Profiling bacterial communities of irrigation water and leafy green vegetables produced by small-scale farms and sold in informal settlements in South Africa

Abstract

Morogo is an African indigenous term used for leafy green vegetables harvested in the wild or cultivated in small-scale farms and consumed by the local populations of the region. Small-scale farmers have gained recognition as important suppliers of morogo to informal settlements. In commercial production systems, leafy green vegetables have increasingly been reported as associated with foodborne pathogens and disease outbreaks. Little is known of the presence of these organisms on leafy green vegetables in the informal unregulated food systems. This study aimed to profile bacterial communities in irrigation water (flooding and overhead irrigation water) and leafy green vegetables (Brassica rapa L. chinensis and Brassica rapa varieties of morogo) to establish the natural bacterial flora at the water-fresh produce interface from five small-scale farms in two provinces in South Africa. Illumina MiSeq high-throughput sequencing showed that each farm exhibited a unique bacterial community composition, with an overall high relative abundance of Proteobacteria, Firmicutes and Actinobacteria, including prominent families such as Burkholderiaceae (48%), Enterobacteriaceae (34%), Bacillales Family XII (8%), Rhodobacteraceae (3%), Micrococcaceae (1.98%) and Pseudomonadaceae (1.79%). Specific Enterobacteriaceae Serratia, Enterobacter, Salmonella, Shigella, Escherichia coli, Buchnera, Citrobacter, Klebsiella and Proteus were identified, in addition to unique communities associated with plant or irrigation water source. These findings suggest that the edible plant microbiome can play an important role as transient contributor to the human gut and has the potential to affect overall health.

Introduction

Small-scale farming in South Africa accounts for 80% of fresh produce supplied to informal settlements (Bunce 2019). Despite this important food security role, small-scale farmers often struggle to produce optimal yields, mainly due to lack of access to adequate water resources and infrastructure, lack of land and funding, as well as technical support and know-how to consistently supply safe, quality food (Hlophe-Ginindza and Mpandeli 2020). The transmission of foodborne pathogens in animal-based food systems such as meat, poultry, eggs and raw milk is well established (Abebe et al. 2020; Heredia and García 2018). However, foodborne pathogens outbreaks have increasingly been associated with fresh and minimally processed leafy green vegetables (Carstens et al. 2019, Herman et al. 2015). Little is known of the prevalence of foodborne pathogens in the informal food production systems. What is known is that these food systems carry an additional burden of being reliant on compromised water resources (Bhagwat 2019) without resources to manage and improve the quality of the water and having poor hygiene and sanitation systems operating in on an unregulated environment (Daniel 2022).

The diverse microbiomes of vegetables may serve as reservoirs for opportunistic and emerging pathogens (Berg et al. 2014). Bacteria in these matrixes occur as both beneficial and pathogenic variants when associated with plant, human and animal health. Pathogenic bacteria such as Campylobacter spp., Clostridium botulinum, Clostridium perfringens, Bacillus cereus, Escherichia coli O157:H7 and other Shiga toxin-producing E. coli, Listeria monocytogenes, Salmonella spp., Shigella spp., enterotoxigenic Staphylococcus aureus, Vibrio cholera, and Yersinia enterocolitica can be significantly detrimental to human health in this one health context. Prevention of source contamination is predominantly advocated through more effective municipal waste management systems, improved hygiene practices and adoption of good agricultural practices within the farm-to-fork continuum. In the process of fresh produce e.g., leafy green vegetable production, contamination with pathogenic microorganisms may result through direct contact with contaminated irrigation water (groundwater, surface water, and human wastewater), improper composted manure, infected soil, or contact with food handlers not having access to basic potable water, hygiene and sanitation systems (Alegbeleye et al. 2018). While agricultural scientists are mindful of potential pathogenic microorganisms found naturally on the plant surface, the awareness of endophytic pathogens is only recently been shown (Dong et al. 2019, Mulaosmanovic et al. 2021). While most endophytes are likely non-pathogenic to humans, several pathogenic bacteria can become internalized and colonise the leaves as temporary endophytes (Mengistu 2020). From the perspective of consumer safety, these temporary pathogenic endophytic inhabitants and other epiphytes present within the vegetable biome can potentially cause disease outbreaks, particularly where no washing or a heat treatment step is practised after harvest.

Indigenous leafy green vegetables, also known as “morogo”, naturally harbour a diverse microbial consortium in the phyllosphere (Taffner et al. 2020). They are considered highly nutritious, are popular with local communities and are mostly traded in informal markets. However, little is known about the water-indigenous leafy green-vegetable nexus in informal food systems in South Africa. The presence of potential human pathogens in the microbial community of indigenous leafy green vegetables is of interest in the context of a product that has a high likelihood of contamination during production and further on in the supply chain during transport, sales and preparation. As such, it is imperative to determine the microbial ecology of the local indigenous leafy green small-scale production systems and informal supply chain. Next-generation 16S rRNA gene sequencing provides a tool to unravel the bacterial diversity of irrigation water, and fresh produce in the informal food system, detect the presence of potential foodborne pathogens amplicons and highlight the core microbiome that is prevalent in the one health context.

Materials and methods

Sample collection and pre-processing

A total of 36 samples of overhead or flood irrigation water was collected from multiple irrigation points across the plot ensuring water was in the proximity of both crops sampled. Brassica rapa L. chinensis, now referred to as chinensis (n = 25), and Brassica rapa, now referred to as rape (n = 25), plants were randomly collected from a single plot per cultivar, once off using systematic random sampling. Several leaves free from soil, from different plants were pooled to represent a single sample. Five small-scale farms in the Brits (North West Province, farming site A, B, C and E) and Delmas (Mpumalanga Province, farming site D) regions, South Africa (Additional file 1: Table S1) were sampled at. Water samples (1 L) were filtered through a 0.20 µm nitrocellulose filter membrane (Sartorius Stedim Biotech, Gottingen, Germany) and aseptically stored at 4 °C until DNA extraction. Chinensis and rape samples (25 g) were placed into stomacher bags containing 225 mL of Peptone Buffered Water (PBW) and subsequently homogenized for five min at maximum speed (level 4) using a Homogenizer Laboratory Blender (Thermo Fischer Scientific, Johannesburg, South Africa). The homogenates were transferred to 50 mL sterile falcon tubes and centrifuged at maximum speed (4000 rpm) for 10 min. The resulting pellets were stored at − 20 °C until DNA extraction.

DNA extractions

Total community DNA from water samples were extracted from the filter membranes using the DNeasy^® PowerWater^® extraction kit (Qiagen, Johannesburg) following the manufacturer’s guidelines. For fresh produce samples, DNA was extracted from fresh produce homogenate pellets using the DNeasy^® PowerSoil^® extraction kit (Qiagen) according to the manufacturer’s guidelines. All DNA samples were quantified and verified for purity using Qubit 2.0 fluorometer with both double stranded broad range and high sensitivity range, according to manufacturer’s guidelines (Life Technologies, Johannesburg) and Nanodrop ND-2000 UV–VIS Spectrophotometer (Nanodrop Technologies, Wilmington, DE, USA).

16S rRNA amplicon sequencing and processing

Library preparation and paired-end sequencing (2 × 300 bp) of the 16S rRNA V4-V5 variable gene region was performed by Molecular Research DNA (MR DNA, Shallowater, TX, USA) on the Illumina MiSeq platform using primers 515F (5’-GTGCCAGCMGCCGCGGTAA-3’) and 909R (5’-CCCCGYCAATTCMTTTRAGT-3’) (Wang and Qian 2009). Generated sequences were returned and pre-processed by demultiplexing and barcode and primer removal using the MR DNA in-house pipeline, and MR DNA freeware for binning and conversion from fastq to fasta and quality files. Sequence data are available at NCBI-SRA under submission numbers SUB12270895 and SUB12272756 for BioProject number PRJNA900001.

The bioinformatics open-source software Quantitative Insights Into Microbial Ecology 2 (QIIME2) version 2020.2 was utilised for raw data processing (Bolyen et al. 2019). Demultiplexing of sequences was performed according to the DADA2 workflow, using the q2-demux plugin (https://github.com/qiime2/q2-demux). Reads were trimmed (forward: 12 bp, end: 220 bp) prior to quality filtering and de-replication using the q2-dada2 plugin (Callahan et al. 2016). This process concurrently removes chimeric sequences and generates amplicon sequence variants (ASVs), using nucleotide quality scores. Taxonomic classification of ASVs was assigned using the Naïve Bayes classifier and the q2-feature-classifier plugin, which assigns taxonomies based on sequence k-mer frequencies (https://github.com/qiime2/q2-feature-classifier), and the self-trained SILVA 138 16S rRNA gene database clustered at 99% similarity (Quast et al. 2013). Mitochondria, chloroplast and archaea sequences were removed. Where necessary, feature tables produced were normalised by rarefaction to a sampling depth of 24,230 reads for downstream diversity analysis.

Diversity measurements and statistical analyses

The alpha (α-) and beta (β-) diversity of bacterial communities was determined using the vegan function in R Statistical Software (R Core Team, 2021). The α-diversity indices measured using non-rarefied data included bacterial community richness, Pielou’s evenness index (J), and Shannon diversity index (H). The bacterial community structure between samples (β-diversity) was visualised using rarefied tables for non-metric multidimensional scaling (nMDS) ordination of the Bray–Curtis dissimilarity matrix (stress values: 0.083–0.165) and verified with permutational multivariate analysis of variance (PERMANOVA, 999 permutations) based on the factors source (overhead water or flooding water; chinensis or rape), area (Brits or Delmas) and farming site (A, B, C, D, E). Homogeneity of multivariate dispersions was determined with a permutational multivariate dispersions (PERMDISP) test using the Bray–Curtis similarity matrix (Anderson et al. 2006).

The Shapiro–Wilk normality test was used to determine the distribution of samples for further statistical testing. Where normal distribution was observed, a T-test or ANOVA was used followed by a post hoc Tukey test. Alternatively, the Kruskal–Wallis H sum test was used for samples not normally distributed, with further analysis using the Dunn test with a Bonferroni correction. Results with p values of ≤ 0.05 were considered statistically significant.

Results

Bacterial community diversity and composition

A total of 2,139,282 reads were recovered from the 16S V4-V5 region of the 68 samples after paired-end alignment, quality filtering and deletion of chimeric sequences. Four samples from farming site D (chinensis samples 3, 4, 5 and rape sample 3) were excluded from the analysis during filtering as these only matched to plant chloroplast and mitochondrial DNA. The 64 remaining samples were subsequently assigned to a total of 2249 amplicon sequence variants (ASVs). Taxonomic composition of bacteria from the 64 samples showed approximately 28 phyla (excluding unidentified and unassigned taxa) composed of 67 classes, 176 orders, 243 families, 298 families, 566 bacterial genera and 260 bacterial species.

The α-diversity analysis of irrigation water (Fig. 1) showed a significant difference in bacterial richness, Shannon diversity and Pielou’s evenness within samples when analysed by farming site, with p = 0.015, p = 0.003 (ANOVA) and p = 0.029 (Kruskal Wallace), respectively. A post hoc TukeyHSD test further highlighted significant community diversity difference for bacterial richness between farming sites A–D (p = 0.022), A–E (p = 0.014), B–E (p = 0.032), C–D (p = 0.016) and C–E (p = 0.011), while the Shannon diversity showed significantly different community compositions between farming sites A–D (p = 0.030), A–E (p = 0.005) and B–E (p = 0.015). Irrigation water source was also determined to be a significant contributor to bacterial richness (ANOVA: p = 0.003) and H (ANOVA: p = 0.018), with the sampling area being significant only for bacterial richness (ANOVA: p = 0.049). The bacterial community showed distinct clustering when evaluated against area (PERMANOVA: R² = 0.145, p = 0.012), farming site (PERMANOVA: R² = 0.523, p = 0.001) and irrigation water source (PERMANOVA: R² = 0.171 p = 0.01,) (Fig. 2).

Fig. 1

Diversity indices of irrigation water and fresh produce based on the area (Brits, Delmas), farming site (A–E) and source (flooding, overhead irrigation, chinensis, rape)

Fig. 2

Non-metric multidimensional scaling (nMDS) ordinations of bacterial communities from irrigation water and fresh produce samples assessed on area (Brits, Delmas), farming site (A–E) and source (flooding, overhead irrigation, chinensis, rape)

Overall, the observed alpha bacterial diversity in chinensis and rape (morogo) showed a significant difference in community composition based on area and farming site for ASV richness (Kruskal–Wallis rank sum test: area, p = 9.737 × 10^–4; farming site, p = 1.161 × 10^–5), Shannon diversity (H, Kruskal–Wallis rank sum test: area, p = 2.006 × 10^–4; farming site, p = 6.736 × 10^–7) and Pielou’s evenness (J, Kruskal–Wallis rank sum test: p = 1.945 × 10^–3; farming site, p = 2.882 × 10^–5) (Fig. 1).

The post hoc Dunn test showed that farming sites A–D (richness: p = 0.017), A–E (richness: p = 0.005), C–E (richness: p = 0.0004; H: p = 0.009), D–E (richness: p = 0.000; H: p = 0.0003) demonstrated the most significant differences between community composition for the measured beta diversity indices. No significant differences in community diversity were observed for chinensis and rape. Non-metric multidimensional scaling (nMDS) ordinations of the fresh produce bacterial communities showed distinct and significant separation for the farming site (PERMANOVA: R² = 0.276, p = 0.001; PERMDISP: F = 3.518, p = 0.016) (Fig. 2), however, no evidence of bacterial clustering by area or source was evident, despite being statistically significant in their composition (PERMANOVA: area, R² = 0.049, p = 0.002; source, R² = 0.045, p = 0.003,). It should be noted that no statistical differences between the dispersions were present for area (PERMDISP: F = 0.000, p = 0.997) and source (PERMDISP: F = 0.258, p = 0.589).

Bacterial community composition of irrigation water and morogo

The predominant phyla observed in all samples (irrigation water, morogo) were Proteobacteria (69.68% relative abundance), Firmicutes (16.47% relative abundance), Actinobacteria (6.06% relative abundance), Bacteroidetes (2.95% relative abundance), Verrucomicrobia (0.75% relative abundance), Cyanobacteria (0.43% relative abundance), Chloroflexi (0.27% relative abundance), Planctomycetes (0.27% relative abundance) and Gemmatimonadetes (0.18% relative abundance) (Fig. 3). Elusimicrobia, Latescibacteria and Nitrospinae were exclusive to farming site B’s irrigation water while Fibrobacteres, Bacteriodetes and Planctomycetes (FBP) and Spirochaetes were only exclusive to farming site C’s irrigation water. Chlamydiae, Dependentiae, Elusimicrobia, Epsilonbacteraeota, FBP, Fibrobacteres, Kiritimatiellaeota, Latescibacteria, Nitrospinae, Patescibacteria and Spirochaetes were only prevalent in irrigation water at all farming sites (Fig. 3). Nitrospirae, Rokubacteria and Tenericutes were only prevalent in irrigation water and rape, while Armatimonadetes were only prevalent in irrigation water and chinensis (Fig. 3). Further analysis of relative abundances at family level in all samples revealed Burkholderiaceae (48.00% relative abundance), Enterobacteriaceae (34.00% relative abundance), Bacillales Family XII (8.00% relative abundance), Rhodobacteraceae (3.00% relative abundance), Micrococcaceae (1.98% relative abundance), Pseudomonadaceae (1.79% relative abundance), Bacillaceae (1.33% relative abundance) and Moxaxellaceae (0.43% relative abundance) as the most dominant bacterial families.

Fig. 3

Relative abundance of different bacterial phyla detected on chinensis, rape and irrigation water samples from informal morogo small-scale farms (farming site A–E) in South Africa

Core analysis was performed at 100% similarity across all irrigation water and morogo samples (Additional file 1: Table S2). A single Micrococcaceae ASV was a common taxon in chinensis samples (15/22) and flooding irrigation water (6/6), while Exiguobacterium was the only common taxon in rape samples (19/24). Flooding irrigation water exhibited a unique core community dominated by 23 taxa (Additional file 1: Table S2), with Burkholderiaceae being the highest abundant of those present, while overhead irrigation water showed a common seven taxa, with Ralstonia (10/12), Staphylococcus (9/12) and Escherichia-Shigella (10/12) as the highest contributing members.

Thirteen species within the Enterobacteriaceae family were identified in the overall community analysis: Serratia (0.25% relative abundance), Raoultella (0.06% relative abundance), Enterobacter (0.27% relative abundance), Salmonella (4.34% relative abundance), Erwinia (0.56% relative abundance), Shigella (0.06% relative abundance), E. coli (26.48% relative abundance), Buchnera (1.60% relative abundance), Citrobacter (23.00% relative abundance), Klebsiella (3.57% relative abundance), Cedecea (2.11% relative abundance), Pantoea (37.48% relative abundance) and Proteus (0.06% relative abundance). Salmonella and E. coli were most prevalent in irrigation water at all farming sites. Overall, Raoultella, Shigella and Proteus were identified in irrigation water samples, while Erwina, Buchnera, Citrobacter and Cedecea were identified in fresh produce.

Upon closer inspection of the ASVs at each farming site, farming site A showed distinctive Enterobacteriaceae species in irrigation water (Salmonella and E. coli). This is in contrast to fresh produce, which showed a dominance of Enterobacter, Erwinia, Buchnera and Pantoea species. In farming site B, Serratia, Salmonella, E. coli and Klebsiella were identified in both irrigation water and fresh produce, while Raoultella was identified in irrigation water only and Erwinia, Buchnera, Cedecea and Pantoea were identified in fresh produce. Farming site C harboured distinctive Enterobacteriaceae species Serratia, Salmonella, E. coli and Klebsiella in irrigation water (Fig. 4). This contrasts with fresh produce, which showed high abundances of Buchnera, Cedecea and Pantoea. In farming site D, E. coli was identified in both irrigation water and fresh produce, while Raoultella, Salmonella, Shigella, Klebsiella and Pantoea were identified in irrigation water. In farming site E, Serratia, Salmonella, Enterobacter, E. coli and Klebsiella were observed in both fresh produce and irrigation water; Shigella and Proteus were present in irrigation water only, and Erwina, Buchnera, Citrobacter, Cedecea and Pantoea were identified in fresh produce only.

Fig. 4

Enterobacteriaceae genera were identified in irrigation water and leafy green vegetables in the five small-scale farms

Discussion

In this study, we used high-throughput Illumina sequencing to characterize the microbial community of irrigation water and fresh produce from regional small-scale farming systems in South Africa. We were further able to identify potentially pathogenic bacterial ASVs from small-scale farming systems and informal settings. Overall, we were able to identify a highly diverse bacterial community that includes copiotrophic phyla such as Proteobacteria, Firmicutes, Bacteriodetes, and Actinobacteria, which are normally detected on other indigenous African plant and crop species (Berg et al. 2016; Bulgarelli et al. 2013; Ho et al. 2017; Taffner et al. 2020). At the genus level, it was evident that both fresh produce and irrigation water harboured unique bacterial communities which may be driven by specific adaptations to distinct environmental factors in the farming area and farming activities at each farming site as well as plant species specificity.

A core community was identified for each source, demonstrating that indigenous leafy green vegetables such as chinensis and rape do not harbour a core microbiome, which is in complete contrast to the irrigation water sources. Flooding irrigation water appeared to possess a core microbiome that consisted of common environmental bacteria isolated from irrigation and soil. However, it is in close proximity to soil, which harbours a complex and dynamic bacterial community (Mäder et al. 2002). Ralstonia and Escherichia-Shigella are amongst the highest contributing core members in overhead irrigation water. Previously reported disease outbreak-associated bacteria in South Africa were matched to genera dominating the bacterial families identified in this study (Additional file 1: Table S3) and include Clostridiaceae and Enterococcaceae known to include genera that indicate human or animal faecal contamination.

In this pioneering study, we identified 28 different phyla, excluding those that were unidentified or unassigned in both fresh produce and irrigation water. Our results correspond with Gu et al. (2019) that shows Proteobacteria, Actinobacteria, Bacteroidetes and Firmicutes are the dominant phyla in irrigation water, regardless of the source (groundwater, wastewater and roof-harvested rainwater). In the present study, we additionally identified Firmicutes, which mostly dominates fresh produce (Leff et al. 2013), and which is a phylum prominent in irrigation water in small-scale farming systems. Leafy green vegetables such as spinach and lettuce are dominated by Proteobacteria, Actinobacteria, Firmicutes and Bacteroidetes (Berg et al. 2014; Leff et al. 2013). Morogo (rape and chinensis) show homologous bacterial communities differing in abundance, complementary to previous studies on lettuce and spinach (Jackson et al. 2015; Leff et al. 2013).

The high diversity observed in all farming sites, production areas and plant sources show the potential for these microbial communities to contain bacteria which have previously been associated with foodborne illnesses. Nevertheless, analysis of the community structure at all sampling points in this study showed the presence of Enterobacteriaceae, known to be associated with vegetables (Berg et al. 2014) and some members are potential foodborne pathogens (Janda and Abbott 2021). In this study, faecal indicators and potential foodborne pathogens such as Enterobacter, E. coli, Klebsiella, Pantoea, Salmonella, Serratia and Shigella were also prevalent.

It is important to take note that flooding water had higher bacterial diversity than overhead irrigation water. The microbiota of crops are directly affected by irrigation application methods (Rodrigues et al. 2020, Banach and van der Fels-Klerx 2020), therefore, contaminated irrigation water could lead to morogo contamination. This is especially relevant when considering the large number of reports associated with foodborne illness outbreaks involving consumption of leafy green vegetables (Browne et al. 1962; Jaklevic 2021; Marshall et al. 2020). This highlights the need for extensions in surveillance of foodborne pathogens in the production chain of small-scale farming environments. Further to this, the World Health Organisation (WHO) lists genera and species of emergent Enterobacteriaceae as potential superbugs (WHO 2017) due to the presence of extended-spectrum beta-lactamases (ESBLs). These bacteria are of clinical importance due to its health care impact rendering antibiotics ineffective (Teklu et al. 2019). In culture-dependent studies performed in parallel with this study, ESBLs or AmpC-producing Enterobacteriaceae were isolated, including E. coli, Serratia fonticola, Klebsiella pneumoniae, Enterobacter spp., Citrobacter spp., Rahnella aquatilis, Cedecea davisae and Proteus vulgaris (Du Plessis et al. 2021). Furthermore, multi-virulent and multidrug resistant S. enterica strains were isolated from source and irrigation water and fresh produce (Du Plessis et al. 2021). These multidrug resistant bacteria are natural in the environment, however, their presence is indicative of human or animal faecal contamination (Delahoy et al. 2018). These microbes are a potential threat to consumers, especially if the leafy green vegetables are consumed unwashed, eaten raw or not cooked. Future studies should focus on the prevalence of foodborne pathogen surveillance systems in unregulated farming systems and informal systems.

Conclusion

In this study we have presented for the first time, a diverse and robust bacterial microbiome profile of irrigation water and indigenous leafy green vegetables produced in small-scale farming systems and sold in informal settings. The current study provides an important framework for further research into the microbial communities of small-scale farming systems and the informal supply chains and its link with public health. Mining this data permitted the detection of potential foodborne pathogen ASVs that arise from the microbial assemblages in irrigation water of small-scale farming systems. The presence of Enterobacteriaceae as well as other putative pathogens commonly associated with human illness in local farming systems demonstrates that irrigation water and morogo cannot be ruled out as potential sources of foodborne pathogens and antimicrobial resistance. From this, plant-associated bacteria play a crucial role in the host microbiome and human health, especially when the product is consumed raw. The maintenance and support of microbial diversity are of key importance to stabilise agroecosystems. In addition, an extension of molecular surveillance in small-scale informal supply chains will be an integral component to establish a "fit-for-purpose" food safety management system to reduce the likelihood of foodborne disease outbreaks.