Taxonomic and Functional Shifts in the Sprout Spent Irrigation Water Microbiome in Response to Salmonella Contamination of Alfalfa Seeds

Interactions of human enteric pathogens like Salmonella with plants and plant microbiomes remain to be elucidated. The rapid development of next-generation sequencing technologies provides powerful tools enabling investigation of such interactions from broader and deeper perspectives.

ABSTRACT

Despite recent advances in Salmonella-sprout research, little is known about the relationship between Salmonella and the sprout microbiome during sprouting. Sprout spent irrigation water (SSIW) provides an informative representation of the total microbiome of this primarily aquaponic crop. This study was designed to characterize the function and taxonomy of the most actively transcribed genes in SSIW from Salmonella enterica serovar Cubana-contaminated alfalfa seeds throughout the sprouting process. Genomic DNA and total RNA from SSIW was collected at regular intervals and sequenced using Illumina MiSeq and NextSeq platforms. Nucleic acid data were annotated using four different pipelines. Both metagenomic and metatranscriptomic analyses revealed a diverse and highly dynamic SSIW microbiome. A “core” SSIW microbiome comprised Klebsiella, Enterobacter, Pantoea, and Cronobacter. The impact, however, of Salmonella contamination on alfalfa seeds influenced SSIW microbial community dynamics not only structurally but also functionally. Changes in genes associated with metabolism, genetic information processing, environmental information processing, and cellular processes were abundant and time dependent. At time points of 24 h, 48 h, and 96 h, totals of 541, 723, and 424 S. Cubana genes, respectively, were transcribed at either higher or lower levels than at 0 h in SSIW during sprouting. An array of S. Cubana genes (107) were induced at all three time points, including genes involved in biofilm formation and modulation, stress responses, and virulence and tolerance to antimicrobials. Taken together, these findings expand our understanding of the effect of Salmonella seed contamination on the sprout crop microbiome and metabolome.

IMPORTANCE Interactions of human enteric pathogens like Salmonella with plants and plant microbiomes remain to be elucidated. The rapid development of next-generation sequencing technologies provides powerful tools enabling investigation of such interactions from broader and deeper perspectives. Using metagenomic and metatranscriptomic approaches, this study identified not only changes in microbiome structure of SSIW associated with sprouting but also changes in the gene expression patterns related to the sprouting process in response to Salmonella contamination of alfalfa seeds. This study advances our knowledge on Salmonella-plant (i.e., sprout) interaction.

INTRODUCTION

Sprouts have been associated with numerous outbreaks caused by various pathogens, including Salmonella enterica and Escherichia coli O157:H7, and a variety of Salmonella enterica serovars have been implicated in Salmonella-associated outbreaks linked to sprouts (http://www.outbreakdatabase.com; https://www.cdc.gov/foodsafety/outbreaks/index.html). Although pathogen contamination of sprouts can occur during the production process, seeds are considered to be the most common source of contamination (1, 2). Consistent with this belief, seed contamination has important implications for the contamination cycle of enteric pathogens in the sprout production environment (2). However, pathogens are usually undetectable in seed lots prior to germination (3). Seed sprouting provides an excellent environment for the growth of microorganisms, including foodborne pathogens. A recent study of microbiological quality in retail alfalfa sprouts revealed a distribution of approximately 7.2 to 7.6 log CFU/g of aerobic plate count (APC) (4). Sprout spent irrigation water (SSIW), i.e., the water that has flowed through the sprouts during production, can provide a representative sample of the entire microbial population, including pathogens, in a batch of sprouts. So much so, microbial counts in the SSIW are usually within 1 log of the counts in sprouts (5). It has been recommended as an analytical sample instead of the sprouts themselves, as an important part of a multihurdle strategy to enhance sprout safety (6).

Factors affecting the growth of Salmonella during sprouting of contaminated seeds have been examined by several research groups (7–10), including initial inoculum level, incubation temperature, length of exposure, contaminated seed storage time, and seed washing frequency, as well as Salmonella serovars and strain virulence. Along with describing fluorescence microscopy observations, Barak et al. (11) demonstrated that the curli phenotype played an important role in the binding of S. enterica to alfalfa sprouts. Additionally, Howard and Hutcheson (8) confirmed that S. enterica can grow saprophytically on soluble organics released from seeds during early phases of germination. Salmonella-sprout and Salmonella-sprout microbiota interactions during the sprouting process, however, are still poorly understood.

Transcriptomics and metatranscriptomics have become powerful tools to better understand the process of disease and other complex biological processes such as biofilm formation, stress response, and pathogen-plant interaction (12–14). Unlike transcriptomics, however, metatranscriptomics can capture gene expression patterns in natural microbial communities (15, 16). Also different from metagenomics, which provides an inventory of the community gene pool, metatranscriptomics identifies the diversity of the active genes in a given ecological context, including under experimentally manipulated conditions (17, 18).

The purpose of this study was to examine the dynamics and functional activity of microbe-microbe interactions in spent irrigation water during sprouting of Salmonella-contaminated alfalfa seeds by shotgun metagenomic and metatranscriptomic approaches.

RESULTS

Shot-gun metagenome analysis reveals temporal patterns of microbial diversity in SSIW.

Shotgun sequencing was performed with spent sprout irrigation water (SSIW) at different time points from alfalfa seeds contaminated with Salmonella enterica serovar Cubana at various levels (0, 0.2, 2, and 10⁴ CFU/g of seed) and also with a tap water control. Sequencing reads were analyzed for identification of microbial DNA at the species level and determination of the organism’s relative abundance using the CosmosID bioinformatics software package (CosmosID, Inc., Rockville, MD).

Shotgun metagenomic analysis revealed a core SSIW microbiome comprising few bacterial genera dominated by Klebsiella, Enterobacter, Pantoea, and Cronobacter, with a strikingly high relative abundance (90.0% ± 6.9%) across all 4-h sampling points and associated inoculation levels, while the tap water control comprised only the genus Afipia (Fig. 1A). As the levels of Salmonella increased, however, a reduction in the relative abundance of Cronobacter could be observed, with the exception of the 24-h sample at the 2-CFU/g seed Salmonella inoculation level. The relative abundance of Pantoea decreased drastically to below 10% in the first 24 h of sprouting at all Salmonella inoculation levels except at the level of 10⁴ CFU/g seed. Salmonella is rarely detectable at concentrations lower than 0.2 CFU per gram of seed across all 4-h and 8-h sampling points (Fig. 1), and no significant change was observed in Salmonella populations over the sampling time at the Salmonella inoculation level of 0.2 CFU/g seed or 10⁴ CFU/g seed (Fig. 1). Interestingly, a peak in Salmonella relative abundance was noted between the 32-h and 36-h time points at an inoculation level of 2 CFU/g seed followed by a decrease in Salmonella relative abundance afterwards (Fig. 1). It is also notable that with a longer sampling period, an increase in the relative abundance of Pseudomonas after 48 h was observed (Fig. 1B).

FIG 1.

Taxonomic profiling of spent sprout irrigation water (SSIW) metagenomes. Classification was performed using CosmosID software (CosmosID, Inc., Rockville, MD). Genus-level taxa representing more than 3% of the annotated reads are named. (A) Alfalfa seeds were inoculated with S. Cubana at three levels (∼0.2 CFU/g, ∼2 CFU/g, and ∼10⁴ CFU/g). SSIW was sampled at 0 h, 8 h, and 24 h and every 4 h after 24 h in triplicate. (B) Alfalfa seeds were inoculated with S. Cubana at two levels (∼0.2 CFU/g and ∼2 CFU/g). SSIW was sampled at 0 h, 8 h, and 24 h and every 8 h after 24 h in triplicate, and tap water used to irrigate the sprouts was included in duplicate.

Metatranscriptome characteristics and annotation.

SSIW samples without Salmonella inoculation at various sampling points were used as controls to examine the role of Salmonella in SSIW microbial community dynamics during sprouting. After quality control, approximately 106.75 million combined metagenomic reads were produced from the Illumina MiSeq, comprising 10.8 billion bp, with an average read length of 101 bp across the 21 samples, ranging from a mean of 2,958,984 to 6,158,197 sequences per replicate samples at each time point (Table 1). After filtering out rRNA reads with the SortMeRNA algorithm against the SILVA database (Microbial Genomics and Bioinformatics Research Group), metatranscriptome data sets with a mean between 2,655,703 and 5,606,201 reads per time point were submitted to the MG-RAST pipeline (Table 1). The process resulted in transcriptional features ranging from 461,101 to 1,307,715 per metatranscriptomic data set at each time point. According to MG-RAST-based lowest common ancestor (LCA) classification of the SSIW metatranscriptomes, 50.55% to 83.67% of the functionally annotated transcriptional features were assigned to Bacteria, while 0.02% to 0.60% were assigned as “unclassified sequences” due to bacteria (Table 1). A large proportion of the transcriptional features were assigned as unclassified sequences due to Plantae from 11.82% to 48.4% per metatranscriptomic data set at each time point.

TABLE 1.

Summary of sequencing and annotation of Salmonella spent sprout irrigation water metatranscriptomes

Parameter	Value for indicated time pointa
	Samples sequenced on MiSeq						Samples sequenced on NextSeq 500
	24hC	24hS	48hC	48hS	96hC	96hS	0hS	24hS	48hS	96hS
Processing of raw sequences (averaged)
Total no. of reads	3,752,463	2,958,984	4,963,610	6,158,197	5,320,040	4,648,767	75,600,887	80,142,881	69,206,224	69,077,254
After SortMeRNA	2,899,686 (77.27%)	2,655,703 (89.75%)	4,449,227 (89.64%)	5,606,201 (91.04%)	4,206,474 (79.07%)	4,109,479 (88.40%)	72,001,722 (95.24%)	67,023,658 (83.63%)	66,010,812 (95.38%)	64,654,545 (93.60%)
Functionally annotated transcriptional feature (from MG-RAST)
Total	461,101	565,145	894,352	1,307,715	881,569	1,012,198	30,841,884	16,419,988	42,164,983	30,436,900
Bacterial (%)	68.80	82.50	80.1	83.67	50.55	73.12	98.21	78.97	99.81	97.63
Eukaryotic (%)	2.21	3.25	0.21	0.93	0.58	0.08	0.32	1.62	0.02	0.40
Unclassified virus (%)	0.15	0.03	0.08	0.11	0.08	0.09	0.07	0.09	0.02	0.04
Unclassified bacteria (%)	0.46	0.60	0.05	0.05	0.03	0.02	0.04	0.07	0.14	0.08
Unclassified plant (%)	28.09	11.82	19.17	14.86	48.4	26.05	1.29	19.05	0.00	1.80

^aEach time point includes quadruplicate samples except 24hC, 24hS, 48hC, and 0hS, which include only triplicate samples after quality control. h, hour; C, control; S, Salmonella.

In another aspect, Salmonella in SSIW at 0 h was used as a control to understand the microbial community effect on changes in Salmonella function during the sprouting process. After quality control, between 50,915,442 and 92,305,144 combined metagenomic reads were recovered from replicate samples at each time point with a total of 15 samples from NextSeq 500 system. Of these, 4 to 24% of reads matched the rRNA database and were removed once again with the SortMeRNA algorithm. Metatranscriptome data sets with a mean of between 64,654,545 and 72,001,722 reads per time point were submitted to the MG-RAST pipeline (Table 1), and averages of 16,419,988 and 30,841,884 transcriptional features per metatranscriptome data set at each time point were obtained from the pipeline. A majority of the features were assigned to Bacteria, averaging from 78.97% to 99.81%. However, various proportions of the transcriptional features were assigned as unclassified sequences due to Plantae, ranging from 0.00% to 19.05% on average.

Taxonomic abundance profiling from SSIW metatranscriptome with four different classification tools.

The taxonomic assignments of the metatranscriptomic data sets sequenced using MiSeq in this study were also predicted at the genus level with the CosmosID, MetaPhlAn, and bactiKmer, in addition to the MG-RAST annotation pipeline. Variations in relative abundance were observed across the samples between biological replicates and among the different taxa classifiers (Fig. 2). Despite the variation in the relative abundances, the same families appeared as active bacterial members residing in SSIW across almost all the classifiers, except MetaPhlAn. Salmonella, Pantoea, Pseudomonas, Cronobacter, and Enterobacter, together with Bacillus, Erwinia, Paenibacillus, and Escherichia, were identified with at least two of the classifiers, representing the most active genera in SSIW (Fig. 2). It should be noted that some of the tools employed here identified additional nonbacterial genus/species due to differences in their reference databases. For example, MetaPhlAn identified two plant virus species (i.e., peanut stunt virus and alfalfa mosaic virus), while MG-RAST detected two genera of Plantae in the legume family (Fabaceae), Glycyrrhiza and Medicago, and one genus of Fungi, Mucor (Fig. 2).

FIG 2.

Taxonomic assignment of SSIW metatranscriptomes. Alfalfa seeds were inoculated with S. Cubana at ∼10⁶ CFU/g seed. Enriched mRNA from SSIW at three time points (24 h, 48 h, and 96 h) was sequenced and annotated using four different classification tools: MetaPhlAn (A), MG-RAST (B), bactiKmer (C), and CosmosID (D). Relative abundance (>0.5% cutoff) for taxa with genus-level assignment is reported.

Changes in SSIW microbial community function associated with S. Cubana seed contamination.

Global functional classification of prokaryotic transcriptional features from MiSeq SSIW metatranscriptomic data sets was performed with SEED subsystems in MG-RAST. Among the functional categories identified by MG-RAST, the five most dominant categories based on the relative abundance of assigned reads across all SSIW samples from the control group, and SSIW samples from Salmonella-contaminated seeds, were protein metabolism (16.1% ± 4.65% and 15.8% ± 3.67%), clustering-based subsystems (functional coupling evidence but unknown function; 12.8% ± 0.35% and 12.6% ± 0.50%), carbohydrates (11.0% ± 1.91% and 11.6% ± 1.91%), amino acids and derivatives (6.6% ± 1.14% and 6.7% ± 0.99%), and cell wall and capsule (5.8% ± 0.90% and 5.9% ± 0.79%) (Fig. 3A). Comparative analysis of the SSIW communities with and without Salmonella based on the full set of replicates showed the same top 10 most enriched functional categories between the two microbial communities (Fig. 3A). Moreover, different temporal patterns were observed within the SSIW microbial community. For example, increased relative abundance over time was found in genes related to protein metabolism and RNA metabolism. In addition, reduced relative abundance over time was found in genes related to DNA metabolism, cell wall and capsule, and fatty acids, lipids, and isoprenoids. Lastly, the relative abundance of genes related to other functions in Fig. 3B displayed “peak” or “valley” patterns over time. However, contamination with S. Cubana did not alter the existing temporal pattern of the genes related to most functional categories in the SSIW microbial community. No effect or even slight effect on the relative abundance was observed compared to the SSIW control microbiome at different time points. Nevertheless, inoculating alfalfa seeds with S. Cubana did change the temporal patterns in seven functional categories, including carbohydrates, clustering-based subsystems, stress response, regulation and cell signaling, motility and chemotaxis, nucleosides and nucleotides, and respiration (Fig. 3C). For instance, compared with the SSIW control microbiome, the increase in relative abundance in genes related to carbohydrate metabolism at 24 h changed the dynamics of carbon metabolism in the SSIW-Salmonella microbiome. Subsystem level 2 analysis highlighted the role of reads encoding sugar alcohols in the temporal change of overall relative abundance of carbohydrate-related reads in the SSIW-Salmonella microbiome (Fig. 3D). The dynamics in relative abundance of annotated reads corresponding to cold shock, osmotic stress, and detoxification contributed largely to the pattern change observed related to stress response (Fig. 3D). It was also noted that the abundance changes in genes associated with quorum sensing and biofilm formation as well as regulation of virulence at different time points may both play an important role in the overall dynamic change in regulation and cell signaling when comparing the SSIW Salmonella microbiome with SSIW control microbiome. (Fig. 3D).

FIG 3.

Changes in SSIW microbial community function associated with S. Cubana seed contamination. (A) Functional categories of the metatranscriptomes from SSIW control community and SSIW-Salmonella community. Functional classification of transcriptional features was done based on SEED subsystem. Bars represent percentage of features (n = 3; means ± standard deviations) that were classified into the first functional category level. (B) Different temporal patterns in average relative abundance observed in selected most transcribed gene functions in SSIW control microbial community. (C) Changes of average relative abundance in selected most transcribed gene functions across time associated with S. Cubana seed contamination. (D) The second functional category levels in carbohydrate, stress response, and regulation and cell signaling functions, shown as average relative abundance.

The same SSIW metatranscriptomic data sets were also annotated by comparison with the Kyoto Encyclopedia of Gene and Genomes (KEGG) database in MG-RAST. The top 10 most enriched KEGG pathways across all of the samples from Salmonella-contaminated alfalfa seeds during sprouting were identified based on the hits of assigned reads to KEGG orthology (KO) accession numbers when comparing the samples with controls at the same time point. Numbers of differentially regulated genes among the top 10 pathways at each time point were grouped based on high-level function, i.e., metabolism, genetic information processing, environmental information processing, and cellular processes (Fig. 4). Overall, the effect of Salmonella contamination of alfalfa seeds on the SSIW microbial community was dynamic. In the cellular process function, flagellar assembly pathways were highly enriched at 24 h, with upregulation of seven genes involved in the assembly. At 48 h, bacterial chemotaxis pathways were significantly enriched, with downregulation of 11 genes involved in the network. Notably, the microbial community was most active at 48 h with pathways involved in metabolism and environmental information processing, while networks within genetic information processing, for instance, homologous recombination, were most active during the first 24 h (Fig. 4).

FIG 4.

Activated genes associated with enriched biological pathways across time in response to S. Cubana seed contamination. The top 10 enriched KEGG pathways in the SSIW metatranscriptomic data set were identified using gene set enrichment analysis. The numbers of genes upregulated (black) and downregulated (gray) under the high-level KEGG pathway category are summarized. Sal, Salmonella; Con, control.

Changes in Salmonella function during interaction with SSIW microbial community.

Totals of 541, 723, and 424 S. Cubana genes at 24 h, 48 h, and 96 h, respectively, were either upregulated or downregulated at least 2-fold (false-discovery rate [FDR] ≤ 0.05) compared with S. Cubana at 0 h in SSIW during sprouting (Fig. 5; see also Table S1 in the supplemental material). Among the three time points sampled during sprouting, most changes in the S. Cubana transcriptome were observed at 48 h. Interestingly, a substantial pool of S. Cubana genes (107) were induced at all three time points, including genes involved with biofilm modulation (bhsA), curli synthesis (csg operon), cellulose biosynthesis (yhjQ), acid adaptation (cad operon), hyperosmotic stress response (osm operon and otsB), superoxide stress response (ibpA), universal stress response (uspF and uspG), lipopolysaccharide biosynthesis (wzzB and manC), type III secretion effector (slrP), toxin synthesis (ldrD and ltxB), a DNA gyrase inhibitor (sbmC), and a transcriptional regulator (yvoA). The transcriptional regulator shared a sequence identity of 38% with the well-studied DasR regulator from the antibiotic-producing soil bacterium Streptomyces, which represents a master switch in a signaling cascade from the nutrient GlcNAc to antibiotic production. However, only three genes were downregulated and shared by all three time points. One of these three genes is metR, encoding a homocysteine-dependent transcriptional activator, which controls methionine biosynthesis and transport (Table S2). It is also important to note that although some genes were upregulated or downregulated at all sampled times points compared with 0 h, the level of expression exhibited temporal dynamics. For example, while cadABC showed upregulation at all three time points, the level of expression decreased substantially from 24 h to 48 h (Fig. 6), indicative of changes in lysin-dependent acid resistance. Also, a decrease in the level of expressions of osmB, osmX, and osmY after 24 h also suggested changes to the levels of osmotic stress over time (Fig. 6).

FIG 5.

Venn diagram of numbers of differentially expressed genes (DEGs) in S. Cubana from the SSIW-Salmonella microbial community at 24 h, 48 h, and 96 h compared to 0 h. Genes in overlapping sets show the differential expression in two or three comparison pairs.

FIG 6.

Mean (log₂) fold changes in cad and osm gene expression in S. Cubana from the SSIW-Salmonella microbial community at 24 h, 48 h, and 96 h compared to 0 h. Different significance levels with adjusted P values are shown as follows: ***, P ≤ 0.001; **, P ≤ 0.01; and *, P ≤ 0.05.

These differentially expressed genes were further compared with the KEGG database to determine enriched KEGG pathways in S. Cubana at each time point during sprouting. The 10 most enriched KEGG pathways in S. Cubana were identified based on the hits of assigned reads to the Salmonella enterica subsp. enterica serovar Cubana KEGG genes database in paired comparisons between different time points during sprouting. Numbers of differentially regulated genes among the top 10 pathways at each time point were grouped together based on high-level function (i.e., metabolism, genetic information processing, environmental information processing, and cellular processes) (Fig. 7A). Metabolically, S. Cubana was most active at 48 h, with the greatest number of genes involved in metabolic pathways compared with those at 24 h and 96 h. However, the genetic profile involved in metabolic pathways at 48 h was strikingly different from those at 24 h and 96 h. At 48 h, 75% of the genes involved in metabolic pathways were downregulated, while 69% and 34% of these genes at 24 h and 96 h, respectively, were upregulated within the networks. Moreover, 63% of genes at 96 h were upregulated compared with genes at 48 h. Two-component systems and ABC transporters were the two main networks under the environmental information processing function observed among the top 10 most enriched pathways (Fig. 7B). A temporal pattern similar to that in metabolic pathways was found in genes associated with two-component systems and ABC transporters. However, it is worth mentioning that the greatest number of upregulated genes was observed at 48 h instead of 24 h in the two-component systems.

FIG 7.

Activated genes associated with enriched biological pathways in S. Cubana across time when interacting with the SSIW microbial community. The top 10 enriched KEGG pathways in S. Cubana from the SSIW metatranscriptome data set were identified using gene set enrichment analysis. (A) The numbers of genes upregulated (black) and downregulated (gray) under the high-level KEGG pathway category are summarized. (B) The numbers of genes upregulated (black) and downregulated (gray) in two-component system and ABC transports, specifically, are summarized.

DISCUSSION

Recent advances in human enteric pathogen-plant interaction insights have provided a better understanding of colonization and persistence of enteric pathogens on and in plant tissues (19–21). However, factors involved in the fitness of enteric pathogens in this ecological niche and their interaction with plants remain to be elucidated. Microbiome profiling of plants has revealed a diverse and highly dynamic plant microbiome, often termed the plant’s “second genome” (22). Several studies have shown that bacterial communities are dynamically shaped by environmental factors, as are the members within that community (23–25). Since sprouts are germinated or partially germinated seeds and traditionally produced entirely in water, the microbial properties of the spent sprout irrigation water (SSIW) should best inform our understanding of the nature and dynamic of the spermosphere (i.e., the short-lived, rapidly changing zone of soil/water surrounding a germinating seeds) microbes (26). The dominant spermosphere bacteria can be recruited from the seed endophytic and epiphytic microbiota or the surrounding water. In this study, the microbial community of SSIW from S. Cubana-contaminated alfalfa seeds was profiled using both metagenomic and metatranscriptomic approaches. After analyzing our data sets with the CosmosID bioinformatics platform, several genera from the Enterobacteriaceae family were revealed to comprise the majority SSIW microbiome compared to tap water controls. Among the taxa, Klebsiella, Enterobacter, Pantoea, Pseudomonas, Paenibacillus, and Bacillus are well-known spermosphere bacteria, dominating bacterial communities not only in most soil but also in endophytic and epiphytic seed communities (26, 27). The metatranscriptome captures real-time functional activities of the microbiome; therefore, greater diversity was observed within the same sample and among individual samples in the metatranscriptome. However, both metagenome and metatranscriptome analyses revealed temporal patterns in the relative abundances of Pantoea Tatumella, and Pseudomonas species in the SSIW control community and the changes in the relative abundances of Cronobacter, Klebsiella, and Enterobacter species in the SSIW Salmonella community as Salmonella inoculation levels increased. These data indicated complex microbe-microbe and microbe-seed interactions during sprouting. Additionally, every-8-h instead of 4-h sampling did not change the core microbiome and associated temporal pattern as well as Salmonella relative abundance. The 8-h sampling scheme, however, did greatly increase the diversity of the SSIW microbiome.

Currently, how much seed endophytic species or environment-inhabiting species contribute to the proliferating spermosphere and/or spermoplane microbiota during seed germination is unknown. Matos et al. compared the culturable portions of the native microflora of various types of sprouts and associated physiological profiles to assess the relative effects of sprout type and inoculum factors on the microbial community structure of sprouts (28). Variability among sprout types was found to be more extensive than any differences between microbial communities associated with sprouts from different sprout-growing facilities and seed lots (28). The sprouting environment, however, may play a role on the microbiota of sprouts. Weiss et al. found that dominating cultivable species were different in hydroponically grown sprouts versus soil-grown samples (29). Using seeds from the same distributor, similar organismal families were found on all final sprout varieties and were primarily composed of Pseudomonadaceae. However, commercially germinated sprout varieties housed more diverse microbial families than laboratory sterile-water-germinated sprouts of all three varieties (30). Asakura et al. (31) reported that seasonal and growth-dependent variation of bacterial community structure were observed in radish sprouts using 16S rRNA sequencing analysis. A predominance of Pseudomonas spp. was found throughout seasons, with summer samples exhibiting an increase in Enterobacteriaceae and decreases in Oxalobacteraceae and Flavobacteriaceae compared with winter samples. Compared with presprouted seeds, an increased proportion of Pseudomonas spp. was observed after sprouting (31). In this study, after ample comparison to tap water controls, different bacterial species were found to be recruited from alfalfa seeds and dominate the spermosphere at different growth stages of the sprouting, suggesting that the spermosphere microbiota is dynamic, not static, and that microbial effects during seed development and dispersal events may be especially important to the microbiota of alfalfa sprouts.

A suite of software tools has been developed to taxonomically classify metagenomic and metatranscriptomic data and estimate taxon abundance profiles. In this study, four different software tools were used to perform taxonomic classification of the SSIW metatranscriptomic data set. Among these tools, the CosmosID bioinformatics platform and BactiKmer metagenome pipeline (32) utilized high-performance k-mer-based algorithms and curated taxonomy databases (GenBook in CosmosID). MetaPhlAn is a computational tool for profiling the composition of microbial communities from metagenomic shotgun sequencing data and relies on unique clade-specific marker genes identified from 3,000 reference genomes (33). The MG-RAST automated analysis pipeline uses a DNA-to-protein classifier and the M5nr (MD5-based nonredundant protein database) for annotation (34). In this study, the same metatranscriptomic data set, after filtering out all the rRNA reads, was subjected to these four different classifiers. It is of note that some variations in the taxonomic assignments revealed in the analytical results were due to the completeness of the precompiled database from each software tool. Others might be caused by reclassification of certain genera. For example, some species of Pantoea were transferred to the genus Tatumella (35), and some species of Enterobacter were transferred to the genera Kosakonia and Lelliottia (36, 37). In addition, bias was observed using MetaPhlAn as a taxonomic classifier due to an uneven distribution of marker sequences among the microbial sequences of interest (38). However, most databases still remain poorly populated below the species level. Thus, depending on the application, a single or multiple classifiers may be chosen.

The release of molecules from germinating seeds into the surrounding water generates a rapid explosion of microbial growth and activity in the spermosphere (39–41). Community-level physiological profile (CLPP) analysis also suggested significant changes in the microbial community metabolic diversity during sprouting for alfalfa sprouts (10). Seed exudates are rich in sugars, amino acids, and organic acids, among other things (39, 41). Results from early studies (42) showing that Enterobacter cloacae, a well-known plant endophyte commonly found in seeds, grew on simple mono- and oligosaccharides but not on polysaccharides paralleled the proliferation of E. cloacae and the increased monosaccharide and the di- and oligosaccharide catabolism in the SSIW microbiome in this study. Moreover, the greater abundance of E. cloacae in the SSIW Salmonella microbiome may explain why contamination of Salmonella in alfalfa seeds did not notably affect seed germination and growth since E. cloacae has been shown to enhance seed germination and seedling growth (43). Recent studies pointed to sugar alcohols, a class of polyols, as having a role in plant-pathogen interaction. It was observed that the tomato pathogen Cladosporium fulvum produced mannitol to suppress reactive oxygen species (ROS)-mediated plant defenses (44). In addition, mannitol production and secretion by a fungal pathogen of tobacco and numerous other plant species, Alternaria alternata, was massively induced by host plant extracts (44, 45). In this study, the greater abundance of genes associated with sugar alcohol catabolism in the SSIW Salmonella microbiome in the first 24 h during sprouting may suggest a Salmonella-sprout interaction. This result thus may also indicate a Salmonella-microbial community interaction, as sugar alcohols may have effects on enzyme activity and microbial community structure (46). Further investigation of the various molecules released during sprouting will help to better understand the roles of exudate molecules in stimulating pathogens and supporting bacterial growth in the spermosphere and also in influencing the interactions that take place in the spermosphere.

Within the SSIW microbiome, ecological competition is often intense, particularly among species with overlapping nutrient requirements. The main functional roles of SSIW microbiota are related to nutrient processing, energy production, and biosynthesis of various secondary metabolites, as suggested in this study. Therefore, microbial members in the SSIW microbiome are likely subjected to various environmental stressors associated with competition and the host, such as exposure to reactive oxygen species and secondary metabolites released from plant defenses and microbial species in the microbiome. In response, related genes and pathways were observed in high abundance in the metatranscriptome of SSIW with or without Salmonella. Interestingly, increased abundance in oxidative stress, osmotic stress and quorum sensing, and biofilm formation in the SSIW-Salmonella microbiome at different time points was observed compared with the control microbiome. In addition, downregulation of antibiotic biosynthesis was observed in the KEGG pathway enrichment analysis. All of these data point to the notion that Salmonella induces stress response, biofilm formation, and antibiotic tolerance in this ecological niche in response to the host and its own competitors, as suggested in a very recent study by Lories et al. (47).

This study also used a metatranscriptomic approach to examine the SSIW microbiome with and without Salmonella. KEGG pathway enrichment analysis suggested that inoculation of S. Cubana in alfalfa seeds altered the function of the SSIW microbial community in metabolism, environmental information processing, and cellular processes. For example, gene functions related to membrane transport and signal transduction (i.e., two-component systems) were highly enriched, suggesting dynamic Salmonella-SSIW microbiome and Salmonella-sprout interactions. Moreover, these enriched pathways recruited the greatest number of genes at 48 h, and the least number of genes at 96 h, supporting temporal dynamic interactions in the SSIW microbiome. In addition, KEGG pathway enrichment analysis in both the metatranscriptome data sets supported temporal change in the regulation of genes related to flagellar assembly and bacterial chemotaxis pathways under cellular processes, suggesting a motility-to-biofilm transition in the SSIW-Salmonella microbial community in the first 48 h of sprouting (48).

In a previous study, Brankatschk et al. (49) examined Salmonella enterica subsp. enterica serovar Weltevreden interaction with alfalfa sprouts during colonization using transcriptome sequencing (RNA-seq). In comparison with M9-glucose medium, the study showed that 177 genes (4.3% of the S. Weltevreden genome) were transcribed at higher levels with sprouts, including the genes coding for proteins involved in attachment, motility, and biofilm formation and proteins of Salmonella pathogenicity island 2, clearly demonstrating some of the commonality shared between bacterial infections of plants and humans. The main caveats of this study were that Salmonella was only inoculated on 5-day-old sprouts and the use of S. Weltevreden in the M9-glucose medium for comparison. In the present study, which focused solely on SSIW, the sprouting process was followed for several days, and the transcriptome profile of S. Cubana at different time points was compared to that of S. Cubana at 0 h in SSIW during sprouting. One possible explanation for this difference may be that this study focused solely on SSIW and not sprouts. Differential gene expression and pathway enrichment analyses showed a marked shift in major transcriptional activities to metabolism and environmental information processing such as two-component systems and ABC transporters, suggesting a dynamic interaction of S. Cubana with the SSIW microbial community. Previously, a shift in the expression patterns of various metabolic pathways was also found in Escherichia coli O157:H7 exposed to lettuce leaves or leaf lysates (50, 51). Constitutive upregulation of the transcriptional regulator yvoA (52) and downregulation of homocysteine-dependent transcriptional activator gene metR (53) further suggest their contribution to metabolism shifts in S. Cubana. Moreover, genes that were transcribed at higher levels at all three time points sampled during sprouting shed more light on the survival and adaptation mechanisms of S. Cubana in the SSIW microbiome. S. Cubana enhanced stress response to oxidative species, hyperosmotic stress, and other stresses by inducing genes including the cad and osm operons, otsB, ibpA, uspF, and uspG (54–57). In response to limited nutrients, in addition to altered metabolism, S. Cubana induced biofilm formation by upregulation of bhsA, the csg operon, and yhjQ (58, 59). In response to the competitors in the SSIW microbiome, S. Cubana increased virulence (slrP and manC), antibiotic and toxin production (yvoA, ldrD, and ltxB), and resistance to antimicrobials produced by other competitors (sbmC) (60–64).

In summary, all of these data demonstrated that the addition of Salmonella in the spermosphere environment may help reshape the SSIW microbiome structurally and functionally. This approach presented a less biased and more real-time and global view of Salmonella-sprout interactions.

MATERIALS AND METHODS

Bacterial strain and growth condition.

S. enterica serovar Cubana strain CFSAN055271 was obtained from the stock culture collection of the Division of Microbiology, Center for Food Safety and Applied Nutrition, U.S. Food and Drug Administration, College Park, MD. It was originally isolated from alfalfa sprouts in 2011 by the U.S. Department of Agriculture Microbiological Data Program (MDP). Stock culture was stored in brain heart infusion (BHI) broth containing 25% glycerol at −80°C and maintained on tryptic soy agar (TSA) plates.

Inoculation of seeds.

A single colony of S. Cubana culture was transferred to 5 ml of tryptic soy broth (TSB) and grown at 36°C for 18 to 20 h. The culture was harvested by centrifugation at 5,000 × g for 10 min, followed by washing with 0.01 M phosphate-buffered saline (PBS) (pH 7.2) three times and then resuspension in 5 ml of TSB. For seed inoculum, the culture was further diluted in sterile double-distilled water (ddH₂O) to four different levels (∼0.2 CFU/g, ∼2 CFU/g, ∼10⁴ CFU/g, and ∼10⁶ CFU/g of seed). Alfalfa seeds (65 g or 85 g) were soaked in each 250-ml inoculum for 20 min, drained, and allowed to air dry at room temperature in a Lumina hood. Seeds were stored at 4°C until use. The inoculation level was determined by plate count immediately following inoculation. To minimize the variations that could be introduced to the experiment, the same batch of seeds was used in the study, and the seeds were inoculated, stored, and sprouted at the same time for a given experiment.

Sprouting of alfalfa seeds.

The seeds were sprouted in triplicate or quadruplicate in an Easy-Sprout sprouter as follows. Twenty grams of seeds per growing vessel was soaked in 250 ml of tap water for 8 h at ambient temperature and drained. Containers were capped with vented lids and incubated at ambient temperature for 4 days. The sprouting seeds were rinsed every 4, 8, or 24 h with 250 ml to 300 ml of tap water to meet various experimental designs.

Sample collection and DNA extraction.

In the control, 0.2-CFU/g inoculation, and 2-CFU/g inoculation groups, 50 ml of sprout irrigation water from each sprouter was collected at 0 h, 8 h, and 24 h, and 20 ml was collected every 4 or 8 h after 24 h in triplicate. In the 10⁴-CFU/g inoculation group, 5 ml was collected at each corresponding time point in triplicate. After collection, all samples were filtered through a MicroFunnel unit with a 0.2-μm Supor membrane (Pall Corporation, Port Washington, NY). In addition, 500 ml of tap water was filtered to be used as a water control. After filtration, the filters were stored at −20°C prior to DNA extraction.

Bacterial DNA was extracted with the DNeasy PowerWater kit (Qiagen, Venlo, Netherlands) following the manufacturer’s recommended protocol, with only one modification: the isolated DNA was eluted in a final volume of 50 μl. All DNA samples were stored at −20°C prior to preparation of sequencing libraries.

Sample collection and RNA extraction.

A total of 250 ml of sprout irrigation water from each sprouter was collected at 0 h, 24 h, 48 h, and 96 h in quadruplicate at a 10⁶-CFU/g inoculation level. Bacterial cells were pelleted with a Sorvall RC 6+ centrifuge (Thermo Fisher Scientific, Waltham, MA) at 12,000 rpm and 4°C for 15 min using an F12S-6 × 500 LEX fixed-angle rotor. Immediately after centrifugation, the bacterial pellets were resuspended in RNAlater (Thermo Fisher Scientific) solution to stabilize RNA in the cells. All samples were stored at −80°C prior to RNA extraction.

Extraction of total RNA and removal of remaining DNA were carried out using the RiboPure-bacteria kit (Thermo Fisher Scientific) following the manufacturer’s instructions. The yield of total RNA was measured using a Qubit fluorometer (Thermo Fisher Scientific), and the integrity of the RNA was verified using an Agilent Bioanalyzer 2100.

Library preparation and sequencing.

For shotgun metagenomic sequencing, DNA sample libraries were constructed with Nextera XT library preparation kits or Nextera DNA Flex library prep kits (Illumina, Inc., San Diego, CA) following the manufacturer’s protocols. Library sequencing (paired-end, 2 × 250 bp) was performed on an Illumina MiSeq (Illumina, Inc.).

For shotgun metatranscriptome sequencing, rRNA was depleted using the Ribo-Zero magnetic kit for bacteria (Illumina, Inc.) following the manufacturer’s instructions. The removal of rRNA was verified with an Agilent Bioanalyzer 2100, and remaining mRNA was resuspended in 18 μl of Elite, Prime, Fragment (EPF) mix from the Illumina TruSeq RNA sample preparation kit v2 (Illumina, Inc.). The following cDNA synthesis and library preparation were performed using the TruSeq RNA sample preparation kit v2 low-sample (LS) protocol. Resultant cDNA libraries were normalized, pooled, and sequenced (2 × 150 bp) on one of the Illumina platforms, a MiSeq or NextSeq 500 system with a high-output flow cell (400 million).

Bioinformatics analysis.

For metagenomic analysis, raw sequence reads were analyzed using the CosmosID bioinformatics software package (CosmosID, Inc., Rockville, MD) to reveal microbial community composition, antibiotic resistance markers, and virulence gene pools. For metatranscriptomic analysis, raw sequence reads were annotated through CosmosID, Metagenomic Phylogenetic Analysis (MetaPhlAn), BactiKmer (an in-house custom k-mer database and C++ search program), and the metagenomics Rapid Annotation using Subsystem Technology (MG-RAST) pipelines for taxonomic profiling. The relative abundance of each bacterial organism per sample was expressed as a percentage of the total number of bacterial reads belonging to that organism, normalized for organism-specific genome length. Reads identified as eukaryotic, viral, or archaeal were excluded depending on the software package. For functional analysis, reads from remaining rRNA after Ribo-Zero treatment were removed through SortMeRNA (65). Functional classification of transcriptional features in the reads from SSIW microbiome was done based on SEED subsystem in MG-RAST (66). For Salmonella differential expression analysis, the raw sequence data from the NextSeq platform was imported into CLC Genomic Workbench (v9) after removal of rRNA reads and mapped to the annotated reference genome (CFSAN055271, with NCBI SRA accession number SRR4175562, and annotated by Prokka v1.12). The expression values for each gene and each transcript within S. Cubana were calculated using the RNA-seq analysis tool in CLC Genomics Workbench (v9). Genes differentially expressed in S. Cubana at different time points were determined using EdgeR (version 3.28.0) with an FDR of ≤0.05.

Data availability.

Metagenomic and metatranscriptomic data are deposited in the NCBI Sequence Read Archive (SRA) database with accession numbers SRR12284435 to SRR12284466 (listed in Table S3).