Harnessing microfluidic streak plate technique to investigate the gut microbiome of Reticulitermes chinensis

Abstract The termite gut microbiome is a model system to investigate microbial interactions and their associations with host. For decades, extensive research with molecular tools and conventional cultivation method has been carried out to define the microbial diversity in termite gut. Yet, many bacterial groups of the termite gut microbiome have not been successfully cultivated in laboratory. In this study, we adapted the recently developed microfluidic streak plate (MSP) technique for cultivation of termite gut microbial communities at both aerobic and anaerobic conditions. We found that 99 operational taxonomic units (OTUs) were cultivable by MSP approach and 18 OTUs were documented first time for termite gut microbiota. Further analysis of the bacterial diversities derived by culture‐dependent MSP approach and culture‐independent 16S rRNA gene typing revealed that both methods have bias in recovery of gut microbiota. In total 396 strains were isolated with MSP technique, and potential new taxa at species and/or genus levels were obtained that were phylogenetically related to Burkholderia, Micrococcus, and Dysgonomonas. Results from this study indicate that MSP technique is applicable for cultivating previously unknown and new microbial groups of termite gut microbiota.

Compared to culture-independent methods, the culturedependent method would better serve the purpose to investigate host-microbe interaction or to recover valuable microbial products (including commercial enzymes) (Keller & Zengler, 2004;Stewart, 2012). However, cultivation of microbes from various samples including termite gut is often hindered as many microbes in nature are resistant to be cultivated in laboratory conditions (Amann, Ludwig, & Schleifer, 1995;Hongoh, 2011;Ohkuma & Brune, 2011). To overcome this obstacle and to cultivate as yet not cultivated microorganisms in laboratory, techniques of high throughput and mimic natural conditions have been developed, such as the high-throughput culturing procedures that utilize the concept of extinction culturing (Colin, Goñiurriza, Caumette, & Guyoneaud, 2013;Colin, Goñi-Urriza, Caumette, & Guyoneaud, 2015;Connon & Giovannoni, 2002), the microencapsulation (Keller & Zengler, 2004;Zhou, Liu, Liu, Ma, & Su, 2008) and the isolation chip (Ichip) (Nichols et al., 2010). Microfluidic devices (Ma et al., 2014;Park, Kerner, Burns, & Lin, 2011;Tandogan, Abadian, Epstein, Aoi, & Goluch, 2014) were also developed for highly parallel cocultivation of symbiotic microbial communities and isolating pure bacterial cultures from samples containing multiple species. The microfluidic streak plate (MSP) technique (Jiang et al., 2016) exploits the advantages of microfluidics to manipulate tiny volume of liquid at several to hundred nanoliters and generate microdroplets for microbial single-cell isolation and cultivation. Superior to the conventional agar plate cultivation, the MSP approach enabled higher throughput of bacterial isolation and better coverage of rare species in community (Jiang et al., 2016).
Reticulitermes chinensis (Snyder) (Isoptera: Rhinotermitidae) is wood-feeding lower termite. In this study, we continued our efforts to cultivate microbes from the gut of from this termite (Chen, Wang, Hong, Yang, & Liu, 2012;Fang, Lv, Huang, Liu, & Yang, 2015;Fang et al., 2016), and adapted the MSP technique for cultivation of gut microbiome at both aerobic and anoxic conditions. With the MSP method, 99 OTUs representing Proteobacteria, Firmicutes, Actinobacteria, Bacteriodetes, Acidobacteria, and Verrucomicrobia were obtained, and 396 bacterial isolates were successfully cultivated in pure cultures. Our results demonstrated that MSP method significantly increased the recovery of various microbial groups and many of them were documented for the first time from termite gut.

| Termite cultivation and retrieving gut microbiota
The termite Reticulitermes chinensis colonies were collected and transferred to laboratory, and were maintained in glass containers on a diet of pinewood and water. Only worker termites were used in this study. The termite's surface was washed three times with 70% ethanol, rinsed with distilled water and blotted dry on sterilized filter papers. The guts from 40 termites were removed aseptically with fine-tipped forceps onto a sterilized glass slide and the gut microbiota were squeezed out of the guts and were transferred into a tube with 1mL of PBS buffer (PBS buffer, g/L: NaCl, 8.00; KCl, 0.20; Na 2 HPO 4 .12H 2 O, 3.58; KH 2 PO 4 , 0.24; pH 7.2). The gut microbiota suspension in the PBS buffer was used subsequently for cell separation and cultivation.

| Operation of microfluidic droplet arrays
Microfluidic streak plate (MSP) was operated according to previously described (Jiang et al., 2016), except that the automated dish driver and the microfluidic device were setup in an anaerobic chamber (ThermoScientific 1029). Droplets were arrayed onto surfacemodified Petri-dish (Jiang et al., 2016), and about 3000 droplets were displayed onto the surface of 9-cm Petri-dish. 2) was used as growth broth and for dilution of gut microbiota samples. In order to prepare samples for MSP, the gut microbiota suspension (see M&M section 1) was diluted with growth broth, either directly from the suspension or after three times washing with Cysteine-reduced (1 g/L) PBS buffer (pH 7.2). The final concentration of diluted gut microbiota suspension was approximately 1 × 10 4-5 cells/ml. This diluted suspension was used for separation and cultivation of the gut microbiota with the MSP method. Petri dishes with droplet arrays were incubated at 30°C under both aerobic and anaerobic condition. After 72 hr incubation, the droplets were individually transferred into 96-well cell-culture plates, each well contained 80 μL of 1/5 R2A medium. After another 72 hr of cultivation at 30°C, the growth of bacterial cells was monitored with a Microplate reader (Biotek SynergyHT). The grown cells were streaked on R2A agar plates, and all bacterial strains obtained were stored at 10°C in cold room until further tests. (5′-GGACTACHVGGGTWTCTAAT-3′) containing barcodes at the 5′ end of the front primer (Werner, Zhou, Caporaso, Knight, & Angenent, 2012). PCR reactions were proceeded in 50 μL volumes, each containing 1.5 μL of 10 μM forward and reverse prim- The full length of 16S rRNA gene from each bacterial strain obtained in this study was amplified with the 27F and 1492R primers (Edwards, Rogall, Blöcker, Emde, & Böttger, 1989;Weisburg, Bars, Pelletier, & Lane, 1991). The 16S rRNA gene sequences of the isolates in this study have been deposited in GenBank databases under the accession numbers MG984070-MG984092.

| 16S rRNA gene-based metagenomic analysis and phylogenetic tree construction
The raw sequences were assigned to individual samples by their unique barcodes. The 16S rDNA primers and barcodes were then removed to generate pair-end (PE) reads. Raw tags were then generated by merging PE reads with FLASH (Magoč & Salzberg, 2011), the raw tags were then filtered and analyzed using QIIME software package (Quantitative Insights Into Microbial Ecology) (Bokulich et al., 2013). Reads from all samples were quality filtered using an average quality value of 20 (Q20) during demultiplexing, sequences with a mean quality score 20 were excluded from analysis, and chimeras were also excluded. For species analysis, 16S rRNA sequences with ≥97% similarity were assigned to the same OTUs using Uparse v7.0.1001 (Edgar, 2013), and similarity hits below 97% were not considered for classification purpose. A representative sequence of each OTU was picked out and the taxonomic information was annotated using RDP classifier (version 2.2) (Wang, Garrity, Tiedje, & Cole, 2007) and GreenGene database (Desantis et al., 2006). Sequences obtained were compared with the published sequences in GenBank using Blast from NCBI (http://www.ncbi.nlm.nih.gov/BLAST).
The 16S rRNA sequences of all the published termite-gut-derived bacteria were mined from NCBI. The OTU sequences of MSP pool sample were blasted with the GenBank of NCBI and the 16S rRNA sequences of type species with the highest similarity to our OTUs were selected. Those sequences together with the extracted termite-gut-derived bacterial 16S rRNA gene sequences were used for the construction of phylogenetic tree. The OTUs from MSP pool samples (accession numbers MH152413-MH152511), the 16S rRNA gene sequences of isolated strains (accession numbers MG984070-MG984092) and the reference sequences (the accession number was available in phylogenetic tree) were aligned using ClustalW (Thompson, Gibson, & Higgins, 2002). Phylogenetic trees were constructed with MEGA6 package based on the alignments of sequences using Neighbor-joining method with p-distance. Bootstrap analysis with 1000 replicates was performed to determine the statistical significance of the branching order.

| Termite gut microbial community revealed with MSP technique and comparison to metagenomic method
We sequenced both the partial 16S RNA gene of the original microbiota from gut sample (hereafter called OMG sample) and DNA extracted from the pooled droplets from cultured MSP plates (hereafter called MSP pool). A total of 38,056 and 37,137 Pair-end reads were retrieved, and after filtering and removing potential erroneous sequences, a total of 28,422 and 29,778 effective tags were obtained from OMG sample and MSP pool, respectively. These sequences represented 58,200 taxon tags that covered 141 genera, 102 families, 57 orders, or 33 classes of 15 phyla. As shown in Figure 1a, the rarefaction curves of OMG and MSP pool reached plateau after 10,000 and 5000 sequences per sample, respectively, indicating that the sequencing depth was adequate to reflect the bacterial diversity in both samples. Data analysis showed that OMG sample had much higher OTU richness than the MSP samples, At the phylum level, the relative abundances of five phyla in OMG samples and two phyla in MSP pool sample were higher than 1%

| Identification of yet-to-be cultured microbial OTUs/taxa from MSP pool
With a cutting edge of 97% sequence similarity, 99 and 353 OTUs from MSP pool and OMG sample, respectively, were recognized. Venn diagram showed that OMG and MSP shared 24 OTUs, but more OTUs were uniquely in either MSP pool or OMG sample ( Figure 2). This is one more example representing that the microbial diversities was differentially reflected with culture-dependent and -independent methods, which is generally acknowledged for that none of the current tools is able to disclose the whole picture of microbial diversity in environments (Lagier et al., 2012;Rettedal, Gumpert, & Sommer, 2014;Sommer, 2015).   (Figure 3a-e). The identification of these bacterial strains is discussed in the following sections.

| Isolation and cultivation of members from Proteobacteria
Analysis of the MSP pool data with RDP databank showed that there were 44 OTUs from Proteobacteria ( Figure 3a). These 44 OTUs were assigned to 24 genera. We found that 12 of the 24 genera had been previously observed (Butera, Ferraro (Cho, Kim, Kim, Kim, & Kim, 2010), whereas the Trabulsiela strain MSP19 was closely related to Trabulsiella strain LB10 (Fang et al., 2016) isolated from R. chinensis and O. formosanus (Chou et al., 2007). In addition, the 16S rRNA genes of strain MSP1b and MSP17b were similar to those of Dyella strains isolated from R. chinensis with sequence divergence less than 2%, and Burkholderia strains MSP23 and MSP32 were similar to strains TM6 and TSB14 from R. chinensis (Fang et al., 2016).

| Isolation and cultivation of members from Actinobacteria, Bacteroidetes, and other phyla
As showed in Table 1, there were 11 OTUs that corresponding to seven genera of Actinobacteria (Figure 3c), and 14 OTUs that corresponding to seven genera of Bacteroidetes (Figure 3d). Four strains were obtained (

| D ISCUSS I ON
In this study, we continued our previous efforts to cultivate microbes from the termite gut (Chen et al., 2012;Fang et al., 2015Fang et al., , 2016 by application of the newly developed MSP method (Jiang et al., 2016).
As a microfluidic technology (Ma et al., 2014;Tandogan et al., 2014), MSP method enables high-throughput single-cell cultivation of diverse bacterial groups and even rare species from environmental samples (Jiang et al., 2016). Comparing with other cultivation tools such as extinction-culturing-based method (Colin et al., 2013(Colin et al., , 2015Connon & Giovannoni, 2002), MSP technology has higher throughput as one culture-plate can harbor thousands of droplets, whereas the extinction cultivation method carried only a few hundreds of     (Keller & Zengler, 2004;Nichols et al., 2010) was reported to be applicable under exclusively aerobic condition, whereas the MSP approach can be used both aerobically and anaerobically. The MSP technique can be further exploited for extended applicability as (a) cocultivation of different microorganisms for the study of symbiotic interaction (Park et al., 2011), (b) recovering functional and rare biosphere members and (c) single-cell sequencing.
We successfully cultivated a range of bacterial strains belonging to the Delftia, Comamonas, Acinetobacter, Moraxella, Luteimona, Sphingomonas, Bosea, Methylobacterium, Corynebacterium, Janibacter, Propionibacterium, and Sphingobacterium with MSP method in this study. Of note, the occurrence of these bacterial taxa in termite gut had been previously detected with molecular tools but they had not been cultivated (Butera et al., 2016;Diouf et al., 2015;Fall et al., 2007;Hongoh et al., 2005;Husseneder, Berestecky, & Grace, 2009;Matsui, Tanaka, Namihira, & Shinzato, 2012;Nakajima, Hongoh, Usami, Kudo, & Ohkuma, 2005;Thong-On et al., 2012;Visser, Nobre, Currie, Aanen, & Poulsen, 2012;Zhu et al., 2012 represented an unclassified member of Proteobacteria, and its 16S RNA gene showed 90% similarity to Mailhella massiliensis (Ndongo et al., 2017). The OTU93 recovered in MSP pool represented a member of Acidobacteria, and its 16S RNA gene showed 94% similarity to Vicinamibacter silvestris (Huber et al., 2016); the OTU97 represented an unclassified member of Phycisphaerae and its closest relative is Tepidisphaera mucosa (Kovaleva et al., 2015) (Their 16S rRNA gene similarity is 91%); More interestingly, the OTU98 recovered from MSP pool in this study together with 16S rRNA gene sequences detected in Formosan subterranean termite (Husseneder et al., 2009) and Reticulitermes speratus  clustered to a unique lineage of Verrucomicrobia (Figure 3e). So far, there is not any bacterial culture showing 16S rRNA gene sequence similarity higher than 80% to this unique lineage. Although we had not obtained pure cultures of those above OTUs, their occurrence in MSP pool indicated they did grow in MSP droplets. Further efforts to optimize their growth in droplets would result in finally obtaining their pure cultures.
Several strains that represent potential novel taxa were isolated with MSP method. Dysgonomonas (Hofstad et al., 2000) belongs to Bacteroidetes, and four species of Dysgonomonas were isolated from clinical specimen. Recently, two new species were isolated and characterized from termite guts, Dysgonomonas macrotermitis (Yang et al., 2014) and Dysgonomonas termitidis (Pramono, Sakamoto, Iino, Hongoh, & Ohkuma, 2015). In this study, we obtained three isolates, as represented by strain MSP50, and they are phylogenetically close to Dysgonomonas. The 16S rRNA gene of MSP50 showed 95% similarity to Dysgonomonas gadei. Whether MSP50 represents a novel species within Dysgonomonas or a novel genus within Bacteroidetes needs additional taxonomic studies.
There are many other bacterial taxa that existed in termite gut but have not been successfully cultivated in this study, such as the members of Spirochaetes and Elusimicrobia (see Supplementary material Table S1). Spirochaetes widely occur in wood-feeding termites and are the most abundant bacterial symbionts in Reticulitermes (Brune & Dietrich, 2015;Graber & Breznak, 2005;Noda, Ohkuma, Yamada, Hongoh, & Kudo, 2003). Elusimicrobia are found almost exclusively in the intestinal tract of animals and are particularly abundant in lower termites, where they reside as intracellular symbionts in the cellulolytic gut flagellates (Geissinger, Herlemann, Mörschel, Maier, & Brune, 2009;Ikeda-Ohtsubo, Faivre, & Brune, 2010;Zheng, Dietrich, Radek, & Brune, 2016). The rarefaction curves of observed taxa in Figure 1a indicated that the achieved isolates can only recover a fraction of total species in termite gut. However, the observed recovery rate might be further improved by utilization of various cultural media and different culture conditions. Specifically, further efforts should be made to apply low but diverse nutrients in culture broth, besides 1/5 R2A, at various aerobic and anoxic levels for harvest of additional bacterial groups that are fastidious to nutrients and sensitive to oxidoreductive states.
In this study, 99 OTUs were identified to be cultivable with culture-dependent MSP method, whereas 353 OTUs were detected from the termite gut microbiota sample (OMG) using the cultureindependent metagenomic method. This result showed that, only a fraction of the cultivable taxa (24 of 99 OTUs) was detectable with metagenomic method. Similar observation was reported by Lagier et al. (2012). In their study, 340 bacterial species were cultured from human gut using the MALDI-TOF-based culturomic strategy, and only 15% (51 species) of these cultivable isolates were detected by metagenomic pyrosequencing. The results manifested that culturomics complemented metagenomic by overcoming the depth bias inherent in metagenomic approaches (Lagier et al., 2012). Later on, Rettedal et al. (2014) demonstrated that the recovery rate and representativeness of culture-dependent approaches in gut microbiota could be further improved by careful design of culture conditions. It is believed that, by optimization of the culture conditions, MSP method would have better performance in microbiome recovery.

ACK N OWLED G M ENT
This research is supported by grant from National Nature Science Foundation of China (Grant No. 31670102).

CO N FLI C T O F I NTE R E S T
None declared.