Testing Different Membrane Filters for 16S rRNA Gene-Based Metabarcoding in Karstic Springs

Abstract

Introduction: Karstic springs are used worldwide by rural communities as sources of fresh water for humans and livestock. In Romania, one-third of the population has no direct access to a public water supply. The present study is part of a country-wide project to develop simple, quick and cheap methods for seasonal environmental and microbiological monitoring of karstic springs used as drinking water by rural populations. Critical steps for monitoring workflow consist of evaluating water quality and selecting suitable membrane filters to efficiently capture environmental DNA for further microbial diversity estimation using 16S rRNA gene-based metabarcoding. Methods: Several commercial membrane filters of different compositions and pore sizes were tested on the water sampled from three karstic springs in Romania, followed by water chemistry and whole community 16S rRNA gene-based metabarcoding analysis. Results: We found that different types of applied membrane filters provide varying recovery in diversity and abundance of both overall and pathogenic bacteria. Conclusions: The result of the experiment with different filters shows that mixed cellulose ester, cellulose acetate, and nitrate membranes of 0.20 and 0.22 µm are the best for amplicon-based metabarcoding monitoring of karst springs.

1. Introduction

A recent review [1] considered the main environmental drivers for 21st century global trends regarding freshwater availability. Changes in freshwater availability not only impact regional food supplies, human and ecosystem health and energy generation but also cause social unrest [1]. Half of the world’s population use groundwater for domestic needs [2], and groundwater provides 38% of the global consumptive irrigation water demand [3]. Yet groundwater monitoring and management are challenging tasks for the vastness and unseen distribution of these freshwater sources, mainly found in soluble rocks such as limestones and dolomites. The high demand for food, in conjunction with climate change, has resulted in economic and social hardships already reported in many parts of the world and are foreseen to accelerate the depletion of available groundwater sources used for drinking and agriculture [4].

The 2006–2017 Eurostat survey (published in 2020, [5]) shows that 5 out of 12 European countries have less than 90% of households connected to a public water supply, with Romania ranked at the lowest percentage of 67.5%. One cause of this shortcoming might be the sparse distribution of houses in most rural areas that makes access to a public water supply problematic if not impossible. Small rural populations use local sources of water that are not under the monitoring programs of water agencies. Usually, such small local sources that come from porous karstic rocks are prone to periods of low or no flow or a combination of low flow and a concentration of contaminants [6]. The primary sources of contamination in small rural localities are from livestock, septic tanks, mining activities in the area and the use of chemicals in agricultural practices [7].

Environmental DNA (eDNA) metabarcoding was proposed as a viable alternative for freshwater quality assessment [8] as it quickly and sensitively reveals the taxonomic inventory of eukaryotic [9,10] or prokaryotic communities [11]. Recently, eDNA metabarcoding has proven to be a useful tool to detect waterborne pathogens in rural surface freshwater impacted by faecal pollution [12] and antibiotic-contaminated groundwater [13]. Surprisingly, little information exists on the microbiological quality of groundwater sources that are used as drinking water in rural areas worldwide. In most, if not all, of the cases, the detection of Escherichia coli as a representative of coliform bacteria is the only microbiological test to check the potential for waterborne diseases [14]. This approach is unable to detect other dangerous bacteria, including pathogenic or opportunistic species bearing antibiotic resistance traits [15]. In this light, eDNA metabarcoding appears to be a novel and feasible approach to assess the microbial diversity of groundwater sources for drinking water in rural areas.

For rural populations that use spring water as their main source of fresh water, our purpose was to test and propose tools and protocols for seasonal water quality monitoring [16]. Within the framework of this country-wide project, we performed a survey on the physicochemical and microbiological characteristics of selected karst springs to establish a fast and cost-effective monitoring protocol. We aimed to evaluate the most suitable membrane filter type and filtration strategy to efficiently recover environmental DNA (eDNA) for further fast and accurate whole community diversity assessment using 16S rRNA gene-based metabarcoding.

2. Materials and Methods

2.1. Sampling and Sample Preparation

Three springs were chosen for their ease of access from Cluj-Napoca (Romania), the headquarters of logistics and the authors. Banpotoc, Băița (Baita) and Rapolțel (Rapoltel) springs are located in the Apuseni Mountains in northwestern Romania (Figure 1). These springs are the only sources of water used by the local populations as drinking water resources for human and livestock consumption and are not included in the national or local water monitoring programs.

In September 2019, samples of spring water were collected from each of the three springs in sterile polyethylene bottles. Bottles were rinsed three times with the sample water and kept refrigerated at 4 °C until chemical analysis. Five liters of water were filtered under aseptic conditions using a vacuum pump and a filtering device, composed of a glass filter holder and a funnel, attached to a collecting Kitasato flask. Nine different types of membrane filters with different pore sizes were tested for their efficiency in retaining biomass that sourced for eDNA from karstic spring water (

Table 1. The tested membrane filters for metabarcoding in karstic spring water.

Membrane Code	Fabric	Producer	Porosity (µm)	Sterile *	Price **
PES	polyethersulfone	Millipore	0.22	sterilized	High
PVDF	polyvinylidene fluoride	Millipore	0.22	sterilized	High
NC	cellulose nitrate	Macherey-Nagel	0.20	sterilized	Low
NYLON	nylon	Fioroni	0.22	sterilized	Low
MCE_a	mixed cellulose ester	Fioroni	0.22	sterile	Low
MCE_b	mixed cellulose ester	Whatman	0.20	sterile	Medium
S-Pak	mixed cellulose ester	Merck	0.20	sterile	Low
PSR009	cellulose acetate and nitrate	Nahita	0.22	sterilized	Low
PSR010	cellulose acetate and nitrate	Nahita	0.45	sterilized	Low

* We termed ‘sterile’ those membrane filters provided sterile by the manufacturer and ‘sterilized’ those membrane filters provided non-sterile and further steam-sterilized at 121 °C for 20 min.** We define the prices as low, medium and high, relative to each other.

). After filtration, each membrane was cut into small pieces and kept in a sterile Petri dish at −20 °C until eDNA extraction.

2.2. Physicochemical Measurements and Laboratory Analysis

The pH and electrical conductivity (EC) were measured in situ using a portable multiparameter monitoring device with built-in temperature correction (Multi 340i, WTW, Germany).

In the laboratory, dissolved organic carbon (DOC) was obtained by subtracting DIC from DC. Dissolved inorganic carbon (DIC) and dissolved carbon (DC) were determined after filtering the samples on 0.45 μm PTFE filters via catalytic combustion and infrared detection of CO₂ using a Multi N/C 2100S Analyzer (Analytik Jena, Germany). Total alkalinity was determined via titration with HCl. The bicarbonate (HCO₃⁻) content was calculated by multiplying the alkalinity expressed in mmol/L by 61. Ammonia was determined with UV-Vis spectrometry (Lambda 25, Perkin Elmer, Waltham, MA, USA) using the salicylate method. Anions were determined with ion chromatography using a 761 Compact IC (Metrohm, Switzerland). The metal concentrations were measured in samples filtered through 0.45 mm cellulose acetate filters and acidulated with 63% HNO₃ using inductively coupled plasma optical emission spectrometry (ICP-OES, Optima 5300 DV, Perkin-Elmer, Waltham, MA, USA) for Na, Mg, Ca and K and inductively coupled plasma mass spectrometry (ICP-MS, ELAN DRC II, Perkin-Elmer, Waltham, MA, USA) for Ba, Sr, Mn, Ni, and Cr.

2.3. 16S rRNA Protocol

MiSeq 16S V3-V4 Metagenome Sequencing was performed using a commercial company (Macrogen Europe). The V3-V4 hypervariable regions of the bacterial and archaeal SSU rRNA gene were amplified via PCR using primers 341F (5′-CCTACGGGNGGCWGCAG-3) and 805R (5′-GACTACHVGGGTATCTAATCC-3′), according to Illumina’s 16S amplicon-based metagenomic sequencing protocol. The amplicon size achieved was between 400 and 500 bp. The pair-end reads were joined in FLASH (1.2.11) [17]. Pre-processing and clustering were performed in CD-HIT-OTU and rDnaTools [18,19]. The taxonomy was assigned for the representative operational taxonomic units (OTUs) in QIIME against the last version of the RDP–16Sr DNA database and using a species-level OTU cut-off at 97% sequence identity [20].

2.4. Statistical Analysis

Principal component analysis (PCA) was conducted with XLSTAT 2020.1.3 (Addinsoft, France). PCA is a multivariate data analysis method in which observations (springs) are described by variables (chemical elements).

The diversity indices combine two independent attributes of communities: the number of species and their relative abundances. Chao1 is a diversity index based on the abundance of individuals (OTUs) belonging to a sample. The Shannon diversity index (H) is commonly used to characterize species diversity, in both abundance and evenness, of those present in a community.

3. Results

3.1. Chemical Analysis

The chemical data quality was assessed by calculating the charge balance between the sum of cations and the sum of anions, expressed in mEq/L. In each case, the charge balance error (CBE) was less than 10% (

Table 2. The physicochemical characteristics of the three studied karstic springs.

Parameter	Unit	Spring
		Banpotoc	Baita	Rapoltel
TDS	mg/L	1210	298	987
DOC	mg/L	17.7	1.4	19.3
pH	-	6.3	7.6	6.5
Bicarbonates	mg/L	1122	256	988
Alkalinity	mmol/L	18.4	4.2	16.2
Na	mg/L	28.5	6	8.1
Ca	mg/L	280	72.9	219
Mg	mg/L	37	4.3	33.9
K	mg/L	5.7	0.45	2.4
Ba	µg/L	671	13.6	109
Sr	µg/L	351	92.7	240
Mn	µg/L	74.6	1.5	25.7
Ni	µg/L	13.6	5.7	13.4
Cr	µg/L	1.48	2.61	1.41
NH4+	mg/L	0.57	0.12	0.04
Cl−	mg/L	4.53	2.55	2.2
SO42−	mg/L	2.2	19.6	6
NO3−	mg/L	<0.2	2.4	<0.2
NO2−	mg/L	0.32	<0.05	<0.05

).

The Piper diagram (Figure S1) shows that Banpotoc, Baita and Rapoltel springs have a Ca-HCO₃ facies. The PCA of the physicochemical variables (Figure 2) separated Baita and Banpotoc along the horizontal axis (F1). The F1 axis explains more than 83.6% of the variation in the analysis. Rapoltel separated from the two other springs along the F2 axis at lower statistical significance (16.4%). The total dissolved solids (TDS) were low in Baita and high in the other two springs, indicating the high mineralization of the latter springs. The Banpotoc and Rapoltel springs were rich in Ca and bicarbonates and had high alkalinity. Ba, Sr, Mn, and Ni were also high in Banpotoc and Rapoltel. Sulfates were high in the Baita spring while all the other elements were found in low concentrations. DOC was one order of magnitude higher in Banpotoc and Rapoltel than in the Baita spring.

In the Romanian legislation regulating the quality of drinking water [21], the maximum allowable concentration (MAC) for ammonia is 0.5 mg/L. This value was slightly exceeded in Banpotoc, possibly indicating the leaching of nitrates from agricultural land use.

3.2. Microbial Diversity Inferred by 16s RNA Gene-Based Metabarcoding

After quality-filtering and operational taxonomic unit (OTU)-clustering at the 97% sequence identity, we obtained 1,025,064 good-quality reads grouped into 1106 OTUs following singleton removal. OTUs pertaining to Archaea and Bacteria were identified, with Bacteria-related sequences dominating in all samples.

The identified taxa show high variations of diversity and abundances between the 28 analyzed filter samples (

Table 3. Collected samples in the three karstic springs with their corresponding diversity indices.

Spring	Code	Fabric	Membrane Code	OTUs	Diversity Indices
					Chao1	Shannon
Banpotoc	AE20	polyethersulfone	PES	186	188.6250	2.8280
	AE21	polyvinylidene fluoride	PVDF	226	264.2778	2.5738
	AE22	cellulose nitrate	NC	228	306.9643	2.6665
	AE23	nylon	NYLON	264	294.3571	2.7375
	AE24	mixed cellulose ester	MCE_a	251	318.0000	2.7688
	AE25	mixed cellulose ester	MCE_b	241	257.5000	2.8318
	AE26	mixed cellulose ester	S-Pak	261	320.2941	2.9322
	AE4	cellulose acetate and nitrate	PSR009	263	267.5000	4.9879
	AE27	cellulose acetate and nitrate	PSR010	115	126.5500	1.2992
Baita	BE11	polyethersulfone	PES	1280	1699.2250	6.4304
	BE12	polyvinylidene fluoride	PVDF	1335	1679.3142	6.8048
	BE18	cellulose nitrate	NC	1368	1652.6980	6.6819
	BE16	nylon	NYLON	1538	1803.8583	7.1422
	BE15	mixed cellulose ester	MCE_a	1271	1569.4933	7.0039
	BE13	mixed cellulose ester	MCE_b	1566	1734.7447	8.5057
	BE19	mixed cellulose ester	S-Pak	1455	1752.9286	7.9813
	BE17	cellulose acetate and nitrate	PSR009	794	1009.5301	4.9305
	BE14	cellulose acetate and nitrate	PSR010	718	915.3356	4.3866
Rapoltel	CE29	polyethersulfone	PES	200	202.0000	4.8509
	CE31	polyvinylidene fluoride	PVDF	221	224.7500	5.1441
	CE32	cellulose nitrate	NC	235	242.2000	4.4465
	CE30	nylon	NYLON	174	174.6000	3.9005
	CE34	mixed cellulose ester	MCE_a	214	217.3333	4.1399
	CE33	mixed cellulose ester	MCE_b	358	360.5000	5.0208
	CE2	mixed cellulose ester	S-Pak	310	312.5000	5.7927
	CE28	cellulose acetate and nitrate	PSR009	260	265.1429	5.1996
	CE35	cellulose acetate and nitrate	PSR009	255	255.4286	5.0427
	CE3	cellulose acetate and nitrate	PSR010	168	169.5000	3.8749

). The highest diversity was in the Baita samples followed by the Rapoltel and Banpotoc samples. AE23 (nylon) and AE4 (cellulose acetate and nitrate) from Banpotoc, BE13 and BE16 from Baita (mixed cellulose ester and nylon, respectively) and CE33 (mixed cellulose ester) from Rapoltel were the most diverse.

Most Archaea-related OTUs (16 in total) belonged to Euryarchaeota, with Methanolobus taylorii, Methanobacterium aggregans and M. palustre as the most frequently encountered species. Crenarchaeota and Thaumarchaeota were also represented in one to four samples, depending on the membrane type. Sample AE25 from Bantopotoc (mixed cellulose ester) was the richest in Archaea-related OTUs.

The identified Bacteria-related OTUs were classified within 34 phyla, where the most abundant were Proteobacteria, Bacteroidetes, Actinobacteria and Firmicutes (Figure 3, Table S1). The other 11.54% of the total OTUs belonged to non-identified phyla. Proteobacteria dominated in all the samples, with Epsilonproteobacteria as the main class in Banpotoc, and Betaproteobacteria as the dominant class in Baita and Rapoltel (Figure 4). In Banpotoc, Actinobacteria was the next diverse phylum, while Bacteroidetes dominated Baita samples. Baita also had the highest diversity in phyla. Rapoltel samples had no other dominant group except for Proteobacteria, by far the most diverse in this spring. Only CE28 had high OTU abundance belonging to Firmicutes phylum.

In a PCA of diversity indices (Figure 5), the samples from Baita separated along the F1 axis (90.85% variance) from Banpotoc and Rapoltel, while these last two separated from each other along the F2 axis (9.05% variance). Banpotoc membranes are more similar to each other and so are the Rapoltel membranes, except for AE4 (0.22 µm porosity) and CE3 (0.45 µm porosity), both consisting of cellulose acetate and nitrate. The Baita membranes are more different from each other. Along the F1 axis, BE17 and BE14 (both cellulose acetate and nitrate membranes, with different porosities) have a relatively separate position from the other Baita membranes; their position is explained by the analysis at the species level (see below).

To infer the efficiency of different filtrating membranes, the results were further analyzed separately for each spring at the species level for more evident differences. The relative abundances of the dominant species are represented in Figure S2. The dominant species in Banpotoc samples were Sulfurovum lithotriphicum and Sulfurimonas autotrophica, both involved in the redox sulfur cycle (Figure 6a). They represented 68.5% of the identified Banpotoc total OTUs. Flavobacterium succinicans and Rhodoferax saidenbachensis represented more than 34% of the identified Baita OTUs (Figure 6b). Sideroxydans lithotrophicus was by far the most abundant species in all Rapoltel samples, representing more than 43% of the total identified OTUs (Figure 6c). The AE4 (0.22 µm) and AE27 (0.45 µm) samples, both cellulose acetate and nitrate membranes, provided different results than the other membranes of Banpotoc, all abundant in Sulfurimonas autotrophica and Sulphurovum lithotrophicum. The AE4 membrane had Acinetobacter junii as the dominant taxa, while AE27 was almost entirely represented by Sulphurovum lithotrophicum. In Baita, two species were dominant, Flavobacterium succinicans and Rhodoferax saidenbachensis, except for the BE13 and BE19 samples, which were both mixed cellulose esters and where no species was particularly dominant. All Rapoltel samples, except for CE28 (0.22 µm cellulose acetate and nitrate), were dominated by Sideroxydans lithotrophicus. CE28 was dominated by Staphylococcus epidermis.

We also checked for the presence of pathogenic bacteria in the different samples (

Table 4. The tested membranes for metabarcoding and the most abundant identified pathogenic bacteria.

Spring	Sample Code	Membrane Code	Biosafety Risk Species for Humans & Animals
			100–500 reads/L	501–1000 reads/L	>1001 reads/L
Banpotoc	AE20	PES
	AE21	PVDF
	AE22	NC
	AE23	NYLON
	AE24	MCE_a			Acinetobacter junii
	AE25	MCE_b
	AE26	S-Pak
	AE4	PSR009	Comamonas testosteroni Enterobacter kobei
	AE27	PSR010
Baita	BE11	PES
	BE12	PVDF
	BE18	NC
	BE16	NYLON
	BE15	MCE_a
	BE13	MCE_b
	BE19	S-Pak
	BE17	PSR009
	BE14	PSR010
Rapolțel	CE29	PES	Staphylococcus epidermidis		Acinetobacter junii
	CE31	PVDF	Acinetobacter junii Escherichiafergusonii	Staphylococcus epidermidis
	CE32	NC	Acinetobacter junii
	CE30	NYLON	Acinetobacter junii
	CE34	MCE_a	Acinetobacter junii
	CE33	MCE_b	Acinetobacter junii
	CE2	S-Pak
	CE28	PSR009	Atopobium vaginae Corynebacterium tuberculostearicum Gardnerella vaginalis Moraxella osloensis		Acinetobacter junii Staphylococcus epidermidis
	CE35	PSR009	Acinetobacter junii Veillonella dispar
	CE3	PSR010	Acinetobacter junii

, Table S2). Baita had no pathogens, while the other two springs had several representatives of known pathogenic bacterial species. The cellulose acetate and nitrate of 0.22 µm and (PSR009) and mixed cellulose ester (MCE_a) membranes were the only ones that retained pathogenic bacteria in Banpotoc. In Rapoltel, all filters, except for the mixed cellulose ester S-Pak type, had pathogens, represented by one or more species. The PSR009 membrane of Rapoltel had the higher number of identified pathogenic species, although there was a difference between the two replicates as CE28 had six species and CE35 had only two (Table 4). Acinetobacter junii and Staphylococcus epidermidis were the pathogenic bacteria found in very high abundances in the Banpotoc and Rapoltel springs.

4. Discussion

Groundwater geochemistry, and that of their surface outlets represented by springs, is regulated by flow paths and groundwater residence time [22,23]. The rocks and deposits crossed underground by the waterway, the length of the underground passages, the speed of transit and the quality of the surface infiltrations shape the chemical composition of karstic spring waters. Banpotoc and Rapoltel waters cross limestone layers, explaining the high concentration of carbonates. The fast flow of groundwater through the limestone levels explains the low mineralization of the Baita spring. Ammonia in Banpotoc slightly exceeded the drinking water quality values [21]. The sources of the nitrogen compounds could be agricultural activities (use of soil fertilizers based on nitrogen), household activities, leaching from septic tanks, and animal debris.

Microbial organisms in groundwater depend mostly on oxidation and reduction of inorganic compounds for energy in this low-nutrient environment [24]. Thus, electron donors and acceptors of water and geological substratum become the major drivers of microbial communities’ composition and diversity [25,26,27,28,29]. Microbial communities can change drastically when contamination is originating from the surface [30,31,32].

In our experiment on the efficiency of different filter membrane types for 16S rRNA metabarcoding, the chemistry was important in shaping the diversity of microorganisms and pathogenic bacteria. Different filtrating membranes provided various OTU diversity in the three springs. Nylon, mixed cellulose ester and cellulose acetate and nitrate (Nylon, MCE-b, S-Pak and PSR009) membranes were the best filters for diversity.

Table 5. The comparative analysis of the different membranes for 16S metabarcoding of karstic springs for diversity and quality analysis. The selected membranes have a porosity of 0.2–0.22 µm.

Spring/ Membrane	Pathogens Identification	Diversity Assessment	Proposed Membrane for Pathogens	Proposed Membrane for Diversity
Banpotoc	cellulose actetate and nitrate (PSR009)	nylon (Nylon)	cellulose actetate and nitrate (PSR009)	S-Pak, cellulose actetate and nitrate (PSR009)
	mixed cellulose ester (MCE_a)	cellulose actetate and nitrate (PSR009)
		mixed cellulose ester (S-Pak)
Baita		mixed cellulose ester (MCE-B)
		nylon (Nylon)
		mixed cellulose ester (S-Pak)
Rapoltel	cellulose actetate and nitrate (PSR009)	mixed cellulose ester (MCE_b)
	polyvinylidene fluoride (PVDF)	mixed cellulose ester (S-Pak)
	polyethersulfone (PES)	cellulose actetate and nitrate (PSR009)

contains the selection of the best filtering membranes for an estimation of the OTU diversity/abundance and presence of pathogens obtained through metabarcoding. We expected to have similar results with the same filter(s) on different spring waters. However, the same filtering membrane gave various results when used on water from springs which are chemically different.

The filtering membranes also performed differently when pathogen identification was considered. However, for the efficiency of different filtering membranes in detecting pathogenic bacteria, most of the species were obtained by filtering water with the cellulose acetate and nitrate (PSR009) membrane. For Banpotoc samples, cellulose acetate and nitrate and mixed cellulose ester (PSR009 and MCE_a) membranes provided good results. Rapoltel samples provided more diverse and abundant pathogens with the cellulose acetate and nitrate (PSR009) and also with the polyvinylidene fluoride (PVDF) and polyethersulfone (PES) membranes. In the three studied springs, all the identified pathogenic bacteria had low or moderate pathogenic risks, and were unlikely to cause human or animal diseases. None of the pathogens classified in high risk groups that can cause serious diseases were found in our samples.

No previous studies were undertaken on the chemical and microbiological composition of the three springs chosen in the present study. We do not know if the ammonia contamination in Banpotoc and the microbiological contamination in Rapoltel are recent or not. However, the presence of pathogenic bacteria, sometimes in high number, in two of the springs (Banpotoc and Rapoltel) that are used as sources of drinking water calls for further quality monitoring. We will continue the monitoring program on these three springs and other springs in Romania and will use the selected filtering membranes as specified in Table 5.

5. Conclusions

For the karstic springs’ quality assessment, we proposed a preliminary evaluation of 16S rRNA diversity and abundance by testing different membranes for water filtering. The testing of different sterile or sterilizable commercial membranes was a prerequisite in establishing monitoring protocols. We based our decision in choosing the best filtering membrane on their efficiency in retaining microorganisms and pathogenic bacteria and price. As a result of the experiment, we proposed two membranes for diversity and one for pathogenic bacteria. These three membranes will be used in the monitoring program of karstic springs used by rural populations as drinking water sources.