105 °C13. The calcium carbonate content was determined by the acid-neutralization method14.
Geostatistical analysis
The geostatistical analysis was chosen to determine the spatial structure of the spatially dependent soil properties15, where a semivariogram was employed to quantify the spatial patterns of the variables. The equation for the semivariogram is16:

$$ {text{R}}left( {text{h}} right) , = frac{1}{{2{text{N}}left( {text{h}} right)}}mathop sum limits_{{{text{i}} = 1}}^{{{text{N}}left( {text{h}} right)}} left[ {{text{Z}}left( {{text{x}}_{{text{i}}} } right){-}{text{Z}}left( {{text{x}}_{{{text{i}} + {text{h}}}} } right)} right]^{{2}} , $$
(1)

where for each site i, N(h) is the number of pairs separated by h, and Z(xi) is the value at location xi and Z(xi+h) for xi+h. There are four semivariogram models (spherical, exponential, linear, and Gaussian), which can be employed to describe the semivariogram, and the best fitting model is selected according to the smallest residual sum of squares (RSS) and the largest coefficient of determination (R2). The equation of each semivariogram model is16:
Exponential model:

$$ {text{R}}left( {text{h}} right) = {text{C}}_{0} + {text{C}}left[ {({1}{-}{text{exp}}( – {text{h}}/{text{A}}_{0} )} right] $$
(2)

Linear Model:

$$ {text{R}}left( {text{h}} right) = {text{C}}_{0} + left[ {{text{h}}left( {{text{C}}/{text{A}}_{0} } right)} right] $$
(3)

Spherical Model:

$$ {text{R}}left( {text{h}} right) = {text{C}}0 + {text{C}}left[ {{1}.{5}left( {{text{h}}/{text{A}}_{0} } right) – 0.{5}left( {{text{h}}/{text{A}}_{0} } right)^{{3}} } right] ;;;;;;;;; {text{h}} le {text{A}}0 $$
(4)

$$ {text{R}}left( {text{h}} right) = {text{C}}_{0} + {text{C}};;;;;;;;;;{text{h}} ge {text{A}}0 $$
(5)

Gaussian Model:

$$ {text{R}}left( {text{h}} right) = {text{C}}_{0} + {text{C}}left[ {{1} – {text{exp}}left( { – {text{h}}^{{2}} /{text{A}}_{0}^{{2}} } right)} right] $$
(6)

where C0 indicates the nugget value, which is the short-range structure that occurs at distances smaller than the sampling interval, microheterogeneity, and experimental error; C0 + C is the sill indicating the random and structural variation, and; A0 is the range indicating the spatial correlation at different distances.
Statistical analysis
Descriptive statistical analyses (maximum, minimum, average, and coefficient of variation), Pearson’s correlation analysis, and linear regression analysis was performed with SPSS 16.0 (IBM SPSS, Chicago, IL, USA). Geostatistical analysis was performed with GS + software (version 7.05). More

1.
Cody, E. M. & Dixon, B. P. Hemolytic uremic syndrome. Pediatr. Clin. North Am. 66, 235–246. https://doi.org/10.1016/j.pcl.2018.09.011 (2019).
Article  PubMed  Google Scholar 
2.
Chiang, Y. N., Penades, J. R. & Chen, J. Genetic transduction by phages and chromosomal islands: The new and noncanonical. PLoS Pathog 15, e1007878. https://doi.org/10.1371/journal.ppat.1007878 (2019).
CAS  Article  PubMed  PubMed Central  Google Scholar 

3.
Penades, J. R. & Christie, G. E. The phage-inducible chromosomal islands: A family of highly evolved molecular parasites. Annu. Rev. Virol. 2, 181–201. https://doi.org/10.1146/annurev-virology-031413-085446 (2015).
CAS  Article  PubMed  Google Scholar 

4.
Valilis, E., Ramsey, A., Sidiq, S. & DuPont, H. L. Non-O157 Shiga toxin-producing Escherichia coli-A poorly appreciated enteric pathogen: Systematic review. Int. J. Infect Dis. 76, 82–87. https://doi.org/10.1016/j.ijid.2018.09.002 (2018).
CAS  Article  PubMed  Google Scholar 

5.
Murinda, S. E. et al. Shiga toxin-producing Escherichia coli in mastitis: An international perspective. Foodborne Pathog. Dis. 16, 229–243. https://doi.org/10.1089/fpd.2018.2491 (2019).
Article  PubMed  Google Scholar 

6.
Feiner, R. et al. A new perspective on lysogeny: prophages as active regulatory switches of bacteria. Nat. Rev. Microbiol. 13, 641–650. https://doi.org/10.1038/nrmicro3527 (2015).
CAS  Article  PubMed  Google Scholar 

7.
Bikard, D. & Marraffini, L. A. Innate and adaptive immunity in bacteria: mechanisms of programmed genetic variation to fight bacteriophages. Curr. Opin. Immunol. 24, 15–20. https://doi.org/10.1016/j.coi.2011.10.005 (2012).
CAS  Article  PubMed  Google Scholar 

8.
Hoskisson, P. A. & Smith, M. C. Hypervariation and phase variation in the bacteriophage “resistome”. Curr. Opin. Microbiol. 10, 396–400. https://doi.org/10.1016/j.mib.2007.04.003 (2007).
CAS  Article  PubMed  Google Scholar 

9.
De Ste Croix, M. et al. Phase-variable methylation and epigenetic regulation by type I restriction-modification systems. FEMS Microbiol. Rev. 41, S3–S15. https://doi.org/10.1093/femsre/fux025 (2017).
Article  PubMed  Google Scholar 

10.
Heredia, N. & Garcia, S. Animals as sources of food-borne pathogens: A review. Anim. Nutr. 4, 250–255. https://doi.org/10.1016/j.aninu.2018.04.006 (2018).
Article  PubMed  PubMed Central  Google Scholar 

11.
Fatima, R. & Aziz, M. in StatPearls (2019).

12.
Mellor, G. E. et al. National Survey of Shiga Toxin-Producing Escherichia coli Serotypes O26, O45, O103, O111, O121, O145, and O157 in Australian Beef Cattle Feces. J. Food Prot. 79, 1868–1874. https://doi.org/10.4315/0362-028X.JFP-15-507 (2016).
Article  PubMed  Google Scholar 

13.
Kampmeier, S., Berger, M., Mellmann, A., Karch, H. & Berger, P. The 2011 German enterohemorrhagic Escherichia coli O104:H4 outbreak-the danger is still out there. Curr. Top. Microbiol. Immunol. 416, 117–148. https://doi.org/10.1007/82_2018_107 (2018).
CAS  Article  PubMed  Google Scholar 

14.
Lee, M. S. & Tesh, V. L. Roles of shiga toxins in immunopathology. Toxins (Basel) https://doi.org/10.3390/toxins11040212 (2019).
Article  PubMed Central  Google Scholar 

15.
Schmidt, H. Shiga-toxin-converting bacteriophages. Res. Microbiol. 152, 687–695. https://doi.org/10.1016/s0923-2508(01)01249-9 (2001).
CAS  Article  PubMed  Google Scholar 

16.
Herold, S., Karch, H. & Schmidt, H. Shiga toxin-encoding bacteriophages–genomes in motion. Int. J. Med. Microbiol. 294, 115–121. https://doi.org/10.1016/j.ijmm.2004.06.023 (2004).
CAS  Article  PubMed  Google Scholar 

17.
Chakraborty, D., Clark, E., Mauro, S. A. & Koudelka, G. B. Molecular mechanisms governing “hair-trigger” induction of shiga toxin-encoding prophages. Viruses https://doi.org/10.3390/v10050228 (2018).
Article  PubMed  PubMed Central  Google Scholar 

18.
Bloch, S. et al. Inhibition of Shiga toxin-converting bacteriophage development by novel antioxidant compounds. J. Enzyme Inhib. Med. Chem. 33, 639–650. https://doi.org/10.1080/14756366.2018.1444610 (2018).
CAS  Article  PubMed  PubMed Central  Google Scholar 

19.
Fang, Y., Mercer, R. G., McMullen, L. M. & Ganzle, M. G. Induction of Shiga Toxin-Encoding Prophage by Abiotic Environmental Stress in Food. Appl. Environ. Microbiol. https://doi.org/10.1128/AEM.01378-17 (2017).
Article  PubMed  PubMed Central  Google Scholar 

20.
Smith, D. L. et al. Comparative genomics of Shiga toxin encoding bacteriophages. BMC Genomics 13, 311. https://doi.org/10.1186/1471-2164-13-311 (2012).
CAS  Article  PubMed  PubMed Central  Google Scholar 

21.
Kakoullis, L., Papachristodoulou, E., Chra, P. & Panos, G. Shiga toxin-induced haemolytic uraemic syndrome and the role of antibiotics: A global overview. J. Infect. 79, 75–94. https://doi.org/10.1016/j.jinf.2019.05.018 (2019).
Article  PubMed  Google Scholar 

22.
Kavanagh, D., Raman, S. & Sheerin, N. S. Management of hemolytic uremic syndrome. F1000Prime Rep 6, 119. https://doi.org/10.12703/P6-119 (2014).
Article  PubMed  PubMed Central  Google Scholar 

23.
James, C. E. et al. Lytic and lysogenic infection of diverse Escherichia coli and Shigella strains with a verocytotoxigenic bacteriophage. Appl. Environ. Microbiol. 67, 4335–4337. https://doi.org/10.1128/aem.67.9.4335-4337.2001 (2001).
CAS  Article  PubMed  PubMed Central  Google Scholar 

24.
Smith, D. L. et al. Multilocus characterization scheme for shiga toxin-encoding bacteriophages. Appl. Environ. Microbiol. 73, 8032–8040. https://doi.org/10.1128/AEM.01278-07 (2007).
CAS  Article  PubMed  PubMed Central  Google Scholar 

25.
Eichhorn, I. et al. Lysogenic conversion of atypical enteropathogenic Escherichia coli (aEPEC) from human, murine, and bovine origin with bacteriophage Phi3538 Deltastx2::cat proves their enterohemorrhagic E. coli (EHEC) progeny. Int. J. Med. Microbiol. 308, 890–898. https://doi.org/10.1016/j.ijmm.2018.06.005 (2018).
Article  PubMed  Google Scholar 

26.
Khalil, R. K., Skinner, C., Patfield, S. & He, X. Phage-mediated Shiga toxin (Stx) horizontal gene transfer and expression in non-Shiga toxigenic Enterobacter and Escherichia coli strains. Pathog. Dis. https://doi.org/10.1093/femspd/ftw037 (2016).
Article  PubMed  Google Scholar 

27.
Smith, D. L. et al. Short-tailed Stx phages exploit the conserved YaeT protein to disseminate Shiga toxin genes among enterobacteria. J. Bacteriol. 189, 7223–7233. https://doi.org/10.1128/JB.00824-07 (2007).
CAS  Article  PubMed  PubMed Central  Google Scholar 

28.
Botos, I., Noinaj, N. & Buchanan, S. K. Insertion of proteins and lipopolysaccharide into the bacterial outer membrane. Philos. Trans. R. Soc. Lond. Ser. B Biol. Sci. https://doi.org/10.1098/rstb.2016.0224 (2017).
Article  Google Scholar 

29.
Knirel, Y. A. et al. Variations in O-antigen biosynthesis and O-acetylation associated with altered phage sensitivity in Escherichia coli 4s. J. Bacteriol. 197, 905–912. https://doi.org/10.1128/JB.02398-14 (2015).
CAS  Article  PubMed  PubMed Central  Google Scholar 

30.
Golomidova, A. K. et al. Branched lateral tail fiber organization in T5-like bacteriophages DT57C and DT571/2 is revealed by genetic and functional analysis. Viruses https://doi.org/10.3390/v8010026 (2016).
Article  PubMed  PubMed Central  Google Scholar 

31.
Kulikov, E. E., Golomidova, A. K., Prokhorov, N. S., Ivanov, P. A. & Letarov, A. V. High-throughput LPS profiling as a tool for revealing of bacteriophage infection strategies. Sci. Rep. 9, 2958. https://doi.org/10.1038/s41598-019-39590-8 (2019).
ADS  CAS  Article  PubMed  PubMed Central  Google Scholar 

32.
van der Ley, P., de Graaff, P. & Tommassen, J. Shielding of Escherichia coli outer membrane proteins as receptors for bacteriophages and colicins by O-antigenic chains of lipopolysaccharide. J. Bacteriol. 168, 449–451. https://doi.org/10.1128/jb.168.1.449-451.1986 (1986).
Article  PubMed  PubMed Central  Google Scholar 

33.
Kunisaki, H. & Tanji, Y. Intercrossing of phage genomes in a phage cocktail and stable coexistence with Escherichia coli O157:H7 in anaerobic continuous culture. Appl. Microbiol. Biotechnol. 85, 1533–1540. https://doi.org/10.1007/s00253-009-2230-2 (2010).
CAS  Article  PubMed  Google Scholar 

34.
Golomidova, A. K., Kulikov, E. E., Babenko, V. V., Kostryukova, E. S. & Letarov, A. V. Complete genome sequence of bacteriophage St11Ph5, which Infects uropathogenic Escherichia coli strain up11. Genome Announcements https://doi.org/10.1128/genomeA.01371-17 (2018).
Article  PubMed  PubMed Central  Google Scholar 

35.
Golomidova, A. K., Naumenko, O. I., Senchenkova, S. N., Knirel, Y. A. & Letarov, A. V. The O-polysaccharide of Escherichia coli F5, which is structurally related to that of E. coli O28ab, provides cells only weak protection against bacteriophage attack. Arch. Virol. 164, 2783–2787. https://doi.org/10.1007/s00705-019-04371-1 (2019).
CAS  Article  PubMed  Google Scholar 

36.
Knirel, Y. A. et al. Structure and gene cluster of the O antigen of Escherichia coli F17, a candidate for a new O-serogroup. Int. J. Biol. Macromol. 124, 389–395. https://doi.org/10.1016/j.ijbiomac.2018.11.149 (2019).
CAS  Article  PubMed  Google Scholar 

37.
Zdorovenko, E. L. et al. Structure of the O-polysaccharide of Escherichia coli O87. Carbohyd. Res. 412, 15–18. https://doi.org/10.1016/j.carres.2015.04.014 (2015).
CAS  Article  Google Scholar 

38.
Zdorovenko, E. L. et al. Corrigendum to “Structure of the O-polysaccharide of Escherichia coli O87” [Carbohydr. Res. 412 (2015) 15–18]. Carbohydrate Res. 464, 1. https://doi.org/10.1016/j.carres.2018.04.013 (2018).

39.
Zdorovenko, E. L. et al. O-Antigens of Escherichia coli strains O81 and HS3–104 are structurally and genetically related, except O-Antigen glucosylation in E. coli HS3–104. Biochemistry 83, 534–541. https://doi.org/10.1134/S0006297918050061 (2018).
CAS  Article  PubMed  Google Scholar 

40.
Kulikov, E. E. et al. Genomic sequencing and biological characteristics of a novel Escherichia coli bacteriophage 9g, a putative representative of a new Siphoviridae genus. Viruses 6, 5077–5092. https://doi.org/10.3390/v6125077 (2014).
CAS  Article  PubMed  PubMed Central  Google Scholar 

41.
Golomidova, A. K. et al. Escherichia coli bacteriophage Gostya9, representing a new species within the genus T5virus. Adv. Virol. 164, 879–884. https://doi.org/10.1007/s00705-018-4113-2 (2019).
CAS  Article  Google Scholar 

42.
Kulikov, E. et al. Isolation and characterization of a novel indigenous intestinal N4-related coliphage vB_EcoP_G7C. Virology 426, 93–99. https://doi.org/10.1016/j.virol.2012.01.027 (2012).
CAS  Article  PubMed  Google Scholar 

43.
Prokhorov, N. S. et al. Function of bacteriophage G7C esterase tailspike in host cell adsorption. Mol. Microbiol. 105, 385–398. https://doi.org/10.1111/mmi.13710 (2017).
CAS  Article  PubMed  Google Scholar 

44.
Zdorovenko, E. L. et al. Corrigendum to “Structure of the O-polysaccharide of Escherichia coli O87”. Carbohydrate Res. 412, 15–18. https://doi.org/10.1016/j.carres.2018.04.013 (2015).
CAS  Article  Google Scholar 

45.
Sambrook, J., Fritsch, E. F. & Maniatis, T. Molecular Cloning: A Laboratory Manual (Cold Spring Harbor Laboratory Press, New York, 1989).
Google Scholar 

46.
Samson, J. E., Magadan, A. H., Sabri, M. & Moineau, S. Revenge of the phages: Defeating bacterial defences. Nat. Rev. Microbiol. 11, 675–687. https://doi.org/10.1038/nrmicro3096 (2013).
CAS  Article  PubMed  Google Scholar 

47.
Pawlak, A. et al. Salmonella O48 serum resistance is connected with the elongation of the lipopolysaccharide O-Antigen containing sialic acid. Int. J. Mol. Sci. https://doi.org/10.3390/ijms18102022 (2017).
Article  PubMed  PubMed Central  Google Scholar 

48.
Coggon, C. F. et al. A novel method of serum resistance by Escherichia coli that causes urosepsis. mBio https://doi.org/10.1128/mBio.00920-18 (2018).
Article  PubMed  PubMed Central  Google Scholar 

49.
Kintz, E. et al. Salmonella enterica Serovar Typhi lipopolysaccharide O-antigen modification impact on serum resistance and antibody recognition. Infect. Immunity https://doi.org/10.1128/IAI.01021-16 (2017).
Article  Google Scholar 

50.
Datsenko, K. A. & Wanner, B. L. One-step inactivation of chromosomal genes in Escherichia coli K-12 using PCR products. Proc. Natl. Acad. Sci. U.S.A. 97, 6640–6645. https://doi.org/10.1073/pnas.120163297 (2000).
ADS  CAS  Article  PubMed  PubMed Central  Google Scholar  More

The experiment was conducted at the INRAE dairy research farm (La boire, Marcenat, https://doi.org/10.15454/1.5572318050509348E12). Procedures were evaluated and approved by the French Ministry of Education and Research (APAFIS #4062-2015043014541577 v5), and carried out in accordance with French and European guidelines and regulations for animal experimentation. Information provided in the manuscript complies with the essential recommendations for reporting of the ARRIVE guidelines.
Animals, diets and experimental design
Eighteen female Holstein (n = 12), Montbéliarde (n = 4) and Holstein x Montbéliarde (n = 2) calves (42.07 ± 3.85 kg birth weight) were enrolled in the study at birth. Calves were kept with their dam for a few hours but systematically received 2 L of warm colostrum of good quality (≥ 50 g IgG per L) that is conserved at − 20 °C until use. Calves were individually housed for the first week of life and bottle fed 3 L of milk twice daily (0700 h and 1800 h). After the first week, calves were group housed according to treatment with ad libitum access to water and hay. Calves were fed up to 8 L of milk per day through the use of an automated milking system (De Laval, Sursee, Switzerland). Similarly, calves had access to calf starter (STARTIVO, Centraliment, Aurillac, France) from four weeks of age with a maximum daily allowance of 2 kg in the pre-weaning period. In the immediate post-weaning period, calves had access to 2 kg of GENIE ELEVAGE (Centraliment, Aurillac, France). Chemical composition of dietary ingredients is presented in Table S1.
Calves were randomly assigned at birth to either a treatment (3-NOP, 3 mg 3-NOP/kg BW, n = 10 up to week 23, a heifer was removed from the herd due to infertility as a result of being born a twin, i.e., a free-martin, n = 9 at week 60) or control (CONT, placebo premix containing SiO2 and propylene glycol only, n = 8, nine calves were recruited but one calf died early in the study) group, such that breed distribution and birthweight were balanced across groups. The 3-NOP supplement contained 10% 3-NOP diluted in propylene glycol and adsorbed on SiO2, such that 30 mg of the supplement was fed per kilogram of body weight to achieve the above target dose of 3-NOP. The 3-NOP and placebo were mixed with water (300 mg/mL water) and administered daily via an oral gavage approximately 2 h after feeding. Calves were treated daily from the day of birth, following consumption of colostrum, until 14 weeks of age. All calves were weaned at week 11 using the step-down method over two weeks. After weaning, all calves were group housed in a single pen to replicate normal production practices.
Calves were weighed weekly. Daily individual milk and concentrate intakes prior to weaning were recorded using automated feeders. Total group intake of hay and concentrate, and all refusals in the post-weaning period were recorded twice weekly.
Sampling
All calves were sampled for rumen fluid and faecal content at 1, 4, 11, 14, 23 and 60 weeks of life. Sampling at week 11 was conducted immediately prior to weaning and sampling at week 14 was conducted just prior to cessation of the treatment. Samples of rumen liquid were obtained via oesophageal tubing at least 2 h after feeding. Aliquots (1 mL) of rumen liquid were immediately frozen in liquid nitrogen and stored at − 80 °C until DNA extraction. Additional rumen liquid aliquots were taken for analysis of volatile fatty acids and ammonia as previously described25,26. At week 60, 3 mL of ruminal fluid was added to 3 mL of methyl green formalin saline (MFS) solution (35 mL/L formaldehyde, 0.14 mM NaCl, and 0.92 mM methyl green) and stored in the dark at room temperature until protozoa were counted. At each sampling period, calves were rectally finger-stimulated with sterile-gloved hand to facilitate the collection of a faecal sample, which was immediately frozen in liquid nitrogen and stored at − 80 °C until DNA extraction. Blood samples were taken via jugular venepuncture into a heparin tube at week 11, 14 and 23 for metabolic analysis. Blood was immediately centrifuged at 1500×g for 10 min at 4 °C. Plasma was stored at − 80 °C until analysis.
Methane measurements
Methane emissions were recorded using the GreenFeed system (C-Lock Inc., Rapid City, South Dakota, USA) during two time periods. Firstly, from weaning (week 11) until week 23, one GreenFeed system was programmed using C-Lock Inc. software to deliver a maximum of six rotations of a feed dispensing cup, delivering ~ 6 g of pellet concentrate GENIE ELEVAGE (as fed) per rotation, with intervals of 30 s between each rotation, so that 36 g of pellet was delivered during each visit. During the second phase of measurement when heifers were 57 to 60 weeks of age, two GreenFeed systems were utilised with software programmed to deliver a maximum of six rotations of a feed dispensing cup, delivering approximately 45 g of pellet (as fed) per rotation, with intervals of 30 s between each rotation, so that 270 g of pellet was delivered during each visit. During the second measurement period, calves were separated according to treatment group and allocated to one of two GreenFeed systems. An adaptation period of one week preceded a 4-week experimental recording period. After two weeks, calves were rotated into the alternate pen to eliminate any possible biases between the two GreenFeed systems. During all measurement periods, a minimum of 3 h was required between visits. The calf starter pellets described above were used as an enticement. Recorded methane measurements were included if the total time spent in the feeder was > 3 min with calves visiting the feeder a minimum of three times per day to ensure repeatability of the recorded measurements27.
DNA extraction and amplicon sequencing
Genomic DNA (gDNA) was extracted from each rumen and faeces sample using a bead-beating and on-column purification28. DNA extracts were quantified on a Nanodrop 1000 Spectrophotometer (Thermo Fisher Scientific, France) and run on a FlashGel System (Lonza, Rockland, Inc.) to check integrity. Approximately 15 µg of rumen or faecal gDNA were sent to Roy J. Carver Biotechnology Center (Urbana, IL61801, USA) for microfluidics PCR amplification (Biomark HD, Fluidigm, South San Francisco, USA) and HiSeq Illumina paired end sequencing. Selected primers for amplification targeting the V3–V5 region of 16S rRNA gene of bacteria, 16S rRNA gene of archaea, fungal ITS2 and protozoal 18S rRNA gene are presented in Table S2. After amplification all samples were pooled. The library was sequenced on one lane of a HiSeq V2 Rapid flowcell for 251 cycles from each end of the fragments using a HiSeq 500-cycle SBS sequencing kit version 2 (Illumina, San Diego, USA).
Bioinformatic analysis
All pipelines have a quality control step, removing sequences with Phred scores of More